text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'Three types of nonlinear Schr[ö]{}dinger models with multiple length scales are considered. It is shown that the length-scale competition universally results into arising of new localized stationary states. Multistability phenomena with a controlled switching between stable states become possible.'
bibliography:
- 'scale.bib'
---
Scale competition in nonlinear Schr[ö]{}dinger models
=====================================================
Introduction
------------
The basic dynamics of deep water and plasma waves, light pulses in nonlinear optics and charge and energy transport in condensed matter and biophysics [@cit1; @cit2; @cit3; @cit4] is described by the fundamental nonlinear Schr[ö]{}dinger (NLS) equation $$i\frac{\partial}{\partial\,t} \psi+L^2\partial^{2}_x\psi+
V|\psi|^{2}\psi+f(x)\psi=0 \; ,$$ where $\psi(x,t)$ is the complex amplitude of quasi-monochromatic wave trains or the wave function of the carriers. The second term represents the dispersion and $L$ is the dispersion length (e.g. in the theory of charge (energy) transfer $L^2=\hbar^2/2m$ with $m$ being an effective mass). The nonlinear term, $V|\psi|^{2}\psi$, describes a self-interaction of the quasiparticle caused either by its interaction with low-frequency excitations (phonons, plasmons, etc.) [@cit9] or by the intensity dependent refractive index of the material (Kerr effect) [@cit10]. The function $f(x)$ is a parametric perturbation which can be a localized impurity potential, a disorder potential, a periodic refractive index, an external electric field, etc. It is well known that as a result of competition between dispersion and nonlinearity nonlinear waves with properties of particles, solitons, arise. One can also say that this competition leads to the appearance of the new length-scale: the width of the soliton $\zeta=L/\sqrt{V}$. The presence of the parametric perturbation $f(x)$ introduces additional interplays between nonlinearity, dispersion and perturbations. In the recent paper by Bishop [*et al.*]{} [@cit5] the concept of competing length-scales and time-scales was emphasized. In particular, Scharf and Bishop [@Scharf1; @Scharf2] have discussed the effects of a periodic potential ($f(x)=\epsilon \cos(2\pi x/\zeta_p)$) on the soliton of the NLS equation, and shown on the basis of an averaged NLS equation that for $\zeta_p/\zeta {\raise.3ex\hbox{$<$\kern-.75em\lower1ex\hbox{$\sim$}}}1$ or ${\raise.3ex\hbox{$>$\kern-.75em\lower1ex\hbox{$\sim$}}}1$ the periodic potential leads to a simple renormalization of the solitons and creates a ’dressing’ of the soliton. But when $\zeta_p \sim \zeta$ there is a crucial length-scale competition which leads to the destruction of the soliton. Another interesting example of the length-scale competition was provided by Ref. [@cit6] where the authors showed that the propagation of intense soliton-like pulses in systems described by the one-dimensional NLS equation may be left practically unaffected by the disorder (when $f(x)$ is a Gaussian $\delta$-correlated process). This theoretical prediction has recently been confirmed experimentally using nonlinear surface waves on a superfluid helium film [@cit7].
The goal of this paper is to extend the concept of the length-scale competition to the essentially non-integrable systems: to systems with nonlocal dispersion and to systems with unstable stationary states.
Excitations in nonlinear Kronig-Penney models
---------------------------------------------
Wave propagation in nonlinear photonic band-gap materials and in periodic nonlinear dielectric superlattices [@phot1; @phot2] consisting of alternating layers of two dielectrics: nonlinear and linear, is governed by the NLS equation $$\begin{aligned}
i\partial_t \psi(x,z,t) &+& \zeta^2\,(\partial^2_x+\partial^2_z)
\psi(x,z,t) \nonumber \\
&+& w\,\sum_n\,\delta (x-x_n)|\psi(x,z,t)|^2\psi(x,z,t)=0 \; ,
\label{eq0}\end{aligned}$$ where $x_n=n\,\ell$ is the coordinate of the n-th nonlinear layer ($\ell$ is the distance between the adjacent nonlinear layers), and it is assumed that the width $w$ of the nonlinear layer is small compared to the soliton width $\zeta$ within the layer. In this case the problem can be described by the nonlinear Kronig-Penney model given by Eq. (\[eq0\]). It was shown in Ref. [@gaid] that the field $\psi(x,z,t)$ can be expressed in terms of the complex amplitudes $\,\psi_n(z,t)\equiv \psi(x_n,z,t)$ at the nonlinear layers. The complex amplitudes $\psi_n(z,t)$ can be found from the set of pseudo-differential equations $$\begin{aligned}
\frac{\zeta^2 \hat{\kappa}}{\sinh{\ell\hat{\kappa}}}
(\psi_{n+1}+\psi_{n-1})- \frac{2 \zeta^2 \hat{\kappa}}{
\tanh{\ell\hat{\kappa}}}\,\psi_n + w |\psi_n|^2 \, \psi_n =0
\label{eq5}\end{aligned}$$ with periodic boundary conditions $\psi_{n+N}=\psi_n$, where $N$ is the number of layers. In Eq. (\[eq5\]) the operator $\hat{\kappa}$ is defined as $\hat{\kappa}\psi= \zeta^{-1} \sqrt{-i\,\partial_t-\zeta^2\,
\partial_z^2}\psi$. Equation (\[eq0\]) has an integral of motion – the norm (in nonlinear optics this quantity is often called the power) $P=\int_{-\infty}^{\infty} |\psi|^2 dx dz$.
For the excitation pattern where the complex amplitudes are the same in all nonlinear layers, $\psi_n(z,t)=\Psi(z,t)$, we get $$\begin{aligned}
\sqrt{-i\partial_t-\zeta^2\,\partial^2_z}\,\tanh\left(
\frac{\ell}{2\zeta}\sqrt{-i\partial_t - \zeta^2 \, \partial^2_z}
\right)\Psi-\frac{w}{2\zeta}|\Psi|^2\Psi=0 \; .
\label{eq7}\end{aligned}$$ Equation (\[eq7\]) clearly shows the existence and competition of two characteristic length-scales: the interlayer spacing $\ell$ and the size of the soliton in the nonlinear layer $\zeta$. When $\ell\ll \zeta$ one can expand the hyperbolic tanhence and Eq. (\[eq7\]) takes the form of usual NLS equation $$\begin{aligned}
\left(i\partial_t+\zeta^2\,\partial^2_z\right)\Psi+
\frac{w}{\ell} |\Psi|^2\Psi=0 \; .
\label{eq8}\end{aligned}$$ In the opposite limit, when $\ell\gg\zeta$ and $\tanh\left(
\frac{\ell}{2\zeta}\sqrt{\cdots}\right)\simeq 1$, equation (\[eq7\]) takes the form $$\begin{aligned}
\sqrt{-i\partial_t-\zeta^2\,\partial^2_z}\Psi-
\frac{w}{2\zeta} |\Psi|^2\Psi=0 \; ,
\label{eq9}\end{aligned}$$ which reduces, for static distribution ($\partial_t \Psi=0$), to the nonlinear Hilbert-NLS equation recently introduced in Ref. [@gmcr]. It is noteworthy that in contrast to usual NLS solitons, the localized solutions of the nonlinear Hilbert-NLS equation have algebraic tails [@gmcr].
It is worth to note the close relation of the problem under consideration to the theory of the long internal gravity waves in a stratified fluid with a finite depth $h$ (see e.g. [@kkd]) which are described by the equation $$\begin{aligned}
\partial_t u+\frac{1}{h}\partial_x u+2u\partial_x u+
T\partial^2_x u=0 \; , \end{aligned}$$ where $T(\cdot)$ is the singular integral operator given by $$\begin{aligned}
(Tf)(x)=\frac{1}{2h} \, p.v. \!\!\! \int\limits_{-\infty}^{\infty}
\coth \left( \frac{\pi(y-x)}{2h} \right) f(y) dy\end{aligned}$$ ($p.v.$ means the principal value integral). In the shallow water limit ($h\rightarrow 0$) the dynamics is described by the Korteweg-de-Vries equation, $\partial_t u+\frac{h}{3}\partial^3_x u+2u\partial_x u=0$, while the Benjamin-Ono equation, $\partial_t u+H\partial^2_x u+2u\partial_x u=0$, governs the water wave motion in the deep-water limit ($h\rightarrow \infty$). Here $(Hf)(x)=\frac{1}{\pi} \, p.v. \!\!\! \int
\limits_{-\infty}^{\infty} dy f(y)/(y-x)$ is the Hilbert transform.
Being interested in stationary states of the system we consider solutions of the form $\Psi(z,t)= \phi(z) \exp(i \Lambda t)$, where $\Lambda$ is the nonlinear frequency and $\phi(z)$ is the real shape function. Since Eq. (\[eq5\]) is Galilean invariant, standing excitations can always be Galileo boosted to any velocity in $z$-direction. For the shape function $\phi(z)$ we obtain a nonlinear eigenvalue problem in the form $$\begin{aligned}
\sqrt{\Lambda -\zeta^2\,\partial^2_z}\,\tanh\left(
\frac{\ell}{2\zeta}\sqrt{\Lambda - \zeta^2\,\partial^2_z}
\right)\phi-\frac{w}{2\zeta}\phi^3=0 \; .
\label{eq11}\end{aligned}$$ Simple scaling arguments show that in the low-frequency limit ($\Lambda \ell^2/ \zeta^2 \rightarrow 0$) the norm behaves in the same way as in the case of usual NLS equation (\[eq8\]): $P\sim \sqrt{\Lambda}$. When $\Lambda \ell^2/ \zeta^2\rightarrow
\infty$ the norm $P$ is a monotonically decreasing function: $P\sim 1/ \sqrt{\Lambda}$. From the analysis of Ref. [@gaid] follows that the norm $P(\Lambda)$ is a nonmonotonic function with a local maximum at $\Lambda_m \approx 1.25 \zeta/ \ell$ (see Fig. \[fig1\]). Thus, the stationary states exist only in a finite interval, $0\leq P \leq P(\Lambda_m)$ and for each value of norm in this interval there are two stationary states. This is an intrinsic property of the nonlinear Schr[ö]{}dinger superlattice system.
Discussing the stability of the stationary states satisfying Eq. (\[eq11\]), there are two sources of instability to be considered: longitudinal and transversal perturbations. The perturbations of the first type are of the same symmetry with respect to transversal degrees of freedom as the stationary states of Eq. (\[eq11\]), while the second type of perturbations breaks this symmetry. It was shown in [@gaid] that stationary states which correspond to the branch with $dP / d \Lambda <0$ are unstable due to the longitudinal perturbations. The states with $\Lambda > (4 \zeta^2/3 \ell^2) \sin^2 (\pi/N)$ are, in their turn, unstable due to the transversal perturbations. Thus, one can expect stable stationary solutions for nonlinear frequencies satisfying the condition $$\Lambda \,<\,\frac{\zeta^2}{\ell^2}\,\mbox{min}\{ \frac{4}{3}
\sin^2(\frac{\pi}{N}),\Lambda_m\} \; .
\label{eq12}$$ In particular, this means that the stationary state $\psi_n(z,t)=e^{i\Lambda t}\phi(z)$ can neither exist in the case of only one nonlinear layer ($\ell \rightarrow \infty $) nor in the quasi-continuum limit ($N \rightarrow \infty$). But in the latter case the system supports stationary states which are localized in both spatial directions (see Ref. [@gaid] for details).
Discrete NLS models with Long-Range dispersive interactions
-----------------------------------------------------------
Determination of the dynamical properties of physical systems with competition between discreteness, nonlinearity and dispersion is very important because of their wide applicability in various physical problems. Examples are coupled optical fibers, nanoscale electronic devices, nonlinear charge and excitation transport in biological macromolecules. As a result of the interplay between discreteness, dispersion and nonlinear interactions, new nonlinear excitations, namely intrinsically localized oscillatory states may appear [@sm83; @st88]. For monoatomic lattices with a nearest-neighbor harmonic interaction and quartic anharmonic interaction the localized states were found [@bkp90; @bm90] to have frequencies lying above the phonon band. In the case of one-dimensional NLS lattice [@ta89] a localized mode lying below the linear excitation band in the small amplitude limit reduces to the one-soliton solution of the continuum NLS equation.
A discrete NLS equation with “tunable” diagonal and off diagonal nonlinearities which includes the integrable Ablowitz-Ladik system [@al76] as a limit was introduced in [@ms92; @cbg94]. The existence of a crossover region which separated the linear (Anderson) localization and the nonlinear self-localization was shown for a disordered discrete NLS equation in Ref. [@cit5]. An interplay of discreteness and nonlinearity in the presence of time-dependent linear electric field perturbations was investigated in Ref. [@cai]. It was shown that nonlinear Bloch oscillations and effects of dynamical localization are generic properties of discrete NLS models. In the main part of the previous studies of the discrete NLS models the dispersive interaction was assumed to be short-ranged and a nearest-neighbor approximation was used. However, there exist physical situations that definitely can not be described in the framework of this approximation. The DNA molecule contains charged groups, with long-range Coulomb interaction ($1/r$) between them. The excitation transfer in molecular crystals [@davydov71] and the vibron energy transport in biopolymers [@scott92] are due to transition dipole-dipole interaction with $1/r^3$ dependence on the distance, $r$. The nonlocal (long-range) dispersive interaction in these systems provides the existence of additional length-scale: the radius of the dispersive interaction. We will show that it leads to the bifurcative properties of the system due to both the competition between nonlinearity and dispersion, and the interplay of long-range interactions and lattice discreteness.
In some approximation the equation of motion is the nonlocal discrete NLS equation of the form $$\begin{aligned}
i\frac{d}{d t}\psi_n+\sum_{m\neq n}J_{n-m}(\psi_m-\psi_n)+
|\psi_n|^2\psi_n=0 \; ,
\label{eq21}\end{aligned}$$ where the long-range dispersive coupling is taken to be either exponentially, $J_n=J\,e^{-\beta|n|}$, or algebraically, $J_n=J\,|n|^{-s}$, decreasing with the distance $n$ between lattice sites. In both cases the constant $J$ is normalized such that $\sum\limits_{n=1}^{\infty} J_n=1$ for all $\beta$ or $s$. The parameters $\beta$ and $s$ are introduced to cover different physical situations from the nearest-neighbor approximation ($\beta \rightarrow \infty, \; s \rightarrow
\infty$) to the quadrupole-quadrupole ($s=5$) and dipole-dipole ($s=3$) interactions. The Hamiltonian $H=\sum\limits_{n,m} J_{n-m} |\psi_n-\psi_m|^2-\frac{1}{2}
\sum\limits_n |\psi_n|^4$, which corresponds to the set of equations (\[eq21\]), and the number of excitations $N=\sum\limits_n |\psi_n|^2$ are conserved quantities.
We are interested in stationary solutions of Eq. (\[eq21\]) of the form $\psi_n(t)=\phi_n \exp (i \Lambda t)$ with a real shape function $\phi_n$ and a frequency $\Lambda$. This gives the governing equation for $\phi_n$ $$\label{23}
\Lambda \phi_n= \sum_{m \neq n} J_{n-m} (\phi_m-\phi_n)+
\phi_n^{3} \; ,$$ which is the Euler-Lagrange equation for the problem of minimizing $H$ under the constraint $N=constant$.
Figure \[fig2\] shows the dependence $N(\Lambda)$ obtained from direct numerical solution of Eq. (\[23\]) for algebraically decaying $J_{n-m}$. A monotonic function is obtained only for $s>s_{cr}$. For $2<s<s_{cr}$ the dependence becomes nonmonotonic (of ${\cal N}$-type) with a local maximum and a local minimum. These extrema coalesce at $s=s_{cr} \simeq 3.03$. For $s<2$ the local maximum disappears. The dependence $N(\Lambda)$ obtained analytically using the variational approach is in a good qualitative agreement with the dependence obtained numerically (see [@gmcr]). Thus the main features of all discrete NLS models with dispersive interaction $J_{n-m}$ decreasing faster than $|n-m|^{-s_{cr}}$ coincide qualitatively with the features obtained in the nearest-neighbor approximation where only one stationary state exists for any number of excitations, $N$. However in the case of long-range nonlocal NLS equation (\[eq21\]), i.e. for $2<s<s_{cr}$, there exist for each $N$ in the interval $[N_{l}(s), N_{u}(s)]$ three stationary states with frequencies $\Lambda_{1}(N) < \Lambda_{2}(N) < \Lambda_{3}(N)$. In particular, this means that in the case of dipole-dipole interaction ($s=3$) multiple solutions exist. It is noteworthy that similar results are also obtained for the dispersive interaction of the exponentially decaying form. In this case the bistability takes place for $\beta\,\leq\,1.67$. According to the theorem which was proven in [@lst94], the necessary and sufficient stability criterion for the stationary states is $dN /d\Lambda > 0$. Therefore, we can conclude that in the interval $[N_{l}(s), N_{u}(s)]$ there are only two linearly stable stationary states: $\Lambda_{1}(N)$ and $\Lambda_{3}(N)$. The intermediate state is unstable since $dN /d\Lambda < 0$ at $\Lambda=\Lambda_2$.
At the end points ($\Lambda(N_l)$ and $\Lambda(N_u)$) the stability condition is violated, since $(dN /d\Lambda)_s$ vanishes. Constructing the locus of the end points we obtain the curve that is presented in Fig. \[fig3\]. This curve bounds the region of bistability. It is analogous to the critical curve in the van der Waals’ theory of liquid-vapor phase transition [@ll59]. Thus in the present case we have a similar phase transition like behavior where two phases are the continuum states and the intrinsically localized states, respectively. The analog of the temperature is the dispersive parameter $s(\beta)$.
The shapes of three stationary states in the interval of bistability differ significantly (see Fig. \[fig4\]). The low frequency states are wide and continuum-like while the high frequency solutions represents intrinsically localized states with a width of a few lattice spacings. It can be obtained [@gmcr] that the inverse widths of these two stable states are $\alpha_1\, \approx \, \left( N/8J \right)^{1/(s-2)}=
\left( N/8J \right)^{\ln \xi/(1-2\ln \xi)}, \; \; \;
\alpha_3\, \approx \, \ln \left( N/J \right)$ with $\xi=\exp(1/s)$ being the characteristic length scale of the dispersive interaction which is defined as a distance (expressed in lattice spacings) at which the interaction decreases in two times. It is seen from these expressions that the existence of two so different soliton states for one value of the excitation number, $N$, is due to the presence of two different length scales in the system: the usual scale of the NLS model which is related to the competition between nonlinearity and dispersion (expressed in terms of the ratio $N/J$ ) and the range of the dispersive interaction $\xi$.
Having established the existence of bistable stationary states in the nonlocal discrete NLS system, a natural question that arises concerns the role of these states in the full dynamics of the model. In particular, it is of interest to investigate the possibility of switching between the stable states under the influence of external perturbations, and to clear up what type of perturbations can be used to control the switching. Switching of this type is important for example in the description of nonlinear transport and storage of energy in biomolecules like the DNA, since a mobile continuum-like excitation can provide action at distance while the switching to a discrete, pinned state can facilitate the structural changes of the DNA [@geor96]. As it was shown recently in [@mj98], switching will occur if the system is perturbed in a way so that an internal, spatially localized and symmetrical mode (’breathing mode’) of the stationary state is excited above a threshold value.
We will in sequel mainly discuss the case when the matrix element of excitation transfer, $J_{n-m}$, decreases exponentially with the distance $|n-m|$. For $\beta=1$ the multistability occurs in the interval $3.23\,\leq\,N\,\leq \,3.78$. It is worth noticing, however, that the scenario of switching described below remains qualitatively unchanged for all values of $\beta\,\leq\,1.67$, and also for the algebraically decaying dispersive coupling with $2\,\leq\,s\,\leq\,3.03$.
An illustration of how the presence of an internal breathing mode can affect the dynamics of a slightly perturbed stable stationary state is given in Figs. \[fig5\] and \[fig6\]. To excite the breathing mode, we apply a spatially symmetrical, localized perturbation, which we choose to conserve the number of excitations in order not to change the effective nonlinearity of the system. The simplest choice, which we have used in the simulations shown here, is to kick the central site $n_0$ of the system at $t=0$ by adding a parametric force term of the form $\theta\delta_{n,n_0}\delta(t)\psi_n(t)$ to the left-hand-side of Eq. (\[eq21\]). As can easily be shown, this perturbation affects only the site $n_0$ at $t=0$, and results in a ’twist’ of the stationary state at this site with an angle $\theta$, i.e. $\psi_{n_0}(0)=\phi_{n_0}\,e^{i\theta}$. The immediate consequence of this kick is, as can been deduced from the form of Eq. (\[eq21\]), that $\frac{d}{dt}\left(|\psi_{n_0}|^2\right)$ will be positive (negative) when $\theta\,>\,0$ ($\theta\,<\,0$). Thus, we choose $\theta\,>\,0$ to obtain switching from the continuum-like state to the discrete state, while we choose $\theta\,<\,0$ investigating switching in the opposite direction. We find that in a large part of the multistability regime there is a well-defined threshold value $\theta_{th}$: when the initial phase torsion is smaller than $\theta_{th}$, periodic, slowly decaying ’breather’ oscillations around the initial state will occur, while for strong enough kicks (phase torsions larger than $\theta_{th}$) the state switches into the other stable stationary state.
It is worth remarking that the particular choice of perturbation is not important for the qualitative features of the switching, as long as there is a substantial overlap between the perturbation and the internal breathing mode. We believe also that the mechanism for switching described here can be applied for any multistable system where the instability is connected with a breathing mode.
Stabilization of nonlinear excitations by disorder
--------------------------------------------------
In this section we discuss disorder effects in NLS models. Usually the investigations of disorder effects have been carried out on systems that are integrable - soliton bearing - in the absence of disorder. A common argument is that the equations, despite their exact integrability, provide a sufficient description of the physical systems to display the essential behavior. However, the more common physical situation is that integrability, and thus the exact soliton, is absent. A relevant example of such an equation is the two-dimensional (or higher-dimensional) NLS equation. The two-dimensional NLS equation is nonintegrable and possesses an unstable ground state solution which, in the presence of perturbations, either collapses or disperses (see e.g. [@af50-a; @af50-b]).
We consider a quadratic two-dimensional lattice with the lattice spacing equal to unity. The model is given by the Lagrangian $$\begin{aligned}
\label{lagr}
L=\frac{i}{2} \sum_{n,m} \left(\psi_{n,m}^{*}\, \frac{d}{dt}
\psi_{n,m}-c.c. \right)-H \; , \end{aligned}$$ where $$\begin{aligned}
H &=& \sum_{n,m}\left(|\psi_{n+1,m}-\psi_{n,m}|^2+|\psi_{n,m+1}-
\psi_{n,m}|^2 \right. \nonumber \\
&-& \left. \frac{1}{2}|\psi_{n,m}|^4 -\epsilon_{n,m}|\psi_{n,m}|^2
\right)
\label{hamil}\end{aligned}$$ is the Hamiltonian of the system. In Eqs. (\[lagr\]) and (\[hamil\]) $(n,m)$ is the lattice vector ($n$ and $m$ are integer). The first two terms in Eq. (\[hamil\]) correspond to the dispersive energy of the excitation, the third term describes a self-interaction of the excitation and the fourth term represents diagonal disorder in the lattice. Here the random functions $\epsilon_{n,m}$ are assumed to have Gaussian distribution with the probability $p(\epsilon_{n,m})=\frac{1}{\eta \sqrt{\pi}}
\exp[-(\epsilon_{n,m}/\eta)^2]$ and have the autocorrelation function $\langle \epsilon_{n,m}\epsilon_{n',m'}\rangle =
\eta^2\delta_{n\,n'}\delta_{m\,m'}$, where the brackets $\langle .... \rangle$ denote averaging over all realizations of the disorder. From the Lagrangian (\[lagr\]) we obtain the equation of motion for the excitation function in the form $$\begin{aligned}
i \frac{d}{dt} \psi_{m,n} &+& (\psi_{m,n-1}+\psi_{m,n+1}+
\psi_{m+1,n} +\psi_{m-1,n}-4\psi_{m,n}) \nonumber \\
&+& |\psi_{m,n}|^{2}\psi_{m,n}+
\epsilon_{m,n}\psi_{m,n}=0 \; .
\label{deq}\end{aligned}$$ Equation (\[deq\]) conserves the norm $N=\sum\limits_{n,m}
|\psi_{n,m}|^2$ and the Hamiltonian $H$.
We are interested in the stationary solutions of Eq. (\[deq\]) of the form $$\begin{aligned}
\psi_{n,m}(t)=\phi_{n,m}\exp(i\Lambda t) \; ,
\label{stat}\end{aligned}$$ with a real shape function $\phi_{n,m}$ and a nonlinear frequency $\Lambda$.
Equation (\[deq\]) together with Eq. (\[stat\]) constitute a nonlinear eigenvalue problem which can be solved numerically using the techniques described in Ref. [@af89]. The dependences $N(\Lambda)$ in the absence and in the presence of disorder are shown in Fig. \[fig7\]. It has previously been shown [@af158; @af159; @lst94; @af47] that the linear stability of the stationary states in the discrete case is determined by the condition $dN/d\Lambda>0$. Thus, in the case without disorder (solid curve in Fig. \[fig7\]) the low-frequency ($0\leq \Lambda\leq
\Lambda_{min}=1.088$) nonlinear excitations in the discrete two-dimensional NLS model are unstable. It is important that in the continuum limit ($\Lambda\rightarrow 0$) the norm $N(\Lambda)$ tends to the non-zero value $N_c\simeq 11.7$.
Other lines in Fig. \[fig7\] show the dependence $N$ on $\Lambda$ for the stationary solutions of Eq. (\[deq\]) in the presence of disorder. The results have been obtained as averages of $150$ realizations of the disorder. Several new features arise as a consequence of the disorder. In the continuum limit ($\Lambda\rightarrow 0$) we no longer have $N=N_c$ with $dN/d\Lambda = 0$. Instead we have $N\rightarrow 0$ with $dN/d\Lambda >0$ signifying that the disorder stabilizes the excitations in the low-frequency limit. The disorder creates a stability window such that a bistability phenomenon emerges. Consequently there is an interval of the excitation norm in which two stable excitations with significantly different widths have the same norm.
Furthermore, we see that the disorder creates a gap at small $\Lambda$ in which no localized excitations can exist, and that the size of this gap apparently is increased as the variance of the disorder is increased. It is also clearly seen that as $\Lambda$ increases (decreasing width) the effect of the disorder vanishes so that the very narrow excitations are in average unaffected by the disorder. It is important to stress that this is an average effect, because for each realization of the disorder the narrow excitation will be affected. The narrow excitation will experience a shift in the nonlinear frequency equal to the amplitude of the disorder at the position of the excitation.
The bistability we observe in Fig. \[fig7\] occurs due to the competition between two different length scales of the system: one length scale being defined by the relation between the nonlinearity and the dispersion, while the other length scale being defined by the disorder. A similar effect was observed by Christiansen [*et al.*]{} [@cg97] for the one-dimensional discrete NLS equation with a quintic nonlinearity. The latter is quite natural because as it is well known (see e.g. [@ob94]) the properties of the two-dimensional NLS model with a cubic nonlinearity are similar to the properties the one-dimensional NLS equation with a quintic nonlinearity.
Having studied the stationary problem it is vital to compare the results to full dynamical simulations. Therefore we carry out a numerical experiment launching a pulse in a system governed by Eq. (\[deq\]). Specifically, stationary solutions (\[stat\]) of Eq. (\[deq\]) with $\Lambda=0.14$ (after reducing the amplitude of these solutions by $5$%) were used as initial conditions of the dynamical simulations. Examples of the described experiment are shown in Fig. \[fig8\]. As is seen the pulse behavior in the absence of disorder and in the presence of disorder (we presented here a realization corresponding to the disorder variance $\eta=0.1$) differs drastically. While the pulse rapidly disperses in the ideal system (the contour plot for $t=250$ is absent because the pulse width is of the system size), the process is arrested in the disordered system. After some transient behavior the excitation stabilizes and attains an approximately stationary width. The dynamical simulations thus support the conclusion that otherwise unstable excitations are stabilized by the presence of disorder in the low frequency limit.
Analytical theory of soliton states in disordered NLS models based on the collective coordinate approach and on the Rice’s theorem from the theory of random processes [@kree] is presented in Refs. [@cg97] and [@gh98].
Summary
-------
In summary we have shown that the presence of competing length scales leads to multistability phenomena in nonlinear Schr[ö]{}dinger models. We have analyzed three types of the NLS models. The nonlinear Schr[ö]{}dinger-Kronig-Penney model presents an example where two competing length scales exist: the width $\zeta$ of the soliton in the nonlinear Schro[ö]{}dinger equation and the interlayer spacing $\ell$. Due to the interplay between these two length scales the localized stationary states exist only in a finite interval of the excitation power. Two branches of stationary states exist but only the low-frequency branch is stable.
In discrete nonlinear Schr[ö]{}dinger models with long-range dispersive interactions there exist three types of length scales: the soliton width, the lattice spacing and the radius of the dispersive interaction. Here the competition of the length scales provides the existence of three branches of stationary states. Two of them: low-frequency branch which contains continuum-like excitations and high-frequency one with intrinsically localized excitations, are stable. It is shown that a controlled switching between narrow, pinned states and broad, mobile states is possible. The particular choice of perturbation is not important for the qualitative features of the switching, as long as there is a substantial overlap between the perturbation and the internal breathing mode. The switching phenomenon could be important for controlling energy storage and transport in DNA molecules.
Considering nonlinear excitations in two-dimensional discrete nonlinear Schr[ö]{}dinger models with disorder it was found that otherwise unstable continuum-like excitations can be stabilized by the presence of the disorder. For the very narrow excitations the disorder has no effect on the averaged behavior. Bistability which was observed in this case is very similar to the bistability that occurs in nonlocal NLS models. Here the bistability arises on similar grounds because of competition between the solitonic length scale and the length scale defined by the disorder.
Yu.B.G. thanks MIDIT, the Technical University of Denmark for the hospitality. Yu.B.G. and S.F.M. acknowledge support from the Ukrainian Fundamental Research Fund under grant 2.4/355.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this article, we show that all admissible rational maps with fixed or period two cluster cycles can be constructed by the mating of polynomials. We also investigate the polynomials which make up the matings that construct these rational maps. In the one cluster case, one of the polynomials must be an $n$-rabbit and in the two cluster case, one of the maps must be either $f$, a “double rabbit”, or $g$, a secondary map which lies in the wake of the double rabbit $f$. There is also a very simple combinatorial way of classifiying the maps which must partner the aforementioned polynomials to create rational maps with cluster cycles. Finally, we also investigate the multiplicities of the shared matings arising from the matings in the paper.'
address: 'University of Warwick, Coventry, CV4 7AL, UK'
author:
- Thomas Sharland
bibliography:
- 'papers.bib'
title: Constructing rational maps with cluster points using the mating operation
---
Introduction
============
The study of holomorphic dynamical systems was first given serious study by Pierre Fatou [@Fatou:1919; @Fatou:1920] and Gaston Julia [@Julia:1918] in the earlier part of the 20th Century. After laying relatively dormant for a number years, the subject was given a new lease of life due to the improvements in technology which allowed mathematicians to view the various sets under investigation. In the mid-1980s, Hubbard and Douady produced their now famous “Orsay lecture notes” [@DouadyHubbard:Orsay1; @DouadyHubbard:Orsay2], and since then the subject has grown from strength to strength.
This paper is a partner to [@Thurstoneq], in which it was shown that, in certain cases, a very simple set of combinatorial data can classify (in the sense of Thurson equivalence) a rational map with a periodic cluster cycle. Like its partner paper, this paper is made up of results found in the author’s PhD thesis [@Mythesis]. Essentially, the former paper showed the uniqueness part of the classification of maps with cluster cycles, whereas the current paper is concerned with the existence part. We will show that all realisable combinatorial data for rational maps can be realised by matings of polynomials and so, when combined with the results of [@Thurstoneq], we will show that all maps with cluster cycles are matings.
Definitions
-----------
Let $f \colon {\widehat{{\mathbb{C} }} }\to {\widehat{{\mathbb{C} }} }$ be a rational map on the Riemann sphere, of degree at least 2. The Julia set $J(f)$ will be the closure of the set of repelling periodic points of $f$, and the Fatou set is the set $F(f) = {\widehat{{\mathbb{C} }} }\setminus J(f)$. The connected components of $F(f)$ are called Fatou components. In the case where the critical orbits are periodic, we call the immediate basins of the (super)attracting orbit critical orbit Fatou components. We concern ourselves in the main with bicritical rational maps with marked critical points and such that the two critical points belong to the attracting basins of two disjoint periodic orbits of the same period. If $f$ is a polynomial, then we define the filled Julia set $K(f)$ to be the set of points that are not attracted by the superattracting fixed point at infinity. It is well-known that $\partial K(f) = J(f)$.
Let $F \colon {\widehat{{\mathbb{C} }} }\to {\widehat{{\mathbb{C} }} }$ be a bicritical rational map. Then a cluster point for $F$ is a point in $J(F)$ which is the endpoint of the angle $0$ internal rays of at least one critical orbit Fatou component from each of the two critical cycles. We will define a cluster to be the union of the cluster point and the Fatou components meeting at it. The period of the cluster will be the period of the cluster point. The star of a cluster will be the union of the cluster point and the associated $0$ internal rays, including the points on the critical orbit.
A standard example of a (fixed) cluster cycle is the rational map formed by the mating of the rabbit with the airplane, see Figure \[f:rabair\]. Notice that, by our restriction to bicritical maps, the rational maps in this paper can have at most one cluster cycle.
![A rational function with a fixed cluster point. This map is formed by the mating of the rabbit with the airplane.[]{data-label="f:rabair"}](rabairmated.eps){width="95.00000%"}
We will define the critical displacement of a rational map with either a fixed cluster point or a period two cluster cycle. Informally, the critical displacement will give a (combinatorial) measure as to how far apart the two critical orbits are in the clusters. In actual fact, we require a different definition of this definition depending on whether we are studying the fixed case or the period two case. The reason for this is due to the result of the following lemma, which was proved in [@Thurstoneq].
\[p:Marysresult\] There does not exist a rational map $F$ with a period 2 cluster cycle such that the critical points are in the same cluster.
The two definitions of critical displacement are the following. It is clear in both cases that the critical displacement is an odd integer, and that the definition is dependent on the labelling of the critical points of the map.
Let $F$ be a rational map with a fixed cluster point. Label the endpoints of the star as follows. Let $e_0$ be the first critical point, and label the remaining arms in anticlockwise order by $e_1,e_2,\ldots,e_{2n-1}$. Then the second critical point is one of the $e_j$, and we call $j$ the critical displacement of the cluster of $F$. We denote the critical displacement by $\delta$.
We will sometimes use the fact that the critical displacement can equally well be calculated as the combinatorial distance between the critical values, as opposed to the critical points. Also, notice that if one choice of marking for the critical points gives $\delta = k$, then the alternative marking will give a critical displacement of $\delta = -k$. Clearly, in light of Lemma \[p:Marysresult\], any attempt to use the above definition to define the critical displacement in the period two case is impossible, hence we give the following definition for the period two case.
Let $F$ be a rational map with a period two cluster cycle. Choose one of the critical points to be $c_1$, and label the cluster containing it to be $\mathcal{C}_{1}$. Then (by Lemma \[p:Marysresult\]) the other critical point $c_2$ is in the second cluster $\mathcal{C}_2$. We define the critical displacement $\delta$ as follows. Label the arms in the star of $\mathcal{C}_1$, starting with the arm with endpoint $c_1$, in anticlockwise order $\ell_0, \ell_1, \ldots, \ell_{2n-1}$. Then $F(c_2)$ is the endpoint of one of the $\ell_k$. This integer $k$ is the critical displacement, which we again denote by $\delta$.
For the period two definition, we will later be using the fact that the critical displacement is also equal to the combinatorial distance between the $F(c_1)$ and $F^{\circ 2}(c_2)$. In contrast to the fixed case, there is not a neat symmetry to the two possible critical displacements a rational map can have. We will discuss this in further detail in Section \[s:shared\].
\[d:CRN\] Let $F \colon {\widehat{{\mathbb{C} }} }\to {\widehat{{\mathbb{C} }} }$ be a bicritical rational map and let $c$ be a cluster point of period $n$ of $F$. Then the combinatorial rotation number is defined as follows. The first return map, $F^{\circ n}$, maps the star of the cluster, $X_{F}$ to itself. Label the arms of the star (the $0$-internal rays) which belong to one of the critical orbits (it does not matter which) cyclically in anticlockwise order by $\ell_1$, $\ell_2, \ldots \ell_n$ (the initial choice of $\ell_1$ is not important). Then for each $k$, there exists $p$ such that $F^{\circ n}$ maps $\ell_{k}$ to $\ell_{k+p}$, subscripts taken modulo $n$. We then say the combinatorial rotation number is $\rho = \rho(F) = p/n$.
We can now define the combinatorial data of a rational map with a cluster cycle to be the pair $(\rho,\delta)$ where $\rho$ is the combinatorial rotation number and $\delta$ is the critical displacement. As shown in [@Thurstoneq], the combinatorial data of a map is enough to classify it (in the sense of Thurston) in the fixed case for all degrees and the period two case if the map is quadratic.
Matings
-------
We will be constructing rational maps using matings. The mating of polynomials was first mentioned by Douady in [@Douady:Bourbaki]. Informally, the construction allows us to take two complex polynomials $f$ and $g$ (along with their filled Julia sets $K(f)$ and $K(g)$) and paste them together to construct branched cover of the sphere. We will informally consider the mating operation to be a map from the ordered pairs $(f,g)$ of polynomials to the space of branched covers of the sphere. For further reading on this subject, see [@TanLei:1990; @Milnor:mating; @Wittner; @TanShish].
Let $f$ and $g$ be two monic degree $d$ polynomials defined on ${\mathbb{C} }$. In this paper, $f$ and $g$ will be unicritical but this is not needed in general. We define $\widetilde{{\mathbb{C} }} = {\mathbb{C} }\cup \{ \infty \cdot e^{2 \pi i t} : t \in {\mathbb{R} }/ {\mathbb{Z} }\}$, the complex plane with the circle at infinity. We then extend the two polynomials to the circle at infinity by defining $$f(\infty \cdot e^{2 \pi i t}) = \infty \cdot e^{2 d \pi i t} \quad \text{and} \quad g(\infty \cdot e^{2 \pi i t}) = \infty \cdot e^{2 d \pi i t}.$$ Label this extended dynamical plane of $f$ by $\widetilde{{\mathbb{C} }}_f$ and the extended dynamical plane of $g$ by $\widetilde{{\mathbb{C} }}_g$. We then create a topological sphere by gluing the two extended planes together along the circle at infinity. More formally, we define $$S^2_{f,g} = \widetilde{{\mathbb{C} }}_f \uplus \widetilde{{\mathbb{C} }}_g / \sim$$ where $\sim$ is the relation which identifies the point $\infty \cdot e^{2 \pi i t} \in \widetilde{{\mathbb{C} }}_f$ with the point $\infty \cdot e^{- 2 \pi i t} \in \widetilde{{\mathbb{C} }}_g$. We now define a new map, the formal mating $f \uplus g$, on this new space $S^2_{f,g}$ by defining $$\begin{aligned}
f \uplus g|_{\widetilde{{\mathbb{C} }}_f} \, =& \, f \quad \textrm{and} \\
f \uplus g|_{\widetilde{{\mathbb{C} }}_g} \, =& \, g. \end{aligned}$$
We now wish to define an alternative type of mating, called the *topological mating*. First we require a brief discussion on the theory of external rays. Suppose the (filled) Julia set of the degree $d \geq 2$ monic polynomial $f \colon {\widehat{{\mathbb{C} }} }\to {\widehat{{\mathbb{C} }} }$ is connected. Recall, that by Böttcher’s theorem, there is a conformal isomorphism $$\phi = \phi_f \colon {\widehat{{\mathbb{C} }} }\setminus \overline{{\mathbb{D} }} \to {\widehat{{\mathbb{C} }} }\setminus K(f)$$ which can be chosen so that it conjugates $z \mapsto z^d$ on ${\widehat{{\mathbb{C} }} }\setminus \overline{{\mathbb{D} }}$ with the map $f$ on ${\widehat{{\mathbb{C} }} }\setminus K(f)$.
\[d:extray\] Consider the radial line $r_t = \{ r \exp{2 \pi i t} : r > 1 \} \subset {\mathbb{C} }\setminus {\mathbb{D} }$. Then we call the set $$R_{f}(t) = \phi_f (r_t)$$ the external ray of angle $t$.
If $K(f)$ (equivalently $J(f)$) is locally connected, we can define the Carathéodory semiconjugacy $\gamma \colon {\mathbb{R} }/ {\mathbb{Z} }$ such that $$\gamma(t) = \gamma_f(t) = \lim_{r \to 1} \phi_f(r \exp(2 \pi i t)).$$ The point $\gamma_f(t) \in K(f)$ is called the landing point of the external ray $R_{f}(t)$. Before moving on, we introduce some terminology. Given a polynomial of the form $f_c(z) = z^d + c$ with locally connected Julia set, it is easy to see that the points $\beta_k = \gamma(k/(d-1))$ must be fixed. We call these the $\beta$-fixed point of $f_c$. There exists at most one other fixed point, which we will call the $\alpha$-fixed point of $f_c$. We can also construct parameter rays analogously in the parameter plane. Let $\Phi$ be the (unique) Riemann map which gives a conformal isomorphism between ${\widehat{{\mathbb{C} }} }\setminus \overline{{\mathbb{D} }}$ and ${\widehat{{\mathbb{C} }} }\setminus {\mathcal{M} }_d$, with $\Phi (\infty) = \infty$ and so that $\Phi$ is asymptotic to the identity at infinity. Then the parameter ray of angle $\theta$ is the set $$R_{{\mathcal{M} }_d}(t) = \Phi( \{ r\exp{2 \pi i t} : r > 1\}).$$ The notion of landing of rays is defined similarly as to the case with external rays.
We now can define the topological mating of two monic degree $d$ polynomials $f_1$ and $f_2$, assuming they have locally connected Julia set. We first define the ray-equivalence relation $\sim$ on $S^2_{f,g}$. The equivalence relation $\sim_f$ on $\widehat{{\mathbb{C} }}_f$ is generated by $x \sim_f y$ if and only if $x,y \in \overline{R}_{f}(t)$ for some $t$. Notice that the closure of the external ray contains both the landing point and the point on the circle at infinity. Define a similar equivalence relation on $\widehat{{\mathbb{C} }}_g$. Then the equivalence relation $\sim$ will be generated $\sim_f$ on $\widehat{{\mathbb{C} }}_f$ and $\sim_g$ on $\widehat{{\mathbb{C} }}_g$. We denote the equivalence class of $x$ under this relation by $[x]$.
Denote the Carathéodory semiconjugacy derived from $f_{j}$ by $\gamma_{j}$, so that $\gamma_{j}$. We see that the ray equivalence relation restricts to an equivalence relation $\sim'$ on the disjoint union of $K(f_1)$ and $K(f_2)$ by $$\gamma_{1}(t) \sim' \gamma_{2}(-t) \quad \textrm{ for each } \quad t \in {\mathbb{R} }/ {\mathbb{Z} }.$$ We define $K(f_{1}) {\perp \! \! \! \perp}K(f_{2})$ to be the quotient topological space $S^2_{f,g} / \sim$, where every equivalence class is identified to a point. Making use of the fact that $\gamma_{j}(dt) = f_{j}(\gamma(t))$, we can piece together $f_{1} |_{K(f_{1})}$ and $f_{2} |_{K(f_{2})}$ to form a continuous map which we call $f_{1} {\perp \! \! \! \perp}f_{2}$. This is the topological mating. In nice cases, this quotient space $K_1 {\perp \! \! \! \perp}K_2$ is a topological sphere. We can think of the topological mating as the formal mating, where the external rays have been “drawn tight” using the ray equivalence relation. We remark that in this paper, we will only be considering the mating of (monic unicritical) hyperbolic polynomials of degree $d \geq 2$. By the results of [@Pilgrim:Jordan], this means the Julia sets will be locally connected and so the topological mating will be well defined. Now suppose we have constructed the topological mating $f_1 {\perp \! \! \! \perp}f_2$. We say that a rational map $F$ is the geometric mating of the two monic polynomials $f_1$ and $f_2$ if there exists a topological conjugacy $h$ which is orientation preserving and holomorphic on $\stackrel{\circ}{K_1}$ and $\stackrel{\circ}{K_2}$ satisfying $$h \circ (f_1 {\perp \! \! \! \perp}f_2) = F \circ h.$$ In this case, we will write $F \cong f_1 {\perp \! \! \! \perp}f_2$. If there exist two (or more) rational maps satisfying the above, we say the mating is shared.
Thurston Equivalence
--------------------
We now discuss how the mating can be used to define a rational map on the Riemann sphere. As it stands the formal mating is a branched covering of the sphere. To provide a conformal structure to the topological sphere, we need to invoke the results of Thurston on the classification of branched covers. Recall a branched covering $F S^2 \to S^2$ is called postcritically finite if the postcritical set $$P_F = \bigcup_{n > 0} F^{\circ n}(\{ z : z \text{ is a critical point of $F$} \}).$$ is finite.
Two postcritically finite orientation preserving branched self-coverings with labelled critical points $(F, P_{F})$ and $(G,P_{G})$ of $S^2$ are said to be *Thurston equivalent* (alternatively, *combinatorially equivalent* or just *equivalent*) if there exists orientation preserving homeomorphisms $\phi_{1}, \phi_{2} \colon S^{2} \to S^{2}$ such that
- [$\phi_{1} |_{P_{F}} = \phi_{2} |_{P_{F}}$]{}
- [The following diagram commutes. $$\xymatrix{ (S^{2}, P_{F}) \ar[r]^{\phi_{1}} \ar[d]_{F} & (S^{2}, P_{G}) \ar[d]^{G}
\\
(S^{2}, P_{F}) \ar[r]_{\phi_{2}} & (S^{2}, P_{G})}$$]{}
- [$\phi_{1}$ and $\phi_{2}$ are isotopic via homeomorphisms $\phi_{t}$, $t \in [0,1]$ satisfying $\phi_{0} |_{P_{F}} = \phi_{t} |_{P_{F}} = \phi_{1} |_{P_{F}}$ for each $t \in [0,1].$]{}
When two branched coverings are equivalent in this way, we write $F \cong G$.
Let $\Gamma = \{ \gamma_{1}, \gamma_{2}, \ldots , \gamma_{n} \}$ be a collection of curves in $S^{2}$. If the $\gamma_{i} \in \Gamma$ are simple, closed, non-peripheral, disjoint and non-homotopic relative to $P_F$ then we say $\Gamma$ is a multicurve. We say the multicurve is $F$-stable if for any $\gamma_{i} \in \Gamma$, all the non-peripheral components of $F^{-1}(\gamma_{i})$ are homotopic rel $S^{2} \setminus P_{F}$ to elements of $\Gamma$. Given an $F$-stable multicurve, we can define a non-negative matrix $F_{\Gamma} = (f_{ij})_{n \times n}$ in the following natural way. For each $i,j$, let $\gamma_{i,j,\alpha}$ be the components (these are all simple, closed curves) of $F^{-1}(\gamma_{j})$ which are homotopic to $\gamma_{i}$ in $S^{2} \setminus P_{F}$. Now define $$F_{\Gamma}(\gamma_{j}) = \sum_{i,\alpha} \frac{1}{\deg F |_{\gamma_{i,j,\alpha}} \colon \gamma_{i,j,\alpha} \to \gamma_{j}} \gamma_{i}$$ where $\deg$ denotes the degree of the map. By standard results on non-negative matrices (see [@Seneta]), this matrix $f_{ij}$ will have a leading non-negative eigenvalue $\lambda$. We write $\lambda(\Gamma)$ for the leading eigenvalue associated to the multicurve $\Gamma$.
The multicurve $\Gamma$ is called a Thurston obstruction if $\lambda(\Gamma) \geq 1$.
A post-critically finite branched covering $F \colon S^{2} \to S^{2}$ of degree $d \geq 2$ with hyperbolic orbifold is equivalent to a rational map $R$ on the Riemann sphere if and only if for any $F$-stable multicurve $\Gamma$ we have $\lambda(\Gamma,F)<1$. In that case the rational function $R$ is unique up to conjugation by an automorphism of the Riemann sphere ${\widehat{{\mathbb{C} }} }$.
The requirement of $F$ having a hyperbolic orbifold in the above theorem will not concern us in this paper, since all of our branched coverings will have hyperbolic orbifolds. Indeed, if a branched covering $F$ has non-hyperbolic orbifold, then it must have $P_F \leq 4$ and the pre-image of the post-critical set must be contained in the union of the postcritical set and the set of critical points of $F$. The interested reader should refer to [@DouadyHubbard:1993] for a proof and discussion of the above result. Though Thurston’s theorem is a very strong result, it has some problems in applications, namely it is difficult to check in general for Thurston obstructions.
A multicurve $\Gamma = \{ \gamma_{1}, \gamma_{2}, \ldots , \gamma_{n} \}$ is a Levy cycle if for each $i =1,\ldots,n$, the curve $\gamma_{i-1}$ (or $\gamma_{n}$ if $i = 1$) is homotopic to some component $\gamma_{i}'$ of $F^{-1}(\gamma_{i})$ (rel $P_{F}$) and the map $F \colon \gamma_{i}' \to \gamma_{i}$ is a homeomorphism.
The following result, the culmination of work by Rees, Shishikura and Tan Lei, greatly simplifies the search for Thurston obstructions in the bicritical case.
In the bicritical case, $F$ has a Levy cycle if and only if it has a Thurston obstruction.
A related result in [@TanLei:1990] is the following.
\[t:Tanequivalences\] Let $f_{1} , f_{2}$ be monic unicritical polynomials of the form $f_{i}(z) = z^d + c_i$ and with $\alpha$-fixed points labelled as $\alpha_{1}$ and $\alpha_{2}$ respectively. Then the following are equivalent.
1. [$F$ has a (good) Levy cycle $\Gamma = \{ \gamma_{1},\ldots,\gamma_{n} \}$.]{}
2. [$F$ has a ray-equivalence class $\tau$ containing closed loops and two distinct fixed points.]{}
3. [$[\alpha_{1}] = [\alpha_{2}]$]{}
4. [$f_{1}$ and $f_{2}$ are in conjugate limbs of the degree $d$ multibrot set.]{}
This result significantly reduces the work required to check whether the mating of two unicritical monic polynomials is obstructed, since it suffices by condition 3 to check that the ray classes of $[\alpha_1]$ and $[\alpha_2]$ are disjoint. We finish with a slight generalisation of Theorem \[t:Tanequivalences\], which we will use in the next section.
\[TanShish\] Let $F$ be a mating. Let $[x]$ be a periodic ray class such that $[x]$ contains a closed loop. Then each boundary curve of a tubular neighbourhood of $[x]$ generates a Levy cycle.
By a result of Rees [@Rees:Param], in the hyperbolic postcritically finite case, if the formal mating of $f_1$ and $f_2$ is Thurston equivalent to the rational map $F$, then $F$ will be the geometric mating of $f_1$ and $f_2$. Hence we can use each of the different notions of mating when discussing a mating, without worrying that a different choice will affect the Thurston class.
### A combinatorial view of matings
In this section we discuss the (periodic) ray classes that occur in the formal matings, which are then collapsed in the topological mating. This discussion is extremely natural and will allow us to focus on the important ray classes (those which become the cluster points) in the later sections.
\[disjointness\] Let $F = f_1 \uplus f_2$ be the formal mating of two hyperbolic polynomials which has no Thurston obstruction. Let $z_0,F(z_0) = z_1,\ldots,F^{\circ(n-1)} = z_{n-1}$ be a period $n$ orbit of $f_1$ which is contained in $J(f_1)$ and has combinatorial rotation number different from $0$. Then the periodic ray classes $[z_0],[z_1],\ldots,[z_{n-1}]$ are pairwise disjoint.
Suppose there exists $k$ with $[z_0] = [z_k]$. Then there exists a path through external rays $\gamma$ from $z_0$ to $z_k$. The map $F^{\circ n}$ will take $z_0$ to $z_0$ and $z_k$ to $z_k$ and takes the path $\gamma$ to some path $\gamma'$, which is a path from $z_0$ to $z_k$. But $\gamma'$ is not equal to $\gamma$, since the first return map to $z_0$ and $z_k$ will permute the external rays landing there. Hence the union $\gamma \cup \gamma'$ contains a loop; and so the mating will be obstructed. This contradiction completes the proof.
\[l:matingrayclass\] If the mating of two hyperbolic polynomials is not obstructed, each periodic ray class contains at most one periodic branch point with non-zero combinatorial rotation number.
Let $w_{0}$ and $z_{0}$ be two periodic branch points with non-zero combinatorial rotation number, such that $[w_{0}] = [z_{0}]$. We will show that the periods of $z_0$ and $w_0$ are equal. Let the period of $z_0$ be $n$. Then the map $F^{\circ n}$ maps $z_0$ to itself, the periodic ray class $[z_0]$ to itself and $w_0$ to $w_n = F^{\circ n}(w_0)$. Then we must have $[w_0] = [z_0] = [w_n]$ and so $w_0 = w_n$ by Lemma \[disjointness\]. Hence the period of $w_0$ is divisible by $n$. An analogous argument shows the period of $w_0$ divides the period of $z_0$, and so the periods are the same. Denote this common period by $n$.
Now let $\gamma$ be a path through external rays from $w_{0}$ to $z_{0}$. Since none of the rays meet a pre-critical point (since the maps $f_1$ and $f_2$ are hyperbolic), the $n$th iterate of $\gamma$, which we call $\gamma'$, will also be a path from $w_{0}$ to $z_{0}$. Since the external rays at $w_{0}$ and $z_{0}$ are permuted under the first return map, $\gamma \neq \gamma'$, and so the curve $\gamma \cup \gamma'$ contains a loop, and so the mating is obstructed.
The importance of this second lemma is it will tell us the nature of the maps which make up the matings that produce maps with cluster cycles. Intuitively, it is clear that, to create a rational map with a cluster cycle of period $p$ and rotation number $\rho$, it is necessary for one of the maps to have a periodic cycle of period $p$ and combinatorial rotation number $\rho$, so that this cycle for the polynomial will become the cluster point cycle. We will show that this intuition is correct, at least in the cases for periods one and two (though analogous proofs exist for all periods). The above lemma, on the other hand, sheds some light on the behaviour of the other polynomial in the mating. In short, it will allow us to say that the second polynomial must belong to a satellite component in the Multibrot set.
Note that if we have a periodic ray class which contains a point $p$ with combinatorial rotation number not equal to $0$, then the ray class must contain a sort of rotational symmetry: each global arm at the point $p$[^1] is homeomorphic to each of the other global arms. Furthermore, if the ray class is periodic, then it cannot contain a strictly pre-periodic point, and all points in the ray class have period dividing the period of the ray class.
\[l:armslemma\] Suppose a periodic ray class contains a point $p$ which has combinatorial rotation number different from $0$. Then given any periodic orbit $\mathcal{O} = \mathcal{O}(z_0)=\{z_0,z_1,\ldots,z_{n-1} \}$, the intersection between any global arm at $p$ and $\mathcal{O}$ contains at most one point.
Clearly if $[p] \neq [z_i]$ for any $i$ then the intersection of $\mathcal{O}(z)$ with any global arm will be empty. So assume without loss of generality that $[p] \cap [z_0] \neq \varnothing$. Let $\ell$ be a global arm at $p$ in the periodic ray class. There exists some $k$ so that $F^{\circ k}(\ell) = \ell$. If $\ell \cap \mathcal{O}$ is empty, then $\ell$ also contains no pre-images of points in $\mathcal{O}$, since otherwise some forward image of $\ell$ will contain a point in $\mathcal{O}$, and so $\ell \cap \mathcal{O}$ would be non-empty, a contradiction. Since $\ell$ contains no pre-images of points in $\mathcal{O}$, no forward iterate of $\ell$ can contain points of $\mathcal{O}$. Since global arms are mapped homeomorphically onto global arms, this means all the global arms have empty intersection with the orbit $\mathcal{O}$. Hence every global arm at $p$ must contain at least one point of $\mathcal{O}$.
Suppose the global arm $\ell$ contains $r>1$ elements of $\mathcal{O}$, $z_0,\ldots,z_{r-1}$. Under the first return map to $\ell$, these elements must be permuted, since the orbit is periodic. Without loss of generality, we can assume the first element of $\ell \cap \mathcal{O}$ on the global arm (in terms of distance from $p$) is $z_0$. Then the second point (again, in terms of distance in the tree from $p$) in $\ell \cap \mathcal{O}$ is $F^{\circ k}(z_0) = z_k$ for some $k$. Let $\gamma$ be the sub-arm from $p$ to $z_0$. Then $\gamma' = F^{\circ k}(\gamma)$ is contained in $\ell$ and will be a path from $p$ to $z_k$, since the global arm $\ell$ maps homeomorphically onto itself under the return maps (since the map $F$ is a homeomorphism on the graphs of the ray equivalence classes). In particular, $z_0 \in \gamma'$. However, this means that there has to be a pre-image of $z_0$ in the path $\gamma$. However, by construction, there are no points in the orbit of $z_0$ in the interior of $\gamma$ and there also does not exist any preperiodic points in $\gamma$. This contradiction means that the global arm $\ell$ must contain at most one point in $\mathcal{O}$.
Main results
------------
The focus of this paper will be the proof of the following two theorems.
\[mainthm1\] Suppose that $F$ is a bicritical rational map with a fixed cluster point and the combinatorial rotation number is $p/q$. Then the following hold.
- [$F$ is the mating of a $q$-rabbit (whose $\alpha$-fixed point has combinatorial rotation number $p/q$) and another map $h$.]{}
- [The map $h$ has an associated angle with angular rotation number $(q-p)/q$.]{}
The theorem in the period two cluster case is only concerned with the quadratic case.
\[mainthm2\] Suppose that $F$ is a quadratic rational map with a period two cluster point and the combinatorial rotation number is $p/q$. Then the following hold.
- [$F$ is a mating.]{}
- [Precisely one of the maps in the mating belongs to the $1/2$-limb of the Mandelbrot set and this map is one of the two period $2q$ maps which belong to the $p/q$-sublimb of the period two component of the Mandelbrot set.]{}
- [The other map has an associated angle with 2-angular rotation number $(q-p)/q$.]{}
We also will prove the following fact about the multiplicity of the shared matings in the period two cluster case.
\[t:shares\] In the period two cluster cycle case, each combinatorial data can be obtained in either two, three or four ways and all of these multiplicities are obtained.
Fixed cluster points {#fixedclust}
====================
Our goal in this section will be to prove Theorem \[mainthm1\]. The results of this section are relatively elementary, but provide an interesting insight into the the differences between the fixed case and the period two case. It will also allow us to introduxw some terminology that will be of use in the period two cluster cycle case.
\[p:adamproof\] Suppose $F$ is a bicritical rational map with a fixed cluster point, where both critical points have period $n$. Then the two critical point components cannot be adjacent. In other words, the critical displacement cannot be $1$ or $2n-1$.
We will assume, to obtain a contradiction, that $U_{0}$ is the adjacent component which is anticlockwise of $V_{0}$ in the cyclic ordering of components around the cluster point $c$, and that $U_0$ and $V_0$ are the components containing the critical points of $F$. We will use the notation that $U_{k} = F^{\circ k}(U_{0})$ and $V_{k} = F^{\circ k}(V_{k})$ to label the critical orbit components. Let $\gamma_1$ be a curve which loops around $v_1$ and $v_2$ as in Figure \[f:Adamproof2\] so that the only critical orbit Fatou components intersecting with $\gamma_1$ are $U_1$ and $U_2$. Now define $\gamma_n = F^{\circ (n-1)}(\gamma_1)$ for $n = 1,\ldots,n-1$. Then $\Gamma = \{ \gamma_1, \ldots, \gamma_n \}$ is a Levy cycle.
(-4,-3.75)(4,3.75) [90]{}[150]{}[4 3 x mul 90 add sin mul]{} [150]{}[210]{}[4 3 x mul 90 add sin mul]{} [30]{}[90]{}[4 3 x mul 270 add sin mul]{} [90]{}[150]{}[4 3 x mul 270 add sin mul]{} (0.5;90)(1.5;138)(2.75;130)(3.75;120) (1.15;90)(3.75;60)(2.75;50)(1.5;42)(0.5;90) (0,0)(1.5;157)(3.5;170)(4;195)(3.5;210)(2.5;185)(0,0) (0,0)(1.5;263)(3.5;250)(4;225)(3.5;210)(2.5;235)(0,0) (0.5;330)(1.5;18)(2.75;10)(3.75;0) (1.15;330)(3.75;300)(2.75;290)(1.5;282)(0.5;330) (0.5;210)(1.5;258)(2.75;250)(3.75;240) (1.15;210)(3.75;180)(2.75;170)(1.5;162)(0.5;210) \[340\](3.25;120)[$\gamma_{1}$]{} \[220\](3.25;0)[$\gamma_{2}$]{} \[210\](1.15;210)[$\gamma_{3}$]{} \[60\](0,0)[$\bullet$]{} \[120\](0,0)[$\bullet$]{} \[180\](0,0)[$\ast$]{} \[240\](0,0)[$\ast$]{} \[0\](0,0)[$\times$]{} \[300\](0,0)[$\times$]{} \[48\](0,0)[$V_1$]{} \[102\](0,0)[$U_1$]{} \[163\](0,0)[$V_0$]{} \[265\](0,0)[$U_0$]{}
This result tells us that the two critical points cannot be in adjacent Fatou components in the cluster. We will actually show that these are the only cases that cannot be obtained as combinatorial data for rational maps. We now prove the first part of our classification: that one of the maps must be an $q$-rabbit. Here, a $q$-rabbit will be any map which belongs to a hyperbolic component which bifurcates off of the (unique) period 1 component in the degree $d$ multibrot set ${\mathcal{M} }_d$.
\[p:nrabbit\] Let $f_1$ and $f_2$ be monic unicritical polynomials with a period $q$ superattracting orbit. Suppose $F \cong f_{1} {\perp \! \! \! \perp}f_2$ is a rational map that has a fixed cluster point. Then one of $f_{1}$ or $f_{2}$ is an $q$-rabbit.
A degree $d$ rational map has $d+1$ fixed points (counting multiplicity), and so we see that one of the $\alpha$-fixed points, $\alpha_1$ or $\alpha_2$, must become the fixed cluster point in the rational map, since otherwise we would have $[\alpha_1] = [\alpha_2]$ and the mating would be obstructed, by Theorem \[t:Tanequivalences\]. Without loss of generality, let the cluster point of $F$ be $[\alpha_1]$.
$\alpha_1$ is the landing point of $k$ external rays and these are permuted cyclically under iteration by the map $f_1$. Since the mating is not obstructed, $[\alpha_1]$ is a tree and the global arms of the tree are permuted cyclically and homeomorphically under $f_1 \uplus f_2$. Furthermore, because the map $f_1 \uplus f_2$ is a homeomorphism on this tree, each global arm contains at most one root point of a critical orbit Fatou component of $f_2$. Since there must be $q$ root points of critical orbit Fatou components in this ray equivalence class, there are $n$ global arms at $\alpha_1$ and so $q$ external rays landing at $\alpha_1$. But then $f_1$ has a period $n$ superattracting cycle and $n$ external rays landing on its $\alpha$-fixed point, so it is an $q$-rabbit.
Indeed, we can say slightly more. If the rational map has a fixed cluster point with combinatorial rotation number $\rho = p/q$, then the rabbit is actually the $p/q$-rabbit, the rabbit whose $\alpha$-fixed point with rotation number $p/q$. This is easy to see by noticing that the combinatorial rotation number at the $\alpha$-fixed point is given by the ordering of the permutation of the external rays, and it is precisely these external rays which form the arms of the periodic ray class which become the cluster point.
Properties of the non-rabbit {#ss:1non-rab}
----------------------------
To complete the classification of the maps which mate to give a rational function with a fixed cluster point, we need to study the properties of the map which partners the $q$-rabbit in the mating. Hence we know turn our attention to the polynomials which partner the rabbits in the matings. Indeed, our classification of this complementary map requires us to take into account the ordering of the external rays which land at the $\alpha$-fixed point of the rabbit. This classification will follow the considerations found in [@Bullett-Sentenac] and [@Blokhetal]. In particular, we need to provide the definition of the rotation number of an angle.
\[d:1-ARN\] Let $\theta \in \mathbb{S}^{1}$ be periodic of period $q$ under the map $\sigma_d \colon t \mapsto dt$, so that $\theta = a/(d^{q}-1)$ for some $a$. Label the angles $d^j\theta$, $j=1,\ldots,q$ cyclically by $\theta_1,\theta_2,\ldots,\theta_n$ with $\theta_1 = \theta$. Then we say that the angle $\theta$ has (angular) rotation number $p/q$ if $$d \theta_k = \theta_{k+p \mod q}$$ for each $k$.
Notice that if $\theta$ has rotation number $p/q$, then the angles $d \theta$, $d^2 \theta, \ldots$ have rotation number $p/q$ also. Hence we can equally well talk of the orbit of angles having angular rotation number $p/q$.
Let $f$ be a hyperbolic polynomial belonging to the degree $d$ multibrot set ${\mathcal{M} }_d$. We say the angle $\theta$ is associated with the polynomial $f$ if the external ray of angle $\theta$ lands at a (not necessarily principle) root point of the critical value component Fatou component of $f$.
The following is well-known and gives an extremely useful link between the dynamical plane and the parameter plane; see for example [@Milnor:mandel].
The angle $\theta$ is associated to $f$ if and only if the parameter ray of angle $\theta$ lands at a (not necessarily principle) root point of the hyperbolic component containing the map $f$.
In the sequel, we use the notation $A_k$ to denote an arc of the form $(k/(d-1),(k+1)/(d-1)) \subset S^1$.
\[hangles\] Suppose $F = f {\perp \! \! \! \perp}h$ is a rational map with a fixed cluster point and $f_1$ is the $p/q$-rabbit. Then one of the angles associated with $h$ has rotation number $(q-p)/q$. Moreover, in degree $d$, all the angles in the forward orbit of this angle lie in an arc $A_j$.
All rays landing on the $\alpha$-fixed point of $f$ have angles $\theta_i$ (in cyclic order) with rotation number $p/q$ and lie in arc $A_k$ for some $k \in \{ 0,1,\ldots,d-2 \}$. Since the only periodic biaccessible point on $J(f_1)$ is $\alpha_1$, ([@Wittner], Claim 10.1.1) the root points of the critical orbit Fatou components must be the landing points of the rays of angles $-\theta_i$. Each of these angles has rotation number $(q-p)/q$, and precisely one of them is the angle of the ray landing at a root point of the critical value Fatou component of $h$. Since it lands at a root point of the critcal value component of $h$, it is one of the angles associated with $h$. The second statement simply follows from the fact that all the angles $\theta$ of the external rays landing at the $\alpha$-fixed point of an $n$-rabbit all lie in some arc $A_k$, and so the angles $-\theta$ must also lie in the arc $A_{d-k-2}$.
\[p:allobtained\] All (admissible) combinatorial data can be obtained by matings in (precisely) $2(d-1)$ ways.
First we will construct a rational map $F \cong f {\perp \! \! \! \perp}h$ with combinatorial data $(\rho, \delta)$. $\rho$ is of the form $p/q$, so we take $f$ to be the $p/n$-rabbit whose corresponding angles lie in the arc $A_0$. We now make a judicious choice for our complementary map $h$. Label the angles of the rays landing at the $\alpha$-fixed point of $f$ in cyclic order by $\theta_1,\ldots,\theta_q$, starting with $\theta_1$ as the angle of the ray which lands anticlockwise from the critical value component of $f$. To get a critical displacement of $\delta = 2k-1$, we choose $h$ to be the rational map corresponding to the angle $-\theta_k$. The map $h$ does not lie in the conjugate limb to $f$ in ${\mathcal{M} }_d$ and so the mating exists and the rational map has combinatorial data $(\rho, \delta)$.
To see that all combinatorial data is obtained with multiplicity $2(d-1)$, observe that the $p/q$-rabbit $f$ that was chosen in the previous paragraph could equally well have been the one whose corresponding angles lay in the arc $A_k$, with $k \in \{0,\ldots,d-2\}$. Then a similar argument allows us to find a complementary map $h$ which means $f {\perp \! \! \! \perp}h$ has combinatorial data $(p/q,\delta)$. Moreover, we can carry out the mating $h' {\perp \! \! \! \perp}f$, where $h'$ is the map for which $f {\perp \! \! \! \perp}h'$ has critical displacement $-\delta$ and again this holds for any choice of the $p/q$-rabbit $f$. This gives us the $2(d-1)$ distinct ways of forming the mating.
To see that $2(d-1)$ is sharp, observe that by results of [@Blokhetal] (in particular, Theorem 2.8) there is precisely one orbit of angles in each arc $A_k$ which has rotation number $(q-p)/q$[^2]. These angles are precisely the angles $-\theta_1,\ldots,-\theta_q$ above. Two of these (namely $-\theta_1$ and $-\theta_n$) correspond to the $(q-p)/q$-rabbit and each of the others construct a rational map with a distinct critical displacement when mated with the appropriate $p/q$-rabbit. So no other maps can be mated with a $p/q$-rabbit to create a rational map with a fixed cluster point.
It follows from the above proof and by a simple counting argument that the maps $h$ such that $f {\perp \! \! \! \perp}h$ has a fixed cluster cycle (for some choice of the $p/q$-rabbit $f$) are precisley those maps $h$ who have a corresponding angle $\theta$ which has angular rotation number $p/n$ and whose orbit $\{ \theta, d\theta, \ldots , d^{q-1}\theta \}$ all lie in an arc $A_k$.
Proof of Theorem \[mainthm1\]
-----------------------------
We now prove the first of our two main theorems. First we state a result from [@Thurstoneq].
\[Fixedcase\]\[Theorem A, [@Thurstoneq]\] Suppose $F$ and $G$ are bicritical rational maps (with labelled critical points) with fixed cluster cycles with the same combinatorial data. Then $F$ and $G$ are equivalent in the sense of Thurston.
By Proposition \[p:nrabbit\], one of the maps must be an $n$-rabbit, $f$, and since combinatorial rotation numbers are preserved by matings and the fact that the $\alpha$-fixed point of $f$ the cluster point, the combinatorial rotation number of the $\alpha$ fixed point of $f$ must be $p/n$ as well. By Lemma \[hangles\], one of the angles associated to $h$ must have rotation number $(n-p)/n$. Since all admissible combinatorial data can be obtained by matings (Proposition \[p:allobtained\]), then by Theorem \[Fixedcase\], all rational maps with fixed cluster points are matings.
Period two cluster cycles
=========================
The simplicity of the results in the fixed cluster point case perhaps could lead the reader to believe that the analogous results hold in the period two case. However, there is actually an increased level of complexity in the period two case, as we will discover. We will only consider the quadratic case in this section: a brief discussion of the higher degree case (and the difficulties in tackling it) is given after Theorem \[t:preciselyone\]. We know from the previous section that if we wanted to construct a rational map with a fixed cluster point that has combinatorial roation number $\rho$, then we need one of the maps to be the $\rho$-rabbit. We do not get such an exact statement. However, we can differentiate between the two polynomials: one of them must belong to the $1/2$-limb.
\[t:preciselyone\] Let $F \cong f_{1} {\perp \! \! \! \perp}f_{2}$ be a degree 2 rational map with a period 2 cycle of cluster points. Then precisely one of the maps $f_{1}$ or $f_{2}$ belongs to ${\mathcal{M} }_{1/2}$, the $1/2$-limb of ${\mathcal{M} }$.
It is clear that both $f_{1}$ and $f_{2}$ cannot belong to the $(1/3,2/3)$-limb, since then the mating would be obstructed by Theorem \[t:Tanequivalences\]. Hence it only remains to show that both of $f_{1}$ and $f_{2}$ cannot lie outside ${\mathcal{M} }_{1/2}$. So assume $f_{1}$ and $f_{2}$ lie outside ${\mathcal{M} }_{1/2}$. If $f_{i}$ is not in ${\mathcal{M} }_{1/2}$, the external rays of angles $1/3$ and $2/3$ must land at distinct points. Since these angles have period 2 under angle doubling, they must land at points with period dividing 2. As $R_{1/3}$ and $R_{2/3}$ land at different points, these landing points must be a period 2 cycle.
Now notice that, under mating, we have the identifications $$\gamma_{f_1}\left( 1/3 \right) \sim \gamma_{f_2} \left( 2/3 \right) \quad \text{ and } \quad \gamma_{f_2} \left( 1/3 \right) \sim \gamma_{f_1} \left( 2/3 \right)$$ and these points are not identified with each other, or any other points. In particular, these points cannot be cluster points. Since $f_{i}(\gamma_{f_i}(1/3)) = \gamma_{f_i}(2/3)$ and $f_{i}(\gamma_{f_i}(2/3)) = \gamma_{f_i}(1/3)$, these pairs form a period 2 cycle for the map $F \cong f_{1} {\perp \! \! \! \perp}f_{2}$. However, since $F$ already has a period 2 cycle (the cluster point cycle) by assumption, we see that this second period 2 cycle cannot exist, since a degree 2 rational map can only have one period 2 cycle. Hence both of the maps cannot lie outside ${\mathcal{M} }_{1/2}$, and so precisely one of them lies in ${\mathcal{M} }_{1/2}$.
We notice that, similarly to the last section, we can separate the classification into an investigation of the map in ${\mathcal{M} }_{1/2}$ and, afterwards, we can study the map that does not belong to ${\mathcal{M} }_{1/2}$. Indeed as we will see, the period two cluster case is far more complicated than the relatively simple fixed cluster case. A further comment is required on the restriction to degree 2. On the one hand, this restriction is motivated by the fact that Thurston equivalence is only possible with the given combinatorial data in the quadratic case. However, it turns out that even a generalisation of Theorem \[t:preciselyone\] does not hold in the higher degree case. In particular, though it remains true that one of the maps must belong to *a* $1/2$-limb[^3] in the degree $d$ multibrot set ${\mathcal{M} }_d$, it is not true that precisely one of them has this property. Examples showing this is no longer the case can be found in [@Thurstoneq] (Section 4) and also in the forthcoming paper [@2SupAtt]. It is hoped a more detailed study of the higher degree case will be the subject of future work.
Properties of the map in ${\mathcal{M} }_{1/2}$
-----------------------------------------------
For each $\rho \in {\mathbb{Q} }/ {\mathbb{Z} }$, there are in fact two maps in the $1/2$ limb that have a period two orbit with combinatorial rotation number $\rho$. It turns out that either of these can be used to create a rational map with a period two cluster cycle using matings.
\[p:2clusttunedrabbit\] Let $f_1$ and $f_2$ be quadratic polynomials with period $2q$ superattracting orbits. Suppose $F \cong f_{1} {\perp \! \! \! \perp}f_2$ is a rational map that has a period 2 cluster cycle. Then one of $f_{1}$ or $f_{2}$ is either the tuning of the basilica by an $q$-rabbit (a“double rabbit”), or the (unique) other period $2q$ component lying in the wake of this double rabbit.
This proof is equivalent to showing that the map in ${\mathcal{M} }_{1/2}$ must belong to a $p/q$-sublimb of the period 2 component of ${\mathcal{M} }$. Notice that maps belonging to this sublimb are precisely those which have a period 2 point with combinatorial rotation number $p/q$. If $f \in {\mathcal{M} }_{1/2}$ then it has a period two cycle $\{ p_1 , p_2 \}$. Under mating, the equivalence classes of these points must also have period two, since otherwise we would have $[p_1] = [p_2]$, contradiciting Lemma \[l:matingrayclass\]. Since a quadratic rational map has at most one period two cycle, the classes $[p_1]$ and $[p_2]$ must become the cluster cycle. It now follows from Lemma \[l:armslemma\] that the combinatorial rotation number of this period two orbit will be $p/q$ for some $p$, and the only polynomials which have a period $2q$ superattracting cycle and a period two point with combinatorial rotation number $p/q$ are those two given in the statment of the proposition.
An example of the position of the two maps in paramter space is given in Figure \[f:twomaps\]. It again follows easily that the combinatorial rotation number of the period two cycle is the same as the combinatorial rotation number of the period two cluster cycle in the resulting rational map. Figure \[f:twomaps\] shows the position of these two maps in the period 8, rotation number 1/4 case.
![The double rabbit component and secondary map component of period 8, rotation number 1/4 case.[]{data-label="f:twomaps"}](twomaps.eps){width="95.00000%"}
We remark for the moment that this classification is the best possible: it is not true in the period two cluster case that one of the maps must be the double rabbit, for example, since there are matings with the secondary maps which create period two cluster cycles. The matings with the double rabbit are the canonical examples, in that they behave similarly to the examples as found in Section \[fixedclust\]. We will consider the matings with the secondary map in Section \[secondmap\]. We denote by $f_{p/q}$ (respectively $g_{p/q}$) the double rabbit (respectively secondary map) with a period two orbit with combinatorial rotation number $p/q$.
Properties of the complementary map
-----------------------------------
We now attempt to prove some analogous results to those in Section \[ss:1non-rab\]. We start with a well-known lemma whose proof, which makes use of results from [@Milnor:mandel], is omitted.
\[l:Wittner2\] Suppose that $z$ is a biaccessible periodic point in $J(f_{p/q})$. Then $z$ is either the $\alpha$-fixed point or belongs to the period 2 orbit $\{p_1,p_2\}$.
As with the previous section, our description of the complementary map (the map which does not belong to the $1/2$-limb of the Mandelbrot set) will take place as a discussion of the associated angles. In fact, there are perhaps more descriptive ways of describing these maps (see [@Mythesis] for a discussion of the combinatorial classification using internal addresses), but the description with associated angles is the more complete for the moment. We first require an analogue to Definition \[d:1-ARN\].
\[d:2-ARN\] Let $\theta \in S^1$ be periodic of period $2q$ under the map $\sigma_2 : t \mapsto 2t$ (so $\theta = a/(2^{2q}-1)$ for some $a$). Consider the disjoint sets $$\begin{aligned}
&A_{0} = A_{0}(\theta) = \left\{ \theta, 2^{2}\theta, 2^{4}\theta, \ldots, 2^{2k}\theta, \ldots , 2^{2(q-1)}\theta \right\} \quad\text{and}\\
&A_{1} = A_1(\theta) = \left\{ 2 \theta, 2^{3} \theta, \ldots, 2^{2k+1}\theta, \ldots, 2^{2q-1}\theta \right\}.\end{aligned}$$ We say $\theta$ has (admissible) ($2$-)angular rotation number $p/q$ if the following conditions are satisfied.
- [$A_0$ and $A_1$ lie in disjoint arcs of $S^1$]{}
- [The sets $A_0$ and $A_1$ have (angular) rotation number (in the sense of Definition \[d:1-ARN\]) $p/q$ under the map $\theta \mapsto 4 \theta$.]{}
Clearly this is not a full generalisation of Definition \[d:1-ARN\], since the first condition above introduces a restriction which was not used previously. It actually turns out that this restriction will be important for us. There are actually orbits of angles which satisfy the second condition but not the first (see [@Blokhetal]), but we do not want to consider such orbits in this paper and so the restriction allows us to ignore them. For the moment we will be assuming that the map in ${\mathcal{M} }_{1/2}$ is the double rabbit. As will be shown later, the set of complementary maps to the secondary map are a subset of the complementary maps to the double rabbit.
\[p:hangles\] Suppose $F \cong f_{p/q} {\perp \! \! \! \perp}h$ is a rational map with a period two cluster cycle. Then one of the angles associated to $h$ has $2$-angular rotation number $(q-p)/q$.
This is similar to Lemma \[hangles\]. The the ray classes of the period two orbit $\{ p_0, p_1 \}$ of $f_{p/q}$ will become the cluster cycle and we note that the angle of any external ray landing on one of the the $p_i$ has 2-angular rotation number $p/q$. Let $p_0$ be the periodic point whose external rays all lie in $(2/3,1/3)$ and label these angles in anticlockwise order (starting anywhere) by $\theta_1,\ldots,\theta_n$. We claim that $R^h_{-\theta_i}$ will land at the root point of a critical orbit Fatou component of $h$. If not, the ray class would have to contain another periodic biaccessible point of $J(f_{p/q})$ and by Lemma \[l:Wittner2\] the only other biaccessible point is $\alpha_f$ and by Lemma \[l:matingrayclass\] $[\alpha_f] \neq [p_1] \neq [p_2] \neq [\alpha_f]$. By Lemma \[p:Marysresult\], one of the rays $R^h_{-\theta_i}$ will land at the root point of the critical value Fatou component of $h$, and so $-\theta_i$ will be one of the angles associated with $h$. Finally, as the angles in the orbit of $\theta_i$ have 2-angular rotation number $p/q$, the angles in the orbit of $-\theta_i$ will have 2-angular rotation number $(q-p)/q$.
\[p:alldata2\] All combinatorial data can be obtained.
We will show that any combinatorial data $(\rho,\delta) = (p/q,2k+1)$ can be obtained by a mating of the form $f_{p/q} {\perp \! \! \! \perp}h$. Clearly, if $h$ is chosen appropriately, then the rational map formed by this mating will have a period two cluster cycle with combinatorial rotation number $p/q$, since matings preserve the combinatorial rotation number. Label the angles of the external rays landing at $p_0 \in J(f_{p/q}))$ by $\theta_0,\ldots,\theta_{q-1}$, where $\theta_0$ is the angle of the external ray which approaches $p_0$ immediately anticlockwise from the critical point component of $f_{p/q}$. Then to get $\delta = 2k+1$, let $h$ be the map associated with the parameter ray of angle $-\theta_k$. Since $\theta_k \in (2/3,1/3)$, then $-\theta_k \in (2/3,1/3)$ also, so the mating is not obstructed by Tan Lei’s theorem. Also, since $-\theta_k$ is associated to $h$, the external ray of angle $-\theta_k$ lands at the base point of the critical value component of $h$. Thus the rational map $F \cong f_{p/q} {\perp \! \! \! \perp}h$ exists and has combinatorial data $(\rho,\delta)$.
A converse to the Proposition \[p:hangles\] also exists, if one places a suitable restriction on the angles that can be associated with the complementary map $h$.
\[p:hanglesconverse\] Suppose $h$ has an associated angle $\theta$ which has $2$-angular rotation number $(q-p)/q$ and that this angle lies in $(2/3,1/3)$. Then $F \cong f_{p/q} {\perp \! \! \! \perp}h$ (and $F' \cong h {\perp \! \! \! \perp}f_{p/q}$) have a period two cluster cycle.
Since $\theta \in (2/3,1/3)$, the set $A_1 = \{ 2^{2k+1}\theta : k = 0,\ldots,q-1 \}$ is a subset of $(1/3,2/3)$ and by definition has $1$-angular rotation number $p/q$ under the map $t \mapsto 4t$. Hence the angles in $A_1$ correspond to the angles of the rays which land on the (unique) $p/q$ rabbit in ${\mathcal{M} }_4$ whose associated angles lie in $(1/3,2/3)$. Hence there are exactly $q$ angles which satisfy the properties of the proposition and since by Proposition \[p:alldata2\], all $q$ different critical displacements can be obtained, we are done.
Proof of Theorem \[mainthm2\] {#Proofmainthm2}
-----------------------------
We now prove the second main theorem. Again, we require a result from [@Thurstoneq].
\[Per2case\]\[Theorem B, [@Thurstoneq]\] Suppose that two quadratic rational maps $F$ and $G$ have a period two cluster cycle with rotation number $p/n$ and critical displacement $\delta$. Then $F$ and $G$ are equivalent in the sense of Thurston.
By Proposition \[p:alldata2\], all combinatorial data can be obtained and by Theorem \[Per2case\], maps with the same combinatorial data are Thurston equivalent, so all maps with period two cluster cycles are matings. By Proposition \[p:2clusttunedrabbit\], one of the maps must be either the double rabbit $f_{p/q}$ or the unique period $2q$ polynomial $g_{p/q}$ that lies in the wake of $f_{p/q}$ in ${\mathcal{M} }$. Finally, by Proposition \[p:hangles\] the complementary map $h$ must have an associated angle with 2-angular rotation number $(q-p)/q$.
Matings with the secondary map {#secondmap}
------------------------------
We now focus on the main difference between the matings in the fixed and period two cases, namely that there exist non-trivial shared matings in the period two case, which do not exist in the fixed case. In the fixed case, since one of the maps had to be a However, it is not true that
We describe the general structure of these Hubbard trees of the secondary maps. This can be calculated from algorithms in [@SchleichBru], for example.
The Hubbard tree of a secondary map can be described as follows. There are two period 2 points $p_1$ and $p_2$ with $n$ arms. One of the global arms at $p_1$ contains the critical point $c_0$ and the other global arms at $p_1$ have endpoints $c_1,c_3,\ldots,c_{2n-3}$. The point $p_2$ has one global arm which contains the critical point $c_0$, and the endpoints of the other global arms are $c_2, c_4, \ldots, c_{2n-2}$. The point $c_{2n-1}$ is on the arc $(p_2, c_{2n-1})$. Finally, there are no branch points on the arc $(p_1,p_2)$, and $c_0$ lies on $(p_1,p_2)$.
As an example, Figure \[f:treepic\] shows the Hubbard tree for the secondary map which corresponds to the angles $(407/1023,408/1023)$.
We now ask which rational maps can be obtained by matings with the secondary map. We will prove the following.
\[l:critdisp\] Suppose $F \cong g_{p/q} {\perp \! \! \! \perp}h$ is a rational map. Then the critical displacement of $F$ will be $1$ or $2n-1$.
For ease of notation we will drop the subscript on $g_{p/q}$. We will break down the proof into a sequence of lemmas, in which we will show that it is not possible for the rational map constructed by the mating $g {\perp \! \! \! \perp}h$ to have any other critical displacement apart from $1$ or $2n-1$. First a comment on the angles of the rays that will be relevant to this discussion. The branch at the period two point $p_1$ in the Hubbard tree containing the critical value of $g$ is separated from the other branches at the period two point by two rays, of angle $\theta$ and $\theta+3$ (we suppress the denominator $2^{2n} -1$). The rays of angles $\theta +1$ and $\theta+2$ land at the root point of the critical value component of $g$, see Figure \[f:critvalrays\].
We claim that the ray graph containing the point $p_1$ (as in the picture) must have (precisely) one of the following two properties. The branch of the graph of the ray equivalence class containing the root point $r$ of the critical value component $U_1$ of $g$ (in other words, the landing point of the rays of angle $\theta+1$ and $\theta+2$) must contain either the ray of angle $\theta$ or the ray of angle $\theta+3$. We will show that every other possibility is impossible.
\[l1\] The branch of the ray equivalence class containing $r$ cannot contain the ray $R_{g}(\phi)$, which lands at $p_1$ and such that $\phi \notin \{ \theta, \theta+3 \}$.
It is clear that each branch of the graph must contain exactly one of the rays landing at the point $p_1$. If it contained more than one then they would form a loop and thus would generate a Levy cycle. Suppose the branch containing $r$ contains a ray of angle $\phi \notin \{ \theta, \theta+3 \}$, and $R_g(\phi)$ lands at $p_1$. Let $\gamma$ denote the (unique) path through external rays from $r$ to $p_1$. Since the only possible biaccessible points in $J(g)$ on this path are $r$ and $p_1$, we see that (using $[p,r]$ as the notation for the regulated arc from $p$ to $r$): $$\Gamma = \gamma \cup [p_1,r]$$ separates the sphere into two pieces. In particular, it separates the root point $r'$ on the branch anticlockwise in the cyclic order from $r$ from the ray of angle $\phi'$ which is the first angle anticlockwise round from $\phi$ in the cyclic ordering of rays at $p$. But since $F$ has to maintain this cyclic ordering, the branch containing $r'$ has to contain the ray of angle $\phi'$, and since rays cannot cross this is a contradiction, see Figure \[f:critpic1\].
So we now know that the branch of the graph of the ray equivalence class containing $r$ must contain the ray $R_g(\theta)$ or $R_g(\theta+3)$. We now study this branch in more detail.
\[l2\]
1. [The rays $R_h(-(\theta+1))$ and $R_h(-\theta)$ do not land at the same point on $J(h)$.]{}
2. [The rays $R_h(-(\theta+2))$ and $R_h(-(\theta+3))$ do not land at the same point on $J(h)$.]{}
Suppose that the ray $R_h(-(\theta+1))$ lands at the same point as $R_h(-\theta)$. This common landing point $\zeta$ cannot be the root point of a critical orbit component, since the size of the sector would mean this would have to be a critical value component, and so the critical values would be in the same cluster, a contradiction of Proposition \[p:Marysresult\]. However, the angular width of the sector bounded by $\zeta$ and the external rays $R_h(-(\theta+1))$ and $R_h(-\theta)$ is the smallest it can possibly be, and so it must contain the root point of the Fatou component which contains critical point of $h$. But this is another contradiction since the root point of the critical value component needs to be in the ray equivalence class of $f(p_1)$ and it is separated from this point. The proof of when $R_h(-(\theta+2))$ lands at the same point as $R_h(-(\theta+3))$ is analogous to the above.
In light of Lemma \[l1\] and Lemma \[l2\], the only remaining possibility is that the rays $R_h(-(\theta+2))$ and $R_h(-\theta)$ land at the same point or the rays $R_h(-(\theta+1))$ and $R_h(-(\theta+3))$ land at the same point. Since both situations are essentially the same will we discuss only the first one. The common landing point $\xi$ of the two rays must be the root point of a critical orbit component. For if not, we notice that $$R_g(\theta+2) \cup R_h(-(\theta+2)) \cup R_h(-\theta) \cup R_g(\theta)$$ separates the ray $R_h(-(\theta+1))$ from all other rays with the necessary denominator (Figure \[f:critpic2\]). Since the root point of a critical orbit component must have (at least) two rays landing on it, this means that this branch of the graph of the ray equivalence class cannot contain a root point of a critical orbit component. But then none of the branches can, as each one maps homeomorphically onto its image and they are periodic. Hence the rays $R_h(-(\theta+2))$ and $R_h(-\theta)$ land at the root point $\hat{r}$ of a critical orbit component. The angular width of the sector containing this component is 2, and so the angular width of its pre-image is 1. This means that the pre-image sector contains the critical value, and so in particular the pre-image of $\hat{r}$ is the root point of the critical value component. Hence $\hat{r}$ is the root point of the component $V_2$ containing the second iterate of the critical point of $h$. Since $V_2$ is adjacent to $U_1$, it follows their pre-images are adjacent, and so the critical point component of $g$ will be adjacent to the critical value component of $h$. This means that the critical displacement of the resultant rational map will have to be $\pm 1$, or equivalently, equal to $1$ or $2n-1$.
This proof shows us that, in order for a rational map to be a mating with the secondary map as one of the participants, one of the clusters has to see a critical point adjacent to the critical value of the other critical point.
\[c:bothmatings\] Suppose $F \cong g_{p/q} {\perp \! \! \! \perp}h$ is a rational map with a period two cluster cycle. Then $F' \cong f_{p/q} {\perp \! \! \! \perp}h$ is also a rational map with a period two cluster cycle.
There are precisely two rays landing at the root points of the critical orbit Fatou components of $h$ and the orbits of these rays are disjoint. Since $F$ is a rational map, the angle $\theta$ associated to $h$ must belong to $(2/3,1/3)$. Furthermore, as $F$ has a period two cluster cycle, one of these ray orbits is made up of angles of the form $-\theta_i$, where $\theta_i$ are the $q$ angles of the rays which land on the period two orbit of $g_{p/q}$. But these are precisely the angles of the rays which land on the period two orbit of $f_{p/q}$. The orbit therefore has $2$-angular rotation number $p/q$ and so one of the angles associated to $h$ satisfies the conditions of Proposition \[p:hanglesconverse\].
The converse to Corollary \[c:bothmatings\] is not true. For example, if $h$ is the quadratic polynomial associated to the angles $(13/63,14/63)$ then $f_{1/3} {\perp \! \! \! \perp}h$ is equivalent to a rational map with combinatorial data $(1/3,3)$. However $g_{1/3} {\perp \! \! \! \perp}h$ is equivalent to a rational map that does not have a period two cluster cycle. We should now specify exactly how the two critical displacements $\delta = 1$ and $\delta = 2q-1$ are obtained under the mating with the secondary map. Indeed, it is easy to see (for example, by considering “drawing the rays shut” in Figure \[f:critpic2\]) that if we are in the case where $R_h(-\theta)$ and $R_h(-(\theta+2))$ land at the same point, then $\delta = 1$. Similarly, if we are in the case where $R_h(-(\theta+1))$ and $R_h(-(\theta+3))$ land at the same point, then we will obtain $\delta = 2q-1$ for the resultant rational map. Conversely, if $\delta =1$ (respectively $\delta = 2q-1$) then the rays $R_h(-\theta)$ and $R_h(-(\theta+2))$ (respectively $R_h(-(\theta+1))$ and $R_h(-(\theta+3))$) land at the same point.
Shared Matings {#s:shared}
==============
A little further work shows that if $g {\perp \! \! \! \perp}h$ has critical displacement 1 (respectively $2n-1$), then $f {\perp \! \! \! \perp}h$ has critical displacement $2n-1$ (respectively 1). The proof of this fact will allow us to discuss how the matings are shared in the period two cluster case. In the case where the clusters were fixed, the nature of the sharing was relatively simple: one just needs to choose the appropriate rabbit and then mate it with the correct choice of a complementary map to get the required critical displacement (depending on whether the rabbit was chosen to be the first or second map in the mating). In this case, the introduction of the secondary map means the discussion of shared matings has an extra degree of complexity.
$G \cong g_{p/q} {\perp \! \! \! \perp}h$ has critical displacement $1$ (respectively $2q-1$) if and only if $F \cong f_{p/q} {\perp \! \! \! \perp}h$ has critical displacement $2q-1$ (respectively $1$).
By the observation at the end of the previous section, we see that if $G \cong g_{p/q} {\perp \! \! \! \perp}h$ has critical displacement $1$ then the rays $R^h_{-\theta}$ and $R^h_{-(\theta+2)}$ land at the same point on $J(h)$, namely the root point of $U$, the Fatou component which contains $h^{\circ 2}(0)$. So when we carry out the mating $f_{p/q} {\perp \! \! \! \perp}h$, we see that since $R^h_{-\theta}$ lands on the root point of $U$, the component $U$ will lie clockwise of the critical value component of the critical point belonging to $f_{p/q}$. This means that the critical displacement will be $2q-1$, as required. The case for critical displacement $2q-1$ is similar.
Conversely, by the fact that the branches of the ray equivalence classes $[p_1]$ and $[p_2]$ are all homeomorphic, it only is necessary to show that if $F \cong f_{p/q} {\perp \! \! \! \perp}h$ has a critical displacement of $1$, then the ray class for $g_{p/q} {\perp \! \! \! \perp}h$ is as in Figure \[f:critpic2\]. By assumption, the ray $R_h(-(\theta +3))$ must land on the root point $r$ of the critical orbit component of $h$ containing $h^{\circ 2}(0)$, which we denote $V_2$. We must show that the partner ray is $R_{h}(-(\theta+1))$. So let the partner ray be $R_h(\phi)$. Clearly, $\phi \neq -(\theta+2)$ since then $V_2$ would have to be the critical value component and $\phi \neq -\theta$ since this would create a loop in the ray equivalence class, meaning the mating would be obstructed. The union $R_{h}(-(\theta+3)) \cup R_{h}(\phi)$ separates $h^{\circ 2}(0)$ from the rest of the critical orbit of $h$ (this is easy to see by, for example, considering the Hubbard tree of $h$). In particular, if $\phi > -(\theta+1)$ then we must have $-(\theta+3) < -(\theta + 1) < \phi$, meaning the ray $R_{h}(-(\theta+1))$ will not land on the root point of a critical orbit Fatou component of $h$. But this would contradict $F$ having a period two cluster cycle, so we see that $\phi = -(\theta+1)$. In particular this means that the ray class is as in Figure \[f:critpic2\] and we are done.
We now have all the information we need to start enumerating examples, save for a simple calculation as was promised after the definition of the period two critical displacement.
Suppose the bicritical rational map $F$ with labelled critical points has a period two cluster cycle with critical data $(p/q,\delta)$. Then under the alternative labelling it has combinatorial data $(p/q,2p-\delta)$.
The critical displacement is by definition the combinatorial distance between $c_1$ and $F(c_2)$. By considering pre-images, this is the same as the distance between $F^{\circ (2q-1)}(c_1)$ and $c_2$. Since the combinatorial rotation number is $p/q$, we see that the distance between $F(c_1)$ and $c_2$ is $\delta - 2p$. Hence the critical displacement when $c_2$ is taken to be the first critical point is $2p-\delta$.
We see now that it is possible to get 2-fold, 3-fold and even 4-fold shared matings in the period two case. For example, we have the following equivalences of matings in the period $8$, rotation number $1/4$ case. The notation is as follows: $$\begin{aligned}
&f = f_{1/4} \text{ is the double rabbit corresponding to the angles $86/255$ and $89/255$.}\\
&g = g_{1/4} \text{ is the secondary map corresponding to the angles $87/255$ and $88/255$.}\\
&h_1 \text{ is the map corresponding to the angles $83/255$ and $84/255$.}\\
&h_3 \text{ is the map corresponding to the angles $77/255$ and $78/255$.}\\
&h_5 \text{ is the map corresponding to the angles $53/255$ and $54/255$.}\\
&h_7 \text{ is the map corresponding to the angles $211/255$ and $212/255$.}\end{aligned}$$
With this notation, we have the following equivalences: $$\begin{aligned}
&(\delta = 1) \quad \quad f {\perp \! \! \! \perp}h_1 \cong g {\perp \! \! \! \perp}h_7 \cong h_1 {\perp \! \! \! \perp}f \cong h_7 {\perp \! \! \! \perp}g \\
&(\delta = 3) \quad \quad f {\perp \! \! \! \perp}h_3 \cong h_7 {\perp \! \! \! \perp}f \cong h_1 {\perp \! \! \! \perp}g \\
&(\delta = 5) \quad \quad f {\perp \! \! \! \perp}h_5 \cong h_5 {\perp \! \! \! \perp}f \\
&(\delta = 7) \quad \quad f {\perp \! \! \! \perp}h_7 \cong g {\perp \! \! \! \perp}h_1 \cong h_3 {\perp \! \! \! \perp}f\end{aligned}$$ Indeed, generally the smallest multiplicity of sharing is $2$, and the greatest is $4$. Using the notation above (so that $h_{\delta}$ is the polynomial such that $F \cong f_{p/q} {\perp \! \! \! \perp}h_{\delta}$ has combinatorial data $(\rho,\delta)$) we always have the equivalence $$\label{e:shared}
f_{p/q} {\perp \! \! \! \perp}h_{\delta} \cong F \cong h_{2p-\delta} {\perp \! \! \! \perp}f_{p/q}.$$ Further, as we saw above, it is also sometimes possible to construct a rational map with a period two cluster cycle using a mating with the map $g_{p/q}$. In some cases (as with the case $\delta = 1$ above), this can give us a shared mating with multiplicity $4$.
All combinatorial data can be obtained by Proposition \[p:alldata2\]. By the observation of equation (\[e:shared\]), each set of data can be obtained in at least two ways and as stated above, this data can be obtained with a mating with the map $g_{p/q}$ in at most two more ways. As shown in the calculation of the case where $\rho = 1/4$ , all the possible multiplicities can be obtained.
Open questions {#s:openprobs}
==============
We now ask how we may generalise the results of this paper. The first question is, even if we restrict ourselves to the quadratic case, what can we say about the matings that produce a map with a period $p$ cluster cycle. Some preliminary calculations are in [@Mythesis] and Saul Schleimer (*personal communication*) has suggested that Thurston equivalence for the period three case should be possible. It is always possible to create such a rational map with the mating with an “order $p$” rabbit (the higher order analogues to the rabbits and “double rabbits” of this paper), but there are in general $2^{p-1}-1$ “secondary maps”in the period $p$ case, each of which having a period $p$ orbit with a well-defined combinatorial rotation number, and so being a viable candidate for a partipant in a mating that will create a rational map with a period $p$ cluster cycle. Such considerations may lead to a better understanding of parameter space: some considerations of the fixed case (and using parabolic, not hyperbolic maps) can be found in [@AdamBuffEcalle]. Moreover, the question remains how else can we create such rational maps and, assuming there are other ways, how prevalent are the shared matings in the higher period cases? Of course, any approach towards answering this question needs to resolve two questions. Firstly, what matings can admit a map with a period $p$ cluster cycle? Secondly, what is the analogue of the Thurston equivalence results as discussed in [@Thurstoneq]? Of course, an attempt to understand the higher degree cases would also be an interesting pursuit, since, as this and the sister paper show, even in the relatively benign period two case, the complications can cause even Thurston equivalence to fail in higher degrees.
*Acknowledgements.* I would like to thank my PhD supervisor, Dr. Adam Epstein, for all the help he has contributed towards my research and, in particular, this article. Furthermore, my thanks to Professors Mary Rees and Anthony Manning for providing a number of useful suggestions in the preparation of the manuscript. This research was funded by a grant from EPSRC.
[^1]: a global arm is a component of the complement to $p$ in the ray class
[^2]: In degree 2, this essentially says the well-known fact that there is a unique $p/q$-rabbit.
[^3]: The notion of the $1/2$-limb is no longer unique in degrees greater than $2$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The effects of laser-induced prealignment on the deflection of paramagnetic molecules by inhomogeneous static magnetic field are studied. Depending on the relevant Hund’s coupling case of the molecule, two different effects were identified: either suppression of the deflection by laser pulses (Hund’s coupling case (a) molecules, such as $ClO$), or a dramatic reconstruction of the broad distribution of the scattering angles into several narrow peaks (for Hund’s coupling case (b) molecules, such as $O_2$ or $NH$). These findings are important for various applications using molecular guiding, focusing and trapping with the help of magnetic fields.'
author:
- 'E. Gershnabel'
- 'M. Shapiro'
- 'I.Sh. Averbukh'
title: 'Stern-Gerlach deflection of field-free aligned paramagnetic molecules'
---
Introduction {#Introduction_magnetic}
============
Manipulating the translational motion of atoms and molecules by means of inhomogeneous external fields has been studied intensively for many years. Since the pioneering work of Stern and Gerlach that demonstrated quantization of atomic trajectories in inhomogeneous magnetic field [@SG], the dynamics of many other systems has been studied both in electric and magnetic fields. An important milestone was, for instance, separation of molecules in different quantum states in order to create a maser, a molecular amplifier of photons [@Townes]. Nowadays, the physics of the deflection of atoms and molecules by inhomogeneous fields is as hot as ever, including studies focused on the motion in the static inhomogeneous electric [@McCarthy; @Benichou; @Loesch; @Antoine; @reduction] and magnetic [@McCarthy; @Kuebler; @even] fields, and even laser fields [@Stapelfeldt; @Zhao1; @Zhao2; @Purcell]. In the case of laser deflection, some novel applications in molecular optics have recently appeared, such as molecular lens [@Stapelfeldt; @Zhao1] and molecular prism [@Zhao2; @Purcell]. The interaction between a molecule and an external field depends upon the orientation of the molecule. The field-molecule interactions become intensity-dependent for strong enough fields due to the field-induced modification of the molecular angular motion [@Zon; @Friedrich]. It was lately shown that the intensity-dependent molecular polarizability-anisotropy provides means for tailoring the dipole force felt by molecules in the laser field [@Purcell]. More recently, a method for controlling the scattering of molecules in external fields by additional ultrashort laser pulses inducing field-free molecular alignment was suggested [@gershnabel1; @gershnabel5; @gershnabel4].
In this work we return to the basics, and study the prospects of the ultrafast laser control of molecular deflection in the Stern-Gerlach (SG) arrangement. It was shown in the past that molecular scattering in magnetic fields is affected by rotational alignment caused, for example, by collisions in seeded supersonic beams [@Aqullanti]. Here we demonstrate that this process can be efficiently and flexibly controlled by novel ultafast optical tools allowing for preshaping the molecular angular distribution before the molecules enter the SG apparatus. This can be done with the help of numerous recent techniques for laser molecular alignment, which use single or multiple short laser pulses (transform limited, or shaped) to temporarily align molecular axes along certain directions (for introduction to the rich physics of laser molecular alignment, see, e.g. [@Zon; @Friedrich; @Stapelfeldt2; @Kumarappan; @Stolow; @rich]). Short laser pulses excite rotational wavepackets, which results in a considerable transient molecular alignment after the laser pulse is over, i.e., at field-free conditions. In the present paper, we will consider only molecules with a permanent magnetic dipole moment, i.e., open shell molecules. The open shell molecules are classified into Hund’s coupling cases according to their angular momenta coupling [@Townes; @Carington]. In the Hund’s coupling case (a), the angular momentum of electrons and their spin are coupled to the internuclear axis, while in the Hund’s coupling case (b), the electronic spin and internuclear axis are not strongly coupled. We will consider magnetic deflection of different paramagnetic molecules subject to a short prealigning laser pulse. In the Hund’s coupling case (a), the magnetic moment is coupled to the internuclear axis, and by rotating the molecule, one rotates the magnetic moment as well. This reduces substantially the Zeeman effect and effectively turns off the interaction between the molecule and magnetic field (like a rotating electric dipole that becomes decoupled from a static electric field [@gershnabel5]). In the Hund’s coupling case (b), the magnetic moment is barely coupled to the internuclear axis. However, laser-induced molecular rotation creates an effective magnetic field which adds to the SG field and modifies the deflection dynamics. As a result, as we show below, a broad and sparse distribution of the scattering angles of deflected molecules collapses into several narrow peaks with controllable positions.
The paper is organized as following. In Sec. \[General Theory\] we outline the general theorical framefork: first, we briefly discuss the SG deflection mechanism (Sec. \[Stern-Gerlach deflection\]), and provide several needed facts on the laser-induced field-free alignment in Sec. \[Prealignment\]. Then, the interaction details for the Hund’s coupling case (a) and Hund’s coupling case (b) are given in Sec. \[Hund case A\] and \[Hund case B\], respectively. Further discussion of the Hund’s coupling case (b) and hyperfine structure appears in the Appendix in Sec. \[NH HFS\]. In Sec. \[Applications to Molecules\] we apply the above theoretical tools to the laser-controlled magnetic scattering of $ClO$ (Sec. \[ClO\]), $O_2$ (Sec. \[$O_2$\]) and $NH$ (Sec. \[NH\]) molecules. Discussion followed by the conclusions, are presented in Sec. \[Discussions\] and \[Summary\], respectively.
General Theory {#General Theory}
==============
Stern-Gerlach deflection {#Stern-Gerlach deflection}
------------------------
Once a beam of molecules enters into a SG magnetic field, the initial eigenstates of the system adiabatically become $|\Psi_i(B)\rangle$:
$$|\Psi_i(B)\rangle=\sum_{j}a_{j}(B)|\Psi_j\rangle, \label{Hund case b diagonalized}$$
where the coefficients $a_{j}(B)$ depend on the magnetic field B as a parameter, and $|\Psi_j\rangle$ is a basis for the free molecule. In this work we consider the magnetic field to be: $\textbf{B}=B(z)\hat{z}$, i.e., pointing in the $z$ direction, with a practically constant gradient along the $z$ direction in the relevant interaction region.
The force acting on the molecule is given by: $$Force=-\nabla E=-\frac{\partial E}{ \partial B}\frac{\partial B}{ \partial z},
\label{Force}$$ where $E$ is the energy of the molecule. The derivative ${\partial E}/{ \partial B}$ may be obtained by means of the Hellman-Feynman theorem, that is, for a system being in the $i$-th energy eigenstate of the system, the force is proportional to:
$$\begin{aligned}
\frac{\partial E_i}{\partial B}&=&\langle\Psi_i(B)|\frac{\partial H}{\partial B}|\Psi_i(B)\rangle\nonumber\\
&=&\langle\Psi_i(B)|\frac{\partial H_z}{\partial B}|\Psi_i(B)\rangle,
\label{Hellman_Feyman}\end{aligned}$$
where $H_z$ is the Zeeman term of the Hamiltonian. Since $H_z$ is proportional to $B$, we conclude that a molecule in an energy eigenstate will be deflected by a force that is proportional to:
$$\begin{aligned}
{\cal A}_{i}\equiv\langle\Psi_i(B)|\frac{H_z}{B}|\Psi_i(B)\rangle.\label{A_force}\end{aligned}$$
Eq. \[A\_force\] will allow us to consider the distribution of forces. In order to take into account the absolute amount of deflection, though, one has to consider the field gradient as well (Eq. \[Force\]). For more details, see, e.g. [@McCarthy].
Laser-induced field-free alignment {#Prealignment}
----------------------------------
If the molecules are subject to a strong linearly polarized femtosecond laser pulse, the corresponding molecule-laser interaction potential is given by: $$H_{ML}=-\frac{1}{4}\epsilon^2\left [ (\alpha_{\parallel}-\alpha_{\perp}) \cos^2\theta+\alpha_{\perp}\right ],\label{pre alignment interaction}$$ where $\theta$ is the angle between the molecular axis and the polarization direction of the pulse, $\alpha_{\parallel},\alpha_{\perp}$ are the parallel and perpendicular polarizability components, and $\epsilon$ is the femtosecond pulse envelope. Since the aligning pulse is short compared to the typical periods of molecular rotation, it may be considered as a delta-pulse. In the impulsive approximation, one obtains the following relationship between the wavefunction before and after the pulse applied at $t=0$ (see e.g. [@Gershnabel3], and references therein): $$\Psi(t=0^+)=\exp{[iP\cos^2\theta]}\Psi(t=0^-),\label{prealignment operator}$$ where the kick strength, $P$ is given by:
$$P=(1/4\hbar)\cdot (\alpha_{\parallel}-\alpha_{\perp})\int_{-\infty}^{\infty}\epsilon^2(t)dt.\label{kick strength}$$
We assume the vertical polarization of the pulse (along the $z$-axis, and parallel to the SG magnetic field). Physically, the dimensionless kick strength $P$, is equal to the typical amount of angular momentum (in the units of $\hbar$) supplied by the pulse to the molecule. In order to find $\Psi(t=0^+)$ for any initial state, we introduce an artificial parameter $\xi$ that will be assigned the value $\xi=1$ at the end of the calculations, and define:
$$\begin{aligned}
\Psi_{\xi}=\exp{\left [ (iP\cos^2\theta)\xi \right ]}\Psi(t=0^-)
=\sum_{i} c_{i}(\xi)|\Psi_i\rangle.\label{artificial}\end{aligned}$$
By differentiating both sides of Eq. \[artificial\] with respect to $\xi$, we obtain the following set of differential equations for the coefficients $c_{i}$:
$$\dot{c}_{i'}=iP\sum_{i}c_{i}\langle \Psi_{i'}|\cos^2\theta|\Psi_i\rangle,
\label{Pre-Alignment Coefficients}$$
where $\dot{c}=dc/d\xi$. Evaluation of the matrix elements in Eq. \[Pre-Alignment Coefficients\] is easily obtained by means of the relationship: $\cos^2\theta=(2D^2_{00}+1)/3$, where $D^k_{pq}$ is the rotational matrix.
Since $\Psi_{\xi=0}=\xi(t=0^-)$ and $\Psi_{\xi=1}=\Psi(t=0^+)$ (see Eq. \[artificial\]), we solve numerically this set of equations from $\xi=0$ to $\xi=1$, and find $\Psi(t=0^+)$. It turns out that the population of rotational levels of the kicked molecules has a maximum at around $\hbar P$.
Finally, we derive the distribution of forces acting on a thermal ensemble of molecules pre-aligned by a laser pulse. For this, we start from a single eigenstate of a free system, apply an alignment pulse in the $z$ direction, and then adiabatically increase the magnetic field (in order to imitate a smooth process of the molecular beam injection into the SG deflector). The distribution will be proportional to:
$$\begin{aligned}
f({\cal A})&=&\sum_{i,j}\frac{\exp\left(-\frac{E_{i}}{k_B T}\right)}{Q_{rot}}\nonumber\\
&\times&|c_{j}|^2\delta_{{\cal A},{\cal A}_{j}},
\label{distribution of forces}\end{aligned}$$
where $k_B$ is the Boltzmann’s constant, $Q_{rot}$ is the partition function, $i$ denotes the quantum numbers associated with the initial eigenstates of free molecules, $c_{j}$ denotes the coefficients of the free eigenstates that were excited by the laser pulse applied to the initial eigenstate $i$, and ${\cal A}_{j}$ are the associated matrix elements given in Eq. \[A\_force\] (proportional to the force), between the states adiabatically correlated with the free states $j$.
Hund’s coupling case (a) {#Hund case A}
------------------------
In this subsection we concentrate on the $^{35}ClO$ molecule, which presents a good example for the Hund’s coupling case (a). Denoted as $^2\Pi$ in its electronic ground state, it has a nuclear spin $I=3/2$, and it was studied well in the past [@Carrington1; @Kakar; @Brian].
In the Hund’s coupling case (a), the electronic angular momentum and spin are strongly coupled to the internuclear axis, and in the case of $ClO$, its effective Hamiltonian is given by [@Carington]:
$$H_{eff}=H_{rso}+H_{hf}+H_Q,\label{ClO molecule Hamiltonian}$$
where $H_{rso}$ is the rotation and spin-orbit coupling, $H_{hf}$ is the magnetic hyperfine interaction, and $H_Q$ is the electric quadrupole interaction. Here
$$H_{rso}=B_r\left\{T^1(\textbf{J})-T^1(\textbf{L})-T^1(\textbf{S})\right\}^2+AT^1(\textbf{L})\cdot T^1(\textbf{S}),\label{rso Hamiltonian}$$
where $T^1()$ is a spherical tensor of rank $1$, $B_r$ is the rotational constant in the lowest vibrational level, and $A$ is the spin-orbit coupling constant. $\textbf{L}$ and $\textbf{S}$ are the electronic angular momentum and spin operators, respectively. The total angular momentum is $\textbf{J}=\textbf{N}+\textbf{L}+\textbf{S}$, where $\textbf{N}$ is the nuclei angular momentum operator. The Hund’s coupling case (a) basis looks like this: $$|\eta,\Lambda;S,\Sigma;J,\Omega,I,F,M_F\rangle,
\label{Hund case A basis}$$ where $\eta$ represents some additional electronic and vibrational quantum numbers, $\Sigma$ and $\Lambda$ are the projections of the electronic spin and angular momentum on the internuclear axis, respectively. For $ClO$ molecule, $S=1/2$, so that $\Sigma=\pm 1/2$ and $\Lambda=1$. The quantity $\Omega$ is $\Omega\equiv\Sigma+\Lambda$ ($\Omega=3/2,1/2$, the $3/2$-state has a lower energy), and $\textbf{F}=\textbf{J}+\textbf{I}$.
The $H_{hf}$ Hamiltonian is given by: $$\begin{aligned}
&&H_{hf}=H_{IL}+H_F+H_{dip}\nonumber\\
&=&aT^1(\textbf{I})\cdot T^1(\textbf{L})+b_FT^1(\textbf{I})\cdot T^1(\textbf{S})\nonumber\\
&-&\sqrt{10}g_S\mu_B g_N \mu_N (\mu_0 /4\pi)T^1(\textbf{I})\cdot T^1(\textbf{S},\textbf{C}^2).\label{H_hf_ClO}\end{aligned}$$
The first term represents the orbital interaction, the second one accounts for the Fermi contact interaction, and the third term describes the dipolar hyperfine interaction. Here $a$ and $b_F$ are constants, $g_N$ and $g_S$ are the nuclear and electron $g$ factors, respectively, $\mu_N$ and $\mu_B$ are the nuclear and electron Bohr magnetons, respectively, and $\mu_0$ is the vacuum permeability.
All the matrix elements for the Hund’s coupling case (a), including those for the quadrupole interaction, are given in [@Carington]. The $ClO$ constants were taken from [@Carington; @Kakar; @Brian].
When considering the Zeeman Hamiltonian, we will concentrate only on the two major terms related with electronic angular momentum and spin: $$H_Z=\mu_B T^1(\textbf{B})\cdot T^1(\textbf{L})+g_S\mu_B T^1(\textbf{B})\cdot T^1(\textbf{S}).\label{ClO Zeeman}$$ The corresponding matrix elements (see Eq. \[A\_force\]) are given in [@Carington].
In order to consider the effect of laser-induced alignment (see Sec. \[Prealignment\], Eq. \[Pre-Alignment Coefficients\]), we have derived the following matrix elements:
$$\begin{aligned}
&\langle& \eta,\Lambda; S,\Sigma;J,\Omega,I,F,M_F|D^{2*}_{00}|\eta,\Lambda;S,\Sigma;J',\Omega,I,F',M_F\rangle\nonumber\\
&=&(-1)^{F-M_F}\left( \begin{array}{ccc} F & 2 & F' \\ -M_F & 0 & M_F \end{array} \right)(-1)^{F'+J+I+2}\nonumber\\
&\times& \sqrt{(2F'+1)(2F+1)}\left\{ \begin{array}{ccc} J' & F' & I \\ F & J & 2 \end{array} \right\}(-1)^{J-\Omega}\nonumber\\
&\times&\left( \begin{array}{ccc} J & 2 & J' \\ -\Omega & 0 & \Omega \end{array} \right)\sqrt{(2J+1)(2J'+1)}.\nonumber\\\label{ClOPulse}\end{aligned}$$
Hund’s coupling case (b) {#Hund case B}
------------------------
We will continue by discussing the Hund’s coupling case (b), and consider the Oxygen molecule, in its predominant isotopomer $^{16}O^{16}O$. This molecule is probably the most important species among $^3\Sigma$ ground state molecules, and it was one of the first molecules studied in detail [@Kuebler; @Tinkham]. It is a homonuclear diatomic molecule, where only odd N’s appear because of the Pauli’s principle and symmetry. This molecule is described well by the Hund’s coupling case (b), with the effective Hamiltonian [@Carington]:
$$H_{eff}=H_{rot}+H_{ss}+H_{sr}.\label{Effective_Hamiltonian}$$
Let us describe separately each term in Eq. \[Effective\_Hamiltonian\]. Here
$$H_{rot}=B_r\textbf{N}^2-D\textbf{N}^4,\label{Rotational}$$
is the energy of the nuclei rotation, where $D$ is the centrifugal distortion coefficient. In addition,
$$H_{ss}=-g_s^2\mu_B^2(\mu_0/4\pi)\sqrt{6}T^2(\textbf{C})\cdot T^2(\textbf{S}_1,\textbf{S}_2),\label{SpinSpin}$$
is the electornic spin-spin dipolar interaction. $T^2()$ is a spherical tensor of rank $2$. $\textbf{S}_1,\textbf{S}_2$ are electronic spin operators. $T^2_q(\textbf{C})=\langle C^2_q(\theta,\phi)R^{-3}\rangle$, where $C^2_q$ is the spherical harmonics, and
$$H_{sr}=\gamma T^1(\textbf{N})\cdot T^1(\textbf{S}),\label{SpinRotation}$$
is the electronic-spin rotation interaction. The Hund’s coupling case (b) basis looks like this: $$|\eta,\Lambda;N,\Lambda;N,S,J,M_J\rangle. \label{Hund case b}$$
Here $N$ is the nuclei rotational quantum number, $\Lambda=0$ in our case, $\textbf{S}$ is the electronic spin, which is $1$ in our case, $\textbf{J}=\textbf{N}+\textbf{S}$, and $M_J$ is the projection of $\textbf{J}$ onto a fixed $z$-direction in space. All the needed matrix elements and constants are given in [@Carington]. The Zeeman Hamiltonian is given by:
$$H_Z=g_S \mu_B T^1(\textbf{B})\cdot T^1(\textbf{S}).\label{Zeeman Oxygen}$$
Its matrix elements (Eq. \[A\_force\]) are given by:
$$\begin{aligned}
&d \langle& \eta ,\Lambda;N,\Lambda;N,S,J,M_J|T^1_0(\textbf{S})|\eta,\Lambda;N',\Lambda;N',S,J',M_J\rangle\nonumber\\
&=& d (-1)^{J-M_J}\left( \begin{array}{ccc} J & 1 & J' \\ -M_J & 0 & M_J \end{array} \right) \delta_{N,N'}(-1)^{J'+S+1+N}\nonumber\\
&\times&\sqrt{(2J'+1)(2J+1)}\left\{ \begin{array}{ccc} S & J' & N \\ J & S & 1 \end{array} \right\}\nonumber\\
&\times& \sqrt{S(S+1)(2S+1)},
\label{Zeeman Deivation}\end{aligned}$$
where $d\equiv g_S \mu_B$. Finally, in order to account for the laser-induced prealignment, we derived the following relation (to be used in Eq. \[Pre-Alignment Coefficients\]):
$$\begin{aligned}
&&\langle \eta ,\Lambda;N,\Lambda;N,S,J,M_J|D^{2*}_{00}|\eta,\Lambda;N',\Lambda;N',S',J',M_J\rangle\nonumber\\
&=& (-1)^{J-M_J}\left( \begin{array}{ccc} J & 2 & J' \\ -M_J & 0 & M_J \end{array} \right) \delta_{S,S'}(-1)^{J'+S+2+N}\nonumber\\
&\times&\sqrt{(2J'+1)(2J+1)}\left\{ \begin{array}{ccc} N' & J' & S \\ J & N & 2 \end{array} \right\}(-1)^{N-\Lambda}\nonumber\\
&\times&\left( \begin{array}{ccc} N & 2 & N' \\ -\Lambda & 0 & \Lambda \end{array} \right)\sqrt{(2N+1)(2N'+1)}.
\label{COS2Derivation}\end{aligned}$$
In the case of the oxygen molecule, there is a relatively strong effect of the spin-spin interaction, which complicates our analysis. Therefore, we have also chosen an additional $^3\Sigma$ molecule, $^{14}NH$ for our study. For this molecule the ratio between the spin-spin and spin-rotation interactions is reduced (compared to the $O_2$ case). This makes $NH$ a simpler candidate to test our rotational effects. The $NH$ molecule was thoroughly studied in the past [@Wayne; @Klaus; @Jesus; @Lewen], and its effective Hamiltonian is: $$H_{eff}=H_{rot}+H_{ss}+H_{sr}+H_{HFS}, \label{Hamiltonian_NH}$$ where $H_{rot}$, $H_{ss}$ and $H_{sr}$ were defined in Eq. \[Rotational\], \[SpinSpin\] and \[SpinRotation\]. Since $NH$ has non-zero nuclei spin ($N$ has nuclear spin $I=1$, $H$ has $I=1/2$), then it has a hyperfine structure described by the Hamiltonian $H_{HFS}$. Further elaboration on the hyperfine structure of NH (including details on the Zeeman term, and the matrix elements related to Eq. \[Pre-Alignment Coefficients\]), is given in the Appendix in Sec. \[NH HFS\].
Laser control of the Stern-Gerlach scattering {#Applications to Molecules}
=============================================
$ClO$ {#ClO}
-----
In this part of the work, we apply the theoretical tools that were presented in the previous sections to the SG scattering of the $ClO$ molecule. This molecule exhibits a good Hund’s coupling case (a), and details about it were already given in Sec. \[Hund case A\]. We will consider here its ground state ($T=0K$), for which $\Lambda=1,\Sigma=1/2,\Omega=3/2,J=3/2,F=0,M_F=0$. In Fig. \[ClO Distribution no kick\] we present the force distribution (Eq. \[distribution of forces\]) for a $ClO$ molecule in the ground state that is deflected by a SG magnetic field. As only single molecular state is occupied, the force has a well defined single value.
![The force distribution for a beam of $ClO$ molecules that are deflected by a magnetic field of $1T$. The temperature is $0K$, therefore only the ground state is considered, and the distribution reduces to a single-value peak.[]{data-label="ClO Distribution no kick"}](ClO_No_Align1.eps){width="70mm"}
As the next step, we assume that the molecules are subject to a short laser pulse with a kick strength of $P=30$ (Eq. \[kick strength\]) before they enter the SG magnetic field. The new force distribution is given in Fig. \[ClO Distribution P30\].
![The force distribution for a beam of prealigned $ClO$ molecules. The temperature is $0K$, and the kick strength of the laser is $P=30$. The prealigned molecule is deflected by a magnetic field of $1T$. This distribution should be compared to the one from Fig. \[ClO Distribution no kick\].[]{data-label="ClO Distribution P30"}](ClO_Align1.eps){width="70mm"}
By comparing Fig. \[ClO Distribution no kick\] to Fig. \[ClO Distribution P30\], it can be observed that the effect of the laser-induced field-free alignment is to effectively turn-off the interaction between the molecule and the magnetic field. This effect is similar to the one discussed by us recently in connection with the scattering of polar molecules by inhomogeneous static electric fields [@gershnabel5]. Moreover, rotation-induced dispersion in molecular scattering by static electric fields was used as a selection tool in recent experiments on laser molecular alignment [@reduction]. A related phenomenon of the reduction of the electric dipole interaction in highly excited stationary molecular rotational states was observed there. Further details and discussion about the $ClO$ magnetic deflection is provided in Sec. \[Discussions\].
$O_2$ {#$O_2$}
-----
In this sub-section we consider the $O_2$ molecule. The $O_2$ molecule is described well by Hund’s coupling case (b) scheme, and the details about it were given in Sec. \[Hund case B\]. First, we consider a beam of $O_2$ molecules at $0K$, i.e., in the ground state ($N=1$, $J=0$, and $M_J=0$). These molecules enter a magnetic field of $1T$, and are deflected by this field. The force distribution for these molecules is given in Fig. \[Results1\].
![The force distribution for a beam of $O_2$ molecules that are deflected by a $1T$ magnetic field. The temperature is $0K$, i.e., only the ground state is populated and therefore the distribution reduces to a single-value peak.[]{data-label="Results1"}](Oxygen0T_No_Alignment1.eps){width="70mm"}
Second, we consider the action of the prealignment pulses of different kick strengths ($P=10, 30, 70$) before the molecules enter the deflecting field. The distribution of forces at $0K$ is given in Fig. \[FinalPlot1\], where two major peaks are observed. As the strength of the pulses is increased, higher rotational states are excited, and the peaks become closer to each other.
![The force distribution for a beam of prealigned $O_2$ molecules. Different kick strengths ($P=10,30,70$) are considered and the magnetic field is $1T$ (temperature is $0K$). As the excitation is increased, the two major peaks become closer to each other. This distribution should be compared to the one from Fig. \[Results1\].[]{data-label="FinalPlot1"}](Oxygen0K1T_Alignment1.eps){width="80mm"}
Third, we consider deflection of thermal molecules without and with prealignment, in Figs \[Results3\] and \[Results4\], respectively. By comparing Fig. \[FinalPlot1\] to Fig. \[Results4\], we observe an additional peak in Fig. \[Results4\]. As the strength of the prealignment pulses is increased, the peaks in Fig. \[Results4\] are changed: they become narrower and the two left peaks become closer to each other. Further discussion on $O_2$ will be provided in Sec. \[Discussions\].
![The force distribution for $O_2$ molecules. The temperature is $5K$ and the magnetic field is $1T$.[]{data-label="Results3"}](Oxygen5K_1T_No_Align1.eps){width="80mm"}
![The force distribution for a beam of $O_2$ molecules, prealigned by a laser field ($P=10,30,70$). The temperature is $5K$, and the magnetic field is $1T$. Here we observe three major peaks. As the laser excitation strength is increased, the peaks become narrower, and the two left peaks become closer to each other.[]{data-label="Results4"}](Oxygen5K1T_Align1.eps){width="60mm"}
$NH$ {#NH}
----
Finally, we consider the $NH$ molecule. This molecule is described well by a Hund’s coupling case (b) scheme, similar to the $O_2$ molecule, however it has a reduced spin-spin to spin-rotation interaction ratio. This makes the $NH$ molecule a simpler candidate for the theoretical analysis. In Fig. \[NH\_distribution\] we plot the force distribution for the ground state $N=0,J=1,F_1=3/2,F=1/2$ molecules that were prealigned by laser pulses of different intensity. $M_F$ was taken to be $1/2$ for certainty, and the case of $M_F=-1/2$ may be considered similarly (with similar consequences, as will be described in Sec. \[Discussions\]).
![The distribution of forces for a beam of $NH$ molecules, that were prealigned (starting from the lowest state $N=0,J=1,F_1=3/2,F=1/2$) by means of laser pulses of different strengths: $P=10$ (green), $P=30$ (blue) and $P=40$ (red). Only $M_F=1/2$ is considered, at a $2T$ magnetic field.[]{data-label="NH_distribution"}](NH_Distribution1.eps){width="80mm"}
One may observe the presence of three major peaks now (for the $O_2$ molecules in the ground state there were only two peaks). As the strength of the prealignment pulse is increased, the major peaks are shifted in position. An additional difference between the $NH$ and the $O_2$ molecules is that now the peak to the right is also shifted with increasing the strength of the prealignment laser pulse. Further elaboration about this molecule, and the difference between it and $O_2$, is given in Sec. \[Discussions\].
Discussion {#Discussions}
==========
$ClO$ {#Discussion, ClO}
-----
First we will discuss the $ClO$ molecule, which exhibits a good Hund’s coupling case (a). Having both electronic angular momentum and spin coupled to the internuclear axis, rotation of the molecule by means of short laser pulses leads to the rotation of the molecular magnetic moment as well. The interaction between the SG magnetic field and the rapidly rotating magnetic moment of the molecule will be thus averaged to zero, leading to the negligible magnetic forces.
$O_2$ {#Discussion, O2}
-----
In Fig. \[RegularLamda1\] we plot the forces vs. magnetic field, for several values of $J$. First, we observe that for a high magnetic field all the curves are separated to form a three SG splitting pattern [@Kuebler]. In the limit of the low magnetic field (and slow rotations), the energy spectrum of the molecule is rather complex due to the spin-spin interaction [@Tinkham]. At around $1T$, though, we are in the regime where the spin-rotation ($H_{sr}$) interaction has a rather strong dynamic control: as $N$ is increased (by the means of prealignment, for instance) then a sizable shift of the force magnitude is observed. This is the origin of the behavior of the distributions of Fig. \[FinalPlot1\] and Fig. \[Results4\]. It can also be observed in Fig. \[RegularLamda1\] that the spin-spin term is larger than the spin-rotation one, and it shifts the curve for $J=N$ from the two other curves. We find also that in the case of $J=N+1,N-1$, the forces are more susceptible to different $N$s, which is observed in the distribution of forces in Fig. \[FinalPlot1\] and Fig. \[Results4\].
![Forces vs. magnetic field for the $O_2$ molecule. The $y$ axis is given in arbitrary units, the $x$ axis is given in the units of Tesla. Blue, green, and red (solid lines) correspond to $N=31$, $J=30,31,32$ ($M_J=0$), respectively. Blue, green, and red (dashed lines) correspond to $N=71$ and $J=70,71,72$ ($M_J=0$), respectively. The effects of the spin-spin interaction reveal themselves in the fact that the upper level $J=N$ is well separated from two almost degenerate levels with $J=N+1,N-1$. Magnetic field near $1T$ is optimal for observing the sensitivity of the deflecting force to the $N$ variation.[]{data-label="RegularLamda1"}](OxygenLevels1.eps){width="60mm"}
Fig. \[RegularLamda1\] also allows us to understand the position of the peaks in Fig. \[FinalPlot1\] and Fig. \[Results4\]. The right peak that appears in Fig. \[Results4\] and does not appear in Fig. \[FinalPlot1\] corresponds to the $J=N$ states. Due to selection rules (Eq. \[COS2Derivation\]), the odd $J$s, i.e., the $J=N$ states, are never excited (if we start from $J=0$, and $M_J=0$ at $0K$). This is why we observe only two peaks, i.e., the $J=N\pm1$ peaks, in Fig. \[FinalPlot1\]. Considering a deflection of the molecules in the ground state alone is important experimentally. Even if one considers an experiment at $T=1K$ ($k_BT=20837MHz$), then the difference between ($N=1, J=0$) and ($N=1, J=2$) (the next energy) is $62486MHz$, which is large enough. Though, for $1K$ we should expect a small peak in the distribution of forces for $J=N$ states. In the case of larger temperature (Fig. \[Results4\]), we start from different $M$s, and also the odd $J$s are present, therefore, we observe the right peak at Fig \[Results4\]. As the prealignment becomes stronger, the distribution transforms into three peaks, each correspond to either $J=N$, $J=N-1$ or $J=N+1$ states.
$NH$ (and the imaginary $\widetilde{O_2}$ molecule!) {#Discussion, NH}
----------------------------------------------------
Before we start with the $NH$ molecule, we consider an imaginary $\widetilde{O_2}$ molecule (!). This molecule is similar to the $O_2$ molecule, only with a spin-spin interaction that is reduced by a factor of $100$. The forces vs. magnetic field for the imaginary $\widetilde{O_2}$ molecule are plotted in Fig. \[LamdaSmall\], where we get a symmetric splitting of the $N$ level into $J=N,N\pm1$ ($J=N$ is in the middle, as is intuitively expected for SG splittings, unlike in the $O_2$ case). Such behavior corresponds to a spin-spin interaction that is negligible compared with the spin-rotation.
![Forces vs. magnetic field for the imaginary $\widetilde{O_2}$ molecule (details in the text). The $y$ axis is given in arbitrary units, the $x$ axis is given in the units of Tesla. Blue, green, and red (solid lines) correspond to $N=31$, $J=30,31,32$ ($M_J=0$), respectively. Blue, green, and red (dashed lines) correspond to $N=71$ and $J=70,71,72$ ($M_J=0$), respectively. At about $1T$ for this molecule, we observe approximately a symmetric splitting into three graphs for $J=N$ and $N\pm1$, where $J=N$ is in the middle (unlike in the $O_2$ case). []{data-label="LamdaSmall"}](ImaginaryMoleculeCurves1.eps){width="70mm"}
As we intuitively suggested in Sec. \[Introduction\_magnetic\], when one applies prealignment to molecules belonging to the Hund’s coupling case (b), the electronic spin feels the SG field combined with the effective magnetic field due to nuclei rotations. The latter field is along the $N$-vector, i.e. perpendicular to the molecular axis. A strong enough vertically polarized laser pulse excites molecular rotations in the vertical planes containing the $z$-axis. As a result, the rotation-induced effective magnetic field is perpendicular to the vertical SG field. Therefore, the force felt by the molecules is given by
$$Force=\frac{K_0B}{\sqrt{B^2+K_1^2}},\label{Fit to curve}$$
where $K_0$ and $K_1$ are constants (the latter is proportional to $N$ or $P$).
Fig. \[LamdaSmall\] presents results of the exact quantum-mechanical calculation of the SG force for our imaginary $\widetilde{O_2}$ molecule. We considered the upper curves in this figure, and tried to fit them to the above analytical expression. We find an excellent agreement between the original data and the fitted curves, and the results of the fit are $K_1=0.61,0.25$ for $N=71,31$, respectively. We also find a good agreement between the ratio of $N's$ (i.e., $71/31=2.3$) and of $K_1's$ (i.e., $0.61/0.25=2.4$).
As we have mentioned before, the $NH$ molecule is also characterized by a reduced value of the spin-spin interaction compared to the spin-rotation interaction. Therefore, its dynamics should be closer to the imaginary $\widetilde{O_2}$ molecule than to the real $O_2$ molecule considered above. In Fig. \[NH\_forces\] we plot some forces vs. the magnetic field for the $NH$ molecule, and indeed, observe a triplet-like structure similar to that one in Fig. \[LamdaSmall\] (with the curve for $J=N$ being in the middle for large enough $N$).
![Forces vs magnetic field for the $NH$ molecule (including the fine and hyperfine details), for $N=10$ (green), $N=30$ (blue) and $N=41$ (red). Only $M_F=1/2$ is considered here, but $M_F=-1/2$ gives the same results (only higher $M_F$’s will modify the spectrum). For $N=10$ the upper/middle/lower curves correspond to $J=N,J=N-1,J=N+1$, respectively, as in the Oxygen case. For $N=30,41$ the upper/middle/lower curves correspond to $J=N-1,J=N,J=N+1$, respectively, as in Fig. \[LamdaSmall\].[]{data-label="NH_forces"}](NH_Curves1.eps){width="70mm"}
By analyzing the results shown in Fig. \[NH\_forces\] we conclude that the hyperfine structure details in $NH$ are not crucially important for our considerations, but the reduced spin-spin to spin-rotation interaction ratio for $NH$ defines the major difference of the deflection dynamics compared to the case of $O_2$. Also one notices the scaling with the magnetic field: here the higher values of the magnetic field are required ($2T$) to observe the collapse of the broad distribution of forces into three narrow groups. This is due to the increased spin-rotation interaction for $NH$ (as compared to $O_2$). Fig. \[NH\_forces\] explains the behavior of the distribution in Fig. \[NH\_distribution\], where we have noticed that the three peaks are shifted as one increases the laser pulse strength.
Conclusions {#Summary}
===========
We considered scattering of paramagnetic molecules by inhomogeneous magnetic field in a Stern-Gerlach-type experiment. We showed that by prealigning the molecules before they interact with the magnetic field, one obtains an efficient control over the scattering process. Two qualitatively different effects were found, depending on the Hund’s coupling case of the molecule. For molecules that belong to the Hund’s coupling case (a), we showed that the deflection process may be strongly suppressed by laser pulses. This may be implemented as an optical switch in the molecular magnetic deceleration techniques [@even]. Furthermore, for the Hund’s coupling case (b) molecules, a sparse distribution of the scattering angles is transformed into a distribution with several compact deflection peaks having controllable positions. Each peak corresponds to a scattered molecular sub-beam with increased brightness. The molecular deflection is considered as a promising route to the separation of molecular mixtures. Narrowing and displacing scattering peaks may substantially increase the efficiency of separating multi-component beams, especially when the prealignment is applied selectively to certain molecular species, such as specific isotopes [@isotopes], or nuclear spin isomers [@isomers1; @isomers2]. One may envision more sophisticated schemes for controlling molecular scattering, which involve multiple pulses with variable polarization for preshaping molecular angular distribution. In particular, molecular rotation may be confined to a certain plane by using the ”optical molecular centrifuge” approach [@centrifuge; @Mullin], double-pulse ignited ”molecular propeller” [@propeller], or permanent planar alignment induced by a pair of delayed perpendicularly polarized short laser pulses [@France1; @France2]. If molecules are prepared like this, a narrow angular peak is expected in their scattering distribution from a magnetic field. The position of the peak is controllable by inclination of the plane of rotation with respect to the deflecting field, similar to a related effect for molecular scattering in inhomogeneous electric fields see [@gershnabel4]). Moreover, further manipulations of the deflection process may be considered, e.g., by using several SG fields with varying directions. Magnetic deflection of $O_2$ molecules subject to laser-induced field-free manipulations, is currently a subject of an ongoing experimental effort.
ACKNOWLEDGMENT {#acknowledgment .unnumbered}
==============
We enjoyed many stimulating discussions with Valery Milner and Sergey Zhdanovich. One of us (IA) appreciates the kind hospitality and support at the University of British Columbia (Vancouver). This work is supported in part by grants from the Israel Science Foundation, and DFG (German Research Foundation). Our research is made possible in part by the historic generosity of the Harold Perlman Family. IA is an incumbent of the Patricia Elman Bildner Professorial Chair.
Appendix: $NH$ (Hund’s coupling case (b)) Hyperfine structure {#NH HFS}
=============================================================
The hyperfine structure for the $NH$ molecule is described by the following Hamiltonian [@Carington; @Klaus; @Jesus; @Lewen]:
$$\begin{aligned}
H_{HFS}&=&\sum_k {b_{F_k} T^1(\textbf{I}_k)\cdot T^1(\textbf{S})}\nonumber\\
&-&\sum_k{t_k\sqrt{10}T^1(\textbf{I}_k)\cdot T^1(\textbf{S},\textbf{C}^2(\omega))}\nonumber\\
&-& eT^2(\nabla\textbf{E}_2)\cdot T^2(\textbf{Q}_2)\nonumber\\
&+&\sum_k c_I(k) T^1(\textbf{N})\cdot T^1(\textbf{I}_k),
\label{hfs hamiltonian}\end{aligned}$$
where the sum over $k=1,2$ represents the terms for both nuclei. The first term is the Fermi contact interaction, the second term is the dipolar interaction, the third one is the quadrupole term (this term exists only for the $^{14}N$), and the last term accounts for the nuclei spin-rotation interaction. In the calculation of matrix elements we first coupled $\textbf{J}=\textbf{S}+\textbf{N}$, $\textbf{F}_1=\textbf{I}_H+\textbf{J}$ and only then $\textbf{F}=\textbf{I}_N+\textbf{F}_1$. All the matrix elements are diagonal in $F$, and the first three terms are given in [@Carington].
The nuclear spin-rotation interactions are given by:
$$\begin{aligned}
&&\langle \eta,\Lambda,N,S,J,I_1,F_1,I_2,F,M_F|T^1(\textbf{N})\cdot T^1(\textbf{I}_1) \nonumber\\
&&|\eta,\Lambda,N',S,J',I_1,F_1',I_2,F,M_F\rangle\nonumber\\
&=& (-1)^{J'+F_1+I_1}\delta_{F_1,F_1'}\left\{ \begin{array}{ccc} I_1 & J' & F_1 \\ J & I_1 & 1 \end{array} \right\}\nonumber\\
&\times&\sqrt{I_1(I_1+1)(2I_1+1)}\delta_{N,N'}(-1)^{J'+N+1+S}\nonumber\\
&\times&\sqrt{(2J+1)(2J'+1)}\left\{ \begin{array}{ccc} N' & J' & I_1 \\ J & N & 1 \end{array} \right\}\nonumber\\
&\times&\sqrt{N(N+1)(2N+1)},\label{spin-rotation1}\end{aligned}$$
where $I_1\equiv I_H$, and
$$\begin{aligned}
&&\langle \eta,\Lambda,N,S,J,I_1,F_1,I_2,F,M_F|T^1(\textbf{N})\cdot T^1(\textbf{I}_2) \nonumber\\
&&|\eta,\Lambda,N',S,J',I_1,F_1',I_2,F,M_F\rangle\nonumber\\
&=& (-1)^{F_1'+F+I_2}\left\{ \begin{array}{ccc} I_2 & F_1' & F \\ F_1 & I_2 & 1 \end{array} \right\}\nonumber\\
&\times&\sqrt{I_2(I_2+1)(2I_2+1)}(-1)^{F_1'+J+1+I_1}\nonumber\\
&\times&\sqrt{(2F_1+1)(2F_1'+1)}\left\{ \begin{array}{ccc} J' & F_1' & I_1 \\ F_1 & J & 1 \end{array} \right\}\nonumber\\
&\times&(-1)^{J'+N+1+S}\sqrt{(2J+1)(2J'+1)}\left\{ \begin{array}{ccc} N' & J' & S \\ J & N & 1 \end{array} \right\}\nonumber\\
&\times&\delta_{N,N'}\sqrt{N(N+1)(2N+1)},\label{spin-rotation2}\end{aligned}$$
where $I_2\equiv I_N$. The constants were taken from [@Jesus]. In the Zeeman Hamiltonian we consider only the contribution due to electronic spin: $$H_{Z}=\mu_B g_s T^1(\textbf{B})\cdot T^1(\textbf{S}), \label{ZeemanHfs}$$ and we neglect other small contributions coming from the nuclei’s rotation and spin, and electronic anisotropy. The Zeeman matrix element is proportional to: $$\begin{aligned}
&&\langle \eta,\Delta,N,S,J,I_1,F_1,I_2,F,M_F|T^1_0(\textbf{S})\nonumber\\
&&|\eta,\Delta,N,S,J',I_1,F_1',I_2,F',M_F\rangle\nonumber\\
&=& (-1)^{F-M_F} \left( \begin{array}{ccc} F & 1 & F' \\ -M_F & 0 & M_F \end{array} \right)\nonumber\\
&\times& (-1)^{F'+F_1+1+I_2}\sqrt{(2F+1)(2F'+1)}\left\{ \begin{array}{ccc} F_1' & F' & I_2 \\ F & F_1 & 1 \end{array} \right\}\nonumber\\
&\times& (-1)^{F_1'+J+1+I_1}\sqrt{(2F_1+1)(2F_1'+1)}\left\{ \begin{array}{ccc} J' & F_1' & I_1 \\ F_1 & J & 1 \end{array} \right\}\nonumber\\
&\times&(-1)^{J'+S+1+N}\sqrt{(2J'+1)(2J+1)}\left\{ \begin{array}{ccc} S & J' & N \\ J & S & 1 \end{array} \right\}\nonumber\\
&\times&\sqrt{S(S+1)(2S+1)},\label{hfs Zeeman}\end{aligned}$$ where it is no more diagonal in F. Finally, for the alignment calculations the following matrix element is useful:
$$\begin{aligned}
&&\langle \eta,\Delta,N,S,J,I_1,F_1,I_2,F,M_F|D^{2 *}_{00} \nonumber\\
&&|\eta,\Delta,N',S,J',I_1,F_1',I_2,F',M_F\rangle\nonumber\\
&=& (-1)^{F-M_F} \left( \begin{array}{ccc} F & 2 & F' \\ -M_F & 0 & M_F \end{array} \right)\nonumber\\
&\times& (-1)^{F'+F_1+2+I_2}\sqrt{(2F+1)(2F'+1)}\left\{ \begin{array}{ccc} F_1' & F' & I_2 \\ F & F_1 & 2 \end{array} \right\}\nonumber\\
&\times& (-1)^{F_1'+J+2+I_1}\sqrt{(2F_1+1)(2F_1'+1)}\left\{ \begin{array}{ccc} J' & F_1' & I_1 \\ F_1 & J & 2 \end{array} \right\}\nonumber\\
&\times&(-1)^{J'+N+2+S}\sqrt{(2J+1)(2J'+1)}\left\{ \begin{array}{ccc} N' & J' & S \\ J & N & 2 \end{array} \right\}\nonumber\\
&\times&(-1)^N \left( \begin{array}{ccc} N & 2 & N' \\ 0 & 0 & 0 \end{array} \right) \sqrt{(2N+1)(2N'+1)}.\label{Prealignment HFS}\end{aligned}$$
W. Gerlach and O. Stern, Z. Phys. **9**, 353 (1922); Ann. Phys. **74**, 673 (1924). C. H. Townes and A. L. Schawlow, Microwave Spectroscopy, 2nd ed. (Dover Publications, Inc., New York, 1975). T. J. McCarthy, M. T. Timko and D. R. Herschbach, J. Chem. Phys. **125**, 133501 (2006). E. Benichou, A. R. Allouche, R. Antoine, M. Aubert-Frecon, M. Bourgoin, M. Broyer, Ph. Dugourd, G. Hadinger and D. Rayane, Eur. Phys. J. D. **10**, 233 (2000). H. J. Loesch, Chem. Phys. **207**, 427 (1996). R. Antoine, D. Rayane, A. R. Allouche, M. Aubert-Frecon, E. Benichou, F. W. Dalby, Ph. Dugourd, M. Broyer and C. Guet, J. Chem. Phys. **110**, 5568 (1999). L. Holmegaard, J. H. Nielsen, I. Nevo, H. Stapelfeldt, F. Filsinger, J. Küpper, and G. Meijer, Phys. Rev. Lett. $\textbf{102}$, 023001 (2009); F. Filsinger, J. Küpper, G. Meijer, L. Holmegaard, J. H. Nielsen, I. Nevo, J. L. Hansen and H. Stapelfeldt, J. Chem. Phys. $\textbf{131}$, 064309 (2009). N. A. Kuebler, M. B. Robin, J. J. Yang, A. Gedanken and D. R. Herrick, Phys. Rev. A **38**, 737 (1988). E. Narevicius, C. G. Parthey, A. Libson, M. F. Riedel, U. Even and M. G. Raizen, New J. Phys., **9**, 96 (2007). H. Stapelfeldt, H. Sakai, E. Constant and P. B. Corkum, Phys. Rev. Lett. **79**, 2787 (1997); H. Sakai, A. Tarasevitch, J. Danilov, H. Stapelfeldt, R. W. Yip, C. Ellert, E. Constant and P. B. Corkum, Phys. Rev. A, **57**, 2794 (1998). B. S. Zhao, H. S. Chung, K. Cho, S. H. Lee, S. Hwang, J. Yu, Y. H. Ahn, J. Y. Sohn, D. S. Kim, W. K. Kang and D. S. Chung, Phys. Rev. Lett. **85**, 2705 (2000); H. S. Chung, B. S. Zhao, S. H. Lee, S. Hwang, K. Cho, S. H. Shim, S. M. Lim, W. K. Kang and D. S. Chung, J. Chem. Phys. **114**, 8293 (2001). B. S. Zhao, S. H. Lee, H. S. Chung, S. Hwang, W. K. Kang, B. Friedrich and D. S. Chung, J. Chem. Phys. **119**, 8905 (2003). S. M. Purcell and P.F. Barker, Phys. Rev. Lett. **103**, 153001 (2009); Phys. Rev. A **82**, 033433 (2010). B. A. Zon and B. G. Katsnelson, Zh. Eksp. Teor. Fiz. **69**, 1166 (1975) \[Sov. Phys. JETP **42**, 595 (1975)\]. B. Friedrich and D. Herschbach, Phys. Rev. Lett. **74**, 4623 (1995); J. Chem. Phys. **111**, 6157 (1999). E. Gershnabel and I. Sh. Averbukh, Phys. Rev. Lett. **104**, 153001 (2010); Phys. Rev. A **82**, 033401 (2010). E. Gershnabel and I. Sh. Averbukh, J. Chem. Phys. **134**, 054304 (2011). J. Floß, E. Gershnabel and I. Sh. Averbukh, Phys. Rev. A **83**, 025401 (2011). V. Aqullanti, D. Ascenzi, D. Cappelletti, and F. Pirani, Nature **371**, 399 (1994); V. Aqullanti, D. Ascenzi, D. Cappelletti, S. Franceschini and F. Pirani, Phys. Rev. Lett. **74**, 2929 (1995). H. Stapelfeldt and T. Seideman, Rev. Mod. Phys. **75**, 543 (2003). V. Kumarappan, S. S. Viftrup, L. Holmegaard, C. Z. Bisgaard and H. Stapelfeldt, Phys. Scr. **76**, C63 (2007). J. G. Underwood, B. J. Sussman, and A. Stolow, Phys. Rev. Lett. **94**, 143002 (2005); K. F. Lee, D. M. Villeneuve, P. B. Corkum, A. Stolow, and J. G. Underwood, Phys. Rev. Lett. **97**, 173001 (2006). D. Daems, S. Guérin, E. Hertz, H. R. Jauslin, B. Lavorel, and O. Faucher, Phys. Rev. Lett. **95**, 063005 (2005); E. Hertz, A. Rouzée, S. Guérin, B. Lavorel, and O. Faucher, Phys. Rev. A **75**, 031403(R) (2007); Guérin, A. Rouzée, and E. Hertz, Phys. Rev. A **77**, 041404(R) (2008). J. Brown and A. Carington, Rotational Spectroscopy of Diatomic Molecules, 1st ed. (Cambridge University Press, New York, 2003). E. Gershnabel, I. Sh. Averbukh and R. J. Gordon, Phys. Rev. A **74**, 053414 (2006). A. Carrington, P. N. Dyer and D. H. Levy, J. Chem. Phys. **47**, 1756 (1967). R. K. Kakar, E. A. Cohen and M. Geller, J. Mol. Spect. **70**, 243 (1978). B. J. Drouin, C. E. Miller, E. A. Cohen, G. Wagner and M. Birk, J. Mol. Spect. **207**, 4 (2001). M. Tinkham and M. W. P. Strandberg, Phys. Rev. **97**, 937 (1955); Phys. Rev. **97**, 951 (1955); M. Tinkham, Ph.D. Thesis, (Massachusetts Institute of Technology, 1954). F. D. Wayne and H. E. Radford, Mol. Phys. **32**, 1407 (1976). T. Klaus, S. Takano and G. Winnewisser, Astron. Astrophys. **322**, L1 (1997). J. Flores-Mijangos J. M. Brown, F. Matsushima, H. Odashima, K. Takagi, L. R. Zink and K. M. Evenson, J. Mol. Spect. **225**, 189 (2004). F. Lewen, S. Brünken, G. Winnewisser, M. Šimečková, and Š. Urban, J. Mol. Spect. **226**, 113 (2004). S. Fleischer, I. Sh. Averbukh and Y. Prior, Phys. Rev. A **74**, 041403(R) (2006). M. Renard, E. Hertz, B. Lavorel, and O. Faucher, Phys. Rev. A **69**, 043401 (2004). S. Fleischer, I. Sh. Averbukh, and Y. Prior, Phys. Rev. Lett. **99**, 093002 (2007); E. Gershnabel and I. Sh. Averbukh, Phys. Rev. A **78**, 063416 (2008). J. Karczmarek, J. Wright, P. Corkum and M. Ivanov, Phys. Rev. Lett. **82**, 3420 (1999); D. M. Villeneuve, S. A. Aseyev, P. Dietrich, M. Spanner, M. Yu. Ivanov, and P. B. Corkum, Phys. Rev. Lett. **85**, 542 (2000). Liwei Yuan, S. W. Teitelbaum, A. Robinson, and A. S. Mullin, PNAS **108**, 6872 (2011). S. Fleischer, Y.Khodorkovsky, Y. Prior, and I. Sh. Averbukh, New J. Phys. **11**, 105039 (2009). M. Lapert, E. Hertz, S. Guérin, and D. Sugny, Phys. Rev. A **80**, 051403(R) (2009). Md. Z. Hoque, M. Lapert, E. Hertz, F. Billard, D. Sugny, B. Lavorel, and O. Faucher, Phys. Rev. A **84**, 013409 (2011).
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'J. van Bree'
- 'M. E. Flatt[é]{}'
title: 'Supplemental Material: Atomic-scale magnetometry of dynamic magnetization'
---
Spin density of $N$-particle state
==================================
In the Letter we asserted that $\langle {\bf J}({\bf x}) \rangle = \langle {\bf J} \rangle {\cal P}({\bf x})$. Here we derive this relation for any $N$-particle wave function $\Psi(1,\dots,N)$. In absence of spin-orbit coupling, this wave function is the product of an orbital part $\psi$ and spin part $\chi$, $$\begin{aligned}
\Psi(1,\dots,N) = \psi({\bf x}_1,\dots,{\bf x}_N) \chi(\sigma_1,\dots,\sigma_N),\end{aligned}$$ where ${\bf x}_i$ and $\sigma_i$ are the spatial and spin coordinates of particle $i$. Computing the expectation value of the $N$-particle spin density operator [@Schwabl2004] $$\begin{aligned}
{\bf J}({\bf x}) &= \sum_{i=1}^N \delta({\bf x}-{\bf x}_{i}) {\bf S}_{i},\end{aligned}$$ we obtain the spin density $$\begin{aligned}
\langle {\bf J}({\bf x}) \rangle &= \sum_{i=1}^N \chi(\sigma_1,\dots,\sigma_N)^* {\bf S}_i \chi(\sigma_1,\dots,\sigma_N) \int \psi({\bf x}_1,\dots,{\bf x}_i,\dots,{\bf x}_N)^* \delta({\bf x}-{\bf x}_i) \psi({\bf x}_1,\dots,{\bf x}_i,\dots,{\bf x}_N)~d^3x_1~\dots~d^3x_N \nonumber \\
&= \sum_{i=1}^N \chi(\sigma_1,\dots,\sigma_N)^* {\bf S}_i \chi(\sigma_1,\dots,\sigma_N) \int \psi({\bf x}_i,\dots,{\bf x}_1,\dots,{\bf x}_N)^* \delta({\bf x}-{\bf x}_1) \psi({\bf x}_i,\dots,{\bf x}_1,\dots,{\bf x}_N)~d^3x_1~\dots~d^3x_N \nonumber \\
&= \sum_{i=1}^N \chi(\sigma_1,\dots,\sigma_N)^* {\bf S}_i \chi(\sigma_1,\dots,\sigma_N) \int \psi({\bf x}_1,\dots,{\bf x}_i,\dots,{\bf x}_N)^* \delta({\bf x}-{\bf x}_1) \psi({\bf x}_1,\dots,{\bf x}_i,\dots,{\bf x}_N)~d^3x_1~\dots~d^3x_N \nonumber \\
&= \chi(\sigma_1,\dots,\sigma_N)^* {\bf J} \chi(\sigma_1,\dots,\sigma_N) \int |\psi({\bf x}_1,\dots,{\bf x}_N)|^2 \delta({\bf x}-{\bf x}_1)~d^3x_1~\dots~d^3x_N \nonumber \\
&\equiv \langle {\bf J} \rangle {\cal P}({\bf x}).\end{aligned}$$ In the first step we swapped the integration variables $i$ and $1$. In the second step we used the coordinate permutation properties of $\psi$: since $\psi$ appears twice in the integral, any interchanging of coordinates $1$ and $i$ does not affect the integral. We then recognize that the integral depends no longer on $i$, so the integral can be taken outside of the summation. The spin density depends therefore solely on the total spin operator ${\bf J} = \sum_{i}^N {\bf S}_i$, and is indeed proportional to the spatial probability density ${\cal P}({\bf x})$. We do point out that ${\cal P}({\bf x})$ is generally not a simple summation of the single particle state probabilities: the anti-symmetrisation of $\Psi$ implies that $\psi$ is a linear combination of [*products*]{} of single particle wave functions, which means ${\cal P}({\bf x})$ depends on the orthogonality of those single particle wave functions.
Magnetic energy Hamiltonian - spherical symmetry
================================================
In the Letter we suggested that the $\eta$-dependence of $E_{\text{mag}}$ is effectively caused by a $J_z^2$ term in spin’s ground state Hamiltonian. Here we derive this Hamiltonian term, taking the quantum-mechanical nature of the spin operator into account. As stated in the Letter, a spin’s magnetic moment density ${\boldsymbol \mu}({\bf x})$ is generated by a current density ${\bf j}({\bf x}) = \nabla \times {\boldsymbol \mu}({\bf x})$. Whereas in the Letter we used the expectation value of the spin-operator, i.e. assumed the spin is classical, here we recognize that ${\bf j}({\bf x})$ is an operator in the spin-subspace, $$\begin{aligned}
{{\bf j}}({\bf x}) = \frac{2\mu_B}{\hbar} \nabla \times {\cal P}({\bf x}) {{\bf J}} = \frac{2\mu_B}{\hbar} \nabla {\cal P}({\bf x}) \times {{\bf J}},\end{aligned}$$ in absence of spin-orbit interaction. Consider now a slightly more general setup than in the Letter: the spin is localized in region I, adjacent to region II with a different magnetic permeability filling half-space ${\bf x}\cdot{\bf n}\geq 1$. The vector potential in region I is then the sum of ${\bf A}({\bf x})$, due to ${\bf j}({\bf x})$, and $\tilde{{\bf A}}({\bf x})$, due to the image current distribution $\tilde{{\bf j}}({\bf x})$ $$\begin{aligned}
\tilde{{\bf j}}({\bf x}) &= \Delta \mu_r \left[{\bf j}(\tilde{\bf x}) - 2 \frac{{\bf j}(\tilde{\bf x}) \cdot {\bf n}}{|{\bf n}|} \frac{\bf n}{|{\bf n}|} \right],\end{aligned}$$ where $\Delta \mu_r = \frac{\mu_r^{\text{II}}-\mu_r^\text{I}}{\mu_r^{\text{II}}+\mu_r^\text{I}}$ is the contrast in magnetic permeability between regions I and II, and $\tilde{\bf x}$ the current image’s coordinates $$\begin{aligned}
\tilde{\bf x} = {\bf x} + 2 {\bf n} - 2\frac{{\bf x}\cdot{\bf n}}{|{\bf n}|} \frac{\bf n}{|{\bf n}|}.\end{aligned}$$ Using the relation $$\begin{aligned}
\nabla_{\tilde{\bf x}} f(\tilde{\bf x}) = \nabla_{\bf x} f(\tilde{\bf x}) - 2 \frac{\nabla_{\bf x} f(\tilde{\bf x}) \cdot {\bf n}}{|{\bf n}|} \frac{\bf n}{|\bf n|},\end{aligned}$$ the image current can be expressed as $$\begin{aligned}
\tilde{\bf j}({\bf x}) &= \frac{2\mu_B}{\hbar} \Delta \mu_r \nabla {\cal P}(\tilde{\bf x}) \times \left(\left[2\frac{\bf n}{|{\bf n}|} \cdot {\bf J}\right] \frac{\bf n}{|{\bf n}|} - {\bf J} \right)\end{aligned}$$ using vector calculus; we take $\nabla$ to act on ${\bf x}$ unless indicated differently. Expressing the vector potential in terms of the current density $$\begin{aligned}
{\bf A}({\bf x}) = \frac{\mu_0}{4 \pi} \int_V \frac{{\bf j}({\bf x'})}{|{\bf x}-{\bf x}'|}~d^3x',\end{aligned}$$ we can write, in the Coulomb gauge, the EM-field Hamiltonian ${\cal H}_{\text{mag}}$ in the spin subspace as [@Bjorken1965; @Jackson1998] $$\begin{aligned}
{\cal H}_{\text{mag}} &= \frac{1}{2} \int_V \left[{\bf A}({\bf x})+\tilde{{\bf A}}({\bf x})\right] \cdot {\bf j}({\bf x})~d^3x \\
&= \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{2\pi \hbar^2} \int_{V} \left[\nabla \zeta({\bf x},{\bf 0}) \times {\bf J}\right] \cdot \left[\nabla {\cal P}({\bf x}) \times {\bf J}\right]~d^3x \nonumber \\
&\hspace{0.5cm} + \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{2\pi \hbar^2} \Delta \mu_r \int_{V} \left[\nabla \zeta({\bf x},{\bf n}) \times \left(\left[2\frac{\bf n}{|{\bf n}|} \cdot {\bf J}\right] \frac{\bf n}{|{\bf n}|} - {\bf J} \right)\right] \cdot \left[\nabla {\cal P}(\bf x) \times {\bf J}\right]~d^3x. \label{eq:H}\end{aligned}$$ Here we have defined $\zeta({\bf x},{\bf n})$ through the relation $$\begin{aligned}
\int_{\tilde V} \frac{\nabla' {\cal P}(\tilde{\bf x}')}{|{\bf x}-{\bf x}'|}~d^3x' &= \left. \frac{{\cal P}(\tilde{\bf x}')}{|{\bf x}-{\bf x}'|} \right|_{{\bf x}' \in \tilde{S}} - \int_{\tilde{V}} {\cal P}(\tilde{\bf x}')\nabla' \left(\frac{1}{|{\bf x}-{\bf x}'|}\right)~d^3x' \\
&= \nabla \left(\int_{\tilde{V}} \frac{{\cal P}(\tilde{\bf x}')}{|{\bf x}-{\bf x}'|}~d^3x'\right) \\
&= \nabla \left(\int_{V} \frac{{\cal P}({\bf x}')}{\left|{\bf x}-\left({\bf x}'+2{\bf n}-2\frac{{\bf x}'\cdot{\bf n}}{|{\bf n}|} \frac{\bf n}{|{\bf n}|}\right)\right|}~d^3x'\right) \\
&\equiv \nabla \zeta({\bf x},{\bf n}),\end{aligned}$$ and assumed the spin to be localized in region I, so that integration volume $V$ ($\tilde{V}$) is limited to region I (II) and ${\cal P}({\bf x})=0$ on their surfaces. By defining ${\bf x}_{\bf n}'={\bf x}'+2{\bf n}-2\frac{{\bf x}'\cdot{\bf n}}{|{\bf n}|} \frac{\bf n}{|{\bf n}|}$, the nominator in $\zeta({\bf x},{\bf n})$ can be expanded in spherical harmonics $Y_l^m(\theta,\phi)$, $$\begin{aligned}
\zeta({\bf x},{\bf n}) &= \sum_{l=0}^{\infty} \sum_{m=-l}^{l} \frac{4\pi}{2l+1} Y_l^m(\theta,\phi) \int_{V} {\cal P}({\bf x}') \frac{r_<^l}{r_>^{l+1}} Y_l^{m}({\theta}_{\bf n}',{\phi}_{\bf n}')^*~d^3x' \equiv \sum_{l=0}^{\infty} \sum_{m=-l}^{l} \frac{4\pi}{2l+1} Y_l^m(\theta,\phi) {\cal R}_l^m(r,{\bf n}),\end{aligned}$$ where $r_<$ ($r_>$) is the smaller (larger) of $r$ and $r_{\bf n}'$. We now revert to the setup of the Letter, i.e. ${\bf n}=d~{\bf e}_z$, so that $$\begin{aligned}
r_{\bf n}' &= \sqrt{4d^2+r'^2-4dr'\cos\theta'} \\
\cos {\theta}_{\bf n}' &= \frac{2d - r' \cos \theta'}{\sqrt{4d^2+r'^2-4dr'\cos\theta'}} \\
{\phi}_{\bf n}' &= \phi',\end{aligned}$$ and assume spherical symmetry, ${\cal P}({\bf x})={\cal P}(r)$. It is straightforward to see that ${\cal R}_l^m(r,{\bf n})$ is only non-zero for $m=0$, and ${\cal R}_l^m(r,{\bf 0})$ is only non-zero for $l=0$. Therefore $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{sph}} &= \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{\sqrt{\pi} \hbar^2} \int_{V} \left[\nabla {\cal R}_0^0(r,{\bf 0}) \times {\bf J}\right] \cdot \left[\nabla {\cal P}(r) \times {\bf J}\right]~d^3x \nonumber \\
&\hspace{0.5cm} + \sum_{l=0}^{\infty} \frac{2 \mu_0 \mu_r^{\text{I}} \mu_B^2}{(2l+1)\hbar^2} \Delta \mu_r \int_{V} \left[\nabla \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf n})\right\} \times \left(2 J_z {\bf e}_z - {\bf J} \right)\right] \cdot \left[\nabla {\cal P}(r) \times {\bf J}\right]~d^3x.\end{aligned}$$ Integrating over $\phi$ and $\theta$ results in $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{sph}} &= \frac{8 \sqrt{\pi}\mu_0 \mu_r^{\text{I}} \mu_B^2}{3 \hbar^2} \int \frac{d{\cal R}_0^0(r,{\bf 0})}{dr} \frac{d{\cal P}(r)}{dr} r^2~dr \left(J_x^2+J_y^2+J_z^2\right) \nonumber \\
&\hspace{0.5cm} - \frac{4 \sqrt{\pi}\mu_0 \mu_r^{\text{I}} \mu_B^2}{15 \sqrt{5} \hbar^2} \Delta \mu_r \int \left(\frac{3}{r} {\cal R}_2^0(r,{\bf n}) + \frac{d{\cal R}_2^0(r,{\bf n})}{dr} + 10 \sqrt{5} \frac{d{\cal R}_0^0(r,{\bf n})}{dr}\right) \frac{d{\cal P}(r)}{dr} r^2~dr \left(J_x^2+J_y^2+J_z^2\right) \nonumber \\
&\hspace{0.5cm} - \frac{4 \sqrt{\pi}\mu_0 \mu_r^{\text{I}} \mu_B^2}{15 \sqrt{5} \hbar^2} \Delta \mu_r \int \left(\frac{3}{r} {\cal R}_2^0(r,{\bf n}) + \frac{d{\cal R}_2^0(r,{\bf n})}{dr} - 20 \sqrt{5} \frac{d{\cal R}_0^0(r,{\bf n})}{dr}\right) \frac{d{\cal P}(r)}{dr} r^2~dr~J_z^2. \label{eq:sph}\end{aligned}$$ Only spherical harmonics with $l=0$ and $l=2$ appear in the part of the Hamiltonian related to region II. As we will show explicitly later on, the $l=0$ term vanishes after radial integration. The $l=2$ contribution does not vanish and its origin can be explained as follows. The Hamiltonian describes the interaction between the spin’s magnetic moment density and the magnetic field it induces in region II. Expanding these in terms of spherical harmonics, they have respectively $l=0$ and $l=2$ in case of spherical symmetry. Their product enters the Hamiltonian, which is proportional to a single spherical harmonic of order $l=0+2$. Therefore only of the $l=2$ term in Eq. \[eq:sph\] will lead to a contribution to the Hamiltonian. Indeed, if we work out the integrals over $r$, we observe that $$\begin{aligned}
{\cal R}_0^0(r,{\bf 0}) &= 2\sqrt{\pi} \left[\frac{1}{r} \int_0^r {\cal P}(r') r'^2~dr' + \int_r^R {\cal P}(r') r'~dr'\right],\end{aligned}$$ so that $$\begin{aligned}
\int \frac{d{\cal R}^0_0(r,{\bf 0})}{dr} \frac{d {\cal P}(r)}{dr} r^2~dr &= 2\sqrt{\pi} \int_0^R {\cal P}(r)^2 r^2~dr,\end{aligned}$$ where, in line with the spin being localized in region I, ${\cal P}(R)=0$ has been assumed. We also find $$\begin{aligned}
{\cal R}_l^0(r,{\bf n}) &= \sqrt{(2l+1)\pi} r^l \int {\cal P}(r') r'^2 \left(\int \frac{P_l^0(\cos \theta_{\bf n}')}{r_{\bf n}'^{l+1}}\sin\theta'~d\theta'\right)dr' = \sqrt{\frac{(2l+1)}{16\pi}} \frac{r^l}{2^l d^{l+1}},\end{aligned}$$ where we used $\int_0^R {\cal P}(r) r^2~dr = \tfrac{1}{4\pi}$. Therefore $$\begin{aligned}
\int \frac{d{\cal R}_l^0(r,{\bf n})}{dr} \frac{d {\cal P}(r)}{dr} r^2~dr &= -\sqrt{\frac{(2l+1)}{16\pi}} \frac{l(l+1)}{2^l d^{l+1}} \int_0^R {\cal P}(r) r^l~dr \\
\int {\cal R}_l^0(r,{\bf n}) \frac{d {\cal P}(r)}{dr} r~dr &= -\sqrt{\frac{(2l+1)}{16\pi}} \frac{l+1}{2^l d^{l+1}} \int_0^R {\cal P}(r) r^l~dr,\end{aligned}$$ where we explicitly see that the $l=0$ contribution vanishes after radial integration. Using these relations, the Hamiltonian becomes $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{sph}} &= \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{\hbar^2} \left[\frac{16 \pi}{3} \int_0^R {\cal P}(r)^2 r^2~dr + \frac{\Delta \mu_r}{16 \pi d^3} \right] \left(J_x^2+J_y^2+J_z^2\right) + \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{16 \pi \hbar^2 d^3} \Delta \mu_r J_z^2.\end{aligned}$$ Indeed we find that the spin’s magnetic energy leads to additional terms in the Hamiltonian. In particular, the introduction of region II leads to an effective Hamiltonian ${\cal H}_{\text{eff}} = D_{\text{mag}}^{\text{sph}} J_z^2$, where $$\begin{aligned}
D_{\text{mag}}^{\text{sph}} = \left(\frac{\mu_r^{\text{II}}-\mu_r^\text{I}}{\mu_r^{\text{II}}+\mu_r^\text{I}}\right) \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{16 \pi \hbar^2 d^3}.\end{aligned}$$ As their pre-factors differ, it might seem ${\cal H}_{\text{mag}}^{\text{sph}}$ and $E_{\text{mag}}$ are incompatible. However, this difference arises from the classical treatment of the spin for $E_{\text{mag}}$ versus the quantum-mechanical treatment for ${\cal H}_{\text{mag}}$. To make a direct comparison, we need to calculate the magnetic energy resulting from ${\cal H}_{\text{mag}}^{\text{sph}}$ for a spin making an angle $\eta$ with respect to the $z$-axis. It is more convenient to use a basis rotated by the angle $\eta$, so that the spin is in the state $|J,J\rangle$ and ${\cal H}_{\text{mag}}^{\text{sph}}$ becomes $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{sph}} &= \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{\hbar^2} \left[\frac{16 \pi}{3} \int_0^R {\cal P}(r)^2 r^2~dr + \frac{\Delta \mu_r}{16 \pi d^3} \right] \left(J_x^2+J_y^2+J_z^2\right) + \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2}{16 \pi \hbar^2 d^3} \Delta \mu_r \left(J_x \sin \eta + J_z \cos \eta\right)^2.\end{aligned}$$ The magnetic energy of this state is then $$\begin{aligned}
\langle J,J | {\cal H}_{\text{mag}}^{\text{sph}} | J,J \rangle = \frac{16}{3} \mu_0 \mu_r^{\text{I}} \mu_B^2 J^2 \pi \int_0^R {\cal P}(r)^2 r^2~dr + \frac{\mu_0 \mu_r^{\text{I}} \mu_B^2 J^2}{32 \pi d^3} \Delta \mu_r \left(3 + \cos 2\eta + \frac{1}{2J} \left[1 - \cos 2\eta\right]\right).\end{aligned}$$ Hence we find $\lim_{J \rightarrow \infty} \langle J,J | {\cal H}_{\text{mag}}^{\text{sph}} | J,J \rangle = E_{\text{mag}}$, as one would expect in the semiclassical (large-J) limit of a quantum system.
Magnetic energy Hamiltonian - cylindrical symmetry
==================================================
In the Letter we calculated the NV$^-$ center’s magnetic energy assuming its probability density ${\cal P}({\bf x})$ has spherical symmetry, instead of its actual $C_{3v}$-symmetry [@Doherty2013]. Here we will validate this approximation. We start by calculating ${\cal H}_{\text{mag}}$ assuming cylindrical $D_{\infty h}$-symmetry, ${\cal P}({\bf x})={\cal P}(\rho,z)$. We take the axial symmetry axis of ${\cal P}(\rho,z)$ perpendicular to the interface between the two regions. It has been demonstrated that such orientation can be realized deterministically in practice [@Michl2014]. Following the same setup as described in the Letter, i.e. ${\bf n}=d~{\bf e}_z$, ${\cal H}_{\text{mag}}$ (Eq. \[eq:H\]) is now $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{cyl}} &= \sum_{l=0}^{\infty} \frac{2 \mu_0 \mu_r^{\text{I}} \mu_B^2}{(2l+1)\hbar^2} \int_{V} \left[\nabla \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf 0})\right\} \times {\bf J}\right] \cdot \left[\nabla {\cal P}(\rho,z) \times {\bf J}\right]~d^3x \nonumber \\
&\hspace{0.5cm} + \sum_{l=0}^{\infty} \frac{2 \mu_0 \mu_r^{\text{I}} \mu_B^2}{(2l+1)\hbar^2} \Delta \mu_r \int_{V} \left[\nabla \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf n})\right\} \times \left(2 J_z {\bf e}_z - {\bf J} \right)\right] \cdot \left[\nabla {\cal P}(\rho,z) \times {\bf J}\right]~d^3x,\end{aligned}$$ since ${\cal R}_l^m(r,{\bf n})$ is only non-zero for $m=0$. The relation $$\begin{aligned}
\iint \frac{\partial \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf n}) \right\}}{\partial \rho} \frac{\partial {\cal P}(\rho,z)}{\partial \rho} \rho~d\rho dz = -\iint \frac{\partial \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf n}) \right\}}{\partial z} \frac{\partial {\cal P}(\rho,z)}{\partial z} \rho~d\rho dz \equiv M_l({\bf n}),\end{aligned}$$ can be derived using partial integration and the property $\nabla^2 \left\{Y_l^0(\theta,\phi) {\cal R}_l^0(r,{\bf n}) \right\} = 0$ (valid by construction). Using this relation and integrating over $\phi$, we obtain $$\begin{aligned}
{\cal H}_{\text{mag}}^{\text{cyl}} &= \frac{2 \pi \mu_0 \mu_r^{\text{I}} \mu_B^2}{\hbar^2} \sum_{l=0}^{\infty} \frac{\Delta \mu_r M_l({\bf n}) - M_l({\bf 0})}{(2l+1)} \left(J_x^2+J_y^2+J_z^2\right) + \frac{2 \pi \mu_0 \mu_r^{\text{I}} \mu_B^2}{\hbar^2} \sum_{l=0}^{\infty} \frac{\Delta \mu_r M_l({\bf n}) + 3 M_l({\bf 0})}{(2l+1)} J_z^2.\end{aligned}$$ Region II leads again to a $J_z^2$-term in the Hamiltonian: lowering the symmetry of ${\cal P}({\bf x})$ to $D_{\infty h}$ does not alter the structure of the Hamiltonian. The $J_z^2$-term is also present when region II is absent, which is caused by the $D_{\infty h}$-symmetry ${\cal P}({\bf x})$. This term is irrelevant, however, as it cannot be distinguished from other mechanisms contributing to the fine structure constant $D_{\text{GS}}$. (If this were possible, it would provide a measure of the aspect ratio of ${\cal P}({\bf x})$.) To quantitatively compare ${\cal H}_{\text{mag}}^{\text{cyl}}$ with ${\cal H}_{\text{mag}}^{\text{sph}}$, we take the spin confined to a cylinder with radius $R$ and height $H$, $$\begin{aligned}
{\cal P}(\rho,z) = N \left[J_0\left(\frac{\rho_{0,1}}{R} \rho\right) \cos \left(\frac{\pi}{H}z\right)\right]^2,\end{aligned}$$ where $N$ is a normalisation constant, and $\rho_{0,1}$ the first zero of the $0^{\text{th}}$-order Bessel function $J_0(x)$. Concentrating on the $J_z^2$ term in the Hamiltonian caused by the presence of region II, we find it can be expressed as $$\begin{aligned}
\frac{\mu_0 \mu_r^{\text{I}}\mu_B^2}{16 \pi \hbar^2 H R^2} \Delta \mu_r \sum_{l=2}^{\infty} f_{2l-2}(\delta,\lambda) J_z^2 = D_{\text{mag}}^{\text{cyl}} J_z^2, \label{eq:DCyl}\end{aligned}$$ where $\delta = d/H$ is the normalized distance of the spin to the interface between regions I and II, $\lambda = H/R$ is the aspect ratio. The numbers $f$ depend only on $\delta$ and $\lambda$, are only non-zero for even $l$ due to the even symmetry of ${\cal P}({\bf x})$ in $z$, and start from $l=2$ for the same reasons as for spherical symmetry (see discussion below Eq. \[eq:sph\]). Interestingly, the relative difference $$\begin{aligned}
\frac{D_{\text{mag}}^{\text{cyl}} - D_{\text{mag}}^{\text{sph}}}{D_{\text{mag}}^{\text{cyl}} + D_{\text{mag}}^{\text{sph}}} = \frac{\delta^3 \lambda^2 \sum_{l=2}^{\infty} f_l(\delta,\lambda) - 1}{\delta^3 \lambda^2 \sum_{l=2}^{\infty} f_l(\delta,\lambda) +1},\end{aligned}$$ depends only on $\delta$ and $\lambda$. From Fig. \[fig:DCylvsSph\](a) it is clear that for either $H > R$ or large enough $\delta$, the difference between spherical and cylindrical symmetry is $\leq 5\%$: in both cases the cylindrical ${\cal P}({\bf x})$ appears as approximately spherical from region II’s perspective. The $l=2$ term is almost entirely responsible for this difference, as can be seen from Fig. \[fig:DCylvsSph\](b). Indeed, the numbers $f$ in Eq. \[eq:DCyl\] depend on the distance as $1/\delta^{l+1}$, so that at fixed distance terms with increasing $l$ become less important.
{width="\textwidth"}
Comparing the NV$^-$ center’s $C_{3v}$-symmetry with cylindrical $D_{\infty h}$-symmetry, the probability density is no longer symmetric in $z$ and has three-fold rotation symmetry around the $z$-axis. This reduction in symmetry allows terms proportional to spherical harmonics with $l=2,3,4,...$ and $m=0,\pm3$ to contribute to $D_{\text{mag}}$. We see that the first correction with respect to $D_{\infty h}$-symmetry has to come from the $l=3$ term. It is $1/\delta$ smaller than the $l=2$ term. Based on DFT calculations of the spin density [@Gali2008], we estimate the NV$^-$-center ground state has $H\approx 2.0$ Å and $R\approx 2.5$ Å, hence $\lambda \approx 0.8$. For $d>1$ nm ($\delta>5$), the corrections of the $l=3$ term are $<20\%$ compared to the $l=2$ term. It is therefore reasonable to approximate the NV$^-$ center’s $C_{3v}$-symmetry with cylindrical symmetry. From Fig. \[fig:DCylvsSph\](a) we see that the deviations of $D_{\text{mag}}^{\text{cyl}}$ from $D_{\text{mag}}^{\text{sph}}$ are less than $1\%$. We therefore conclude that taking the NV$^-$ center’s wave function as spherically symmetric is a good approximation.
[6]{}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 @noop [**]{} (, ) @noop [**]{} (, ) @noop [**]{}, ed. (, , ) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We prove that two-step analytic sub-Riemannian structures on a compact analytic manifold equipped with a smooth measure and Lipschitz Carnot groups satisfy measure contraction properties.'
author:
- 'Z. Badreddine[^1]'
- 'L. Rifford[^2]'
title: 'Measure contraction properties for two-step analytic sub-Riemannian structures and Lipschitz Carnot groups'
---
Introduction {#SECintroduction}
============
\[Intro\] The aim of this paper is to provide new examples of sub-Riemannian structures satisfying measure contraction properties. Let $M$ be a smooth manifold of dimension $n\geq 3$ equipped with a sub-Riemannian structure $(\Delta,g)$ of rank $m<n$, whose geodesic distance $d_{SR}$ is supposed to be complete. We refer the reader to Appendix \[Notations\] for the notations used throughout the paper. As in the previous paper of the second author on the same subject [@rifford13], we restrict our attention to the notion of measure contraction properties in metric measured spaces with negligeable cut loci (if $A\subset M$ is a Borel set then $\mathcal{L}^n (A) =0$ means that $A$ has vanishing $n$-dimensional Lebesgue measures in charts):
We say that the sub-Riemannian structure $(\Delta,g)$ on $M$ has negligeable cut loci if for every $x\in M$, there is a measurable set $\mathcal{C}(x) \subset M$ with $$\mathcal{L}^n \left( \mathcal{C}(x)\right) =0,$$ and a measurable map $\gamma_{x} \, : \, \left( M \setminus \mathcal{C}(x)\right) \times [0,1] \longrightarrow M$ such that for every $y \in M \setminus \mathcal{C}(x)$ the curve $$s \in [0,1] \longmapsto \gamma_x (s,y)$$ is the unique minimizing geodesic from $x$ to $y$.
Measure contraction properties consists in comparing the contraction of volumes along minimizing geodesics from a given point with what happens in classical model spaces of Riemannian geometry. We recall that for every $K\in \R$, the comparison function $s_K: [0,+\infty) \rightarrow [0,+\infty)$ ($s_K: [0,\pi/\sqrt{K})\rightarrow [0,+\infty)$ if $K>0$) is defined by $$s_K(t) := \left\{ \begin{array}{lll}
\frac{\sin(\sqrt{K}t)}{\sqrt{K}} & \mbox{ if } K>0\\
t & \mbox{ if } K=0 \\
\frac{\sinh(\sqrt{-K}t)}{\sqrt{-K}} & \mbox{ if } K<0.
\end{array}
\right.$$ In our setting, the following definition is equivalent to the notion of measure contraction property introduced by Ohta in [@ohta07] for more general measured metric spaces (see also [@sturm06b]).
Let $(\Delta,g)$ be a sub-Riemannian structure on $M$ with negligeable cut loci, $\mu$ a measure absolutely continuous with respect to $\mathcal{L}^n$ and $K\in \R, N> 1$ be fixed. We say that $(\Delta,g)$ equipped with $\mu$ satisfies $\MCP(K,N)$ if for every $x\in M$ and every measurable set $A\subset M \setminus \mathcal{C}(x)$ (provided that $A\subset B_{SR}(x,\pi\sqrt{N-1/K})$ if $K>0$) with $0< \mu(A)<\infty$, $$\mu \left(A_s\right) \geq \int_{A} s \left[ \frac{s_K\left(sd_{SR}(x,z)/\sqrt{N-1}\right)}{s_K\left(d_{SR}(x,z)/\sqrt{N-1}\right)} \right]^{N-1} \, d\mu(z) \qquad \forall s \in [0,1],$$ where $A_s$ is the $s$-interpolation of $A$ from $x$ defined by $$A_s := \Bigl\{ \gamma_x(s,y) \, \vert \, y \in A \setminus \mathcal{C}(x) \Bigr\} \qquad \forall s \in [0,1].$$ In particular, $(\Delta,g)$ equipped with $\mu$ satisfies $\MCP(0,N)$ if for every $x\in M$ and every measurable set $A\subset M\setminus \mathcal{C}(x)$ with $0< \mu(A)<\infty$, $$\mu \left(A_s\right) \geq s^N \mu (A) \qquad \forall s \in [0,1].$$
To our knowledge, the first study of measure contraction properties in the sub-Riemannian setting has been performed by Juillet in his thesis. In [@juillet09], Juillet proved that the $n$-th Heisenberg group $\H^n$ (with $n\geq 1$) equipped with its sub-Riemannian distance and the Lebesgue measure $\mathcal{L}^{2n+1}$ (in this case the ambiant space is $\R^{2n+1}$) satisfies $\mbox{MCP}(0,2n+3)$. This result is sharp for two reasons. First, Juillet proved that $\H^n$ does not satisfy any other stronger notion of “Ricci curvature bounded from below” in metric measured spaces such as for example the so-called curvature dimension property (see [@lv07; @sturm06a; @sturm06b; @villanibook]). Secondly, Juillet showed that $2n+3$ is the optimal dimension for which $\H^n$ satisfies $\mbox{MCP}(0,N)$, there is no $N<2n+3$ such that $\H^n$ (equipped with $d_{SR}$ and $\mathcal{L}^{2n+1}$) satisfies $\mbox{MCP}(0,N)$. The Juillet’s Theorem, which settled the case of the simplest sub-Riemannian structures, paved the way to the study of measure contraction properties for more general sub-Riemannian structures. In [@al14], Agrachev and Lee investigated the case of sub-Riemannian structures associated with contact distributions in dimension $3$. In [@lee16; @lcz16], Lee and Lee, Li and Zelenko studied the particular case of Sasakian manifolds. In [@rifford13], the second author proved that any ideal Carnot group satisfy $\mbox{MCP}(0,N)$ for some $N>1$ (it has been shown later by Rizzi [@rizzi16] that a Carnot group is ideal if and only if it is fat). In [@rizzi16], Rizzi showed that any co-rank $1$ Carnot group of dimension $k+1$ (equipped with the sub-Riemannian distance and a left-invariant measure) satisfies $\mbox{MCP}(0,k+3)$. Finally, more recently, Barilari and Rizzi [@br17] proved that $H$-type Carnot groups of rank $k$ and dimension $n$ satisfy $\mbox{MCP}(0,k+3(n-k))$. The purpose of the present paper is to pursue the qualitative approach initiated by the second author in [@rifford13]. We aim to show that some assumptions on the sub-Riemannian structure insure that the sub-Riemannian distance enjoyes some properties which guarantee that some measure contraction property of the form $\mbox{MCP}(0,N)$ is satisfied for some $N>1$ (in fact $N$ has to be greater or equal to the geodesic dimension of the sub-Riemannian structure as introduced by Rizzi [@rizzithesis]). Our approach is purely qualitative, we do not compute any curvature type quantity in order to find the best exponents. Our results are concerned with two-step analytic sub-Riemannian structures and Lipschitz Carnot groups.\
Given a (real) analytic manifold $M$, we say that $(\Delta,g)$ is analytic if both $\Delta$ and $g$ are analytic on $M$. Moreover, we recall that a distribution $\Delta$, or a sub-Riemannian structure $(\Delta,g)$, is [*two-step*]{} if $$[\Delta,\Delta](x):= \Bigl\{ [X,Y](x) \, \vert \, X, Y \mbox{ smooth sections of } \Delta \Bigr\} =T_xM \qquad \forall x \in M.$$ A measure on $M$ is called [*smooth*]{} if it is locally defined by a positive smooth density times the Lebesgue measure $\mathcal{L}^n$, our first result is the following:
\[THMgen\] Every two-step analytic sub-Riemannian structure on a compact analytic manifold equipped with a smooth measure satisfies $\mbox{MCP}(0,N)$ for some $N>0$.
In the case of Carnot groups which are, as Lie groups equipped with left-invariant sub-Riemannian structures, analytic manifolds with analytic sub-Riemannian structures, the homogeneity allows us to extended the above result to left-invariant Lipschitz distributions.\
Following [@riffordbook], we say that a sub-Riemannian structure $(\Delta,g)$ or a Carnot group whose first layer $\Delta$ is equipped with a left-invariant metric, is [*Lipschitz*]{} if it is complete and the associated geodesic distance $d_{SR} : M\times M \rightarrow \R$ is locally Lipschitz outside of the diagonal $D=\{(x,y) \in M\times M\, \vert \, x=y \}$. Examples of Lipschitz sub-Riemannian structures include two-step distributions and more generally medium-fat distributions. A distribution $\Delta$ (or a sub-Riemannian structure with distribution $\Delta$ or a Carnot group whose first layer $\Delta$ is equipped with a left-invariant metric) is called *medium-fat* if, for every $x\in M$ and every smooth section $X$ of $\Delta$ with $X(x) \neq 0$, there holds $$\begin{aligned}
\label{EQmediumfat}
T_xM = \Delta(x)+ [\Delta,\Delta](x) + \bigl[X,[\Delta,\Delta] \bigr](x),\end{aligned}$$ where $$\bigl[X,[\Delta,\Delta] \bigr] (x) := \Bigl\{ \bigl[X,[Y,Z]\bigr] (x) \, \vert \, Y, Z \mbox{ smooth sections of } \Delta \Bigr\}.$$ The notion of medium-fat distribution has been introduced by Agrachev and Sarychev in [@as99]. Of course, in the case of a Carnot group the property of being medium-fat depends only on the properties of its Lie algebra. Our second result is the following:
\[THMCarnot\] Any Lipschitz Carnot group whose first layer is equipped with a left-invariant metric and equipped with Haar measure satisfies $\mbox{MCP}(0,N)$ for some $N>0$.
The proofs of Theorem \[THMgen\] and \[THMCarnot\] are based on the fact that squared sub-Riemannian pointed distances $d_{SR}(x,\cdot)^2$ satisfy a certain property of horizontal semiconcavity. Note that for the moment, we are only able to prove this property in the analytic case under an assumption of compactness of length minimizers. The property of horizontal semiconcavity together with the lipschitzness of $d_{SR}(x,\cdot)^2$ allows us to give an upper bound for divergence of horizontal gradients of $f^x$ which implies the desired measure contraction property.\
It is worth to notice that, thanks to a seminal result by Cavaletti and Huesmann [@ch15], measure contraction properties are strongly connected with the well-posedness of the Monge problem for quadratic geodesic distances. We refer the interested reader to [@badreddine17; @badreddinethesis] for further details.\
We recall that all the notations used throughout the paper are listed in Appendix \[Notations\]. The material required for the proof of the two theorems above is worked out in Section \[SECprel\]. The proofs of Theorems \[THMgen\] and \[THMCarnot\] are respectively given in Sections \[SECproofTHMgen\] and \[SECproofTHMCarnot\].\
[**Acknowledgement.**]{} The authors are grateful to the referee for useful remarks and for pointing out a gap in the initial proof of Proposition 12. The authors are also indebted to Adam Parusinski for fruitful discussions and the reference to the paper by Denef and Van den Dries [@dvdd88].
Preliminaries {#SECprel}
=============
Throughout all this section, $(\Delta,g)$ denotes a complete sub-Riemannian structure on $M$ of rank $m\leq n$.
The minimizing Sard conjecture
------------------------------
The minimizing Sard conjecture is concerned with the size of points that can be reached from a given point by singular minimizing geodesics. Following [@riffordbourbaki], given $x\in M$, we set $$\mathcal{S}^x_{\Delta,min^g} := \Bigl\{ \gamma(1) \, \vert \, \gamma \in W^{1,2}_{\Delta}([0,1],M), \gamma \mbox{ sing., } d_{SR}(x,\gamma(1))^2=\mbox{energy}_{g} (\gamma) \Bigr\}.$$ Note that for every $x \in M$, the set $\mathcal{S}^x_{\Delta,min^g}$ is closed and contains $x$ (because $m<n$). Let us introduce the following definition.
\[DEFSardmin\] We say that $(\Delta,g)$ satisfies the minimizing Sard conjecture at $x\in M$ if the set $\mathcal{S}^x_{\Delta,min^g}$ has Lebesgue measure zero in $M$. We say that it satisfies the minimizing Sard conjecture if this property holds for any $x\in M$.
It is not known if all complete sub-Riemannian structures satisfy the minimizing Sard conjecture (see [@agrachev14; @riffordbourbaki]). The best general result is due to Agrachev who proved in [@agrachev09] that all closed sets $\mathcal{S}^x_{\Delta,min^g}$ have empty interior. As the next result shows, the minimizing Sard conjecture is related to regularity properties of pointed distance functions. Following Agrachev [@agrachev09], we call smooth point of the function $y\mapsto d_{SR}(x,y)$ (for a fixed $x\in M$) any $y\in M$ for which there is $p\in T_x^*M$ which is not a critical point of the exponential mapping $\exp_x$ and such that the projection $\gamma_{x,p}$ of the normal extremal $\psi:[0,1] \rightarrow T^*M$ starting at $(x,p)$ is the unique minimizing geodesic from $x$ to $y=\gamma_{x,p}(1)$. By Agrachev’s Theorem, the set $\mathcal{O}_x$ of smooth points is always open and dense in $M$. The following holds:
\[PROPSardminEQ\] Let $x\in M$ be fixed, the following properties are equivalent:
- the structure $(\Delta,g)$ satisfies the minimizing Sard conjecture at $x\in M$,
- the function $y\mapsto d_{SR}(x,y)$ is differentiable almost everywhere in $M$,
- the set of smooth points $\mathcal{O}_x$ is an open set with full measure in $M$.
Furthermore, the function $y\mapsto d_{SR}(x,y)$ is smooth on $\mathcal{O}_x$ and if $M$ and $(\Delta,g)$ are analytic, then the set $\mathcal{O}_x$ is geodesically star-shaped at $x$, that is $$\begin{aligned}
\label{starshaped}
\gamma(s,y)\in \mathcal{O}_x \qquad \forall s\in (0,1], \, \forall y \in \mathcal{O}_x, \end{aligned}$$ where $\gamma_x (\cdot,y)\in W_{\Delta}^{1,2}([0,1],M)$ is the unique minimizing geodesic from $x$ to $y$.
Let $x\in M$ be fixed. The part (iii) $\Rightarrow$ (ii) is immediate. Let us prove that (ii) $\Rightarrow$ (i). By assumption the set of differentiability $D$ of $f:=d_{SR}(x,\cdot)$ has full measure in $M$. Recall that for every $y\in D$, there is a unique minimizing geodesic from $x$ to $y$ which is given by the projection of the normal extremal $\psi:[0,1] \rightarrow T^*M$ such that $\psi(1)=(y,d_{SR}(x,y)d_yf)$ (see [@riffordbook Lemma 2.15 p. 54]). By Sard’s Theorem, the set $S$ of $\exp_x(p)$ with $p\in T_x^*M$ critical has Lebesgue measure zero in $M$. Therefore, the set $D\setminus S$ has full measure and for every $y\in D\setminus S$ there is a unique minimizing geodesic from $x$ to $y$ and it is not singular, which shows that $y$ does not belong to $\mathcal{S}^x_{\Delta,min^g}$. Let us now show that (i) $\Rightarrow$ (iii). By definition of $\mathcal{S}^x_{\Delta,min^g}$, for every $y\notin \mathcal{S}^x_{\Delta,min^g}$ all minimizing horizontal paths between $x$ and $y$ are not singular. So repeating the proof of [@riffordbook Theorem 3.14 p. 98] (see also [@cr08]), we can show that the function $f:y\mapsto d_{SR}(x,y)$ is locally semiconcave and so locally Lipschitz on the open set $U:=M\setminus \mathcal{S}^x_{\Delta,min^g}$. Thus for every compact set $K \subset U$, there is a compact set $\mathcal{P}_K\subset T^*_xM$ such that for every $y\in K$, there is $p\in \mathcal{P}_K$ with $\exp_x(p)=y$ and $H(x,p)=d_{SR}(x,y)^2/2$ (in other words $\gamma_{x,p}:[0,1]\rightarrow M$ is a minimizing geodesic from $x$ to $y$). By Sard’s Theorem, the set $S_K$ of $\exp_x(p)$ with $p\in \mathcal{P}_K$ critical is a closed set of Lebesgue measure zero. For every positive integer $k$, set (here the diameter of the convex set $d_y^+f$ is taken with respect to some geodesic distance on $T^*M$) $$\Sigma^{k} (f) := \Bigl\{y \in U \, \vert \, \mbox{diam} (d^+_yf) \geq 1/k\Bigr\}.$$ By local semiconcavity of $f$ in $U$, each set $\Sigma^{k} (f)$ is a closed set in $U$ with Lebesgue measure zero (see [@cs04 Proposition 4.1.3 p. 79]). We claim that $$S_K' := K \cap \overline{ \bigcup_{k>0} \Sigma^k(f)} \subset \left( K \cap \bigcup_{k>0} \Sigma^k(f) \right) \cup S_K.$$ As a matter of fact, if $y\in K$ belongs to $ \overline{ \cup_{k>0} \Sigma^k(f)} \setminus \cup_{k>0} \Sigma^k(f)$, then $d_y^+f$ is a singleton and there is a sequence $\{y_l\}_l$ converging to $y$ such that all $d_{y_l}^+f$ have dimension at least one and tend to $d_y^+f$. This implies that the covector $p$ such that $\exp_x(p)=y$ and $H(x,p)=d_{SR}(x,y)^2/2$ is critical, which shows that $y$ belongs to $S_K$. By construction, every point in $K\setminus S_K'$ is a smooth point. We conclude easily.
It remains to prove the second part. The smoothness of $f:y\mapsto d_{SR}(x,y)$ is an easy consequence of the inverse function theorem. As a matter of fact, we can show easily that for every $y\in \mathcal{O}_x$ such that $y=\exp_x(p)$ with $H(x,p)=d_{SR}(x,y)^2/2$ and $p\in T_x^*M$ non-critical, there is a neighborhood $U$ of $y$ in $\mathcal{O}_x$ such that $$f(z)^2= 2H(x,\exp_x(z)^{-1}) \qquad \forall z \in U,$$ where $\exp_x^{-1}$ denotes a local inverse of the exponential mapping from a neighborhood of $p$ to $U$. To prove (\[starshaped\]), we argue by contradiction. If there are $x\in M$, $y\in \mathcal{O}_x$ and $s\in (0,1)$ such that $z:=\gamma(s,y)\in \mathcal{O}_x$ then either there are two distinct minimizing geodesics from $x$ to $z$ or there is only one minimizing geodesic from $x$ to $y$ which is singular. In the first case, we infer the existence of two distinct minimizing geodesics from $x$ to $y$, which contradicts the smoothness of $y$. In the second case, we deduce that the minimizing geodesic $\gamma_x(\cdot,y)$ is the projection of a normal extremal which is regular and whose restriction to $[0,s]$ is singular. This cannot happen under the assumption of analyticity of the datas because an abnormal extremal above $\gamma_x(\cdot,y)$ over $[0,s]$ could be extended to an abnormal extremal over $[0,1]$ (see [@riffordbook Proposition 1.11 p.21]).
By Proposition \[PROPSardminEQ\], any (complete) sub-Riemannian structure satisfying the minimizing Sard conjecture has negligeable cut loci.
\[REMample\] As pointed out by the referee, the set $\mathcal{O}_x$ could be replaced by the set $\mathcal{A}_x\subset \mathcal{O}_x$ of ample points from $x$. This set is defined as the set of $y \in \mathcal{O}_x$ where the unique minimizing geodesic from $x$ to $y$ is ample, that is whose growth vector saturates the tangent space. It can be shown to be open with full measure and geodesically star-shaped in the smooth case, we refer the interested reader to the monograph [@abr18] for further details. In our case, since we need the analyticity to prove other results, we prefer to work with the simpler $\mathcal{O}_x$ which is geodesically star-shaped in the analytic case.
Two characterizations for $\mbox{MCP}(0,N)$ {#SECMCP}
-------------------------------------------
The following result was implicit in the previous paper [@rifford13] of the second author (it is also the case in [@lee15 Page 5] and [@ohta14 Section 6.2]). The measure contraction property $\mbox{MCP}(0,N)$ is equivalent to some upper bound on the divergence of the horizontal gradient of the squared pointed sub-Riemannian distance. This result holds at least whenever the horizontal gradient is well-defined and the sets $\mathcal{O}_x$ are geodesically star-shaped.
\[PROPversus\] Assume that $(\Delta,g)$ satisfies the minimizing Sard conjecture and that all its sets $\mathcal{O}_x$ are geodesically star-shaped, and let $\mu$ be a smooth measure on $M$ and $N>0$ be fixed. Then $(\Delta,g)$ equipped with $\mu$ satisfies $\mbox{MCP}(0,N)$ if and only if $$\begin{aligned}
\label{divfxN}
\mbox{div}_y^{\, \mu} \left( \nabla^h f^x \right) \leq N \qquad \forall y \in \mathcal{O}_x, \, \forall x \in M,\end{aligned}$$ where $f^x:M\rightarrow \R$ is the function defined by $f^x(y):=d_{SR}(x,y)^2/2$.
Let $x\in M$ be fixed, the vector field $Z:=-\nabla^h f^x $ is well-defined and smooth on $\mathcal{O}_x$. Moreover by assumption, every solution of $\dot{y}(t)=Z(y(t))$ with $y(0)\in \mathcal{O}_x$ remains in $\mathcal{O}_x$ for all $t\geq 0$, we denote by $\{\varphi_t\}_{t\geq0}$ the flow of $Z$ on $\mathcal{O}_x$. For every $y\in \mathcal{O}_x$, the function $\theta : t\in [0,+\infty) \mapsto d_{SR}(\varphi_t(y),y)$ satisfies $$\theta(0)=0 \quad \mbox{and} \quad \theta(t) = \mbox{length}^g \left( \varphi_{[0,t]}(y)\right) = \int_0^t \left| Z(\varphi_s(y) \right|\, ds.$$ So that, for all $t\geq 0$, $$\dot{\theta}(t)= \left| Z(\varphi_t(y) \right| = d_{SR} \left(x,\varphi_t(y)\right) = d_{SR}(x,y) - d_{SR}\left(y,\varphi_t(y)\right) = d_{SR}(x,y) - \theta(t),$$ which yields $$\theta(t) = d_{SR}(x,y) \left( 1-e^{-t}\right) \qquad \forall t \geq 0.$$ Consequently, if $A\subset \mathcal{O}_x$ is a Borel set and $s\in (0,1]$, then we have $$A_s = \Bigl\{ \gamma_x(s,y) \, \vert \, y \in A\Bigr\} = \varphi_t(A) \quad \mbox{with} \quad t=- \ln (s).$$ Let us now assume that (\[divfxN\]) is satisfied. By definition of $\mbox{div}^{\mu}Z$, for every $x\in M$ and any measurable set $A\subset \mathcal{O}_x$, we have for every $t\geq 0$ (see for example, see [@br16 Proposition B.1]), $$\mu\left(\varphi_t(A) \right) = \int_A \exp \left( \int_0^t \mbox{div}_{\varphi_s(y)}^{\mu} (Z) \, ds\right) \, d\mu(y),$$ which by (\[divfxN\]) implies with $s=e^{-t}$, $$\mu \left( A_s\right) = \mu\left(\varphi_t(A) \right) \geq \int_A \exp \left(-Nt\right) \, d\mu(y) = s^N \mu(A).$$ This shows that (\[divfxN\]) implies $\mbox{MCP}(0,N)$. Conversely, if $(\Delta,g)$ equipped with $\mu$ satisfies $\mbox{MCP}(0,N)$ then for every $x\in M$ and every small ball $B_{\delta}(y)\subset \mathcal{O}_x$ (say a Riemannian ball with respect to the Riemannian extension $g$), we have $$\mu\left(\varphi_t \left(B_{\delta}(y)\right) \right) = \int_{B_{\delta}(y)} \exp \left( \int_0^t \mbox{div}_{\varphi_s(y)}^{\mu} (Z) \, ds\right) \, d\mu(y)\geq e^{-Nt}\, \mu \left(B_{\delta} (y) \right) \qquad \forall t \geq 0.$$ For every $t\geq 0$, letting $\delta$ go to $0$ yields $$\exp \left( \int_0^t \mbox{div}_{\varphi_s(y)}^{\mu} (Z) \, ds\right)\geq e^{-Nt}.$$ We infer (\[divfxN\]) by dividing by $t$ and letting $t$ go to $0$.
In the case of Carnot groups, the invariance of the divergence of $\nabla^h f^x$ by dilation allows us to characterize $\mbox{MCP}(0,N)$ in term of a control on the divergence over a compact set not containing the origin.
\[PROPCarnoteq\] Let $\G$ be a Carnot group whose first layer is equipped with a left-invariant metric satisfying the minimizing Sard conjecture and $N>0$ fixed. Then the metric space $(\G,d_{SR})$ with Haar measure $\mu$ satisfies $\mbox{MCP}(0,N)$ if and only if $$\begin{aligned}
\label{divfxNsph}
\mbox{div}_y^{\, \mu} \left( \nabla^h f^0 \right) \leq N \qquad \forall y \in \mathcal{O}_0 \cap S_{SR}(0,1),\end{aligned}$$ where $f^0:M\rightarrow \R$ is the function defined by $f^0(y):=d_{SR}(0,y)^2/2$.
Since Carnot groups are indeed analytic, by the second part of Proposition \[PROPSardminEQ\] and Proposition \[PROPversus\], it is sufficient to show that (\[divfxNsph\]) is equivalent to $$\begin{aligned}
\label{21dec1}
\mbox{div}_y^{\, \mu} \left( \nabla^h f^0 \right) \leq N \qquad \forall y \in \mathcal{O}_0.\end{aligned}$$ Recall that by taking a set of exponential coordinates $(x_1, \ldots,x_n)$, we can identify $\G$ with its Lie algebra $\mathfrak{g}\simeq \R^n$ and indeed consider that we work with the Lebesgue measure in $\R^n$ and that the sub-Riemannian structure is globally parametrized by an orthonormal family of analytic vector fields $X^1, \ldots, X^n$ in $\R^n$ satisfying $$\begin{aligned}
\label{homo}
X^i \left( \delta_{\lambda}(x)\right) = \lambda^{-1} \, \delta_{\lambda} \left( X^i(x)\right) \qquad \forall x \in \R^n, \, \forall i=1, \ldots,n,\end{aligned}$$ where $\{\delta_{\lambda}\}_{\lambda > 0}$ is a family of dilations defined as ($d_1, \ldots, d_n$ are positive integers) $$\delta_{\lambda} \left(x_1,\ldots, x_n\right) = \left( \lambda^{d_1} x_1, \lambda^{d_2} x_2, \ldots, \lambda^{d_n} x_n \right) \qquad \forall x \in \R^n.$$ By the homogeneity property, we have $d_{SR} \left(0,\delta_{\lambda}(x)\right) = \lambda \, d_{SR}(0,x)$ for all $x \in \R^n$ and $\lambda >0$. Then we have $$\begin{aligned}
\label{homo2}
f^0\left( \delta_{\lambda}(x)\right) = \lambda^2 \, f^0(x) \quad \mbox{and} \quad d_{\delta_{\lambda}(x)}f^0\circ \delta_{\lambda}= \lambda^2 \, d_xf^0 \qquad \forall x \in \R^n, \, \forall \lambda >0.\end{aligned}$$ Recall that the horizontal gradient $\nabla^h f^0$ is given by $$\nabla^h_x f^0 = \sum_{i=1}^m \left( X^i \cdot f^0\right)(x) X^i(x) \qquad \forall x \in \R^n.$$ Therefore, by (\[homo\])-(\[homo2\]), we infer that for every $x\in \R^n$ and $\lambda >0$, $$\begin{aligned}
\nabla^h_{\delta_{\lambda}(x)} f^0 &= & \sum_{i=1}^m d_{\delta_{\lambda}(x)} f^0 \left( X^i \left( \delta_{\lambda}(x)\right) \right) \, X^i \left( \delta_{\lambda}(x)\right) \\
& = & \lambda^{-2} \sum_{i=1}^m d_{\delta_{\lambda}(x)} f^0 \left( \delta_{\lambda} \left( X^i (x) \right) \right) \, \delta_{\lambda} \left( X^i (x)\right)\\
& = & \sum_{i=1}^m d_x f^0 \left( X^i (x) \right) \, \delta_{\lambda} \left( X^i (x)\right) = \delta_{\lambda} \left( \nabla^h_x f^0\right).\end{aligned}$$ We deduce that $$\mbox{div}_{\delta_{\lambda}(x)}^{\, \mu} \left( \nabla^h f^0 \right) = \mbox{div}_x^{\, \mu} \left( \nabla^h f^0 \right) \qquad \forall x \in \R^n, \, \forall \lambda >0,$$ which shows that (\[divfxNsph\]) and (\[21dec1\]) are equivalent and concludes the proof.
Nearly horizontally semiconcave functions {#SECnhscf}
-----------------------------------------
Recall that without loss of generality, we can assume that the metric $g$ over $\Delta$ is the restriction of a global Riemannian metric on $M$. This metric allows us to define the $C^2$-norms of functions from $\R^m$ to $M$. In the following statement, $(e_1, \ldots, e_m)$ stands for the canonical basis in $\R^m$.
Let $C>0$ and $U$ an open subset of $M$, a function $f:U \rightarrow \R$ is said to be $C$-nearly horizontally semiconcave with respect to $(\Delta,g)$ if for every $y\in U$, there are an open neighborhood $V^y$ of $0$ in $\R^m$, a function $\varphi^y: V^y \subset \R^m \rightarrow U$ of class $C^2$ and a function $\psi^y : V^y \subset \R^m \rightarrow \R$ of class $C^2$ such that $$\begin{aligned}
\label{DEFnhscf1}
\varphi^y(0)=y, \quad \psi^y(0)= f(y), \quad f\left(\varphi^y(v)\right) \leq \psi^y(v) \, \, \, \forall v \in V^y, \end{aligned}$$ $$\begin{aligned}
\label{DEFnhscf2}
\Bigl\{ d_0\varphi^y (e_1), \ldots, d_0\varphi^y(e_m)\Bigr\} \mbox{ is an orthonormal family of vectors in } \Delta(y),\end{aligned}$$ and $$\begin{aligned}
\label{DEFnhscf3}
\left\|\varphi^y \right\|_{C^2}, \, \left\|\psi^y \right\|_{C^2} \leq C,\end{aligned}$$ where $\left\|\varphi^y \right\|_{C^2}, \, \left\|\psi^y \right\|_{C^2}$ denote the $C^2$-norms of $\varphi^y$ and $\psi^y$.
We refer the reader to [@mm16] for an other notion of horizontal semiconcavity of interest that has been investigated by Montanari and Morbidelli in the framework of Carnot groups.
If $m$ were equal to $n$ that is if we were in the Riemannian case, the above definition would coincide with the classical definitions of semiconcave functions (see [@cs04; @riffordbook]). Here, in the case $m<n$, the definition tells that at each point, there is a support function from above of class $C^2$ which bounds the function along a $C^2$ submanifold which is tangent to the distribution. This type of mild horizontal semiconcavity will allows us, at least in certain cases, to bound the divergence of the horizontal gradient of squared pointed sub-Riemannian distance functions.
Before stating the main result of this section, we recall that a minimizing geodesic $\gamma : [0,1] \rightarrow M$ from $x$ to $y$ is called normal if it is the projection of a normal extremal, that is a trajectory $\psi :[0,1] \rightarrow T^*M$ of the Hamiltonian vector field associated with $(\Delta,g)$. We refer the reader to Appendix \[Notations\] for more details. Here is our result ($|p|^*=|p|_x^*$ for every $(x,p)\in T^*M$ is the dual norm associated with $g$):
\[PROPhsc\] Assume that $M$ and $(\Delta,g)$ are analytic and let $K$ be a compact subset of $M$, $U\subset M$ a relatively compact open set of $M$, and $A>0$ satisfying the following property: For every $x\in K$, every $y \in \bar{U}$ and every minimizing geodesic $\gamma:[0,1] \rightarrow M$ from $x$ to $y$, there is $p\in T_x^*M$ with $|p|^*\leq A$ such that $\gamma$ is the projection of the normal extremal $\psi:[0,1] \rightarrow T^*M$ starting at $(x,p)$. Then there is $C>0$ such that for every $x\in K$, the function $y\mapsto d_{SR}(x,y)^2$ is $C$-nearly horizontally semiconcave in $U$.
Let $K$ be a compact set of $M$, $U$ be a relatively compact open set of $M$ and $\bar{x} \in K$ fixed, let us first show how to construct functions $\varphi^{\bar{y}}, \psi^{\bar{y}}$ of class $C^2$ satisfying (\[DEFnhscf1\])-(\[DEFnhscf2\]) for $\bar{y} \in \bar{U}$. Pick a minimizing geodesic $\bar{\gamma}:[0,1] \rightarrow M$ from $\bar{x}$ to $\bar{y}=\bar{\gamma}(1)$. There is an open neighborhood $U_{\bar{\gamma}}$ of $\bar{\gamma}([0,1])$ and a family $\mathcal{F}_{\bar{\gamma}}$ of $m$ analytic vector fields $X^1_{\bar{\gamma}}, \ldots, X_{\bar{\gamma}}^m$ in $M$ such that for every $z\in U_{\bar{\gamma}}$ the family $\{X_{\bar{\gamma}}^1(z), \ldots, X_{\bar{\gamma}}^m(z)\}$ is orthonormal with respect to $g$ and parametrize $\Delta$ (that is $\mbox{Span} \{X_{\bar{\gamma}}^1(z), \ldots, X_{\bar{\gamma}}^m(z)\}=\Delta(z)$) and for every $z \in M\setminus U_{\bar{\gamma}}$, $X_{\bar{\gamma}}^1(z), \ldots, X_{\bar{\gamma}}^m(z)$ belongs to $\Delta(z)$. Consider the End-Point mapping from $\bar{x}$ in time $1$ associated with the family $\mathcal{F}_{\bar{\gamma}}=\{X^1_{\bar{\gamma}}, \ldots, X_{\bar{\gamma}}^m\}$, it is defined by $$E^{\bar{x},1}_{\mathcal{F}_{\bar{\gamma}}}: u\in L^{2}([0,1];\R^m) \, \longmapsto \, \gamma_u^{\mathcal{F}_{\bar{\gamma}}}(1)\in M,$$ where $\gamma_u^{\mathcal{F}_{\bar{\gamma}}}(1) :[0,1] \rightarrow M$ is the solution to the Cauchy problem $$\begin{aligned}
\label{CPproof}
\dot{\gamma} (t) = \sum_{i=1}^m u_i(t) X_{\bar{\gamma}}^i(\gamma(t)) \, \mbox{ for a.e. } t \in [0,1], \quad \gamma(0)= \bar{x}.\end{aligned}$$ Note that taking the vector fields $X_{\bar{\gamma}}^1, \ldots, X_{\bar{\gamma}}^m$ equal to zero outside of an neighborhood of $U_{\bar{\gamma}}$, we may assume without loss of generality that $E^{x,1}_{\mathcal{F}_{\bar{\gamma}}}$ is well-defined on $L^{2}([0,1];\R^m)$. Recall that the function $E^{x,1}_{\mathcal{F}_{\bar{\gamma}}}$ is smooth and satisfies (see [@riffordbook Proposition 1.10 p. 19]) $$\begin{aligned}
\label{benin}
\Delta \left( \bar{y} \right) \subset \mbox{Im} \left(d_{u_{\bar{\gamma}}}E_{\mathcal{F}_{\bar{\gamma}}}^{\bar{x},1}\right),\end{aligned}$$ where $u_{\bar{\gamma}}$ is the unique control $u\in L^2([0,1],\R^m)$ such that $\gamma_u^{\mathcal{F}_{\bar{\gamma}}}=\bar{\gamma}$. Therefore, there are $v_{\bar{\gamma}}^1, \ldots, v_{\bar{\gamma}}^m \in L^{2}([0,1],\R^m)$ such that $$\begin{aligned}
\label{3-12Proof1}
d_{u_{\bar{\gamma}}}E^{\bar{x},1}_{\mathcal{F}_{\bar{\gamma}}}\left(v^i_{\bar{\gamma}}\right) = X_{\bar{\gamma}}^i \left( \bar{y} \right) \qquad \forall i=1, \ldots,m.\end{aligned}$$ Define $\varphi_{\bar{\gamma}}:\R^m \rightarrow M$ by $$\varphi_{\bar{\gamma}} (v) := E^{\bar{x},1}_{\mathcal{F}_{\bar{\gamma}}} \left( u_{\bar{\gamma}} + \sum_{i=1}^m v_i v_{\bar{\gamma}}^i\right) \qquad \forall v = (v_1, \ldots, v_m) \in \R^m.$$ By construction, $\varphi_{\bar{\gamma}}$ is smooth and satisfies $$\varphi_{\bar{\gamma}}(0)=E^{\bar{x},1}_{\mathcal{F}_{\bar{\gamma}}} \left( u_{\bar{\gamma}} \right)= \bar{\gamma}(1)= \bar{y},$$ and $$d_0 \varphi_{\bar{\gamma}}(e_i) = d_{u_{\bar{\gamma}}}E^{ \bar{x},1}_{\mathcal{F}_{\bar{\gamma}}} \left(v^i_{\bar{\gamma}}\right)=X^i_{\bar{\gamma}} \left( \bar{y} \right) \qquad \forall i=1, \ldots,m.$$ Moreover, for every $v \in \R^m$ such that the solution to (\[CPproof\]) associated with the control $u_{\bar{\gamma}} + \sum_{i=1}^m v_i v_{\bar{\gamma}}^i$ remains in $U_{\bar{\gamma}}$, we have $$d_{SR} \left( \bar{x},\varphi_{\bar{\gamma}}(v) \right)^2 \leq \left\| u_{\bar{\gamma}} + \sum_{i=1}^m v_i v_{\bar{\gamma}}^i\right\|_{L^2}^2 =:\psi_{\bar{\gamma}}(v).$$ By construction, $\varphi_{\bar{\gamma}}$ and $\psi_{\bar{\gamma}}$ are smooth, defined in a neighborhood of $0\in \R^m$ and satisfy (\[DEFnhscf1\])-(\[DEFnhscf2\]). It remains to show that the $C^2$ norms of $\varphi^{\bar{\gamma}}$ and $\psi^{\bar{\gamma}}$ can be taken to be uniformly bounded, this is the purpose of the next lemma.
\[LEMaims\] There are neighborhoods $U_{\bar{x}}$ and $U_{\bar{y}}$ respectively of $\bar{x}$ and $\bar{y}$ in $M$, a neighborhood $\mathcal{U}_{\bar{\gamma}}$ of $u_{\bar{\gamma}}$ in $L^2([0,1],\R^m)$ and $C_{\bar{\gamma}}>0$ such that for every $x \in K\cap U_{\bar{x}}$, $y\in \bar{U} \cap U_{\bar{y}}$ and every control $u_{x,y}$ associated with a minimizing geodesic $\gamma_{x,y}:[0,1] \rightarrow M$ from $x$ to $y$ with $u_{x,y} \in \mathcal{U}_{\bar{\gamma}}$, there are $v^1_{u_{x,y}}, \ldots, v^m_{u_{x,y}} \in L^{2}([0,1],\R^m)$ such that $$\begin{aligned}
\label{aims1}
d_{u_{x,y}}E^{x,1}_{\mathcal{F}_{\bar{\gamma}}}\left(v^i_{u_{x,y}}\right) = X_{\bar{\gamma}}^i (y) \qquad \forall i=1, \ldots,m\end{aligned}$$ and $$\begin{aligned}
\label{aims2}
\left\| v_{u_{x,y}}^i\right\|_{L^2} \leq C_{\bar{\gamma}} \qquad \forall i=1, \ldots,m.\end{aligned}$$
Note that if we prove for each $i\in \{1, \ldots, m\}$ the existence of neighborhoods $U_x^i, U_y^i$ and $\mathcal{U}_{\bar{\gamma}}^i$ such that (\[aims1\])-(\[aims2\]) are satisfied for $i$, then the result follows by taking the intersections of the neighborhoods $U_{\bar{x}}^i, U_{\bar{y}}^i$ and $\mathcal{U}_{\bar{\gamma}}^i$. So, let us fix $i$ in $\{1, \ldots,m\}$. By taking a chart on a neighborhood of $\bar{\gamma}([0,1])$ (that we still denote by $U_{\bar{\gamma}}$) we may assume that the restriction of $E^{\bar{x},1}_{\mathcal{F}_{\bar{\gamma}}}$ to a neighborhood of $u_{\bar{\gamma}}$ is valued in $\R^n$ and doing a change of coordinates $(y_1, \ldots, y_n)$ in a neighborhood $U_{\bar{y}}^0$ we may also assume that there are analytic vector fields $Y^1,\ldots, Y^m$ such that $$Y^1(y)= X^i (y) = \partial_{y_1}, \quad Y^j (y)= \partial_{y_j} + \sum_{l=m+1}^n a^j_l (y) \, \partial_{y_l} \qquad \forall j=2, \ldots,m, \, \forall y \in U_{\bar{y}}^0$$ and $$\Delta (z) = \mbox{Span} \Bigl\{ Y^1(z),\ldots, Y^m (z)\Bigr\} \qquad \forall z \in U_{\bar{\gamma}}.$$ Observe that by construction, there are analytic mappings $f^j_k$ on $U_{\bar{\gamma}}$ for $j\in \{1, \ldots,m\}$ and $k\in \{1, \ldots, m\}$ such that $$Y^j(z) = \sum_{k=1}^m f^j_k(z) \, X^k(z) \qquad \forall z \in U_{\bar{\gamma}}, \, \forall j =1, \ldots, m.$$ As a consequence, if $\gamma :[0,1] \rightarrow U_{\bar{\gamma}}$ is solution to $$\begin{aligned}
\label{aimsday2}
\dot{\gamma}(t) = \sum_{j=1}^m w_j(t) \, Y^j(\gamma(t)) \, \mbox{ for a.e. } t \in [0,1],\end{aligned}$$ for some $w\in L^2([0,1],\R^m)$, then there holds for a.e. $t\in [0,1]$, $$\begin{aligned}
\dot{\gamma}(t) = \sum_{j=1}^m w_j(t) \, \sum_{k=1}^m f_k^j(\gamma(t)) \, X^j(\gamma(t)) = \sum_{k=1}^m \left( \sum_{j=1}^m w_j(t) \, f_k^j(\gamma(t)) \right) \, X^k(\gamma(t)).\end{aligned}$$ This shows that if we denote by $\mathcal{F}$ the family $\{Y^1, \ldots, Y^m\}$, then any horizontal curve parametrized by $\mathcal{F}$ can be parametrized by $\mathcal{F}_{\bar{\gamma}}$ as above. On the other hand, any parametrization with respect to $\mathcal{F}_{\bar{\gamma}}$ leads to a parametrization with respect to $\mathcal{F}$. Consequently, there is a control $\bar{w}\in L^2([0,1),\R^m)$ such that the solution of (\[aimsday2\]) starting at $\bar{x}$ is equal to $\bar{\gamma}$ and there are neighborhoods $U_{\bar{x}}^0$ of $\bar{x}$ and $\mathcal{W}$ of $\bar{w}$ together with a mapping $G: U_{\bar{x}}^0 \times \mathcal{W} \rightarrow L^2([0,1],\R^m)$ of class $C^1$ such that $$E^{x,1}_{\mathcal{F}}(w) = E^{x,1}_{\mathcal{F}_{\bar{\gamma}}} ( G(x,w)) \qquad \forall x\in U_{\bar{x}}^0, \, \forall w \in \mathcal{W},$$ where $E^{x,1}_{\mathcal{F}}$ denotes the End-Point mapping from $x$ in time $1$ associated with the family $\mathcal{F}$. Therefore, in order to prove our result it is sufficient to prove that there is $\bar{C}>0$ such that for every $x$ in $K$ close to $\bar{x}$, every $y$ in $\bar{U}$ close to $\bar{y}$ and any $\gamma$ minimizing geodesic from $x$ to $y$ close to $\bar{\gamma}$ associated to the control $w\in \mathcal{W}$ with respect to $\mathcal{F}$, there is $\omega\in L^2([0,1],\R^m)$ such that $$\begin{aligned}
\label{aimsday2_1}
d_wE^{x,1}_{\mathcal{F}} (\omega) = X^i (y) = \partial_{y_1} \quad \mbox{and} \quad \left\| \omega\right\|_{L^2} \leq \bar{C}.\end{aligned}$$
Let $x\in U_{\bar{x}}^0$ and $w\in \mathcal{W}$ be fixed, the differential of $E^{x,1}_{\mathcal{F}}$ at $w$ is given by (see [@riffordbook Remark 1.5 p.15]) $$d_{w}E^{x,1}_{\mathcal{F}}(v) = S(1) \int_0^1 S(t)^{-1} B(t) v(t) \, dt \qquad \forall v \in L^2([0,1],\R^m),$$ where $S:[0,1] \rightarrow M_n(\R)$ is solution to the Cauchy problem $$\dot{S}(t) =A(t) S(t) \quad \mbox{for a.e. } t \in [0,1], \quad S(0)=I_n$$ and where the matrices $A(t)\in M_n(\R), B(t) \in M_{n,m}(\R)$ are defined by (note that $J_{Y^{1}}=J_{X^i}=0$) $$A(t) := \sum_{j=2}^m w_j(t) \, J_{Y^j} \left( \gamma(t) \right) \qquad \mbox{a.e. } t \in [0,1],$$ $$B(t) = \left( Y^1 \left( \gamma(t) \right), \ldots, Y^m \left( \gamma (t) \right) \right) \qquad \forall t \in [0,1].$$ By construction of $Y^1, \ldots, Y^m$, there is $\tau^0 \in (0,1)$ (which does not depend upon $x, y$ but only upon $\bar{x}, \bar{y}$) such that for almost every $t\in [1-\tau^0,1]$ the matrix $A(t)$ has the form $$A(t) =
\begin{pmatrix}
0 & \vline & \begin{matrix} 0 & \cdots & \cdots & 0 \end{matrix} &\vline & \begin{matrix} 0 & \cdots & 0 \end{matrix} \\
\hline
\begin{matrix}
0 \\
\vdots \\
\vdots \\
0
\end{matrix}
& \vline &
{\mbox{\normalfont\Large\bfseries 0}}_{m-1,m-1}
& \vline & {\mbox{\normalfont\Large\bfseries 0}}_{m-1,n-m} \\
\hline
\alpha(t)
& \vline & \beta(t) & \vline & \delta(t)
\end{pmatrix},$$ where $\alpha(t)$ is in $\R^{n-m}$, $\beta(t)=(\beta(t)_{k,l})$ is a $(n-m)\times (m-1)$ matrix and $\delta(t)$ is a $(n-m)$ square matrix. Therefore, as the solution of the Cauchy problem $\dot{\tilde{S}}(t)=\tilde{S}(t) A(1-t), \tilde{S}(0)=I_n$, the first column of the matrix $\tilde{S}(t):=S(1)S(1-t)^{-1}$ with $t\in [0,1]$ has the form $$\begin{aligned}
\label{aimsday2_3}
\begin{pmatrix}
\tilde{s}_1(t) \\
\vdots \\
\tilde{s}_n(t)
\end{pmatrix}
\quad \mbox{with} \quad
\left\{
\begin{array}{l}
\tilde{s}_1(t) = 1 \\
\tilde{s}_j(t) = 0 \quad \forall j=2, \ldots, m
\end{array}
\right.
\quad \forall t \in [0,\tau_0]\end{aligned}$$ and the column vector $\tilde{s}(t)$ with coordinates $(\tilde{s}_{m+1}(t), \ldots, \tilde{s}_n(t))$ is solution of $$\begin{aligned}
\label{aimsday2_2}
\dot{\tilde{s}}(t) = D(t) \, \alpha(1-t) \quad \forall t \in [0,\tau_0], \quad \tilde{s}(0)=0\end{aligned}$$ with $D(t)$ the $(n-m)$ square matrices satisfying $$\begin{aligned}
\label{aimsday2_22}
\dot{D}(t) = D(t) \delta(1-t) \quad \forall t \in [0,\tau_0], \quad D(0)=I_{n-m}.\end{aligned}$$ Thus, a way to solve $d_w E^{x,1}_{\mathcal{F}}(\omega)=\partial_{y_1}$ is to take $\omega \in L^2([0,1], \R^m)$ of the form $$\begin{aligned}
\label{johan1}
\omega(t) = \left(\omega_1(1-\cdot),0, \ldots, 0\right)\qquad \mbox{a.e. } t \in [0,1]\end{aligned}$$ with $$\begin{gathered}
\label{johan2}
\mbox{Supp}(\omega_1) \subset [0,\tau^0], \quad \int_0^1 \omega_1(t) \, dt = 1 \\
\quad \mbox{and} \quad \int_0^1 \omega_1(t) \, \tilde{s}_l(t)\, dt = 0 \quad \forall l=m+1, \ldots, n.\end{gathered}$$
Remember now that by assumption, $M$ and $(\Delta,g)$ are analytic and for every $x\in K$ and $ y \in \bar{U}$ all minimizing geodesics joining $x$ to $y$ are normal with initial covector bounded by $A$. Let $\bar{p}\in T_{\bar{x}}^*M$ be such that $\bar{\gamma}=\gamma_{\bar{x},\bar{p}}$ where for any $(x,p)\in T^*M$, $\gamma_{x,p}:[0,1] \rightarrow M$ denotes the projection of the normal extremal $\psi_{x,p}:[0,1] \rightarrow T^*M$ starting at $(x,p)$ (see Appendix \[Notations\]). Then, the desired result will follow if we show that there are a neighborhood $\mathcal{T}$ of $(\bar{x},\bar{p})$ in $T^*M$ and $\bar{C}>0$ such that for every $(x,p)\in \mathcal{T}$, the point $x$ belongs to $U_{\bar{x}}^0$, the curve $\gamma_{x,p}([0,1])$ is contained in $U_{\bar{\gamma}}$, the associated control $w=w^{x,p}$ belongs to $\mathcal{W}$ and there is $\omega= \omega^{x,p} \in L^2([0,1],\R^m)$ satisfying (\[johan1\])-(\[johan2\]) (with the function $\tilde{s}=\tilde{s}^{x,p}=(\tilde{s}_{m+1}(t), \ldots, \tilde{s}_n(t)) :[0,1]\rightarrow \R^{n-m}$ given by the first column of $\tilde{S}(t)=\tilde{S}^{x,p}(t):=S(1)S(1-t)^{-1}$ associated with $\gamma_{x,p}$ and $w_{x,p}$ which satisfies (\[aimsday2\_3\])-(\[aimsday2\_2\]) with $\alpha=\alpha^{x,p}$ and $\delta=\delta^{x,p}$ associated with $\gamma_{x,p}$ and $w_{x,p}$ as well) such that $$\begin{aligned}
\label{johan3}
\left\| \omega^{x,p} \right\|_{L^2} \leq \bar{C}.\end{aligned}$$ We need the following lemma whose proof is postponed to Appendix \[PROOFbenin\].
\[benin4j\] Let $\mathcal{K} \subset \R^l$ be a compact set and $h:[0,1] \times \mathcal{K} \rightarrow \R$ be an analytic mapping such that $h(0,\kappa)=0$ for all $\kappa\in \mathcal{K}$. Then there are $\tau>0$ as small as desired and $\nu \in (0,1)$ such that $$\begin{aligned}
\label{eq15oct}
\left( \int_0^{\tau} h(t,p) \, dt \right)^2 \leq \nu \, \tau \, \int_0^{\tau} h(t,p)^2 \, dt \qquad \forall \kappa \in \mathcal{K}.\end{aligned}$$
Let $\mathcal{T}$ be a compact neighborhood of of $(\bar{x},\bar{p})$ in $T^*M$ such that for every $(x,p)\in \mathcal{T}$, the point $x$ belongs to $U_{\bar{x}}^0$, the curve $\gamma_{x,p}([0,1])$ is contained in $U_{\bar{\gamma}}$ and the associated control $w=w^{x,p}$ belongs to $\mathcal{W}$. Then we note that the mapping $$(t,x,p,\lambda) \in [0,1] \times \mathcal{T} \times [-1,1]^{n-m} \longmapsto h(t,x,p,\lambda) := \sum_{k=1}^{n-m} \lambda_k \tilde{s}_k^{x,p}(t)$$ is analytic. Therefore, by Lemma \[benin4j\], there are $\tau\in(0,\tau^0)$ and $\nu\in (0,1)$ such that $$\begin{aligned}
\label{johan6}
\left( \int_0^{\tau} h(t,x,p,\lambda) \, dt \right)^2 \leq \nu \tau \, \int_0^{\tau} h(t,x,p,\lambda)^2 \, dt \qquad \forall (x,p) \in \mathcal{T}, \, \forall \lambda \in [-1,1]^{n-m}.\end{aligned}$$ Define $I\in L^2([0,1],\R)$ by $I(t):= 1_{[0,\tau]}(t)$ for all $t\in [0,1]$ and for every $(x,p) \in \mathcal{T}$ denote by $P\in L^2([0,1],\R)$ the orthogonal projection of $I$ in $L^2([0,1],\R)$ over the vector space $$V^{x,p}:= \Bigl\{ f \in L^2([0,1],\R) \, \vert \, \langle f, 1_{[0,\tau]} \, \tilde{s}_l^{x,p} \rangle_{L^2}=0, \, \forall l=m+1, \ldots,n \Bigr\}.$$ Let $(x,p)\in \mathcal{T}$ be fixed, then there are $\Lambda_1^{x,p}, \ldots, \Lambda_{n-m}^{x,p} \in \R$ such that $$P= I + \sum_{k=1}^{n-m} \Lambda_k \, 1_{[0,\tau]} \, \tilde{s}^{x,p}_{m+k} =: I+ J$$ and there holds $$\begin{gathered}
\langle P, 1_{[0,\tau]} \, \tilde{s}^{x,p}_l \rangle_{L^2} = \int_0^1 P(t) 1_{[0,\tau]} (t) \, \tilde{s}_l^{x,p}(t) \, dt =\\
0 = \int_0^{\tau} \left[ 1_{[0,\tau]} (t) P(t) \right] \, \tilde{s}_l^{x,p}(t) \, dt \qquad \forall l=m+1, \ldots, n-m\end{gathered}$$ and $$\begin{aligned}
\int_0^1 1_{[0,\tau]} (t) P(t) (t) \, dt =\langle P,I\rangle_{L^2} = \|P\|^2_{L^2} & = & \|I\|_{L^2}^2 + \|J\|_{L^2}^2 + 2 \langle I,J\rangle_{L^2} \\
& \geq & \|I\|_{L^2}^2 + \|J\|_{L^2}^2 - 2 \sqrt{\nu} \|I\|_{L^2} \, \|J\|_{L^2}\\
& \geq & \left(1- \sqrt{\nu}\right) \|I\|_{L^2}^2 = \nu \left(1-\sqrt{\nu}\right),\end{aligned}$$ where we used that setting $\Lambda_{\bar{k}} = \max \{\Lambda_1, \ldots, \Lambda_{n-m}\} $ (note that $\Lambda_{\bar{k}} \neq0$) thanks to (\[johan6\]) yields $$\begin{aligned}
\left( \langle I,J\rangle_{L^2} \right)^2 = \left(\int_0^{\tau} \sum_{k=1}^{n-m} \Lambda_k \tilde{s}_k^{x,p}(t) \, dt \right)^2& = & \Lambda_{\bar{k}}^2 \left(\int_0^{\tau} \sum_{k=1}^{n-m} \frac{\Lambda_k}{\Lambda_{\bar{k}}} \, \tilde{s}_k^{x,p}(t) \, dt \right)^2 \\
& \leq & \Lambda_{\bar{k}}^2 \, \nu \tau \, \int_0^{\tau} \left( \sum_{k=1}^{n-m} \frac{\Lambda_k}{\Lambda_{\bar{k}}} \, \tilde{s}_k^{x,p}(t) \right)^2 \, dt \\
& = & \nu \, \|I\|_{L^2}^2 \, \|J\|_{L^2}^2.\end{aligned}$$ In conclusion, we deduce that for every $(x,p)\in \mathcal{T}$, the function $\omega=\omega^{x,p} \in L^2([0,1], \R^m)$ of the form (\[johan1\]) with $\omega_1=\omega_1^{x,p}$ given by $$\omega_1^{x,p}(t) := \frac{1_{[0,\tau]} (t) P(t)}{ \|P\|^2_{L^2} } \qquad \forall t \in [0,1]$$ satisfies (\[johan2\]) and $$\left\| \omega_1^{x,p} \right\|_{L^2} \leq \frac{1}{\sqrt{ \nu \left(1-\sqrt{\nu}\right)}}.$$ The proof of Lemma \[LEMaims\] is complete.
We conclude easily by compactness of $K$ and $\bar{U}$.
Proof of Theorem \[THMgen\] {#SECproofTHMgen}
===========================
Let $M$ be an analytic compact manifold, $(\Delta,g)$ a two-step analytic sub-Riemannian structure and $\mu$ a smooth measure on $M$. The following result, due to Agrachev and Lee [@al09] (see also [@riffordbook]), is a consequence of the fact that $\Delta$ is two-step (and the compactness of $M$). We refer the reader to [@al09; @riffordbook] for the proof. In fact, it is worth mentioning that Agrachev and Lee prove that a sub-Riemannian structure is two-step if and only if $d_{SR}$ is locally Lipschitz in charts.
\[LEMgen2\] The function $d_{SR}^2 : M\times M \rightarrow \R$ is locally Lipschitz in charts. In particular, there is $L>0$ such that $\left| \nabla_yf^x\right| \leq L$ for all $x\in M$ and $y \in \mathcal{O}_x$. Furthermore, there is $A>0$ such that for every $x, y \in M$ and every minimizing geodesic $\gamma:[0,1] \rightarrow M$ from $x$ to $y$, there is $p\in T_x^*M$ with $|p|^*\leq A$ such that $\gamma$ is the projection of the normal extremal $\psi:[0,1] \rightarrow T^*M$ starting at $(x,p)$.
By the above lemma and Proposition \[PROPhsc\], there is $C>0$ such that for every $x\in M$ the function $f^x: y\mapsto d_{SR}(x,y)^2/2$ is $C$-nearly horizontally semiconcave in $M$.
\[LEMgen1\] There is $B>0$ such that for every $x\in M$ the following property holds: for every $y\in \mathcal{O}_x$, there is a neighborhood $U^y \subset \mathcal{O}_x$ of $y$ along with an orthonormal family of smooth vector fields $X^1,\ldots, X^m$ which parametrize $\Delta$ in $U^y$ such that $$\begin{aligned}
\label{22dec6}
\left\| X^i\right\|_{C^1} \leq B \qquad \forall i=1, \ldots,m,\end{aligned}$$ and $$\begin{aligned}
\label{22dec5}
\left[X^i \cdot (X^i \cdot f^x)\right](z) \leq B \left|\nabla_z f^x\right| +B \qquad \forall z \in U^y, \, \forall i=1, \ldots,m,\end{aligned}$$ where $\nabla_z f^x$ stands for the gradient of $f^x$ at $z$ with respect to the global Riemannian metric $g$.
First of all, we notice that there is $A>0$ such that if $v^1, \ldots, v^m$ is an orthonormal family of tangent vectors in $\Delta(z)$ for some $z\in M$ then there is an orthonormal family of smooth vector fields $X^1, \ldots, X^m$ which generates the distribution $\Delta$ in a neighborhood of $z$ and such that $\|X^i\|_{C^1}$ is bounded by $A$ for all $i=1, \ldots, m$. Let $x\in M$ be fixed, then by $C$-nearly horizontal semiconcavity of $f^x$, for every $y\in M$, there are an open neighborhood $V^y$ of $0$ in $\R^m$, a function $\varphi^y: V^y \subset \R^m \rightarrow U$ of class $C^2$ and a function $\psi^y : V^y \subset \R^m \rightarrow \R$ of class $C^2$ such that (\[DEFnhscf1\]) (with $f=f^x$), (\[DEFnhscf2\]) and (\[DEFnhscf3\]) are satisfied. Fix $y\in \mathcal{O}_x$ and define the function $F^y:U^y \rightarrow \R$ by $F^y:=f^x\circ \varphi^y - \psi^y$, it is of class $C^2$ and satisfies $$d_0F^y=0 \quad \mbox{and} \quad \mbox{Hess}_0 F^y \leq 0.$$ Taking a chart near $y$ we can assume that we work in $\R^n$. Let $\varphi^y=(\varphi_1^y, \ldots, \varphi_n^y)$ and $(x_1, \ldots, x_n)$ and $(v_1, \ldots, v_m)$ the coordinates respectively in $\R^n$ and $\R^m$. Then, we have $$\frac{ \partial F^y}{\partial v_i}(0) = \left( \sum_{k=1}^n \frac{\partial f^x}{\partial x_k}(y) \frac{\partial \varphi_k^y}{\partial v_i}(0) \right) - \frac{\partial \psi^y}{\partial v_i} (0)=0 \qquad \forall i=1, \ldots, m$$ and for every $i=1, \ldots, m$, $$\begin{gathered}
\frac{\partial^2 F^y}{\partial v_i^2} (0) = \left( \sum_{k,l =1}^n \frac{\partial^2 f^x}{\partial x_l \partial x_k} (y) \frac{\partial \varphi_k^y}{\partial v_i}(0) \frac{\partial \varphi_l^y}{\partial v_i}(0) \right) \\
+ \left( \sum_{k=1}^n \frac{\partial f^x}{\partial x_k} (y)\frac{\partial^2 \varphi_k^y}{\partial v_i^2} (0)\right) - \frac{\partial^2 \psi^y}{\partial v_i^2}(0) \leq 0,\end{gathered}$$ which yields $$\begin{aligned}
\label{22dec2}
\sum_{k,l =1}^n \frac{\partial^2 f^x}{\partial x_l \partial x_k} (y) \frac{\partial \varphi_k^y}{\partial v_i}(0) \frac{\partial \varphi_l^y}{\partial v_i}(0) & \leq & \frac{\partial^2 \psi^y}{\partial v_i^2}(0) - \sum_{k=1}^n \frac{\partial f^x}{\partial x_k} (y)\frac{\partial^2 \varphi_k^y}{\partial v_i^2} (0) \nonumber\\
& \leq & C+ C \left| \nabla_y f^x\right|.\end{aligned}$$ By (\[DEFnhscf2\]) and the observation made at the very beginning of this proof, there is an orthonormal family of smooth vector fields $X^1, \ldots, X^m$ which generates the distribution $\Delta$ in a neighborhood of $z$ and such that $$\begin{aligned}
\label{22dec4}
\left\| X^i\right\|_{C^1} \leq A \quad \mbox{and} \quad d_0\varphi^y(e_i)=\frac{\partial \varphi^y}{\partial v_i}(0)=X^i(y) \qquad \forall i=1, \ldots, m.\end{aligned}$$ Setting $X^i=\sum_{k=1}^n a^i_k \partial_k$, we check easily that $$X^i\cdot f^x = \sum_{k=1}^n a^i_k \, \frac{\partial f^x}{\partial x_k}$$ and $$X^i \cdot \left(X^i \cdot f^x\right) = \sum_{k=1}^n \left( \sum_{l=1}^n a^i_l \frac{\partial a^i_k}{\partial x_l} \right) \, \frac{\partial f^x}{\partial x_k} + \sum_{k=1}^n a^i_k \left( \sum_{l=1}^n a^i_l \frac{\partial^2 f^x}{ \partial x_l \partial x_k}\right).$$ The last expression at $y$ yields, thanks to (\[22dec2\]) and (\[22dec4\]) (which implies $a^i_k(y)=\frac{\partial \varphi^y_k}{\partial v_i}(0)$ for all $i=1, \ldots, m$ and $k=1, \ldots, m$) $$\left[X^i \cdot (X^i \cdot f^x)\right](y) \leq A^2 \left| \nabla_y f^x\right| + C+ C \left| \nabla_y f^x\right| \qquad \forall i=1, \ldots,m.$$ We conclude easily by smoothness of $f^x$ in $\mathcal{O}_x$ with $U^y$ sufficiently small and $B>0$ sufficiently large.
In order to prove Theorem \[THMgen\], we need to bound from above the divergence of $f^x$ over $\mathcal{O}_x$ for all $x$ in $M$. The following holds:
\[LEMgen3\] There is $N>0$ such that the following property holds: $$\begin{aligned}
\label{divfxNgen}
\mbox{div}_y^{\, \mu} \left( \nabla^h f^x \right) \leq N \qquad \forall y \in \mathcal{O}_x, \, \forall x \in M.\end{aligned}$$
Let $x\in M$ and $y\in \mathcal{O}_x$ be fixed, by Lemma \[LEMgen1\] there is a neighborhood $U^y \subset \mathcal{O}_x$ of $y$ along with an orthonormal family of smooth vector fields $X^1,\ldots, X^m$ which parametrize $\Delta$ in $U^y$ such that (\[22dec5\]) holds. The horizontal gradient of $f^x$ in $U^y$ is given by $$\nabla^h_y f^x = \sum_{i=1}^m \left(X^i \cdot f^x\right) (y) X^i(y).$$ So, we have $$\mbox{div}_y^{\mu} \left( \nabla^h f^x \right) = \sum_{i=1}^m \left(X^i \cdot f^x\right) (y) \, \mbox{div}_y^{\mu} \left(X^i\right) + \sum_{i=1}^m \left[ X^i \cdot (X^i \cdot f)\right] (y).$$ The second term above (in the right-hand side) is bounded thanks to (\[22dec5\]) and Lemma \[LEMgen2\] and the first term is bounded by (\[22dec6\]) and Lemma \[LEMgen2\] (the quantities $\left(X^i \cdot f^x\right) (y) $ are indeed bounded by the fact that $d_{SR}(x,\cdot)$ is solution to the horizontal eikonal equation, see [@fr10]). The proof of Lemma \[LEMgen3\] is complete.
To conclude the proof of Theorem \[THMgen\], we observe that as the functions $f^x=d_{SR}(x,\cdot)^2/2$ are Lipschitz on $M$ the minimizing Sard conjecture is satisfied (by Proposition \[PROPSardminEQ\]) and we note that by analyticity the the sets $\mathcal{O}_x$ are geodesically star-shaped. Then we can apply Proposition \[PROPversus\] together with Lemma \[LEMgen3\].
Proof of Theorem \[THMCarnot\] {#SECproofTHMCarnot}
==============================
By Proposition \[PROPCarnoteq\], it is sufficient to show that (\[divfxNsph\]) holds. By assumption the function $f^0:y\rightarrow d_{SR}(0,y)^2/2$ is locally Lipschitz on $\G\setminus \{0\}$. Thus for every relatively compact open neighrborhood $U$ of $S_{SR}(0,1)$ with $\bar{U} \subset \G\setminus \{0\}$, there is $A>0$ such that for every $y \in \bar{U}$ and every minimizing geodesic $\gamma:[0,1] \rightarrow M$ from $0$ to $y$, there is $p\in T_0^*M$ with $|p|^*\leq A$ such that $\gamma$ is the projection of the normal extremal $\psi:[0,1] \rightarrow T^*M$ starting at $(0,p)$. Moreover, a Carnot group whose first layer is equipped with a left-invariant metric is an analytic manifold equipped with an analytic sub-Riemannian structure. Consequently, by Proposition \[PROPhsc\], the function $f^0:y\rightarrow d_{SR}(0,y)^2/2$ is $C$-nearly horizontally semiconcave in $\bar{U}$ and we can repeat the arguments used in the proof of Theorem \[THMgen\] for $y\in \mathcal{O}_0\cap S_{SR}(0,1)$.
Notations {#Notations}
=========
We list below the notations used throughout this paper, we refer the reader to the monographs [@abb12; @cs04; @montgomery02; @riffordbook] for further details:
- $M$ is a smooth manifold of dimension $n\geq 3$.
- $\Delta$ is a smooth totally nonholonomic distribution of rank $m< n$.
- $g$ is a smooth metric over $\Delta$. Sometimes, we see $g$ as the restriction of a global Riemannian metric $g$ on $M$. We use the notation $\langle \cdot, \cdot \rangle$ instead of $g_x(\cdot,\cdot)$ and we denote the norm associated with $g$ by $|\cdot|$ (instead of $|\cdot|_x=g_x(\cdot,\cdot)^{1/2}$). $B_{r}(x)$ stands for the open geodesic ball of radius $r>0$ centered at $x$. The dual norm associated with the Riemannian metric $g$ on each $T_x^*M$ is denoted by $|p|^*=|p|_x^*$ for every $(x,p)\in T^*M$.
- We call horizontal path any $\gamma:[0,1] \rightarrow M$ in $W^{1,2}$ which is almost everywhere tangent to $\Delta$. We denote by $W^{1,2}_{\Delta}([0,1],M)$ the set of horizontal paths $\gamma:[0,1] \rightarrow M$ endowed with the $W^{1,2}$-topology.
- For every $\gamma\in W^{1,2}_{\Delta}([0,1],M)$, we define the length of $\gamma$ (w.r.t. $g$) by $\mbox{length}^g(\gamma)= \int_0^1 |\dot{\gamma}(t)|\, dt$ and its energy (w.r.t. $g$) by $\mbox{energy}^g(\gamma)= \int_0^1 |\dot{\gamma}(t)|^2\, dt$.
- For any $x,y \in M$, we denote by $d_{SR}(x,y)$ (resp. $e_{SR}(x,y)$) the infimum of lengths (resp. energies) of horizontal paths joining $x$ to $y$. We note that $e_{SR}=d_{SR}^2$. We denote the open ball and the sphere centered at $x$ with radius $r>0$ respectively by $B_{SR}(x,r)$ and $S_{SR}(x,r)$. The geodesic distance $d_{SR}$ is said to be complete if the metric space $(M,d_{SR})$ is complete. In this case, all closed balls $\bar{B}_{SR}(x,r)$ are compact (for any $x\in M$ and any $r>0$).
- We call minimizing geodesic from $x$ to $y$ any $\gamma\in W^{1,2}_{\Delta}([0,1],M)$ with $\gamma(0)=x, \gamma(1)=y$ which minimizes the energy $e_{SR}(x,y)$ (and so the distance $d_{SR}(x,y)$), that is such that $\mbox{energy}^g(\gamma)=e_{SR}(\gamma)$. We note that if $d_{SR}$ is complete, then there exist minimizing geodesics between any pair of points.
- For every $x\in M$, we denote by $W^{1,2}_{\Delta,x}([0,1],M)$ the set of paths in $W^{1,2}_{\Delta}([0,1],M)$ starting at $x$ (that is $\gamma(0)=x$) and we define the end-point map $$E_{\Delta}^x \, : \, W^{1,2}_{\Delta,x}([0,1],M) \, \longrightarrow \, M$$ by $E_{\Delta}^x(\gamma)=\gamma(1)$. The infinite dimensional space $W^{1,2}_{\Delta,x}([0,1],M)$ has a smooth manifold structure and the end-point map $E_{\Delta}^x$ is smooth.
- An horizontal path $\gamma \in W^{1,2}_{\Delta,x}([0,1],M)$ is called singular if it is singular with respect to the end-point map $E_{\Delta}^x$, that is if the differential $d_{\gamma}E_{\Delta}^{x,1}$ is not surjective. It is convenient to rewrite the definition of singular curves in term of singular controls. If the distribution $\Delta$ is parametrized by a family $\mathcal{F}$ of $k$ smooth vector fields $X^1, \ldots, X^k$ in a open neighborhood of $\gamma([0,1])$ and if $u\in L^2([0,1],\R^k)$ satisfies $$\dot{\gamma} (t) = \sum_{i=1}^k u_i(t) \, X^i(\gamma(t)) \, \mbox{ for a.e. } t \in [0,1],$$ then $\gamma$ is singular if and only if the control $u$ is a singular point of the smooth mapping (well-defined in an open set $\mathcal{U}$) $$E_{\mathcal{F}}^{x,1} \, : \, \mathcal{U} \subset L^{2}([0,1],\R^k) \, \longrightarrow \, M$$ defined by $$E_{\mathcal{F}}^{x,1}(v) := \gamma_v(1) \qquad \forall v \in L^2([0,1],\R^k),$$ where $\gamma_v$ is the curve in $W^{1,2}_{\Delta,x}([0,1],M)$ solution to the Cauchy problem $$\dot{\gamma}_v (t) = \sum_{i=1}^k v_i(t) \, X^i \left(\gamma_v(t)\right) \, \mbox{ for a.e. } t \in [0,1], \quad \gamma_v(0)=x.$$
- An horizontal path $\gamma \in W^{1,2}_{\Delta,x}([0,1],M)$ is singular if and only if it is the projection of an abnormal extremal $\psi: [0,1] \rightarrow T^*M$ that never intersects the zero section of $T^*M$, such that $$\dot{\psi}(t) = \sum_{i=1}^k u_i(t) \vec{h}^i(\psi(t)) \qquad \mbox{for a.e. } t \in [0,1],$$ where $\mathcal{F}$ is a family of $k$ smooth vector fields $X^1, \ldots, X^k$ which parametrizes $\Delta$ in a open neighborhood of $\gamma([0,1])$ and $\vec{h}^1, \ldots, \vec{h}^k$ are the Hamiltonian vector fields associated canonically with $h^i(x,p)=p\cdot Xi(x)$ in $T^*M$. The curve $\psi$ is called an abnormal lift of $\gamma$ and $\gamma$ is said to be abnormal.
- The Hamiltonian $H:T^*M \rightarrow \R$ associated with $(\Delta,g)$ is defined by $$H(x,p) := \frac{1}{2} \max \left\{ \frac{p(v)^2}{g_x(v,v)} \, \vert \, v \in \Delta(x) \setminus \{0\} \right\} \qquad \forall (x,p)\in T^*M,$$ which coincides with $$\frac{1}{2} \sum_{i=1}^m \left( p\cdot X^i(x)\right)^2,$$ if $\Delta$ is parametrized locally by an orthonormal family $X^1, \ldots, X^m$. The Hamiltonian vector field $\vec{H}$ associated with $(\Delta,g)$ is the Hamiltonian vector field given by $H$ with respect to the canonical symplectic form on $T^*M$. In local coordinates $(x,p)$ the trajectories $\psi=(x,p)$ of $\vec{H}$ are solution to $$\dot{x} = \frac{\partial H}{\partial p}(x,p), \quad \dot{p}=- \frac{\partial H}{\partial x}(x,p),$$ we call them normal extremals. Any projection of a normal extremal is an horizontal path that is said to be normal.
- An horizontal path $\gamma$ is called strictly abnormal if it is abnormal (singular) and not normal.
- For every $x\in M$, the exponential mapping $\exp_x:T_x^*M \rightarrow M$ associated with $(\Delta,g)$ at $x$ is defined by $\exp_x:=\pi (\psi_{x,p}(1))$ where $\psi_{x,p}$ is the trajectory of $\vec{H}$ starting at $(x,p)$ and $\pi :T_x^*M \rightarrow M$ is the canonical projection.
- A Carnot group $(\G, \star )$ of step $s$ is a simply connected Lie group whose Lie algebra $\mathfrak{g}=T_0\G$ (we denote by $0$ the identity element of $\G$) admits a nilpotent stratification of step $s$, [*i.e.*]{} $$\begin{aligned}
\label{carnot0}
\mathfrak{g} = V_1 \oplus \cdots \oplus V_s,\end{aligned}$$ with $$\begin{aligned}
\label{carnot1}
\bigl[ V_1, V_j\bigr] = V_{j+1} \quad \forall 1\leq j \leq s, \quad V_s \neq\{0\}, \quad V_{s+1}= \{0\}.\end{aligned}$$ By simple-connectedness of $\G$ and nilpotency of $\mathfrak{g}$, $\exp_{\G}$ is a smooth diffeomorphism, which allows to identify $\G$ with its Lie algebra $\mathfrak{g} \simeq \R^n$. If the first layer $V_1$ of $\G$ is equipped with a left-invariant metric, then there is a set of coordinates $(x_1, \ldots, x_n)$, a one-parameter family of dilations $\{\delta_{\lambda}\}_{\lambda >0}$ of the form $$\delta_{\lambda} \left(x_1,\ldots, x_n\right) = \left( \lambda^{d_1} x_1, \lambda^{d_2} x_2, \ldots, \lambda^{d_n} x_n \right) \qquad \forall x \in \R^n,$$ and a orthonormal family of left-invariant vector fields generating $V_1$ satisfying $$X^i \left( \delta_{\lambda}(x)\right) = \lambda^{-1} \, \delta_{\lambda} \left( X^i(x)\right) \qquad \forall \lambda >0, \, x \in \R^n.$$
- A function $f:U\rightarrow \R$ on a open set $U\subset M$ is called locally semiconcave if for every $x\in U$ there are a open neighborhood $V\subset U$ of $x$ and $C>0$ such that for any $y\in V$ there is a function $\psi:M\rightarrow \R$ with $\|\psi\|_{C^2}\leq C$ such that $f\leq V$ on $M$ and $f(y)=\psi(y)$. For every $y\in U$, $d_y^+f$ denotes the set of super-differentials of $f$ at $y$, it is the set of $\alpha \in T_x^*M$ for which there is a function of class $C^1$, $\psi:M\rightarrow \R$ such that $\psi\geq f$ on $M$, $\psi(y)=f(y)$ and $d_y\psi=\alpha$.
- If $f:U\rightarrow M$ is smooth on the open set $U\subset M$, $\nabla^hf$ denotes its horizontal gradient with respect to $(\Delta,g)$. For every $y\in U$, $\nabla^h_yf$ is defined as the unique $v\in \Delta(y)$ such that $d_yf (w)=\langle v,w\rangle$ for all $w\in \Delta(y)$. If $\Delta(y)$ is generated by an orthonormal family $X^1(y), \ldots, X^m(y)$, then $\nabla^h_yf= \sum_{i=1}^m (X^i \cdot f)(y) \, X^i(y)$.
Proof of Lemma \[benin4j\] {#PROOFbenin}
==========================
Let $\mathcal{K} \subset \R^l$ be a compact set and $h:[0,1] \times \mathcal{K} \rightarrow \R$ an analytic mapping such that $h(0,\kappa)=0$ for all $\kappa\in \mathcal{K}$. Let $\bar{\kappa}\in\mathcal{K}$ be fixed, then by [@dvdd88 Lemma 4.12 p.126], there are an integer $d>0$ and $\rho>0$ together with analytic functions $a_1, \ldots, a_d$ on $B(\bar{\kappa},\rho)$ and $b_1, \ldots, b_d$ on $ [0,\rho] \times B(\bar{\kappa},\rho) $ such that $$h(t,\kappa) = \sum_{k=1}^{d} a_k(\kappa) \, b_k (t,\kappa) \, t^k \qquad \forall (t,\kappa) \in [0,\rho] \times B(\bar{\kappa},\rho)$$ and $$b_k (0,\bar{\kappa})=1 \qquad \forall k =1, \ldots, d.$$ Hence, by compactness of $\mathcal{K}$ we infer that there are an integer $\bar{d}>0$ and $\bar{\rho}>0$ such that for every $\bar{\kappa} \in \mathcal{K}$ there are analytic functions $a_1^{\bar{\kappa}}, \ldots, a_{\bar{d}}^{\bar{\kappa}}$ on $B(\bar{\kappa},\bar{\rho})$ and $b_1^{\bar{\kappa}}, \ldots, b_{\bar{d}}^{\bar{\kappa}}$ on $[0,\bar{\rho}] \times B(\bar{\kappa},\bar{\rho})$ such that $$h(t,\kappa) = \sum_{k=1}^{\bar{d}} a_k^{\bar{\kappa}}(\kappa) \, b_k^{\bar{\kappa}} (t,\kappa) \, t^k \qquad \forall (t,\kappa) \in [0,\bar{\rho}] \times B(\bar{\kappa},\bar{\rho})$$ and $$b_k (t,\kappa) \in [1/2,1] \qquad \forall (t,\kappa) \in [0,\bar{\rho}] \times B(\bar{\kappa},\bar{\rho}) , \, \forall k =1, \ldots, \bar{d}.$$
Let $\tau\in (0,\bar{\rho})$ be fixed. By Cauchy-Schwarz inequality, for every $\kappa \in \mathcal{K}$ we have $$\left( \int_0^{\tau} h(t,\kappa) \, dt \right)^2 \leq \tau \, \int_0^{\tau} h(t,\kappa)^2 \, dt$$ with equality only if $\lambda_1 h^2(\cdot,\kappa)=\lambda_2 h(\cdot,\kappa)$ for some nonzero $(\lambda_1,\lambda_2) \in \R^2$. Since $h(0,\kappa)=0$ the case of equality may only happen whenever $h(\cdot, \kappa)\equiv0$ where there holds $\left( \int_0^{\tau} h(t,\kappa) \, dt \right)^2 \leq \nu \tau \, \int_0^{\tau} h(t,\kappa)^2 \, dt$ for all $\nu \in (0,1)$. In conclusion, by compactness of $\mathcal{K}$ if there is no $\nu \in (0,1)$ such that $$\left( \int_0^{\tau} h(t,\kappa) \, dt \right)^2 \leq \nu \, \tau \, \int_0^{\tau} h(t,\kappa)^2 \, dt \qquad \forall \kappa \in \mathcal{K},$$ then there is a sequence $\{\kappa_l\}_l$ in $\mathcal{K}$ converging to some $\kappa \in \mathcal{K}$ such that $$h \left(\cdot,\kappa_l\right) \not\equiv 0 \, \, \forall l \quad \mbox{and} \quad \lim_{l \rightarrow \infty} \frac{\left( \int_0^{\tau} h \left(t,\kappa_l\right) \, dt \right)^2}{ \tau \int_0^{\tau} h \left(t,\kappa_l\right) ^2 \, dt} = 1.$$ Then there are analytic functions $a_1^{\kappa}, \ldots, a_{\bar{d}}^{\kappa}$, $b_1^{\kappa}, \ldots, b_{\bar{d}}^{\kappa}$ and a sequence $\{\bar{k}_l\}_l$ in $\{1, \ldots, \bar{d} \}$ such that for all $l$, $$a_{\bar{k}_l}^{\kappa} \left( \kappa_l\right) \neq 0 \quad \mbox{and} \quad \left| a_{k}^{\kappa} \left(\kappa_l\right) \right| \leq \left| a_{\bar{k}_l}^{\kappa} \left( \kappa_l\right) \right| \quad \forall k \in \left\{1, \ldots, \bar{d}\right\},$$ which allows to write $$\lim_{l \rightarrow \infty} \frac{\left( \int_0^{\tau} \sum_{k=1}^{\bar{d}} \frac{a_k^{\kappa} \left(\kappa_l\right)}{a_{\bar{k}_l}^{\kappa}\left(\kappa_l\right)} \, b_k^{\kappa} \left(t,\kappa_l\right) \, t^k \, dt \right)^2}{ \tau \int_0^{\tau} \left( \sum_{k=1}^{\bar{d}} \frac{a_k^{\kappa} \left(\kappa_l\right)}{a_{\bar{k}_l}^{\kappa}\left(\kappa_l\right)} \, b_k^{\kappa} \left(t,\kappa_l\right) \, t^k \right) ^2 \, dt} = 1.$$ Therefore, since all the quantities $ a_k^{\kappa} \left(\kappa_l\right) /a_{\bar{k}_l}^{\kappa}\left(\kappa_l\right)$ belong to $[-1,1]$ for all $k$ and are equal to $1$ for $k=\bar{k}_l$, and since the functions $b_1(\cdot,\kappa), \ldots, b_{\bar{d}}(\cdot, \kappa)$ are analytic, there are $c_1, \ldots c_{\bar{d}}$ in $[-1,1]$ and $\bar{k} \in \{1, \ldots, \bar{d}\}$ with $c_{\bar{k}}=1$ such that $$\left( \int_0^{\tau} \sum_{k=1}^{\bar{d}} c_k \, b_k^{\kappa} \left(t,\kappa \right) \, t^k \, dt \right)^2 = \tau \int_0^{\tau} \left( \sum_{k=1}^{\bar{d}} c_k \, b_k^{\kappa} \left(t,\kappa \right) \, t^k \right)^2 \, dt.$$ This means that there is a nonzero pair $(\lambda_1,\lambda_2) \in \R^2$ such that $$\lambda_1 \, \left( \sum_{k=1}^{\bar{d}} c_k \, b_k^{\kappa} \left(t,\kappa \right) \, t^k \right)^2 = \lambda_2 \, \left( \sum_{k=1}^{\bar{d}} c_k \, b_k^{\kappa} \left(t,\kappa \right) \, t^k \right) \qquad \forall t \in [0,\tau].$$ which implies that $$\sum_{k=1}^{\bar{d}} c_k \, b_k^{\kappa} \left(t,\kappa \right) \, t^k = 0 \qquad \forall t\in [0,\tau].$$ Let $k_0 \in \{1, \ldots, \bar{d}\}$ be such that $c_k=0$ for all $k\in \{1, \ldots, \bar{d}-1\}$ with $k<k_0$. Then we have $$c_{k_0} \, b_{k_0} \left(t,\kappa \right) \, t^{k_0} + t^{k_0} \left( \sum_{k=k_0+1}^{\bar{d}} c_k \, b_k \left(t,\kappa \right) \, t^{k-k_0}\right) = 0 \qquad \forall t\in [0,\tau],$$ which is impossible because $b_0(0,\kappa)\in [1/2,1]$. The proof is complete.
[99]{}
A. Agrachev. Compactness for Sub-Riemannian length-minimizers and subanalyticity. Control theory and its applications (Grado, 1998). , 56(4):1–12, 2001.
A. Agrachev. Any sub-Riemannian metric has points of smoothness. , 424(3): 295–298, 2009 Translation in [*Dokl. Math.*]{}, 79(1):45–47, 2009.
A. Agrachev. Some open problems. , 1-13. Springer INdAM, Ser. 5, Springer, Cham, 2014.
A. Agrachev, D. Barilari and U. Boscain. . Book in preparation, 2017.
A. Agrachev, D. Barilari and L. Rizzi. . Memoirs of the American Mathematical Society, 256, no 1225, 2018.
A. Agrachev and P. Lee. Optimal transportation under nonholonomic constraints. , 361(11):6019–6047, 2009.
A. Agrachev and P. Lee. Generalized Ricci curvature bounds for three dimensional contact subriemannian manifolds. , 360(1-2):209–253, 2014.
A. Agrachev and A. Sarychev. Sub-Riemannian metrics: minimality of singular geodesics versus subanalyticity. , 4:377–403, 1999.
Z. Badreddine. Mass transportation on sub-Riemannian structures of rank two in dimension four. Preprint, 2017.
Z. Badreddine. . Thesis, Université de Bourgogne - Franche Comté, 2017.
D. Barilari and L. Rizzi. Sharp measure contraction property for generalized $H$-type Carnot groups. , to appear.
A. Belotto da Silva and L. Rifford. The Sard conjecture on Martinet surfaces. , to appear.
P. Cannarsa and L. Rifford. Semiconcavity results for optimal control problems admitting no singular minimizing controls. , 25(4):773–802, 2008.
P. Cannarsa and C. Sinestrari. . Progress in Nonlinear Differential Equations and their Applications, 58. Birkhäuser Boston Inc., Boston, MA, 2004.
F. Cavalletti and M. Huesmann. Existence and uniqueness of optimal transport maps. , 32(6):1367–1377, 2015.
J. Denef and L. Van den Dries. $p$-adic and real subanalytic sets. , 128(1):79–138, 1988.
A. Figalli and L. Rifford. Mass Transportation on sub-Riemannian Manifolds. , 20(1):124–159, 2010.
N. Juillet. Geometric inequalities and generalized Ricci bounds in the Heisenberg group. , 13:2347–2373, 2009.
P.W.Y. Lee. Ricci curvature lower bounds on Sasakian manifolds. ArXiv e-prints, 2015.
P.W.Y. Lee. On measure contraction property without Ricci curvature lower bound. , 44:(1):27–41, 2016.
P.W.Y. Lee, C. Li and I. Zelenko. Ricci curvature type lower bounds for sub-Riemannian structures on Sasakian manifolds. , 36:(1):303–321, 2016.
J. Lott and C. Villani. Weak curvature conditions and functional inequalities. , 245(1):311–333, 2007.
R. Montgomery. . , Vol. 91. American Mathematical Society, Providence, RI, 2002.
A. Montanari and D. Morbidelli. On the lack of semiconcavity of the subRiemannian distance in a class of Carnot groups. , 444(2):1652–1674, 2016.
S. Ohta. On the measure contraction property of metric measure spaces. , 82(4):805–828, 2007.
S. Ohta. On the curvature and heat flow on Hamiltonian systems. , 2:81–114, 2014.
L. Rifford. . Springer Briefs in Mathematics, Springer, New York, 2014.
L. Rifford. Ricci curvatures in Carnot groups. , 3(4):467–487, 2013.
L. Rifford. Singulières minimisantes en géométrie sous-Riemannienne. Séminaire Bourbaki, Vol. 2015/2016. Exposés 1101-1119. , Vol. 390, Exp. No. 1179: 277-301, 2017.
L. Rifford and E. Trélat. Morse-Sard type results in sub-Riemannian geometry. , 332(1):145–159, 2005.
L. Rizzi. . Thesis, Scuola Internazionale di Studi Avanzati, 2014.
L. Rizzi. Measure contraction properties of Carnot groups. , 55(3), Art. 60, 20pp., 2016.
K. -T. Sturm. On the geometry of metric measure spaces. I. , 196(1):65–131, 2006.
K. -T. Sturm. On the geometry of metric measure spaces. II. , 196(1):133–177, 2006.
C. Villani. . Grundlehren der Mathematischen Wissenschaften, 338. Springer-Verlag, Berlin, 2009.
[^1]: Université Côte d’Azur, CNRS, Inria, Labo. J.-A. Dieudonné, UMR CNRS 7351, Parc Valrose 06108 Nice, Cedex 2, France ([[email protected]]{})
[^2]: Université Côte d’Azur, CNRS, Inria, Labo. J.-A. Dieudonné, UMR CNRS 7351, Parc Valrose 06108 Nice, Cedex 2, France ([ [email protected]]{})
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate sub-gap transport through a single-level quantum dot tunnel coupled to one superconducting and two normal-conducting leads. Despite the tendency of a large charging energy to suppress the equilibrium proximity effect, a finite Andreev current through the dot can be achieved in non-equilibrium situations. We propose two schemes to identify non-local Andreev transport. In one of them, the presence of strong Coulomb interaction leads to negative values of the non-local conductance as a clear signal of non-local Andreev transport.'
author:
- David Futterer$^1$
- Michele Governale$^1$
- 'Marco G. Pala$^2$'
- 'Jürgen König$^{1,3}$'
title: 'Non-local Andreev transport through an interacting quantum dot'
---
Introduction
============
The entanglement of electrons in Cooper pairs of a superconductor can generate non-local transport effects. A prominent example is *Crossed Andreev Reflection* (CAR) at the contact between a superconductor and two normal leads: there the two electrons with opposite spins and symmetric energies with respect to the Fermi level of the superconductor, that are transferred through the normal-superconductor interface via Andreev reflection,[@andreev64] originate from or end up in different normal leads. This is a non-local transport phenomenon which has been extensively studied both theoretically[@falci01; @lesovik01; @yamashita03; @sanchez03; @melin04; @morten06; @brinkmann06; @kalenkov07; @golubev07; @kalenkov07_2] and experimentally.[@bozhko82; @benistant83; @beckmann04; @russo05; @cadden-zimansky06]
The main problem to identify the non-locality of CAR in a transport measurement is to separate it from other transport channels. In this paper, we study non-local Andreev transport through a single-level quantum dot. The quantum-dot level energy can be tuned by a gate voltage, which opens the possibility to control the Andreev transport channels. At first glance, strong Coulomb interaction in the quantum dot seems to be counterproductive: the formation of a finite equilibrium superconducting pair amplitude is suppressed since a large charging energy prevents the equilibrium state to be a coherent superposition of dot states with particle numbers differing by two, and Cooper pairs can be transferred through the dot by higher-order tunneling processes (cotunneling) only. On the other hand, a finite *non-equilibrium* pair amplitude in the dot can be achieved with a bias voltage.[@pala07; @governale08] A large charging energy provides even the key ingredient for identifying non-local Andreev transport in one of the schemes we propose.
Andreev reflection through quantum dots, a problem which combines Coulomb interaction, superconducting correlations and non-equilibrium, has been extensively studied theoretically.[@fazio98; @fazio99; @kang98; @schwab99; @clerk00; @cuevas01; @pala07; @governale08] Here, we apply the diagrammatic real-time transport theory of Ref. . The relevance from the experimental point of view is proven by the recent success in coupling superconductors to quantum dots in either a carbon nanotube[@buitelaar02; @cleuziou06; @jarillo-herrero06; @jorgensen06] or a semiconductor nanowire.[@van_dam06; @sand-jespersen07; @buizert07] In particular, we propose to investigate non-local effects in Andreev transport through a single-level quantum dot with one superconducting and two normal-conducting leads, which may be (i) ferromagnetic with collinear magnetization or (ii) non-magnetic (see Fig. \[setup\]). Thereby we identify non-locality either (i) by the dependence of the Andreev current on the relative orientation of the two ferromagnets that are biased with the same voltage, or (ii) by the response of the current in one normal lead to the applied voltage in the other one.
![(color online) Setup: a quantum dot is tunnel coupled to one superconducting and two normal-conducting leads. The latter may be ferromagnetic with magnetization directions $\mathbf{\hat{m}_{\rm{L}}}$ and $\mathbf{\hat{m}_{\rm{R}}} = \pm \mathbf{\hat{m}_{\rm{L}}}$. \[setup\]](fig1.eps){width="2.6in"}
Model
=====
The Hamiltonian of the system is $H=H_{\rm{dot}}+\sum_{\eta}(H_{\eta}+H_{\rm{tunn},\eta})$. The label $\eta= \rm{L},\rm{R},\rm{S}$ corresponds to the left, right and the superconducting lead, respectively. The dot is described by the Anderson-impurity model: $$H_{\rm{dot}}=\sum_{\sigma} \epsilon d_{\sigma}^{\dagger} d_{\sigma} + U n_{\uparrow}
n_{\downarrow},$$ where $d_{\sigma}^{(\dagger)}$ is the annihilation (creation) operator for a dot electron with spin $\sigma$, $n_{\sigma}=d_{\sigma}^{\dagger} d_{\sigma} $ is the corresponding number operator, $\epsilon$ the energy of the spin-degenerate single-particle level, and $U$ the on-site Coulomb repulsion. The Hamiltonian of the lead $\eta$ reads $$H_{\eta}=\sum_{k \sigma} \epsilon_{\eta k\sigma}
c_{\eta k \sigma}^\dagger c_{\eta k \sigma}- g_\eta \sum_{k,k'} c_{\eta k \uparrow}^\dagger
c_{\eta -k \downarrow}^\dagger c_{\eta -k' \downarrow}
c_{\eta k' \uparrow},$$ where the single-particle energies $\epsilon_{\eta k\sigma}$ are spin dependent in the case of ferromagnetic leads (with the quantization axis for $\sigma$ being along the magnetization direction of the left lead), and the BCS pairing-interaction strength $g_\eta$ is non-vanishing only for $\eta={\rm S}$. The lead-electron operators are $c_{\eta k \sigma}$ and $c_{\eta k \sigma}^\dagger$. We treat the superconductor on a mean-field level, which introduces the notion of fermionic quasiparticles and a Cooper-pair condensate. The coupling between the dot and the leads is taken into account by the tunneling Hamiltonians $$H_{\rm{tunn},\eta}=
V_{\eta} \sum_{k \sigma} \left( c_{\eta k \sigma}^\dagger d_\sigma +
{\rm H.c.} \right),
\label{htun}$$ where for simplicity the tunnel matrix elements $V_{\eta}$ are considered to be spin and wavevector independent.
The spin-resolved tunnel-coupling strengths are defined as $\Gamma_{\eta\sigma}=2 \pi |V_{\eta}|^2 \sum_k \delta (\omega - \epsilon_{k\sigma})=2\pi |V_\eta|^2 \rho_{\eta\sigma}$, with $\rho_{\eta\sigma}$ being the density of states of the spin species $\sigma$ in lead $\eta$, which we assume to be constant. We also define the mean level-broadening $\Gamma_\eta=\frac{1}{2}\sum_{\sigma}\Gamma_{\eta\sigma}$. The spin polarization of lead $\eta$ is defined as $p_\eta=(\rho_{\eta+}-\rho_{\eta-})/(\rho_{\eta+}+\rho_{\eta-})$, with $+(-)$ denoting the majority(minority) spins.
Method
======
We integrate out the leads’ degrees of freedom to obtain an effective description of the quantum dot, whose Hilbert space is spanned by the four basis states $|\chi\rangle \in \{ | 0 \rangle, | \uparrow \rangle,
| \downarrow \rangle, | {\rm{D}} \rangle\equiv d^{\dagger}_{\uparrow}
d^{\dagger}_{\downarrow}|0\rangle \}$, with energies $E_0$, $E_\uparrow=E_\downarrow$, and $E_{\rm{D}}$, corresponding to an empty, singly and doubly occupied dot. It is useful to define the detuning $\delta=E_{\rm{D}}-E_0=2\epsilon+U$. The dot is described by its reduced density matrix, $\rho_{\rm{red}}$, whose matrix elements are $P_{\chi_2}^{\chi_1} \equiv\langle \chi_1|\rho_{\rm{red}}
|\chi_2\rangle$. The superconducting proximity effect induces a finite pair amplitude on the dot, expressed by the off-diagonal matrix element $P_{0}^{\rm D} = (P_{\rm D}^{0})^*$.
In the stationary limit, the elements of the reduced density matrix obey the generalized master equation $$i(E_{\chi_1}-E_{\chi_2}) P^{\chi_1}_{\chi_2}=\sum_{\chi_1' \chi_2'} W^{\chi_1 \chi_1'}_{\chi_2 \chi_2'}P^{\chi_1'}_{\chi_2'} \,$$ with generalized rates $ W^{\chi_1 \chi_1'}_{\chi_2 \chi_2'}$, that can be computed by means of a real-time diagrammatic technique.[@koenig96; @koenig96_2; @koenig03; @braun04; @pala07; @governale08] The stationary current out of lead $\eta$ can be expressed as $$J_{\eta} = -\frac{e}{\hbar} \sum_{\chi \chi_1' \chi_2'}
W_{\chi \chi_2'}^{\chi \chi_1' \eta}
P_{\chi_2'}^{\chi_1'},
\label{current3}$$ where $W_{\chi \chi_2'}^{\chi \chi_1' \eta} \equiv
\sum_s s W_{\chi \chi_2'}^{\chi \chi_1' s \eta}$, and $W_{\chi \chi_2'}^{\chi \chi_1' s \eta}$ is the sum of all generalized rates that describe transitions in which in total $s$ electrons are removed from lead $\eta$.
The diagrammatic rules to compute the diagrams contributing to the rates in the presence of superconducting and ferromagnetic leads are given in Refs. and , respectively. In the following, we assume small and equal tunnel-coupling strengths $\Gamma_{\rm{L}} =\Gamma_{\rm{R}} \equiv \Gamma_{\rm{N}} < T$ to the left and right lead, which we keep up to first order in the calculation of the current. The chemical potential of the superconductor is chosen as the reference for the energies, i.e. $\mu_{\rm{S}}=0$. To study subgap transport, we consider the limit of a large superconducting order parameter $\Delta \rightarrow \infty$, which we choose to be real. In this case, all orders in the tunnel-coupling strength with the superconductor $\Gamma_{\rm{S}}$ can be resummed exactly.[@governale08; @rozhkov00] In the absence of a superconducting lead, the excitation energies of the dot, $\epsilon$ and $\epsilon+U$, are split by the charging energy $U$. Due to the tunnel coupling to a superconductor the particle and hole sector of the Hilbert space are mixed, leading to four Andreev bound-state energies, defined as poles of the retarded Green’s function of the dot for $\Gamma_{\rm N} =0$: $$\label{ABS}
E_{\rm{A}, \gamma', \gamma} = \gamma' \frac{U}{2} + \gamma \sqrt{\left( \epsilon + \frac{U}{2} \right)^{2} + \frac{\Gamma_{\rm S}^{2}}{4}},$$ with $\gamma', \gamma = \pm 1$ (see Fig. \[boundstates\]). In the following, we also make use of the definition $\epsilon_{\rm A} \equiv \sqrt{(\epsilon + \frac{U}{2})^{2} + \frac{\Gamma_{\rm S}^{2}}{4}}$. The Andreev transport is largest for small detuning $|\delta|$. We consider the regime $|\delta| < \sqrt{U^2 - \Gamma_{\rm{S}}^2}$, for which the inequality $E_{\rm{A},+,+} > E_{\rm{A},+,-} > 0 > E_{\rm{A},-,+} > E_{\rm{A},-,-}$ holds.
![ Andreev bound-state energies as a function of the level position $\epsilon$, with $\Gamma_{\rm {S}}/U=0.2$ and $\Gamma_{\rm{N}}=0$. \[boundstates\]](fig2.eps){width="3.1in"}
The Andreev channel supporting transport from a normal lead to a superconductor through a quantum dot can be switched between different states by an applied bias voltage $\mu_{\rm{N}}$.[@governale08] In the following we consider $\mu_{\rm{N}} > 0$ (the case $\mu_{\rm{N}} < 0$ is obtained from the symmetry transformation $\mu_{\rm{N}} \rightarrow -\mu_{\rm{N}}$, $\delta \rightarrow - \delta$, and $J_{\rm{N}} \rightarrow -J_{\rm{N}}$). The different states are characterized by how they influence and probe the state of the quantum dot.
For small bias voltage, $E_{\rm{A},+,-} > \mu_{\rm{N}}$, all rates (to first order in $\Gamma_{\rm{N}}$) involving an electron transfer from or to the normal lead are not affected by the superconductor. The current can be written in the simple form $J_{\rm{N}} = \frac{2e}{\hbar} \Gamma_{\rm{N}} \left( 1 + Q/e \right)$, independent of the pair amplitude $P_{0}^{\rm D}$, where $Q/(-e) = \langle \sum_{\sigma} d_{\sigma}^{\dagger}d_{\sigma} \rangle =P_{\uparrow}+P_{\downarrow}+2P_{\rm{D}}$ (with $-2e \le Q \le 0$) is the average quantum-dot charge. The latter can be affected by the proximity effect in the quantum dot. In the stationary limit, the dot is singly occupied, $Q=-e$, and the Andreev channel is Coulomb blocked, $J_{\rm{N}} = 0$. A current can only flow for $Q \neq -e$, which. e.g., could be achieved by attaching a voltage-biased third lead. For large bias voltage, $\mu_{\rm{N}} > E_{\rm{A},+,+}$, the Andreev channel is also independent of the dot pair amplitude, with $J_{\rm{N}} = \frac{2e}{\hbar} \Gamma_{\rm{N}} \left( 1 + Q/2e \right)$. In both cases, the sub-gap transport involves Andreev processes at the interface between quantum dot and the superconducting lead only. As a consequence, the current is, in both cases, insensitive to the sign of the detuning $\delta$. This is different for the regime of intermediate voltages, $E_{\rm{A},+,+} > \mu_{\rm{N}} > E_{\rm{A},+,-}$. In this case the rates involving an electron transfer to or from the normal lead do depend on $\delta$ and $\Gamma_{\rm{S}}$, the rates $W_{0 \sigma}^{\rm{D} \sigma}$ and $W_{\rm{D} \sigma}^{0 \sigma}$ describing proximization of the quantum dot are non-vanishing, and the current in the normal lead also depends on the pair amplitude $P_{0}^{\rm D}$. An interesting feature of this regime is that a positive detuning drives the dot to an average occupation of less than one electron, thus overcompensating the effect of the finite bias voltage.[@governale08]
Crossed Andreev Reflection
==========================
We first consider the case of ferromagnetic leads with equal polarization strengths $|p_{\rm{L}}|=|p_{\rm{R}}|=p$, kept at the same chemical potential $\mu_{\rm{N}}$ and characterized by the same Fermi distribution $f_{\rm{N}}(\epsilon)=\left[ 1 + \exp{(\epsilon - \mu_{\rm{N}})/k_{\rm{B}}T} \right]^{-1}$. Crossed Andreev transport is identified by its dependence on the relative orientation of the ferromagnets, quantified by the Tunneling Magneto Resistance (TMR) ${\rm{TMR}}\equiv (J_{\rm{S}}^{\rm{AP}}-J_{\rm{S}}^{\rm{P}})/ J_{\rm{S}}^{\rm{AP}}$, where $J_{\rm{S}}^{\rm{P(AP)}} = (2 e\Gamma_{\rm{S}}/\hbar) \rm{Im} P_{0}^{\rm{D}}$ is the current in the superconductor for parallel (antiparallel) alignment of the magnetizations. The pair amplitude $P_{0}^{\rm{D}}=\langle d_{\downarrow} d_{\uparrow} \rangle$ is calculated in the stationary limit from the master equation $P_{0}^{\rm{D}} = \sum_{\chi} W_{0\chi}^{\rm{D}\chi} P_{\chi}^{\chi} / \left[ W_{00}^{\rm{DD}} - i(2\epsilon + U) \right]$. In the inset of Fig. \[car1\] we show the current in the superconducting lead for both the parallel and antiparallel alignment of the magnetizations as a function of the gate voltage for $\mu_{\rm{N}}=U$ (large-bias regime). The current shows a peak around zero detuning, $\epsilon=-U/2$, with a width given by $\Gamma_{\rm{S}}$. The sub-gap current for the parallel alignment is clearly suppressed, indicating the presence of CAR.
![ TMR as a function of the polarization $p$, for different values of the detuning $\delta=2\epsilon+U$. Inset: Current $J_{\rm{S}}$ in the superconducting lead as a function of $\epsilon$, for polarization $p=0.5$ and parallel and antiparallel alignment of the leads’ magnetizations. For both graphs, we chose $\mu_{\rm{L}}=\mu_{\rm{R}}=\mu_{\rm{N}}=U$, $\mu_{\rm{S}}=0$, $\Gamma_{\rm{S}}=0.2U$, $\Gamma_{\rm{L}}=\Gamma_{\rm{R}}=0.001 U$, and $k_{\rm{B}}T=0.01 U$. \[car1\]](fig3.eps){width="3.1in"}
The TMR is plotted in Fig. \[car1\] as a function of the polarization strength $p$ for different values of the detuning $\delta$. For small values of $p$ the TMR exhibits a quadratic dependence on $p$, $${\rm{TMR}}\approx \frac{\left[1-\Pi_{\gamma=\pm}\left(\sum_{\gamma'=\pm}\gamma' f_{\rm{N}} (E_{\rm{A},\gamma',\gamma})\right)\right] \Gamma_{\rm{S}}^2 p^2} {4\epsilon_{\rm{A}}^2-
\left[\sum_{\gamma,\gamma'=\pm} \gamma \gamma'(\epsilon+\frac{U}{2}+\epsilon_{\rm{A}}) f_{\rm{N}}(E_{\rm{A},\gamma',\gamma})\right]^2},$$ which simplifies in the regime of large bias voltage to ${\rm{TMR}} \approx p^2$ for $|\delta| \ll \Gamma_{\rm{S}}$ and ${\rm{TMR}} \approx p^2 \Gamma_{\rm{S}}^2 / \delta^2$ for $|\delta| \gg \Gamma_{\rm{S}}$, and in the regime of intermediate (positive) bias voltage to ${\rm{TMR}} \approx 4p^2/3$ for $|\delta| \ll \Gamma_{\rm{S}}$, ${\rm{TMR}} \approx p^2 \Gamma_{\rm{S}}^2/\delta^2$ for $-\delta \gg \Gamma_{\rm{S}}$, and ${\rm{TMR}} \approx 2p^2$ for $\delta \gg \Gamma_{\rm{S}}$.
![(color online) Current $J_{\rm{R}}$ in the right lead as a function of the chemical potential $\mu_{\rm{L}}$ of the left lead for different values of the detuning $\delta$ and $\Gamma_{\rm{S}}=0.2 U$, $\Gamma_{\rm{L}}=\Gamma_{\rm{R}}=0.001U$, $p_{\rm{L}}=p_{\rm{R}}=0$, $k_{\rm{B}}T = 0.01 U$, and (a) $\mu_{\rm{R}}=0$ (small bias), (b) $\mu_{\rm{R}}=0.5 U$ (intermediate bias), and (c) $\mu_{\rm{R}}=U$ (large bias). \[rev1\]](fig4a.eps "fig:"){width="3.1in"}\
![(color online) Current $J_{\rm{R}}$ in the right lead as a function of the chemical potential $\mu_{\rm{L}}$ of the left lead for different values of the detuning $\delta$ and $\Gamma_{\rm{S}}=0.2 U$, $\Gamma_{\rm{L}}=\Gamma_{\rm{R}}=0.001U$, $p_{\rm{L}}=p_{\rm{R}}=0$, $k_{\rm{B}}T = 0.01 U$, and (a) $\mu_{\rm{R}}=0$ (small bias), (b) $\mu_{\rm{R}}=0.5 U$ (intermediate bias), and (c) $\mu_{\rm{R}}=U$ (large bias). \[rev1\]](fig4b.eps "fig:"){width="3.1in"}\
![(color online) Current $J_{\rm{R}}$ in the right lead as a function of the chemical potential $\mu_{\rm{L}}$ of the left lead for different values of the detuning $\delta$ and $\Gamma_{\rm{S}}=0.2 U$, $\Gamma_{\rm{L}}=\Gamma_{\rm{R}}=0.001U$, $p_{\rm{L}}=p_{\rm{R}}=0$, $k_{\rm{B}}T = 0.01 U$, and (a) $\mu_{\rm{R}}=0$ (small bias), (b) $\mu_{\rm{R}}=0.5 U$ (intermediate bias), and (c) $\mu_{\rm{R}}=U$ (large bias). \[rev1\]](fig4c.eps "fig:"){width="3.1in"}
Negative non-local conductance
==============================
We now look for non-local effects in Andreev transport for non-magnetic leads, $p_{\rm{L}}=p_{\rm{R}}=0$, by studying the current in the right lead, $J_{\rm{R}}$, as a function of the voltage applied to the left lead $\mu_{\rm{L}}$. For a three-terminal device, we define a non-local conductance, $G^{\rm nl} \equiv J_{\rm{R}} / (\mu_{\rm{S}} - \mu_{\rm{L}})$, as the current response in the right lead to a voltage bias between superconductor and left lead. In particular, we will consider the non-local differential conductance $G^{\rm nl}_{\rm diff} \equiv -\partial J_{\rm{R}} / \partial \mu_{\rm{L}}$. Direct transport between the two normal leads contributes with a positive sign to $G^{\rm nl}_{\rm diff}$. Non-local transport channels such as CAR may contribute with a negative sign. For a large class of such three-terminal devices, however, it has been shown[@morten07] that the sum of all contributions to $G^{\rm nl}_{\rm diff}$ remains positive. In contrast, we find for our system regimes with negative values of $G^{\rm nl}_{\rm diff}$ and, even more striking, negative values of $G^{\rm nl}$.
In Fig. \[rev1\]a, b, and c, we show the current in the right lead as a function of the electrochemical potential in the left lead for the three cases of a small, intermediate and large voltage $\mu_{\rm{R}}$ applied to the current probe, respectively. Several features of these current-voltage characteristics indicate the strong coupling of the quantum dot to a superconducting lead. First, there are four steps (instead of two) associated with the four Andreev bound-state energies. Second, the height of many of the plateaus is sensitive to the detuning $\delta$. In the cases displayed in Figs. \[rev1\]a and c, the plateau height of the current, given by $J_{\rm{R}} = 2e \Gamma_{\rm{N}} \left( 1 + Q/e \right)$ and $J_{\rm{R}} = 2e \Gamma_{\rm{N}} \left( 1 + Q/2e \right)$, respectively, directly reflects the average quantum-dot charge, which is strongly influenced by the proximity effect.[@blatter07; @governale08] For $\delta=0$, the proximity effect is maximal, with $Q=-e$ for all values of the bias voltages, which leads to $J_{\rm{R}}=0$ and $J_{\rm{R}}=e \Gamma_{\rm{N}}$ in Fig. \[rev1\]a and c, respectively. With increasing $|\delta|$, the proximity effect decreases, and the current approaches the value expected in the absence of the superconducting lead. Third, the striking feature indicating non-local Andreev transport is the non-monotonic dependence of $J_{\rm{R}}$ on $\mu_{\rm{L}}$, i.e., the appearance of a negative non-local differential conductance, $G^{\rm nl}_{\rm diff} < 0$. Even more remarkable is for $\mu_{\rm{R}}=0$ the negative non-local conductance $G^{\rm nl} < 0$ that occurs for positive/negative detuning $\delta $ at positive/negative $\mu_{\rm{L}}$ in the intermediate-bias regime. To understand this behavior, we realize that in this regime there are combined Andreev processes that involve both the interfaces from the quantum dot to the superconductor and the left lead, while there are no Andreev processes involving electron transfer between quantum dot and right lead. The intermediate-voltage Andreev transport between left lead and superconductor yields an average dot charge that is determined by the dot level’s position relative to the chemical potential of the superconductor rather than that of the normal lead: for positive (negative) detuning the probability of double occupation decreases (increases), and the average occupation of the dot is smaller (larger) than one. This deviation from single occupancy is probed by the right lead. Changing the sign of detuning leads to a sign change in the current measured in the right lead.
We remark that the negative non-local conductance is not due to CAR. In fact, making the normal-conducting leads ferromagnetic suppresses the negative non-local conductance, independent of whether the ferromagnets are aligned parallel or antiparallel, since a finite spin accumulation on the quantum dot reduces the dot pair amplitude. For CAR an enhanced (reduced) effect would be expected for the antiparallel (parallel) alignment. The effect that we predict is rather a consequence of combined Andreev processes between left lead and superconductor. The negative non-local conductance can only be probed because of a large charging energy that prohibits direct transport between the normal-conducting leads.
Conclusions
===========
We investigated non-local Andreev transport through an interacting quantum dot in a three-terminal setup with one superconducting and two normal-conducting leads. We considered two different biasing schemes. In the first one, the normal-conducting leads are ferromagnetic with collinear magnetizations and they are kept at the same chemical potential. The key results for this case is that CAR occurs due to the non-equilibrium proximity effect in the dot and it is characterized by a finite TMR. In the second scheme, the two normal-conducting leads are non magnetic, and the response of the current in one lead to the voltage applied to the other one is studied. In that case, non-local Andreev transport is identified by negative values of the non-local differential conductance. Even more strikingly, we find regimes with negative values of the full non-local conductance. The virtue of employing quantum dots lies, first, in the possibility to tune the Andreev channels by a gate voltage and, second, in the presence of a large charging energy which generates specific transport regimes characterized by a negative non-local conductance. Both aspects are advantageous for a clear identification of non-local Andreev transport.
We acknowledge useful discussions with W. Belzig, M. Eschrig, A. Zaikin, and financial support from DFG via SFB 491.
[99]{} A. F. Andreev. Sov. Phys. JETP **19**, 1228 (1964).
G. Falci, D. Feinberg, and F. W. J. Hekking, Europhys. Lett. **54**, 255 (2001). G.B. Lesovik, T. Martin, and G. Blatter, Eur. Phys. J. B **24**, 287 (2001). T. Yamashita, S. Takahashi, and S. Maekawa, Phys. Rev. B **68**, 174504 (2003). D. Sánchez, R. López, P. Samuelsson, and M. Büttiker, Phys. Rev. B **68**, 214501 (2003). R. Mélin, and D. Feinberg, Phys. Rev. B **70**, 174509 (2004). J. P. Morten, A. Brataas, and W. Belzig, Phys. Rev. B **74**, 214510 (2006). A. Brinkman, and A.A. Golubov, Phys. Rev. B **74**, 214512 (2006). M. S. Kalenkov, and A. D. Zaikin, Phys. Rev. B **75**, 172503 (2007). D.S. Golubev, and A.D. Zaikin, Phys. Rev. B **76**, 184510 (2007). M. S. Kalenkov, and A. D. Zaikin, Phys. Rev. B **76**, 224506 (2007).
S. I. Bozhko, V. S. Tso[ĭ]{}, and E. Yakolev, JETP Lett **36**, 153 (1982). P. A. M. Benistant, H. van Kempen, and P. Wyder, Phys. Rev. Lett **51**, 817 (1983). D. Beckmann, H. B. Weber, and H. v. Löhneysen, Phys. Rev. Lett. **93**, 197003 (2004). S. Russo, M. Kroug, T. M. Klapwijk, and A. F. Morpurgo, Phys. Rev. Lett. **95**, 027002 (2005). P. Cadden-Zimansky and V. Chandrasekhar, Phys. Rev. Lett. **97**, 237003 (2006).
M. G. Pala, M. Governale, and J. König, New. J. Phys. **9**, 278 (2007).
M. Governale, M. G. Pala, and J. König, Phys. Rev. B [**77**]{}, 134513 (2008).
R. Fazio and R. Raimondi, Phys. Rev. Lett. [**80**]{}, 2913 (1998). R. Fazio and R. Raimondi, Phys. Rev. Lett. [**82**]{}, 4950 (1999). K. Kang, Phys. Rev. B [**58**]{}, 9641 (1998). P. Schwab and R. Raimondi, Phys. Rev. B [**59**]{}, 1637 (1999). A. A. Clerk, V. Ambegaokar, and S. Hershfield, Phys. Rev. B [**61**]{}, 3555 (2000). J. C. Cuevas, A. Levy Yeyati, and A. Martín-Rodero, Phys. Rev. B [**63**]{}, 094515 (2001).
M. R. Buitelaar, T. Nussbaumer, and C. Schönenberger, Phys. Rev. Lett. **89**, 256801 (2002). J.-P. Cleuziou, W. Wernsdorfer, V. Bouchiat, T. Ondarçuhu, and M. Monthioux, Nature Nanotechnology **1**, 53 (2006). P. Jarillo-Herrero, J. A. van Dam, and L. P. Kouwenhoven, Nature **439**, 953 (2006). H. I. J[ø]{}rgensen, K. Grove-Rasmussen, T. Novotný, K. Flensberg, and P. E. Lindelof, Phys. Rev. Lett. **96**, 207003 (2006).
J. A. van Dam, Y. V. Nazarov, E. P. A. M. Bakkers, S. De Franceschi, and L.P. Kouwenhoven, Nature [**442**]{}, 667 (2006). T. Sand-Jespersen, J. Paaske, B. M. Andersen, K. Grove-Rasmussen, H. I. J[ø]{}rgensen, M. Aagesen, C. B. S[ø]{}rensen, P. E. Lindelof, K. Flensberg, and J. Nyg[å]{}rd, Phys. Rev. Lett. **99**, 126603 (2007). C. Buizert, A. Oiwa, K. Shibata, K. Hirakawa, and S. Tarucha, Phys. Rev. Lett. **99**, 136806 (2007).
J. König, H. Schoeller, and G. Schön, Phys. Rev. Lett. [**76**]{}, 1715 (1996). J. König, J. Schmid, H. Schoeller, and G. Schön, Phys. Rev. B [**54**]{} 16820 (1996).
J. König and J. Martinek, Phys. Rev. Lett. [**90**]{}, 166602 (2003). M. Braun, J. König, and J. Martinek, Phys. Rev. B [**70**]{}, 195345 (2004).
A. V. Rozhkov and D. P. Arovas, Phys. Rev. B **62**, 6687 (2000).
J. P. Morten, A. Brataas, and W. Belzig, Appl. Phys. A **89**, 609 (2007).
I. A. Sadovskyy, G. B. Lesovik, and G. Blatter, Phys. Rev. B **75**, 195334 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The metal-insulator transition temperature in CMR manganites has been altered and brought close to the room temperature by preparing La$_{0.67}$Sr$_{0.33}$MnO$_{3}$ (LSMO)/ Nd$_{0.67}$Sr$_{0.33}$MnO$_{3}$ (NSMO) multilayers with ultra thin individual layers of LSMO and NSMO. The LSMO/NSMO multilayers with ultra thin individual layers of thickness of about $10\AA$ exhibits $150\%$ magnetoresistance at $270$ K whereas LSMO/NSMO multilayers with moderate individual layer thickness of about $40\AA$ each exhibits a mere $15\%$ magnetoresistance at the same temperature. We have shown that the reduction in thickness of the individual layers leads to increased spin fluctuation which results in the enhancement of magnetoresistance.'
author:
- Soumik Mukhopadhyay
- 'I. Das'
title: 'Giant enhancement of room-temperature magnetoresistance in La$_{0.67}$Sr$_{0.33}$MnO$_{3}$/Nd$_{0.67}$Sr$_{0.33}$MnO$_{3}$ multilayers'
---
Ever since the discovery of colossal magnetoresistance (CMR) in perovskite manganites [@Helmholt], extensive research on the magneotransport properties of manganite films, multi-layers, tunnel junctions etc. has been initiated due to its potential for technological applications. The CMR manganites like La$_{0.67}$Sr$_{0.33}$MnO$_{3}$ (LSMO), La$_{0.67}$Ca$_{0.33}$MnO$_{3}$ (LCMO), Nd$_{0.67}$Sr$_{0.33}$MnO$_{3}$ (NSMO) exhibit transition from a high temperature paramagnetic insulator to a low temperature ferromagnetic metal. In the ground state, these manganites are fully spin polarized [@Viret] at the Fermi level. Encouraged by this high spin polarization, numerous efforts have been made to achieve room temperature MR using the extrinsic magnetoresistive properties such as the tunneling magnetoresistance (TMR) [@Viret] phenomenon observed in manganite tunnel junctions or polycrystalline manganites. However it has been observed that for manganite tunnel junctions, the TMR falls off rapidly with increasing temperature [@Sun] and generally vanishes even before reaching the room temperature. Recently large room temperature magnetoresistance has been achieved in magnetic tunnel junctions with MgO as the tunnel barrier and Fe or CoFe as the electrodes [@yuasa; @parkin]. However, in such cases the insulating layer MgO needs to be a highly oriented single crystal, which is very difficult from the fabrication point of view. Moreover, the possibility of the transition metal electrodes having an oxidized and amorphous under-layer cannot be ruled out. Coming back to the half-metallic CMR manganites, although such materials produce high intrinsic MR around the Curie temperature, the fact that their critical temperatures are well away from the room temperature (for LSMO $T_{c} \sim 360$ K ; for NSMO $T_{c} \sim 220$ K; for LCMO $T_{c} \sim 250$ K) comes in the way of obtaining large MR around room temperature.
We have studied the magneto-transport properties of LSMO/NSMO multi-layers where the thickness of alternate LSMO and NSMO layers have been varied. This study attempts to bring the metal-insulator transition temperature, which is usually nearly coincident with the curie temperature, closer to the room temperature so that large MR can be obtained around the room temperature. We have obtained huge enhancement in magnetoresistance near room temperature by reducing the thickness of alternate layers in LSMO/NSMO magnetic multi-layer. Four sets of samples have been prepared. Sample $1$ is a LSMO film of thickness $3000\AA$ nm; sample $2$ is a NSMO film of same thickness $3000\AA$ nm; sample $3$ is the LSMO/NSMO multilayer with alternate layer thickness $10\AA$ \[LSMO($10\AA$)/NSMO($10\AA$)\] and sample $4$ is another LSMO/NSMO multilayer with alternate layer thickness of $40\AA$ \[LSMO($40\AA$)/NSMO($40\AA$)\] and with LSMO as the top layer. Sample $4$ will be used from now on as a reference. The total thickness of all the samples is about $3000\AA$ nm each. All the samples were prepared by pulsed laser deposition under identical conditions and deposited on LaAlO$_{3}$ substrate. X-ray diffraction confirms the epitaxial nature of the samples. The magnetotransport properties were studied using standard four probe method. The magnetic field was applied parallel to the applied electric field.
For sample $3$, the resistivity curve in absence of magnetic field (Fig: \[fig:res\]) shows a distinct peak associated with metal-insulator transition at around $280$ K. Considering the $T_{C}$ of bulk LSMO and NSMO, this is a remarkable shift in transition temperature. Sample $1$ and sample $4$ show more or less the same feature with no metal-insulator transition below $300$ K whereas sample $2$ exhibits metal insulator transition at $220$ K. The absolute value of resisitivities at $3$ K in samples $1$, $2$, $3$ and $4$ are $46\, \mu\Omega$-cm, $200\, \mu\Omega$-cm, $3950\, \mu\Omega$-cm and $300\, \mu\Omega$-cm, respectively. Since sample $1$ and sample $4$ exhibit almost similar magnetotransport properties and since sample $3$ and sample $4$ are similar samples with variation in individual layer thickness only, we shall primarily compare the properties of sample $3$ with that of sample $4$.
The resistivity data was analyzed using a polynomial expansion in temperature T. We have fitted the resistivity data for sample $3$ and $4$ with the following function $$\frac{\rho(T)}{\rho(0)} = 1 + \beta T^{2} + \gamma T^{4.5}$$ In the present case, the $T^2$ term comes from thermal spin fluctuation [@Moriya] (of course electron-electron scattering can also give rise to such a term according to Fermi liquid theory [@fermi] for systems with high density of states at the fermi level). The $T^{4.5}$ temperature dependence has been predicted for electron-magnon scattering in the double exchange theory [@magnon]. But the fitting is not very satisfactory at very low temperature where the resistivity is almost temperature independent with a very minor rise with lowering of temperature below $5$ K. We also tried to fit with $T^{2.5}$ term instead of $T^{4.5}$ term to include electron-phonon scattering instead of electron magnon scattering but no improvement was observed. Henceforth we excluded the $T^{4.5}$ and $T^{2.5}$ term and the resistivity data was fitted considering the function $$\frac{\rho(T)}{\rho(0)} = 1 + \beta T^{2}$$ in the temperature range above $5$ K and below approximately $T_{MI}/2$. Most of the studies concerning transport properties of thin films of manganites have observed a dominant $T^{2}$ term at low temperature [@electron], which is attributed to electron-electron interaction. Chen et. al. [@chen] have fitted the ferromagnetic metallic region of the resistivity curve using small polaronic transport term $\sinh (C/T)$, C being a constant and a spin wave scattering term $T^{3.5}$. But in our case, the fitting is evidently poor when those terms are incorporated. To ascertain whether $T^{2}$ term is due to the predominance of spin fluctuations, we fitted the resistivity data for sample $3$ and $4$ with the above polynomial at different magnetic fields and studied the variation in $\beta$. The coefficient $\beta$ decreases substantially with increasing magnetic field (Fig: \[fig:coeff\]). This suggests the suppression of spin fluctuation by applied magnetic field. The observed decrease in $\beta$ with increasing magnetic field is more pronounced for sample $3$ in comparison with sample $4$. This indicates that the reduction in thickness of individual layers has resulted in enhanced spin fluctuation in sample $3$ compared to sample $4$.
The magnetic field dependence of magnetoresistance (MR) \[$\{\rho(H)-\rho(0)\}/\rho(H) (\%)$\] for sample $3$ and $4$ at $3$ K shows contrasting behavior in the low and high magnetic field region (Fig: \[fig:mrh\]: Inset). In the low magnetic field region, a sharp drop in resistivity with increasing magnetic field, associated with the suppression of domain wall scattering, is observed for sample $4$ whereas sample $3$ shows no low field magnetoresistance. In the high field region the MR for sample $4$ almost saturates, whereas the MR for sample $3$ exhibits linear magnetic field dependence. However, near the metal insulator transition temperature, enhancement of low field MR has been observed for sample $3$ as compared to sample $4$. About $8\%$ MR around $270$ K has been observed for sample $3$ in a magnetic field of $5$ kOe compared to a mere $1\%$ for sample $4$. The comparison of the temperature dependence of high field MR between sample $1$, $3$ and $4$ is shown in Fig: \[fig:mrt\]. Enhancement of high field MR for sample $3$ as compared to sample $1$ and $4$ is observed over the entire temperature range. For sample $3$, the MR peaks at $270$ K and then decreases with increasing temperature whereas the MR for sample $1$ and $4$ weakly increases up to $300$ K, indicating that both the samples are still well below the ferromagnetic transition temperature. In contrast to about $15\%$ MR at $270$ K for samples $1$ and $4$, the MR for sample $3$ at $270$ K is about $150\%$ in $80$ kOe magnetic field. The MR peak at $270$ K suggests that the curie temperature is very close to the metal-insulator transition temperature for sample $3$ and that the magnetoresistive properties exhibited by sample $3$ is purely intrinsic in nature. The temperature dependence of magnetization for sample $3$ shows that indeed the curie temperature is at around $270$ K (Fig: \[fig:mrt\],inset). The separate transition temperatures at about $210$ K and $350$ K for the NSMO and LSMO layers are evident from the temperature dependence of magnetization for sample $4$ (Fig: \[fig:mrt\],inset). At $300$ K, the MR for sample $3$ is about $75\%$ whereas it is only about $25\%$ for samples $1$ and $4$. We have also studied the magnetic field dependence of MR for sample $3$ and $4$ (Fig: \[fig:mrh\]) at $300$ K. While the MR of sample $4$ shows linear magnetic field dependence, the MR for sample $3$ shows distinct $-H^{2}$ dependence up to about 30kOe. This suggests that the origin of enhanced MR in sample $3$ is the increased spin fluctuation due to reduction in thickness of individual layers and the eventual suppression of spin fluctuation by external magnetic field as already enunciated at the beginning of the discussion.
To summarize, we have compared the magnetotransport properties between the two LSMO/NSMO multi-layers : LSMO($10\AA$)/NSMO($10\AA$) and LSMO($40\AA$)/NSMO($40\AA$), fabricated by us. The magnetotransport properties observed in both the multi-layers do not mimic the extrinsic inter-granular transport properties shown by ferromagnetic metal-insulator composites but are rather intrinsic in nature. Analyzing the resistivity data in the presence as well as in absence of magnetic field, we conclude that the reduction in thickness of the individual layers leads to increased spin fluctuation in LSMO($10\AA$)/NSMO($10\AA$) and the enhanced magnetoresistance is a consequence of suppression of spin fluctuation by applied magnetic field. Although no extrinsic magnetoresistive properties due to scattering by domain walls or spin polarized tunneling across grain boundaries has been observed, still enhancement of low field MR has also been achieved over a wide temperature range near room temperature for sample $3$. At $270$ K and $80$ kOe magnetic field, LSMO($10\AA$)/NSMO($10\AA$) shows $150\%$ magnetoresistance which becomes $75\%$ at $300$ K. The enhancement of high field magnetoresistance has been observed in LSMO($10\AA$)/NSMO($10\AA$) over the entire temperature range compared to LSMO($40\AA$)/NSMO($40\AA$) multi-layer and LSMO film. The results suggest that it is possible to achieve much higher MR around room temperature by tuning the individual layer thickness and exploring suitable materials.
[99]{}
R. von Helmolt, J. Wecker, B. Holzapfel, L. Schultz, and K. Samwer, Phys. Rev. Lett. **71**, 2331 (1993) M. Viret , M. Drouet , J. Nassar , J. P. Contour , C. Fermon and A. Fert, Europhys. Lett. **39**, 545 (1997) J. Z. Sun, D. W. Abraham, K. Roche, S. S. P. Parkin, Appl. Phys. Lett. **73**, 1008 (1998) Shinji Yuasa, Taro Nagahama, Akio Fukushima, Yoshishige Suzuki, and Koji Ando, Nature Materials **3**, 868 (2004) Stuart S. S. Parkin, Christian Kaiser, Alex Panchula, Phillip M. Rice, Brian Hughes, Mahesh Samant and See-Hun Yang, Nature Materials **3**, 862 (2004) Moriya T, Jour. Magn. Magn. Mater. **31-34**, 10 (1983) L. D. Landau, Sov. Phys. JETP **3**, 920 (1957) K. Kubo and N. Ohata J. Phys. Soc. Jpn. **33**, 21 (1972) G. Jeffrey Snyder, Ron Hiskes, Steve DiCarolis, M. R. Beasley, and T. H. Geballe Phys. Rev. B **53**, 14434 (1996) X. J. chen, H.-U. Habermeier, C. L. Zhang, and C. C. Almasan Phys. Rev. B **67**, 134405 (2003)
$~~~~~~~~~~~~$
$~~~~~~~~~~~~~~~~~~~~~~~~$
$~~~~~~~~~~~~~~~$
$~~~~~~~~~~~~~~~$
$~~~~~~~~~~~~~~~$
| {
"pile_set_name": "ArXiv"
} |
The recently discovered alkali-fullerides AC$_{60}$[@dec] (A=K, Rb, Cs) exhibit a phase transition from a high temperature cubic phase [@trans] to an orthorhombic one in which the molecules form one-dimensional polymer chains[@chauvet; @chainpekk; @stephens; @fox], at about $350K$. Despite extensive studies, the physical properties of the latter phase are still the subject of controversy. They have been investigated by ESR[@chauvet; @bommeli; @auban; @janossy], $\mu $SR[@muon] , NMR[@auban; @alloul; @brouet; @rachdi], and optical and electrical conductivity measurements[@bommeli; @hone]. RbC$_{60}$ and CsC$_{60}$ possess a magnetic transition towards an insulating phase below $\sim 50K$, whereas KC$
_{60}$ does not. The exact nature of the magnetic low temperature phase is not yet understood; several scenarios for quasi-one-dimensional [@chauvet; @chainpekk; @bommeli; @janossy; @brouet] or three dimensional [@auban; @erwin; @kuzmany] magnetic ordering are being debated. The different behavior of KC$_{60}$ relative to RbC$_{60}$ and CsC$_{60}$ is also not understood. Recent theoretical calculations show that the chain orientations influence the dimensionality of the electronic properties[@tanaka]. However, powder diffraction studies[@stephens] revealed no definite structural difference between KC$_{60}$ and RbC$_{60}$. Accordingly, a better knowledge of the details of the AC$_{60}$ structures, especially the chain orientations, is needed. We present the first * *single crystal ** diffraction study of KC$_{60}$ and RbC$_{60}$, and we show that KC$_{60}$ and RbC$_{60}$ possess different relative chain orientations.
The main structural results[@chauvet; @stephens; @fox] obtained for KC$_{60}$ and RbC$_{60}$ are summarized in the following. The unit cell is orthorhombic, with parameters $a$, $b$ and $c$ equal to $9.11,9.95,14.32$ Å and to $9.14,10.11,14.23$Å for KC$_{60}$ and RbC$_{60}$, respectively [@stephens]. C$_{60}$ molecules are centered at $(0,0,0)$ and $(1/2,1/2,1/2)$ positions, and alkali ions at $(0,0,1/2)$ and $ (1/2,1/2,0)$. Polymerization occurs via \[2+2\] cycloaddition along the shortest parameter $\bf{a}$ (fig.1(a)). The most plausible orthorhombic space groups compatible with the molecular symmetry are Immm
0.2cm
and Pmnn. The orientation of a C$_{60}$ chain about its axis $\bf{a}$ can be characterized by the angle $\mu $ of the planes of cycloaddition with $\bf{c}$. For the body-centered space group Immm, two configurations must be considered: an ordered one, with $\mu =0$ or $90{^{\circ }} $ (fig.1(b)) where chains passing through the origin and the center of the unit cell have the same orientation, and a disordered one if $\mu \neq 0$ or $90{
{}^{\circ }}$: the mirror planes perpendicular to $b$ and $c$ constrain the chains to take orientations $\mu $ or $-\mu $, with equal probabilities (fig.1(c)). Such a disorder would give rise to diffuse planes perpendicular to $\bf{a}^{*}$ in reciprocal space. The Pmnn structure has glide planes, so that if the orientation of the chain passing through the unit cell origin is $\mu $, the orientation of that passing through its center is $-\mu $ (fig.1(d)). In diffraction, Immm can be distinguished from Pmnn by the extinction of the reflections for $h+k+l$ odd. From Rietveld refinements, Stephens et al.[@stephens] found $\mu =45\pm 5{ {}^{\circ }}$ for both KC$_{60}$ and RbC$_{60}$ samples, but they could not discriminate between Immm and Pmnn. A pair distribution function analysis performed by Fox et al.[@fox] indicated a possible orientational chain disorder.
For the present study, KC$_{60}$ and RbC$_{60}$ crystals (typical size: $10^{-2}mm^{3}$) were prepared by stoichiometric doping of C$_{60}$ single crystals at $400{
^{\circ }}C$; the polymer phase was obtained by subsequent slow cooling (the detailed procedure is described in ref.[@hone]). The samples were first characterized by X-ray powder diffraction[@hone](b) and electron beam analysis[@chopra]; their electrical transport properties were reported in ref.[@hone]. The sample crystallinity suffered from the insertion and polymerization process, but it was still acceptable for our diffraction studies (mosaic spread $\approx 2{{}^{\circ }}$, full-width at half-maximum, for KC$_{60}$ and $\approx 2.5{{}^{\circ }}$ for RbC$_{60}$).
The single-crystal X-ray experiments combined photographic (precession and fixed-crystal) and diffractometer techniques. The precession method, which gives undistorted sections of the reciprocal space, enabled us to check the body-centered extinctions mentioned above and to study the domain structure of the crystal. The latter information cannot be accessed by powder diffraction techniques. The fixed-crystal fixed-film monochromatic technique was employed to detect the X-ray diffuse scattering possibly due to orientational disorder of the C$_{60}$ chains. This technique, where the crystal is under vacuum, is particularly efficient as it maximizes the signal/noise ratio. Finally, the diffractometer technique enabled us to make quantitative measurements of the Bragg peak intensities used for structure refinements.
Precession photographs (CuK$\alpha $ radiation) have been taken on different crystals to ascertain general results. Complex precession patterns similar to those presented for pressure polymerized C$ _{60}$ in ref.[@moret] were obtained. They show the coexistence of orientational variants due to the cubic-orthorhombic symmetry lowering.These variants are related by the lost symmetry operations. We have determined the orientational relationships between the variants, which gives information regarding the structural polymerization mechanism. In KC$_{60}$, the polymerization involves the sliding of dense $(111)_{c}$ cubic planes whose orientation is preserved, as in pressure polymerized C$_{60}$[@moret]. The situation is somewhat different for RbC$_{60}$ because the orientation of $(111)_{c}$ planes is not preserved. The structural polymerization mechanism observed for pressure polymerized C$_{60}$ and for KC$_{60}$ does not apply to RbC$_{60}$ possibly due to steric constraints imposed by the larger Rb ions. The symmetry elements preserved by the cubic-orthorhombic transformation are (i) in KC$ _{60}$ the $\left[ 110\right] _{c}=\bf{b}$ 2-fold axis and (ii) in RbC$_{60}$ the $\left[ 1\overline{1}0\right] _{c}=\bf{a}$ and the $\left[ 110\right] _{c}=\bf{b}$ 2-fold axes. This transformation generates $12$ variants for KC$_{60}$ and only $6$ for RbC$ _{60}$. A careful analysis of the precession patterns reveals that KC$_{60}$ presents a primitive (P) lattice while RbC$_{60}$ is body-centered (I, absence of $h+k+l=2n+1$ reflections).
Experiments using the fixed-crystal fixed-film method were performed for KC$_{60}$ and RbC$_{60}$. They revealed no diffuse scattering. Calculations showed that the diffuse scattering intensity expected for chain disorder should have the same order of magnitude as that produced by the rotating molecules in pure C$_{60}$, which can be easily detected by the fixed-crystal fixed-film method. We thus conclude that the polymer chains are ordered in both KC$_{60}$ and RbC$_{60}$.
The Bragg peak intensity measurements on KC$_{60}$ and RbC$_{60}$ were performed on a three-circle diffractometer, using CuK$\alpha $ radiation. AC$_{60} $ crystals often present $\left\{ 111\right\} _{c}$ twins originating from the parent C$_{60}$ crystals and we selected samples with negligible twin volumes ($\leq 1\%$). The unit cell parameters are found equal to those in ref. [@stephens] within experimental accuracy. In order to refine the structure, we had to measure diffraction peaks from a single domain. This task required a selection procedure to exclude overlapping reflections. First we computed the Bragg peak positions for all variants and selected isolated reflections, then we scanned these reflections towards neighboring ones (typically within $0.8$ Å$^{-1}$), to ensure the absence of contamination. The remaining Bragg peaks were fitted using gaussian profiles, yielding peak intensities $I_{0}\left( hkl\right) $. We obtained the following data set of unique and isolated reflections: $177$ reflections, among which $107$ reflections with $h+k+l=2n+1$ and $70$ reflections with $h+k+l=2n$, for KC$_{60}$, $82$ reflections with $h+k+l=2n$ for RbC$ _{60}$. Their intensities are relatively weak: only $111$ peaks for KC$_{60}$ ( $63$ peaks with $h+k+l=2n+1$ and $48$ peaks with $h+k+l=2n$) and $39$ peaks for RbC$_{60}$ verify the relation $I>\sigma $[@stout]. We also checked i) the $h+k+l=2n+1$ extinctions for RbC$_{60}$, ii) the glide plane extinctions in KC$_{60}$.
The structural analysis is based on the minimization of the reliability factor [@stout] $$R=\sum
\left| \frac{\left| F_{obs}\right| }{\sum \left|
F_{obs}\right| }- \frac{\left| F_{calc}\right| }{\sum \left|
F_{calc}\right| }\right| \label{Rdef}$$ where $F_{obs}$ and $F_{calc}$ are observed and calculated structure factors. $\left| F_{obs}\left( h,k,l\right)
\right| $ is the square-root of the integrated intensity given by $I=I_{0}\left( hkl\right)\cdot Q^{2}$ (the mosaic broadening of the reflection is considered to be proportional to $Q^{2}$), and corrected for polarization effects. The limited intensity data lead us to restrict the number of refined parameters to the chain orientation angle $\mu $, an isotropic carbon Debye-Waller (DW) parameter $U_{C}$ , anisotropic alkali DW parameters $U_{11},U_{22},U_{33}$, and the molecular distortion; we relaxed the C$_{1}$ atom positions only, which most affects the reliability factor[@stephens]. We varied these parameters simultaneously over broad ranges using a step by step procedure which required reasonably small computing times due to the limited data set. The set of parameter values that gave the lowest R ($R_{\min }$) was retained. In comparison with the usual least-squares refinement method, this procedure enables us to test all possible combinations of the parameters, but its drawback is that uncertainties are not easily evaluated.
The R values obtained by minimization over the molecular distortion and over the DW factors are plotted versus orientation angle $\mu$ in fig.2. For Pmnn KC$_{60}$, we obtain $R_{\min }\simeq 0.16$ for $ U_{C}\simeq 0.01$Å$^{2}$, $U_{11},U_{22},U_{33}\simeq 0.24,0.02,0.3$Å$
^{2}$, $d_{C_{1}-inter}\simeq d_{C_{1}-intra}\simeq 1.55$Å, and $\mu \simeq 51 {{}^{\circ }}$. The DW factor for carbon is normal, while those of potassium are unusually large; preliminary experiments as a function of temperature indicate that they probably conceal alkali ion displacements[@ulter].
0.2cm
For RbC$_{60}$ a refinement was first attempted within the Immm space group, leading to $R_{\min }\simeq 0.24$ for $\mu \simeq 48{{ {}^{\circ
}}}$(fig.2). However this implies, as indicated before, a $
+\mu /-\mu $ orientational chain disorder which is ruled out by the absence of diffuse scattering. Aside from Immm, no body-centered orthorhombic space group is convenient to describe the orientational ordering of C$_{60}$ polymer chains. We were thus forced to consider monoclinic body-centered arrangements. The space group compatible with the symmetry of C$_{60}$ chains is I2/m, with the chain axis parallel to the 2-fold axis; the corresponding chain orientations are depicted in fig.1(e). In this case, the $(\bf{b},\bf{c})$ angle is not constrained to $90{{}^{\circ
}}$. However, we have not detected a deviation of this angle value from $90{{}^{\circ }}$ (within an estimated experimental uncertainty of $0.5{{}^{\circ}}$). It may be very weak because of the relatively homogeneous distribution of the atoms on a chain around its axis (if it were fully homogeneous, the $(\bf{b},\bf{c})$ angle would be equal to $90{{}^{\circ }}$). Within the I2/m hypothesis, the reliability factor minimum $R_{\min }\simeq 0.06$ corresponds to: $U_{C}\simeq 0.01$Å$^{2}$, $U_{11},U_{33}\simeq 0.16,0.09$ Å$^{2}$[@absorb]($U_{22}$ cannot be determined because all measured $h,k,l$ peaks have small $k$ values), $d_{C_{1}-inter}\simeq 1.5$Å, $ d_{C_{1}-intra}\simeq
1.6$Å, and* *$\mu \simeq 47{{}^{\circ }}$ (fig.2)[@comment]. This model is highly attractive because the chain orientations are ordered (in agreement with the absence of diffuse scattering) and it greatly improves $R_{\min }$, as compared to Immm. The chain orientations in KC$_{60}$ and RbC$_{60}$ ($\mu \simeq
51{{}^\circ}$ and $47{{}^\circ}$) are in agreement with the results of Stephens et al. ($45\pm 5
{{}^\circ}$)[@stephens], and the molecular distortions are found to be similar in KC$_{60}$ and RbC$_{60}$[@fox]. However the different space groups obtained for KC$_{60}$ and RbC$_{60}$, namely Pmnn and I2/m, imply completely different relative orientations of the chains.
0.2cm
It is interesting to compare the structural environments in KC$_{60}$ and RbC$_{60}$. The first neighbor C$_{60}$-C$_{60}$ interchain ($8.73$Å) and intermolecular ($9.85$Å) distances are remarkably similar for both compounds while the second neighbor distance (equal to b) increases from $9.95$Å (KC$_{60}$) to $10.11$Å(RbC$_{60}$). The A-C$_{60}$ distance increases for the first neighbors ($6.74$Å for KC$_{60}$ and $6.81$Å for RbC$_{60}$) while it decreases slightly for the second neighbors ($7.16$Å for KC$_{60}$ and $7.12 $Å for RbC$_{60}$). The above distances depend on the unit cell parameters only. The chain orientation comes into play for the interatomic A-C or C-C environments. In both compounds the alkali ions roughly face carbon hexagons from the first neighbors C$_{60}$ (along $\left[ 110\right]$ and $\left[1\overline{1}0\right]$) and ’single’ C-C bonds from their second neighbors (along $\left[001\right]$)[@molec]. The influence of the different values of $\mu$ ($\mu_{K} \simeq 51{{ {}^{\circ
}}}$ and $\mu_{Rb} \simeq 47{{ {}^{\circ }}}$) is small and no clear distinction between KC$_{60}$ and RbC$_{60}$ can be identified at this point. In contrast the intermolecular C$_{60}$ environments are different. There is only one type of environment in KC$_{60}$ (along $\left[ 111\right]$ and $\left[1\overline{1}1\right]$) where a ’double’ bond approximately face a pentagon from the neighboring C$_{60}$, as shown in fig.3(a). In RbC$_{60}$ the lower space group symmetry implies that the $\left[111\right]$ and $\left[1\overline{1}1\right]$ intermolecular environments are different, as shown in fig.3(b) and (c). It is possible that alkali-C$_{60}$ interactions favor roughly the same chain orientation angle in KC$_{60}$ and RbC$_{60}$ ($\sim
45-50{{}^\circ}$), while the C$_{60}$-C$_{60}$ interactions determine the relative chain orientations and thus the space group symmetry. Preliminary calculations[@ulter] have indeed shown that the C$_{60}$-C$_{60}$ intermolecular potential varies appreciably with the chain orientations, the symmetry of their arrangement and the unit cell parameters. This should be kept in mind when analyzing pressure effects and pressure induced transitions in AC$_{60}$ compounds[@auban; @hone; @simovic].
MAS-NMR spectra of RbC$_{60}$ and CsC$_{60}$ are very similar and differ from that of KC$_{60}$[@alloul; @rachdi]. Alloul et al.[@alloul] suggested that the distribution of spin density along a chain is influenced by its neighbors. With the present results, we can now attribute the difference between the KC$_{60}$ and RbC$_{60}$ spectra to the distinct relative chain orientations in the two compounds (fig.1(d) and (e)). The similarity of the RbC$_{60}$ and CsC$_{60}$ spectra suggests that the chain orientations are likely the same in RbC$_{60}$ and CsC$_{60}$. As discussed in the introduction, the physical properties of RbC$_{60}$ and CsC$_{60}$ are very similar and differ markedly from those of KC$_{60}$. A strong correlation between physical properties and relative chain orientations can thus be inferred in polymerized AC$_{60}$. Electronic structure calculations have already been performed by Erwin et al.[@erwin] and by Tanaka et al.[@tanaka] for I2/m RbC$_{60}$. Further theoretical investigations taking into account the distinct chain orientations in AC$_{60}$ are much awaited.
J.H. and A.Z. acknowledge support from the U.S. Department of Energy under Contract No. DE-AC03-76SF00098.
J. Winter and H. Kuzmany, Solid State Comm. **84**, 935 (1992).
Q. Zhu et al., Phys. Rev. B**47**, 13948 (1993).
O. Chauvet et al., Phys. Rev. Lett. **72**, 2721 (1994).
S. Pekker et al., Solid State Comm. **90**, 349 (1994).
P.W. Stephens et al., Nature **370**, 636 (1994).
J.R. Fox et al., Chem. Phys. Lett. **249**, 195 (1996).
F. Bommeli et al., Phys. Rev. B **51**, 14794 (1995).
P. Auban-Senzier et al., J. Phys. IFrance **6**, 2181 (1996).
A. Jánossy et al., Phys. Rev.Lett. **79**, 2718 (1997).
Y.J. Uemura et al., Phys. Rev. B **52**, 6991 (1995); W.A. MacFarlane et al., Phys.Rev. B **52**, 6995 (1995); L. Cristofolini et al., J. Phys.: Condens. Matter **7**, L567 (1995).
H. Alloul et al., Phys. Rev. Lett.**76**, 2922 (1996); V. Brouet et al., in *Molecular Nanostructures*, H. Kuzmany et al. Ed., World Scientific, 1998, p.328.
V. Brouet et al., Phys. Rev. Lett. **76**, 3638 (1996).
F. Rachdi et al., Appl. Phys. A **64**, 295 (1997).
\(a) J. Hone et al., Phys. Rev. B **52**, 8700 (1995); K. Khazeni et al., Phys. Rev. B **56**, 6627 (1997); (b) K. Khazeni et al., Appl.Phys. A **64**, 263 (1997).
S.C. Erwin, G.V. Krishna and E.J. Mele, Phys. Rev. B **51**, 7345 (1995).
H. Kuzmany et al., Physica B **244**, 186 (1998).
K. Tanaka et al., Chem. Phys. Lett. **272**, 189 (1997).
N.G. Chopra, J. Hone and A. Zettl, Phys. Rev. B **53**, 8155 (1996).
R. Moret et al., Europh. Lett. **40**, 55 (1997); P. Launois et al., in *Molecular Nanostructures*, H. Kuzmany et al. Ed., World Scientific, 1998, p. 348.
G.H. Stout and L.H. Jensen, *X-ray diffraction structure determination, a practical guide,* John Wiley & Sons, Inc., 1989.
P. Launois et al., Proceedings of the 1998 International Conference on Science and Technology of Synthetic Metals, to appear in Synthetic Metals.
The DW values may be slightly underestimated since the data were not corrected for absorption.
In principle the monoclinic distortion should lead to more than 6 orientational variants. However the distortion is not observed within the resolution of the present study, and its effect can be safely neglected. We also mention that we have assumed equivalent domains with chain orientations $\mu $ and $-\mu $ to calculate the structure factors $F_{calc}$.
In the usual C$_{60}$ nomenclature, ’single’ C-C bonds fuse a hexagon and a pentagon and ’double’ C-C bonds fuse two hexagons.
B. Simovic et al., to appear in the Proceedings of the Kirchberg Conference, 1998.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a procedure for determining independently the lattice structure and the vibrational properties of the same individual nano-object. For the example of an individual single-walled carbon nanotube we demonstrate the determination of the structural indices (n,m) of the nanotube by electron diffraction and of the frequencies of vibrational modes by micro-Raman spectroscopy. The precise and independent determination of both structure and mode frequencies allows for direct and unambiguous verification of molecular dynamical calculations and of conclusions drawn from Raman-only experiments.'
author:
- 'Jannik C. Meyer'
- Matthieu Paillet
- 'Jean-Louis Sauvajol'
- 'Georg S. Duesberg'
- Siegmar Roth
bibliography:
- 'Raman.bib'
- 'TEM.bib'
- 'diverse.bib'
- 'books.bib'
- 'transport.bib'
title: 'Lattice structure and vibrational properties of the same nano-object'
---
[^1]
[^2]
Modeling the dynamics of a physical system at the atomic level requires knowledge, or assumptions, about the lattice structure or the positions of the individual atoms within the object. The properties of nanoscopic objects often depend critically on the position of each atom as finite-size and quantization effects play an important role. For carbon nanotubes, for example, the electronic, mechanical, and vibrational properties vary significantly depending on their structure. Therefore a comparison of experimental data from single objects with theoretical predictions will be directly possible if the structure of the object is known and independently determined.
A carbon nanotube can be considered as a graphene sheet rolled into a cylinder. The nanotube indices (n,m) describe the possible nanotube structures for single-walled nanotubes (SWNTs) [@IijimaSWNT1993]. Depending on the indices, the nanotube can be either metallic or semiconducting with varying band-gaps. Also the characteristic features of the Raman spectrum, like the so-called radial breathing mode (RBM), usually in the 100-300 cm$^{-1}$ frequency range, and the so-called tangential modes (TM), usually within 1500-1600 cm$^{-1}$, depend on the nanotube indices [@DresselhausAdvPhys00]. As long as modelization is used to derive the structural information from the spectroscopic data, a verification of the model itself is limited. We present an independent determination of the nanotube structure in combination with Raman spectroscopy on the same object.
The structure, that is the indices, can be determined by electron diffraction in a transmission electron microscope (TEM). However combining the electron microscopic investigation with other measurements on the same object is an experimental challenge. A few groups, including our own, have achieved high-resolution transmission electron microscopy and transport measurements on the same individual single-walled carbon nanotube [@MeyerTEMTransport04; @KasumovSWNTSupercurrent99]. A combination with electron diffraction has been achieved only for nanotubes with multiple walls [@LinkingNMandTransportDWNTwPiezo]. Transport measurements, however, probe not only the properties of the nanotube but also of the contacts, making the interpretation much more difficult.
Raman spectroscopy, on the other hand, directly yields information about the vibrational properties of the investigated object. The frequencies of the Raman-active modes of the system can be determined with high accuracy, and can be directly compared with the results of the modelizations of the vibrational dynamics of the investigated nano-object.
We show a simple [\[]{}but completely new[\]]{} procedure to create arbitrary nanostructures by electron beam lithography in such a way that access by TEM is possible. The structures, with the carbon nanotubes embedded, are created on the edge of a cleaved substrate and made free-standing with an etching process. Fig. \[cap:SEM\] shows two samples with a free-standing section and embedded carbon nanotubes. The free-standing structure can be freely designed by electron beam lithography.
Single-walled carbon nanotubes are grown by chemical vapour deposition (CVD) on highly doped silicon substrates with a 200 nm Silicon dioxide layer [@PailletCVDJPCB04]. A metal structure consisting of 3 nm Cr and 110 nm Au is created by electron beam lithography on top of the as-grown carbon nanotubes. The substrate is then cleaved through the metal grid structure. An etching process, as illustrated in Fig. \[cap:Schematic-illustration\], is used to obtain freestanding nanotubes: The sample is etched in a 30% KOH bath at 60°C for 7 hours. This removes quickly the bulk Si substrate, and slowly the oxide layer. The etch rate of the doped silicon substrate can be controlled by biasing it with respect to the bath. Since the oxide layer initially acts as a mask, the structure is undercut mainly from the side of the cleaved edge. An undercut of 10 $µ$m can be achieved, and the etching process has to be stopped just when the oxide layer is completely removed. After the etching process, the sample is transferred into deionized water, isopropanol, and acetone before a critical point drying step.
Since the substrate is no longer in the way, TEM is possible on the free-standing part of the structure on the edge of the substrate. The carbon nanotubes are held in place by the metal structure. Before the micro-Raman experiment, overview TEM images are obtained at low dose and voltage (60 kV) to obtain the position and orientation of the carbon nanotubes with respect to the metal structure (Fig. \[cap:SEM\]).
Since the metal structure is visible in the optical microscope and the overview images show the position of the nanotubes and their orientation, it is possible to carry out micro-Raman experiments on an oriented single tube. Room-temperature Raman spectra were measured using the Ar/Kr laser lines at 488 nm (2.54 eV), 514.5 nm (2.41 eV) and 647.1 nm (1.92 eV) in the back-scattering geometry on a triple substractive Jobin-Yvon T64000 spectrometer equipped with a liquid nitrogen cooled charge coupled device (CCD) detector. The instrumental resolution is 2 cm$^{-1}$. A precise positioning of the tubes under the laser spot was monitored with a piezoelectric nano-positioner. In our experimental configuration, the incident light polarization is along the SWNT axis (the Z axis), and no analysis of the polarization of the scattered light is done.
After measuring the Raman spectra, high-resolution TEM images and diffraction images of the Raman-active nanotubes are obtained. We obtain high-resolution images using a Philips CM200 microscope operated at 120 kV, and we record diffraction patterns on image plates in a Zeiss 912$\Omega$ microscope operated at 60 kV. It is operated in the Köhler illumination condition with a condensor aperture of 20 $µ$m. The demagnified image of the condensor aperture illuminates an area of 900 nm in diameter of the sample. We move the nanotube into the illuminated region. Now, only this one nanotube and a small part of the contact structure is illuminated. To exclude the contact structure from the diffraction pattern, the selected area aperture is used additionally to further limit the effective area to approximately 300 nm in diameter. The energy filter is set to a width 15 eV. The diffraction pattern is exposed onto the image plate for 5 minutes with an illumination angle of 0.1 mrad.
Fig. \[cap:RAM\]a and b show the Raman spectrum, and Fig. \[cap:TEM\]c and d show a diffraction pattern and a high-resolution TEM image obtained from the exact same carbon nanotube.
The high-resolution image proves that the investigated tube is indeed individual and not a bundle. From the high-resolution images alone, the nanotube diameter can be estimated only with limited precision. The apparent diameter depends on the defocus value, and the apparent diameter is always smaller than the actual tube diameter [@DiameterMeasurementFromTEMImg02]. A comparison with image simulations would be possible in principle for a known defocus value, however also the point of zero defocus can not be obtained with the desired precision.
From the diffraction pattern however, it is possible to unambiguously identify the nanotube structure. This nanotube is identified from the diffraction pattern alone as (27,4). By taking into account a C-C distance of 1.42 $\pm{0.02}$ Å, the diameter of the (27,4) tube is 2.29 $\pm{0.03}$nm. Its chiral angle is 6.8$^{\circ}$. Fig. \[cap:DIFF\]c shows the diffraction pattern and a simulated diffraction pattern of a (27,4) nanotube. We can rule out other indices for which the simulated patterns clearly do not match the measured one, and its diameter of 2.29 $\pm{0.03}$ nm is consistent with the high-resolution images.
Figures \[cap:RAM\]a and b show the RBM and TM ranges of the Raman spectrum measured on the SWNT using a 1.92 eV excitation ($\lambda$=647.1 nm), with the polarization in the direction of the tube axis [@DuesbergPRL00]. The (27,4) tube is semiconducting, and with regards to the resonance conditions calculated using a nearest-neighbor tight binding approach [@KatauraSynthMet99], its Raman response is expected to be enhanced at 1.7 eV excitation. In agreement with this prediction a detectable signal is only observed for the 1.92 eV excitation in our Raman experiments (available energies were 1.92 eV, 2.41 eV, and 2.54 eV). As expected for the Raman spectrum of an individual SWNT, the spectrum is featured by a single narrow RBM. For the (27,4) tube under investigation, this RBM has a center frequency of 119.5 cm$^{-1}$ (Fig. \[cap:RAM\]a). In our scattering geometry, a number of lines greater than two in the TM bunch of the Raman spectrum is predicted due to the chiral character of the (27,4) tube [@JorioPRL03]. In agreement with this, the profile of the TM bunch, displayed in Fig. \[cap:RAM\]b, is well fitted by using three main Lorentzian components located at 1593 cm$^{-1}$, 1579 cm$^{-1}$ and 1573 cm$^{-1}$. As expected for a semiconducting SWNT, no line broadening as predicted for metallic SWNTs [@DresselhausAdvPhys00] in the TM bunch is observed .
In the inset of figure \[cap:RAM\]b, the D band range is shown in an expanded scale, and no band is observed in this region. The strength of the D band is expected to be indicative of the amount of defects in the nanotube. The quality of the diffraction pattern indicates a high structural integrity of the nanotube: The structure is periodic, i.e. the indices do not change within the illuminated region and the amount of defects is low. The high-resolution TEM image provides an upper limit for the amount of amorphous carbon on the tube, since additional carbon is clearly deposited during the TEM analysis. This is in agreement with the interpretation of the D-band as an indicator of defects and amorphous carbon coating. The nanotubes in our samples are completely free-standing, separated and sufficiently clean; most environmental influences on the nanotube spectra can be excluded.
A huge number of experiments and modelization efforts were made to relate the radial breathing mode (RBM) frequency to the nanotube structure. A review and summary of various models and experiments is given in [@sreich_CNTs-concepts+prop2004]. Our experimental approach gives access to a first accurate and independent determination of a specific SWNT diameter, at the same time with its RBM frequency. For a free-standing individual SWNT, the RBM frequency $\nu$ is expected to be inversely proportional to the nanotube diameter $d$, i.e. $d=\textrm{{A}}/\nu$ [@sreich_CNTs-concepts+prop2004]. From the results obtained on the (27,4) SWNT, a proportionality constant of A=273$\pm$10 nm.cm$^{-1}$ is found. This surprisingly high value either indicates that the models are underestimating the RBM frequency (see table 8.2 in Ref.[@sreich_CNTs-concepts+prop2004]), or that this simple RBM vs. diameter relationship can not be extrapolated up to the relatively large diameter of this tube. However for a precise estimation of the RBM vs. tube diameter relationship, a diffraction analysis of many tubes with a wide distribution of diameters is required.
As a conclusion, we have obtained Raman spectra for a precisely known structure, determined by electron diffraction and high-resolution TEM imaging. A measurement of the vibrational modes for a precisely known structure provides the ultimate test for molecular-dynamics simulations. Both the micro-Raman spectroscopy and the electron microscopic investigation are done on a freely suspended object without an influence from an environment, substrate or contact. We expect that this procedure, due to the freely designable freestanding structure, can be adopted to various nano-objects or macromolecules to combine electron microscopic structural analysis with Raman spectroscopy and potentially other investigations (transport, AFM) on the same object. Also, the possibility to create arbitrary free-standing structures may facilitate novel in-situ experiments in the TEM.
The authors acknowledge financial support by the EU project CARDECOM and the BMBF project INKONAMI. We thank xlith.com for lithography services. We thank P. Poncharal, A. Zahab, C. Koch, K. Hahn, M. Kelsch, F. Phillipp and M. Rühle for support and helpful discussions.
This article has been submitted to Applied Physics Letters (http://apl.aip.org)
[^1]: These authors contributed equally to this work
[^2]: These authors contributed equally to this work
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Junjie Cao,'
- 'Yangle He,'
- 'Liangliang Shang,'
- 'Wei Su,'
- Yang Zhang
title: Natural NMSSM after LHC Run I and the Higgsino dominated dark matter scenario
---
Introduction {#sec:introduction}
============
In the supersymmetric models such as the Minimal Supersymmetric Standard Model (MSSM) [@MSSM-1; @MSSM-2] and the Next-to-Minimal Supersymmetric Standard Model (NMSSM) [@NMSSM-1; @NMSSM-2], the $Z$ boson mass is given by [@Baer:2012uy] $$\begin{aligned}
m^2_{Z}=\frac{2 (m^2_{H_d}+\Sigma_{d})- 2 (m^2_{H_u}+
\Sigma_{u})\tan^{2}\beta}{\tan^{2}\beta-1}- 2 \mu^{2},
\label{minimization}\end{aligned}$$ where $m^2_{H_d}$ and $m^2_{H_u}$ represent the weak scale soft SUSY breaking masses of the Higgs fields $H_d$ and $H_u$ respectively, $\Sigma_{d}$ and $\Sigma_{u}$ are their radiative corrections, $\mu$ is the Higgsino mass and $\tan\beta\equiv v_u/v_d$. As was shown in [@Baer:2012cf], the corrections $\Sigma_{d}$ and $\Sigma_{u}$ can be obtained from the effective Higgs potential at loop level, and in case of a large $\tan \beta$, their largest contributions arise from the Yukawa interactions of third generation squarks, which are $$\begin{aligned}
\Sigma_u &\simeq & \sum_{i=1}^2 \frac{3Y_t^2}{16\pi^2}\times m^{2}_{\tilde{t}_i}
\left( \log\frac{m^{2}_{\tilde{t}_i}}{Q^2}-1\right), \nonumber \\
\Sigma_d &\simeq & \sum_{i=1}^2 \frac{3Y_b^2}{16\pi^2}\times m^{2}_{\tilde{b}_i}
\left( \log\frac{m^{2}_{\tilde{b}_i}}{Q^2}-1\right). \nonumber
\label{rad-corr}\end{aligned}$$ In above formulae, $Q$ denotes the renormalization scale in getting the effective potential, and its optimized value is usually taken as $Q = \sqrt{m_{\tilde{t}_1} m_{\tilde{t}_2}}$ with $\tilde{t}_1$ and $\tilde{t}_2$ being the light and heavy top squarks (stop) respectively. Obviously, if the observed value of $m_Z$ is obtained without resorting to large cancelations, each term on the right hand side of Eq. (\[minimization\]) should be comparable in magnitude with $m_Z^2$, and this in return can put non-trivial constraints on the magnitudes of $\mu$ and $m_{\tilde{t}_{1,2}}$. Numerically speaking, we find that requiring the individual term to be less than $10 m_Z^2$ leads to $\mu$ and $m_{\tilde{t}_{1,2}}$ upper bounded by about 200 GeV and 1.5 TeV respectively. In history, the scenario satisfying the bounds is dubbed as Natural SUSY (NS) [@Baer:2012uy].
In the MSSM, the NS scenario is theoretically unsatisfactory due to at least three considerations. First, since the Higgsino mass $\mu$ is the only dimensionful parameter in the superpotential of the MSSM, its typical size should be of the order of the SUSY breaking scale. Given that the LHC searches for supersymmetric particles have pushed the masses of gluinos and first generation squarks up to above $1 {\rm TeV}$ [@ATLAS-Multi-jets; @CMS-Multi-jets], $\mu \lesssim 200 {\rm GeV}$ seems rather unnatural. Second, the relic density of the dark matter (DM) predicted in the NS scenario is hardly to coincide with its measured value. Explicitly speaking, it has been shown that if the DM $\tilde{\chi}_1^0$ is Higgsino-dominated [^1], its density is usually about one order smaller than its measured value [@Baer:2012uy], alternatively if it is Bino-dominated, the correct density can be achieved only in very limited parameter regions of the MSSM [@Cao:2015efs]. These features make the NS scenario disfavored by DM physics. Third, the NS scenario is further exacerbated by the uncomfortably large mass of the recently discovered Higgs particle [@ATLAS-2012; @CMS-2012]: its value $m_h \simeq 125 {\rm GeV}$ lies well beyond its tree-level upper bound $m_h \leq m_Z$, and consequently stops heavier than about $1 {\rm TeV}$ must be present to provide a large radiative correction to the mass [@MSSM-Early-1; @MSSM-Early-2; @MSSM-Early-3; @MSSM-Early-4; @MSSM-Early-5; @MSSM-Early-6]. This requirement seems in tension with the naturalness argument of Eq. (\[minimization\]). In fact, all these problems point to the direction that the NS scenario should be embedded in a more complex framework. Remarkably, we note that the NMSSM is an ideal model to alleviate these problems.
The NMSSM extends the MSSM by one gauge singlet superfield $\hat{S}$, and it is the simplest SUSY extension of the Standard Model (SM) with a scale invariant superpotential (i.e. its superpotential does not contain any dimensionful parameters) [@NMSSM-1; @NMSSM-2]. In this model, the Higgsino mass $\mu$ is dynamically generated by the vacuum expectation value of $\hat{S}$, and given that all singlet-dominated scalars are lighter than about $v \simeq 174 {\rm GeV}$, its magnitude can be naturally less than $200 {\rm GeV}$. These additional singlet-dominated scalars, on the other hand, can act as the mediator or final states of the DM annihilation [@Cao:2015loa], and consequently the NS scenario in the NMSSM with a Singlino-dominated DM can not only predict the correct relic density, but also explain the galactic center $\gamma$-ray excess [@Cao:2015loa; @Cao:2014efa; @Guo:2014gra; @Bi:2015qva]. Moreover, in the NMSSM the interaction $\lambda \hat{S} \hat{H}_u \cdot \hat{H}_d $ can lead to a positive contribution to the squared mass of the SM-like Higgs boson, and if the boson corresponds to the next-to-lightest CP-even Higgs state, its mass can be further lifted up by the singlet-doublet Higgs mixing. These enhancements make the large radiative correction of the stops unnecessary in predicting $m_h \simeq 125 {\rm GeV}$, and thus stops can be relatively light [@NMSSM-Early-1; @NMSSM-Early-2; @NMSSM-Early-3; @NMSSM-Early-4; @NMSSM-Early-5; @NMSSM-Early-6; @NMSSM-Early-7; @NMSSM-Early-8].
So far studies on the NS scenario in the NMSSM are concentrated on the assumption that $\tilde{\chi}_1^0$ is Singlino-dominated [@Cao:2015loa; @Cao:2014efa; @Guo:2014gra; @Bi:2015qva; @Singlino-NMSSM-1; @Singlino-NMSSM-2; @Singlino-NMSSM-3; @Singlino-NMSSM-4; @Singlino-NMSSM-5; @Singlino-NMSSM-6; @Singlino-NMSSM-7; @Singlino-NMSSM-8; @Singlino-NMSSM-9; @Singlino-NMSSM-10; @Singlino-NMSSM-11; @Singlino-NMSSM-12; @Singlino-NMSSM-13; @Singlino-NMSSM-14; @Singlino-NMSSM-15; @Singlino-NMSSM-16; @Singlino-NMSSM-17]. In this case, the branching ratio of the golden channel $\tilde{t}_1 \to t \tilde{\chi}_1^0$ in the LHC search for a moderately light stop is highly suppressed. Instead, $\tilde{t}_1$ mainly decays into the Higgsino-dominated $\tilde{\chi}_1^+$ and $\tilde{\chi}_{2,3}^0$ in following way [@Singlino-NMSSM-13] $$\begin{aligned}
\tilde{t}_1 \to b \tilde{\chi}_1^{+} \to b W^{+ (\ast)} \tilde{\chi}_1^0, \quad \tilde{t}_1 \to t \tilde{\chi}_{2,3}^0 \to t X^{0 (\ast)} \tilde{\chi}_1^0, \label{Singlino-LSP}\end{aligned}$$ where $X^0$ denotes either $Z$ boson or a neutral Higgs boson. These lengthened decay chains can generate softer final particles in comparison with the golden channel, and consequently weaken the LHC bounds in the stop search. This feature is also applied to other sparticle searches, and it has been viewed as an advantage of the NMSSM in circumventing the tight constraints from the LHC searches for SUSY. In this work, we consider another realization of the NS scenario where $\tilde{\chi}_1^0$ is Higgsino-dominated. In our scenario, the Higgsinos and the Singlino are degenerated in mass at $50\%$ level, and consequently they mix rather strongly to form mass eigenstates $\tilde{\chi}_{1,2,3}^0$ with $\tilde{\chi}_{1,2}^0$ and $\tilde{\chi}_3^0$ being Higgsino-dominated and Singlino-dominated respectively. Since the role of the Singlino component in $\tilde{\chi}_1^0$ is to decrease the DM annihilation rate, $\tilde{\chi}_1^0$ may achieve the relic density measured by Planck and WMAP experiments [@Planck; @WMAP] without contradicting the DM direct search experiments such as LUX [@LUX; @LUX-1]. The phenomenology of our scenario is somewhat similar to that of the popular NS scenario in the MSSM, which was proposed in [@Baer:2012uy], but our scenario has following advantages
- In the parameter regions allowed by current LHC searches for SUSY, it may have a lower fine tuning in getting the $Z$ boson mass. Meanwhile, it has broad parameter regions to predict the right relic density (see our discussions in Sect. III).
- The mass gaps of $\tilde{\chi}_1^\pm$ and $\tilde{\chi}_2^0$ from $\tilde{\chi}_1^0$ are sizable, e.g. $\Delta_{\pm} \equiv m_{\tilde{\chi}_1^\pm} - m_{\tilde{\chi}_1^0} \gtrsim 30 {\rm GeV}$ and $\Delta_{0} \equiv m_{\tilde{\chi}_2^0} - m_{\tilde{\chi}_1^0}
\gtrsim 50 {\rm GeV}$, and consequently the leptons from the decay chains $\tilde{\chi}_1^\pm \to W^{(\ast)} \tilde{\chi}_1^0 \to l \nu \tilde{\chi}_1^0$ and $\tilde{\chi}_2^0 \to Z^{(\ast)} \tilde{\chi}_1^0 \to l l \tilde{\chi}_1^0$ are usually energetic. As a result, our scenario can be tested at future LHC experiments by the process $p p \to \tilde{\chi}_1^\pm \tilde{\chi}_2^0 \to 3 l + E_{T}^{miss}$. By contrast, in the NS scenario of the MSSM the leptons are very soft and hardly detectable due to the small mass splittings: $\Delta_\pm, \Delta_0 \sim {\cal{O}} (1 {\rm GeV})$ [@NS-SUSY-Compressed].
- Since all the light particles in our scenario, i.e. $\tilde{\chi}_1^\pm$, $\tilde{\chi}_{1}^0$, $\tilde{\chi}_{2}^0$ and $\tilde{\chi}_{3}^0$, have sizable $\tilde{H}_u$ component, the main decay modes of $\tilde{t}_1$ include $$\begin{aligned}
\tilde{t}_1 \to b \tilde{\chi}_1^+ \to b W^{\ast} \tilde{\chi}_1^0, \quad \tilde{t}_1 \to t \tilde{\chi}_{1}^0, \quad \tilde{t}_1 \to t \tilde{\chi}_{2,3}^0 \to t X^{0 \ast} \tilde{\chi}_1^0, \label{Higgsino-LSP}\end{aligned}$$ and each mode corresponds to different signals. Because the signal of $\tilde{t}_1$ pair production is shared by rich final states, the LHC bounds on $m_{\tilde{t}_1}$ are usually weakened.
Moreover, we remind that the phenomenology of our scenario is different from that of the NS scenario with a Singlino-dominated DM. This can be seen for example from the decay modes of $\tilde{t}_1$, which are presented in Eq.(\[Singlino-LSP\]) and Eq.(\[Higgsino-LSP\]) respectively. We also remind that our scenario was scarcely discussed in literatures. In fact, within our knowledge only the work [@SNMSSM-Higgsino] briefly commented that $\tilde{\chi}_1^0$ may be Higgsino-dominated in the constrained NMSSM.
This work is organized as follows. In Sec. II, we briefly recapitulate the framework of the NMSSM, then we scan its parameter space by considering various constraints to get the NS scenarios in the NMSSM. Especially, we take great pains to implement the constraints from the LHC searches for SUSY by multiple packages and also by detailed Monte Carlo simulations, like what the work [@Natural-NMSSM-Simulation] did. After these preparation, we exhibit in Sec. III the features of the NS scenario with a Higgsino-dominated DM, including its favored spectrum and the properties of the neutralinos and stops, and subsequently in Sec. IV we take several benchmark points as examples to show the detection of our scenario in future experiments. Finally, we draw our conclusions in Sec.V. The details of our treatment on the LHC searches for SUSY are presented in the Appendix.
The Structure of the NMSSM and Our Scan Strategy
================================================
The Structure of the NMSSM
--------------------------
The NMSSM extends the MSSM by adding one gauge singlet superfield $\hat{S}$, and since it aims at solving the $\mu$ problem of the MSSM, a $Z_3$ discrete symmetry under which the Higgs superfields $\hat{H}_{u,d}$ and $\hat{S}$ are charged is adopted to avoid the appearance of dimensionful parameters in its superpotential. Consequently, the superpotential of the NMSSM can be written as [@NMSSM-1] $$\begin{aligned}
% \nonumber to remove numbering (before each equation)
W_{\rm NMSSM} &=& W_F + \lambda\hat{H_u} \cdot \hat{H_d} \hat{S}
+\frac{1}{3}\kappa \hat{S^3},
\end{aligned}$$ where $W_F$ is the superpotential of the MSSM without the $\mu$-term, and the dimensionless parameters $\lambda$, $\kappa$ describe the interactions among the Higgs superfields.
The Higgs potential of the NMSSM is given by the usual F-term and D-term of the superfields as well as the soft breaking terms, which are given by $$\begin{aligned}
V_{\rm NMSSM}^{\rm soft} &=& m_{H_u}^2 |H_u|^2 + m_{H_d}^2|H_d|^2
+ m_S^2|S|^2 +( \lambda A_{\lambda} SH_u\cdot H_d
+\frac{1}{3}\kappa A_{\kappa} S^3 + h.c.), \label{input-parameter1}\end{aligned}$$ with $H_u$, $H_d$ and $S$ representing the scalar component fields of $\hat{H}_u$, $\hat{H}_d$ and $\hat{S}$ respectively. Considering that the physical implication of the fields $H_u$ and $H_d$ is less clear, one usually introduces following combinations [@NMSSM-1] $$\begin{aligned}
H_1=\cos\beta H_u + \varepsilon \sin\beta H_d^*, ~~
H_2=\sin\beta H_u - \varepsilon \cos\beta H_d^*, ~~H_3 = S,\end{aligned}$$ where $\varepsilon$ is second-order antisymmetric tensor with $\varepsilon_{12}=-\varepsilon_{21}=1$ and $\varepsilon_{11}=\varepsilon_{22}=0$, and $\tan \beta \equiv v_u/v_d$ with $v_u$ and $v_d$ denoting the vacuum expectation values of $H_u$ and $H_d$ respectively. In this representation, the redefined fields $H_i$ ($i=1,2,3$) are given by $$\begin{aligned}
H_1 = \left ( \begin{array}{c} H^+ \\
\frac{S_1 + i P_1}{\sqrt{2}}
\end{array} \right),~~
H_2 & =& \left ( \begin{array}{c} G^+
\\ v + \frac{ S_2 + i G^0}{\sqrt{2}}
\end{array} \right),~~
H_3 = v_s +\frac{1}{\sqrt{2}} \left( S_3 + i P_2 \right).
\label{fields}\end{aligned}$$ These expressions indicate that the field $H_2$ corresponds to the SM Higgs doublet with $G^+$ and $G^0$ denoting the Goldston bosons eaten by $W$ and $Z$ bosons respectively, and the field $H_1$ represents a new $SU(2)_L$ doublet scalar field which has no tree-level couplings to the W/Z bosons. These expressions also indicate that the Higgs sector of the NMSSM includes three CP-even mass eigenstates $h_1$, $h_2$ and $h_3$ which are the mixtures of the fields $S_1$, $S_2$ and $S_3$, two CP-odd mass eigenstates $A_1$ and $A_2$ which are composed by the fields $P_1$ and $P_2$, as well as one charged Higgs $H^+$. In the following, we assume $m_{h_3} > m_{h_2} > m_{h_1}$ and $m_{A_2} > m_{A_1}$, and call the state $h_i$ the SM-like Higgs boson if its dominant component comes from the field $S_2$.
In the NMSSM, the squared mass of the filed $S_2$ is given by $$\begin{aligned}
m_{S_2 S_2}^2 = m_Z^2 \cos^2 2 \beta + \lambda^2 v^2 \sin^2 2 \beta, \nonumber\end{aligned}$$ where the last term on the right side is peculiar to any singlet extension of the MSSM[@NMSSM-1], and its effect is to enhance the mass of the SM-like Higgs boson in comparison with the case of the MSSM. Moreover, if the inequation $m_{S_3 S_3}^2 < m_{S_2 S_2}^2$ holds, the mixing of the field $S_2$ with the field $S_3$ in forming the SM-like Higgs boson can further enhance the mass. In this case, $h_1$ is a singlet-dominate scalar, $h_2$ acts as the SM-like Higgs boson, and due to the enhancement effects the requirement $m_{h_2} \simeq 125 {\rm GeV}$ does not necessarily need the large radiative correction of stops [@NMSSM-Early-1; @NMSSM-Early-4]. We remind that the singlet-dominated physical scalars (i.e. the mass eigenstates mainly composed by $S_3$ and $P_2$ respectively) are experimentally less constrained, and in case that they are lighter than about $200 {\rm GeV}$, $\mu = \lambda v_s$ naturally lies within the range from $100 {\rm GeV}$ to $200 {\rm GeV}$.
In practice, it is convenient to use [@NMSSM-1] $$\begin{aligned}
\lambda, \quad \kappa, \quad \tan \beta, \quad \mu, \quad M_A, \quad A_\kappa, \label{input-parameter}\end{aligned}$$ as input parameters, where $m_{H_u}^2$, $m_{H_d}^2$ and $m_S^2$ in Eq.(\[input-parameter1\]) are traded for $m_Z$, $\tan \beta$ and $\mu $ by the potential minimization conditions, and $A_\lambda$ is replaced by the squared mass of the CP-odd field $P_1$, which is given by $$\begin{aligned}
M^2_A \equiv m_{P_1 P_1}^2 = \frac{2\mu}{\sin2\beta}(A_{\lambda}+\kappa v_s).\end{aligned}$$ Note that $M_A$ represents the mass scale of the new doublet $H_1$, and it is preferred by current experiments to be larger than about $300 {\rm GeV}$.
When we discuss the naturalness of the NMSSM, we consider two fine tuning quantities defined by [@Baer:2013gva] $$\begin{aligned}
\Delta_Z = \max_i |\frac{\partial \log m_Z^2}{\partial \log p_i}|, \quad \Delta_h = \max_i |\frac{\partial \log m_h^2}{\partial \log p_i}|,\end{aligned}$$ where $h$ represents the SM-like Higgs boson, $p_i$ denotes SUSY parameters at the weak scale, and it includes the parameters listed in Eq.(\[input-parameter\]) and top quark Yukawa coupling $Y_t$ with the latter used to estimate the sensitivity to stop masses. Obviously, $\Delta_Z$ ($\Delta_h$) measures the sensitivity of $m_Z$ ($m_h$) to SUSY parameters at weak scale, and the larger its value becomes, the more tuning is needed to get the corresponding mass [^2]. In our calculation, we calculate $\Delta_Z$ and $\Delta_h$ by the formulae presented in [@Ulrich-Fine-Tuning] and [@lambd-SUSY-recent-2] respectively.
The NMSSM predicts five neutralinos, which are the mixtures of the fields Bino $\tilde{B}^0$, Wino $\tilde{W}^0$, Higgsinos $\tilde{H}_{d,u}^0$ and Singlino $\tilde{S}^0$. In the basis $\psi^0 = (-i \tilde{B}^0, - i \tilde{W}^0, \tilde{H}_{d}^0, \tilde{H}_{u}^0, \tilde{S}^0)$, the neutralino mass matrix is given by [@NMSSM-1] $${\cal M} = \left(
\begin{array}{ccccc}
M_1 & 0 & -\frac{g_1 v_d}{\sqrt{2}} & \frac{g_1 v_u}{\sqrt{2}} & 0 \\
& M_2 & \frac{g_2 v_d}{\sqrt{2}} & - \frac{g_2 v_u}{\sqrt{2}} &0 \\
& & 0 & -\mu & -\lambda v_u \\
& & & 0 & -\lambda v_d\\
& & & & \frac{2 \kappa}{\lambda} \mu
\end{array}
\right). \label{eq:MN}$$ If $|M_1|, |M_2| \gg |\mu|$ and $2 \kappa/\lambda \sim 1$, the Bino and Wino fields are decoupled from the rest fields. In this case, the remaining three light neutralinos $\tilde{\chi}_i^0$ ($i=1,2,3$) can be approximated by $$\begin{aligned}
\tilde{\chi}_i^0 & \approx & N_{i3} \tilde{H}_d^0 + N_{i4}\tilde{H}_u^0 + N_{i5} \tilde{S}^0, \label{neutralino-mass}\end{aligned}$$ where the elements of the rotation matrix $N$ roughly satisfy $$\begin{aligned}
N_{i3}:N_{i4}:N_{i5} \simeq \lambda (v_d \mu - v_u m_{\tilde{\chi}_i^0}):
\lambda (v_u \mu - v_d m_{\tilde{\chi}_i^0}): (m_{\tilde{\chi}_i^0}^2 - \mu^2) \label{neutralino-mixing}\end{aligned}$$ with $m_{\tilde{\chi}_i^0}$ denoting the mass of $\tilde{\chi}_i^0$. In the following, we are interested in the parameter region featured by $\tan \beta \sim 10$, $2 \kappa/\lambda \sim (1 - 1.5)$ and $|\mu| \sim (100-200) {\rm GeV}$, which is hereafter dubbed as the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. From Eq.(\[neutralino-mass\]) and Eq.(\[neutralino-mixing\]), one can conclude that this scenario has following characters
- The lightest two neutralinos $\tilde{\chi}_1^0$ and $\tilde{\chi}_2^0$ are Higgsino-dominated, and $\tilde{\chi}_3^0$ is Singlino-dominated. Their masses should satisfy following relations: $m_{\tilde{\chi}_1^0} < |\mu|$, $m_{\tilde{\chi}_2^0} \sim |\mu|$ and $m_{\tilde{\chi}_3^0} > |\frac{2 \kappa}{\lambda} \mu| > |\mu| $.
- As far as $\tilde{\chi}_1^0$ is concerned, its largest component comes from $\tilde{H}_{u}^0$ field. If the splitting between $m_{\tilde{\chi}_1^0}$ and $|\mu|$ is significant, its Singlino component may also be quite large. The importance of the Singlino component is that it can dilute the couplings of $ \tilde{\chi}_1^0$ with $W$ and $Z$ bosons, Higgs scalars and SM fermions, and consequently the density of $ \tilde{\chi}_1^0$ can coincide with the DM density measured by WMAP and Planck experiments.
- The $\tilde{H}_u^0$ and $\tilde{H}_d^0$ components in $ \tilde{\chi}_2^0$ should be comparable, and they are usually much larger than $\tilde{S}^0$ component of $ \tilde{\chi}_2^0$, i.e. $|N_{23}| \sim |N_{24}| \gg |N_{25}|$.
- As for $ \tilde{\chi}_3^0$, the relation $|N_{35}| > |N_{33}| > |N_{34}|$ usually holds.
Strategy in Scanning the Parameter Space of the NMSSM
-----------------------------------------------------
In this part, we perform a comprehensive scan over the parameter space of the NMSSM by considering various experimental constraints. Especially, we take great pains to implement the constraints from the direct searches for SUSY at the LHC. After this procedure, we get the NS scenario with a Higgsino-dominated $\tilde{\chi}_1^0$.
----- -------------------------- -------------------------- --------------------------- ---------------------------
$0.\thicksim0.41$ $0.\thicksim0.68$ $0.\thicksim0.75$ $0.18\thicksim 0.71$
$0.\thicksim0.41$ $0.\thicksim0.68$ $0.\thicksim0.75$ $0.18\thicksim 0.71$
$ 0.\thicksim0.66 $ $ 0.\thicksim0.52 $ $ 0.\thicksim0.27 $ $ 0.\thicksim0.51 $
$ 0.\thicksim0.66 $ $ 0.\thicksim0.52 $ $ 0.\thicksim0.27 $ $ 0.\thicksim0.51 $
$ 3\thicksim 60$ $ 4\thicksim 38 $ $ 3\thicksim 60$ $ 3\thicksim 18 $
$ 3\thicksim 60 $ $ 4\thicksim 38 $ $ 3\thicksim 60$ $ 3\thicksim 18 $
$ 150\thicksim 400 $ $ 115\thicksim 370 $ $ 105\thicksim 315 $ $ 110 \thicksim 175 $
$ 180 \thicksim 400 $ $ 115\thicksim 370 $ $ 105\thicksim 315 $ $ 110 \thicksim 165 $
$ -2000\thicksim 0$ $ -750\thicksim 0 $ $ -350 \thicksim 40 $ $ -650\thicksim -20 $
$ -2000\thicksim 0 $ $ -750 \thicksim 0 $ $ -350\thicksim 40$ $ -650\thicksim -30 $
$ 390 \thicksim 2000 $ $ 550 \thicksim 2000 $ $ 450 \thicksim 2000 $ $ 450 \thicksim 2000 $
$ 570 \thicksim 2000 $ $ 680 \thicksim 2000 $ $ 450 \thicksim 2000 $ $ 620 \thicksim 2000 $
$ 480 \thicksim 2000 $ $ 530 \thicksim 2000 $ $ 480 \thicksim 2000 $ $490 \thicksim 2000 $
$ 610 \thicksim 2000 $ $ 680\thicksim 2000 $ $ 580 \thicksim 2000 $ $ 560 \thicksim 2000$
$ 100 \thicksim 1000 $ $ 100 \thicksim 640 $ $ 100 \thicksim 730 $ $ 100 \thicksim 950 $
$ 100 \thicksim 1000 $ $ 100 \thicksim 640 $ $ 100 \thicksim 730 $ $ 100 \thicksim 950 $
$ -4700 \thicksim 5000 $ $ -4500 \thicksim 4500 $ $ -5000 \thicksim 4800 $ $ -5000\thicksim 4500 $
$ -4400 \thicksim 4850 $ $ -4500 \thicksim 4500 $ $ - 5000 \thicksim 4600 $ $ -5000 \thicksim 4500 $
$ 40 \thicksim 350$ $ 20 \thicksim 175 $ $ 45 \thicksim 500 $ $ 120 \thicksim 500 $
$ 40 \thicksim 350 $ $ 40 \thicksim 175$ $ 45 \thicksim 500 $ $ 120 \thicksim 500 $
$ 105 \thicksim 1000 $ $ 150 \thicksim 1000 $ $ 150 \thicksim 1000$ $ 160 \thicksim 1000 $
$ 105 \thicksim 1000$ $ 150 \thicksim 1000 $ $ 155 \thicksim 1000 $ $ 165 \thicksim 1000 $
$ 200 \thicksim 2000 $ $ 900 \thicksim 2000 $ $ 500 \thicksim 2000 $ $ 430 \thicksim 2000 $
$ 320 \thicksim 2000 $ $ 900 \thicksim 2000 $ $ 500 \thicksim 2000 $ $ 430 \thicksim 2000 $
$R$ $52\%$ $54\%$ $71\%$ $65\%$
----- -------------------------- -------------------------- --------------------------- ---------------------------
: Favored parameter ranges for different types of samples. In each item, the range in the first row is for the samples that survive the constraints (1), (2) and (3) presented in the text, and that in the second row corresponds to the samples that further satisfy the constraints from the direct search for sparticles at the LHC Run-I, i.e. the constraints (4) and (5). The quantity $R$ in the last row represents the retaining ratio of the samples before and after considering the direct search constraints in our scan. For Type I samples, $h_1$ corresponds to the SM-like Higgs boson and $\tilde{\chi}_1^0$ is Bino-dominated, while for Type II, III and IV samples, $h_2$ acts as the SM-like Higgs boson with $\tilde{\chi}_1^0$ being Bino-, Singlino- and Higgsino-dominated respectively.\[table-1\]
We begin our study by making following assumptions about some unimportant SUSY parameters:
- We fix all soft breaking parameters for the first two generation squarks at $2 {\rm TeV}$. Considering that the third generation squarks can affect significantly the mass of the SM-like Higgs boson, we vary freely all soft parameters in this sector except that we assume $m_{U_3} = m_{D_3}$ for right-handed soft breaking masses and $A_t = A_b$ for soft breaking trilinear coefficients.
- Considering that we require the NMSSM to explain the discrepancy of the measured value of the muon anomalous magnetic moment from its SM prediction, we treat the common value for all soft breaking parameters in the slepton sector (denoted by $m_{\tilde{l}}$ hereafter) as a free parameter.
- We fix gluino mass at $2 {\rm TeV}$, and treat the Bino mass $M_1$ and the Wino mass $M_2$ as free parameters since they affect the properties of the neutralinos.
Then we use the package NMSSMTools-4.9.0 [@NMSSMTools] to scan following parameter space: $$\begin{aligned}
\label{NMSSM-scan}
&& 0 <\lambda\leq 0.75,\quad 0 <\kappa \leq 0.75, \quad 2 \leq \tan{\beta} \leq 60,\quad 100{\rm ~GeV}\leq m_{\tilde{l}} \leq 1 {\rm ~TeV}, \nonumber \\
&& 100 {\rm GeV} \leq \mu \leq 1 {\rm TeV}, \quad 50 {\rm ~GeV}\leq M_A \leq 2 {\rm ~TeV}, \quad |A_{\kappa}| \leq 2 {\rm TeV}, \nonumber\\
&& 100{\rm ~GeV}\leq M_{Q_3},M_{U_3} \leq 2 {\rm ~TeV}, \quad |A_{t}|\leq {\rm min}(3 \sqrt{M_{Q_3}^2 + M_{U_3}^2}, 5 {\rm TeV}), \nonumber\\
&& 20 {\rm GeV} \leq M_1 \leq 500 {\rm GeV}, \quad 100 {\rm GeV} \leq M_2 \leq 1 {\rm TeV},\end{aligned}$$ where all the parameters are defined at the scale of $1 {\rm TeV}$. During the scan, we use following constraints to select physical parameter points:
- All the constraints implemented in the package NMSSMTools-4.9.0, which include the $Z$-boson invisible decay, the LEP search for sparticles (i.e. the lower bounds on various sparticle masses and the upper bounds on the chargino/neutralino pair production rates), the $B$-physics observables such as the branching ratios for $B \to X_s \gamma$ and $B_s \to \mu^+ \mu^-$, the discrepancy of the muon anomalous magnetic moment, the dark matter relic density and the LUX limits on the scattering rate of dark matter with nucleon. In getting the constraint from a certain observable which has an experimental central value, we use its latest measured result and require the NMSSM to explain the result at $2\sigma$ level.
- Constraints from the direct searches for Higgs bosons at LEP, Tevatron and LHC. Especially we require that one CP-even Higgs boson acts as the SM-like Higgs boson discovered at the LHC. We implement these constraints with the packages HiggsSignal for $125 {\rm GeV}$ Higgs data fit [@HiggsSignals] [^3] and HiggsBounds for non-standard Higgs boson search at colliders [@HiggsBounds].
- Constraints from the fine-tuning consideration: $\Delta_Z \leq 50$ and $\Delta_h \leq 50$.
- Constraints from the preliminary analyses of the ATLAS and CMS groups in their direct searches for sparticles at the LHC Run-I. We implement these constraints by the packages FastLim [@Papucci:2014rja] and SModelS [@Kraml:2013mwa]. These two packages provide cut efficiencies or upper bounds on some sparticle production processes in simplified model framework, and thus enable us to impose the direct search bounds in an easy and fast way. In the appendix, we briefly introduce the two packages.
- Constraints from the latest searches for electroweakinos and stops by the ATLAS collaboration at the LHC Run-I. We implement these constraints by detailed Monte Carlo simulation. Since we have to treat more than twenty thousand samples at this step, this process is rather time consuming in our calculation by clusters. In the appendix, we provide details of our simulation.
![$\Delta_Z$ and $\Delta_h$ predicted by the Type III samples (upper panels) and Type IV samples (lower panels) respectively. Samples in the left panels survive the constraints (1), (2) and (3) presented in the text, and those in the right panels further satisfy the constraints from the direct searches for sparticles at LHC Run I, i.e. the constraints (4) and (5). The panels in each row adopt same color convention for $\mu$, which is presented on the right side of the row. \[fig1\]](fig1-a.jpg "fig:"){height="8cm" width="16cm"}\
![$\Delta_Z$ and $\Delta_h$ predicted by the Type III samples (upper panels) and Type IV samples (lower panels) respectively. Samples in the left panels survive the constraints (1), (2) and (3) presented in the text, and those in the right panels further satisfy the constraints from the direct searches for sparticles at LHC Run I, i.e. the constraints (4) and (5). The panels in each row adopt same color convention for $\mu$, which is presented on the right side of the row. \[fig1\]](fig1-b.jpg){height="8cm" width="16cm"}
After analyzing the samples that survive the constraints, we find that they can be classified into four types: for Type I samples, $h_1$ corresponds to the SM-like Higgs boson and $\tilde{\chi}_1^0$ is Bino-dominated, while for Type II, III and IV samples, $h_2$ acts as the SM-like Higgs boson with $\tilde{\chi}_1^0$ being Bino-, Singlino- and Higgsino-dominated respectively. In Table \[table-1\], we list the favored parameter ranges for each type of samples before and after considering the constraints from the direct search experiments, i.e. the constraints (4) and (5). Note that in the last row of this table, the retaining ratio $R$ is defined by $R = N_{after}/N_{before}$ where $N_{before}$ is the number of the samples that satisfy the constraints (1), (2) and (3) in our scan, and $N_{after}$ is the number of the samples that further satisfy the constraints (4) and (5). This table indicates that although the direct search experiments are effective in excluding parameter points encountered in the scan, they scarcely change the ranges of the input parameters where $\Delta_Z$ and $\Delta_h$ take rather low values, or equivalently speaking the NMSSM can naturally predict $m_Z$ and $m_h$ even after considering the direct search constraints from LHC Run-I. The underlying reason for this phenomenology is that the exclusion capability of the direct search experiments depends not only on sparticle production rate, but also on the decay chain of the sparticle and the mass gap between the sparticle and its decay product. We checked that a large portion of the excluded samples are characterized by $M_2 \leq 250 {\rm GeV}$. In this case, there exist one moderately light neutralino and one moderately light chargino with both of them being Wino-dominated, and their associated production rate at the LHC is quite large so that the $3 l + E_{T}^{miss}$ signal of the production after cuts may exceed its experimental upper bound (see appendix for more information). Among the four types of points, we also find that the lowest fine-tuning comes from type III and type IV samples, for which $\Delta_Z$ and $\Delta_h$ may be as low as about 2. This character is shown in Fig.\[fig1\], where we project type III and IV samples on $\Delta_Z-\Delta_h$ plane. We emphasize that for the type IV samples with $\Delta_Z, \Delta_h \lesssim 10$, $\mu$ is upper bounded by about $145 {\rm GeV}$. In this case, it is compressed spectrum among the Higgsino-dominated particles $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_1^\pm$ that helps the samples evade the direct search experiments.
Because the type IV samples were scarcely studied in previous literatures and also because they have similar phenomenology to that of the NS scenario in the MSSM, we in the following focus on this type of samples. In order to make the essential features of the samples clear, we only consider those that satisfy additionally the condition $M_1, M_2, m_{\tilde{l}} \geq 300 {\rm GeV}$. Hereafter we call such samples collectively as the NS scenario with a Higgsino-dominated $\tilde{\chi}_1^0$. As we will show below, $|\mu|$ in this scenario is upper bounded by about $160 {\rm GeV}$, so the condition is equivalent to $M_1, M_2, m_{\tilde{l}} \gtrsim 2 |\mu| $. In this case, gauginos and sleptons affect little on the properties of the lightest three neutralino, and the lighter chargino.
[|c|c|c|c|c|c|]{} para & range & para & range & para & range\
& $7\thicksim 18$ & & $65\thicksim85$ & & $-400\thicksim -60$\
& $7\thicksim 18$& & $65\thicksim85$& & $-400\thicksim -60$\
& $0.15\thicksim0.49$ & & $125\thicksim195$ & & $45\thicksim 120 $\
& $0.15\thicksim0.49$& & $125\thicksim195$ & & $45\thicksim 120 $\
& $0.28\thicksim 0.68$& & $150 \thicksim260 $& & $120\thicksim 350$\
& $0.28\thicksim 0.68$& & $150 \thicksim260$ & & $120\thicksim 350$\
& $110 \thicksim 160 $ & & $105 \thicksim 150 $& & $800 \thicksim 2000 $\
& $110 \thicksim 160 $& & $105\thicksim 150$ & & $800 \thicksim 2000 $\
& $-5000\thicksim 4500 $ & & $380 \thicksim 2050 $ & & $1050 \thicksim 3100 $\
& $-5000\thicksim 4500 $ & & $500 \thicksim 2050 $ & & $1100 \thicksim 3100 $\
Key features of the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated
================================================================================
In this section, we investigate the features of the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. We are particulary interested in neutralino-chargino sector and stop sector since they play an important role in determining the fine tunings of the theory. In Table \[table-2\], we show the favored ranges of some quantities such as $\mu$ and stop masses. This table indicates that our scenario is featured by $\mu/{\rm GeV} \in [ 110, 160 ]$, $m_{\tilde{\chi}_1^0}/{\rm GeV} \in [ 65, 85 ]$, $ m_{\tilde{\chi}_2^0}/{\rm GeV} \in [ 125, 195 ]$, $ m_{\tilde{\chi}_3^0}/{\rm GeV} \in [ 150, 260 ]$, $m_{\tilde{\chi}_1^\pm}/{\rm GeV} \in [105, 150 ]$ and $m_{\tilde{t}_1}/{\rm GeV} \in [ 500, 2050 ]$, and among the ranges, only the lower bound of $m_{\tilde{t}_1}$ is shifted from $380 {\rm GeV}$ to $500 {\rm GeV}$ by the constraints from the direct search experiments. Moreover, we checked that in our scenario, the ratio $2 \kappa/\lambda$ is restricted in the range from about 1 to 1.5. In this case, Higgsinos and Singlino are approximately degenerated in mass, and consequently they mix strongly to form mass eigenstates.
![Mass spectrum of $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$, $\tilde{\chi}_3^0$ and $\tilde{\chi}_1^\pm$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that the mass splittings among these particles are induced by the strong mixings between Higgsinos and Singlino, and significantly larger than those in the NS scenario of the MSSM. \[fig2\]](fig2-a.png "fig:"){height="6cm" width="5.2cm"} ![Mass spectrum of $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$, $\tilde{\chi}_3^0$ and $\tilde{\chi}_1^\pm$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that the mass splittings among these particles are induced by the strong mixings between Higgsinos and Singlino, and significantly larger than those in the NS scenario of the MSSM. \[fig2\]](fig2-b.png "fig:"){height="6cm" width="5.2cm"} ![Mass spectrum of $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$, $\tilde{\chi}_3^0$ and $\tilde{\chi}_1^\pm$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that the mass splittings among these particles are induced by the strong mixings between Higgsinos and Singlino, and significantly larger than those in the NS scenario of the MSSM. \[fig2\]](fig2-c.png "fig:"){height="6cm" width="5.2cm"}
![Components of the physical states $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_3^0$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that these results can be understood by Eq.(\[neutralino-mixing\]). \[fig3\]](fig3-a.png "fig:"){height="6cm" width="5.2cm"} ![Components of the physical states $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_3^0$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that these results can be understood by Eq.(\[neutralino-mixing\]). \[fig3\]](fig3-b.png "fig:"){height="6cm" width="5.2cm"} ![Components of the physical states $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_3^0$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that these results can be understood by Eq.(\[neutralino-mixing\]). \[fig3\]](fig3-c.png "fig:"){height="6cm" width="5.2cm"}
In Fig.\[fig2\], we show the mass spectrum of $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$, $\tilde{\chi}_3^0$ and $\tilde{\chi}_1^\pm$ in our scenario. This figure indicates that the mass splittings among the particles satisfy $ 30 {\rm GeV} \lesssim \Delta_{\pm} \lesssim
70 {\rm GeV}$, $ 50 {\rm GeV} \lesssim \Delta_{0} \lesssim 110 {\rm GeV}$ and $ 80 {\rm GeV} \lesssim
(m_{\tilde{\chi}_3^0} - m_{\tilde{\chi}_1^0}) \lesssim 160 {\rm GeV}$. We remind that these splittings are induced by the strong mixings between Higgsinos and Singlino, and significantly larger than those among $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_1^\pm$ in the NS scenario of the MSSM [@Baer:2012uy] . In Fig.\[fig3\], we show the field components of the states $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_3^0$ respectively. As is expected, the $\tilde{H}_u^0$, $\tilde{S}^0$ and $\tilde{H}_d^0$ components in $\tilde{\chi}_1^0$ are comparable in magnitude with the largest one coming from the $\tilde{H}_u^0$ component. We emphasize again that the large Singlino component, i.e. $N_{15} \sim 0.5$, can dilute the interactions of the Higgsino-dominated $\tilde{\chi}_1^0$ with other fields, and consequently $\tilde{\chi}_1^0$ can reach its right relic density. In this case, we checked that the main annihilation channels of $\tilde{\chi}_1^0$ in early universe include $\tilde{\chi}_1^0 \tilde{\chi}_1^0 \to W^+ W^-, Z Z, Z h_1, h_1 h_1, h_1 h_2, q \bar{q}$. As for $\tilde{\chi}_2^0$, its largest component comes from either $\tilde{H}_d^0$ field (for most cases) or $\tilde{H}_u^0$ field (in rare cases), and in general the two components are comparable, which can be learned from the figure and also from Eq.(\[neutralino-mixing\]). We checked that due to the spectrum and the mixings, the dominant decay of $\tilde{\chi}_1^\pm$ is $\tilde{\chi}_1^\pm \to \tilde{\chi}_1^0 W^\ast$, and that of $\tilde{\chi}_2^0$ is usually $\tilde{\chi}_2^0 \to \tilde{\chi}_1^0 Z^{(\ast)}$. By contrast, the possible dominant decay modes of $ \tilde{\chi}_3^0$ are rather rich, which include $\tilde{\chi}_1^0 Z^{(\ast)}, \tilde{\chi}_1^0 h_1, \tilde{\chi}_1^\pm W^{(\ast)} $. Since $ \tilde{\chi}_3^0$ is Singlino-dominated, its production rate is rather low, and consequently its phenomenology is of less interest.
![Correlations among the main decay modes of $\tilde{t}_1$ in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Note that if $\tilde{t}_1$ is $\tilde{t}_R$ dominated and meanwhile $|N_{14}| \simeq |N_{24}|$, we have $Br(\tilde{t}_1 \to \tilde{\chi}_1^+ b): Br(\tilde{t}_1 \to \tilde{\chi}_1^0 t):
Br(\tilde{t}_1 \to \tilde{\chi}_2^0 t) \simeq 2:1:1$. On the other hand, if $\tilde{t}_1$ is $\tilde{t}_L$ dominated, $\tilde{t}_1$ prefers to decay into the Higgsino-dominated $\tilde{\chi}_{1,2}^0$ with $Br(\tilde{t}_1 \to \tilde{\chi}_1^0 t) \simeq Br(\tilde{t}_1 \to \tilde{\chi}_2^0 t)$. \[fig4\]](fig4.png "fig:"){height="6cm" width="15cm"}\
Next we turn to the properties of $\tilde{t}_1$. From the interactions of $\tilde{t}_1$ presented in [@MSSM-1], one can infer that if $\tilde{t}_1$ is $\tilde{t}_R$ dominated and meanwhile $|N_{14}| \simeq |N_{24}|$, the relation $Br(\tilde{t}_1 \to \tilde{\chi}_1^+ b): Br(\tilde{t}_1 \to \tilde{\chi}_1^0 t): Br(\tilde{t}_1 \to \tilde{\chi}_2^0 t) \simeq 2:1:1$ should hold. On the other hand, if $\tilde{t}_1$ is $\tilde{t}_L$ dominated, $\tilde{t}_1$ prefers to decay into the Higgsino-dominated $\tilde{\chi}_{1,2}^0$ with $Br(\tilde{t}_1 \to \tilde{\chi}_1^0 t) \simeq Br(\tilde{t}_1 \to \tilde{\chi}_2^0 t)$. These features are exhibited in Fig.\[fig4\], where we show the correlations between different decay rates of $\tilde{t}_1$ in our scenario. From Fig.\[fig4\], one can also learn that the branching ratio of $\tilde{t}_1 \to \tilde{\chi}_3^0 t$ is less than $10\%$. This is because $\tilde{\chi}_3^0$ is Singlino-dominated and its $\tilde{H}_u^0$ component is small. Moreover, we note that in our scenario $m_{\tilde{t}_1}$ is lower bounded by about $500 {\rm GeV}$, which is about $100 {\rm GeV}$ less than that in the NS scenario of the MSSM [@Kobakhidze:2015scd; @Cao:2012rz; @Han:2013kga]. One reason for the difference is that $\tilde{t}_1$ in our scenario has richer decay modes.
Future detection of our scenario
================================
Detection at 14 TeV LHC
-----------------------
From the analysis in last section, one can learn that the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated is characterized by predicting $\mu \leq 160 GeV$ and sizable mass splittings among the Higgsino-dominated neutralinos and chargino, i.e. 30 GeV $\leq \Delta_{\pm} \leq $ 70 GeV and 50 GeV $\leq \Delta_{0} \leq $ 110 GeV. Although this kind of spectrum is allowed by the direct searches for the electroweakinos at LHC Run I, it is expected to be tightly constrained at the upgraded LHC.
$m_{\tilde{\chi}_{1}^{0}}$ $m_{\tilde{\chi}_{2}^{0}}$ $m_{\tilde{\chi}_{3}^{0}}$ $m_{\tilde{\chi}_{1}^{\pm}}$ BR($\tilde{\chi}_{2}^{0}\to \tilde{\chi}_{1}^{0}Z^{(*)}$) BR($\tilde{\chi}_{3}^{0}\to \tilde{\chi}_{1}^{0}Z^{(*)}$) BR($\tilde{\chi}_{1}^{\pm}\to \tilde{\chi}_{1}^{0}W^*$)
---- ---------------------------- ---------------------------- ---------------------------- ------------------------------ ----------------------------------------------------------- ----------------------------------------------------------- ---------------------------------------------------------
P1 80.2 129.1 158.4 108.0 94.2% 9.08% 100%
P2 67.3 142.6 180.0 110.2 94.7% 7.55% 100%
P3 82.5 165.5 219.2 135.9 98.1% 26.3% 100%
P4 74.9 193.4 220.6 147.6 96.2% 6.50% 100%
: Mass and decay information of the benchmark points P1, P2, P3 and P4 in our study.[]{data-label="table-3"}
SR0$\tau$a $m_{SFOS}$ $m_T$ $E_T^{miss}$ $m_{3l}$ Background P1 P2 P3 P4
------------ ------------ -------- -------------- ---------- ------------ ------- ------- ------- ------- -- --
1 12-40 0-80 50-90 no 2.41 0.652 0.423 0.183 0.007
2 12-40 0-80 $>$90 no 0.45 0.273 0.176 0.108 0.003
3 12-40 $>$80 50-75 no 1 0.070 0.054 0.040 0.001
4 12-40 $>$80 $>$75 no 1.08 0.064 0.074 0.074 0.008
5 40-60 0-80 50-75 yes 1.37 0.131 0.365 0.170 0.006
6 40-60 0-80 $>$75 no 0.76 0.119 0.509 0.302 0.013
7 40-60 $>$80 50-135 no 1.49 0.122 0.240 0.183 0.011
8 40-60 $>$80 $>$135 no 0.2 0.008 0.022 0.022 0.002
9 60-81.2 0-80 50-75 yes 2.4 0.032 0.156 0.218 0.040
10 60-81.2 $>$80 50-75 no 1.51 0.027 0.074 0.087 0.015
11 60-81.2 0-110 $>$75 no 2.98 0.062 0.312 0.438 0.094
12 60-81.2 $>$110 $>$75 no 0.63 0.039 0.072 0.082 0.017
13 81.2-101.2 0-110 50-90 yes 66.41 0.024 0.415 0.146 0.870
14 81.2-101.2 0-110 $>$90 no 21.62 0.016 0.303 0.107 0.744
15 81.2-101.2 $>$110 50-135 no 5.98 0.031 0.086 0.047 0.157
16 81.2-101.2 $>$110 $>$135 no 0.59 0.006 0.018 0.025 0.032
17 $>$101.2 0-180 50-210 no 7.65 0.066 0.136 0.091 0.032
18 $>$101.2 $>$180 50-210 no 0.44 0.008 0.026 0.017 0.001
19 $>$101.2 0-120 $>$210 no 0.24 0.002 0.003 0.002 0.005
20 $>$101.2 $>$120 $>$210 no 0.09 0.002 0.005 0.007 0.000
: Cross sections of the four benchmark points in each bin of the signal region SR0$\tau$a, which are obtained from our simulations and presented in the last four columns of this table. Quantities from the second column to the fifth column define the bins of the SR0$\tau$a, and their physical meanings are explained in Appendix B.1. The sixth column corresponds to the backgrounds of the bins, which are taken from [@Cao:2015efs]. All quantities with mass dimension and cross sections are in units of GeV and fb respectively.[]{data-label="table-4"}
---- ------------------- ------------------- ------------------- ------------------- -- -- -- --
P1 $S$(bin2) = 2.09 $S$(bin1) = 1.75 $S$(bin2) = 4.59 $S$(bin1) = 2.54
P2 $S$(bin6) = 2.88 $S$(bin5) =1.44 $S$(bin6) = 5.58 $S$(bin2) = 2.96
P3 $S$(bin6) = 1.71 $S$(bin11) = 1.01 $S$(bin6) = 3.31 $S$(bin2) = 1. 82
P4 $S$(bin14) = 0.32 $S$(bin15) = 0.21 $S$(bin16) = 0.42 $S$(bin14) = 0.34
---- ------------------- ------------------- ------------------- ------------------- -- -- -- --
: The best two signal bins and corresponding significances for the benchmark points with 30 fb$^{-1}$ and 300 fb$^{-1}$ integrated luminosity data respectively at 14 TeV LHC.[]{data-label="table-5"}
We investigate this issue by considering the neutralino and chargino associated production processes at 14 TeV LHC. For simplicity we adopt 4 benchmark points listed in Table \[table-3\], which are discriminated by the values of $\mu$ (or equivalently $m_{\tilde{\chi}_1^\pm}$), $\Delta_{0}$ and $\Delta_{\pm}$. Since $\tilde{\chi}_1^\pm$ for these points decays into $\tilde{\chi}_1^0$ plus an off-shell $W$ boson, and $\tilde{\chi}_2^0$ decays mainly into $\tilde{\chi}_1^0$ plus a $Z$ boson (on-shell or off-shell), the signal region SR0${\tau a}$ in the ATLAS direct searches for electroweakinos by trileptons and large $E_T^{miss}$ signal [@Aad:2014nua], which was proposed in the analysis [@Baer:1985at] and also briefly introduced in the appendix of this work, is most pertinent to explore those points. In our analysis, we simulate the processes $ p p \to \tilde{\chi}_1^\pm \tilde{\chi}_2^0, \tilde{\chi}_1^\pm \tilde{\chi}_3^0 \to 3 l + E_T^{miss} $ to get their summed rate in each bin of the signal region, and present the result in the last four columns of Table \[table-4\]. We also present in the table the backgrounds of the bins at 14 TeV LHC, which were obtained by detailed simulations done in [@Cao:2015efs]. With these results, we evaluate the significance $S = s /\sqrt{b+(\epsilon b)^2 }$ for each bin, where $s$ and $b$ correspond to the number of signal and background events and $\epsilon=10\%$ is the assumed systematical uncertainty of the backgrounds. Assuming 30 fb$^{-1}$ and 300 fb$^{-1}$ integrated luminosity data at 14 TeV LHC, we present the best two signal bins and corresponding expected significances for P1, P2, P3 and P4 in Table \[table-5\]. This table reveals following information:
- With 30 fb$^{-1}$ integrated luminosity data, P1 and P2 can be excluded at 2$\sigma$ confidence level, and with 300 fb$^{-1}$ data P3 can also be excluded. In any case, the point P4 is hard to be excluded.
- For each point, which signal bin is best for exclusion depends on the mass splittings among the neutralinos and chargino. For example, since $\Delta_{\pm}< m_W$ for all the four points, the most effective bins usually require $m_T<$ 80 or 110 GeV. For points P1, P2 and P3, the bins satisfying $m_{SFOS} < m_Z$ are preferred for exclusion since $\Delta_0 < m_Z$, and by contrast the bins with $|m_{SFOS}-m_Z|<$ 10 GeV (such as bins 14 and 15) are favored by point P4 since in this case $\tilde{\chi}_2^0$ can decay into an on-shell $Z$ boson. Note that for bins 14 and 15, the backgrounds are relatively large, and that is why the point P4 can not be excluded at 14 TeV LHC after including the systematic uncertainties.
- With 300 fb$^{-1}$ data, the point P2 can be discovered at 14 TeV LHC. This is partially because $\tilde{\chi}_2^0$ and $\tilde{\chi}_1^\pm$ are relatively light so that the rate of their associated production is large, and also because they have sizable mass splittings from $\tilde{\chi}_1^0$ to result in moderately energetic decay products.
- Since the bins in the SR0$\tau$a are disjoint, in principle their results can be statistically combined to maximize the significance. We did this, but we found that the improvement is not significant.
Since the four benchmark points stand for the typical situation in our scenario, we conclude that the future LHC experiments can exclude most part of the parameter space for the scenario, and consequently the fine-tuning of the NMSSM can be pushed to higher level. We will discuss such an issue extensively in our forthcoming work.
![Spin-independent (SI) and Spin-dependent (SD) DM-nucleon scattering cross sections versus DM mass in the NS scenario with $\tilde{\chi}_1^0$ being Higgsino-dominated. Capabilities of future DM direct detection experiments in detecting the scattering are also plotted. \[fig5\] ](fig5.png){width="17cm"}
Dark Matter Direct Search
-------------------------
Since $\tilde{\chi}_1^0$ as the DM candidate has quite large $\tilde{H}_u^0$, $\tilde{H}_d^0$ and $\tilde{S}^0$ components in our scenario, its interactions with the Higgs bosons and $Z$ boson are unsuppressed. As a result, the spin-independent (SI) and spin-dependent (SD) cross sections of the $\tilde{\chi}_1^0$-nucleon scattering are sizable, and may reach the sensitivities of future DM direct detection experiments such as XENON-1T and LZ-7.2T experiments [@direct:future]. In this section, we investigate such an issue. In Fig.\[fig5\], we project the samples of our scenario on $m_{\tilde{\chi}_1^0}-\sigma^{SI}_{\tilde{\chi}-p}$ and $m_{\tilde{\chi}_1^0}-\sigma^{SD}_{\tilde{\chi}-p}$ planes with $\sigma^{SI}_{\tilde{\chi}-p}$ and $\sigma^{SD}_{\tilde{\chi}-p}$ denoting the SI and SD cross sections respectively. The blue lines, red lines and green lines are the sensitivities of LUX, XENON-1T and LZ experiments respectively.
For the SI cross section, one can learn from Fig.\[fig5\] that the future XENON-1T experiment is able to probe a large portion of the samples, and the LZ experiment can test even more. Anyhow, there still exist some samples remaining untouched by these future experiments. We numerically checked that for the untouched samples, there exists rather strong cancelation among the contributions induced by different CP-even Higgs bosons. On the other hand, the story for $\sigma^{SD}_{\tilde{\chi}-p}$ is quite different. From the right panel of Fig.\[fig5\] one can see that the future XENON-1T experiment can test nearly all of the samples, let alone the more sensitive LZ experiment. The underlying reason is that in the NMSSM with heavy sfermions, the SD cross section gets contribution mainly from the $t$-channel $Z$-mediated diagram. As a result, the size of the cross section is determined by the $Z \tilde{\chi}_1^0 \tilde{\chi}_1^0$ coupling, which is given by $$\begin{aligned}
g_{Z \tilde{\chi}_1^0 \tilde{\chi}_1^0} = \frac{m_Z}{\sqrt{2} v} ( N_{13}^2 - N_{14}^2 ) \sim 0.2 \times \frac{m_Z}{\sqrt{2} v}\end{aligned}$$ In the last step of the equation, we have used the information of $N_{13}$ and $N_{14}$ presented in Fig.\[fig3\]. Moreover, it is interesting to see that although the benchmark point P4 in Table \[table-3\] is hard to be excluded by sparticle direct search experiments at the LHC, its SD cross section is quite large and so the point will be tested by future dark matter direct search experiments.
In getting Fig.\[fig5\], we use the package micrOMEGAs [@micrOMEGA] to calculate the cross sections. We choose its default setting $\sigma_{\pi N} = 34 {\rm MeV}$ and $\sigma_0 = 42 {\rm MeV}$ as input. We checked that if we take $\sigma_{\pi N} = 59 {\rm MeV}$ from [@SI-piN] and $\sigma_0 = 58 {\rm MeV}$ from [@SI-pi0-1; @SI-pi0-2; @SI-pi0-3], the SI cross section will be enhanced by a factor from $20\%$ to $40\%$, and this does not affect the conclusions presented in this work.
Conclusions
===========
With the great discovery of the 125 GeV Higgs boson and the increased bounds on sparticle masses at LHC Run I, the NS scenario in MSSM has become theoretically unsatisfactory. By contrast, the situation may be improved greatly in the NMSSM. This motivates us to scrutinize the impact of the direct searches for SUSY at LHC Run I on the naturalness of the NMSSM.
We start our study by scanning the vast parameter space of the NMSSM to get the region where the fine tuning measures $\Delta_Z$ and $\Delta_h$ at electroweak scale are less than about 50. In this process, we considered various experimental constraints such as DM relic density, LUX limits on the scattering rate of DM with nucleon and the 125 GeV Higgs data on the model. We classify the surviving samples into four types: for Type I samples, $h_1$ corresponds to the SM-like Higgs boson and $\tilde{\chi}_1^0$ is Bino-dominated, while for Type II, III and IV samples, $h_2$ acts as the SM-like Higgs boson with $\tilde{\chi}_1^0$ being Bino-, Singlino- and Higgsino-dominated respectively. After these preparations, we specially study the influence of the direct searches for SUSY on the samples. We implement the direct search constraints by the packages FastLim and SModelS and also by simulating the electroweakino and stop production processes. Our results indicate that although the direct search experiments are effective in excluding the samples, the parameter intervals for the region and also the minimum reaches of $\Delta_Z$ and $\Delta_h$ are scarcely changed by the constraints, which implies that, contrary to general belief, the fine tuning of the NMSSM does not get worse after LHC Run I. Our results also indicate that the lowest fine-tuning comes from type III and type IV samples, for which $\Delta_Z$ and $\Delta_h$ may be as low as about 2 without conflicting with any experimental constraints.
Considering that the type IV samples were scarcely studied in previous literatures and that they have similar phenomenology to that of the NS scenario in the MSSM, we investigate the essential features of this kind of samples. We find that they are characterized by strong mixings between Higgsinos and Singlino in forming mass eigenstates called neutralinos. As a result, the lightest neutralino $\tilde{\chi}_1^0$ as the DM candidate has significant Singlino component so that it can easily reach the measured DM relic density, and meanwhile the mass splittings among the Higgsino-dominated particles $\tilde{\chi}_1^0$, $\tilde{\chi}_2^0$ and $\tilde{\chi}_1^\pm$ are usually larger than $30 {\rm GeV}$. These features make the samples rather special. For example, we show that due to the rich decay products of the lighter scalar top quark, its lower mass bound is decreased by about 100GeV in comparison with that in the NS scenario of the MSSM, and that the neutralino-chargino sector of the samples can be readily tested either through searching for $3 l + E_T^{miss}$ signal at 14 TeV LHC or through future dark matter direct detection experiments.
In summary, we conclude that so far the fine tuning of the NMSSM is scarcely affected by the direct searches for SUSY at LHC Run I, and it can still predict $Z$ boson mass and the SM-like Higgs boson mass in a natural way. This conclusion, however, may be altered by the upcoming 14 TeV LHC experiments and DM matter direct search experiments.
Acknowledgement {#acknowledgement .unnumbered}
===============
This work was supported by the National Natural Science Foundation of China (NNSFC) under grant No. 11575053, 11275245 and 11305050.
Fastlim and SModelS
===================
Name Description $\sqrt{s}$ $\mathcal{L}_{\text{int}}(\text{fb}^{-1})$ Ref.
--------------------- ---------------------------------------------------------------- ------------ -------------------------------------------- ------------------
ATLAS-CONF-2013-062 1-2 leptons + 3-6 jets + $E_T^{miss}$ (squarks and gluino) 20.3 [@atlas:2013062]
ATLAS-CONF-2013-061 0-1 lepton + $\ge$3 b-jets + $E_T^{miss}$ (3rd gen. squarks) 20.1 [@atlas:2013061]
ATLAS-CONF-2013-054 0 lepton + $\ge$ 7-10 jets + $E_T^{miss}$ (squarks and gluino) 20.3 [@atlas:2013054]
ATLAS-CONF-2013-053 0 lepton + 2 b-jets + $E_T^{miss}$ (sbottom and stop) 20.1 [@atlas:2013053]
ATLAS-CONF-2013-048 2 leptons (+ jets) + $E_T^{miss}$ (stop) 20.3 [@atlas:2013048]
ATLAS-CONF-2013-047 0 lepton + 2-6 jets + $E_T^{miss}$ (squarks and gluino) 20.3 [@atlas:2013047]
ATLAS-CONF-2013-037 1 lepton + 4(1 b-)jets + $E_T^{miss}$ (stop) 20.7 [@atlas:2013037]
ATLAS-CONF-2013-024 0 lepton + (2 b-)jets + $E_T^{miss}$ (stop) 20.5 [@atlas:2013024]
: Experiments in the Fastlim database.[]{data-label="t-fastlim"}
In this section, we briefly introduce the packages Fastlim [@Papucci:2014rja] and SModelS [@Kraml:2013mwa], which can be used to implement the constraints from the direct searches for sparticles at LHC Run I in an easy and fast way.
Fastlim
-------
Fastlim is a package that limits SUSY parameter space using the LHC-8TeV data. It incorporates in its database cut efficiency tables for different sparticle production processes in simplified model framework. Now it supports the processes of gluino pair production and third generation squark pair productions, and it involves many decay modes of gluino and the squarks, such as $\tilde{g}\to \tilde{t}_{1,2}^\ast t \to \tilde{\chi}_1^0 t \bar{t}$ or $\tilde{g}\to q \bar{q} \tilde{\chi}_1^0$ and $\tilde{b}_{1,2} \to b \tilde{\chi}_1^0$ where $q$ stands for the first two generation quarks. Any processes that contain the same final states at the detector level are combined by Fastlim to improve the signal significance.
In this work, we only use the experiments of searching for gluino and squarks, which are listed in Table \[t-fastlim\]. This is because that Fastlim gives better limits for strong SUSY productions in comparison with the package SModelS [@Kraml:2013mwa].
SModelS
-------
SModelS have the same function as that of Fastlim, but it could give better limits for slepton productions and electroweakino productions. SModelS has implemented the information about following experiments:
- Direct slepton searches (ATLAS): ATLAS-CONF-2013-049 [@atlas:2013049].
- Direct slepton searches (CMS): SUS-12-022 [@cms:12022], SUS-13-006 [@cms:13006].
- Electroweakino searches (ATLAS): ATLAS-CONF-2013-028 [@atlas:2013028], ATLAS-CONF-2013-035 [@atlas:2013035], ATLAS-CONF-2013-036 [@atlas:2013036], ATLAS-CONF-2013-093 [@atlas:2013093].
- Electroweakino searches (CMS): SUS-12-022 [@cms:12022], SUS-13-006 [@cms:13006], SUS-13-017 [@cms:13017],
and it contains the cross section limits at 95% confidence level for above analyses. If one parameter point predicts cross sections larger than those presented in the analyses, it will be excluded. Otherwise, it is allowed.
Note that the efficiencies or the upper bounds on SUSY signals in the database of the two packages are usually based on certain assumptions, which may not be applied to some parameter points encountered in our scan. In this case, the encountered point is considered to be experimentally allowed. Also note that the packages are based on the preliminary analyses of the ATLAS and CMS groups, which were done in 2013. Given that most of these analyses have been updated in past two years, a more powerful exclusion capability may be obtained if one repeats the updated analyses by detailed Monte Carlo simulation. Anyhow, these two packages can serve as useful tools to exclude some SUSY parameter points.
Details of our simulation
=========================
In our study, any point that passed Fastlim and SModelS is further tested by simulations to see whether it survives the constraints from the direct search experiments. In detail, we first use MadGraph/MadEvent [@Alwall:2011uj] to generate parton level events for certain sparticle production processes, and feed them into Pythia [@Sjostrand:2006za] for parton showering and hadronization. Then we use the package CheckMATE [@Drees:2013wra] where a well-tuned Delphes [@deFavereau:2013fsa] is provided for the detector simulation and analyses. We define $R\equiv \text{max}\left\{S_i/S_{i, obs}^{95}\right\}$ to decide whether the point survives the analysis, where $S_i$ stands for the simulated signal events in the $i$th signal region of the analysis, and $S_{i, obs}^{95}$ represents the $95\%$ C.L. upper limit of the event number in the signal region. If $R>1$, the parameter point is excluded by the analysis and otherwise it is allowed.
Since the fine tuning of the NMSSM at the electroweak scale is mainly affected by its chargino-neutralino sector and stop sector, we repeat by simulation the ATLAS analyses in [@Aad:2014vma], [@Aad:2014nua], [@Aad:2014kra] and [@Aad:2014mha]. All these analyses are based on 20.3fb$^{-1}$ data at the LHC-8TeV with the former two presenting so far the strictest limits on electroweakino productions, and the latter two providing the tightest constraints on the pair productions of third generation squarks. In the following, we briefly introduce these analyses.
Search for Electroweakino at the LHC
------------------------------------
The analysis [@Aad:2014vma] targets final states with two leptons and large $E_T^{miss}$. In our simulation of the analysis, we mainly focus on the signal region named “SR-Zjets”, which is specifically designed for the process $p p \to \tilde{\chi}_{2}^0\tilde{\chi}_1^{\pm}\to Z\tilde{\chi}_1^0 W\tilde{\chi}_1^0
\to \ell\ell\tilde{\chi}_1^0 q q \tilde{\chi}_1^0$. This signal region requires that the two leading leptons should be same flavor but opposite sign (SFOS), and that their invariant mass locates at $Z$-peak. As was pointed out in [@Aad:2014vma], it provides the strongest constraints on the chargino-neutralino sector among the anlayses with dilepton final state.
The analysis in [@Aad:2014nua] also searches for electroweakino productions but with final states of three leptons and large $E_T^{miss}$. Here we concentrate on the signal region named “SR0$\tau$a” which is optimized for processes $p p \to \tilde{\chi}_{2}^0\tilde{\chi}_1^{\pm}\to Z^{(\ast)} \tilde{\chi}_1^0 W^{(\ast)} \tilde{\chi}_1^0
\to \ell\ell\tilde{\chi}_1^0 \ell\nu \tilde{\chi}_1^0$. This region requires a pair of SFOS leptons in its signal, and utilizes the transverse mass $m_T=\sqrt{2\left|\vec p_T^{~\ell}\right|\left|\vec E_T^{miss}\right|
-2\vec p_T^{~\ell} \cdot \vec E_T^{miss}}$ (here $p_T^l$ is the transverse momentum of the lepton not forming the SFOS lepton pair) to suppress the SM background. It considers twenty bins, which are categorized by the SFOS leptons’ invariant mass, $m_T$ and $E_T^{miss}$, to maximize its sensitivity to different mass spectrums of $\tilde{\chi}_{2}$, $\tilde{\chi}_1^{\pm}$ and $\tilde{\chi}_1^0$.
We remind that in the NS scenarios of the NMSSM, all $\tilde{\chi}_{i}^0\tilde{\chi}_1^{\pm}$ associated production processes with $i=2,3,4,5$ may contribute sizably to the trilepton signal, so in our simulation we include all these contributions. By contrast, SModelS does not combine processes that have the same final states at the detector level. Moreover, in our analysis we combine the signal region “SR-Zjets” with the bins in “SR0$\tau$a” to maximize the discovery significance by the $CL_s$ method in RooStats as we did in [@Cao:2015efs].
Search for stops at the LHC
---------------------------
In the NS scenario of the NMSSM, $\tilde{t}_{1}$ usually decays like $\tilde{t}_1 \to \tilde{\chi}_{2,3}^0 t \to \tilde{\chi}_{1}^0 Z^{(\ast)} t$ or $\tilde{t}_1 \to \tilde{\chi}_{1}^+ b \to \tilde{\chi}_{1}^0 W^{(\ast)} b$. Considering that these two-step topologies haven’t been included in the database of Fastlim, we repeat recent ATLAS analyses on stop pair productions, which were presented in [@Aad:2014kra] and [@Aad:2014mha].
The analysis in [@Aad:2014kra] searches for stops in final states containing exactly one isolated lepton, at least two jets and a large $E_T^{miss}$. It contains fifteen signal regions targeting a large number of stop pair production scenarios, where stop may decay like $\tilde{t}_1\to t\tilde{\chi}_1^0$, $\tilde{t}_1\to bW^{(\ast)}\tilde{\chi}_1^0$ and $\tilde{t}_1\to b\tilde{\chi}_1^{\pm}\to b W^{(\ast)}\tilde{\chi}_1^0$. Especially, nine of these signal regions are designed for the latter decay with different mass relations among $\tilde{t}_1$, $\tilde{\chi}_1^{\pm}$ and $\tilde{\chi}_1^0$, which determines the kinematic properties of the process. These nine regions are particularly relevant to our study. In general, the signal regions are discriminated by different kinematic cuts on the leptons and (b-)jets, $E_T^{miss}$, $m_T$, b-jet multiplicity and the asymmetric transverse mass $am_{T2}$ [@Bai:2012gs].
The analysis in [@Aad:2014mha] targets the process $p p \to \tilde{t}_2^\ast \tilde{t}_2$ with $\tilde{t}_2\to\tilde{t}_1 Z\to\tilde{\chi}_1^0 Z t$, and it searches for the signal that contains a SFOS pair of leptons with their invariant mass near $m_Z$, at least one b-jet and a large $E_T^{miss}$. Obviously, if $\tilde{t}_2$ is significantly heavier than $\tilde{t}_1$, the $Z$ boson is highly boosted. As a result, the transverse momentum of the dilepton system $p_T(\ell\ell)$ tends to be high and the azimuthal separation $\Delta \phi({\ell\ell})$ prefers to be low. On the other side, if the mass splitting between $\tilde{t}_1$ and $\tilde{\chi}_1^0$ is large, high jet multiplicity is expected. These facts motivates physicists to define five signal regions, which are called “SR2(A,B,C)” and “SR3(A,B)” respectively, by the number of leptons in the signal, the jet multiplicity and whether the Z boson is boosted. For example, the signal regions “SR2(A,B,C)” require that the signal events contain exactly two signal leptons and the $Z$ boson is boosted. The regions “SR2A” and “SR2B” are optimized for low jet multiplicity, while the region “SR2C” is designed for high jet multiplicity case.
[32]{} H. E. Haber and G. L. Kane, Phys. Rept. [**117**]{}, 75 (1985). A. Djouadi, Phys. Rept. [**459**]{}, 1 (2008) \[hep-ph/0503173\]. U. Ellwanger, C. Hugonie and A. M. Teixeira, Phys. Rept. [**496**]{}, 1 (2010) doi:10.1016/j.physrep.2010.07.001 \[arXiv:0910.1785 \[hep-ph\]\]. M. Maniatis, Int. J. Mod. Phys. A [**25**]{}, 3505 (2010) doi:10.1142/S0217751X10049827 \[arXiv:0906.0777 \[hep-ph\]\]. H. Baer, V. Barger, P. Huang and X. Tata, JHEP [**1205**]{}, 109 (2012) doi:10.1007/JHEP05(2012)109 \[arXiv:1203.5539 \[hep-ph\]\]. H. Baer, V. Barger, P. Huang, D. Mickelson, A. Mustafayev and X. Tata, Phys. Rev. D [**87**]{}, no. 11, 115028 (2013) doi:10.1103/PhysRevD.87.115028 \[arXiv:1212.2655 \[hep-ph\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], ATLAS-CONF-2013-047.
S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], JHEP [**1406**]{}, 055 (2014) \[arXiv:1402.4770 \[hep-ex\]\]. J. Cao, Y. He, L. Shang, W. Su and Y. Zhang, JHEP [**1603**]{}, 207 (2016) doi:10.1007/JHEP03(2016)207 \[arXiv:1511.05386 \[hep-ph\]\].
G. Aad [*et al.*]{} \[ATLAS Collaboration\], Phys. Lett. B [**716**]{}, 1 (2012) \[arXiv:1207.7214 \[hep-ex\]\]. S. Chatrchyan [*et al.*]{} \[CMS Collaboration\], Phys. Lett. B [**716**]{}, 30 (2012) \[arXiv:1207.7235 \[hep-ex\]\]. H. Baer, V. Barger and A. Mustafayev, Phys. Rev. D [**85**]{}, 075010 (2012) \[arXiv:1112.3017 \[hep-ph\]\]; S. Heinemeyer, O. Stal and G. Weiglein, Phys. Lett. B [**710**]{}, 201 (2012); A. Arbey [*et al.*]{}, Phys. Lett. B [**708**]{}, 162 (2012); P. Draper [*et al.*]{}, Phys. Rev. D [**85**]{}, 095007 (2012); M. Carena, [*et al.*]{}, JHEP [**1203**]{}, 014 (2012); J. Cao, [*et al.*]{}, Phys. Lett. B [**710**]{}, 665 (2012) \[arXiv:1112.4391 \[hep-ph\]\]; J. Cao, L. Shang, P. Wu, J. M. Yang and Y. Zhang, JHEP [**1510**]{}, 030 (2015) doi:10.1007/JHEP10(2015)030 \[arXiv:1506.06471 \[hep-ph\]\]. J. Cao, L. Shang, P. Wu, J. M. Yang and Y. Zhang, Phys. Rev. D [**91**]{}, no. 5, 055005 (2015) doi:10.1103/PhysRevD.91.055005 \[arXiv:1410.3239 \[hep-ph\]\]. J. Guo, J. Li, T. Li and A. G. Williams, Phys. Rev. D [**91**]{}, no. 9, 095003 (2015) doi:10.1103/PhysRevD.91.095003 \[arXiv:1409.7864 \[hep-ph\]\]. X. J. Bi, L. Bian, W. Huang, J. Shu and P. F. Yin, Phys. Rev. D [**92**]{}, 023507 (2015) doi:10.1103/PhysRevD.92.023507 \[arXiv:1503.03749 \[hep-ph\]\].
U. Ellwanger, JHEP [**1203**]{}, 044 (2012) \[arXiv:1112.3548 \[hep-ph\]\]; Z. Kang, J. Li and T. Li, JHEP [**1211**]{}, 024 (2012) \[arXiv:1201.5305 \[hep-ph\]\]. S. F. King, M. Muhlleitner and R. Nevzorov, Nucl. Phys. B [**860**]{}, 207 (2012) \[arXiv:1201.2671 \[hep-ph\]\]; J. J. Cao, [*et al.*]{}, JHEP [**1203**]{}, 086 (2012) \[arXiv:1202.5821 \[hep-ph\]\]; K. Choi, [*et al.*]{}, JHEP [**1302**]{}, 090 (2013) \[arXiv:1211.0875 \[hep-ph\]\]; S. F. King, [*et al.*]{}, Nucl. Phys. B [**870**]{}, 323 (2013) \[arXiv:1211.5074 \[hep-ph\]\]; M. Badziak, M. Olechowski and S. Pokorski, JHEP [**1306**]{}, 043 (2013) \[arXiv:1304.5437 \[hep-ph\]\]; S. Moretti, S. Munir and P. Poulose, Phys. Rev. D [**89**]{}, no. 1, 015022 (2014) doi:10.1103/PhysRevD.89.015022 \[arXiv:1305.0166 \[hep-ph\]\]. D. Das, U. Ellwanger and A. M. Teixeira, JHEP [**1204**]{}, 067 (2012) doi:10.1007/JHEP04(2012)067 \[arXiv:1202.5244 \[hep-ph\]\]. D. A. Vasquez, [*et al.*]{}, Phys. Rev. D [**86**]{}, 035023 (2012) \[arXiv:1203.3446 \[hep-ph\]\]; U. Ellwanger, JHEP [**1311**]{}, 108 (2013) doi:10.1007/JHEP11(2013)108 \[arXiv:1309.1665 \[hep-ph\]\]. J. S. Kim and T. S. Ray, Eur. Phys. J. C [**75**]{}, 40 (2015) doi:10.1140/epjc/s10052-015-3281-4 \[arXiv:1405.3700 \[hep-ph\]\]. U. Ellwanger and C. Hugonie, JHEP [**1408**]{}, 046 (2014) doi:10.1007/JHEP08(2014)046 \[arXiv:1405.6647 \[hep-ph\]\]. U. Ellwanger and A. M. Teixeira, JHEP [**1410**]{}, 113 (2014) doi:10.1007/JHEP10(2014)113 \[arXiv:1406.7221 \[hep-ph\]\]. B. Dutta, Y. Gao and B. Shakya, Phys. Rev. D [**91**]{}, no. 3, 035016 (2015) doi:10.1103/PhysRevD.91.035016 \[arXiv:1412.2774 \[hep-ph\]\]. B. Allanach, M. Badziak, C. Hugonie and R. Ziegler, Phys. Rev. D [**92**]{}, no. 1, 015006 (2015) doi:10.1103/PhysRevD.92.015006 \[arXiv:1502.05836 \[hep-ph\]\]. A. Chakraborty, D. K. Ghosh, S. Mondal, S. Poddar and D. Sengupta, Phys. Rev. D [**91**]{}, 115018 (2015) doi:10.1103/PhysRevD.91.115018 \[arXiv:1503.07592 \[hep-ph\]\]. J. Beuria, A. Chatterjee, A. Datta and S. K. Rai, JHEP [**1509**]{}, 073 (2015) doi:10.1007/JHEP09(2015)073 \[arXiv:1505.00604 \[hep-ph\]\].
C. T. Potter, Eur. Phys. J. C [**76**]{}, no. 1, 44 (2016) doi:10.1140/epjc/s10052-015-3867-x \[arXiv:1505.05554 \[hep-ph\]\]. R. Enberg, S. Munir, C. Pérez de los Heros and D. Werder, arXiv:1506.05714 \[hep-ph\]. D. Barducci, G. BÅ¡Å langer, C. Hugonie and A. Pukhov, arXiv:1510.00246 \[hep-ph\].
J. S. Kim, D. Schmeier and J. Tattersall, Phys. Rev. D [**93**]{}, no. 5, 055018 (2016) doi:10.1103/PhysRevD.93.055018 \[arXiv:1510.04871 \[hep-ph\]\]. M. Badziak, M. Olechowski and P. Szczerbiak, JHEP [**1603**]{}, 179 (2016) doi:10.1007/JHEP03(2016)179 \[arXiv:1512.02472 \[hep-ph\]\]. J. Beuria, A. Chatterjee and A. Datta, arXiv:1603.08463 \[hep-ph\]. Qian-Fei Xiang, Xiao-Jun Bi, Peng-Fei Yin, Zhao-Huan Yu, arXiv:1606.02149 \[hep-ph\]. P. A. R. Ade [*et al.*]{} \[Planck Collaboration\], Astron. Astrophys. [**571**]{}, A16 (2014) doi:10.1051/0004-6361/201321591 \[arXiv:1303.5076 \[astro-ph.CO\]\]. G. Hinshaw et al. \[WMAP Collaboration\], Astrophys. J. Suppl. 208, 19 (2013).
D. S. Akerib [*et al.*]{} \[LUX Collaboration\], Phys. Rev. Lett. [**116**]{}, no. 16, 161301 (2016) doi:10.1103/PhysRevLett.116.161301 \[arXiv:1512.03506 \[astro-ph.CO\]\]. D. S. Akerib [*et al.*]{} \[LUX Collaboration\], Phys. Rev. Lett. [**116**]{}, no. 16, 161302 (2016) doi:10.1103/PhysRevLett.116.161302 \[arXiv:1602.03489 \[hep-ex\]\]. For recent summary in this field, see H. Baer, V. Barger, M. Savoy and X. Tata, arXiv:1604.07438 \[hep-ph\]. U. Ellwanger and C. Hugonie, Adv. High Energy Phys. [**2012**]{}, 625389 (2012) doi:10.1155/2012/625389 \[arXiv:1203.5048 \[hep-ph\]\]. T. Cheng, J. Li, T. Li and Q. S. Yan, Phys. Rev. D [**89**]{}, no. 1, 015015 (2014) doi:10.1103/PhysRevD.89.015015 \[arXiv:1304.3182 \[hep-ph\]\]. H. Baer, V. Barger and D. Mickelson, Phys. Rev. D [**88**]{}, no. 9, 095013 (2013) doi:10.1103/PhysRevD.88.095013 \[arXiv:1309.2984 \[hep-ph\]\].
U. Ellwanger, G. Espitalier-Noel and C. Hugonie, JHEP [**1109**]{}, 105 (2011) \[arXiv:1107.2472 \[hep-ph\]\]. M. Farina, M. Perelstein and B. Shakya, JHEP [**1404**]{}, 108 (2014) \[arXiv:1310.0459 \[hep-ph\]\].
U. Ellwanger and C. Hugonie, Comput. Phys. Commun. [**175**]{}, 290 (2006); U. Ellwanger, J. F. Gunion and C. Hugonie, JHEP [**0502**]{}, 066 (2005). P. Bechtle, S. Heinemeyer, O. St[å]{}l, T. Stefaniak and G. Weiglein, Eur. Phys. J. C [**74**]{}, no. 2, 2711 (2014) \[arXiv:1305.1933 \[hep-ph\]\]; JHEP [**1411**]{}, 039 (2014) \[arXiv:1403.1582 \[hep-ph\]\]; O. St[å]{}l and T. Stefaniak, PoS EPS [**-HEP2013**]{}, 314 (2013) \[arXiv:1310.4039 \[hep-ph\]\]. M. D. Goodsell, K. Nickel and F. Staub, Phys. Rev. D [**91**]{}, 035021 (2015) doi:10.1103/PhysRevD.91.035021 \[arXiv:1411.4665 \[hep-ph\]\]. F. Staub, P. Athron, U. Ellwanger, R. Gr?ber, M. Mš¹hlleitner, P. Slavich and A. Voigt, Comput. Phys. Commun. [**202**]{}, 113 (2016) doi:10.1016/j.cpc.2016.01.005 \[arXiv:1507.05093 \[hep-ph\]\]. P. Bechtle, O. Brein, S. Heinemeyer, G. Weiglein and K. E. Williams, Comput. Phys. Commun. [**181**]{}, 138 (2010) \[arXiv:0811.4169 \[hep-ph\]\]; P. Bechtle, O. Brein, S. Heinemeyer, G. Weiglein and K. E. Williams, Comput. Phys. Commun. [**182**]{}, 2605 (2011) \[arXiv:1102.1898 \[hep-ph\]\]. M. Papucci, K. Sakurai, A. Weiler and L. Zeune, Eur. Phys. J. C [**74**]{}, no. 11, 3163 (2014) doi:10.1140/epjc/s10052-014-3163-1 \[arXiv:1402.0492 \[hep-ph\]\]. S. Kraml, S. Kulkarni, U. Laa, A. Lessa, W. Magerl, D. Proschofsky-Spindler and W. Waltenberger, Eur. Phys. J. C [**74**]{}, 2868 (2014) doi:10.1140/epjc/s10052-014-2868-5 \[arXiv:1312.4175 \[hep-ph\]\]. A. Kobakhidze, N. Liu, L. Wu, J. M. Yang and M. Zhang, Phys. Lett. B [**755**]{}, 76 (2016) doi:10.1016/j.physletb.2016.02.003 \[arXiv:1511.02371 \[hep-ph\]\]. J. Cao, C. Han, L. Wu, J. M. Yang and Y. Zhang, JHEP [**1211**]{}, 039 (2012) doi:10.1007/JHEP11(2012)039 \[arXiv:1206.3865 \[hep-ph\]\]. C. Han, K. i. Hikasa, L. Wu, J. M. Yang and Y. Zhang, JHEP [**1310**]{}, 216 (2013) doi:10.1007/JHEP10(2013)216 \[arXiv:1308.5307 \[hep-ph\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], JHEP [**1404**]{}, 169 (2014) doi:10.1007/JHEP04(2014)169 \[arXiv:1402.7029 \[hep-ex\]\]. H. Baer and X. Tata, Phys. Lett. B [**155**]{}, 278 (1985). doi:10.1016/0370-2693(85)90654-9 P. Cushman [*et al.*]{}, arXiv:1310.8327 \[hep-ex\]. G. Belanger, F. Boudjema, P. Brun, A. Pukhov, S. Rosier-Lees, P. Salati and A. Semenov, Comput. Phys. Commun. [**182**]{}, 842 (2011) \[arXiv:1004.1092 \[hep-ph\]\]. J. M. Alarcon, J. Martin Camalich and J. A. Oller, Phys. Rev. D [**85**]{}, 051503 (2012) doi:10.1103/PhysRevD.85.051503 \[arXiv:1110.3797 \[hep-ph\]\]. J. M. Alarcon, L. S. Geng, J. Martin Camalich and J. A. Oller, Phys. Lett. B [**730**]{}, 342 (2014) doi:10.1016/j.physletb.2014.01.065 \[arXiv:1209.2870 \[hep-ph\]\]. X.-L. Ren, L. S. Geng, J. Martin Camalich, J. Meng and H. Toki, JHEP [**1212**]{}, 073 (2012) doi:10.1007/JHEP12(2012)073 \[arXiv:1209.3641 \[nucl-th\]\]. X. L. Ren, L. S. Geng and J. Meng, Phys. Rev. D [**91**]{}, no. 5, 051502 (2015) doi:10.1103/PhysRevD.91.051502 \[arXiv:1404.4799 \[hep-ph\]\].
“Search for squarks and gluinos in events with isolated leptons, jets and missing transverse momentum at $\sqrt{s}=8$TeV with the ATLAS detector”, ATLAS-CONF-2013-062.
“Search for strong production of supersymmetric particles in final states with missing transverse momentum and at least three b-jets using 20.1fb$^{-1}$ of pp collisions at $\sqrt{s}=8$TeV with the ATLAS Detector”, ATLAS-CONF-2013-061.
“Search for new phenomena using final states with large jet multiplicities and missing transverse momentum with ATLAS in 20 fb$^{-1}$ of $\sqrt{s}=8$TeV proton-proton collisions”, ATLAS-CONF-2013-054.
“Search for direct third generation squark pair production in final states with missing transverse momentum and two b-jets in $\sqrt{s}=8$TeV pp collisions with the ATLAS detector”, ATLAS-CONF-2013-053.
The ATLAS collaboration, “Search for direct top squark pair production in final states with two leptons in $\sqrt{s}=8$TeV pp collisions using 20fb$^{-1}$ of ATLAS data”, ATLAS-CONF-2013-048.
“Search for squarks and gluinos with the ATLAS detector in final states with jets and missing transverse momentum and 20.3fb$^{-1}$ of $\sqrt{s}=$8TeV proton-proton collision data”, ATLAS-CONF-2013-047.
“Search for direct top squark pair production in final states with one isolated lepton, jets, and missing transverse momentum in $\sqrt{s}=8$TeV pp collisions using 21fb${-1}$ of ATLAS data”, ATLAS-CONF-2013-037.
“Search for direct production of the top squark in the all-hadronic ttbar + etmiss final state in 21fb$^{-1}$ of p-pcollisions at $\sqrt(s)=8$TeV with the ATLAS detector”, ATLAS-CONF-2013-024.
“Search for direct-slepton and direct-chargino production in final states with two opposite-sign leptons, missing transverse momentum and no jets in 20fb$^{-1}$ of pp collisions at $\sqrt{s}=8$ TeV with the ATLAS detector”, ATLAS-CONF-2013-049.
“Search for electroweak production of charginos, neutralinos and sleptons using leptonic final states in pp collisions at $\sqrt{s}=8$TeV”, SUS-12-022.
“Searches for electroweak production of charginos, neutralinos, and sleptons decaying to leptons and W, Z, and Higgs bosons in pp collisions at $\sqrt{s}=8$TeV”, SUS-13-006.
“Search for direct slepton and gaugino production in final states with hadronic taus and missing transverse momentum with the ATLAS detector in pp collisions at $\sqrt{s}=8$TeV”, ATLAS-CONF-2013-028.
“Search for direct gaugino production in events with three or more leptons and MET”, ATLAS-CONF-2013-035.
“Search for supersymmetry in events with four or more leptons and MET”, ATLAS-CONF-2013-036.
“Search for chargino and neutralino production in final states with one lepton, two b-jets consistent with a Higgs boson, and missing transverse momentum with the ATLAS detector in 20.3fb$^{-1}$ of $\sqrt{s}=8$TeV pp collisions”, ATLAS-CONF-2013-093.
“Search for electroweak production of charginos and neutralinos in final states with a Higgs boson in pp collisions at $\sqrt{s}=8$TeV”, SUS-13-017. J. Alwall [*et al.*]{}, JHEP [**1407**]{}, 079 (2014) doi:10.1007/JHEP07(2014)079 \[arXiv:1405.0301 \[hep-ph\]\]. J. Alwall, M. Herquet, F. Maltoni, O. Mattelaer and T. Stelzer, JHEP [**1106**]{}, 128 (2011) doi:10.1007/JHEP06(2011)128 \[arXiv:1106.0522 \[hep-ph\]\]. T. Sjostrand, S. Mrenna and P. Z. Skands, JHEP [**0605**]{}, 026 (2006) doi:10.1088/1126-6708/2006/05/026 \[hep-ph/0603175\]. J. de Favereau [*et al.*]{} \[DELPHES 3 Collaboration\], JHEP [**1402**]{}, 057 (2014) doi:10.1007/JHEP02(2014)057 \[arXiv:1307.6346 \[hep-ex\]\]. M. Drees, H. Dreiner, D. Schmeier, J. Tattersall and J. S. Kim, Comput. Phys. Commun. [**187**]{}, 227 (2014) doi:10.1016/j.cpc.2014.10.018 \[arXiv:1312.2591 \[hep-ph\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], JHEP [**1405**]{}, 071 (2014) doi:10.1007/JHEP05(2014)071 \[arXiv:1403.5294 \[hep-ex\]\].
G. Arnison [*et al.*]{} \[UA1 Collaboration\], Phys. Lett. B [**122**]{}, 103 (1983). doi:10.1016/0370-2693(83)91177-2
G. Aad [*et al.*]{} \[ATLAS Collaboration\], JHEP [**1411**]{}, 118 (2014) doi:10.1007/JHEP11(2014)118 \[arXiv:1407.0583 \[hep-ex\]\]. Y. Bai, H. C. Cheng, J. Gallicchio and J. Gu, JHEP [**1207**]{}, 110 (2012) doi:10.1007/JHEP07(2012)110 \[arXiv:1203.4813 \[hep-ph\]\]. G. Aad [*et al.*]{} \[ATLAS Collaboration\], Eur. Phys. J. C [**74**]{}, no. 6, 2883 (2014) doi:10.1140/epjc/s10052-014-2883-6 \[arXiv:1403.5222 \[hep-ex\]\].
[^1]: Throughout this work, we denote the mass eigenstates of the neutralinos by $\tilde{\chi}_i^0$ with $i$ ranging from 1 to 4 (5) for MSSM (NMSSM), and assume an ascending mass order for the $\tilde{\chi}_i^0$ by convention. With such an assumption, the lightest neutralino $\tilde{\chi}_1^0$ as the lightest supersymmetric particle (LSP) is regarded as the DM candidate.
[^2]: As was pointed out in [@Baer:2013gva], if the NMSSM is considered as the low energy realization of an (unknown) overarching ultimate theory, $\Delta_Z$ and $\Delta_h$ can be thought of as providing a lower bound on electroweak fine-tuning. Any parameter point with a low $\Delta_Z$ and $\Delta_h$ implies that the ultimate theory may be low fine-tuned at high energy scale. By contrast, if the point correspond to large $\Delta_Z$ and $\Delta_h$, the underlying theory must be fine-tuned.
[^3]: In our fit, we adopt a moderately wider range of the SM-like Higgs boson mass, $122 {\rm GeV} \leq m_h \leq 128 {\rm GeV}$, in comparison with the default uncertainty of $2 {\rm GeV}$ for $m_h$ in the package HiggsSignal. This is because $\lambda \gtrsim 0.4$ may induce a ${\cal{O}}(1 {\rm GeV})$ correction to $m_h$ at two-loop level [@Goodsell:2014pla; @Staub:2015aea], which is not considered in the NMSSMTools.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A nonlocal-in-time problem for the abstract Schrödinger equation is considered. By exploiting the linear nature of nonlocal condition we derive an exact representation of the solution operator under assumptions that the spectrum of Hamiltonian is contained in the horizontal strip of complex plane. The derived representation permits us to establish the necessary and sufficient conditions for the problem’s well-posedness and the existence of its mild, strong solutions. Furthermore, we present new sufficient conditions for the existence of solution which extend the available results to the case when some nonlocal parameters are unbounded. Two examples are provided.
**Keywords:** nonlocal problem, abstract time-dependent Schrödinger equation, well-posedness, solution operator, Dunford-Cauchy formula, zeros of polynomial, quantum computing, driven quantum systems.
author:
- |
Dmytro Sytnyk\
Institute of mathematics, National Academy of Sciences, Ukraine,\
[email protected]
- |
Roderick Melnik\
Wilfrid Laurier University, Waterloo, Canada,\
[email protected]
bibliography:
- |
%
\\bibpath/sytnyk.bib
- |
%
\\bibpath/nonlocal.bib
- |
%
\\bibpath/nonlocal\_schrodinger.bib
- |
%
%\\bibpath/ill\_posed.bib
- |
%
\\bibpath/matan.bib
- |
%
\\bibpath/root\_finding.bib
- |
%
\\bibpath/abstract\_schrodinger.bib
- |
%
%\\bibpath/AbstractCauchy.bib
- |
%
\\bibpath/operator\_calculus.bib
- |
%
\\bibpath/driven\_quantum\_systems.bib
- |
%
%%\\bibpath/comp\_math.bib
- |
%
%%\\bibpath/GMV.bib
- |
%
%%\\bibpath/denis.bib
- |
%
%%\\bibpath/caley\_transform.bib
- |
%
%%\\bibpath/software.bib
- |
%
%%\\bibpath/diff\_schemes.bib
- |
%
%%\\bibpath/fem.bib
- |
%
%%\\bibpath/lin\_sys.bib
- |
%
%%\\bibpath/algo.bib
- |
%
%%\\bibpath/kp.bib
- |
%
%%\\bibpath/led.bib
- '%.bib'
title: 'Linear nonlocal problem for the abstract time-dependent non-homogeneous Schrödinger equation'
---
Introduction
============
In the abstract setting the evolution of quantum system is governed by differential equation $$\label{SchrodEqt}
{i}\psi'_t-H\psi= i v(t), \quad t \in [0, T],
$$ which is called time-dependent non-homogeneous Schrödinger equation. Standard axiomatic approach to the quantum mechanics ensures that the state of the system described by a wave function $\psi(t) \in X$, is uniquely determined by and a given initial state $\psi_0$ $$\label{eq:SEIC}
\psi(0) = \psi_0.$$ This is achieved by requiring that the linear operator (Hamiltonian) $H:\ X \rightarrow X$ is self-adjoint in the Hilbert space $X$ and its domain $D\left (H\right ) \subseteq X$ is dense. Stone’s theorem states that in such a case there exists a strongly continuous unitary group $U(t) = {{\rm e}}^{-iHt}$ with generator $iH$ [@Fattorini1985]. The function $\psi(t)$ is called a mild solution of , , if it satisfies the equation $$\label{eq:MildSolSchrodCP}
\psi(t) = U(t) \psi(0) + {\int\limits}_{0}^{t}U\left(t-s\right) v(s)ds, \quad t \in [0, T].$$ Substitution of the initial data from into this general solution representation lead us to the usual propagator formula $\psi(t) = U(t) \psi_0$ for the solution of , with $v(t) \equiv 0$.
In this work we consider a nonlocal generalization of condition : $$\label{eq:linear_nc}
\psi(0)+{\sum\limits}_{k=1}^{n}\alpha_k \psi(t_k) =\psi_1.$$ For the fixed state $\psi_1 \in X$ this condition is determined by the set of parameters $ 0<t_1<t_2<\ldots < t_n \leq T,\ \alpha_k \in \C$ which will be called the *parameters of nonlocal condition*. Aside from the standard initial condition it generalizes other important condition types, such as periodic conditions $\psi(0) = \psi(t_1)$ and Bitsadze-Samarskii conditions $\psi(0)+\alpha_1 \psi(t_1)=\alpha_2 \psi(t_2)$ (e.q. [@Ashyralyev2008]). Formula can be also viewed as approximation to a more general nonlinear condition $\psi(0) + g\left (t_1, \ldots t_k, \psi\left ( \cdot \right )\right ) = 0$ for a suitably defined function $g\left (t_1, \ldots t_k, \cdot \right ):\ X \rightarrow X$. Among other applications, nonlocal problem , is important for the theory of driven quantum systems, where one is interested in a way to recapture specific nonlocal behavior of solution $\psi(0) =\alpha_1 \psi(t_1)+\psi_1$ by changing the properties of driving potential $p(t)$ from the Hamiltonian $H = H_0 + p(t)$. To stay within the classical formulation , this theory operates upon assumption that $p(t)$ is periodic [@KuwaharaMoriSaito2016; @LazaridesDasMoessner2015; @VerdenyMintert2015]. Then a predictable nonlocal-in-time behavior of the system follows from the Floquet theorem [@daners1992abstract]. The case of non-periodic $p(t)$ is much harder to treat, since the Floquet theory can not be applied. It is our belief that the nonlocal formulation is a viable alternative to other proposed generalizations of periodic quantum driving that are currently under research [@VerdenyPuigMintert2016; @YanLiYangEtAl2015]. The above mentioned two-point nonlocal condition with $\psi_1\neq0$ can also be thought of as a generalization of the renowned Rabi problem [@Gardas2013; @LeHur2016] used in the modern quantum computing for state preparation and information processing [@LeBellac2006]. A nonlinear problem similar to , was analyzed in [@bunoiu2016vectorial], where the authors were motivated by the study of Bose-Einstein condensates.
In spite of the increasing importance, a surprisingly little is known about the solution of , . This problem was studied in [@ashyralyev2008nonlocal], using the Hilbert space methods. For the self-adjoint $H$ it was proved that the condition $$\label{eq:NCNS_ash_sufff_cond}
{\sum\limits}_{k=1}^{n}\left |\alpha_k \right | <1$$ is sufficient for the existence of solution to , , when $\psi_1$ is a smooth enough vector with respect to $H$ (see [@ashyralyev2008nonlocal] for the details). The same condition appeared earlier in [@Byszewski1992], where a more general nonlocal problem for the first order equation with sectorial operator coefficient in a Banach space was considered. In the current work we focus on a following generalization of the condition from [@ashyralyev2008nonlocal; @Byszewski1992]: $$\label{estLiang2002}
{\sum\limits}_{k=1}^{n} |\alpha_k|e^{d t_k} \leq 1,$$ developed in [@NonlocalAbsNonLinNtouyas1997]. Here $d$ is a half-height of the strip containing the spectrum of operator $H$ defined in the Banach space $X$. In the course of the work we show that inequality represents only a fraction of the parameter space where problem , is well-posed and have a mild solution defined by . More generally, we establish new necessary and sufficient conditions for the existence of solution to , which can be verified for any given set of $\alpha_k,t_k$ from . In addition to that, we derive several versions of sufficient conditions for the solvability of the given nonlocal problem which extend the region of admissible $\alpha_k$ outside the manifold governed by .
The paper is organized as follows. In Section \[sec:hs\_operator\_calculus\] we introduce a notion of strip-type operators $H$ acting on Banach space $X$ and review the functional calculus of such operators. Our aim is to specify the class of $H$ such that the propagator $U(t)$ is well-defined and can be represented via the Dunford-Cauchy formula. Section \[sec:NCNS\_reduction\] is devoted to the analysis of solution existence. We start with the reduction of nonlocal problem , to classical Cauchy problem , . Then the operator calculus of Section \[sec:hs\_operator\_calculus\] is applied to study the obtained solution operator of nonlocal problem. Theorem \[thm:NCNS\_exist\_mild\] gives the necessary and sufficient conditions for the existence and uniqueness of mild solution to , . Corollaries \[thm:NCNS\_exist\_strong\] and \[thm:NCNS\_well\_posed\] concern the existence of strong solution and the well-posedness of the given problem. Apart from several simple cases, the conditions on parameters $\alpha_k,t_k$, mentioned in Theorem \[thm:NCNS\_exist\_mild\], can be verified if the values of these nonlocal parameters are specified. In Section \[sec:NCNS\_polynomial\_reduction\] we further adopt the technique from [@nonloc_exMVS2014], which, when suited with the properly chosen conformal mapping (adjusted to the spectral-strip parameter $d$), permits us to reduce the question of solution’s existence to the question about the location of roots for a certain polynomial associated with the nonlocal condition. This, in turn, enables us to obtain the conditions for the existence and uniqueness of the solution to , stated in terms of the constraints on $t_k,\ \alpha_k$ (Theorems \[thm:NCNS\_suf\_cond\_root\_est\], \[thm:NCNS\_criteria\]). Finally, we compare newly derived conditions against , using the three-point nonlocal problem as a model example.
Functional calculus of strip-type operators {#sec:hs_operator_calculus}
===========================================
With intent to study problem , in a Banach space setting, in this section, we review necessary facts from the holomorphic functional calculus for operators with the spectrum in a horizontal strip [@Haase2007]. A densely defined closed linear operator $H$ with the domain $D(H) \subseteq X$, whose spectrum belongs to the set $$\label{eq:SpHalfStrip}
\Sigma_d = \left\{
z=x + i y \middle|\ x,y \in \R,\ |y|\leq d
\right\},$$ and the resolvent $R\left(z,H\right) \equiv (zI-H)^{-1}$ satisfies $$\label{eq:ResHalfStrip}
\left \|R\left(z,H\right)\right \|\leq \frac{M}{|\Im{z}|-d},\quad z \in \Omega\setminus \Sigma,\ \Sigma \subset \Omega,$$ is called a strip-type operator of the height $2d >0$. Next we define the rule to interpret operator functions. Let $f(z)$ be a complex valued function analytic in the neighborhood $\Omega$ of the spectrum $\Sigma(H) \subset \C$ and $|f(z)|<c_f\left (1+|z|\right )^{-1-\delta}$, for $\delta >0$. Suppose that there exists a closed set $\Phi \subset \Omega$ with the boundary $\Gamma$ consisting of a finite number of rectifying Jordan curves, then the operator function $f(H)$ can be defined as follows $$\label{reprDunford}
f(H) x=\frac{1}{2\pi i} {\int\limits}_{\Gamma}f(z) R(z,H)x dz.$$ This formula yields an algebra homomorphism between the mentioned class of holomorphic functions and the algebra of bounded operators on $X$. Besides, any two valid functions of the same operator commute. Unfortunately, Dunford-Cauchy integral can not be used straight away to define the propagator, because $|{{\rm e}}^{-iz}|$ will not vanish as $z\rightarrow\infty$ on $\Gamma$. Assume that there exists a so-called *regularizer* function $\epsilon(z)$ such, that both $e(H)$ and $ef (H)$ are well defined in terms of and $e(H)$ is injective. Then the formula $$\label{eq:func_calc_regularizer}
f(A) = e^{-1}(H) ef (H)$$ is used to define $f(H)$ for a class of functions wider than the natural function calculus defined by alone. By setting $e(z)=(\lambda - z)^{-1-\delta}$ with $|\Im{\lambda}|>d$ we ensure that $f(H)x$ is well defined and bounded, whenever $f(z)$ is bounded in $\Omega$ and $e^{-1}(H)x$ exists. In other words, the propagator $U(t)$ is bounded linear operator with the domain $x \in D(H^{1+\delta})$. By using the closed graph theorem [@Munkres2005], $U(t)$, $t \in \R$ can be extended to the bounded operator on $X$ when the set $D(H^{1+\delta})$ is dense in $X$. For more details on the construction and properties of the functional calculus for strip-type operator we direct the reader to [@haase2006functional Chapter 4].
Reduction of nonlocal problem to classical Cauchy problem {#sec:NCNS_reduction}
=========================================================
We depart from the general solution formula , with $\psi(0)$ supplied by $$\label{eq:linear_NCNS}
\psi(t)=U(t)\left(\psi_1-{\sum\limits}_{k=1}^{n}\alpha_k \psi(t_k) \right)+{\int\limits}_{0}^{t}U\left(t-s\right) v(s)ds,$$ that is valid for the strip-type operator $H$ under assumptions of Section \[sec:hs\_operator\_calculus\]. To get the exact representation for $\psi(t)$ one needs to factor out the unknown $\psi(t_k)$, $k=\overline{1,n}$ from the above formula. We define $w \equiv {\sum\limits}_{k=1}^{n}\alpha_k \psi(t_k)$ and then formally evaluate this expression using as a representation for $\psi(t)$. It leads to the equation $$w = {-{\sum\limits}_{i=1}^n \alpha_i U(t_i) w} +{\sum\limits}_{i=1}^n \alpha_i U(t_i) \psi_1+ {\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds.$$ By denoting $B = I +{\sum\limits}_{i=1}^n \alpha_i U(t_i)$ we rewrite this equation as follows $$\label{eqnoneq1}
B w = B \psi_1 -\psi_1 +{\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds.$$
At this point it is clear that equation can be solved for $w$ with any combination of $\psi_1$ and $v(t)$ if and only if the operator function $B$ posses the inverse $B^{-1}$. In such case the substitution $$\label{eq:w_repr}
w = \psi_1- B^{-1} \psi_1 + B^{-1}{\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds$$ into yields a representation of the general (mild) solution to nonlocal problem , $$\label{bp1IntRed}
\begin{split}
\psi(t)=& U(t)\left(B^{-1}\psi_1 -B^{-1}{\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds\right)+{\int\limits}_{0}^{t}U\left(t-s\right) v(s)ds.\\
\end{split}$$ Now we can formalize our previous analysis as a theorem.
\[thm:NCNS\_exist\_mild\] Let $H$ be a strip-type operator with the spectrum $\Sigma$, having nonempty point-spectrum component, and the domain $D(H^\delta)$ is dense in $X$ for some $\delta > 1$. The mild solution of nonlocal problem , exists and is unique for any $\psi_1 \in X$, $v \in L^1((0;T),X)$ if and only if all the zeros of the entire function $$\label{zerosExpI}
b(z)=1+\sum_{k=1}^n{\alpha_k e^{(-i t_k z)}}, $$ associated with , are contained in the interior of the set $\mathbb{C} \backslash \Sigma$.
We prove necessity first. A solution to the given nonlocal problem satisfies differential equation , hence general representation is valid for such solution with any given combination of $\psi (0)$, $v(t)$. Upon setting $v(t) = 0$ in this representation, we substitute it into to get the equation $$\label{eq:nc_zero_v}
B\psi(0) = \psi_1$$ with respect to $\psi(0)$. Suppose that the function $b(z)$ has a root $z_0 \in \Sigma$ which belongs to the point spectrum of $H$, with $\varphi \neq 0$ being the corresponding eigenstate. Now, we pick a bounded sequence $\left\{ \psi_{1k}\right\}_{k=1}^\infty$, so that $\psi_{1k} \in D(H^\delta)$, $\psi_{1k} \neq \varphi$ and $\psi_{1k} \rightarrow \varphi$ strongly. Such sequence always exists since the domain $D(H^\delta)$ is dense in $X$. By the theorem’s premise, for any $\psi_1$ there should exist a corresponding bounded state $\psi(0)$ satisfying . To show that this is not true for $\psi_1 = \varphi$ we, first, evaluate $B\varphi$ via the Dunford-Cauchy integral $$B\varphi =\frac{1}{2\pi i}{\int\limits}_{\Gamma} b\left (z\right )R(z,H) \varphi dz = \frac{1}{2\pi i}{\int\limits}_{\Gamma} \frac{b\left (z\right )}{z-z_0} \varphi dz = b(z_0) = 0,$$ and then apply the general inequality $\|B^{-1}\| \geq \frac{1}{\|B\|}$ to $B^{-1}\psi_{1k}$: $$\lim\limits_{k \rightarrow \infty}\|B^{-1}\psi_{1k}\| \geq \lim\limits_{k \rightarrow \infty}\frac{1}{\|B\psi_{1k}\|} = \infty .$$ Next we prove sufficiency. Assume that all the zeros of $b(z)$ belong to the interior of $\mathbb{C} \backslash \Sigma$. By using the operator function calculus from Section \[sec:hs\_operator\_calculus\] we define $$\label{reprDelta}
B^{-1}\varphi =\frac{1}{2\pi i}{\int\limits}_{\Gamma} \frac{1}{b\left (z\right )} R(z,H) \varphi dz,$$ for any $\varphi \in X$. The contour $\Gamma$ satisfying the requirements of exists, since $1/b(z)$ is holomorphic in the neighborhood of $\Sigma$. Formula , the condition $v \in L^1((0;T),X)$ and Lemma 5.2 from [@Fattorini1985] guaranty that the state $\psi_0$ given by $$\label{eq:initial_state_NCNS}
\psi_0 = B^{-1}\psi_1 -B^{-1}{\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds,$$ is well-defined for any combination of $v(t)$ and $\psi_1$ fulfilling the theorem’s assumptions. That, in turn, implies a well-definiteness of $\psi(t)$ given by formula . To prove that $\psi(t)$ is a solution to nonlocal problem , we need to check if it satisfies . This is trivially true, since is transformed into via the direct manipulation with initial state using : $$\psi_0 = \psi_1 - \psi_1 + \psi_0 = \psi_1 - w = \psi_1-{\sum\limits}_{k=1}^{n}\alpha_k \psi(t_k).$$ The uniqueness of solution to the given nonlocal problem follows from the linear nature of both differential equation and nonlocal condition as well as from the fact that $\psi(t) \equiv 0$ when $\psi_1$ and $v(t)$ are equal to zero simultaneously.
We highlight, that the proof of sufficiency relies only on the assumptions needed for the existence of operator function $B(H)\psi$ for any $\psi \in X$. These assumptions do not include the requirement of $H$ having at least one eigenvector, that is essential to prove the necessity of Theorem \[thm:NCNS\_exist\_mild\]. The theorem concerns the existence and uniqueness of the solution for any possible combination of $\psi_1$ and $v(t)$. It does not discount the existence of solutions other than for some specific combination of $\psi_1$ and $v(t)$. Namely, if the nonzero initial data $\psi_1, v(t)$ is chosen in such a way that the right-hand side of $$\label{eq:nc_nonzero_v}
B\psi(0) = \psi_1 - {\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds$$ is zero and there is a non-empty intersection between the set of roots of $b(z)$ and the spectrum of $H$, then one can construct a whole family of non-trivial solutions to , . Because, as we have shown in the proof, every eigenstates of $H$ for which the corresponding eigenvalue coincides with the root of $b(z)$, will satisfy with the zero right-hand side.
It should also be noted, that by its structure, formula resembles representation of the solution to classical Cauchy problem , . More precisely, the following is true.
\[thm:NCNS\_equiv\_Cauchy\] Assume that the requirements of Theorem \[thm:NCNS\_exist\_mild\] are fulfilled, then the mild solution of nonlocal problem , is equivalent to the solution of classical Cauchy problem , represented by , with the initial state $\psi_0$ defined by .
The correspondence between the solution of nonlocal problem and the solution of the classical Cauchy problem permits us to establish other important properties of , .
\[thm:NCNS\_exist\_strong\] Assume that in addition to the requirements of Theorem \[thm:NCNS\_exist\_mild\] on $H$, both $b(z)$, $\psi_1$ belong to $D(H)$ and either one of two following conditions is satisfied:
[a)]{} $v(t) \in D(H)$ and $v(t)$, $H v(t)$ are continuous on $[0, T]$, or
[b)]{} $v(t)$ is continuously differentiable on $[0, T]$.
Then is a strong (genuine) solution of nonlocal problem , .
We proceed by reducing the proof to the corresponding results on the genuine solution of the classical Cauchy problem [@Fattorini1985 Lemma 5.1]. In order to achieve that it is enough to show that the theorem’s assumptions imply $\psi_0 \in D(H)$ or, which is equivalent, that $H\psi_0$ is well defined. We depart from and use the above-mentioned properties of function calculus for strip-type operators: $$\begin{split}
H\psi_0 = &HB^{-1}\left (\psi_1 - {\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)v(s) ds\right ) \\
= &B^{-1}H\psi_1 -B^{-1}{\sum\limits}_{i=1}^n \alpha_i {\int\limits}_0^{t_i} U(t_i-s)Hv(s) ds.
\end{split}$$ The first term in the last formula is well defined because $\psi_1 \in D(H)$ and there always exists a sequence of states from $D(H^{\delta})$ with $\psi_1$ as a limit, such that $B^{-1}H$ is bounded on the elements of that sequence. By the same token we can show the well-definiteness of the second term, under assumption that *a)* is true. The case of *b)*, as well as the rest of the proof, literally repeats the proof of the mentioned Lemma 5.1 from [@Fattorini1985], and thus will be omitted here.
The conditions necessary for the existence of the strong solution are closely related to the well-posendess of , . The evolution problem is called uniformly well-posed in $t \in [0, T]$ (see Section 1.2 of [@Fattorini1985]), if and only if the strong solution exists for a dense subspace of the initial data and the solution operator is uniformly bounded in $t$ on compact subsets of $[0, T]$.
\[thm:NCNS\_well\_posed\] Let $H$ be an operator satisfying the assumptions of Theorem \[thm:NCNS\_exist\_mild\]. The nonlocal problem , is uniformly well-posed in $t \in \R$ for any bounded $t_k\in [0, T]$, $\alpha_k \in \C$ if and only if all the zeros of $b(z)$ defined by are separated from $\Sigma$ .
In Corollary \[thm:NCNS\_exist\_strong\] we’ve already identified the dense subset $D(H)$ of $X$ such that for any $\psi_1 \in D(H)$ there exists a genuine solution of , . Assumptions on the parameters of nonlocal condition imply the boundedness of $B^{-1}$. In Section \[sec:hs\_operator\_calculus\] we mentioned that $U(t)$ is bounded as well, thus the solution operator from is bounded. To conclude the proof we recall that the propagator $U(t)$ forms the group for $t \in \R$, hence the bounded solution operator is also uniformly bounded [@Fattorini1985 Theorem 2.1].
\[ex1\] Let us consider a two point version of nonlocal problem , . In such simple case, nonlocal condition takes the form $$\label{ex:nc_two_point}
u(0)+\alpha_1 u(t_1) =u_0,\quad t_1>0.
$$ Here we assume that $H$ has all the properties mentioned in Theorem \[thm:NCNS\_exist\_mild\]. To determine the location of zeros of $b(z)$ we need to solve the equation $$1 +\alpha_1e^{-zi t_1} = 0,$$ assuming that $\alpha_1 \in \C$ and $t_1 \in [0, T]$ are given. It has an infinite number of solutions $z_m$ $$\begin{split}\label{kerB1p}
z_m = & -\frac{1}{it_1}\ln\left(-\frac{1}{\alpha_1}\right)=
\\
= &\frac{1}{t_1}\left[\operatorname{Arg}\left(\frac{1}{\alpha_1}\right)+2\pi m +i\ln\left|\frac{1}{\alpha_1}\right|\right], \quad m\in \Z.
\end{split}$$ Here $\operatorname{Arg}\left(\cdot\right)$ stands for a principal value of the argument of complex number. The zeros $z_m$ are situated on the line, where the imaginary part $\Im{z}= \ln{\left |1/\alpha_1\right |}/t_1$ is constant. They will belong to $\C\setminus\Sigma$ if $\left |\Im{z}\right |$ is greater than the half-height $d$ of the spectrum $\Sigma$ defined by . Consequently, the solution of , exists if and only if $$\label{eq:NCN_est_nc}
|\alpha_1| < e^{-t_1d},\quad \text{or}\quad |\alpha_1| > e^{t_1d}.\\$$ The given nonlocal problem is well-defined for any $\alpha_1 \in \C$, except for the complex numbers lying in the annulus $e^{-t_1d}\leq |\alpha_1| \leq e^{t_1d}$.
It is important to note that constraints enforce $|\alpha_1|\neq 1$. That requirement can be relaxed for some $\psi_1, v(t)$ if the spectrum of $H$ is disjoint in the neighborhood of $\R$. Another unique feature of the two-point problem , is expressed by one’s ability to write the closed-form solution , without specifying $\alpha_1$ beforehand. It becomes impossible for the general case of multi-point nonlocal condition , where one must rely on the numerical procedures to solve $b(z)=0$ and for that reason predefine the parameters of nonlocal condition. For many applications of , with $n>1$ this is not enough as one still would like to have some apriori information about the admissible set of $\alpha_k$ rather than simply check the existence of solution for a fixed sequence $\alpha_k$, $k=\overline{1,n}$.
Zeros of $b(z)$ and associated problem for polynomials {#sec:NCNS_polynomial_reduction}
======================================================
To find a way around the direct solution of $b(z)=0$, $n>1$ we start with a general observation suggested by Example \[ex1\]. The function $b(z)$ can be arbitrary closely approximated by a periodic function ${b^{\star}(z)}\equiv 1+\sum_{k=1}^n{\alpha_k e^{(-i {t^{\star}_k} z)}}$, where each ${t^{\star}_k}$ is the rational approximation to the corresponding real number $t_k$, $k=\overline{1,n}$. The function $b^{\star}(z)$ better suits our needs than $b(z)$, because the equation $b^{\star}(z)=0$ can always be reduced to the polynomial root finding problem.
Let $$t_k= \frac{\lambda_k}{\mu_k}, \quad \lambda_k \in \mathbb{Z},\quad \mu_k \in \mathbb{N},$$ we set $c_k=\frac{Q \lambda_k}{\mu_k}$, where $$Q = \frac{\mbox{LCM}(\mu_1,\mu_2, \ldots, \mu_n)}{\mbox{GCD}(\lambda_1,\lambda_2, \ldots, \lambda_n)}$$ is the ratio of the least common multiple (LCM) and the greatest common divisor (GCD) of the numerators and denominators of $t_k$ correspondingly. A substitution $$\label{eq:NCNS_tranfs}
\Phi:\ u=\exp{\left (-iz/Q \right )}$$ transforms the original problem about the location of zeros of $b(z)$ in $\C\setminus\Sigma$ into the problem about the location of zeros of a polynomial $$\label{eq:NCNS_zeros_pol}
r(u)=1+{\sum\limits}_{k=1}^{n}\alpha_k u^{c_k}$$ in the exterior of an annulus $$\Upsilon:\quad e^{-d/Q}\leq |u| \leq e^{d/Q}, \quad u \in \C.$$ The polynomial root finding problem for $r(u)=0$ is extensively studied (see [@Milovanovic2000a; @Milovanovic2000], as well as [@Sendov1994625; @Henrici1974v1]). Polynomial $r(u)$ has exactly $c_n$ roots $u_k$ over $\C$. Their closed form representation exists for $c_n\leq 4$. So, now we technically can write the exact solvability conditions for , in terms of $\alpha_k$ for $k$ up to 4. More importantly, it is possible to avoid the full solution of $r(u)=0$ altogether whilst checking $u_k \in \C\setminus\Upsilon$: $$\label{eq:NCNS_ext_annulus}
|u_k|< e^{-d/Q} \lor |u_k| > e^{d/Q}, \quad k=\overline{1,c_n}.$$ The shape of $\Upsilon$ suggests that we focus on a subclass of available root finding methods with results stated in the form of bounds . Among those, we choose three effective complex root bounds (see [@Batra2016; @Rump2003; @PhdThSytnykEN] for the discussion and comparisons) for $P(u)={\sum\limits}_{k=0}^{N}a_k u^k$. We have ordered them by the increasing computational complexity. Each of the following bounds has been reformulated as a double estimate to better fit .
([@Milovanovic2000a], Theorem 2.4)\[lem:zpol\_2\] The zeros of $P(u)$ satisfy the following inequalities: $$\begin{split}
|u|\leq \left(1+\left(\frac{M_s}{|a_N|}\right)^q \right)^{1/q},\quad
|u|\geq \frac{|\a_0|}{\left(|\a_0|+M_s^q \right)^{1/q}}, \\
M_s=\left(\sum\limits_{k=1}^N |\a_k|^s\right)^{1/s},\quad s,\ q \in \R_{>1},\quad \frac{1}{s}+\frac{1}{q}=1
\end{split}$$
The next estimate is due to M. Fujiwara [@Fujiwara1916]. It is the nearly optimal homogeneous bound in the space of polynomials [@Batra2016]:
\[lem:zpol\_3\] All zeros of $P(u)$ satisfy the inequalities $$\begin{split}
|u| \leq 2\max\left \{\left |\frac{\a_0}{2\a_N}\right |^{1/N}, \left |\frac{\a_1}{\a_N}\right |^{1/(N-1)}, \ldots, \left |\frac{\a_{N-1}}{\a_N}\right |\right \},\\
|u| \geq \frac{1}{2}\min\left \{\left |\frac{2\a_N}{\a_0}\right |^{1/N}, \left |\frac{\a_N}{\a_1}\right |^{1/(N-1)}, \ldots, \left |\frac{\a_N}{\a_{N-1}}\right |\right \},
\end{split}$$ where $1/0=+\infty$.
The third estimate, originally proved by H. Linden [@Linden1998], gives bounds on the real and imaginary part of zeros separately. It has been adapted in [@PhdThSytnykEN] to fit within the framework studied here.
\[lem:zpol\_4\] All zeros of $P(u)$ satisfy the double estimate $\max\{V_1^{-1},V_2^{-1}\} \leq |u|\leq \min\{V'_1,V'_2\}$, where $$V_1=\cos{\frac{\pi}{N+1}}+\frac{|\alpha_N|}{2|\alpha_0|}\left(\left|\frac{\alpha_1}{\alpha_N}\right|+\sqrt{1+{\sum\limits}_{k=1}^{N-1}\left|\frac{\alpha_k}{\alpha_N}\right|^2}\right)$$ $$\begin{gathered}
V_2=\frac{1}{2}\left(\left|\frac{\alpha_1}{\alpha_0}\right|+\cos\frac{\pi}{N}\right)\\
+\frac{1}{2}\left[\left(\left|\frac{\alpha_1}{\alpha_0}\right|-\cos\frac{\pi}{N}\right)^2+
\left(1+\left|\frac{\alpha_N}{\alpha_0}\right|\sqrt{1+{\sum\limits}_{k=2}^{N-1}\left|\frac{\alpha_k}{\alpha_N}\right|^2}\right)^2\right]^{1/2}
\end{gathered}$$ and $V'_i$ is obtained from $V_i$ by the substitution $\a_k =\a_{N-k}$, $k=\overline{0,N}$, $i=1,2$.
Now we are in the position to formulate our next result.
\[thm:NCNS\_suf\_cond\_root\_est\] Suppose that operator $H$ from satisfies the assumptions of Theorem \[thm:NCNS\_exist\_mild\] and all $t_k$ in are rational numbers. If at least one bound from Lemmas \[lem:zpol\_2\] - \[lem:zpol\_4\] for polynomial induce , then the nonlocal problem , has the following properties:
1. \[NCNS\_prop1\] it is uniformly well-posed in $t \in \R$;
2. for any $\psi_1 \in X$, $v \in L^1((0;T),X)$ there exists mild solution with the characteristics mentioned in Theorem \[thm:NCNS\_exist\_mild\];
3. \[NCNS\_prop3\]solution will also be strong if $\psi,v(t)$ satisfy either of the requirements a) or b) from Corollary \[thm:NCNS\_exist\_strong\].
If the zeros $u_k$ of obey , their images $$z_k = \Phi^{-1}(u_k) = Q\left[\operatorname{Arg}\left(u_k\right)+2\pi m +i\ln\left|u_k\right|\right],$$ are clearly in the interior of $\C\setminus\Sigma$ no matter what is the value of $m\in \Z$. The application of Theorem \[thm:NCNS\_exist\_mild\] and Corollaries \[thm:NCNS\_exist\_strong\],\[thm:NCNS\_well\_posed\] concludes the proof.
The result of Theorem \[thm:NCNS\_suf\_cond\_root\_est\] can be turned into criteria by enforcing the necessary and sufficient conditions for the validity of derived via the Schur-Cohn algorithm [@Henrici1974v1 p. 493]. For a given polynomial $r(u)$ the algorithm produces a set of up to $2 c_n$ inequalities, that are polynomial in $\alpha_k$, $k=\overline{1,n}$. These inequalities need to be valid simultaneously in order for the Schur-Cohn test to pass [@Henrici1974v1 Thm. 6.8b]. The precise result is stated as follows.
\[thm:NCNS\_criteria\] Suppose that operator $H$ from satisfies the assumptions of Theorem \[thm:NCNS\_exist\_mild\] and all $t_k$ in are rational numbers. Nonlocal problem , has properties \[NCNS\_prop1\]–\[NCNS\_prop3\] of Theorem \[thm:NCNS\_suf\_cond\_root\_est\] if and only if the polynomials $b(e^{d/Q}u)$, $u^{c_n} b(e^{-d/Q}u)$ pass the Schur-Cohn test for the given set of parameters $\alpha_k \in \C$, $k=\overline{1,n}$ from .
The substitution $u = e^{d/Q}u'$ ($u = u'^{c_n} b(e^{-d/Q}u')$) transforms right (left) inequality from into the inequality $u'_k > 1$. In both cases the validity of the last inequality is checked by the Schur-Cohn test [@Henrici1974v1 Thm. 6.8b]. “If” clause of Theorem \[thm:NCNS\_exist\_mild\] along with Corollaries \[thm:NCNS\_exist\_strong\],\[thm:NCNS\_well\_posed\] assures the sufficiency. Mapping $\Phi$ is a bijection of the vertical strip $|\Im{z}|\leq \pi \frac{\mbox{GCD}(\mu_1,\mu_2, \ldots, \mu_n)}{\mbox{LCM}(\lambda_1,\lambda_2, \ldots, \lambda_n)}$ onto $\C$. The strip’s height equals to the period of $b(z)$. This fact guaranties the necessity via application of the “only if” clause of Theorem \[thm:NCNS\_exist\_mild\] and Corollaries \[thm:NCNS\_exist\_strong\],\[thm:NCNS\_well\_posed\].
It remains to study the following question: what happens when some of $t_k$ are irrational? Consider an approximation $b^{\star}(z)$ of $b(z)$ mentioned above. If $t^{\star}_k \rightarrow t_k$, $k=\overline{1,n}$ the function $b^{\star}(z)$ uniformly converges to $b(z)$ on the compact subsets of the open set containing $\Sigma$. Hurwitz theorem [@Henrici1974v1 Corollary 4.10f] provides the means to claim that all zeros of $b(z)$ lies in the interior of $\C\setminus\Sigma$, if that is true for $b^{\star}(z): t^{\star}_k \rightarrow t_k$. The degree of a polynomial $r^{\star}(u)$ corresponding to $b^{\star}(z)$ grows to $\infty$ when $t^{\star}_k \rightarrow t_k$ and this $t_k$ is irrational. But, its coefficients $\alpha_k$ are not affected by the increase of $c_n^{\star}$. This keeps the root estimates from Lemmas \[lem:zpol\_2\]–\[lem:zpol\_4\] meaningful. As a result we have arrived at the following corollary.
Assume that for every $k=1,\ldots, n$ the sequence of rational numbers $\left \{t^{\star}_{kl}\right \}_{l=1}^\infty$, is such that $\lim\limits_{l \rightarrow \infty} t^{\star}_{kl}=t_k$. If the conditions of Theorem \[thm:NCNS\_suf\_cond\_root\_est\] regarding the roots of $r^{\star}_l(u)$ associated with $t^{\star}_{kl}$, $k=\overline{1,n}$ are fulfilled for all $l>0$, then the rest of theorem’s statement remains valid for $t_k \in \R$.
Now, we would like to compare the conditions on $\alpha_{1},\alpha_{2} \in \R$ obtained with help of Theorems \[thm:NCNS\_suf\_cond\_root\_est\], \[thm:NCNS\_criteria\] against the previously known condition .
Let us consider a three point nonlocal condition $$u(0)+\alpha_1 u(t_1)+\alpha_2 u(t_2)=0,\quad t_1,t_2 >0.$$ For simplicity we set $t_1=1,\ t_2=2$ and consider the non-zero spectral half-height $d=\pi/40$. Then, the equation $b(z)=0$ is reduced to $1+\alpha_1 u + \alpha_2 u^2=0$. As shown in Fig. \[fig:NCNS\_est\_comp\] b), the exact conditions on $\alpha_{1},\alpha_{2}$ calculated by Theorem \[thm:NCNS\_criteria\] (the Schur-Cohn algorithm) $$\label{eqintshura2}
\left[
\begin{array}{l}
\begin{cases}
|a_{{2}}|^{2} < {{\rm e}}^{-4d},\\
{{\rm e}}^{4d}|a_{{1}}|^{2}|a_{{2}}|^{2}-{{\rm e}}^{6d}|a_{{2}}|^{4}-2 {{\rm e}}^{4d}|a_{{2}}|\left (|a_{{1}}|^{2}-|a_{
{2}}|\right) +|a_{{1}}|^{2} < {{\rm e}}^{-2d}
\end{cases}
\\
\begin{cases}
|a_{{2}}|^{2} > {{\rm e}}^{4d},\\
{{\rm e}}^{-4d}|a_{{1}}|^{2}|a_{{2}}|^{2}-{{\rm e}}^{-6d}|a_{{2}}|^{4}-2 {{\rm e}}^{-2d}|a_{{2}}|\left (|a_{{1}}|^{2}-|a_{
{2}}|\right) +|a_{{1}}|^{2} > {{\rm e}}^{2d}
\end{cases}
\end{array}
\right.$$ lead to a considerably wider class of admissible pairs $\left (\alpha_{1},\alpha_{2}\right )$ than those obtained by .
[c]{}
[schrod\_eq\_3\_point\_t1\_1\_t2\_2\_d\_pi\_40\_thm2\_est\_color\_large1.pdf]{} (5,5)[**a)**]{} (90,52)[$\alpha_1$]{} (53,90)[$\alpha_2$]{}
[schrod\_eq\_3\_point\_t1\_1\_t2\_2\_d\_pi\_40\_thm3\_est\_color\_large.pdf]{} (5,5)[**b)**]{} (90,52)[$\alpha_1$]{} (53,90)[$\alpha_2$]{}
In fact, the second system of inequalities from gives rise to the unbounded region (union of two unbounded sets depicted in Fig. \[fig:NCNS\_est\_comp\] b) in the space of parameters $\alpha_{1},\alpha_{2} \in \R$, meanwhile the solutions of are strictly bounded in $|\alpha_1|, |\alpha_2|$ (the interior of the rhombic region in Fig. \[fig:NCNS\_est\_comp\] b). They lay within the isosceles triangle which acts as graphical solution of the first system of inequalities in . The gap between this triangle and the two other regions containing the solutions of shortens when $d\rightarrow 0$, and in the limit is described by $|\alpha_2|=1$. Comparison of generalized condition from [@ashyralyev2008nonlocal; @Byszewski1992] and the sufficient conditions provided by Theorem \[thm:NCNS\_suf\_cond\_root\_est\] (depicted in Fig. \[fig:NCNS\_est\_comp\] a) unveils that performs better than the inner circle estimates of Lemmas \[lem:zpol\_2\]–\[lem:zpol\_4\] (the part $\C \setminus \Upsilon$ defined by the first inequality in ). Therefore, when it comes to the apriori estimates on the parameters of nonlocal condition, we advice to use the combination of and the part of Theorem \[thm:NCNS\_suf\_cond\_root\_est\] which implies $|u_k| > e^{d/Q}$.
Conclusions {#conclusions .unnumbered}
===========
We established exact dependence of the solution to problem , on the parameters of nonlocal condition, derived the well-posedness criteria, and proved the theorems regarding existence of the problem’s mild (strong) solution. The conditions on existence of the solution to the given nonlocal problem obtained here, generalize other available results [@ashyralyev2008nonlocal; @NonlocalAbsNonLinNtouyas1997] beyond the case of $\alpha_k$ bounded by .
To illustrate the applicability of the obtained results we use them to study the existence of solutions to the general two- and three- point nonlocal problems. For each model problem we were able to describe analytically the entire manifold of admissible parameters of the corresponding nonlocal condition. Aside from the case when $H$ is a classical Schrödinger operator, our method of analysis covers the situation when $H$ in is non-Hermitian. The later situation occurs in the modeling of open-quantum systems, where the anti-Hermitian part of Hamiltonian describes the interaction of quantum system with the environment (see [@Zloshchastiev2014] and the references therein). The technique used to prove the main result of this paper relies on the linear nature of the problem and the existence of exact representation for the solution operator via the Dunford-Cauchy formula. The results of the paper, therefore, can be generalized to other linear nonlocal problems for Schrödinger equation. This will be the subject of our future work.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Acquaah and Konyagin showed that if $N$ is an odd perfect number with prime factorization $N= p_1^{a_1}p_2^{a_2} \cdots p_k^{a_k}$ where $p_1 < p_2 \cdots < p_k$, then one must have $p_k < 3^{1/3}N^{1/3}$. Using methods similar to theirs, we show that $p_{k-1}< (2N)^{1/5}$ and that $p_{k-1}p_k < 6^{1/4}N^{1/2}.$ We also show that if $p_k$ and $p_{k-1}$ are close to each other then these bounds can be further strengthened.'
author:
- Joshua Zelinsky
title: Upper bounds on the second largest prime factor of an odd perfect number
---
Iowa State University\
Email:[email protected], [email protected]
Throughout this paper we will assume $N$ is an odd perfect number with $N= p_1^{a_1}p_2^{a_2} \cdots p_k^{a_k}$ where $p_1 < p_2 \cdots < p_k$ are all prime. Acquaah and Konyagin [@AK] showed that one must have $$\label{p_k AK bound} p_k < 3^{1/3}N^{1/3}.$$
We recall Euler’s result that if $N$ is an odd perfect number we may write $N=q^eM^2$ where $q$ is prime, $q \equiv e \equiv 1$ (mod 4), and $(q,M)=1$. Equivalently, Euler’s result states that all the $p_k$ are raised to an even power, except for a single prime $p_i$ where $p_i \equiv a_i \equiv i$ (mod 4). We will refer to this single prime raised to a $1$ (mod 4) power as the special prime.[^1]
Acquaah and Konyagin proved their result by showing that for any prime $p_i$ which is not the special prime one has $p_i < 2^{1/4}N^{1/4}$, and most of their work focuses on the possibility that $p_k$ is the special prime. Our situation is similar, although we will need to examine both the situation where $p_k$ is the special prime as well as the situation where $p_{k-1}$ is the special prime.[^2]
It also follows immediately from their results that one has that
$$\label{p_k-1 AK bound} p_{k-1} < 2^{1/4}N^{1/4}.$$
We are interested in proving similar bounds about the second largest prime factor, $p_{k-1}$. Note that a lower bound of $p_{k-1} > 10^4$ is due to Iannucci [@Iannucci].
In this article we prove $$\label{New p_k-1 bound} p_{k-1}< (2N)^{1/5},$$ and $$\label{p_kp_k-1 bound} p_{k-1}p_k < 6^{1/4}N^{1/2}.$$
Note that Inequality \[p\_kp\_k-1 bound\] is not a direct consequence of Inequality \[p\_k AK bound\] and Inequality \[New p\_k-1 bound\] since those two equations together would just yield $$p_{k-1}p_k < 2^{1/5}3^{1/3}N^{\frac{8}{15}}.$$ It is also interesting to compare Inequality \[p\_kp\_k-1 bound\] to the bound of Luca and Pomerance [@LucaPomerance] who proved that
$$\label{Pom Luca bound}p_1p_2p_3 \cdots p_{k-1}p_k < 2N^{\frac{17}{26}}.$$
Of course, $\frac{1}{2}$ is less than $\frac{17}{26}$, but the left hand of Inequality \[p\_kp\_k-1 bound\] only has $p_{k-1}p_k$ as opposed to the product of all the primes dividing $N$ which appears in Inequality \[Pom Luca bound\].\
As with Acquaah and Konyagin’s result, the fact that for any prime $p$, $\sigma(p^a)$ and $p^a$ must be relatively prime, will play a critical role in our results
Before proving our main theorem we need the following lemma:
\[p|q+1 and q|p\^2+p+1 lemma\] If $p$ and $q$ are positive odd integers such that $q \mid (p^2 + p + 1)$, and $p \mid (q + 1)$, then we must have $(p, q) = (1, 1)$ or $(p, q) = (1, 3)$.
Assume that $p$ and $q$ are positive odd integers $q$ such that $q|p^2+p+1$ and $p|q+1$. So there is an $m$ such that $pm=q+1$ and we may then write $q=pm-1$. By assumption, We have $$pm-1|p^2+p+1$$ and hence $$pm-1|p^2 + p+1 + pm-1 = p(p+m+1).$$ Since $(pm-1,p)=1$ we have then that $$pm-1|p+m+1.$$ Note that $pm-1=q$ is odd, and hence $m$ is even. Thus $p+m+1$ is even, and so $2|p+m+1$. We have $$2(pm-1)|p+m+1.$$ Hence, $2pm-2 \leq p+m+1$ and so $$p \leq \frac{m+3}{2m-1} \leq 4.$$ From the last inequality, we must have $p=1$ or $p=3$. If $p=1$, then $q|1^2+1+1=3$, and so $q=1$ or $q=3$. If $p=3$, then $q|3^2+3+1$, and hence $q=1$ or $q=13$, neither of which satisfies $p|q+1$.
We will first prove Inequality \[p\_kp\_k-1 bound\] and then prove Inequality \[New p\_k-1 bound\].
We have $$p_{k-1}p_k < 2^{1/4}3^{1/4}N^{1/2}.$$
We will split our proof into two cases: Case I is where $p_k$ is the special prime; Case II is where $p_k$ is not the special prime.
We first consider Case I where $p_k$ is the special prime. Note that if $a_k>1$, then one must have $a_k \geq 5$ and hence $p_k^5 < N$. From this we have that $p_{k-1}<N^{1/5}$, and hence $$p_{k-1}p_k < p_k^2< N^{2/5} < 2^{1/4}3^{1/4}N^{1/2}.$$ Thus, we may assume that $a_k=1$. Since $p_k$ is the special prime, we must have that $a_{k-1}$ is even and that $a_{k-1} \geq 2$. If $a_{k-1} \geq 4$, since $(\sigma(p_{k-1}^{a_{k-1}}),p_{k-1}^{a_{k-1}})=1$ we conclude $$p_{k-1}^8< p_{k-1}^4\sigma(p_{k-1}^{4}) < 2N,$$ and hence that $p_{k-1} < 2^{1/8}N^{1/8}$ which combining with Inequality \[p\_k AK bound\] would yield that $$p_{k-1}p_k < 2^{1/8}3^{1/3}N^{\frac{11}{24}} < 2^{1/4}3^{1/4}N^{1/2}.$$ Hence, we may assume that $a_{k-1} = 2 $. Then by Lemma \[p|q+1 and q|p\^2+p+1 lemma\], we have either $p_{k-1} \not | (p_k+1)$ or $p_k \not|(p_{k-1}^2 + p_{k-1} + 1) $. We shall label the first of these two Case Ia, and the second Case Ib.
In Case Ia, since $p_{k-1} \not | (p_k+1)$ we have $$p_{k-1}^2(p_k+1)\sigma(p_{k-1}^2)|2N,$$ and hence $$p_{k-1}^4p_k < p_{k-1}^2(p_k+1)\sigma(p_{k-1}^2) \leq 2N.$$
We have then $p_{k-1}^4p_k<2N$. If we then set $p_k = N^{\alpha}$ for some real number $\alpha$ we have that $p_{k-1}^4 < 2N^{1-\alpha}$, and hence $$p_{k-1}< 2^{\frac{1}{4}}N^{\frac{1-\alpha}{4}}.$$ We have then that
$$p_{k-1}p_k < 2^{\frac{1}{4}}N^{\frac{1-\alpha}{4}}N^{\alpha} = 2^{1/4}N^{\frac{1}{4} + \frac{3}{4}\alpha}.$$
The far right-hand side of the above equation is increasing in $\alpha$, and hence largest when $\alpha$ as large as possible, namely when we have $N^\alpha= 3^{1/3}N^{1/3}$, in which case we obtain that $p_{k-1}p_k < 2^{1/4}3^{1/4}N^{1/2}$.
In Case Ib, we have that $p_k \not | (p_{k-1}^2 + p_{k-1} + 1)$, and hence $$\sigma({p_{k-1}}^2){p_k}{p_{k-1}}^2|2N$$ and hence $$p_{k-1}^4p_k < 2N$$ and we then continue just as in Case Ia.\
We now consider Case II, where $p_k$ is not the special prime. Then following logic similar to that in [@AK], since $(\sigma(p_k^{a_k}),p_k^{a_k})=1$ one has $$\sigma(p_k^{a_k})p_k^{a_k} | 2N,$$ and hence $p_k^4 < 2N$, and so $p_k < 2^{1/4}N^{1/4}$. Since $p_{k-1} < p_k$, we have that $$p_{k-1}p_k < (2^{1/4}N^{1/4})^2 = 2^{1/2}N^{1/2} < 2^{1/4}3^{1/4}N^{1/2}.$$
We now prove
$$p_{k-1}< (2N)^{1/5}.$$
Our method of proof is very similar to our method of proof above, but with a slightly different case breakdown. Case I will be when $p_{k-1}$ is the special prime. In Case II, when $p_{k-1}$ is not special, we will break this down into two cases. Case IIa is when $p_k$ is special and Case IIb is when neither $p_k$ nor $p_{k-1}$ is the special prime.\
Let us consider Case I, where $p_{k-1}$ is special. Since $p_k$ is not special, $a_k$ must be even. If $a_k \geq 4$ then since we have that $\sigma(p_k^{a_k})p_k^{a_k}|2N$, this gives us $$p_k^8 < \sigma(p_k^{a_k})p_k^{a_k} \leq 2N.$$ We have then that $p_k<2^{1/8}N^{1/8},$ and so the desired bound on $p_{k-1}$ follows. We may thus assume that $a_k=2$. We again use Lemma \[p|q+1 and q|p\^2+p+1 lemma\] to conclude that either $p_{k-1} \not| (p_k^2 +p_k +1)$ or $p_k \not | (p_{k-1}+1)$. In the case that $p_k \not| p_{k-1}+1$, we have that $$p_{k-1}^5 < (p_{k-1}+1)\sigma(p_k^2)p_{k-1}p_k|\sigma(2N),$$ from which the desired inequality follows. Similarly, in the case that $p_{k-1}$ does not divide $p_k^2 +p_k +1$, then we have that $p_{k-1}^5 < \sigma(p_k^2)p_{k-1}p_k^2|2N $.\
Case IIa where $p_k$ is the special prime is nearly identical to Case I.\
Now consider Case IIb, where neither $p_k$ nor $p_{k-1}$ is special. By logic similar to that in Case I, we may assume that $$a_{k-1}=a_k=2.$$ We note that $$p_k^2 > p_{k-1}^2 + p_{k-1} +1,$$ and hence $p_k^2 \not | p_{k-1}^2 + p_{k-1} + 1$. We have $$p_k\sigma(p_{k-1}^2)p_{k-1}^2 |2N,$$ and so we have that $$p_{k-1}^5 < p_k\sigma(p_{k-1}^2)p_{k-1}^2 \leq 2N$$ from which the desired inequality follows.
Based on the ease of proving stronger inequalities for $p_{k-1}$ than for $p_k$ it seems natural to ask if we can prove that $p_k$ must be much larger than $p_{k-1}$. Right now, it is not obvious how to even rule out a situation like $p_k=p_{k-1}+2$. However we can prove the following weak result:
If $p_k < p_{k-1} + \sqrt{3p_{k-1}} -2,$ then we have $$p_{k-1} < 2^{\frac{1}{6}}N^{\frac{1}{6}}.$$ \[p\_k and p\_k-1 close result\]
To prove this theorem we need the following lemma:
Let $p$ be an odd prime and assume that $q>3$ is a prime such that $q|(p^2+p+1)$ then one has $|p-q|> \sqrt{3p}-2.$ \[p\^2 + p +1 does not have any divisors near its square root\]
Assume that $p$ and $q$ are primes such that $q|p^2+p+1$. The case of $p=3$ is easy to check so assume that $p>3$. Set $p=q+k$, so $q=p-k$. We note then that $$k^2+k+1= p^2+p+1 - (p+k+1)(p-k),$$ and hence $p-k|k^2+k+1.$ If $p-k=k^2+k+1$ then we have $p=k^2+2k+1=(k+1)^2$. But $p$ is prime and so cannot be a perfect square. We must then have $m(p-k)=k^2+k+1$. We note that $m$ is odd and greater than 1. So we have $m \geq 3$. We thus have $$3(p-k) \leq k^2+k+1.$$ and hence $3p \leq k^2 +4k + 1 < (k+1)^2$. Therefore, $$\sqrt{3p} < |k+1|$$ from which the desired bound follows.
We are now in the position to prove Theorem \[p\_k and p\_k-1 close result\].
Assume that $p_k < p_{k-1} + \sqrt{3p_{k-1}} -1.$ We wish to show that $p_{k-1} < 2^{\frac{1}{6}}N^{\frac{1}{6}}.$ Note that this result follows easily if we have either $p_k^4|N$ or have $p_{k-1}^4|N$. Thus, we may assume that we have $a_k$ and $a_{k-1}$ are either 1 or 2. Let us consider the case where $a_k=a_{k-1}=2$. We have then by Lemma \[p\^2 + p +1 does not have any divisors near its square root\] that $$p_{k-1} \not|p_k^2+p_k +1=\sigma(p_k^2)$$ and we get the same for swapping $p_k$ and $p_{k-1}$. Hence we have that $$(p_k^2+p_k+1)(p_{k-1}^2+p_{k-1}+1)p_{k-1}^2p_k^2 |2N.$$ We thus have $p_{k-1}^8 < 2N$, and hence $$p_{k-1} < 2^{\frac{1}{8}}N^{\frac{1}{8}} < 2^{\frac{1}{6}}N^{\frac{1}{6}}.$$
We thus may assume that one of $p_k$ and $p_{k-1}$ is special and the other is raised to the second power. We will look at the case where $p_k$ is special (the other case is nearly identical). We note that we cannot have $p_{k-1}|p_k+1$ since this would force $p_{k-1} \leq \frac{p_k+1}{2}$ which would contradict our assumption.
From the above note and Lemma \[p\^2 + p +1 does not have any divisors near its square root\], we have $$(p_{k-1}^2+p_{k-1}+1)p_{k-1}^2p_k(p_k+1)|2N,$$ and so $$p_{k-1}^6 < (p_{k-1}^2+p_{k-1}+1)p_{k-1}^2p_k(p_k+1) \leq 2N.$$ From this last inequality our desired inequality immediately follows.
We also have the following assertion as an easy corollary of Theorem \[p\_k and p\_k-1 close result\]
If $p_k < p_{k-1} + \sqrt{3p_{k-1}} -2,$ then $$p_1p_2p_3 \cdots p_k < 3N^{\frac{7}{12}},$$ and $$p_{k-1}p_k < 2N^{\frac{1}{3}}.$$
Note that $\frac{7}{12} < \frac{17}{26}$, so the inequality for the radical in this corollary is stronger than the Luca and Pomerance bound albeit it requires the additional hypothesis that $p_k$ and $p_{k-1}$ are close.\
There are a variety of avenues for future work. We are optimistic that careful refinements of these sorts of arguments can further improve bounds on $p_k$ and $p_{k-1}$. We also suspect that recent results showing that an odd perfect number must have many repeated prime factors such as [@OchemRao1] and [@Zelinsky] may be used to improve either these results or the results of Acquaah and Konyagin for $p_k$. The primary barrier to doing so appears to be that one may have an odd perfect number with a prime $p$ and many distinct $p_i$ such that $\sigma(p_i^{a_i})$ is a power of $p$. This situation corresponds to systems of extremely restrictive Diophantine equations, so limiting how many such $p_i$ one can have may be possible. It also may be possible to combine these sorts of bounds with the sort of estimates of the index of an odd perfect number as introduced by Dris[@Dris][@DrisLuca].
One other potential direction to go in is to obtain non-trivial estimates for $p_{k-2}$, or more generally to give a bound on $p_{k-i}$ for $1\leq i\leq k$. One easily has from Euler’s theorem about an odd perfect number that $$p_{k-i} < N^{\frac{1}{2i+1}}.$$ A bound for $p_{k-2}$ better than $N^{\frac{1}{5}}$ then is a natural goal since $N^{1/5}$ is the bound one gets immediately from applying Euler’s theorem.
[**Acknowledgements**]{} The author would like to acknowledge the helpful comments from the referee. The author would also like to acknowledge Douglas McNeil who noted that a prior version of this paper had an incorrect version of Lemma \[p\^2 + p +1 does not have any divisors near its square root\]. Helpful comments were also provided by Carl Pomerance.
[99]{}
P. Acquaah and S. Konyagin, “On Prime Factors of Odd Perfect Numbers,” [*International Journal of Number Theory.*]{} [**[8]{}**]{} (6) (2012) 1537–1540. J. Dris, “Solving the odd perfect number problem: some old and new approaches,” M. Sc. Thesis, De La Salle University, Philippines, 2008. J. Dris and F. Luca, “A note on odd perfect numbers,” [*Fibonacci Quart.*]{} [**54**]{} (2016), 291-?295. D Iannucci, “The second largest prime divisor of an odd perfect number exceeds ten thousand,” [*Math Comp.*]{} [**68**]{}(228) (1999), 1749?-1760. F. Luca and C. Pomerance, “On the radical of a perfect number,” [*New York J. Math.*]{} P. Ochem and M. Rao, “On the number of prime factors of an odd perfect number,” [*Math Comp.*]{} [**83**]{} (2014), 2435–2439. [**16**]{} (2010) 23-?30. P. Starni, “On Dris conjecture about odd perfect numbers,” [*Notes on Number Theory and Discrete Mathematics.*]{} [**24**]{} (1) (2018) 5–9. J. Zelinsky “An improvement of an inequality of Ochem and Rao concerning odd perfect numbers,” [*Integers.*]{} [**18**]{} (2018) –A48.
[^1]: Some authors call $q$ the “Euler prime.” A better name than the Euler prime in fact would be the Cartesian prime since prior to Euler’s result Descartes proved that an odd perfect number needed to have exactly one prime factor raised to an odd power. In any event, the term special prime avoids any issues of priority.
[^2]: Starni’s recent paper [@Starni] claims a very tight upper bound on the size of the special prime in general, but the author was unable to follow the proof of Corollary 2.2 in his paper which gives that result. Therefore, none of the results in this paper rely on that one.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent works have proposed neural models for dialog act classification in spoken dialogs. However, they have not explored the role and the usefulness of acoustic information. We propose a neural model that processes both lexical and acoustic features for classification. Our results on two benchmark datasets reveal that acoustic features are helpful in improving the overall accuracy. Finally, a deeper analysis shows that acoustic features are valuable in three cases: when a dialog act has sufficient data, when lexical information is limited and when strong lexical cues are not present.'
address: |
Institute for Natural Language Processing (IMS)\
University of Stuttgart, Germany\
[{daniel.ortega, thang.vu}@ims.uni-stuttgart.de]{}
bibliography:
- 'ICASSP.bib'
title: 'Lexico-acoustic Neural-based Models for Dialog Act Classification'
---
dialog act, lexico-acoustic features
Introduction {#sec:intro}
============
Every utterance in a conversation has a level of illocutionary force [@austin1962tw] whose meaning induces an effect over the course of the dialog. That meaning can be categorized into dialog acts (DAs) taking into account the relationship between the words being used and the force of the utterance [@book:kentBach2000]. A DA is the expression of the speaker’s attitude or intention at every utterance in a conversation. Kent Bach [@book:kentBach2000] illustrates this by pointing out that a statement expresses a belief, a request expresses a desire, and an apology expresses a regret. In this manner, dialogs can be studied and modeled by analyzing their sequence of DAs.
Automatic DA tagging is an important preprocessing step for semantic extraction in natural language understanding and dialog systems. This task has been approached using two main information sources: lexical cues from dialog transcripts and acoustic cues from speech signals. For the former, traditional statistical algorithms have been employed, such as hidden Markov models (HMMs) [@Stolcke:2000], conditional random fields (CRFs), [@Zimmermann:CRF2009] and support vector machines (SVMs) [@Henderson:SVM2010]. Recently, deep learning (DL) techniques, such as convolutional neural networks (CNNs) [@RCNN:KalchbrennerB13; @lee:CNN_RNN_DA_2016], recurrent neural networks (RNNs) [@lee:CNN_RNN_DA_2016; @DBLP:journals/corr/JiHE16] and long short-term memory (LSTM) models [@AttentionCNN:ShenL16], have attained the state-of-the-art results in DA classification. DAs can be ambiguous if only lexical information is considered. For example, a *Declarative Question* like “*this is your car(?)*” is hard to distinguish from a *Statement* if the question mark is not present, and can easily be misclassified due to word order. In this case, acoustic information can help disambiguate. Moreover, in real applications that involve automatic speech recognition (ASR), a DA classifier can help deal with noisy transcriptions. Hence, some researchers [@Stolcke:2000; @DBLP:shiberg2000] have explored acoustic and prosodic cues from the speech signal as a potential knowledge source for DA classification.
Other works [@Arsikere2016; @Ondas2015] have showed improvements exploring combinations of lexico-acoustic features. Inspired by these works, we present a neural hybrid model that takes both lexical and acoustic features as input in order to classify dialog utterances into DAs. Our model is a combination of two neural-based models: one, which processes lexical features of the utterances and their context (based on [@ortega-vu:2017:SIGDIAL]), and a second, which processes acoustic features. Our experiments show that acoustic features are helpful for improving overall accuracy and attaining state-of-the-art results on two benchmark datasets: the ICSI Meeting Recorder Dialog Act Corpus (MRDA) and the NXT-format Switchboard Corpus (SwDA). We also include an analysis of the acoustic features contribution for DA classification in three circumstances: when a DA has sufficient data, when strong lexical cues are missing and for single-word utterances.
Model {#sec:model}
=====
The architecture of the lexico-acoustic model (LAM) proposed in this paper is depicted in Figure \[fig:LAM\]. It contains two main parts: on the left side is the lexical model (LM) and on the right side the acoustic model (AM). Models are detailed in sections \[sec:LM\]-\[sec:LAM\].
![Architecture of the lexico-acoustic model. $\oplus$ represents a concatenation.[]{data-label="fig:LAM"}](lex_ac_model){width="8.5cm"}
Lexical model {#sec:LM}
-------------
The LM, based on [@ortega-vu:2017:SIGDIAL], takes the concatenation of grid-like representations of the current utterance and its $n$ previous utterances in the dialog as input to be processed by a CNN, generating a vector representation for each of those utterances.
The CNN performs a discrete convolution using a set of different filters on an input matrix, where each column of the matrix is the word embedding of the corresponding word. We use 2D filters $f$ (with width $|f|$) spanning over all embedding dimensions $d$ as described by the following equation: $$(w \ast f)(x,y) = \sum_{i=1}^{d}\sum_{j = -|f|/2}^{|f|/2}w(i,j) \cdot f(x-i,y-j)$$
After convolution, an utterance-wise max pooling operation is applied. Then, the feature maps are concatenated, resulting in one vector per utterance. These are represented in Figure \[fig:LAM\] as $p_{t-2}, p_{t-1}$ and $p_t$.
The vector representations of the utterances are then processed by a context learning method, the RNN-Output-Attention (ROA) proposed in [@ortega-vu:2017:SIGDIAL], in order to model the relation between each utterance and its context. ROA consists of an RNN with LSTM units followed by a weighted sum of the RNN’s hidden states using an attention mechanism [@Attn_ML:Bahdanau2014].
For each of the hidden state vector $h(t-i)$ at time step $t-i$ in a dialog, where $t$ is the current time step. The attention weights $\alpha_{i} $ are computed as follows
$$\alpha_i = \frac{exp(f(h(t-i)))}{\sum_{j} {exp(f(h(t-j))}}$$
where $f$ is the scoring function. In our work, $f$ is the linear function of the input $h(t-i)$ $$f(h(t-i))= W^T h(t-i)$$ where $W$ is a trainable parameter. The output $l_t$ is the weighted sum of the hidden states sequence. $$l_t= \sum_{i} \alpha_{i} h(t-i)$$
Finally, the context representation $l_t$ is fed into a softmax layer that outputs a probability distribution over the DA set, given the current dialog utterance.
Acoustic model {#sec:AM}
--------------
We propose a CNN-based model to process acoustic features, because the speech signal of the utterances encodes important information for DA classification that is not contained in the transcripts. The acoustic features from the speech signal are not taken at word or utterance level, but at frame level, i.e. the speech signal is divided into frames of 25 ms with a shift of 10 ms, and 13 Mel-frequency-cepstral coefficients (MFCC) per frame are extracted using the openSMILE toolkit [@tools:openSMILE]. MFCC features are stacked sequentially in order to obtain a grid-like input representation of the acoustic signal.
The input is processed by a one-layer CNN using filters that span over the 13 MFCC features and 5 frames a time, with a max pooling layer in order to obtain a fixed-length vector representation. This is fed into a softmax layer for DA classification, as explained in Section \[sec:LM\].
Lexico-acoustic model {#sec:LAM}
---------------------
The core of this work is the LAM (depicted in Figure \[fig:LAM\]), a bi-CNN that employs lexical and acoustic cues simultaneously as input. The LAM combines a LM and a AM by concatenating the vector representations ($l_t$ and $a_t$) obtained from the context processing method in the LM and the pooling layer in the AM. Both vectors represent the current utterance, and can therefore be joined at this level and passed to the softmax function to output a final probability distribution over the DA set.
Experiments {#sec:experiments}
===========
Data
----
We test our model on two DA datasets: 1) **MRDA**: ICSI Meeting Recorder Dialog Act Corpus [@corpus:ICSI_annotated], a dialog corpus of *multiparty meetings*. The 5-tag-set used in this work was introduced by [@ICSI:tagsetAng]. 2) **SwDA**: NXT-format Switchboard Corpus [@NXT:2010], a dialog corpus of *2-speaker conversations*. NXT-format Switchboard Corpus was preferred over the original Switchboard Dialog Act Corpus [@Corpus:Switchboard; @jurafsky97switchboard] because the former provides utterance transcripts and DA annotations as well as the time stamps at word level that were useful to extract acoustic features. Nonetheless, this corpus only provides DA annotation for roughly 50% of the original dataset.
Train, validation and test splits on MRDA were taken as defined in [@lee:CNN_RNN_DA_2016]. However, on SwDA the splits were built by taking the annotated conversations from NXT-format Switchboard Corpus that appear in the split lists published in [@lee:CNN_RNN_DA_2016]. The new train, validation and test splits are roughly the half of the conversations on each original split. Summary statistics are shown in Table \[tab:datasets\]. In both datasets, the classes are highly unbalanced; the majority class is $59.1 $% on MRDA and $34.7$ % on SwDA.
**Dataset** **C** **$\mid$V$\mid$** **Train** **Validation** **Test**
------------- ------- ------------------- ----------- ---------------- ----------
MRDA 5 12k 78k 16k 15k
SwDA 42 16k 98k 8.5k 2.5k
: \[tab:datasets\] Data statistics: **C** is the number of classes, **$\mid$V$\mid$** is the vocabulary size and Train/Validation/Test are \#utterances.
Hyperparameters and Training
----------------------------
The hyperparameters of the three models for both datasets are summarized in Table \[tab:hyperparams\]. The LM’s hyperparameters were taken from [@ortega-vu:2017:SIGDIAL], while the AM’s hyperparameters were obtained by varying one hyperparameter at a time while keeping the others fixed. Training was done for 25 epochs with averaged stochastic gradient descent [@ASGD:Polyak1992] over mini-batches. The learning rate was initialized at 0.11 and reduced 10% every 2000 parameter updates. Word2vec pretrained embeddings [@WordEmbeddings:word2vec] were employed and tuned during training. The context length $n$ was taken from the original the LM, i.e. $n=3$ for MRDA and $n=2$ for SwDA.
Results {#sec:Results}
-------
Table \[tab:results\] shows the results obtained from the three models on both datasets. As expected, the LM is superior to the AM, i.e. the lexical features yield more valuable information than the acoustic features for our task. On both datasets, the LM’s accuracy is significantly higher than the AM’s accuracy. However, for both datasets, the combined model yields improvements over both constituent models. It indicates that both cue sources complement each other.
**Model** **MRDA** **SwDA**
----------------- ---------- ----------
Lexical 84.1 73.6
Acoustic 67.8 50.9
Lexico-acoustic **84.7** **75.1**
: Accuracy (%) of the three models on both datasets.[]{data-label="tab:results"}
Analysis {#sec:analysis}
========
This section’s goal is to analyze the impact of joining both models, and to report and discuss which DAs benefit and which are impacted negatively, by applying a LAM versus a LM. Moreover, we also investigate the effect of the acoustic features when the question mark (*?*) is removed from transcripts and when utterances are very short. On MRDA, as reported in the previous section, the LAM yielded an improvement of 0.6% over the LM. However, the improvement is not uniform over the five classes. While the prediction of the DAs *Statement*, *Disruption* and *Backchannel* obtains a benefit from the acoustic features, *Filter* is impacted negatively and *Question* stays the same. Nonetheless, in general terms, the LAM benefits the overall DA classification, specially for those DAs with a higher presence in the training set, and the degradation caused by the model does not hurt its overall performance.
On SwDA, the LAM also outperformed the LM by 1.5%. Five DAs benefited by adding acoustic features: *Statement*, *Backchannel*, *Opinion*, *Abandon* and *Agree*, *Wh\_question* and *Acknowledge* were negatively affected in a minimal extent, and the remaining 35 DAs were not impacted. These results are again highly correlated to the DA distribution in the corpus – the 5 most frequent DAs obtained an improvement that is reflected in the overall accuracy. Therefore, we argue that the LAM helps when a large number of examples per DA is available. One possible reason is that we have enough training data for these particular DAs to properly train the AM part of the LAM.
#### Effect of removing the question mark {#effect-of-removing-the-question-mark .unnumbered}
Contrary to our initial hypothesis that acoustic features would improve the accuracy of classifying *Question*, no improvement was noted. Therefore, we analyzed how the LM and the LAM performed on this particular DA more deeply. The question mark *?* in the manual transcripts plays a fundamental role for the DA *Question* in the LM; 97.7% of the utterances with question marks which are labeled as *Question* are correctly predicted (see Table \[tab:question\]) by the LM. For that reason, the acoustic features are not able to provide any useful information. Consequently, we retrained and tested the LM and the LAM using transcripts from which the question mark was removed. This change also makes the transcripts more similar to transcripts from an ASR, where punctuation is not available or is not highly accurate. As expected, the overall accuracy dropped, from 84.1% to 80.8% in the LM and from 84.7% to 81.9% in the LAM. Although both models were affected by this modification, the LAM performed 1.1% better than the LM, versus the improvement of 0.6% with the original transcripts. Acoustic features slightly dampen the negative effect on the accuracy of removing the question mark.
Table \[tab:question\] shows the accuracy of the LM and the LAM exclusively on utterances whose DA is *Question* and which have a question mark in the manual transcript. The second column corresponds to the models which were trained and tested on the original transcripts and the third column to the models which were trained on transcripts with question marks removed. As mentioned above, the LM has a high accuracy at correctly predicting *Question* if the utterance has the question mark. Moreover, when the acoustic features are added, the accuracy decreases by 1.6%. Nonetheless, if question marks are not present in the data, the LM’s accuracy drops to 46.6%. This shows that this character is the most important cue at lexical level. The LAM’s accuracy drops to 50.2%, but this time it is superior to the LM by 3.6%. This indicates that acoustic information is an important source of cues for tasks that use DA classification over data that lacks these important lexical cues, such as spoken language understanding.
**Model** **With *’?’*** ***’?’* removed**
----------------- ---------------- -------------------
Lexical **97.7** 46.6
Lexico-acoustic 96.1 **50.2**
: Accuracy (%) of *Question* utterances on MRDA with question mark and when the question mark is removed.[]{data-label="tab:question"}
#### Single-word utterances {#single-word-utterances .unnumbered}
There exist utterances like *Right* or *Yeah* that are very common across several DAs. One of their characteristics is that they are very short and consequently they do not yield much information for classification. [@ortega-vu:2017:SIGDIAL; @lee:CNN_RNN_DA_2016; @Liu:DA] have successfully explored the use of context as a way to differentiate these type of utterances. In line with these works, both the LM and the LAM (in its lexical component) encode the context.
We have shown in Section \[sec:Results\] that the LAM outperforms the LM on both datasets, however, we explored particularly the effect of using acoustic features on the utterances *Right* and *Yeah* that are frequently tagged as *Statement* and *Backchannel* on MRDA. For our analysis purposes, we extracted the predictions of the utterances that exclusively contained one word that is either *Right* or *Yeah*, from which we can artificially define four subclasses: *Statement-Right*, *Backchannel-Right*, *Statement-Yeah* and *Backchannel-Yeah*.
Table \[tab:class\_right\] shows the precision, recall and F~1~ score of the LM and the LAM for the utterances *Right*. On the one hand, for the DA *Statement* the LAM achieves a higher F~1~ score than the LM, while on the other hand, the F~1~ score for *Backchannel* decreases slightly. This means that using acoustic features improves the classification of utterances *Right* as *Statement* without affecting those utterances tagged as *Backchannel*. A similar phenomenon is observed with utterances *Yeah*, however, in this case, the LAM improves the F~1~ score for both DAs *Statement* and *Backchannel* (see Table \[tab:class\_yeah\]).
**DA-Right** **Model** **P** **R** **F~1~**
-------------- ----------------- ------- ------- ----------
Lexical 0.62 0.35 0.45
Lexico-acoustic 0.60 0.45 **0.52**
Lexical 0.56 0.85 **0.67**
Lexico-acoustic 0.56 0.77 0.65
: Precision, recall and F~1~ score for the utterances *Right*[]{data-label="tab:class_right"}
**DA-Yeah** **Model** **P** **R** **F~1~**
------------- ----------------- ------- ------- ----------
Lexical 0.65 0.36 0.46
Lexico-acoustic 0.67 0.50 **0.57**
Lexical 0.60 0.89 0.72
Lexico-acoustic 0.64 0.87 **0.74**
: Precision, recall and F~1~ score for the utterances *Yeah*[]{data-label="tab:class_yeah"}
Comparison with other works {#sec:comparison}
===========================
We present a comparison between different works and our model in Table \[tab:comparison\]. On MRDA, as we used the setup proposed by [@lee:CNN_RNN_DA_2016], our results can only be compared accurately to [@ortega-vu:2017:SIGDIAL] and [@lee:CNN_RNN_DA_2016], and the LAM outperforms both works. On SwDA, as we used the data available in the NXT format, and, to the best of our knowledge, no other model has been trained and tested on this subset of SwDA, our results cannot be strictly compared with other works.
**Model** **MRDA** **SwDA**
----------------- ---------- ----------
LAM (Our model) **84.7** **75.1**
NCRL 84.3 73.8
CNN-FF 84.6 73.1
HBM 81.3 —
CNN+DAP — 79.9
HCNN — 73.9
HMM — 71.0
Majority class 59.1 34.7
: Comparison of accuracy (%). *NCRL*: Neural context representation learning proposed in [@ortega-vu:2017:SIGDIAL], *CNN-FF*: proposed in [@lee:CNN_RNN_DA_2016], *HBM*: hidden backoff model [@HiddenBackoff:Ji2006]. *CNN+DAP*:proposed by [@Liu:DA]. *HCNN*: hierarchical CNN [@RCNN:KalchbrennerB13]. *HMM* [@Stolcke:2000]. *Majority class* is the most frequent class.
\[tab:comparison\]
Conclusion {#ssec:conclusion}
==========
We proposed an approach to incorporate lexical and acoustic features in a neural model for DA classification. Our experiments on two benchmark datasets reveal that adding acoustic information to the model improves the overall accuracy attaining state-of-the-art results. A deeper analysis showed that acoustic features specially help when the data for a particular DA is large enough, when lexical information is limited, as in single-word utterances, and when strong lexical cues are not present.
Acknowledgements {#ssec:Ak}
================
This work was funded by the National Council of Science and Technology of Mexico (CONACyT), the German Academic Exchange Service (DAAD) and the German Science Foundation (DFG), Sonderforschungsbereich 732, Project A8, at the University of Stuttgart.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss the stability properties of an autonomous system in loop quantum cosmology. The system is described by a self-interacting scalar field $\phi$ with positive potential $V$, coupled with a barotropic fluid in the Universe. With $\Gamma=VV''''/V''^2$ considered as a function of $\lambda=V''/V$, the autonomous system is extended from three dimensions to four dimensions. We find that the dynamic behaviors of a subset, not all, of the fixed points are independent of the form of the potential. Considering the higher-order derivatives of the potential, we get an infinite-dimensional autonomous system which can describe the dynamical behavior of the scalar field with more general potential. We find that there is just one scalar-field-dominated scaling solution in the loop quantum cosmology scenario.'
author:
- Kui Xiao
- 'Jian-Yang Zhu'
title: Stability analysis of an autonomous system in loop quantum cosmology
---
[^1]
Introduction
============
The scalar field plays an important role in modern cosmology. Indeed, scalar-field cosmological models are of great importance in the study of the early Universe, especially in the investigation of inflation. The dynamical properties of a scalar fields also make an interesting research topic for modern cosmological studies [@Copeland-IJMPD; @Coley-Book]. The dynamical behavior of scalar field coupled with a barotropic fluid in a spatially flat Friedmann-Robertson-Walker universe has been studied by many authors (see [@Copeland-IJMPD; @Copeland-scalar; @Leon-Phase], and the first section of [@Coley-Book]).
The phase-plane analysis of the cosmological autonomous system is an useful method for studying the dynamical behavior of a scalar field. One always considers the dynamical behavior of a scalar field with an exponential potential in the classical cosmology [@Copeland-exponential; @Hao-PRD1; @Hao-PRD2] or modified cosmology [@Samart-Phantom; @Li-O(N)]. And, if one considers the dynamical behavior of a scalar field coupled with a barotropic fluid, the exponential potential is also the first choice [@Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom]. The exponential potential $V$ leads to the fact that the variables $\Gamma=VV''/V'^2$ equal 1 and that $\lambda=V'/V$ is also a constant. Then the autonomous system is always two dimensional in classical cosmology [@Copeland-exponential], and three dimensional in loop quantum cosmology (LQC) [@Samart-Phantom]. Although one can also consider a more complex case with $\lambda$ being a dynamically changing quantity [@Copeland-IJMPD; @Macorra-lambda; @Nunes-lambda], the fixed point is not a real one, and this method is not exact. Recently, Zhou *et al* [@Zhou-plb; @Fang-CQG] introduced a new method by which one can make $\Gamma$ a general function of $\lambda$. Then the autonomous system is extended from two dimensions to three dimensions in classical cosmology. They found that this method can help investigate many quintessence models with different potentials. The goal of this paper is to extend this method for studying the dynamical behavior of a scalar field with a general potential coupled with a barotropic fluid in LQC.
LQC [@Bojowald-Living; @Ashtekar-overview] is a canonical quantization of homogeneous spacetime based on the techniques used in loop quantum gravity (LQG) [@Rovelli-Book; @Thiemann-Book]. Owing to the homogeneity and isotropy of the spacetime, the phase space of LQC is simpler than that of LQG. For example, the connection is determined by a single parameter $c$ and the triad is determined by $p$. Recently, it has been shown that the loop quantum effects can be very well described by an effective modified Friedmann dynamics. Two corrections of the effective LQC are always considered: the inverse volume correction and the holonomy correction. These modifications lead to many interesting results: the big bang can be replaced by the big bounce [@Ashtekar], the singularity can be avoided [@Singh], the inflation can be more likely to occur (e.g., see [@Bojowald; @Germani-inflation; @Copeland-superinflation; @Ashtekar-inflation; @Corichi-measure]), and more. But the inverse volume modification suffers from gauge dependence which cannot be cured and thus yields unphysical effects. In the effective LQC based on the holonomy modification, the Friedmann equation adds a $-\frac{\kappa}{3}\frac{\rho^2}{\rho_c}$ term, in which $\kappa=8\pi G$, to the right-hand side of the standard Friedmann equation [@Ashtekar-improve]. Since this correction comes with a negative sign, the Hubble parameter $H$, and then $\dot{a}$ will vanish when $\rho=\rho_c$, and the quantum bounce occurs. Moreover, for a universe with a large scalar factor, the inverse volume modification to the Friedmann equation can be neglected and only the holonomy modification is important.
Based on the holonomy modification, the dynamical behavior of dark energy has recently been investigated by many authors [@Samart-Phantom; @Samart-dy; @Xiao-dynamical]. The attractor behavior of the scalar field in LQC has also been studied [@Copeland-superinflation; @Lidsey-attractor]. It was found that the dynamical properties of dark-energy models in LQC are significantly different from those in classical cosmology. In this paper, we examine the background dynamics of LQC dominated by a scalar field with a general positive potential coupled with a barotropic fluid. By considering $\Gamma$ as a function of $\lambda$, we investigate scalar fields with different potentials. Since the Friedmann equation has been modified by the quantum effect, the dynamical system will be very different from the one in classical cosmology, e.g., the number of dimensions of autonomous system will change to four in LQC. It must be pointed out that this method cannot be used to describe the dynamical behavior of scalar field with arbitrary potential. To overcome this problem, therefore, we should consider an infinite-dimensional autonomous system.
The paper is organized as follows. In Sec. \[sec2\], we present the basic equations and the four dimensional dynamical system, and in Sec. \[sec3\], we discuss the properties of this system. In Sec. \[sec4\], we discuss the autonomous system in greater detail, as well as an infinite-dimensional autonomous system. We conclude the paper in the last section. The Appendix contains the analysis of the dynamical properties of one of the fixed points, $P_3$.
Basic equations {#sec2}
===============
We focus on the flat Friedmann-Robertson-Walker cosmology. The modified Friedmann equation in the effective LQC with holonomy correction can be written as [@Ashtekar-improve] $$\begin{aligned}
H^2=\frac{1}{3}\rho\left(1-\frac{\rho}{\rho_c}\right),\label{Fri}\end{aligned}$$ in which $\rho$ is the total energy density and the natural unit $\kappa=8\pi G=1$ is adopted for simplicity. We consider a self-interacting scalar field $\phi$ with a positive potential $V(\phi)$ coupled with a barotropic fluid. Then the total energy density can be written as $\rho=\rho_\phi+\rho_\gamma$, with the energy density of scalar field $\rho_\phi=\frac12\dot{\phi}^2+V(\phi)$ and the energy density of barotropic fluid $\rho_\gamma$. We consider that the energy momenta of this field to be covariant conserved. Then one has $$\begin{aligned}
&&\ddot{\phi}+3H\dot{\phi}+V'=0,\label{ddotphi}\\
&&\dot{\rho}_\gamma+3\gamma H\rho_\gamma=0, \label{dotrg}\end{aligned}$$ where $\gamma$ is an adiabatic index and satisfies $p_\gamma=(\gamma-1)\rho_\gamma$ with $p_\gamma$ being the pressure of the barotropic fluid, and the prime denotesthe differentiation with respect to the field $\phi$. Differentiating Eq. (\[Fri\]) and using Eqs. (\[ddotphi\]) and (\[dotrg\]), one can obtain $$\begin{aligned}
\dot{H}=-\frac12\left(\dot{\phi}^2+\gamma\rho_\gamma\right)\left[1-\frac{2(\rho_\gamma+\rho_\phi)}{\rho_c}\right].
\label{Fri4}\end{aligned}$$
Equations (\[Fri\])-(\[dotrg\]) and (\[ddotphi\])-(\[Fri4\]) characterize a closed system which can determine the cosmic behavior. To analyze the dynamical behavior of the Universe, one can further introduce the following variables [@Copeland-exponential; @Samart-Phantom]: $$\begin{aligned}
x\equiv\frac{\dot{\phi}}{\sqrt{6}H},\quad
y\equiv\frac{\sqrt{V}}{\sqrt{3}H},\quad
z\equiv\frac{\rho}{\rho_c},\quad \lambda\equiv\frac{V'}{V},
\label{new-v}\end{aligned}$$ where the $z$ term is a special variable in LQC \[see Eq. (\[Fri\])\]. In the LQC scenario, the total energy density $\rho$ should be less than or equal to the critical energy density $\rho_c$, and thus $0\leq z\leq 1$. Notice that, in the classical region, $z=0$ for $\rho\ll\rho_c$. Using these new variables, one can obtain $$\begin{aligned}
&&\frac{\rho_\gamma}{3H^2}=\frac{1}{1-z}-x^2-y^2,\label{rg-xyz}\\
&&\frac{\dot{H}}{H^2}=-\left[3x^2+\frac{3\gamma}{2}\left(\frac{1}{1-z}-x^2-y^2\right)\right]\left(1-2z\right)\nonumber\\
\label{dH-HH}.\end{aligned}$$
Using the new variables (\[new-v\]), and considering Eqs. (\[rg-xyz\]) and (\[dH-HH\]), one can rewrite Eqs. (\[Fri\])-(\[dotrg\]) in the following forms: $$\begin{aligned}
\frac{dx}{dN} &=&-3x-\frac{\sqrt{6}}{2}\lambda y^2+x\left[3x^2+\frac{3\gamma}{2}\left(\frac{1}{1-z}-x^2-y^2\right)\right]\nonumber\\
&&\times(1-2z),\label{x'}\\
\frac{dy}{dN}&=&\frac{\sqrt{6}}{2}\lambda x
y+y\left[3x^2+\frac{3\gamma}{2}
\left(\frac{1}{1-z}-x^2-y^2\right)\right]\nonumber\\
&&\times(1-2z),\label{y'}\\
\frac{dz}{dN} &=&-3\gamma z-3z\left(1-z\right)\left(2x^2-\gamma x^2-\gamma y^2\right),\label{z'}\\
\frac{d\lambda}{dN} &=&\sqrt{6}\lambda^2
x\left(\Gamma-1\right),\label{l}\end{aligned}$$ where $N=\ln a$ and $$\begin{aligned}
\Gamma\equiv \frac{VV''}{V'^2}.\label{Gamma}\end{aligned}$$ Note that the potential $V(\phi)$ is positive in this paper, but one can also discuss a negative potential. Just as [@Heard-negative] has shown, the negative scalar potential could slow down the growth of the scale factor and cause the Universe to be in a collapsing phase. The dynamical behavior of the scalar field with the positive and negative potential in brane cosmology has been discussed by [@Copeland-scalar]. In this paper we are concerned only with an expanding universe, and both the Hubble parameter and the potential are positive.
Differentiating $\lambda$ with respect to the scalar field $\phi$, we obtain the relationship between $\lambda$ and $\Gamma$, $$\begin{aligned}
\frac{d\lambda^{-1}}{d\phi}=1-\Gamma. \label{l-G}\end{aligned}$$ If we only consider a special case of the potential, like exponential potential [@Copeland-exponential; @Hao-PRD1; @Hao-PRD2; @Samart-Phantom; @Li-O(N); @Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom], then $\lambda$ and $\Gamma$ are both constants. In this case, the four dimensional dynamical system, Eqs. (\[x’\])-(\[l\]), reduces to a 3-dimensional one, since $\lambda$ is a constant. (In the classical dynamical system, the $z$ term does not exist, and then the dynamical system is reduced from three dimensions to two dimensions.) The cost of this simplification is that the potential of the field is restricted. Recently, Zhou *et al* [@Zhou-plb; @Fang-CQG] considered the potential parameter $\Gamma$ as a function of another potential parameter $\lambda$, which enables one to study the fixed points for a large number of potentials. We will follow this method in this section and the sections that follow to discuss the dynamical behavior of the scalar field in the LQC scenario, and we have $$\begin{aligned}
\Gamma(\lambda)=f(\lambda)+1.\label{G-l}\end{aligned}$$ In this case, Eq. (\[G-l\]) can cover many scalar potentials.
For completeness’ sake, we briefly review the discussion of [@Fang-CQG] in the following. From Eq. (\[l-G\]), one can obtain $$\begin{aligned}
\frac{d\lambda}{\lambda f(\lambda)}=\frac{dV}{V}.\label{l-f-V}\end{aligned}$$ Integrating out $\lambda=\lambda(V)$, one has the following differential equation of potential $$\begin{aligned}
\frac{dV}{V\lambda(V)}=d\phi.\label{l-V}\end{aligned}$$ Then, Eqs. (\[l-f-V\]) and (\[l-V\]) provide a route for obtaining the potential $V=V(\phi)$. If we consider a concrete form of the potential (e.g., an exponential potential), the dynamical system is specialized (e.g., the dynamical system is reduced to three dimensions if one considers the exponential potential for $d\lambda/dN=0$). These specialized dynamical systems are too special if one hopes to distinguish the fixed points that are the common properties of scalar field from those that are just related to the special potentials [@Fang-CQG]. If we consider a more general $\lambda$, then we can get the more general stability properties of scalar field in the LQC scenario. We will continue the discussion of this topic in Sec. \[sec4\]. In this case, Eq. (\[l\]) becomes $$\begin{aligned}
\frac{d\lambda}{dN}&=&\sqrt{6}\lambda^2 xf(\lambda).\label{l'}\end{aligned}$$ Hereafter, Eqs. (\[x’\])-(\[z’\]) along with Eq. (\[l’\]) definitely describe a dynamical system. We will discuss the stability of this system in the following section.
Properties of the autonomous system {#sec3}
===================================
Obviously, the terms on the right-hand side of Eqs. (\[x’\])-(\[z’\]) and (\[l’\]) only depend on $x,y,z,\lambda$, but not on $N$ or other variables. Such a dynamical system is usually called an autonomous system. For simplicity, we define $
\frac{dx}{dN}=F_1(x,y,z,\lambda)\equiv F_1,
\frac{dy}{dN}=F_2(x,y,z,\lambda)\equiv F_2,
\frac{dz}{dN}=F_3(x,y,z,\lambda)\equiv F_3$, and $\frac{d\lambda}{dN}=F_4(x,y,z,\lambda)\equiv F_4.$ The fixed points $(x_c,y_c,z_c,\lambda_c)$ satisfy $F_i=0, i=1,2,3,4$. From Eq. (\[l’\]), it is straightforward to see that $x=0, \lambda=0$ or $f(\lambda)=0$ can make $F_4(x,y,z,\lambda)=0$. Also, we must consider $\lambda^2f(\lambda)=0$. Just as [@Fang-CQG] argued, it is possible that $\lambda^2 f(\lambda)\neq0$ and $\frac{d\lambda}{dN}\neq0$ when $\lambda=0$. Thus the necessary conditionfor the existence of the fixed points with $x\neq 0$ is $\lambda^2f(\lambda)=0$. Taking into account these factors, we can easily obtain all the fixed points of the autonomous system described by Eqs. (\[x’\])-(\[z’\]) and (\[l’\]), and they are shown in Tab. **I**.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Fixed-points $x_c$ $y_c$ $z_c$ $\lambda_c$ Eigenvalues Stability
-------------- ------------------------------------------- ------------------------------------------------- ------- ------------- ----------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------- --
$P_1$ $0$ $0$ $0$ $0$ $\mathbf{M}^T=(0,-3\gamma,\frac32\gamma,-3+\frac32\gamma)$ U, for all $\gamma$
$P_2$ $0$ $0$ $0$ $\lambda_*$ $\mathbf{M}^T=(0,\frac32\gamma,-3\gamma,-3+\frac32\gamma)$ U, for all $\gamma$
S, for $\gamma=1,f_1(0)\geq0$
$P_3$ $0$ $1$ 0 $0$ $\mathbf{M}^T=(-3,-3\gamma,0,0)$ U, for $\gamma=\frac43$, if $f_1(0)\neq 0$
S, for $\gamma=\frac43$, if $f_1(0)=0$
$P_4$ $1$ $0$ $0$ $0$ $\mathbf{M}^T=(0,-6,0,6-3\gamma)$ U,for all $\gamma$
$P_5$ $-1$ $0$ $0$ $0$ $\mathbf{M}^T=(0,-6,0,6-3\gamma)$ U,for all $\gamma$
$P_6$ $0$ $0$ $0$ $\lambda_a$ $\mathbf{M}^T=(0,\frac32\gamma,-3\gamma,-3+\frac32\gamma)$ U,for all $\gamma$
$P_7$ $1$ $0$ $0$ $\lambda_a$ $\mathbf{M}^T= U,for all $\gamma$
\left(-6,6-3\gamma,\frac12\sqrt{6}\lambda_a+3,\sqrt{6}\lambda_a A\right)$
$P_8$ $-1$ $0$ $0$ $\lambda_a$ $\mathbf{M}^T= U,for all $\gamma$
\left(-6,6-3\gamma,-\frac12\sqrt{6}\lambda_a+3,-\sqrt{6}\lambda_aA\right)$
$P_9$ $-\frac{\sqrt{6}}{6}\lambda_a$ $\sqrt{1-\frac{\lambda^2_a}{6}}$ 0 $\lambda_a$ $\mathbf{M}^T=\left(-\lambda_a^2,-3+\frac12\lambda_a^2,\lambda_a^2-3\gamma,-\lambda^3_a-f_1(\lambda_a) \right)$ S, for $f_1(\lambda_a)>\lambda_a$, and $\lambda_a<3\gamma$
U, for $f_1(\lambda_a)<\lambda_a$ and/or $\lambda_a>3\gamma$
$P_{10}$ $-\sqrt{\frac32}\frac{\gamma}{\lambda_a}$ $\sqrt{\frac{3}{2\lambda_a^2}\gamma(2-\gamma)}$ 0 $\lambda_a$ See the Eq. (\[p10\]) U, for all $\gamma$
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
: The stability analysis of an autonomous system in LQC. The system is described by a self-interacting scalar field $\phi$ with positive potential $V$ coupled with a barotropic fluid $\rho_\gamma$. Explanation of the symbols used in this table: $P_{i}$ denotes the fixed points located in the four dimensionsal phase space, which are earmarked by the coordinates $(x_c,y_c,z_c,\lambda_c)$. $\lambda_*$ means that $\lambda$ can be any value. $\lambda_a$ is the value that makes $f(\lambda)=0$. $\mathbf{M}^T$ means the inverted matrix of the eigenvalues of the fixed points. $f_1(\Lambda)=\left.\frac{df(\lambda)}{d\lambda}\right|_{\lambda=\Lambda}$ with $\Lambda=0, \lambda_a$. $A=\left[2f(\lambda_a)+\lambda_a\left(
\left.\frac{df(\lambda)}{d\lambda}\right|_{\lambda_a}\right)\right]$. U stands for unstable, and S stands for stable.
The properties of each fixed points are determined by the eigenvalues of the Jacobi matrix $$\begin{aligned}
\mathcal{M}= \left. \begin{pmatrix}
\frac{\partial F_1}{\partial x}&\frac{\partial F_1}{\partial y}&\frac{\partial F_1}{\partial z}&\frac{\partial F_1}{\partial \lambda}\\
\frac{\partial F_2}{\partial x}&\frac{\partial F_2}{\partial y}&\frac{\partial F_2}{\partial z}&\frac{\partial F_2}{\partial \lambda}\\
\frac{\partial F_3}{\partial x}&\frac{\partial F_3}{\partial y}&\frac{\partial F_3}{\partial z}&\frac{\partial F_3}{\partial \lambda}\\
\frac{\partial F_4}{\partial x}&\frac{\partial F_4}{\partial
y}&\frac{\partial F_4}{\partial z}&\frac{\partial F_4}{\partial
\lambda}
\end{pmatrix}
\right|_{(x_c,y_c,z_c,\lambda_c)}.\end{aligned}$$ According to Lyapunov’s linearization method, the stability of a linearized system is determined by the eigenvalues of the matrix $\mathcal{M}$ (see Chapter 3 of [@Slotine-book]). If all of the eigenvalues are strictly in the left-half complex plane, then the autonomous system is stable. If at least one eigenvalue is strictly in the right-half complex plane, then the system is unstable. If all of the eigenvalues are in the left-half complex plane, but at least one of them is on the $i\omega$ axis, then one cannot conclude anything definite about the stability from the linear approximation. By examining the eigenvalues of the matrix $\mathcal{M}$ for each fixed point shown in Table **I**, we find that points $P_{1,2,4-8,10}$ are unstable and point $P_{9}$ is stable only under some conditions. We cannot determine the stability properties of $P_{3}$ from the eigenvalues, and we will give the full analysis of $P_3$ in the Appendix.
Some remarks on Tab.**I**:
1. Apparently, points $P_2$ and $P_6$ have the same eigenvalues, and the difference between these two points is just on the value of $\lambda$. As noted in the caption of Table **I**, $\lambda_*$ means that $\lambda$ can be any value, and $\lambda_a$ is just the value that makes $f(\lambda)=0$. Obviously, $\lambda_a$ is just a special value of $\lambda_*$, and point $P_6$ is a special case of point $P_2$. $P_6$ is connected with $f(\lambda)$, but $P_2$ is not. From now on, we do not consider separately the special case of $P_6$ when we discuss the property of $P_2$. Hence the value of $\lambda_a$ is contained in our discussion of$\lambda_*$.
2. It is straightforward to check that, if $x_c=\lambda_c=0$, $y_c$ can be any value $y_*$ when it is greater than or equal $1$. But, if $y_*>1$, then $z_c=1-1/y_*^2<1$, and this means that the fixed point is located in the quantum-dominated regions. Although the stability of this point in the quantum regions may depend on $f(\lambda)$, it is not necessary to analyze its dynamical properties, since it does not have any physical meanings. The reason is the following: If the Universe is stable, it will not evolve to today’s pictures. If the Universe is unstable, it will always be unstable. We will just focus on point $P_3$ staying in the classical regions. Then $y_c=y_*=1,z_c=1-1/y_*^2=0$, i.e., for the classical cosmology region, $\rho\ll\rho_c$.
3. Since the adiabatic index $\gamma$ satisfies $0<\gamma< 2$ (in particular, for radiation $\gamma=\frac43$ and for dust $\gamma=1$), all the terms that contain $\gamma$ should not change sign. A more general situation of $\gamma$ is $0\leq \gamma\leq 2$ [@Billyard-scaling]. If $\gamma=0$ or $\gamma=2$, the eigenvalues corresponding to points $P_{1,2,4,5,9}$ will have some zero elements and some negative ones. To analyze the stability of these points, we need to resort to other more complex methods, just as we do in the Appendix for the dynamical properties of point $P_3$. In this paper, we just consider the barotropic fluid which includes radiation and dust, and $\gamma\neq 0,2$. Notice that if one considers $\gamma=0$, the barotropic fluid describes the dark energy. This is an interesting topic, but will not be considered here for the sake of simplicity.
4. $-\sqrt{6}<\lambda_a<\sqrt{6},\lambda_a\neq 0$ should hold for point $P_9$, hence $-3+\frac12\lambda^2_a<0$.
5. $\lambda_a>0$ should hold, since $y_c>0$ for point $P_{10}$. The eigenvalue of this point is $$\begin{aligned}
\label{p10}
\mathbf{M}=\begin{pmatrix}
-3\gamma\\-3\lambda_a\gamma f_1(\lambda_a)\\-\frac32+\frac34\gamma+\frac{3}{4\lambda_a}\sqrt{(2-\gamma)(\lambda^2_a(2-\gamma)+8\gamma+24\gamma^2)}\\
-\frac32+\frac34\gamma-\frac{3}{4\lambda_a}\sqrt{(2-\gamma)(\lambda^2_a(2-\gamma)+8\gamma+24\gamma^2)}
\end{pmatrix}.\nonumber\\\end{aligned}$$ Since we just consider $0<\gamma<2$ in this paper, it is easy to check that $(2-\gamma)(\lambda^2_a(2-\gamma)+8\gamma+24\gamma^2)>0$ is always satisfied. And this point is unstable with $f_1(\lambda_a)=\left.\frac{df(\lambda)}{d\lambda}\right|_{\lambda=\lambda_a}$ being either negative or positive, since $-\frac32+\frac34\gamma+\frac{3}{4\lambda_a}\sqrt{(2-\gamma)(\lambda^2_a(2-\gamma)+8\gamma+24\gamma^2)}$ is always positive.
Based on Table **I** and the related remarks above, we have the folloing conclusions:
1. Points $P_{1,2}$: The related critical values, eigenvalues and stability properties do not depend on the specific form of the potential, since $\lambda_c=0$ or $\lambda$ can be any value $\lambda_*$.
2. Point $P_3$: The related stability properties depend on $f_1(0)=\left.\frac{df(\lambda)}{d\lambda}\right|_{\lambda=0}$.
3. Points $P_{4,5}$: The related eigenvalues and stability properties do not depend on the form of the potential, but the critical values of these points should satisfy $\lambda^2 f(\lambda)=0$ since $x_c\neq 0$.
4. Point $P_6$: It is a special case of $P_2$, but $f(\lambda_a)=0$ should be satisfied.
5. Points $P_{7,8}$: Same as $P_6$, they would not exist if $f(\lambda_a)\neq 0$.
6. Point $P_{9,10}$: $f(\lambda_a)=0$ should hold. The fixed values and the eigenvalues of these two points depend on $f_1(\lambda_a)=
\left.\frac{df(\lambda)}{d\lambda}\right|_{\lambda=\lambda_a}$.
Thus, only points $P_{1,2}$ are independent of $f(\lambda)$.
Comparing the fixed points in LQC and the ones in classical cosmology (see the Table **I** of [@Fang-CQG]), we can see that, even though the values of the coordinates $(x_{c},y_{c},\lambda_{c})$ are the same, the stability properties are very different. This is reasonable, because the quantum modification is considered, and the autonomous system in the LQC scenario is very different from the one in the classical scenario, e.g., the autonomous system is four dimensional in LQC but three dimensional in the classical scenario. Notice that all of the fixed points lie in the classical regions, and therefore the coordinates of fixed points remain the same from classical to LQC, which we also pointed out in an earlier paper [@Xiao-dynamical].
Now we focus on the late time attractors: point $P_{3}$ under the conditions of $\gamma=1, f_1(0)\geq 0$ and $\gamma=4/3, f_1(0)=0$, and point $P_{9}$ under the conditions of $\lambda_a^2<6,
f_1(\lambda_a)>\lambda_a,\lambda_a<3\gamma$. Obviously, these points are scalar-field dominated, since ${\rho_\gamma}=H^2(1/(1-z_c)-x_c^2-y_c^2)=0$. For point $P_{3}$, the effective adiabatic index $\gamma_\phi=(\rho_\phi+p_\phi)/\rho_\phi=0$, which means that the scalar field is an effective cosmological constant. For point $P_{9}$, $\gamma_\phi=\lambda^2_a/2$. This describes a scaling solution that, as the universe evolves, the kinetic energy and the potential energy of the scalar field scale together. And we can see that there is not any barotropic fluid coupled with the scalar-field-dominated scaling solution. This is different from the dynamical behavior of scalar field with exponential potential $V=V_0
\exp(-\lambda\kappa\phi)$ in classical cosmology [@Copeland-exponential; @Hao-PRD1; @Hao-PRD2; @Samart-Phantom; @Li-O(N); @Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom], and also is different from the properties of the scalar field in brane cosmology [@Copeland-scalar], in which $\lambda=\rm{const.}$ (notice that the definition of $\lambda$ in [@Copeland-scalar] is different from the one in this paper) and $\Gamma$ is a function of $L(\rho(a))$ and $|V|$. In these models, the Universe may enter a stage dominated by a scalar field coupled with fluid when $\lambda,\gamma$ satisfy some conditions [@Copeland-exponential; @Copeland-scalar].
We discuss the dynamical behavior of the scalar field by considering $\Gamma$ as a function of $\lambda$ in this and the preceding sections. But $\Gamma$ can not always be treated as a function of $\lambda$. We need to consider a more general autonomous system, which we will introduce in the next section.
Further discussion of the autonomous system {#sec4}
===========================================
The dynamical behavior of the scalar field has been discussed by many authors (e.g., see [@Copeland-IJMPD; @Coley-Book; @Copeland-exponential; @Hao-PRD1; @Hao-PRD2; @Samart-Phantom; @Li-O(N); @Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom]). If one wants to get the potentials that yield the cosmological scaling solutions beyond the exponential potential, one can add a $\frac{d\phi}{dN}$ term into the autonomous system [@Nunes-scaling]. All of these methods deal with special cases of the dynamical behavior of scalar fields in backgrounds of some specific forms. By considering $\Gamma$ as a function of $\lambda$, one can treat potentials of more general forms and get the common fixed points of the general potential, as shown in [@Zhou-plb; @Fang-CQG] and in the two preceding sections. However, as is discussed in [@Fang-CQG], sometimes $\Gamma$ is not a function of $\lambda$, and then the dynamical behaviors of the scalar fields discussed above are still not general in the strict sense. For a more general discussion, we must consider the higher-order derivatives of the potential. We define $$\begin{aligned}
&&{}^{(1)}\Gamma=\frac{VV_3}{V'^2}, \quad {}^{(2)}\Gamma
=\frac{VV_4}{V'^2},\quad{}^{(3)}\Gamma=\frac{VV_5}{V'^2},\nonumber\\
&&\cdots\quad{}^{(n)}\Gamma=\frac{VV_{n+2}}{V'^2},\quad \cdots\end{aligned}$$ in which $V_{n}=\frac{d^n V}{d\phi^n},n=3,4,5,\cdots$. Then we can get $$\begin{aligned}
\frac{d\Gamma}{d N}
&=&\sqrt{6}x\left[\Gamma\lambda+{}^{(1)}\Gamma-2\lambda\Gamma^2\right],\label{G'}\\
\frac{d\left({}^{(1)}\Gamma\right)}{dN}
&=&\sqrt{6}x\left[{}^{(1)}\Gamma\lambda +{}^{(2)}\Gamma
-2\lambda\Gamma\left({}^{(1)}\Gamma\right)\right],\label{G3'}\\
\frac{d\left({}^{(2)}\Gamma\right)}{dN} &=&\sqrt{6}x\left[
{}^{(2)}\Gamma\lambda+{}^{(3)}\Gamma
-2 \lambda\Gamma\left({}^{(2)}\Gamma\right)\right],\label{G4'}\\
\frac{d\left({}^{(3)}\Gamma\right)}{dN}
&=&\sqrt{6}x\left[{}^{(3)}\Gamma\lambda+{}^{(4)}\Gamma
-2\lambda\Gamma\left({}^{(3)}\Gamma\right)\right],\label{G5'}\\
&& \cdots\cdots\nonumber\\
\frac{d\left({}^{(n)}\Gamma\right)}{dN}
&=&\sqrt{6}x\left[{}^{(n)}\Gamma\lambda+{}^{(n+1)}\Gamma
-2\lambda\Gamma \left({}^{(n)}\Gamma\right)\right],\label{Gn'}\\
&& \cdots\cdots\nonumber\end{aligned}$$ To discuss the dynamical behavior of scalar field with more general potential, e.g., when neither $\lambda$ nor $\Gamma$ is constant, we need to consider a dynamical system described by Eqs. (\[x’\])-(\[l\]) coupled with Eqs. (\[G’\])-(\[Gn’\]). It is easy to see that this dynamical system is also an autonomous one. We can discuss the values of the fixed points of this autonomous system. Considering Eq. (\[l\]), we can see that the values of fixed points should satisfy $x_c=0$, $\lambda_c=0$, or $\Gamma_c=1$. Then, we can get the fixed points of this infinite-dimensional autonomous system.
1. If $x_c=0$, considering Eqs. (\[x’\])-(\[z’\]), one can get $(y_c,z_c,\lambda_c)=(0,0,0)$ or $(y_c,z_c,\lambda_c)=(0,0,\lambda_*)$, and $\Gamma_c,{}^{(n)}\Gamma_c$ can be any values.
2. If $\lambda_c=0$, considering Eqs. (\[x’\])-(\[z’\]), one can see that the fixed points of $(x,y,z)$ are $(x_c,y_c,z_c)=(0,y_*,1-1/y_*^2)$and $(x_c,y_c,z_c)=(\pm1,0,0)$. If $x_c=0$, $\Gamma_c$ and ${}^{(n)}\Gamma_c$ can be any values, and if $x_c=\pm 1$, ${}^{(n)}\Gamma_c=0$.
3. If $\Gamma_c=1$, considering Eqs. (\[x’\])-(\[z’\]), one can get that the fixed points of $(x,y,z,\lambda)$ are $(x_c,y_c,z_c,\lambda_c)=(0,0,0,\lambda_*)$ and $(x_c,y_c,z_c,\lambda_c)=(\pm 1,0,0,\lambda_*)$. And ${}^{(n)}\Gamma_c$ should satisfy ${}^{(n)}\Gamma_c=\lambda_*^n$. There are other fixed points, which will be discussed below.
Based on the above analysis and Table **I**, one can find that points $P_{1-10}$ are just special cases of the fixed points of an infinite-dimensional autonomous systems. Considering the definition of $\Gamma$ (see Eq. (\[Gamma\])), the simplest potential is an exponential potential when $\Gamma_c=1$. The properties of these fixed points have been discussed by many authors [@Copeland-exponential; @Hao-PRD1; @Hao-PRD2; @Samart-Phantom; @Li-O(N); @Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom]. If $x_c=0$ and $y_c=0$, this corresponds to a fluid-dominated universe, which we do not consider here. If $x_c=\pm 1$, $\Gamma_c=0$ and ${}^{(n)}\Gamma_c=0$, we do not need to consider the $\Gamma$ and the $^{(n)}\Gamma$ terms. Then the stability properties of these points are the same as points $P_{4,5}$ in Table **I**, and there are unstable points. The last case is $(x_c,y_c,z_c,\lambda_c)=(0,y_*,1-1/y_*^2,0)$ and $\Gamma,{}^{(n)}\Gamma$ can be any value. To analyze the dynamical properties of this autonomous system, we need to consider the ${}^{(n)}\Gamma_c$ terms. We will get an infinite series. In order to solve this infinite series, we must truncate it by setting a sufficiently high-order ${}^{(M)}\Gamma$ to be a constant, for a positive integer $M$, so that $d\left({}^{(M)}\Gamma\right)/dN=0$. Thus we can get an $(M+4)$-dimensional autonomous system. One example is the quadratic potential $V=\frac12m^2\phi^2$ with some positive constant $m$ that gives a five dimensional autonomous system, and another example is the Polynomial (concave) potential $V=M^{4-n}\phi^n$ [@Lindle-potential] that gives an $(n+3)$-dimensional autonomous system. Following the method we used in the two preceding sections, we can get the dynamical behavior of such finite-dimensional systems.
In the remainder of this section, we discuss whether this autonomous system has a scaling solution.
If $x_c=0$, then $\Gamma_c\neq 0,{}^{(n)}\Gamma_c\neq 0$, and the stability of the fixed points may depend on the truncation. As an example, if we choose ${}^{(2)}\Gamma=0$, then we can get a six dimensional autonomous system. The eigenvalues for the fixed point $(x_c,y_c,z_c,\lambda_c,\Gamma_c,{}^{(1)}\Gamma_c)=(0,0,0,\lambda_b,\Gamma_*,{}^{(1)}\Gamma_*)$, where $\lambda_b=0$ or $\lambda_b=\lambda_*$, is $$\begin{aligned}
\mathbf{M}^T=(0,0,0,\frac32\gamma,-3\gamma,-3+\frac32\gamma).\nonumber\end{aligned}$$ Obviously, this is an unstable point, and it has no scaling solution. The eigenvalues for the fixed point $(x_c,y_c,z_c,\lambda_c,\Gamma_c,{}^{(1)}\Gamma_c)=(0,1,0,0,\Gamma_*,{}^{(1)}\Gamma_*)$ is $$\begin{aligned}
\mathbf{M}^T=(0,0,0,0,-3\gamma,-3-3\gamma).\nonumber\end{aligned}$$ According to the center manifold theorem (see Chapter 8 of [@Khalil-non], or [@DynamicalReduction]), there are two nonzero eigenvalues, and we need to reduce the dynamical system to two dimensions to get the stability properties of the autonomous system. This point may have scaling solution, but we need more complex mathematical method. We discuss this problem in another paper [@Xiao-scaling].
We discuss the last case. If $\Gamma_c=1$, we can consider an exponential potential. Then the autonomous system is reduced to three dimensions. It is easy to check that the values $(x_{ec},y_{ec},z_{ec})$ of the fixed points are just the values $(x_c,y_c,z_c)$ of points $P_{6-10}$ in Table **I**. We focus on the two special fixed points: $$\begin{aligned}
&& F_1: (x_{ec},y_{ec},z_{ec})=(-\lambda/\sqrt{6}, \sqrt{1-\lambda^2/6}, 0),\nonumber\\
&&F_2: (x_{ec},y_{ec},z_{ec})=(-\sqrt{3/2}\gamma/\lambda,
\sqrt{3\gamma(2-\gamma)/(2\lambda^2)}, 0).\nonumber\end{aligned}$$ Using Lyapunov’s linearization method, we can find that $F_2$ is unstable and $F_1$ is stable if $\lambda<3\gamma$. It is easy to check that $\rho_\gamma=H^2[1/(1-z_{ec})-x_{ec}^2-y_{ec}^2]=0$ when $(x_{ec},y_{ec},z_{ec})=(-\lambda/\sqrt{6}, \sqrt{1-\lambda^2/6},
0)$. From the above analysis, we find that there is just the scalar-field-dominated scaling solution when we consider the autonomous system to be described by a self-interacting scalar field coupled with a barotropic fluid in the LQC scenario.
Conclusions {#sec5}
===========
The aim of this paper is two-fold. We discuss the dynamical behavior of scalar field in the LQC scenario following [@Fang-CQG; @Zhou-plb]. To further analyze the dynamical properties of scalar field with more general potential, we introduce an infinite-dimensional autonomous system.
To discuss the dynamical properties of scalar field in the LQC scenario, we take $\Gamma$ as a function of $\lambda$, and extend the autonomous system from three dimensions to four dimensions. We find this extended autonomous system has more fixed points than the three dimensional one does. And we find that for some fixed points, the function $f(\lambda)$ affects either their values, e.g., for points $P_{4-10}$, or their stability properties, e.g., for points $P_{3,9}$. In other words, the dynamical properties of these points depend on the specific form of the potential. But some other fixed points, e.g., points $P_{1,2}$,are independent of the potential. The properties of these fixed points are satisfied by all scalar fields. We also find that there are two later time attractors, but the Universe is scalar-field dominated since $\rho_\gamma=0$ at these later time attractors.
The method developed by [@Fang-CQG; @Zhou-plb] can describe the dynamical behavior of the scalar field with potential of a more general form than, for example, an exponential potential [@Copeland-exponential; @Hao-PRD1; @Hao-PRD2; @Samart-Phantom; @Li-O(N); @Ferreira-scaling; @Hoogen-scaling; @Billyard-interaction; @Fu-phantom]. But it is not all-encompassing. If one wants to discuss the dynamical properties of a scalar field with an arbitrary potential, one needs to consider the higher-order derivatives of the potential $V(\phi)$. Hence the dynamical system will extend from four dimensions to infinite-dimensions. This infinite-dimensional dynamical system is still autonomic, but it is impossible to get all of its dynamical behavior unless one considers $\Gamma_c=1$ which just gives an exponential potential. If one wants to study as much as possible the dynamical properties of this infinite-dimensional autonomous system, one has to consider a truncation that sets ${}^{(M)}\Gamma=\rm{Const.}$, with $M$ above a certain positive integer. Then the infinite-dimensional system can be reduced to $M+4)$ dimensions. And we find that there is just the scalar-field-dominated scaling solution for this autonomous system. We only give out the basic properties of this infinite-dimensional autonomous system in this paper, and will continue the discussion in the paper in [@Xiao-scaling].
We only get the scalar-field-dominated scaling solutions, whether we consider $\Gamma$ as a function of $\lambda$ or consider the higher-order derivatives of the potential. This conclusion is very different from the autonomous system which is just described by a scalar field with an exponential potential [@Samart-Phantom]. This is an interesting problem that awaits further analysis.
K. Xiao thanks Professor X. Li for his help with the center manifold theorem. This work was supported by the National Natural Science Foundation of China, under Grant No. 10875012 and the Fundamental Research Funds for the Central Universities.
The stability properties of the Point $P_3$
===========================================
In Sec. \[sec3\], we point out that it is impossible to get the stability properties of the fixed point if at least one of the eigenvalues of $\mathcal{M}$ is on the $i\omega$ axis with the rest being in the left-half complex plane. The fixed point $P_3$ is exactly such a point. In this appendix, we use the center manifold theorem (see Chapter 8 of [@Khalil-non] , or [@DynamicalReduction]) to get the condition for stability of $P_3$. The coordinates of $P_3$ are $(0,1,0,0)$ and the eigenvalues are $(-3,-3\gamma,0,0)$. First, we transfer $P_3$ to $P'_3$ $(x_c=0,\bar{y}_c=y-1=0,z_c=0, \lambda_c=0)$. In this case, Eqs. (\[x’\])-(\[z’\]) and (\[l’\]) become $$\begin{aligned}
\frac{dx}{dN}&=&-3\,x-\frac12\,\sqrt {6}\lambda\, \left( \bar{y}+1
\right) ^{2}+x \left[ 3\,{x}^
{2}+\frac32\gamma\, \left((1+z)\right.\right.\nonumber\\
&&\left.\left.-{x}^{2}-\left( \bar{y}+1 \right) ^{2} \right) \right]\left( 1-2z\right)\label{X'},\\
\frac{d \bar{y}}{dN}&=&\frac12\,\sqrt {6}\lambda\,x \left( \bar{y}+1
\right) + \left( \bar{y}+1 \right)
\left[3\,{x}^{2}+\frac32\gamma\, \left((1+z)\right.\right.\nonumber\\
&&\left.\left.-{x}^{2}-\left( \bar{y}+1 \right) ^{2} \right)
\right] \left( 1-2z
\right)\label{Y'}, \\
\frac{dz}{dN}&=&-3\gamma z-3z\left(1-z\right)\left[2x^2-\gamma
x^2-\gamma
(\bar{y}+1)^2\right],\label{Z'}\\
\frac{d\lambda}{dN}&=& \sqrt
{6}{\lambda}^{2}\left(f(0)+f_1(0)\lambda\right) x \label{L'},\end{aligned}$$ where we have considered that the related variables $(x,\bar{y},z,\lambda)$ are small around point $(x_c,\bar{y}_c,z_c,\lambda_c)=\left(0,0, 0,0\right)$. Therefore the following Taylor series $$\begin{aligned}
&&\frac{1}{1-z}=1+{z}+\cdots,\nonumber\\
&&f(\lambda)=f(0)+f_1(0)\lambda+\cdots,\nonumber\end{aligned}$$ can be used, where $f_1(0)=\frac{df(\lambda)}{d\lambda}\left.\right|_{\lambda=0}$.
We can get the Jacobi matrix $\mathcal{M'}$ of the dynamical system Eqs. (\[X’\])-(\[L’\]) as $$\begin{aligned}
\mathcal{M'}= \begin{pmatrix}
-3& 0& 0&-\frac{\sqrt{6}} 2\\
0&-3\gamma& \frac32\gamma & 0\\
0&0& 0&0\\
0&0&0&0
\end{pmatrix}.\end{aligned}$$ It is easy to get the eigenvalues and eigenvectors of $\mathcal{M}'$. Let $\mathcal{A}$ denote the matrix whose columns are the eigenvalues, and $\mathcal{S}$ denote the matrix whose columns are the eigenvectors, and then we have $$\begin{aligned}
\mathcal{A}=\begin{pmatrix} -3\\ -3\gamma\\ 0\\ 0
\end{pmatrix},\quad
\mathcal{S}=\begin{pmatrix}
1& 0& -\frac{\sqrt{6}}{6}& 0\\
0& 1& 0 & \frac{1}{2}\\
0 & 0&0& 1\\
0 & 0& 1 & 0
\end{pmatrix}.\end{aligned}$$ With the help of $\mathcal{S}$, we can transform $\mathcal{M}'$ into a block diagonal matrix $$\begin{aligned}
\mathcal{S}^{-1} \mathcal{M}' \mathcal{S}=\begin{pmatrix}
-3 & 0&0&0\\
0& -3\gamma& 0& 0\\
0&0&0&0\\
0&0&0&0
\end{pmatrix}=\begin{pmatrix}
\mathcal{A}_1 &0\\
0& \mathcal{A}_2
\end{pmatrix},\end{aligned}$$ where all eigenvalues of $\mathcal{A}_1$ have negative real parts, and all eigenvalues of $\mathcal{A}_2$ have zero real parts.
Now we change the variables to be $$\begin{aligned}
\begin{pmatrix}
X\\Y\\Z\\ \bar{\lambda}
\end{pmatrix}=
\mathcal{S}^{-1}\begin{pmatrix}
x\\ \bar{y}\\ {z}\\ \lambda
\end{pmatrix}=
\begin{pmatrix}
x+\frac{\sqrt{6}}{6}\lambda\\ \bar{y}-\frac12z\\ \lambda \\z
\end{pmatrix}.\end{aligned}$$ Then, we can rewrite the autonomous system in the form of the new variables: $$\begin{aligned}
\begin{pmatrix}\label{A9}
\frac{dX}{dN}\\ \frac{dY}{dN}\\ \frac{dZ}{dN}\\ \frac{d\bar{\lambda}}{dN}
\end{pmatrix}=
\begin{pmatrix}
-3 & 0&0&\\
0&-3\gamma&0&0\\
0&0&0&0\\
0&0&0&0
\end{pmatrix}
\begin{pmatrix}
X\\Y\\Z\\ \bar{\lambda}
\end{pmatrix}+
\begin{pmatrix}
G_1\\G_2\\G_3\\G_4
\end{pmatrix},\nonumber\\\end{aligned}$$ where $G_i=G_i(X,Y,Z,\bar{\lambda}),(i=1,2,3,4)$ are functions of $X,Y,Z$, and $\bar{\lambda}$. It is easy to get $G_i$ by substituting the transformations $x=X-\frac{\sqrt{6}}{6}Z,
\bar{y}=Y+\frac12\bar{\lambda}, z=\bar{\lambda}, \lambda=Z$ into the R.H.S. of Eqs. (\[X’\])-(\[L’\]).
According to the center manifold theorem [@DynamicalReduction], there exists a $C^\infty$-center manifold $$\begin{aligned}
W^c_{loc}&=&\left\{(X,Y,Z,\bar{\lambda}): X\equiv h_1(Z,\bar{\lambda}), Y\equiv h_2(Z,\bar{\lambda}),\right.\nonumber\\
&&\left.h_i(0,0)=0, J_{h_i}(0,0)=0\right\}\nonumber\end{aligned}$$ such that the dynamics of (\[A9\]) can be restricted to the center manifold. $J_{h_i}$ is the Jacobi matrix of $h_i$, and $h_1(Z,\bar{\lambda}), h_2(Z,\bar{\lambda})$ are $$\begin{aligned}
h_1(Z,\bar{\lambda})=A_1 Z^2+A_2 Z \bar{\lambda}+A_3 \bar{\lambda}^2+\cdots,\label{h1}\\
h_2(Z,\bar{\lambda})=B_1 Z^2+B_2 Z \bar{\lambda}+B_3
\bar{\lambda}^2+\cdots.\label{h2}\end{aligned}$$ We just consider the quadratic forms of $Z$ and $\bar{\lambda}$ in this appendix.
Considering the center manifold theorem, we have $$\begin{aligned}
\frac{dX}{dN}&=&\frac{\partial h_1(Z,\bar{\lambda})}{\partial Z}\frac{dZ}{dN}+\frac{\partial h_1(Z,\bar{\lambda})}{\partial\bar{\lambda}}\frac{d\bar{\lambda}}{dN},\label{X''}\\
\frac{dY}{dN}&=&\frac{\partial h_2(Z,\bar{\lambda})}{\partial Z}\frac{dZ}{dN}+\frac{\partial h_2(Z,\bar{\lambda})}{\partial\bar{\lambda}}\frac{d\bar{\lambda}}{dN}.\label{Y''}\end{aligned}$$ Inserting the Eqs. (\[h1\]) and (\[h2\]) into $dX/dN,dY/dN$ in Eq. (\[A9\]) and Eqs. (\[X”\])-(\[Y”\]), and comparing the coefficients of $dX/dN$ and $dY/dN$, we get $$\begin{aligned}
&&A_1=0,\quad A_2=\frac{\sqrt{6}}{6},\quad A_3=0, \quad B_1=\frac{1}{12},\nonumber\\
&& B_2=0, \quad B_3=\frac{1}{8}.\end{aligned}$$ Then, the dynamics near the origin is governed by the following equations, $$\begin{aligned}
\frac{dZ}{dN}&=&-{Z}^{3}f_1(0),\label{ZF'}\\
\frac{d\bar{\lambda}}{dN}&=&-{Z}^{2}\bar{\lambda}+\gamma\,{Z}^{2}\bar{\lambda}-\frac32\,\gamma\,{\bar{\lambda}}^{3}.\label{LF'}\end{aligned}$$ We consider two different values of $\gamma$ to get the stability properties of this system. This is because a different $\gamma$ will give a different dynamical systems. The first one to be considered is dust, which has $\gamma=1$. Then, we have $$\begin{aligned}
\frac{dZ}{dN}&=&-{Z}^{3}f_1(0)\label{ZF''},\\
\frac{d\bar{\lambda}}{dN}&=&-\frac32\bar{\lambda}^3.\label{LF''}\end{aligned}$$ According to Lyapunov’s theorem, we can define a Lyapunov function to analyze the stability properties of a dynamical system. Different dynamical systems have different Lyapunov functions, and one dynamical system can also have different Lyapunov functions. But all the Lyapunov functions $U$ should satisfy $U(\mathbf{x})\geq 0$ at the original point (Chapter 2 of [@Khalil-non]). Then we can define $$\begin{aligned}
U_1=\frac12\left(Z^2+\bar{\lambda}^2\right).\end{aligned}$$ Using Eqs.(\[ZF”\]) and (\[LF”\]), wehave $$\begin{aligned}
\frac{dU_1}{dN}=-f_1(0)Z^4-\frac32Z^2\bar{\lambda}^4.\end{aligned}$$ According to Lyapunov¡¯s stability theorems, the system is stable if $f_1(0)\geq 0$.
Now we turn to considering radiation, which has $\gamma=\frac43$. Eqs. (\[ZF’\]) and (\[LF’\]) become $$\begin{aligned}
\frac{dZ}{dN}&=&-{Z}^{3}f_1(0),\label{zc}\\
\frac{d\bar{\lambda}}{dN}&=&-2\bar{\lambda}^3+\frac13Z^2\bar{\lambda}.\label{lc}\end{aligned}$$ We need to consider three possible cases: (a) $f_1(0)\neq 0$, (b) $f_1(0)=0, Z(N=0)=0$, and (c) $f_1(0)=0, Z(N=0)\neq 0$, since these three different cases will bring out three different dynamical systems.
If $f_1(0)\neq 0$, the Lyapunov function can be defined as $$\begin{aligned}
U_2=\frac{1}{1+Z^2/(6A)+\bar{\lambda}^2},\end{aligned}$$ where $A=f_1(0)$ if $f(0)>0$, and $A=-f_1(0)$ if $f_1(0)<0$. Then one can get $$\begin{aligned}
\frac{dU_2}{dN}=\frac{12A^2\left[(Z^2-\bar{\lambda}^2)^2+5\bar{\lambda}^4 \right]}{\left[6A+6A\bar{\lambda}^2+Z^2\right]^2}>0.\end{aligned}$$ Then this point is an unstable one.
If $f_1(0)=0$ and $Z(N=0)=0$, Eq. (\[lc\]) becomes $d\bar{\lambda}/{dN}=-2\bar{\lambda}^3$. Defining Lyapunov function, $$\begin{aligned}
U_3=1+\bar{\lambda}^2,\end{aligned}$$ then $$\begin{aligned}
\frac{dU_3}{dN}=-4\bar{\lambda}^4\leq 0.\end{aligned}$$
If $f_1(0)=0$ and $Z(N=0)\neq 0$, one can get $Z=C$ from Eq. (\[zc\]), with a non-zero constant $C$. Equation (\[lc\]) becomes $$\begin{aligned}
\frac{d\bar{\lambda}}{dN}=-2\bar{\lambda}^3+\frac13C^2\bar{\lambda},\end{aligned}$$ The Lyapunov function can be defined as $$\begin{aligned}
U_4=\left(1-\frac{6}{C^2}\bar{\lambda}^2\right)^2,\end{aligned}$$ Then we have $$\begin{aligned}
\frac{dU_4}{dN}=-\frac{8}{C^4}\bar{\lambda}^2\left(C^2-6\bar{\lambda}^2\right)^2\leq 0.\end{aligned}$$
Obviously, according to Lyapunov¡¯s stability theorem, this point is stable as long as $f_1(0)=0$, regardless of $Z(N=0)= 0$ or $Z(N=0)\neq 0$.
[99]{} E. J. Copeland, M. Sami, and S. Tsujikawa, Int. J. Mod. Phys. D **15**, 1753(2006). A. A. Coley, *Dynamical systems and cosmology*, Kluwer Academic Publishers, Dordrecht/Boston/London, 2003. E.J. Copenland, S. Mizuno, and M. Shaeri, Phys. Rev. D **79**, 103515(2009). G. Leon, P. Silveira and C. R. Fadragas, arXiv: 1009.0689. E.J. Copeland, A.R. Liddle, and D. Wands, Phys. Rev. D **57**, 4686(1998). J.G. Hao, and X.Z. Li, Phys. Rev. D **67**, 107303(2003). J.G. Hao, and X.Z. Li, Phys. Rev. D **70**, 043529(2004). D. Samart and B. Gumjudpai, Phys. Rev. D **76**, 043514(2007). X.Z. Li, and J.G. Hao, Phys. Rev. D **69**, 107303(2004). P. G. Ferreira, and M. Joyce, Phys. Rev. D **58**, 023503(1998). R.J. van den Hoogen, A.A. Coley, and D. Wands, Class. Quantum Grav.**16**, 1843(1999). A.P. Billyard, and A.A. Coley, Phys. Rev. D **61**, 083503(2000). X.Y. Fu, H.W. Yu, and P.X. Wu, Phys. Rev. D **78**, 063001(2008). A. de la Macorra, and G. Piccinelli, Phys. Rev. D **61**, 123503(2000). S.C.C. Ng, N.J. Nunes, and F. Rosati, Phys. Rev. D **64**, 083510(2001). S.Y. Zhou, Phys. Lett. B **660**, 7(2008). W. Fang, Y. Li, K. Zhang, and H.Q. Lu, Class. Quant. Grav. **26**, 155005(2009). Martin Bojowald, Living Rev. Rel. **11**, 4(2008). AbhayAshtekar, J. Phys .Conf. Ser. **189**, 012003(2009). Gen. Rel. Grav. **41**, 707(2009). C. Rovelli, *Quantum Gravity*, Cambridge university press, Cambridge, 2004. T. Thiemann, *Modern Canonical Quantum General Relativity*, Cambridge university press, Cambridge, 2007. A. Ashtekar, T. Pawlowshik, and P. Singh, Phys. Rev. Lett. **96**, 141301(2006). Phys. Rev. D **73**, 124038(2006). P. Singh and A. Toporensky, Phys. Rev. D **69**, 104008(2004). G.V. Vereshchagin, J. Cosmol. Astropart.Phys.**0407**, 013(2004). G. Date and G.M. Hossain, Phys. Rev. Lett. **94**, 011302(2005). M. Bojowald, Phys. Rev. Lett. **89**, 261301(2002). M. Bojowald and K. Vandersloot, Phys. Rev. D **67**, 124023(2003). M. Bojowald, J.E. Lidsey, D.J. Mulryne, P. Singh, and R. Tavakol, Phys. Rev. D **70**, 043530(2004). C. Germani, W. Nelson, M. Sakellariadou, Phys. Rev. D **76**, 043529(2007). E.J. Copeland, D.J. Mulryne, N.J. Nunes, and M. Shaeri, Phys. Rev. D **77**, 023510(2008). A. Ashtekar, D. Sloan, Phys. Lett. B **694**, 108(2010). A. Corichi, A, Karami, *On the measure problem in slow roll inflation and loop quantum cosmology*, arXiv:1011.4249. A. Ashtekar, T. Pawlowski, and P. Singh, Phys. Rev. D **74**, 084003(2006). H. Wei and S.N. Zhang, Phys. Rev. D **76**, 063005(2007). X.Y. Fu, H.W. Yu and P.X. Wu, Phys. Rev. D **78**, 063001(2008). S. Li and Y.G. Ma, Eur. Phys. J. C **68**, 227(2010). Raphael Lamon, Andreas J. Woehr, Phys.Rev.D **81**, 024026(2010). K. Xiao, and J.Y. Zhu, Int. J. Mod. Phys.A, **25**, 4993(2010). J.E. Lidsey, J. Cosmol. Astropart.Phys. **0412**, 007(2004). I.P.C. Heard and D. Wands, Classical Quantum Gravity, **19**, 5435(2002). J.J.E. Slotine and W.P. Li, *Applied nonlinear control*, Englewood Cliffs, NJ: Prentice-Hall, 1991. A.P. Billyard, A.A. Coley, and R.J. van den Hoogen, Phys. Rev. D **58**, 123501(1998). A. Nunes, J.P. Mimoso, Phys. Lett. B **488**, 423(2000). H.K. Khalil, *Nonlinear system*, Englewood Cliffs, NJ: Prentice Hall, 1996. W.O. Bray, *Dynamcial System Reduction: The center manifold*, URL: http://www.math.umaine.edu/$\sim$bray/\
Archive$\_$res. R. Kallosh, J. Kratochvil, A. Linde, E.V. inder and M. Shmakova, J. Cosmol. Astropart.Phys. **0310**, 015(2003). A. D. Linde, Phys. Lett. B **259**, 38(1991); Phys. Rev. D. **49**, 748(1994). K. Xiao and J.Y. Zhu, *Scaling solution in loop quantum cosmology*.
[^1]: Author to whom correspondence should be addressed
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $T_{f}$ be a circle homeomorphism with two break points $a_{b},c_{b}$ and irrational rotation number $\varrho_{f}$. Suppose that the derivative $Df$ of its lift $f$ is absolutely continuous on every connected interval of the set $S^{1}\backslash\{a_{b},c_{b}\}$, that $DlogDf \in L^{1}$ and the product of the jump ratios of $ Df $ at the break points is nontrivial, i.e. $\frac{Df_{-}(a_{b})}{Df_{+}(a_{b})}\frac{Df_{-}(c_{b})}{Df_{+}(c_{b})}\neq1$. We prove that the unique $T_{f}$- invariant probability measure $\mu_{f}$ is then singular with respect to Lebesgue measure $l$ on $S^{1}$.'
title: |
Singular measures of circle homeomorphisms\
with two break points
---
\[section\] \[theo\][Definition]{} \[theo\][Lemma]{} \[theo\][Proposition]{} \[theo\][Remark]{} \[theo\][Example]{} \[theo\][Corollary]{} Ł
[**Singular measures of circle homeomorphisms with two breakpoints**]{} [^1]\
Introduction
============
Circle homeomorphisms constitute one important class of one-dimentional dynamical systems.The investigation of their properties was initiated by Poincaré [@Po1885], who came across them in his studies of differential equations more than a century ago. Since then interest in these maps never diminished. Circle maps are also important because of their applications to natural sciences (see for instance [@Cv1989]). Let $T_{f}$ be an orientation preserving circle homeomorphism with lift $f:\mathbb{R}\rightarrow \mathbb{R}$, $f$ continuous, strictly increasing and $f(x+1)=f(x)+1$, $x\in \mathbb{R}$. We identify the unit circle $S^{1}=\mathbb{R}/ \mathbb{Z}$ with the half open interval $[0,1)$. The circle homeomorphism $T_{f}$ is then defined by $T_{f}x=f(x)$ mod 1, $x\in S^{1}$. An important conjugacy invariant characteristic of orientation preserving homeomorphisms is the rotation number $\varrho(f)$. If $T_{f}$ is a circle homeomorphism with lift $f$, then the rotation number $\varrho=\varrho(f)$ is defined by $$\rho(f)=\left( \overset{}{\underset{n\rightarrow \infty}{\lim
}}\frac{f^{n}(x)}{n}\right) \text{mod}\ 1,$$ with $f^{n}$ the $n$-th iterate of $f$. This limit exists and is independent of the choice of the lift and the point $x\in \mathbb{R}$. If $\varrho$ is irrational, then for sufficiently smooth diffeomorphisms the trajectory of an arbitrary point is dense on the circle, and the diffeomorphism itself can be reduced to the pure rotation $T_{\varrho}x=(x+\varrho) $ mod 1 by an angle $\varrho$ through a change of coordinates. This result was proved by Denjoy [@De1932]. More precisely, Denjoy proved that if $f\in C^{1}(R^{1})$ and $var(\text{log}Df)<\infty$, then there exists a circle homeomorphism $T_{\varphi}$ such that $$\label{eq1}
T_{f}\circ T_{\varphi}=T_{\varphi}\circ T_{\varrho}.$$ It is a well known fact that a circle homeomorhism $T_{f}$ with irrational rotation number $\varrho$ is strictly ergodic i.e. admits an unique $T_{f}$-invariant probability measure $\mu_{f}$. Note, that the conjugating map $T_{\varphi}$ and the invariant measure $\mu_{f}$ are related by $T_{\varphi}x=\mu_{f}([0,x])$ (see [@CFS1982]). This last relation implies that regularity properties of the conjugating map $T_{\varphi}$ are closely related to the existence of an absolutely continuous invariant measure $\mu_{f}$ with a regular density.
The problem of smoothness of the conjugacy of smooth diffeomorphisms is now very well understood(see for instance [@Ar1961; @Mo1966; @He1979; @KO1989; @KS1989; @Yo1984]).
An important result is the one by M. Herman [@He1979]:
\[Theorem 1.\] If $T_{f}$ is a $C^{2}$-diffeomorphism with rotation number $\varrho=\varrho(f)$ of bounded type (that means the entries in the continued fraction expansion of $\varrho$ are bounded) and $T_{f}$ is close to $T_{\varrho}$ then $\mu_{f}$ is absolutely continuous with respect to Lebesgue measure.
Katznelson-Ornstein [@KO1989] and Khanin-Sinai [@KS1989] gave new proofs and an improved global version of this theorem in showing that it is not necessary to assume that $T_{f}$ is close to $T_{\varrho}$:
(Katznelson-Ornstein)\[KO\]. Let $T_{f}$ be an orientation preserving $C^{1}$-circle diffeomorphism. If $f$ is absolutely continuous, $D(logDf)\in L^{p}$ for some $p>1$ and the rotation number $\rho=\rho(f)$ is of bounded type, then the invariant measure $\mu(f)$ is absolutely continuous with respect to Lebesgue measure.
The result proved by Khanin and Sinai in [@KS1989] is the following:
(Khanin-Sinai)\[KS\]. Let $T_{f}$ be a $ C^{2+\varepsilon}$ circle diffeomorphism with $\varepsilon>0$, and let the rotation number $\rho=\rho(f)$ be a Diophantine number with exponent $\delta\in(0,\varepsilon)$, i.e., there is a constant $c(\varrho)$ such that $$|\rho-\frac{p}{q}|\geq\frac{c(\rho)}{q^{2+\delta}}\ \text{for any}\ \frac{p}{q}\in\mathbb{Q}.$$ Then the conjugating map $T_{\varphi}$ belongs to $C^{1+\varepsilon-\delta}$.
Note, that the condition $T_{f}\in C^{2+\varepsilon}$ is sharp, because there is a set of full Lebesgue measure in $[0,1]$ such that for any rotation number in this set there are $C^{2}$-diffeomorphisms for which the conjugating map $T_{\varphi}$ is singular [@HS1982].
An important and interesting class of circle homeomorphisms are homeomorphisms with singularities. The simplest among them are critical circle homeomorphisms and homeomorphisms with break points. We call this latter class [**[$P$-homeomorphisms]{}**]{}. In general their ergodic properties like the invariant measures , their renormalization and also their rigidity properties are different from the properties of diffeomorphisms (see [@dMvS1993] chapter I and IV, [@He1979] chapter VI, [@KhKm2003], [@dFdM1999]).
The invariant measures of critical circle homeomorphisms, that means $C^{3}-$ smooth circle homeomorphisms with a finite number of critical points of polynomial type have been studied in ([@GrSw1993]):
(Graczik-Swiatek)\[GS\]. Critical circle homeomorphisms with irrational rotation number have an invariant measure singular with respect to Lebesgue measure.
The class of $P$-homeomorphisms consists of orientation preserving circle homeomorphisms $T_{f}$ which are differentiable away from countable many points, the so called break points, at which left and right derivatives, denoted respectively by $Df_{-}$ and $Df_{+}$, exist such that
- there exist constants $0<c_{1}<c_{2}<\infty$ with $c_{1}<Df(x)<c_{2}$ for all $x\in S^{1}\setminus BP(f)$,\
$c_{1}<Df_{-}(x_{b})<c_{2}<$ and $c_{1}<Df_{+}(x_{b})<c_{2}$ for all $x_{b}\in BP(f)$, the set of break points of $f$;
- $logDf$ has bounded variation.
In this case $logDf$, $logDf_{-}$, $logDf_{+}$ and $logDf^{-1}$, $logDf_{-}^{-1}$, $logDf_{+}^{-1}$ all have the same total variation denoted by $v=Var(logDf)$.
The ratio $\sigma_{f}(x_{b})= \frac{Df_{-}({c_{b}})}{Df_{+}({c_{b}})}$ is called the jump ratio of $T_{f}$ in $x_{b}$ .
Piecewise linear (PL) orientation preserving circle homeomorphisms with piecewise constant derivatives are the simplest examples of class $P$-homeomorphisms. They occur in many other areas of mathematics such as group theory, homotopy theory and logic via the Thompson groups (see [@St1992]. PL-homeomorphisms were considered first by Herman in [@He1979] as examples of homeomorphisms of arbitrary irrational rotation number which admit no invariant $\sigma -$finite measure equivalent to Lebesgue measure.
(Herman)\[H\]. A PL-circle homeomorphisms with two break points and irrational rotation number has an invariant measure absolutely continuous with respect to Lebesgue measure if and only if its break points belong to the same orbit.
General (non PL) class $P$-homeomorphisms with one break point have been studied by Dzhalilov and Khanin in [@DK1998]. The character of their results for such circle maps is quite different from the one for $C^{2+\varepsilon}$ diffeomorphisms. The main result of [@DK1998] is the following:
\[DK\]. Let $T_{f}$ be a class $P$-homeomorphism with one break point $c_{b}$. If the rotation number $\rho_{f}$ is irrational and $T_{f}\in C^{2+\varepsilon}(S\backslash\{c_{b}\})$ for some $\varepsilon>0$, then the $T_{f}$-invariant probability measure $\mu_{f}$ is singular with respect to Lebesgue measure l on $S^{1}$, i.e. there exists a measurable subset $A\subset S^{1}$ such that $\mu_{f}(A)=1$ and $l(A)=0$.\
I. Liousse proved in [@Li2005] the same result for “generic” PL-homeomorphisms with irrational rotation number of bounded type. In a next step Dzhalilov and I. Liousse studied in [@DL2006] circle homeomorphisms with two break points. Their result is the following:
\[DL\]. Let $T_{f}$ be a class $P$-homeomorphism satisfying the following conditions:
- $T_{f}$ has irrational rotation number $\rho_{f}$ of bounded type;
- there exists constants $k_{i}>0$ such that $|Df(x)-Df(y)|\le k_{i}|x-y|$ on every continuity interval of $Df$;
- $T_{f}$ has two break points not on the same orbit of $T_{f}$.
Then the $T_{f}$- invariant probability measure $\mu_{f}$ is singular with respect to Lebesgue measure.
In the present paper we continue our study of invariant measures for circle homeomorphisms $T_{f}$ with two break points and arbitrary irrational rotation number $\rho_{f}$. The main result of our paper is the following:
\[DLM\]. Let $T_{f}$ be a class $P$-homeomorphism satisfying the following conditions:
- the rotation number $\rho=\rho_{f}$ of $T_{f}$ is irrational;
- $T_{f}$ has two break points $a_{b}$, $c_{b}$ and the product of the jump ratios of $Df$ at the break points is nontrivial i.e. $\sigma_{f}(a_{b})\cdot\sigma_{f}(c_{b})\ne 1$.
- $Df(x)$ is absolutely continuous on every connected interval of $S^{1}\backslash\{a_{b}, c_{b}\}$ and the second derivative $D^{2}f\in L^{1}$;
Then the $T_{f}$- invariant probability measure $\mu_{f}$ is singular with respect to Lebesgue measure.
[Obviously condition (c) is weaker than a Lipschiz condition for $Df$. In the case when $T_{f}$ has two break points on the same orbit our Theorem \[DLM\] gives a new proof of the result in [@DK1998], but with a weaker condition than $C^{2+\varepsilon}$.]{} on $T_{f}$.
A direct consequence of our Theorem \[DLM\] is
\[DLM1\] Let $T_{f}$ be a circle homeomorphism satisfying condition (c) of Theorem \[DLM\] and the conditions
- the rotation number $\rho_{f}$ of $T_{f}$ is irrational of bounded type;
- $T_{f}$ has two break points $a_{b}$, $c_{b}$ with disjoint orbits and $\sigma(a_{b})\sigma(a_{b})=1.$
Then the $T_{f}-$invariant measure $\mu_{f}$ is singular with respect to Lebesgue measure.
Preliminaries and Notations
===========================
Let $T_{f}$ be an orientation preserving homeomorphism of the circle with lift $f$ and irrational rotation number $\rho=\rho_{f}$. We take an arbitrary point $x_{0}\in S^{1}$ and consider the trajectory of this point under the action of $T_{f}$, i.e., the set of points $\{x_{i}=T_{f}^{i}x_{0}, \
i\in\mathbb{Z} \}$. According to a classical theorem of Poincaré (see [@CFS1982]), the order of the points along the trajectory is the same as in the case of the linear rotation $T_{\rho}$ of the circle, i.e. for the sequence $\{\overline{x}_{i}=\{x_{0}+i\rho\}, mod 1, \, i\in\mathbb{Z} \}$. This important property allows one to define a sequence of natural partitions of the circle closely related to the continued fraction expansion of the number $\rho$ .
We denote by $\left\lbrace k_{n}, n \in \mathbb{N} \right\rbrace $ the sequence of entries in the continued fraction expansion of $\rho$, so that $\varrho=\left[k_{1},k_{2},...,k_{n},...\right]=
\left( 1/k_{1}+\left( 1/\left( k_{2}+...+1/k_{n}+...\right) \right) \right) $. For $n \in \mathbb{N}$ we denote by $p_{n}/q_{n}=[k_{1},k_{2},...,k_{n}] $ the convergents of $\rho$. Their denominators $q_{n}$ satisfy the recursion relation $q_{n+1}=k_{n+1}q_{n}+q_{n-1}, \ n\geq 1, \ q_{0}=1, \ q_{1}=k_{1}$.
For an arbitrary point $x_{0}\in S^{1}$ denote by $\Delta_{0}^{(n)}(x_{0})$ the closed interval with endpoints $x_{0}$ and $x_{q_{n}}.$ For $n$ odd $x_{q_{n}}$ is to the left of $x_{0}$, for $n$ even it is to the right. Denote by $\Delta_{i}^{(n)}(x_{0})$ the iterates of the interval $\Delta_{0}^{(n)}(x_{0})$ under $T_{f}$: $\Delta_{i}^{(n)}(x_{0})=T^{i}_f\Delta_{0}^{(n)}(x_{0}),i\ge1$.
It is well known (since Denjoy) that the system of intervals $$\label{eq2}
\xi_{n}(x_{0})=\left\lbrace \Delta_{i}^{(n-1)}(x_{0}), \ 0\leq i<q_{n}; \ \Delta_{j}^{(n)}(x_{0}), \ 0\leq j<q_{n-1}\right\rbrace$$ cover the whole circle and that their interiors are mutually disjoint. The partition $\xi_{n}(x_{0})$ is called the $n$-th **dynamical partition** of the point $x_{0}$ with **generators** $\Delta_{0}^{(n-1)}(x_{0})$ and $\Delta_{0}^{(n)}(x_{0})$. We briefly recall the structure of the dynamical partitions. The passage from $\xi_{n}(x_{0})$ to $\xi_{n+1}(x_{0})$ is simple: all intervals of order $n$ are preserved and each of the intervals $ \Delta_{i}^{(n-1)}(x_{0}), \ 0\leq
i<q_{n},$ is divided into $k_{n+1}+1$ intervals: $$\begin{aligned}
\Delta_{i}^{(n-1)}(x_{0})=\Delta_{i}^{(n+1)}(x_{0})\cup\bigcup_{s=0}^{k_{n+1}-1}\Delta_{i+q_{n-1}+sq_{n}}^{(n)}(x_{0}).\end{aligned}$$ The following Lemma plays a key role for studying metrical properties of the homeomorphism $T_{f}$.
\[lem2.1\] Let $T_{f}$ be a $P$- circle homeomorphism with a finite number of break points $z^{(i)},$ $i=1,2,...,m$ and irrational rotation number $\rho_{f}$. If $x_{0}\in S^{1},$ $ n\ge 1$ and $z^{(i)}\notin\left\lbrace T^{i}x_{0}, 0\leq
i<q_{n},\right\rbrace $ then $$\label{eq3}
e^{-v}\leq \prod_{i=0}^{q_{n}-1}DT_{f}^{i}(x_{0})\leq e^{v},$$ where $v=Var(logDf)$.
Inequality (\[eq3\]) is called the **Denjoy inequality**. The proof of Lemma \[lem2.1\] is just like in the case of diffeomorphisms (see for instance [@KS1989]). Using Lemma \[lem2.1\] it can be shown easily that the lenghts of the intervals of the dynamical partition $\xi_{n}$ in (\[eq2\]) are exponentially small:
\[cor2\] Let $\Delta^{(n)}$ be an arbitrary element of the dynamical partition $\xi_{n}(x_{0})$. Then $$\label{eq4}
l(\Delta^{(n)}) \leq const \,\, \lambda^{n},$$ where $\lambda=(1+e^{-v})^{-1/2}<1.$
\[defi3\] Two homeomorphisms $T_{1}$ and $T_{2}$ of the circle are said to be topologically equivalent if there exists a homeomorphism $T_{\varphi}:S^{1}\rightarrow S^{1}$ such that $T_{\varphi}(T_{1}(x))=T_{2}(T_{\varphi(x)})$ for any $x\in S^{1}$.We call the homeomorphism $T_{\varphi}$ a conjugating map.
From Corollary \[cor2\] it follows that the trajectory of each point is dense in $S^{1}$. This together with the monotonicity of the homeomorphism $T_{f}$ implies the following generalization of the classical Denjoy theorem.\
\[Denjoy\] Suppose that a homeomorphism $T_f$ with an irrational rotation number $\rho$ satisfies the conditions of Lemma \[lem2.1\].Then $T_f$ is topologically equivalent to the linear rotation $T_{\rho}$.
(see [@KO1989])\[defKO\] An interval $I=(\tau,t)\subset S^1$ is $q_{n}$-small and its endpoints $\tau,t$ are $q_{n}$-close if the system of intervals $T_{f}^{i}(I),\ 0\leq i<q_{n}$ are disjoint.
It is known that the interval $(\tau,t)$ is $q_{n}$-small if, depending on the parity of n, either $t\leq\tau\leq
T_{f}^{q_{n-1}}(t)\ \text{or} \ T_{f}^{q_{n-1}}(\tau)\leq
t\leq\tau.$\
\[Def2.3\] [ Let $C>1$ . We call two intervals of $S^{1}$ **C-comparable** if the ratio of their lengts is in $[C^{-1}, C].$]{}
Lemma \[lem2.1\] then implies
\[cor2.2\] Suppose that a homeomorphism $T_f$ with an irrational rotation number $\rho$ satisfies the conditions of Lemma \[lem2.1\]. Then for any interval $I\subset S^{1}$ the intervals $I$ and $T_{f}^{q_{n}}I$ are $e^{v}$-comparable. If the interval $I$ is $q_{n}-$small then $l(T_{f}^{i}I)< const \, \lambda^{n} $ for all $i=1,2, ...,q_{n}-1.$
\[lemm2.2\] Suppose, that a homeomorphism $T_f$ with an irrational rotation number $\rho$ satisfies the conditions of Lemma \[lem2.1\] and $x,y\in S^{1}$ are $q_{n}$-close. Then for any $0\leq l \leq q_{n}$ the following inequality holds: $$\label{eq5}
e^{-v}\leq \frac {Df^l(x)}{Df^l(y)}\leq e^{v}.$$
Take any two $q_{n}$-close points $x,y\in S^{1}$ and $0\leq l\leq\ q_{n}-1$. Denote by $I$ the open interval with endpoints $x$ and $y$. Because the intervals $T_{f}^{i}(I),\ 0\leq i<q_{n}$ are disjoint, we obtain $$|log Df^{q_n}(x)-log Df^{q_n}(y)|\le \sum_{s=0}^{q_{n}-1}|log f(T_{f}^{s}x)-log f(T_{f}^{s}y)|\le v,$$ from which inequality (\[eq5\]) follows immediately.
Note that P-homeomorphisms $T_f$ satisfying the conditions of Lemma \[lem2.1\] are ergodic with respect to Lebesgue measure , i.e. every $T_f$-invariant set has measure zero or one.
Cross-ratio tools
=================
\[crossratio\] [The **cross-ratio** $Cr(z_1,z_2,z_3,z_4)$ of four points $z_i\in \mathbb{R}^{1}$, $i=1,2,3,4$, $z_1<z_2<z_3<z_4$ is defined as $$Cr(z_1,z_2,z_3,z_4)=\frac{(z_2-z_1)(z_4-z_3)}{(z_3-z_1)(z_4-z_2)}.$$]{}
\[distortion\] [The **cross-ratio distortion** $Dist(z_1,z_2,z_3,z_4;f)$ of four points $z_i\in \mathbb{R}^{1}$, $i=1,2,3,4$, $z_1<z_2<z_3<z_4$ with respect to a strictly increasing function $f$ on $ \mathbb{R}$ is defined as $$Dist(z_1,z_2,z_3,z_4;f)=\frac{Cr(f(z_1),f(z_2),f(z_3),f(z_4))}{Cr(z_1,z_2,z_3,z_4)}.$$]{}
Consider then a function $f:[a,b]\rightarrow R^{1}$, $[a,b]\subset S^{1}$ satisfying the following conditions:
1. $f \in C^{1}([a,b])$, $Df(x)\ge const >0$, $\forall x\in [a,b]$;
2. $D^{2}f\in L^{1}([a,b])$.
Fix an arbitrary $\varepsilon >0$. Since $D^{2}f\in L^{1}([a,b])$, it can be written in the form $$\label{eq6}
D^{2}f(x)=g_{\varepsilon}(x)+\theta_{\varepsilon}(x),\ x\in [a,b],$$ where $g_{\varepsilon}$ is a continuous function on $[a,b]$ and $\Vert\theta_{\varepsilon}\Vert_{L^{1}}<\varepsilon$.
\[theo3.1\] Suppose, the function $f=f(x)$ satisfies the above conditions ${i}),{ii})$. For $z_i\in [a,b]$, $i=1,2,3,4$, with $z_1<z_2<z_3<z_4$, the following estimate holds:
$$\begin{aligned}
\label{eq6a}
|Dist(z_1,z_2,z_3,z_4;f)-1|&\leq& C_1|z_4-z_1|\underset{x,t\in
[a,b]}{\max}|g_{\varepsilon}(x)-g_{\varepsilon}(t)|+\notag \\
&+&C_1\overset{z_1}{\underset{z_2}{\int}}|\theta_{\varepsilon}(y)|dy+
C_{1}\Big(\overset{z_2}{\underset{z_1}{\int}}|D^{2}f(y)|dy\Big)^{2}\end{aligned}$$
where the constant $C_{1}>0$ depends only on the function $f$.\
Take $z_i\in [a,b]\subset S^{1}$, $i=1,2,3,4$, with $z_1<z_2<z_3<z_4$. The following equalities are easy to check: $$\begin{aligned}
f(z_k)-f(z_1)=Df(z_1)(z_k-z_1)+\overset{z_k}{\underset{z_1}{\int}}D^{2}f(y)(z_k-y)dy,\
k=2,3;\\
f(z_4)-f(z_l)=Df(z_4)(z_4-z_l)-\overset{z_4}{\underset{z_l}{\int}}D^{2}f(y)(y-z_l)dy,\
l=2,3.\end{aligned}$$ Using these relations we obtain: $$\begin{aligned}
\label{eq8}
&&Cr(f(z_1),f(z_2),f(z_3),f(z_4))=
\frac{f(z_2)-f(z_1)}{f(z_3)-f(z_1)}\cdot \frac{f(z_4)-f(z_3)}{f(z_4)-f(z_2)}= \notag \\
&=& Cr(z_1,z_2,z_3,z_4)\cdot\frac{1+\frac{1}{Df(z_1)(z_2-z_1)}
\overset{z_2}{\underset{z_1}{\int}}D^{2}f(y)(z_2-y)dy}{1+\frac{1}{Df(z_1)(z_3-z_1)}
\overset{z_3}{\underset{z_1}{\int}}D^{2}f(y)(z_3-y)dy}\times \\
&\times& \frac{1-\frac{1}{Df(z_4)(z_4-z_3)}
\overset{z_4}{\underset{z_3}{\int}}D^{2}f(y)(y-z_3)dy}{1-\frac{1}{Df(z_4)(z_4-z_2)}
\overset{z_4}{\underset{z_2}{\int}}D^{2}f(y)(y-z_2)dy}.\notag \end{aligned}$$ Setting\
$
A(a,b)=\frac{1}{Df(a)(b-a)}
\overset{b}{\underset{a}{\int}}D^{2}f(y)(b-y)dy,
$and $B(a,b)=\frac{1}{Df(b)(b-a)}
\overset{b}{\underset{a}{\int}}D^{2}f(y)(y-a)dy
$\
we can hence rewrite $Dist(z_1,z_2,z_3,z_4;f)$ in the following form:
$$\begin{aligned}
Dist(z_1,z_2,z_3,z_4;f)&=&\frac{1+A(z_1,z_2)}{1+A(z_1,z_3)}\times\frac{1-B(z_3,z_4)}{1-B(z_2,z_4)}= \\
&=&(1+A(z_1,z_2))\cdot(1-A(z_1,z_3)+O(A^2(z_1,z_3))\cdot(1-B(z_3,z_4))\times \\
&\times&(1+B(z_2,z_4)+O(B^2(z_2,z_4))).\end{aligned}$$
Therefore $$\begin{aligned}
\label{eq9}
Dist(z_1,z_2,z_3,z_4;f)&=&1+A(z_1,z_2)-A(z_1,z_3)+B(z_2,z_4)-B(z_3,z_4)+\\
&+&O\left(\Big(\overset{z_3}{\underset{z_2}{\int}}|D^{2}f(y)|dy\Big)^2
\right)\notag.\end{aligned}$$ Set $M_{1}=0.5\Big(\overset{}{\underset{x\in
(z_{1},z_{4})}{\inf }} Df(x)\Big)^{-1}. $\
To continue the proof of Theorem \[theo3.1\] we need the following\
\[lemm3.1\] Assume, that the function $f$ satisfies the conditions of Theorem \[theo3.1\]. Then for any $a,b\in S^{1}$, $a<b$ the following identities hold: $$A(a,b)=\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(y)}dy+G_1(a,b),\,\,\,\,
B(a,b)=\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(y)}dy+G_2(a,b),$$ where $$\begin{aligned}
\label{eq10}
|G_i(a,b)|\leq M_1(b-a)\underset{x,t\in
[a,b]}{\max}|g_{\varepsilon}(x)-g_{\varepsilon}(t)|+ \notag\\
+ M_1
\overset{b}{\underset{a}{\int}}|\theta_{\varepsilon}(y)|dy+
2M_1^{2}\left(\overset{b}{\underset{a}{\int}}|D^{2}f(y)|dy\right)^{2},\
i=1,2.\end{aligned}$$
We prove only the identity for $A(a,b)$, the one for $B(a,b)$ is similar. Set $$G_1(a,b)=A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(y)}dy.$$\
It is clear that $$\begin{aligned}
|G_1(a,b)|&\le& \Big|A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(a)}dy\Big|+
\Big|\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(a)}dy-\overset{b}
{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(y)}dy\Big|=\\
&=&\Big|A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(A)}dy\Big|+
\frac{1}{2}\Big|\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{Df(y)Df(a)}dy
\overset{y}{\underset{a}{\int}}D^{2}f(t)dt\Big|.\end{aligned}$$ using this and the bound $(Df(x))^{-1} \leq 2\, M_1$ we get : $$\begin{aligned}
\label{eq11}
|G_1(a,b)| \leq \Big|A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(a)}dy\Big|+2M_{1}^{2}
\Big(\overset{b}{\underset{a}{\int}}|D^{2}f(y)|^{2}dy\Big).\end{aligned}$$
To get finally the estimate (\[eq10\]) for $G_1(a,b)$ it is sufficient to estimate the difference $A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(a)}dy$. Using the definition of $A(a,b)$ and the decomposition (\[eq6\]) we obtain: $$\begin{aligned}
&&\Big|A(a,b)-\overset{b}{\underset{a}{\int}}\frac{D^{2}f(y)}{2Df(a)}dy\Big|
=\Big|\frac{1}{Df(a)}\overset{b}{\underset{a}{\int}}
D^{2}f(y)\left(\frac{b-y}{b-a}-\frac{1}{2}\right)dy\Big|=\\
&=&\frac{1}{Df(a)}\Big|\overset{b}{\underset{a}{\int}}
\Big(g_{\varepsilon}(y)+\theta_{\varepsilon}(y))
\left(\frac{b-y}{b-a}-\frac{1}{2}\right)dy\Big|
\leq\frac{1}{Df(a)}\Big|g_{\varepsilon}(a)\overset{b}{\underset{a}{\int}}
\left(\frac{b-y}{b-a}-\frac{1}{2}\right)dy\Big|+\\
&+&\frac{1}{Df(a)}\Big|\overset{b}{\underset{a}{\int}}|g_{\varepsilon}(y)-g_{\varepsilon}(a)|
\Big|\frac{b-y}{b-a}-\frac{1}{2}\Big|dy\Big|
+\frac{1}{Df(a)}\overset{b}{\underset{a}{\int}}
|\theta_{\varepsilon}(y)|\Big|\frac{b-y}{b-a}-\frac{1}{2}\Big|dy\leq\\
&\leq& M_1(b-a)\underset{x,t\in
[a,b]}{\max}|g_{\varepsilon}(x)-g_{\varepsilon}(t)|+M_1
\overset{b}{\underset{a}{\int}}|\theta_{\varepsilon}(y)|dy.\end{aligned}$$ Combining this with the estimate (\[eq11\]) we obtain the estimate (\[eq10\]) for $G_1(a,b)$ in the Lemma..
We can now finish the proof of Theorem \[theo3.1\]. Combining (\[eq9\]) with the representations of $A(a,b)$ and $B(a,b)$ in Lemma \[lemm3.1\] we obtain:\
$Dist(z_1,z_2,z_3,z_4;f)=1+\overset{z_2}{\underset{z_1}{\int}}
\frac{D^{2}f(y)}{2Df(y)}dy+G_1(z_1,z_2)-\overset{z_3}{\underset{z_1}{\int}}
\frac{D^{2}f(y)}{2Df(y)}dy-G_1(z_1,z_3)$\
$+\overset{z_4}{\underset{z_2}{\int}}
\frac{D^{2}f(y)}{2Df(y)}dy+G_2(z_3,z_4)
-\overset{z_4}{\underset{z_3}{\int}}
\frac{D^{2}f(y)}{2Df(y)}dy-G_2(z_3,z_4)
+O\left(\left(\overset{z_4}{\underset{z_1}{\int}}
|D^{2}f(y)|dy\right)^2\right)$\
$=1+G_1(z_1,z_2)-G_1(z_1,z_3)+G_2(z_2,z_4)-G_2(z_3,z_4)+
+O\left(\left(\overset{z_4}{\underset{z_1}{\int}}
|D^{2}f(y)|dy\right)^2\right).$\
Applying next (\[eq10\]) for the intervals $[z_{s},z_{s+1}]\in [z_{1},z_{4}],$$ s=1,2,3 $ we obtain $$|G_1(z_{s},z_{s+1})|\leq \frac{M_1}{2}|z_4-z_1|\underset{[z_1,z_4]}{\max}
|g_{\varepsilon}(x)-g_{\varepsilon}(t)|
+\frac{M_1}{2}\overset{z_4}{\underset{z_1}{\int}}|\theta_{\varepsilon}(y)|dy+
\frac{M_1^2}{2}\overset{z_4}{\underset{z_1}{\int}}|f''(y)|dy.$$ from which the assertion of Theorem \[theo3.1\] follows immedately.
Next we consider the case when the interval $[z_1,z_4]$ contains just one break point $x=x_b$. We estimate the distortion of the cross ratio when the break point lies outside the middle interval $[z_2,z_3]$ i.e. $x_{b}\in [z_1,z_2]\cup[z_3,z_4] $.
For $z_i\in S^1$, $i=1,2,3,4$ with $z_1<z_2<z_3<z_4$ and $x_b\in [z_1,z_2]$ we set $$\label{eq12}
\alpha:=z_{2}-z_{1},\,\, \beta:=z_{3}-z_{2}, \,\,\gamma:=z_{4}-z_{3},\,\, \tau:=z_{2}-x_{b}, \,\,\xi:=\frac{\beta}{\alpha},\,\,z:=\frac{\tau}{\alpha}.$$
\[lemm3.2\] Assume, the function $f$ is defined on $[z_{1},z_{4})]$, its derivative $Df$ is continuous on every connected interval of the set $[z_1,z_4]\backslash\{x_{b}\}$ and $D^{2}f\in L^{1}[z_{1},z_{4})]$. Choose $z_i\in S^1$, $i=1,2,3,4$, such that $z_1<z_2<z_3<z_4$ and $x_b\in[z_1,z_2]$. Then $$\begin{aligned}
\label{eq13} &|Dist(z_1,z_2,z_3,z_4;f)-\frac{[\sigma(x_{b})+(1-\sigma(x_{b}))z](1+\xi)}{\sigma(x_{b})+(1-\sigma(x_{b}))z+\xi}|\leq
K_1 \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,\end{aligned}$$ where the constant $K_1>0$ depends only on the function $f$.
. By assumption $x_b\in[z_1,z_2]$. Let the jump ratio of $Df(x)$ at the point $x_{b}$ be $\sigma(x_{b})=\frac{Df_{-}(x_{b})}{Df_{+}(x_{b})}$. Rewrite then $Dist(z_1,z_2,z_3,z_4;f)$ in the form: $$\begin{aligned}
Dist(z_1,z_2,z_3,z_4;f)&=&\frac{Cr(f(z_1),f(z_2),f(z_3),f(z_4))}{Cr(z_1,z_2,z_3,z_4)}\\
&=&\Big(\frac{f(z_2)-f(z_1)}{z_2-z_1}:\frac{f(z_3)-f(z_1)}{z_3-z_1}\Big)\Big(\frac{f(z_4)-f(z_3)}{z_4-z_3}:\frac{f(z_4)-f(z_2)}{z_4-z_2}\Big).\end{aligned}$$ It is easy to check that $$\begin{aligned}
&&f(z_{2})-f(z_{1})=(f(x_{b})-f(z_{1}))+(f(z_{2})-f(x_{b}))=\\
&=&\left\lbrace Df_{-}(x_{b})(x_{b}-z_{1})-
\overset{x_{b}}{\underset{z_{1}}{\int}}
D^{2}f(y)(y-z_1)dy\right\rbrace + \left\lbrace Df_{+}(x_{b})(z_{2}-x_{b})+
\overset{z_{2}}{\underset{x_{b}}{\int}}
D^{2}f(y)(z_{2}-y)dy\right\rbrace =\\
&=&Df_{+}(x_{b})(z_{2}-z_{1})\left[ \sigma(x_{b})+(1-\sigma(x_{b}))\frac{\tau}{\alpha})\right]+\\
&+& Df_{+}(x_{b})\alpha\left \lbrace\frac{1}{Df_{+}(x_{b})} \overset{z_{2}}{\underset{x_{b}}{\int}}
D^{2}f(y)\frac{z_{2}-y}{z_{2}-z_{1}}dy-\frac{1}{Df_{+}(x_{b})} \overset{x_{b}}{\underset{z_{1}}{\int}}
D^{2}f(y)\frac{y-z_{1}}{z_{2}-z_{1}}dy\right\rbrace.\\\end{aligned}$$ Hence $$\label{eq14}
f(z_{2})-f(z_{1})=Df_{+}(x_{b})(z_{2}-z_{1}) \left[ \sigma(x_{b})+(1-\sigma(x_{b}))\frac{\tau}{\alpha}+r_{1}(x_{b},z_{1},z_{2})\right];$$ in analogy we find $$\begin{aligned}
&&f(z_{3})-f(z_{1})=(f(x_{b})-f(z_{1}))+(f(z_{3})-f(x_{b}))=\\
&=&\left\lbrace Df_{-}(x_{b})(x_{b}-z_{1})-
\overset{x_{b}}{\underset{z_{1}}{\int}}
D^{2}f(y)(y-z_1)dy\right\rbrace + \left\lbrace Df_{+}(x_{b})(z_{3}-x_{b})+
\overset{z_{3}}{\underset{x_{b}}{\int}}
D^{2}f(y)(z_{3}-y)dy\right\rbrace=\\
&=&Df_{+}(x_{b}) (z_{3}-z_{1})\left[\frac{z_{3}-x_{b}}{z_{3}-z_{1}} +\sigma(x_{b})\frac{z_{1}-x_{b}}{z_{3}-z_{1}})\right]+\\ &+&Df_{+}(x_{b})(z_{3}-z_{1})\left \lbrace\frac{1}{Df_{+}(x_{b})} \overset{z_{3}}{\underset{x_{b}}{\int}}
D^{2}f(y)\frac{z_{3}-y}{z_{2}-z_{1}}dy-\frac{1}{Df_{+}(x_{b})} \overset{x_{b}}{\underset{z_{1}}{\int}}
D^{2}f(y)\frac{y-z_{1}}{z_{2}-z_{1}}dy\right\rbrace.\end{aligned}$$ respectively $$\label{eq15}
f(z_{3})-f(z_{1})=
Df_{+}(x_{b})(z_{3}-z_{1})
\left[\frac{\tau+\beta}{\alpha+\beta} +\sigma(x_b)\frac{\alpha-\tau}{\alpha+\beta}+r_{2}(x_{b},z_{1},z_{3})\right].$$ For $|r_{1}(x_{b},z_{1},z_{2})|$ and $|r_{2}(x_{b},z_{1},z_{3})|$ then the following estimates hold: $$\label{eq16}
|r_{1}(x_{b},z_{1},z_{2})|,|r_{2}(x_{b},z_{1},z_{3})|\le \frac{1}{Df_{+}(x_{b})} \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy$$ Using this, (\[eq14\]) and (\[eq15\]) we get:\
$$\begin{aligned}
\label{eq17}
\Big|\frac{f(z_2)-f(z_1)}{z_2-z_1}:\frac{f(z_3)-f(z_1)}{z_3-z_1}-
\frac{[\sigma(f)+(1-\sigma(f))z](1+\xi)}{\sigma(f)+\xi+(1-\sigma(f))z}\Big|\nonumber\\
\leq
K_{2}\overset{z_3}{\underset{z_1}{\int}} |D^{2}f(y)|dy,\end{aligned}$$ with $\xi$ and $z$ as defined in (\[eq12\]) and where the constant $K_{2}>0$ is depending only on the function $f$.
Since the interval $[z_2,z_4]$ does not contain the break point $x_{b}$, it can easily be shown that $$\begin{aligned}
\Big|\frac{f(z_4)-f(z_3)}{z_4-z_3}:\frac{f(z_4)-f(z_2)}{z_4-z_2}-1\Big|\leq
K_{3}\overset{z_4}{\underset{z_2}{\int}} |D^{2}f(y)|dy,\end{aligned}$$ where also the constant $K_{3}>0$ depends only on $f$. The last inequality together with the bounds (\[eq16\]) and (\[eq17\]) imply the assertion of Lemma \[lemm3.2\].
If the break point $x=x_{b}$ belongs to the right interval $[z_{3},z_{4}]$, then one can prove the following estimate: $$|Dist(z_{1},z_{2},z_{3},z_{4};f)-\frac{[\sigma(x_b)+(1-\sigma(x_b))\vartheta](1+\eta)}{\sigma(x_b)+(1-\sigma(x_b))\vartheta+\eta}|\leq
K_4 \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,$$ where $\eta=\frac{z_{3}-z_{2}}{z_{4}-z_{3}}$, $\vartheta=\frac{x_{b}-z_{3}}{z_{4}-z_{3}}$ and the constant $K_4>0$ depends only on the function $f$.
The proofs of Theorem \[DLM\] and Theorem \[DLM1\]
==================================================
For the proofs of Theorem \[DLM\] and Theorem \[DLM1\] we need several Lemmas which we formulate next and whose proofs will be given later.
\[lemm4.1\] Assume that the lift $\varphi$ of the conjugating homeomorphism $T_{\varphi}(x)$ has a positive derivative $D \varphi(x_{0})=\omega$ at the point $x=x_{0}\in S^{1}$, and the following conditions hold for $z_i\in S^1,\, i=1,..,4$ with $z_1<z_2<z_3<z_4$ and some constant $R_{1}>1:$
- $R_{1}^{-1}|z_{3}-z_{2}| \leq|z_{2}-z_{1}|\leq
R_{1}|z_{3}-z_{2}|$, $R_{1}^{-1}|z_{3}-z_{2}| \leq|z_{4}-z_{3}|\leq
R_{1}|z_{3}-z_{2}|;$
- $\underset{1\leq i\leq 4}{max}|z_{i}-x_{0}|\leq
R_{1}|z_{2}-z_{1}|.$
Then for any $\varepsilon>0$ there exists $\delta=\delta(\varepsilon)>0$ such that $$\begin{aligned}
\label{eq18}
|Dist(z_1,z_2,z_3,z_4;T_{\varphi})-1|\leq C_{2}\varepsilon,\end{aligned}$$ if $z_{i}\in (x_{0}-\delta,\ x_{0}+\delta)$ for all $i=1,2,3,4$, where the constant $C_{2}>0$ depends only on $R_{1},$ $\omega$ and not on $\varepsilon.$
Suppose that $D\varphi(x_{0})=\omega$ for some point $x=x_{0},\ x_{0}\in S^{1}$. Consider its $n-$th dynamical partition $$\xi_{n}(x_{0})=\left\lbrace \Delta_{i}^{(n-1)}(x_{0}), \ 0\leq
i<q_{n}; \ \Delta_{j}^{(n)}(x_{0}), \ 0\leq j<q_{n-1}\right\rbrace.$$ For definitness suppose, that $n$ is odd. Then $\Delta_{0}^{(n)}(x_{0})=[T_{f}^{q_{n}}x_{0},x_{0}]$ and $\Delta_{0}^{(n-1)}(x_{0})=[x_{0},T_{f}^{q_{n-1}}x_{0}]$ are its two generators. Denote by $\overline{a}_{b}$ and $\overline{c}_{b}$ the preimages of $a_{b}$ and $c_{b}$ in the interval $\left[ T_{f}^{q_{n}}x_{0}, T_{f}^{q_{n-1}}x_{0}\right] $ such that $\overline{a}_{b}=T_{f}^{-l}a_{b}$ and $\overline{c}_{b}=T_{f}^{-p}c_{b}$ for some $l,p \in [0,q_{n}) $.
Define next for $m \in [0,q_{n}]$ $$\begin{aligned}
\label{eq19}
\xi(m):=\frac{T_{f}^{m}z_{3}-T_{f}^{m}z_{2}}{T_{f}^{m}z_{2}-T_{f}^{m}z_{1}},\,\,\, \,\,\,
z(m):=\frac{T_{f}^{m}z_{2}-T_{f}^{m}\overline{c}_{b}}{T_{f}^{m}z_{2}-T_{f}^{m}z_{1}},\nonumber\\
\eta(m):=\frac{T_{f}^{m}z_{3}-T_{f}^{m}z_{2}}{T_{f}^{m}z_{4}-T_{f}^{m}z_{3}},\,\,\,\,\,\,
\vartheta(m):=\frac{T_{f}^{m}\overline{c}_{b}-T_{f}^{m}z_{3}}{T_{f}^{m}z_{4}-T_{f}^{m}z_{3}}\end{aligned}$$
The numbers $z(m)$ (if $\overline{c}_{b}\in \left[z_{1}, z_{2} \right]$) and $\vartheta(m)$ (if $\overline{c}_{b}\in \left[z_{3}, z_{4} \right] $) are called normalized coordinates of the point $T_{f}\overline{c}_{b}$. It is clear that the normalized coordinates $z(m)$ (respectively $\vartheta(m)$) change from 0 to 1, when the break point $c_{b}$ is moving from $T_{f}^{p}z_{2}$ to $T_{f}^{p}z_{1}$ (respectively from $T_{f}^{p}z_{3}$ to $T_{f}^{p}z_{4}$).
\[defi4.1\] The intervals $ \left\lbrace T_{f}^{j}[z_{1},z_{2}], T_{f}^{j}[z_{2},z_{3}], T_{f}^{j}[z_{3},z_{4}]: 0\le j\le q_{n}-1\right\rbrace $ cover the break points $a_{b}$, $c_{b}$ regularly with constants $C\geq 1$, $\zeta\in [0,1]$, if
- the intervals $ \left\lbrace T_{f}^{j}[z_{1},z_{4}], 0\le j\le q_{n}-1\right\rbrace$ cover every break point only once;
- either $z_{2}=T_{f}^{-l}a_{b}$ and $T_{f}^{-p}c_{b}\in \left[z_{1}, z_{2} \right] $ or $z_{3}=T_{f}^{-l}a_{b}$ and $T_{f}^{-p}c_{b}\in \left[z_{3}, z_{4} \right] $ for some $l, p\in \left[ 0, q_{n}\right) $;
- $\xi(0)\geq C$ and $z(0)\in [0, \zeta]$ if $\overline{c}_{b}=T_{f}^{-l}c_{b}\in \left[z_{1}, z_{2} \right], $\
$\eta(0) \geq C$ and $\vartheta(0)\in [0, \zeta]$ if $\overline{c}_{b}=T_{f}^{-p}c_{b}\in \left[z_{3}, z_{4} \right], $
or if conditions 1)-3) hold for $a_{b}$ and $c_{b}$ interchanged.\
In order to formulate the next Lemma we introduce the following functions for $x>0$ and $0\leq t\leq 1$: $$\begin{aligned}
\label{eq20}
G(x)=\frac{\sigma(a_{b})(1+x)}{\sigma (a_{b})+x},\,\,
F(x,t)=\frac{[\sigma(c_{b})+(1-\sigma(c_{b}))t](1+x)}{\sigma(c_{b})+(1-\sigma(c_{b}))t+x},\end{aligned}$$
\[lemm4.2\] Suppose that the homeomorphism $T_{f}$ satisfies the conditions of Theorem \[DLM\]. If the intervals $ \left\lbrace T_{f}^{j}[z_{1},z_{2}], T_{f}^{j}[z_{2},z_{3}], T_{f}^{j}[z_{3},z_{4}]: 0\le j\le q_{n}-1\right\rbrace $ cover the break points $a_{b}$, $c_{b}$ such that either $z_2=T_f^{-l}a_b$ and $T_f^{-p}c_b\in[z_1,z_2]$ or $z_3=T_f^{-l}a_b$ and $T_f^{-p}c_b\in[z_3,z_4]$ then
- $ Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})
=[G(\xi(l))+\chi_{1}][F(\xi(p),z(p))+\chi_{2}]\times $\
$\times\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f})
$ if $z_{2}=\overline{a}_{b},$ $ \overline{c}_{b}\in \left[z_{1}, z_{2} \right]; $
- $ Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})
=[G(\eta(l))+\chi_{3}][F(\eta(p),\vartheta(p))+\chi_{4}]\times$\
$\times\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f})
$ if $z_{3}=\overline{a}_{b},$ $ \overline{c}_{b}\in \left[z_{3}, z_{4} \right] $
where $$\label{eq21}
|\chi_{j}|=|\chi_{j}(z_{1},z_{2},z_{3},z_{4})|\leq
K_1 \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,\,\, 1\leq j \leq4.$$ and the constant $R_{5}$ does not depend on $n$ and $\varepsilon$
Using Lemma \[lemm2.2\] we get the following inequalities for all $1\leq m\leq q_n$: $$\begin{aligned}
\label{eq22}
e^{-v}\xi(0)\leq \xi(m)\leq e^{v} \xi(0),\,\,\,\,
e^{-v}z(0)\leq z(m)\leq e^{v}z(0),\nonumber\\
e^{-v}\eta(0)\leq \eta(m)\leq e^{v} \eta(0),\,\,\,\,
e^{-v}\vartheta(0)\leq \vartheta(m)\leq e^{v}\vartheta(0)\end{aligned}$$ From this it follows, that the normalized coordinates $\xi(m)$, $\eta(m)$, $z(m)$, and $\vartheta(m)$ are uniformly (with respect to $x_{0}$ and $m$) comparable with the initial normalized coordinates $\xi(0)$, $\eta(0)$, $z(0)$, and $\vartheta(0)$ respectively.
\[lemm4.3\]If a circle homeomorphism $T_{f}$ satisfies the conditions of Lemma \[lemm2.2\] then there exist for any $x_{0}\in S^{1}$ and any $\delta>0$ constants $C_{0}=C_{0}(f,\sigma(a_{b}),\sigma(c_{b}))>1$ and $\zeta_{0}=\zeta_{0}(f,\sigma(a_{b}),\sigma(c_{b}))\in (0,1)$, such that for all triple of intervals $[z_{s},z_{s+1}]\subset(x_{0}-\delta, \ x_{0}+\delta),$ $s=1,2,3,$ covering the break points $a_{b}$,$c_{b}$ regularly with constants $C_{0}$ and $\zeta_{0}$ the following relations hold: $$\begin{aligned}
|G(\xi(l))F(\xi(p),z(p))-1|\geq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4} \,\,
if \,\, z_{2}=\overline{a}_{b},\,0\le \frac{z_{2}-\overline{c}_{b}}{z_{2}-z_{1}}\le \zeta_{0}\end{aligned}$$ respectively $$\begin{aligned}
|G(\eta(l))F(\eta(p),\vartheta(p))-1|\geq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4}
\,\, if \,\, z_{3}=\overline{a}_{b}, \, 0\le \frac{\overline{c}_{b}-z_{3}}{z_{4}-z_{3}}\le \zeta_{0}.\end{aligned}$$
\[lemm4.4\] If the circle homeomorphism $T_{f}$ satisfies the conditions of Lemma \[lemm2.2\] there exists for any $x_{0}\in S^{1}$ and any $\delta>0$ a number $N=N(\delta, x_{0})>1,$ such that for all $n>N(\delta, x_{0})$, there is a triple of intervals $[z_{s},z_{s+1}]\subset(x_{0}-\delta, \ x_{0}+\delta),$ $s=1,2,3$ with the following properties:
- the interval $[z_{1},z_{4}]$ is $q_{n}$ small;
- the intervals $[z_{s},z_{s+1}]$ and $ [T_{f}^{q_{n}}z_{s}, T_{f}^{q_{n}}z_{s+1}]$ $s=1,2,3$ satisfy conditions a) and b) of Lemma 4.1 with some constant $R_{1}>1$ depending on $C_{0}$, $\zeta_{0}$ and $v$;
- the intervals $\left\lbrace T_{f}^{i}[z_{1},z_{2}],\,\,T_{f}^{i}[z_{2},z_{3}]\,\, T_{f}^{i}[z_{3},z_{4}],\,\,\,0\leq i \leq q_{n}-1 \right\rbrace $ either cover both break points $a_{b}$, $c_{b}$ regularly with constants $C_{0}$ and $\zeta_{0}$, or cover only the break point $a_{b}$ such that its preimage $\overline{a}_{b}$ coincides with $z_{2}$ or $z_{3};$
\[lemm4.5\] Suppose, that the circle homeomorphism $T_{f}$ satisfies the conditions of Theorem \[DLM\] and the intervals $[z_{s},z_{s+1}],\,\, s=1,2,3$ satisfy conditions 1)-3) of Lemma \[lemm4.4\]. Then the following inequality holds for sufficiently large $n$: $$|Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})-1|>const >0$$ where the constant depends only on the function $f$.
After these preparations we can now proceed to the proof of Theorem \[DLM\]. Let $T_{f}$ be a class $P$-homeomorphism satisfying the conditions of Theorem \[DLM\]. Since its rotation number $\rho_{f}$ is irrational the $T_{f}$-invariant measure $\mu_{f}$ is nonatomic i.e. every one point subset of the circle has zero $\mu_{f}$-measure. The conjugating map $T_{\varphi}$ related to $\mu_{f}$ by $T_{\varphi}x=\mu_{f}([0,x])$ , $x\in S^{1}$, is a continuous and monotone increasing function on $S^{1}$. Hence $T_{\varphi}$ has a finite derivative almost everywhere (w.r.t. Lebesgue measure) on the circle. We show that $D\varphi(x)=0$ at all points at which the derivative is defined. Choose an $\varepsilon>0$ and a triple of intervals $[z_{s},z_{s+1}]\subset(x_{0}-\delta, \ x_{0}+\delta),$ $s=1,2,3,$ satisfying the conditions of Lemma \[lemm4.1\]. It follows from this Lemma and Lemma \[lemm4.4\] that $$\begin{aligned}
\label{eq23}
|Dist(z_1,z_2,z_3,z_4;T_{\varphi})-1|\leq C_{2}\varepsilon,\end{aligned}$$ and $$\begin{aligned}
\label{eq24}
|Dist(T_{f}^{q_{n}}z_{1},T_{f}^{q_{n}}z_{2}, T_{f}^{q_{n}}z_{3},T_{f}^{q_{n}}z_{4}; T_{\varphi})-1|\leq C_{2}\varepsilon.\end{aligned}$$ By definition $$\begin{aligned}
\label{eq25}
&&Dist(T_{f}^{q_{n}}z_{1},T_{f}^{q_{n}}z_{2}, T_{f}^{q_{n}}z_{3},T_{f}^{q_{n}}z_{4}; T_{\varphi})=\nonumber\\
&=&\frac{Cr(T_{\varphi}(T_{f}^{q_{n}}z_{1}), T_{\varphi}(T_{f}^{q_{n}}z_{2}), T_{\varphi}(T_{f}^{q_{n}}z_{3}), T_{\varphi}(T_{f}^{q_{n}}z_{4}) )}{Cr(T_{f}^{q_{n}}z_{1},T_{f}^{q_{n}}z_{2}, T_{f}^{q_{n}}z_{3},T_{f}^{q_{n}}z_{4})}.\end{aligned}$$\
Since $T_{\varphi}$ conjugates $T_{f}$ with the linear rotation $T_{\rho}$, we can readily see that $$Cr(T_{\varphi}(T_{f}^{q_{n}}z_{1}),(T_{\varphi}(T_{f}^{q_{n}}z_{2}), (T_{\varphi}(T_{f}^{q_{n}}z_{3}),(T_{\varphi}(T_{f}^{q_{n}}z_{4}) ))=Cr(T_{\varphi}z_{1},T_{\varphi}z_{2}, T_{\varphi}z_{3},T_{\varphi}z_4) ))$$ and hence $$Dist(T_{f}^{q_{n}}z_{1},T_{f}^{q_{n}}z_{2}, T_{f}^{q_{n}}z_{3},T_{f}^{q_{n}}z_{4}; T_{\varphi})
=\frac{Cr(T_{\varphi}z_{1},T_{\varphi}z_{2}, T_{\varphi}z_{3},T_{\varphi}z_4) ))}{Cr(T_{f}^{q_{n}}z_{1},T_{f}^{q_{n}}z_{2}, T_{f}^{q_{n}}z_{3},T_{f}^{q_{n}}z_{4})}$$ This together with , and implies $$\begin{aligned}
\label{eq26}
|Dist(z_1,z_2,z_3,z_4;T_{f}^{q_{n}})-1|\leq C_{3}\varepsilon,\end{aligned}$$\
where the constant $C_{3}>0$ does not depend on $\varepsilon$ and $n$. But this contradicts Lemma \[lemm4.5\] according to which $$| Dist(z_1,z_2,z_3,z_4;T_f^{q_n}-1|>const >0$$ for sufficiently large $n$. This contradiction proves Theorem \[DLM\].
The proofs of Lemmas \[lemm4.1\]-\[lemm4.5\]
============================================
**Proof of Lemma \[lemm4.1\]** Suppose, that the derivative $D\varphi(x_{0})$ exists and $D\varphi(x_{0})=\omega >0$. By the definition of the derivative there exists for any $\varepsilon>0$ a number $\delta=\delta(x_{0} a \ \varepsilon)>0$, such that for all $x\in(x_{0}-\delta, \ x_{0}+\delta).$ $$\begin{aligned}
\label{eq27}
\omega-\varepsilon<\frac{\varphi(x)-\varphi(x_{0})}{x-x_{0}}<\omega+\varepsilon.\end{aligned}$$ Now take four points $z_{i}\in(x_{0}-\delta,x_{0}+\delta)$ satisfying conditions (a) and (b) of Lemma \[lemm4.1\]. Assume that $z_{i}<x_{0}, \,1\leq i\leq 4$. For the other cases Lemma \[lemm4.1\] can be proved similarly. Relation (\[eq27\]) implies for $x=z_{i},$ $ i=1,2,3,4$ $$(\omega-\varepsilon)(x_{0}-z_{i})<\varphi(x_{0})-\varphi(z_{i})<
(\omega +\varepsilon)(x_{0}-z_{i}).$$
This yields the following inequalities:
$$\begin{aligned}
\label{eq28}
\omega-\varepsilon\frac{(x_{0}-z_{s+1})+(x_{0}-z_{s})}{z_{s+1}-z_{s}}&\leq&
\frac{\varphi(z_{s+1})-\varphi(z_{s})}{z_{s+1}-z_{s}}\nonumber\\
&\leq&\omega+\varepsilon
\frac{(x_{0}-z_{s+1})+(x_{0}-z_{s})}{z_{s+1}-z_{s}}\end{aligned}$$
for $s=1,2,3$, and $$\begin{aligned}
\label{eq29}
\omega -\varepsilon\frac{(x_{0}-z_{s+2})+(x_{0}-z_{s})}{z_{s+2}-z_{s}}&\leq&
\frac{\varphi(z_{s+2})-\varphi(z_{s})}{z_{s+2}-z_{s}}\nonumber\\
&\leq&\omega+\varepsilon\frac{(x_{0}-z_{s+2})+(x_{0}-z_{s})}{z_{s+2}-z_{s}}\end{aligned}$$ for $s=1,2.$\
From conditions (a) and (b) of Lemma \[lemm4.1\] on the other hand it follows that $$\begin{aligned}
\label{eq30}
\underset{1\leq i\leq 4}{max}\Big\{\frac{x-z_{i}}{z_{2}-z_{1}},
\frac{x_{0}-z_{i}}{z_{3}-z_{1}}, \frac{x_{0}-z_{i}}{z_{4}-z_{2}}, \frac{x_{0}-z_{i}}{z_{4}-z_{3}}
\Big\}\leq K_{1}\end{aligned}$$ where the constant $K_{1}>0$ depends on $R_{1}$ and does not depend on $\varepsilon$.\
We rewrite $Dist(z_1,z_2,z_3,z_4;T_{\varphi})$ in the following form: $$Dist(z_1,z_2,z_3,z_4;T_{\varphi})
=\frac{T_{\varphi}z_{2}-T_{\varphi}z_{1}}{z_{2}-z_{1}}\cdot
\frac{T_{\varphi}z_{4}-T_{\varphi}z_{3}}{z_{4}-z_{3}}\cdot
\frac{z_{3}-z_{1}}{T_{\varphi}z_{3}-T_{\varphi}z_{1}}\cdot
\frac{z_{4}-z_{2}}{T_{\varphi}z_{4}-T_{\varphi}z_{2}}.$$ The inequalities (\[eq28\])-(\[eq30\]) then imply the assertion of Lemma \[lemm4.1\].\
**Proof of Lemma \[lemm4.2\]**. We consider the case $z_{2}=T_{f}^{-l}a_{b}$, $ T_{f}^{-p}c_{b}\in \left[z_{1}, z_{2} \right]$, $0 \leq l,p\leq q_n$, the case $z_{3}=T_{f}^{-l}a_{b},$ $ T_{f}^{-p}c_{b}\in \left[z_{3}, z_{4} \right],$ $0 \leq l,p\leq q_n$, can be treated similarly. Rewrite the distortion $Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})$ in the following form $$\begin{aligned}
\label{eq31}
&&Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})
=Dist (T_{f}^{l}z_{1},T_{f}^{l}z_{2}, T_{f}^{l}z_{3},T_{f}^{l}z_{4};T_{f})\times \notag\\
&\times& Dist(T_{f}^{p}z_{1},T_{f}^{p}z_{2}, T_{f}^{p}z_{3},T_{f}^{p}z_{4};T_{f})
\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f}).\end{aligned}$$ By assumption only the two intervals $[T_{f}^{l}z_{1},T_{f}^{l}z_{2}]$ and $[T_{f}^{p}z_{1},T_{f}^{p}z_{2}]$ contain the break points: namely $a_{b}=T_{f}^{l}z_{2}=a_{b}$, and $ c_{b}\in [T_{f}^{p}z_{1}, T_{f}^{p}z_{2} ]$ for some $l,p\in [0,q_{n}).$
Using Lemma \[lemm3.2\] and the definitions of the functions $G(x)$,$F(x,t)$ we get $$\begin{aligned}
Dist (T_{f}^{l}z_{1},T_{f}^{l}z_{2}, T_{f}^{l}z_{3},T_{f}^{l}z_{4};T_{f})&=&\frac{\sigma(a_{b})(1+\xi(l))}{\sigma (a_{b})+\xi(l)}+\chi_{1}=G(\xi(l)+\chi_{1},\\
Dist (T_{f}^{p}z_{1},T_{f}^{p}z_{2}, T_{f}^{p}z_{3},T_{f}^{p}z_{4};T_{f})&=&\frac{[\sigma(c_{b})+(1-\sigma(c_{b}))z(p)](1+\xi(p))}{\sigma(c_{b})+(1-\sigma(c_{b}))z(p)+\xi(p)}+\chi_{2}=\\
&=&F(\xi(p),z(p))+\chi_{2},\end{aligned}$$ with $|\chi_{j}|=|\chi_{j}(z_{1},z_{2},z_{3},z_{4})|\leq
K_1 \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,$ $j=1,2.$\
This together with (\[eq31\]) imply the assertion of Lemma \[lemm4.2\].
**Proof of Lemma \[lemm4.3\]**. We prove only the bound for $G(\xi(l)F(\xi(p),z(p))$. The one for $G(\eta(l)F(\eta(p),\vartheta(p))$ can be proved similarly. We start rewriting $G(\xi(l)F(\xi(p),z(p))$ in the following form: $$\begin{aligned}
\label{eq32}
G(\xi(l)F(\xi(p),z(p))=\frac{\sigma(a_{b})(1+\xi(l))}{\sigma (a_{b})+\xi(l)}\cdot
\frac{[\sigma(c_{b})+(1-\sigma(c_{b}))z(p)](1+\xi(p)]}{\sigma(c_{b})+(1-\sigma(c_{b}))z(p)+\xi(p)}\nonumber =\\
=[\sigma(a_{b})\sigma(c_{b})+(1-\sigma(c_{b}))\sigma(a_{b})z(p)]
\times \big [\frac{(1+\xi(l))}{\sigma (a_{b})+\xi(l)}\nonumber\\ \cdot\frac{(1+\xi(p))}{\sigma(c_{b})+(1-\sigma(c_{b}))z(p)+\xi(p)}\big ] \equiv \Phi_{1}(z(p))\times \Phi_{2}(\xi(l),\xi(p),z(p)) \end{aligned}$$ where $z(p)\in [0,1])$ and $\xi(l),\, \xi(p)>0.$
It is clear, that $\Phi_{1}(0)=\sigma(a_{b})\sigma(c_{b})$ and $ \Phi_{2}(\xi(l),\xi(p),z(p))$ tends to 1 as $\xi(l), \xi(p)$ tend to $\infty.$ Recall that $\sigma{a_{b}}\sigma{c_{b}}\neq 1$ by assumption.\
Next we discuss the conditions under which the expression $\Phi_{1}(z(p))\Phi_{2}(\xi(l),\xi(p),z(p))$ stays away from 1. Obviously $$\begin{aligned}
\label{eq57}
|\Phi_{1}\Phi_{2}-1|=|(\Phi_{1}-1)+\Phi_{1}(\Phi_{2}-1)|\geq||\Phi_{1}-1|-\Phi_{1}|\Phi_{2}-1|.\end{aligned}$$ Using the bounds for $z(m)$ in (\[eq22\]) we get $$\begin{aligned}
|\Phi_{1}-1|&=&|\sigma(a_{b})\sigma(c_{b})+(1-\sigma(c_{b}))\sigma(a_{b})z(p)-1|\\
&\geq& |\sigma(a_{b})\sigma(c_{b})-1|-|1-\sigma(c_{b})|\sigma(a_{b})z(p)\\
&\geq& |\sigma(a_{b})\sigma(c_{b})-1|-
|1-\sigma(c_{b})|\sigma(a_{b})e^{v}z(0).\end{aligned}$$ If next $z(0)$ fulfills the inequality $$|\sigma(a_{b})\sigma(c_{b})-1|-
|1-\sigma(c_{b})|\sigma(a_{b})e^{v}z(0)\geq
\frac{|\sigma(a_{b})\sigma(c_{b})-1|}{2},$$ and hence $$z(0)\leq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{2e^{v}|\sigma(a_{b})\sigma(c_{b})-\sigma(a_{b})|},$$ then we conclude that $$\begin{aligned}
\label{eq34}
|\Phi_{1}-1| \geq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{2},\,\,\,\, if\,\,\,0\le z(0)\leq
\zeta_{0},\end{aligned}$$ where $$\label{35}
\zeta_{0}:=min \left\lbrace \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{2e^{v}|\sigma(a_{b})\sigma(c_{b})-\sigma(a_{b})|}, 1\right\rbrace .$$
Next we determine, under which condition on $\xi(0)$ the inequality: $$\begin{aligned}
\label{eq36}
\Phi_{1}|\Phi_{2}-1|\le \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4}\end{aligned}$$ holds true. Obviously, for $z(p)\in [0,1].$ one has $\Phi_{1}(z(p))\leq max\left\lbrace \sigma(a_{b})\sigma(c_{b}),\sigma(c_{b})\right\rbrace:= m_{\sigma} ,$ for $z(p)\in [0,1].$ Inequality (\[eq36\]) then follows, if $$\begin{aligned}
\label{eq37}
|\Phi_{2}-1| \le \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4m_{\sigma}}.\end{aligned}$$ Now, if $\xi(l)$ and $\xi(p)$ are sufficiently large, then, since $$\begin{aligned}
\label{eq38}
\Phi_{2}-1=\frac{(1+\xi(l))}{\sigma (a_{b})+\xi(l)}\cdot\frac{(1+\xi(p))}{\sigma(c_{b})+(1-\sigma(c_{b}))z(p)+\xi(p)}-1,\end{aligned}$$ the right hand side of (\[eq38\]) behaves like $(1+O(\frac{1}{\xi(l)}))(1+O(\frac{1}{\xi(p)}))-1$, which can be bounded by $R_{6}\left( \frac{1}{\xi(l)}+\frac{1}{\xi(p)}\right) $ for some constant $R_{6}>1$ not depending on $\xi(l)$ and $\xi(p).$ On the other hand, according to relations (\[eq22\]), $\xi(m)$ is for $ m\in (0,q_{n}]$ comparable with $\xi(0)$, and hence $$\begin{aligned}
\label{eq39}
|\Phi_{2}-1|\leq R_{6}\left( \frac{1}{\xi(l)}+\frac{1}{\xi(p)}\right) \leq 2R_{6}e^{v}\frac{1}{\xi(0)}.\end{aligned}$$ Hence if $$2R_{6}e^{v}\frac{1}{\xi(0)}\leq\frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4m_{\sigma}}$$ respectively $$\begin{aligned}
\label{eq40}
\xi(0)\geq \frac{4R_{6}e^{v}m_{\sigma}}{|\sigma(a_{b})\sigma(c_{b})-1|},\end{aligned}$$ then inequality (\[eq37\]) holds true. Finally, we define the constant $C_{0}$ by $$\label{eq41}
C_{0}:=\max \left\lbrace \frac{4R_{6}e^{v}m_{\sigma}}{|\sigma(a_{b})\sigma(c_{b})-1|},1\right\rbrace .$$ From (\[eq34\])-(\[eq41\]) the assertion of Lemma \[lemm4.3\] then follows immediately.\
[**Proof of Lemma \[lemm4.4\]**]{}. Let $D\varphi(x_{0})=\omega >0.$ Fix $n\geq1$. W.l.o.g. we consider the case $n $ odd, the case $n$ even can be deduced from the odd case by reversing the orientation of the circle. From the structure of the dynamical partition $\xi_n(x_{0})$ it follows, that both preimages $\overline{a}_{b}$ , $\overline{c}_{b}$ are in the interval $[T_{f}^{q_{n}}x_{0},T_{f}^{q_{n-1}}x_{0}]$ with $\overline {a}_{b}=T_{f}^{-l}{a}_{b}$, $\overline{c}_{b}=T_{f}^{-p}{c}_{b}$ for some $0\leq l,p\leq q_{n}$. Take the point $\overline{a}_{b}$ and consider its neighbourhood $[T_{f}^{-q_{n-1}}\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}] $. By Corollary \[cor2.2\] for any $a,b\in S^{1}$ all the intervals $[a,b]$, $T_{f}^{q_{n}}[a,b]$, $T_{f}^{-q_{n}}[a,b]$ are $e^{v}$- comparable. Since $\overline{a}_{b}\in [T_{f}^{q_{n}}x_{0},T_{f}^{q_{n-1}}x_{0}] $, it can easily be shown, that the pairs of intervals $([T_{f}^{-q_{n-1}}x_{0},x_{0}], [T_{f}^{-q_{n-1}}\overline{a}_{b},\overline{a}_{b}])$, $([x_{0},T_{f}^{q_{n-1}}x_{0}], [\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}]) $ and $([T_{f}^{-q_{n-1}}x_{0},T_{f}^{q_{n-1}}x_{0}], [T_{f}^{-q_{n-1}}\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}]) $ are $e^{v}$- comparable.
Let $\tau_{0}$ be the middle point of the interval $[T{f}^{-q_{n-1}}\overline{a}_{b},\overline{a}_{b} ].$ Since $[\overline{a}_{b},T_{f}^{q_{n-1}}\tau_{0}]=\\ T_{f}^{q_{n-1}}[T_{f}^{-q_{n-1}}\overline{a}_{b},\tau_{0}]$ and $l([T_{f}^{-q_{n-1}}\overline{a}_{b},\tau_{0}])=l([\tau_{0},\overline{a}_{b}])$ we conclude, that the intervals $[\tau_{0},\overline{a}_{b}]$ and $[\overline{a}_{b},T_{f}^{q_{n-1}}\tau_{0}]$ are $e^{v}$- comparable (see figure $1$).
![Comparison of the intervals $[x_{0},T_{f}^{q_{n-1}} x_0 ]$ and $[\bar{a}_{b},T_{f}^{q_{n-1}}\bar{a}_{b} ]$](AD.IL.DM.-1.eps){width="14cm"}
![The neighbourhoods $U_{n}(\bar{a}_{b})$ and $V_{n}(a_{b})$ are comparable with $[T_{f}^{-q_{n-1}} \bar{a}_{b},T_{f}^{q_{n-1}}
\bar{a}_{b}]$. $\tau_0$ is the middle point of the interval $[T_{f}^{-q_{n-1}}\bar{a}_{b},\bar{a}_{b} ]$](AD.IL.DM.-2.eps){width="14cm"}
Set $$d_{n}:=
\frac{1}{2} min\left\lbrace l([\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}]),l([T_{f}^{-q_{n-1}}\overline{a}_{b},\overline{a}_{b}]) \right\rbrace.$$ Using this and Corollary \[cor2.2\] we get $$\label{eq43}
e^{-v}\frac{1}{2} l([\overline{a}_b,T_f^{q_{n-1}}\overline{a}_b])\leq d_n\leq e^v\frac{1}{2}\,l([\overline{a}_b,T_f^{q_{n-1}}\overline{a}_b]).$$
Notice, that the interval $[\tau_{0},T_{f}^{q_{n-1}}\tau_{0}]$ is one of the two generators of the partition $\xi_{n}(\tau_{0}).$ Hence the intervals $T_{f}^{i}[\tau_{0},T_{f}^{q_{n-1}}\tau_{0}],$$i\in [0,q_{n})$ cover the break point ${a}_{b}$ only once. Using the constants $C_{0}$ and $\zeta_{0}$ in Lemma \[lemm4.3\] we define two neighbourhoods(see figure 2) of the point $\overline{a}_{b}$: $$V_{n}(\overline{a}_{b})=(\overline{a}_{b}-\frac{1}{2}e^{-v}C_{0}^{-1}d_{n},\,\,\overline{a}_{b}+\frac{1}{2}e^{-v}C_{0}^{-1}d_{n}),$$ $$U_{n}(\overline{a}_{b})= [\overline{a}_{b}-\frac{1}{2}\zeta_{0}l(V_{n}(\overline{a}_{b}),\,\,\overline{a}_{b}+\frac{1}{2}\zeta_{0}l(V_{n}(\overline{a}_{b})].$$ It is clear that $U_{n}(\overline{a}_{b})\subset V_{n}(\overline{a}_{b})\subset [\tau_{0},T_{f}^{q_{n-1}}\tau_{0}].$ The construction of the intervals $[z_{s},z_{s+1}]$ will depend of the location of $ \overline{c}_{b}$ in the interval $V_{n}(\overline{a}_{b}). $ There are two possibilities to consider: $$\begin{aligned}
\label{eq44}
{either}\,\,\, \overline{c}_{b}\notin U_{n}(\overline{a}_{b}), i.e.\,\, \overline{c}_b\in[T_f^{q_n}\tau_0,T_f^{q_{n-1}}\tau_0]\setminus U_n(\overline{a}_b)\,\,\, or\,\,\overline{c}_{b}\in U_{n}(\overline{a}_{b})).\end{aligned}$$
Consider the first case, when $ \overline {c}_{b}\in V_{n}(\overline{a}_{b})\setminus U_{n}(\overline{a}_{b})).$ In this case we set $$z_{2}=\overline{a}_{b}, z_{3}=\overline{a}_{b}+\frac{1}{4}l(U_{n}(\overline{a}_{b})),z_{4}=\overline{a}_{b}+\frac{1}{2}l(U_{n}(\overline{a}_{b}))\,\,and
\,\,z_{1}=\overline{a}_{b}-\frac{1}{4}l(U_{n}(\overline{a}_{b})).$$
It is easy to see, that the interval $[z_{1},z_{4}]$ is a subset of $[\tau_{0},T_{f}^{q_{n-1}}\tau_{0}]$ and it does not contain the break point $\overline{c}_{b}.$ Next we check, that the intervals $[z_{s},z_{s+1}]$, $s=1,2,3$ satisfy properties 1)-3) in Lemma \[lemm4.4\]. The interval $[z_1,z_4]$ is $q_n$-small, because $[z_1,z_4]\subset [\tau_0,T_f^{q{n-1}}\tau_0]$ which is one of the generators of the dynamical partition $\xi_n(\tau_0)$. By construction, the length of $[z_{1},z_{4}]$ is equal to $\frac{3}{4}e^{-v}C_{0}^{-1}\zeta_{0}d_{n},$ but $d_{n}$ is half the length of one of the intervals $[T_{f}^{-q_{n-1}}\overline{a}_{b},\overline{a}_{b}]$ or $[\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}] .$ Consequently, the length of $[z_{1},z_{4}]$ is $4e^{v}C_{0}\zeta_{0}^{-1}$- comparable with $[x_{0},T_{f}^{q_{n-1}}x_{0}] $. Next we check, that the assumptions of Lemma \[lemm4.1\] hold for both intervals $[z_{s},z_{s+1}]$ and $[T_{f}^{q_{n}}z_{s},T_{f}^{q_{n}}z_{s+1}]$. Note first, that the lengths of the intervals $[z_{s},z_{s+1}]$, $s=1,2,3 $ are equal to $\frac{1}{4}e^{-v}C_{0}^{-1}\zeta_{0}d_{n},$ and that the intervals $[z_{s},z_{s+1}]$ and $T_{f}^{q_{n-1}}[z_{s},z_{s+1}]$ are $e^{v}$- comparable for $s=1,2,3 .$ Hence assumption a) of Lemma \[lemm4.1\] holds true for these intervals with constant $e^{v}.$
Next we check assumption b) of Lemma \[lemm4.1\]. It is easy to see that for all $i=1,2,3,4$ $$\begin{aligned}
\label{eq45}
|x_{0}-z_{i}|\le |x_{0}-z_{2}|+|z_{4}-z_{1}|=|x_{0}-z_{2}|+d_{n},\nonumber\\
|x_{0}-T_{f}^{q_{n}}z_{i}|\le |x_{0}-z_{2}|+|z_{2}-T_{f}^{q_{n}}z_{2}|+|T_{f}^{q_{n}}z_{4}-T_{f}^{q_{n}}z_{1}|.\end{aligned}$$ The point $z_{2}$ belongs to $[T_{f}^{q_{n}}x_{0},T_{f}^{q_{n-1}}x_{0}]\subset[T_{f}^{-q_{n-1}}x_{0},T_{f}^{q_{n-1}}x_{0}]$, which is $e^{v}$ comparable with $[T_{f}^{-q_{n-1}}\overline{a}_{b},T_{f}^{q_{n-1}}\overline{a}_{b}]).$ But the length of this last interval is $4e^{v}$-comparable with $d_{n}.$ In complete analogy we can estimate the second expression in (\[eq45\]) by $12e^{v}C_{0}\zeta_{0}^{-1}d_{n}$.\
By assumption, the intervals $T_{f}^{i}[z_{1},z_{1}]$ do not cover the break point $c_{b},$ but they cover the point $a_{b}$ exactly once with $z_{2}=T_{f}^{-l}a_{b}.$
Next consider the second case, when $ \overline {c}_{b}\in U_{n}(\overline{a}_{b}).$ There are again two possibilities: If $ \overline {c}_{b}\in [\overline{a}_{b}-\frac{1}{2}\zeta_{0}l(V_{n}(\overline{a}_{b})),\,\overline{a}_{b}],$ we set $$z_{1}=\overline{a}_{b}-\frac{1}{2}l(V_{n}(\overline{a}_{b})),z_{2}=\overline{a}_{b}, z_{3}=\overline{a}_{b}+\frac{1}{2}C_{0}l(V_{n}(\overline{a}_{b})),z_{4}=\overline{a}_{b}+C_{0}l(V_{n}(\overline{a}_{b})).$$ Then the length of the first interval $[z_{1},z_{2}]$ is $\frac{1}{2}e^{-v}C_{0}^{-1}d_{n}$ and of the other ones it is equal to $\frac{1}{2}e^{-v}d_{n}$. Hence the lengths of these intervals are $2e^{v}C_{0}$-comparable with $d_{n}$. The first two statements of Lemma \[lemm4.4\] for these intervals can be checked in complete analogy to the first case in (\[eq44\]).
Next we show, that in the present case the intervals\
$\left\lbrace T_{f}^{i}[z_{1},z_{2}],\,\,T_{f}^{i}[z_{2},z_{3}],\,\, T_{f}^{i}[z_{3},z_{4}],\,\, 0\leq i \leq q_{n} \right\rbrace $ cover both break points $a_{b}$, $c_{b}$ regularly with constants $C_{0}$ and $\zeta_{0}$. By construction, these intervals cover both break points exactly once. Moreover we have $z_{2}=\overline{a}_{b}$ and $ \overline {c}_{b}\in [z_{1},z_{2}].$ It is easy to see, that $\xi(0)=\frac{z_{3}-z_{2}}{z_{2}-z_{1}}=C_{0}.$ Since $ \overline {c}_{b}\in [\overline{a}_{b}-\frac{1}{2}\zeta_{0}l(V_{n}(\tau_{0})),\,\overline{a}_{b}]$, we find that $z(0)=\frac{z_{2}-\overline {c}_{b}}{z_{2}-z_{1}}\le \zeta_{0}.$ So the intervals $[z_{s},z_{s+1}]$, $s=1,2,3$ satisfy the statements of Lemma \[lemm4.4\].
Consider finally the case, when the break point $\overline{c}_{b}$ is in the interval $ [ \overline{a}_{b},\overline{a}_{b}+\frac{1}{2}\zeta_{0}l(V_{n}(\tau_{0})]$. In this case we set $$z_{1}=\overline{a}_{b}-C_{0}l(V_{n}(\overline{a}_{b})), z_{2}=\overline{a}_{b}-\frac{1}{2}C_{0}l(V_{n}(\overline{a}_{b})), z_{3}=\overline{a}_{b}, z_{4}=\overline{a}_{b}+\frac{1}{2}l(V_{n}(\overline{a}_{b})).$$ The proof of Lemma \[lemm4.4\] for these intervals $[z_{s},z_{s+1}]$, $s=1,2,3$ proceeds now exactly as in the previous case. This concludes the proof of Lemma \[lemm4.4\].\
**Proof of Lemma \[lemm4.5\]**. Assume, that the circle homeomorphism $T_{f}$ satisfies the assumptions of Theorem \[DLM\] and the intervals $ [z_{s},z_{s+1}]$, $s=1,2,3$ satisfy Lemma \[lemm4.4\]. Consider first the case when the intervals $\left\lbrace T_{f}^{i}[z_{1},z_{2}],\,\,T_{f}^{i}[z_{2},z_{3}]\,
T_{f}^{i}[z_{3},z_{4}],\,\,\,0\leq i \leq q_{n}-1 \right\rbrace $ cover both break points $a_{b}$, $c_{b}$ regularly with constants $C_{0}$ and $\zeta_{0}.$ Suppose that $z_{2}=\overline{a}_{b}=T_{f}^{-l}{a}_{b} $ and $\overline{c}_{b}=T_{f}^{-p}{c}_{b},$ for some $0\leq l,p\leq q_{n}$ and $$\begin{aligned}
\label{eq46}
\frac{z_{3}-z_{2}}{z_{2}-z_{1}}\le C_{0},\,\,\ 0\le \frac{z_{3}-\overline{c}_{b}}{z_{2}-z_{1}}\le \zeta_{0}.\,\,\end{aligned}$$ Lemma \[lemm4.2\] shows that $$\begin{aligned}
\label{eq47}
Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})
=[G(\xi(l)+\chi_{1}][F(\xi(p),z(p))+\chi_{2}]\times \nonumber\\
\times\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f})\end{aligned}$$ if $z_{2}=\overline{a}_{b},$ and $\overline{c}_{b}\in \left[z_{1}, z_{2} \right] $ where $$\begin{aligned}
\label{eq48}
|\chi_{j}|=|\chi_{j}(z_{1},z_{2},z_{3},z_{4})|\leq
K_{1} \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,\,\, j=1,2.\end{aligned}$$ with some constant $K_{1}$ not depending on $n$ and $\varepsilon.$ Next we estimate the right hand side in equation \[eq47\]. Fix some $\varepsilon>0.$ By assumption, the second derivative $D^{2}f$ of the lift $f$ belongs to $L^{1}(S^{1},dl).$ Hence it can be written in the form $D^{2}f(x)=g_{\varepsilon}(x)+\theta_{\varepsilon}(x)$ with $g_{\varepsilon}$ a continuous function on $S^{1}$ and $\parallel\theta_{\varepsilon}(x)\parallel_{L^{1}}< \varepsilon.$ By assumption, among the intervals $T_{f}^{i}[z_{s},z_{s+1}]$, $0\leq i\leq q_{n}$, only the intervals $T_{f}^{l}[z_{1},z_{4}]$ and $T_{f}^{p}[z_{1},z_{4}]$ contain the break points ${a}_{b}$ respectively ${c}_{b}$.
Obviously $$|\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist(T_{f}^{i}z_{1},T_{f}^{i}z_{2},T_{f}^{i}z_{3},T_{f}^{i}z_{4};
T_{f}) -1|=$$
$$\begin{aligned}
\label{eq49}
=|exp\{\overset{q_{n}-1}{\underset{i=0,\;i\neq l,b}
{\sum}}
\text{log}(1+(Dist(T_{f}^{i}z_{1},T_{f}^{i}z_{2},T_{f}^{i}z_{3},T_{f}^{i}z_{4};
T_{f})-1))\}-1|\end{aligned}$$
Next applying Theorem \[theo3.1\] we obtain $$\begin{aligned}
\label{eq50}
|Dist(T_{f}^{i}z_{1},T_{f}^{i}z_{2},T_{f}^{i}z_{3},T_{f}^{i}z_{4};
T_{f})-1|\le C_{1}|T_{f}^{i}z_{4}-T_{f}^{i}z_{1}|\times\nonumber\\
\overset{}{\underset{x,t\in
[T_{f}^{i}z_{1},T_{f}^{i}z_{4}]}
{\max}|g_{\varepsilon}(x)-g_{\varepsilon}(t)|}
+C_{1}\overset{T_{f}^{i}z_{4}}{\underset{T_{f}^{i}z_{1}}{\int}}|\theta_{\varepsilon}(y)|dy+C_{1}
\Big(\overset{T_{f}^{i}z_{4}}{\underset{T_{f}^{i}z_{1}}{\int}}|D^{2}f(y)|dy\Big)^{2}\end{aligned}$$ where the constant $C_{1}>0$ depends only on the function $f.$
For $D^{2}f\in L^{1}([0,1])$ the function $
\Psi(x)=\overset{x}{\underset{0}{\int}}<D^{2}f(y)|dy
$ is absolutely continuous on $S^{1}$. Note that the functions $\Psi(x)$ and $g_{\varepsilon}(x)$ are uniformly continuous on $ S^{1}$ because they are continuous on $S^{1}$. Hence there exists $\delta_{0}=\delta_{0}(\varepsilon)>0$ such that for any $x,t\in S^{1}$ with $|x-t|<\delta_{0},$ the inequalities $$\begin{aligned}
\label{eq51}
|\Psi(x)-\Psi(t)|<\varepsilon,\,\,
|g_{\varepsilon}(x)-g_{\varepsilon}(t)|<\varepsilon,\end{aligned}$$ hold true.\
By assumption, the interval $[z_{1},z_{4}]$ is $q_{n}-$small. Hence by Corollary \[cor2.2\] for all $ 0\leq i\leq q_{n}$ we have $
l(T_{f}^{i}[z_{1},z_{4}])\leq const\,\lambda^{n},\ 0 \leq i
\leq q_n-1,
$ with $\lambda=(1+e^{-v})^{\frac{-1}{2}}<1.$ Consequently there exists a number $N_{0}=N_{0}(\delta)>0$ such that for $n>N_{0}$ and all $0 \leq i\leq q_{n}$ one has $
l(T_{f}^{i}[z_{1},z_{4}])\leq \delta_{0}.
$ This together with (\[eq51\]) implies that for $n>N_{0}$ the following inequalities $$\begin{aligned}
\label{eq52}
|\Psi(x)-\Psi(y)|<\varepsilon,\,\,\,\,
|g_{\varepsilon}(x)-g_{\varepsilon}(y)|<\varepsilon,\end{aligned}$$ hold for all $x,y\in [T_{f}^{i}z_{1},T_{f}^{i}z_{4}],$ and all $0 \leq i\leq q_{n}$.
On the other hand, since the interval $[z_{1},z_{4}]$ is $q_{n}-$ small, the intervals $T_{f}^{i}[z_{1},z_{4}]$, $0 \leq i\leq q_{n}$, are non intersecting and trivially $$\sum_{i=0}^{q_{n}}l(T_{f}^{i}[z_{1},z_{4}])\le1.$$ Since $\parallel\theta_{\varepsilon}\parallel_{L^{1}}< \varepsilon$, we find, using relations (\[eq49\])-(\[eq52\]), $$|\prod_{\substack{0\leq i<q_{n} \\ i\neq l,p }}Dist(T_{f}^{i}z_{1},T_{f}^{i}z_{2},T_{f}^{i}z_{3},T_{f}^{i}z_{4};
T_{f})-1| \le$$ $$\le
C_{2}\sum_{i=0}^{q_{n}-1}
|Dist(T_{f}^{i}z_{1},T_{f}^{i}z_{2},T_{f}^{i}z_{3},T_{f}^{i}z_{4};
T_{f})-1|\le C_{2}\varepsilon \sum_{i=0}^{q_{n}-1}
|T_{f}^{i}z_{4}-T_{f}^{i}z_{1}|+$$$$+C_{2} \sum_{i=0}^{q_{n}-1}\overset{T_{f}^{i}(z_4)}
{\underset{T_{f}^{i}(z_1)}{\int}}|\theta_{\varepsilon}(y)|dy
+C_{2} \sum_{i=0}^{q_{n}-1}
|\Psi(T_{f}^{i}z_{4})-\Psi(T_{f}^{i}z_{1})|\overset{T_{f}^{i}(z_4)}
{\underset{T_{f}^{i}(z_1)}{\int}}|D^{2}f(y)|dy\le$$ $$\begin{aligned}
\label{eq53}
\,\,\,\,\,\,\,\,\le C_{2}\left\{2\varepsilon+\overset{1}
{\underset{0}{\int}}|\theta_{\varepsilon}(y)|dy+\varepsilon\overset{1}
{\underset{0}{\int}}|D^{2}f_{1}(y)|dy\right\}
\le C_{2}(3+||D^{2}f||_{L^{1}})\varepsilon \end{aligned}$$ where the constant depends only on $f.$.
Next we estimate the expression $(G(\xi(l))+\chi_{1})(F(\xi(p),z(p))+\chi_{2})$. where $|\chi_{i}|$ is bounded above by $K_{1} \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy$, for $j=1,2$ with some constant $K_{1}$ not depending on $n$ and $\varepsilon$. For $n>N_{0}$ we have $\overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy=\Psi(z_{4})-\Psi(z_{1})<\varepsilon.$ Since $G(x)$ is bounded for $x>0$ and $F(x,t),$ is bounded for $x>0$ and $1\leq t \leq 1$, it is hence sufficient to estimate the term $G(\xi(l))F(\xi(p),z(p))$ in the above product. By assumption, the intervals $\left\lbrace T_{f}^{i}[z_{1},z_{2}],\,\,T_{f}^{i}[z_{2},z_{3}]\,
T_{f}^{i}[z_{3},z_{4}],\,\,\,0\leq i \leq q_{n}-1 \right\rbrace $ cover both break points $a_{b}$, $c_{b}$ regularly with constants $C_{0}$ and $\zeta_{0}.$ Applying Lemma \[lemm4.3\] we conclude $$|G(\xi(l))F(\xi(p),z(p))-1|\geq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{4}>0,$$ since by assumption the product of the jump ratios of $Df$ at the break points is nontrivial i.e. $\sigma(a_{b})\sigma(c_{b})\neq 1$. It then follows, that for sufficiently large $n$ and small $\varepsilon$ the inequality $$|[G(\xi(l))+\chi_{1}][F(\xi(p),z(p))+\chi_{2}]-1|\geq \frac{|\sigma(a_{b})\sigma(c_{b})-1|}{8}$$ holds true. This together with (\[eq44\]) and (\[eq53\]) imply the assertion of Lemma \[lemm4.5\] in the case of a regular covering of the two break points.\
W.l.o.g. we assume next that the intervals $\left\lbrace T_{f}^{i}[z_{1},z_{2}],\,\,T_{f}^{i}[z_{2},z_{3}]\,
T_{f}^{i}[z_{3},z_{4}],\,\,\,0\leq i \leq q_{n}-1 \right\rbrace $ cover only the break point $a_{b}$ with $z_{2}=\overline{a}_{b}=T_{f}^{-l}a_{b}$ for some $0\leq l\leq q_{n} $ and satisfy properties 1), 2) of Lemma \[lemm4.4\].
We write $Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})$ in the following form $$\begin{aligned}
\label{eq54}
Dist(z_{1},z_{2},z_{3},z_{4};T_{f}^{q_{n}})
=Dist (T_{f}^{l}z_{1},T_{f}^{l}z_{2}, T_{f}^{l}z_{3},T_{f}^{l}z_{4};T_{f})\times \nonumber\\
\times\prod_{\substack{0\leq i<q_{n} \\ i\neq l }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f}).\end{aligned}$$ For sufficiently large $n$ and any $\varepsilon >0$ the product over $i\neq l, p$ in (\[eq54\]) takes it value in an $\varepsilon- $neighbourhood of 1. By assumption, only the interval $T_{f}^{l}[z_{1},z_{4}]$ contains the break point $a_{b}$ with $a_{b}=T_{f}^{l}z_{2}.$ Using Lemma \[lemm3.2\] we find $$\begin{aligned}
\label{eq55}
|Dist (T_{f}^{l}z_{1},T_{f}^{l}z_{2}, T_{f}^{l}z_{3},T_{f}^{l}z_{4};T_{f})- \frac{\sigma(a_{b})(1+\xi(l))}{\sigma(a_{b})+\xi(l)}|\nonumber\\
\le K_1 \overset{z_4}{\underset{z_1}{\int}}|D^{2}f(y)|dy,\end{aligned}$$ where the constant $K_1>0$ depends only on the function $f$ and where $\xi(l)=\frac{T_{f}^{l}z_{3}-T_{f}^{l}z_{2}}{T_{f}^{l}z_{2}-T_{f}^{l}z_{1}}.$ Obviously $$\frac{\sigma(a_{b})(1+\xi(l))}{\sigma(a_{b})+\xi(l)}-1=\frac{(\sigma(a_{b})-1)\xi(l)}{\sigma(a_{b})+\xi(l)}.$$ Using this and the inequalities $R_{2}^{-1}\le \xi(l)\le R_{2}$, following from (\[eq22\]), and the comparability of the intervals $[z_{s}, z_{s+1}]$ for $s=1,2,3$, we obtain $$R_{3}^{-1} \le \frac{(\sigma(a_{b})-1)\xi(l)}{\sigma(a_{b})+\xi(l)}\le R_{3}$$ where the constants $R_{i}>0,\,\,i=1,2$ depend only on $f.$ Finally we obtain $$\begin{aligned}
\label{eq56}
|Dist (T_{f}^{l}z_{1},T_{f}^{l}z_{2}, T_{f}^{l}z_{3},T_{f}^{l}z_{4};T_{f})-1|\geq const>0,\end{aligned}$$ where the constant again depends only on $f$. As in the first case, the inequality\
$$|\prod_{\substack {0\leq i<q_{n} \\ i\neq l }}Dist (T_{f}^{i}z_{1},T_{f}^{i}z_{2}, T_{f}^{i}z_{3},T_{f}^{i}z_{4};T_{f})-1|\le const\,\, \varepsilon.$$ holds true also in the present case. This together with (\[eq54\]) and (\[eq56\]) proves Lemma \[lemm4.5\].
**Proof of Theorem \[DLM1\]**. The idea of the proof of Theorem \[DLM1\] is completely similar to the one of Theorem \[DLM\]. Hence we will only give the construction of the intervals $[z_{s}, z_{s+1}]$, $s=1,2,3$, which play the key role in the proof.\
Consider the $n-$th dynamical partition $\xi_{n}(\overline{a}_{b})$ of the preimage $\overline{a}_{b}$ of the break point $a_{b}$ in the interval $[T_{f}^{q_{n}}x_{0}, T_{f}^{q_{n}-1}x_{0}]$ around the point $x_{0}$, at which there exists a positive derivative $DT_{\varphi}(x_0)$ of the conjugating homeomorphism $T_{\varphi}(x).$ Since the rotation number $\rho_{f}$ is irrational of bounded type there exists a subsequence $\left\lbrace n_{k},k=1,2,...\right\rbrace \in\mathbb{N} $ such that for every $n_{k}$ the interval $[\overline{a}_{b}, T_{f}^{q_{n{k}-1}}\overline{a}_{b}]$ respectively the interval $[T_{f}^{q_{n{k}}}\overline{a}_{b}, \overline{a}_{b}]$ contains the point $\overline{c}_{b}=T_{f}^{-p}c_{b}$ for some $p\in [0,q_{n_{k}})$ and such that furthermore $K_{3}^{-1}\le \frac{\overline{c}_{b}-\overline{a}_{b}}{T_{f}^{q_{n_{k}-1}}\overline{c}_{b}-\overline{a}_{b}}\le K_{3}$ respectively $K_{3}^{-1}\le \frac{\overline{a}_{b}-\overline{c}_{b}}{\overline{c}_{b}-T_{f}^{q_{n_{k}}}\overline{a}_{b}}\le K_{3}$ holds, where the constant $K_{3}$ depends only on $f.$ Set $d_{n_{k}}=min\left\lbrace|\overline{c}_{b}-\overline{a}_{b}|, |T_{f}^{q_{n_{k}-1}}\overline{a}_{b}-\overline{c}_{b}| \right\rbrace $ and define the points $$z_{2}=\overline{c}_{b}, z_{1}=\overline{c}_{b}-\frac{1}{2}d_{n_{k}}, z_{3}=\overline{c}_{b}+\frac{1}{2}d_{n_{k}}, z_{4}=\overline{c}_{b}+d_{n_{k}}.$$ As in the proof of Lemma \[lemm4.4\] it can be checked that the intervals $[z_{s}, z_{s+1}]$, $s=1,2,3$ then satisfy the statements of this Lemma.\
[**[Acknowledgement]{}**]{}\
The work of Akhtam Dzhalilov during a stay at the University of Clausthal was supported by the German Research Council (DFG) under project Ma 633/18-1.
[24]{} Arnol’d, V. I.: Small denominators: I. Mappings from the circle onto itself. Izv. Akad. Nuak SSSR, Ser. Mat., **25**,21-86(1961).
Denjoy, A.: Sur les courbes définies par les équations différentielles à la surface du tore. J. Math. Pures Appl., **11**,333-375 (1932).
de Faria, E., de Melo, W.: Rigidity of critical circle mappings. I. J. Eur. Math. Soc. (JEMS), **1**,(4),339-392(1999).
de Melo, W., van Strien, S.: One dimensional Dynamics. Springer Verlag Berlin, p.3-25 (1993).
Dzhalilov, A.A. , Khanin, K.M.: On invariant measure for homeomorphisms of a circle with a point of break., Funct. Anal. Appl., **32**,(3)153-161(1998).
Dzhalilov, A.A., Liousse, I.: Circle homeomorphisms with two break points. Nonlinearity, **19**, 1951-1968(2006).
Cornfeld, I.P., Fomin, S.V., Sinai, Ya.G.: Ergodic Theory, Springer Verlag, Berlin (1982).
Cvitanovic,P.: (Ed.) Universality in Chaos. Second ed. , Adam Hilger, Bristol (1989).
Graczyk, J., Swiatek, G.: Singular measures in circle dynamics. Commun. Math. Phys., **157**, 213-230(1993).
Herman, M.: Sur la conjugaison différentiable des difféomorphismes du cercle à des rotations. Inst. Hautes Etudes Sci. Publ. Math., **49**, 225-234(1979).
Herman; M.: Sur les difféomorphismes du cercle de nombre de rotation de type constant. Conf. on Harmonic Analysis in Honor of A. Zygmund, vol. II, 707-725.
Hawkins, J., Schmidt, K.: On $C^{2}$-Diffeomorphisms of the circle which are of type $III_1$. Invent. Math., **66**, 511-518(1982). Katok, A., Hasselblatt,B.: Introduction to the Modern Theory of Dynamical Systems. Cambridge University Press, Cambridge (1995).
Katznelson, Y., Ornstein, D.: The absolute continuity of the conjugation of certain diffeomorphisms of the circle. Ergod. Theor. Dyn. Syst.,**9**,681-690 (1989).
Khanin, KM., Sinai, Ya. G.: Smoothness of conjugacies of diffeomorphisms of the circle with rotations. Russ. Math. Surv., **44**,69-99(1989), translation of Usp. Mat. Nauk, **44**,57-82(1989).
Khanin, K.M., Khmelev, D.: Renormalizations and Rigidity Theory for Circle Homeomorphisms with Singularities of the Break Type. Commun. Math. Phys.,**235**,69-124(2003)
Khanin, K.M., Vul, E.B.: Circle homeomorphisms with weak discontinuities. Advances in Soviet Mathematics, **3**,57-98 (1993).
Liousse, I.: Nombre de rotation, mesures invariantes et ratio set des homéomorphisms affines par morceaux du cercle, Ann.Inst. Fourier, **55**, 431-482(2005).
Moser, J.: A rapid convergent iteration method and non-linear differential equations. II. Ann. Scuola Norm. Sup. Pisa, **20**(3),499-535(1966).
Poincaré, H.: Sur le courbes définies par les équations différentielles. J. Math. Pures et Appl., **1**,167-244(1885), Reprinted in Ouvres de Henri Poincaré), Tome I, Gauthier-Villars, Paris, (1928).
Stein, M.: Groups of piecewise linear homeomorphisms. Trans. A.M.S. **32**, 477-514, (1992).
Teplinskii, A. Yu, Khanin, K.M.: Rigidity for circle diffeomorphisms with simgularities. Russ. Math. Surv.,**759**(2),329-353.(2004), translation of Usp. Mat. Nauk, **59**(2),137-160(2004).
Yoccoz, J. C.: Il n’y a pas de contre-exemple de Denjoy analytique. C. R. Acad. Sci. Paris T., **298**(7),141-144 (1984).
[^1]: MSC: 37E10, 37C15, 37C40
Keywords: Circle homeomorphism, break points, rotation number, invariant measures
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the energy dependence of global polarization of $\Lambda$ hyperons in peripheral $Au-Au$ collisions. We combine the calculation of vorticity and strange chemical potential in the framework of kinetic Quark-Gluon String Model with the anomalous mechanism related to axial vortical effect. We pay special attention to the temperature dependent contribution related to gravitational anomaly and found that the preliminary RHIC data are compatible with its suppression discovered earlier in lattice calculations.'
author:
-
-
-
-
title: 'Hyperon polarization in Heavy-Ion Collisions and gravity-related anomaly'
---
Introduction
============
The experimental evidences for polarization of hyperons in heavy-ion collisions found by STAR collaboration [@Lisa] attracted recently much attention [@Becattini:2016gvu; @Karpenko:2016jyx; @Karpenko:2016idy; @Xie:2015xpa; @Fang:2016uds].
The studies of polarization are often performed [@Becattini:2013vja] in the framework of approach exploring local equilibrium thermodynamics [@Becattini:2014yxa] and hydrodynamical calculations of vorticity [@Betz:2007kg; @Csernai:2013bqa; @Csernai:2014hva].
There is another (although related [@Prokhorov:2017atp]) approach to polarization first proposed in [@Rogachevsky:2010ys] and independently in [@Gao:2012ix]. The so-called axial vortical effect (see e.g. [@Kalaydzhyan:2014bfa] and references therein) being the macroscopic manifestation of axial anomaly [@Son:2009tf] leads to induced axial current of strange quarks which may be converted to polarization of $\Lambda$-hyperons [@Rogachevsky:2010ys; @Gao:2012ix].
The effect is proportional to vorticity and helicity of the strong interacting medium, and, in particular, to helicity separation effect discovered [@Baznat:2013zx] in the kinetic Quark-Gluon-String Model(QGSM)[@toneev83; @toneev90; @amelin91] and confirmed [@Teryaev:2015gxa] in HSD [@Cassing:1999es] model. This helicity separation effect receives [@Baznat:2013zx; @Teryaev:2015gxa] the significant contribution $ \sim \vec v_y \vec \omega_y$ from the transverse component of velocity and vorticity. It is easily explained [@Baznat:2013zx] by the same signs of transverse vorticities in the “upper” and “lower” (w.r.t. reaction plane) half-spaces, combined with the opposite signs of velocities. At the same time, even larger contribution [@Baznat:2013zx; @Teryaev:2015gxa] of longitudinal components of velocity and vorticity $\sim \vec v_z \vec \omega_z$ implies the appearance of the “quadrupole” structure of longitudinal vorticity, recently found [@Becattini:2017gcx] in the hydrodynamical approach.
Indeed, the opposite values of longitudinal velocities in the “left” and “right” (w.r.t. to “vertical” plane $x=0$ normal to reaction one and containing the beams direction) require exactly the quadrupole structure of longitudinal vorticities in the quater-spaces formed by the intersection of reaction and vertical planes: $$\begin{aligned}
h =h_x+h_y+h_z \sim sign(y); \\
v_z \sim sign (x); \\
\omega_z \sim sign(x) sign (y), \end{aligned}$$ where $h_i = v_i \omega_i$ is the contribution of the respective component of velocity and vorticity to the helicity density. It is this quadrupole structure of vorticity that leads to the up-down mirror structure of helicity after multiplication by the left-right mirror structure of velocity: $$h_z = \omega_z v_z \sim (sign(x))^2 sign(y) = sign(y).$$
To make (2) applicable and observe the quadrupole picture one needs to average the longitudinal velocity and vorticity over the cylinder symmetric w.r.t. the plane $z=0$. Otherwize, taking the different slices $z=const$, longitudinal velocity is not, generally speaking, changing sign with $x$. The dependence of the quadrupole picture over the height of tghis cylinder is represented at Figure 1.
Later more detailed calculations were performed [@Baznat:2015eca], including the structure of emergent vortex sheets, as well as spatial and temporal dependence of strange chemical potential which is also the ingredient of anomalous approach to polarization.
Anomalous mechanism of hyperon polarization
===========================================
Anomalous mechanism of polarization makes this effect qualitatively similar to the local violation [@Fukushima:2008xe] of discrete symmetries in strongly interacting QCD matter. The most well known here is the Chiral Magnetic Effect (CME) which uses the (C)P-violating (electro)magnetic field emerging in heavy ion collisions in order to probe the (C)P-odd effects in QCD matter.
The polarization, in turn, is similar to Chiral Vortical Effect (CVE)[@Kharzeev:2007tn] due to coupling to P-odd medium vorticity leading to the induced electromagnetic and all conserved-charge currents [@Rogachevsky:2010ys], in particular the baryonic one.
Recently the pioneering preliminary results on global polarization of $\Lambda$ and $\bar \Lambda$ hyperons in $Au-Au$ collisions in Beam Energy Scan at RHIC were released [@Lisa] and the qualitative tendency of polarization decrease with energy in agreement with the prediction [@Rogachevsky:2010ys] was revealed. The recent theoretical analysis [@Sorin:2016smp] suggested that decrease of polarization with energy is related to the suppression of Axial Magnetic effect contribution in strongly correlated QCD matter found in lattice simulations.
Indeed, the chiral vorticity coefficient describing the axial vortical effect $$\label{cv} c_V=\frac{\mu_s^2+\mu_A^2}{2 \pi^2}+\frac{T^2}{6},\quad
%Q_5^s=N_c \int d^3 x \,c_V \gamma^2 \epsilon^{i j k}v_{i}
%\partial_{j}v_ k$$ contains temperature dependent term related to gravitational anomaly [@Landsteiner:2011iq], and naively it can be quite substantial and increase with energy. However, lattice simulations [@Braguta:2016pwq] lead to the zero result in the confined phase and to the suppression by one order of magnitude at high temperatures. Neglecting axial chemical potential, the coefficient $c_V$ takes the form $$\label{cvl} c_V=\frac{\mu_s^2}{2 \pi^2}+k \frac{T^2}{6},\quad
%Q_5^s=N_c \int d^3 x \,c_V \gamma^2 \epsilon^{i j k}v_{i}
%\partial_{j}v_ k$$
.
As soon as for free fermion gas the $T^2/6$ term is recovered [@Buividovich:2013jba] for large lattice volume at fixed temperature, the above-mentioned suppression should be attributed to the correlation effects. It was suggested[@Sorin:2016smp], that the accurate measurements of polarization energy dependence may serve a sensitive probe of strongly correlated QCD matter. In the current paper we perform numerical simulations to implement this suggestion.
The polarization is related [@Baznat:2013zx; @Sorin:2016smp] to the strange axial charge $$\label{q5}
Q_5^s=N_c \int d^3 x \,c_V \gamma^2 \epsilon^{i j k}v_{i}
\partial_{j}v_ k,$$ and as a result the quark and hadronic observables are related, that is of special importance in the confined phase. Another approach to confined phase is provided by consideration [@Teryaev:2017nro] of the vortices in pionic superfluid, whose cores are associated with polarized baryons
Numerical simulations of axial anomaly contributions to (anti)hyperon polarization
==================================================================================
We performed the numerical simulations in QGSM model [@toneev83; @toneev90; @amelin91]. We decomposed the space-time to the cells, allowing to define velocity and vorticity in the kinetic model, as described in detail in [@Baznat:2013zx]. To define the strange chemical potential (assuming that $\Lambda$ polarization is carried by strange quark) we used the matching procedure [@Baznat:2015eca] of distribution functions to its (local) equilibrium values. In this paper, we also determine in this way the values of temperature. In general, let us stress that we realized in our particular case the relation between kinetics, hydrodynamics and thermodynamics.
We first neglect the gravitational anomaly contribution and start by considering the energy dependence of polarization (described in detail in [@Sorin:2016smp]) for three values of impact parameter. The results are presented at Figure 2. The curves correspond to $b=8.0 fm, 6.4 fm, 4.8 fm$.
{width="80.00000%"}
We continue by the inclusion of contribution related to gravitational anomaly, which is the central issue of this paper. The results are presented at Figure 3. We consider as a starting point the original value of anomaly coefficient[@Landsteiner:2011iq] $T^2/6$ which is reproduced for large lattice volume at fixed temperature[@Buividovich:2013jba]. We present the curves following from the coefficients suppressed by factor $k$ (\[cvl\]) resulting from the lattice calculations [@Braguta:2016pwq]. We compare values of $k=1$ with $k=0, 1/15,
1/10$. As one can see, the lattice-supported value $1/15$ is most close to the behavior of preliminary data which may be considered as a signal of strongly correlated matter formation. The closeness of $k=0$ curve to the experimental points may be related to the contribution of confinement phase, where lattice calculations [@Braguta:2016pwq] lead to zero temperature-dependent effect. At the same time, already $k=1/10$ leads to the curve growing with energy, while $k=1$ leads to extremely strong growth.
{width="80.00000%"}
The $\bar \Lambda$ polarization is emerging due to the polarization of $\bar s-$quarks, which has the same sign, as the axial current and charge are C-even operators. The magnitude of the $\bar \Lambda$ is larger as the same axial charge is distributed between the polarizations of the smaller number of particles [@Sorin:2016smp]. It is mandatory to take into account the contribution of $K^*$ mesons. In the case of $\Lambda$ the $K^{*-}, \bar K^0$ mesons contain two sea(anti)quarks and does not change the polarization significantly. At the same time, for $\bar \Lambda$ the relevant $K^{*+}, K^0$ mesons are more numerous and make the excess of $\bar \Lambda$ polarization less pronounced.
Note that this excess is larger for smaller energies, where suppression of $\bar \Lambda$ is larger. This differs from the (C-odd) effect of magnetic field, which is increasing with energy, although more detailed studies taking into account the finite time of magnetic field existence are required.
The quantitative analysis of these effects, taking into account the gravitational anomaly contribution, is presented at Figure 4.
{width="80.00000%"}
The result is in reasonable agreement with STAR data, although further analysis is required.
Conclusions and Outlook
=======================
We numerically studied the generation of polarization by the anomalous mechanism (Axial Vortical Effect) and compared it with the observed data.
First we neglected the gravitational anomaly related temperature dependent contribution when the decrease of chemical potential with energy leads, in turn, to the decrease of polarization. We considered this effect for three impact parameters.
We also included the contribution related to gravitational anomaly proportional to $T^2$ and studied its possible suppression in strongly correlated matter. We found that the preliminary data are in accordance with suppression effect found on the lattice.
We also considered the polarization of $\bar \Lambda$ hyperons, taking into account the contribution of $K^*$ mesons. We found that the $\bar \Lambda$ polarization is larger than that of $\Lambda$ and is growing at smaller energies.
The further more accurate measurements of $\Lambda$ polarization should provide the additional check of gravitational anomaly related contribution.
The useful discussions with F. Becattini, V. Braguta, S. Voloshin and V.Zakharov are gratefully acknowledged. This work was supported in part by the Russian Foundation for Basic Research, Grant No. 17-02-01108.
[99]{} L. Adamczyk [*et al.*]{} \[STAR Collaboration\], arXiv:1701.06657 \[nucl-ex\].
F. Becattini, I. Karpenko, M. Lisa, I. Upsal and S. Voloshin, Phys. Rev. C [**95**]{}, no. 5, 054902 (2017) doi:10.1103/PhysRevC.95.054902 \[arXiv:1610.02506 \[nucl-th\]\].
I. Karpenko and F. Becattini, Eur. Phys. J. C [**77**]{}, no. 4, 213 (2017) doi:10.1140/epjc/s10052-017-4765-1 \[arXiv:1610.04717 \[nucl-th\]\].
I. Karpenko and F. Becattini, J. Phys. Conf. Ser. [**779**]{}, no. 1, 012068 (2017) doi:10.1088/1742-6596/779/1/012068 \[arXiv:1611.08136 \[nucl-th\]\].
Y. Xie, R. C. Glastad and L. P. Csernai, Phys. Rev. C [**92**]{} (2015) no.6, 064901 doi:10.1103/PhysRevC.92.064901 \[arXiv:1505.07221 \[nucl-th\]\].
R. h. Fang, J. y. Pang, Q. Wang and X. n. Wang, arXiv:1611.04670 \[nucl-th\].
F. Becattini, L. Csernai and D. J. Wang, “$\Lambda$ polarization in peripheral heavy ion collisions,” Phys. Rev. C [**88**]{}, no. 3, 034905 (2013) \[arXiv:1304.4427 \[nucl-th\]\].
F. Becattini, L. Bucciantini, E. Grossi and L. Tinti, Eur. Phys. J. C [**75**]{}, no. 5, 191 (2015) doi:10.1140/epjc/s10052-015-3384-y \[arXiv:1403.6265 \[hep-th\]\].
B. Betz, M. Gyulassy and G. Torrieri, Phys. Rev. C [**76**]{} (2007) 044901 doi:10.1103/PhysRevC.76.044901 \[arXiv:0708.0035 \[nucl-th\]\]. L. P. Csernai, V. K. Magas and D. J. Wang, Phys. Rev. C [**87**]{} (2013) no.3, 034906 doi:10.1103/PhysRevC.87.034906 \[arXiv:1302.5310 \[nucl-th\]\].
L. P. Csernai, D. J. Wang and T. Csorgo, Phys. Rev. C [**90**]{} (2014) no.2, 024901 doi:10.1103/PhysRevC.90.024901 \[arXiv:1406.1017 \[hep-ph\]\]. G. Prokhorov and O. Teryaev, arXiv:1707.02491 \[hep-th\]. O. Rogachevsky, A. Sorin and O. Teryaev, “Chiral vortaic effect and neutron asymmetries in heavy-ion collisions,” Phys. Rev. C [**82**]{}, 054910 (2010) \[arXiv:1006.1331 \[hep-ph\]\].
J. -H. Gao, Z. -T. Liang, S. Pu, Q. Wang and X. -N. Wang, “Chiral Anomaly and Local Polarization Effect from Quantum Kinetic Approach,” Phys. Rev. Lett. [**109**]{}, 232301 (2012) \[arXiv:1203.0725 \[hep-ph\]\].
T. Kalaydzhyan, “Temperature dependence of the chiral vortical effects,” Phys. Rev. D [**89**]{}, no. 10, 105012 (2014) \[arXiv:1403.1256 \[hep-th\]\].
D.T. Son and P. Surowka, “Hydrodynamics with Triangle Anomalies,” Phys. Rev. Lett. [**103**]{}, 191601 (2009) \[arXiv:0906.5044 \[hep-th\]\]. M. Baznat, K. Gudima, A. Sorin and O. Teryaev, “Helicity separation in Heavy-Ion Collisions,” Phys. Rev. C [**88**]{}, 061901 (2013) \[arXiv:1301.7003 \[nucl-th\]\].
V.D. Toneev, K.K. Gudima, Nucl. Phys. A [**400**]{}, 173c (1983).
V.D. Toneev, N.S. Amelin, K.K. Gudima, S.Yu. Sivoklokov, Nucl. Phys. A [**519**]{}, 463c (1990).
N.S. Amelin, E.F. Staubo, L.S. Csernai et al., Phys.Rev. C [**44**]{}, 1541 (1991).
O. Teryaev and R. Usubov, Phys. Rev. C [**92**]{}, no. 1, 014906 (2015). doi:10.1103/PhysRevC.92.014906 W. Cassing and E. L. Bratkovskaya, Phys. Rept. [**308**]{}, 65 (1999). doi:10.1016/S0370-1573(98)00028-3 Plenary talks of F. Becattini and S. Voloshin at SQM-2017, Utrecht, July 11-15; F. Becattini and I. Karpenko, arXiv:1707.07984 \[nucl-th\].
M. I. Baznat, K. K. Gudima, A. S. Sorin and O. V. Teryaev, Phys. Rev. C [**93**]{}, no. 3, 031902 (2016) doi:10.1103/PhysRevC.93.031902 \[arXiv:1507.04652 \[nucl-th\]\].
K. Fukushima, D.E. Kharzeev and H.J. Warringa, “The Chiral Magnetic Effect,” Phys. Rev. D [**78**]{}, 074033 (2008) \[arXiv:0808.3382 \[hep-ph\]\]. D. Kharzeev and A. Zhitnitsky, “Charge separation induced by P-odd bubbles in QCD matter,” Nucl. Phys. A [**797**]{}, 67 (2007) \[arXiv:0706.1026 \[hep-ph\]\]. A. Sorin and O. Teryaev, Phys. Rev. C [**95**]{}, no. 1, 011902 (2017) doi:10.1103/PhysRevC.95.011902 \[arXiv:1606.08398 \[nucl-th\]\].
K. Landsteiner, E. Megias, L. Melgar and F. Pena-Benitez, “Holographic Gravitational Anomaly and Chiral Vortical Effect,” JHEP [**1109**]{} (2011) 121
V. Braguta, M.N. Chernodub, K. Landsteiner, M.I. Polikarpov, M.V. Ulybyshev Phys.Rev. D88 (2013) 071501 DOI: 10.1103/PhysRevD.88.071501 e-Print: arXiv:1303.6266 \[hep-lat\]; V. Braguta, M.N. Chernodub, V.A. Goy, K. Landsteiner, A.V. Molochkov, M.I. Polikarpov, Phys.Rev. D89 (2014) no.7, 074510 DOI: 10.1103/PhysRevD.89.074510 e-Print: arXiv:1401.8095 \[hep-lat\]; V. Braguta, M. N. Chernodub, V. A. Goy, K. Landsteiner, A. V. Molochkov and M. Ulybyshev, “Study of axial magnetic effect,” AIP Conf. Proc. [**1701**]{}, 030002 (2016); doi:10.1063/1.4938608; V. Goy, “Investigation of SU(2) gluodynamics in the framework of the latice approach”, PhD Thesis (in Russian), Vladivostok, 2015.
P. V. Buividovich, “Axial Magnetic Effect and Chiral Vortical Effect with free lattice chiral fermions,” J. Phys. Conf. Ser. [**607**]{}, no. 1, 012018 (2015) doi:10.1088/1742-6596/607/1/012018 \[arXiv:1309.4966 \[hep-lat\]\]. O. V. Teryaev and V. I. Zakharov, “From the chiral vortical effect to polarization of baryons: A model,” Phys. Rev. D [**96**]{}, no. 9, 096023 (2017). doi:10.1103/PhysRevD.96.096023
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The origin of GRBs have been a mystery for almost 30 years. The afterglow observed in the last few years enabled redshift determination for a handful of bursts, and the cosmological origin is now firmly established. Though the distance scale is settled, there still remains orders of magnitude uncertainty in their rate and in the total energy that is released in the explosion due to the possibility that the emission is not spherical but jet-like. Contrary to the GRB itself, the afterglow can be measured up to months and even years after the burst, and it can provide crucial information on the geometry of the ejecta. We review the theory of afterglow from jets and discuss the evidence that at least some of the bursts are not spherical. We discuss the prospects of polarization measurements, and show that this is a powerful tool in constraining the geometry of the explosion.'
address: 'Theoretical Astrophysics 130-33, California institute of Technology, Pasadena CA 91125'
author:
- 'Re’em Sari'
title: Beaming and Jets in Gamma Ray Bursts
---
Jets? - A fundamental question
==============================
The study of $\gamma$-ray bursts was revolutionized when the Italian-Dutch satellite BeppoSAX delivered arcminutes positioning of some GRBs, within a few hours after the event. This enabled other ground and space instruments to monitor the relatively narrow error boxes. Emission in X-ray, infrared, optical and radio, so called “afterglow”, was observed by now for more than a dozen of bursts.
The current understanding of the GRBs phenomenon is that a compact source emits relativistic flow with Lorentz factor $\gamma$ of at least a few hundreds. This flow emits, probably by internal shocks (see e.g. [@SP97; @FMN96]), the GRB. After these internal shocks have produced the GRB, the ultra relativistic flow interacts with the surrounding medium and decelerates. Synchrotron radiation is emitted by the heated surrounding matter. As more and more of the surrounding mass is accumulating, the flow decelerates and the emission shifts to lower and lower frequencies. Excitingly, the afterglow theory is relatively simple. It deals with the emission on timescales much longer than those of the GRBs. The details of the complex initial conditions are therefore forgotten and the evolution depends only on a small number of parameters.
We begin by clarifying some of the confusing terminology. There are two distinct, but related, effects. The first, [**“jets”**]{}, describes scenarios in which the relativistic flow emitted from the source is not isotropic but collimated towards a finite solid angle. The term jet refers to the geometrical shape of the relativistic flow emitted from the inner engine. The second effect is that of [**“relativistic beaming”**]{}. The radiation from any object that is radiating isotropically in its own rest frame, but moving with a large Lorentz factor $\gamma$ in the observer frame, is collimated into a small angle $1/\gamma$ around its direction of motion. This is an effect of special relativity. It has nothing to do with the ejecta’s geometry (spherical or jet) but only with the fact that the ejecta is moving relativisticly. The effect of relativistic beaming allows an observer to see only a small angular extent, of size $1/\gamma$ centered around the line of sight. Unfortunately, the term beaming was also used for “jets” by many authors (including myself). We will keep a clear distinction between the two in this paper. Since we know the flow is ultra-relativistic (initially $\gamma>100$), there is no question that the relativistic beaming is always relevant for GRBs. The question we are interested in is that of the existence of “jets”.
The idealized description of a jet is a flow that occupies only a conical volume with half opening angle $\theta_0$. In fact the relativistic dynamics is such that the width of the matter in the direction of its propagation is much smaller than its distance from the source by a factor of $1/\gamma^2$. The flow, therefore, does not fill the whole cone. Instead it occupies only a thin disk at its base, looking more like a flying pancake [@P99] - see figure \[grbjets\]. If the “inner engine” emits two such jets in opposite directions then the total solid angle towards which the flow is emitted is $\Omega=2\pi \theta_0^2$. The question whether the relativistic flow is in the form of a jet or a sphere has three important implications.
[**The Total Emitted Energy.**]{} Optical observations of afterglows enabled redshift determination, and therefore a reasonably accurate estimate of the distance, $D$, to these events (the uncertainty is now in the cosmological parameters of the universe). The so called “isotropic energy” can then be inferred from the fluence $F$ (the total observed energy per unit area at earth) as $E_{iso}=4\pi D^2 F$ (taking cosmological corrections into account, $D=D_L/\sqrt{1+z}$ where $D_L$ is the luminosity distance and $z$ is the redshift). The numbers obtained in this way range from $10^{51}$erg to $10^{54}$erg with the record of $3\times 10^{54}$erg held by the famous GRB 990123. These huge numbers approach the equivalent energy of a solar mass, all emitted in a few tens of seconds!
These calculations assumed that the source emitted the same amount of energy towards all directions. If instead the emission is confined to some solid angle $\Omega$ then the true energy is $E=\Omega D^2 F$. As we show later $\Omega$ is very weakly constrained by the GRB itself and can be as low as $10^{-6}$. If so the true energy in each burst $E \ll E_{iso}$. We will show later that interpretation of the multi-wavelength afterglow lightcurves indeed indicates that some bursts are jets with solid angles considerably less than $4 \pi$. The isotropic energy estimates may be fooling us by a few orders of magnitudes! Clearly this is of fundamental importance when considering models for the sources of GRBs.
[**The Event Rate.**]{} BATSE sees about one burst per day. With a few redshifts measured this translates to about $10^{-7}$ bursts per year per galaxy. However, if the emission is collimated to $\Omega \ll 4\pi$ then we do not see most of the events. The true event rate is then larger than that measured by BATSE by a factor of $4\pi /\Omega$. Again this is of fundamental importance. Clearly, the corrected GRB event rate must not exceeds that of compact binary mergers or the birth rate of massive stars if these are to produce the majority of the observed GRBs.
[**The Physical Ejection Mechanism.**]{} Clearly, different physical models are needed to explain collimated and isotropic emission. For example, in the collapsar model (e.g. [@MW99]), relativistic ejecta that is capable of producing a GRB is produced only around the rotation axis of the collapsing star with half opening angle of about $\theta_0 \cong 0.1$. Such models would have difficulties to explain isotropic bursts as well as very narrow jets.
With these uncertainties we are therefore left with huge ignorance in how, how much and how many GRBs are produces. The question as to whether the emission of GRBs is spherical or collimated in jets is fundamental to almost all aspects of the GRB phenomenon.
Afterglow Spectrum - Basic Theory
=================================
When the ejecta interacts with the surrounding medium, a shock waves (so called the forward shock) is going through the cold ambient medium and heating it up to relativistic temperatures. The basic afterglow model assumes that electrons are accelerated by the shock into a powerlaw distribution of their Lorentz factor $\gamma_e$: $N(\gamma
_{e})\sim \gamma _{e}^{-p}$ for $\gamma _{e}>\gamma _{m}$. The lower cutoff of this distribution is assumed to be a fixed fraction of equipartition. It is also assumed that a considerable magnetic field is being built behind the shock, again characterized by a certain fraction of equipartition. The relativistic electrons then emit synchrotron radiation and produce the observed afterglow. The broad band spectrum of such emission was given by Sari, Piran & Narayan [@SPN98] (see figure \[SPNspec\]).
At each instant, there are three characteristic frequencies: (I) $\nu
_{m}$ which is the synchrotron frequency of the minimal energy electron, having a Lorentz factor $\gamma _{m}$. (II) The cooling time of an electron is inverse proportional to its Lorentz factor $\gamma
_{e}$. Therefore, electrons with a Lorentz factor higher than some critical value $\gamma _{e}>\gamma _{c}$ can cool on the dynamical timescale of the system. This characteristic Lorentz factor corresponds to the “cooling frequency” $\nu _{c}$. (III) Below some critical frequency $\nu _{a}$ the flux is self absorbed and is given by the Rayleigh-Jeans portion of a black body spectrum. The broad band spectrum of the well studied GRB 970508 [@G+98] is in very good agreement with the theoretical picture.
We stress that the spectrum given above is quite robust. The only assumption is synchrotron radiation from a powerlaw distribution of relativistic electrons. The same spectrum will hold whether the shocks propagates into a constant density interstellar medium or a decreasing surrounding density produced earlier by the progenitor’s wind. It will be valid whether the ejecta is spherical or jet-like, whether the equipartition parameters are constant with time or not.
On the contrary, the temporal evolution of the spectrum is more subtle. The simplest evolution, which well describes the data of some bursts, is the spherical adiabatic model with a constant density ambient medium. In this scenario, $\gamma \sim R^{-3/2}$ or in terms of the observer time, $t=R/\gamma^2c$, $\gamma \sim
t^{-3/8}$. Given the evolution of $\gamma(t)$ one can derive the temporal evolution of the break frequencies and the results are indicated in figure \[SPNspec\]. The peak flux, in the adiabatic, spherical constant ambient density model is constant with time.
Hydrodynamics of Jets
=====================
Interestingly, due to the effect of relativistic beaming (which is independent of jets) we are only able to see an angular extent of $1/\gamma <0.01$ during the GRB itself where the Lorentz factor $\gamma
>100$. Moreover, it is only regions of size $1/\gamma$ that are causally connected. Therefore, each fluid element evolves as if it is part of a sphere as long as $1/\gamma<\theta_0$. Combining these two facts, we cannot distinguish a jet from spherical ejecta as long as $1/\gamma<\theta_0$.
However, as the afterglow evolves, $\gamma$ decreases and it will eventually fall below the initial inverse opening angle of the jet. The observer will notice that some of the sphere is missing from the fact that less radiation is observed. This effect alone, will produce a significant break, steepening the lightcurve decay by a factor of $\gamma^2 \sim
t^{-3/4}$ even if the dynamics of each fluid element has not changed. The transition should occur at the time $t_{jet}$ when $1/\gamma \cong \theta_0$. Observing this time can therefore provide an estimate of the jet’s opening angle according to $$\label{t_jet}
t_{{\rm jet}}\approx 6.2 (E_{52}/n_{1})^{1/3}(\theta _{0}/0.1)^{8/3}
{\rm hr}.$$
Additionally, Rhoads [@R99] has shown that at about the same time (see however [@PM99; @MR99; @MSB99]), the jet will begin to spread laterally so that its opening angle $\theta (t\grave{)}\sim 1/\gamma$. The ejecta now encounters more surrounding matter and decelerates faster than in the spherical case. The Lorentz factor now decays exponentially with the radius and as $\gamma\sim t^{-1/2}$ with observed time. Taking this into account, the observed break is even more significant. The slow cooling spectrum given in figure \[SPNspec\] evolves now with decreasing peak flux $F_{\nu,m} \sim t^{-1}$ and the break frequencies evolve as $\nu_m \sim t^{-2}$, $\nu_c \sim t^0$ and $\nu_a \sim t^{-1/5}$. This translate to a temporal decay in a given frequency as listed in table \[t:afterglow\].
------------------------------ ---------------------------------------- ----------------------------- ---------------------
spectral index
$\beta$, $F_{\nu}\propto \nu^{-\beta}$ sphere jet
$\nu<\nu_a$ $\beta=-2$ $\alpha=-1/2$ $\alpha=0$
$\nu_a<\nu<\nu_m$ $\beta=-1/3$ $\alpha=-1/2$ $\alpha=1/3$
$\alpha=3(p-1)/4\cong 1.05$ $\alpha=p\cong 2.4$
\[0pt\][$\nu_m<\nu<\nu_c$]{} \[0pt\][$(p-1)/2 $\alpha=3\beta/2$ $\alpha=2\beta+1$
\cong0.7$]{}
$\alpha=(3p-2)/4\cong 1.3$ $\alpha=p\cong 2.4$
\[0pt\][$\nu>\nu_c$]{} \[0pt\][$ p/2 $\alpha=3\beta/2-1/2$ $\alpha=2\beta$
\cong1.2$]{}
------------------------------ ---------------------------------------- ----------------------------- ---------------------
: The spectral index $\beta$ and the temporal index $\alpha$ as function of $p$ for a spherical and a jet-like evolution. Typical values are quoted using $p=2.4$. The parameter free relation between $\alpha$ and $\beta$ is given for each case (eliminating $p$). The difference in $\alpha$ between a jet and a sphere is always substantial at all frequencies.[]{data-label="t:afterglow"}
The jet break is a hydrodynamic one. It should therefore appear at the same time at all frequencies - an achromatic break. Though an achromatic break is considered to be a strong signature of a jet, one should keep in mind that any other hydrodynamic transition will also produce an achromatic break. To name a few: the transition from relativistic to non-relativistic dynamics, a jump in the ambient density or the supply of new energy from slower shells that catch up with the decelerated flow. However, the breaks produced by the transition from a spherical like evolution (when $1/\gamma<\theta_0$) to a spreading jet has a well defined prediction for the change in the temporal decay indices. The amount of break depends on the spectral regime that is observed. It can be seen from table \[t:afterglow\] that the break is substantial $\Delta
\alpha >0.5$ in all regimes and should be easily identified.
Finally we note that if jet’s opening angle is of order unity, the total energy may still be about an order of magnitude lower than the isortropic estimate. However, in this case the break will be “hidden” as it will overlap the transition to non-relativistic dynamics. It was suggested that this is the case for GRB 970508 [@FWK99]
Observational Evidence for Jets
===============================
Evidence of a break from a shallow to a steep power law was first seen in GRB 990123 [@K+99a; @F+99]. Unfortunately the break was observed only in one optical band while the infrared data was ambiguous. Yet, the strongest evidence for this burst being a jet does not come from this optical break but rather from radio observations, as explained below. A famous and exciting event this year was the first detection of a bright (9th magnitude) optical emission simultaneous with GRB 990123 [@A99]. Another new ingredient in GRB 990123 is a radio flare [@K+99b]. Contrary to previous afterglows, where the radio peaks around few weeks and then decays slowly, this burst had a fast rising flare, peaking around a day and then decaying quickly. Sari and Piran [@SP99c] have shown that the bright optical flash and the radio flare are related. Within a day the emission from the adiabatically cooling ejecta, that produced the $60$s optical flash shifts into the radio frequencies. Given this interpretation, the regular forward shock emission should have come later, on the usuall few weeks timescale. The fact that this “usual” forward shock radio emission did not show up is in agreement with the interpretation of this burst as a “jet” which causes the emission to considerably weaken by the time the typical frequency $\nu_m$ arrives to radio frequencies.
GRB 990510 had a very clear break simultaneously in all optical bands and in radio [@S+99; @H+99]. In GRB 990123 and GRB 990510 the transition times were about $2.1$ days and $1.2$ days reducing the isotropic energy estimate by a factor of $\sim 200$ and $\sim 300$, respectively. The total energy is now well below a solar rest mass!
Sari, Piran & Halpern [@SPH99] have noted that the observed decays in GRB afterglows that do not show a break are either of a shallow slope of $\sim t^{-1.2}$ or a very steep slope of $\sim
t^{-2.1}$. They argued that the rapidly decaying bursts are those in which the ejecta was a narrow jet and the break in the light curve was before the earliest observation. Interestingly, evidence for jets are found when the inferred energy $E_{iso}$ (which does not take jets into account) is the largest. This implies that jets may account for a considerable fraction of the wide luminosity distribution seen in GRBs, and that the true energy distribution is less wide than it seems to be.
An alternative explanation for these afterglows with fast decline is propagation into a medium with decreasing density, i.e. a wind produced earlier by the progenitor [@CL99]. We favor the jet interpretation for two reasons: (I) decreasing density only enhance the decay by $t^{-1/2}$ for $\nu_m<\nu<\nu_c$ and does not enhance the decay at all for $\nu>\nu_c$ (with typical parameters the optical and certainly the x-ray bands are above $\nu_c$). The rest of the needed effect, in the wind interpretation, is associated with a higher value of the electron powerlaw distribution index $p$ ($p \cong 3$ instead of $p\cong 2.2-2.4$). Why should the value of $p$ be different for shocks propagating into winds? With the jet interpretation one can explain all afterglows with a single value of $p$, as in [@SPH99]. (II) The jets interpretation makes the luminosity distribution of GRBs more narrow, since evidence for jets are found in bright events. Clearly, these are circumstantial evidence. A more clear cut between these two possible interpretations can be done with the use of early afterglow observation, preferably at radio frequencies (see [@FKS+99]).
In summary, there are several kind of afterglows:
[**Shallow decline**]{} $\sim t^{-1.2}$ for as long as the afterglow can be observed. These are probably spherical or at least have a large opening angle (e.g. GRB 970508).
[**Fast decline**]{} $\sim t^{-2.1}$ (e.g. GRB 980519 and GRB 980326). These are either narrow jets, in which the break was very early or they have high values of $p$ and propagate into decreasing density medium.
[**Breaks**]{}: Initially slow decline that changes into a fast decline. These are the best candidates for jets (e.g. GRB 990510).
Polarization - A promising tool
===============================
An exciting possibility to further constrain the models and obtain a more direct proof of the geometrical picture of “jets” is to measure linear polarization. High levels of linear polarization are usually the smoking gun of synchrotron radiation. The direction of the polarization is perpendicular to the magnetic field and can be as high as $70\%$. Gruzinov & Waxman and Medvedev & Loeb [@GW99; @ML99] considered the emission from spherical ejecta which by symmetry should produce no polarization on the average, except for fluctuations of order of a few percent. Polarization is more natural if the ejecta is a “jet” and the line of sight from the observer is with in the jet but does not coincide with its axis. In this case, the spherical symmetry is broken [@G99; @GL99; @S99], and the natural polarization produced by synchrotron radiation should not vanish. For simplicity, lets assume that the magnetic field behind the shock is directed along the shock’s plane (the results hold more generally, unless the magnetic field has no preferred direction). The synchrotron polarization from each part of the shock front, which is perpendicular to the magnetic field, is therefore directed radially.
As long as the relativistic beaming angle $1/\gamma$ is narrower than the physical size of the jet $\theta_0$, one is able to see a full ring and therefore the radial polarization averages to zero (the first frame, with $\gamma\theta_0=4$ of the left plot in figure \[polfig\]). As the flow decelerates, the relativistic beaming $1/\gamma$ becomes comparable to $\theta_0$ and only a part of the ring is visible; net polarization is then observed. Note that due to the radial direction of the polarization from each fluid element, the total polarization is maximal when a quarter ($\gamma\theta_0=2$ in figure \[polfig\]) or when three quarters ($\gamma\theta_0=1$ in figure \[polfig\]) of the ring are missing (or radiate less efficiently) and vanishes for a full and for half ring. The polarization when more than half of the ring is missing is perpendicular to the polarization direction when less than half of it is missing.
At late stages the jet expands and since the offset of the observer from the physical center of the jet is constant, spherical symmetry is regained. The vanishing and re-occurrence of significant parts of the ring results in a unique prediction: there should be three peaks of polarization, with the polarization position angle during the central peak rotated by $90^{\circ }$ with respect to the other two peaks. In case the observer is very close to the center, more than half of the ring is always observed, and therefore only a single direction of polarization is expected. A few possible polarization light curve are presented in figure \[polfig\].
Summary
=======
Now when redshifts for GRBs are routinely measured, the largest uncertainty in their energy budget and event rate is the possibility that the emission is not spherical but jet-like. We discussed the theory of afterglow from jet-like event. These should produce a substantial break at all frequencies. The time where this break occurs is an indication of the jets opening angle. GRB 990510 seems to be a perfect example for this behavior. The inferred opening angle is about $0.1$ consistent with upper limits from searches of orphan X-ray afterglows [@GHV+99]. Several other candidate for jets are bursts with fast decline, where the break presumably took place before the earliest observation. This question will be settled when more frequent early observations are available. We have shown that afterglow from jets should show a unique signature of polarization, at detectable levels. Observing such signature will confirm the jet interpretation and the synchrotron model in general.
[**Acknowledgements**]{} I thank Titus Galama for very useful comments, and the Sherman Fairchild foundation for support.
MacFadyen, A. I., Woosley, S. E. 1999, ApJ, 526, 152.
Sari, R., & Piran T. 1997, ApJ, 485, 270
Fenimore, E. E., Madras, C., & Nayakshine, S. 1996, ApJ, 473, 998
Piran, T. 1999, in the proceedings of the Gräft[å]{}vallen workshop ‘Gamma Ray Bursts: The First Three Minutes’, Ed. Juri Poutanen.
Sari, R., Piran, T. & Narayan, R. 1998, 497, L17
Galama, T. J. et al. 1998, ApJ, 500, 101
Rhoads, J. E. 1999, ApJ, 525, 737
Panaitescu, A. & Mészáros, P., 1999, ApJ, 503, 314
Mészáros, P., & Rees M. J. 1999, MNRAS, 299, L10
Moderski, R., Sikora, M., Bulik, T. 1999, astro-ph/9904310
Frail, D. A., Waxman, E. & Kulkarni, S. R. 1999, astro-ph/9910319
Kulkarni, S. R., et al. 1999, Nature, 398, 389
Fruchter, A. S., et al., 1999, ApJ, 519, L13
Akerlof, C. et al., 1999, Nature. 398, 400
Kulkarni, S. R., et al. 1999, ApJ, 522, L97
Sari, R., & Piran T. 1999c, ApJ, 517, L109
Stanek, K. Z., Garnavich, P. M., Kaluzny, J., Pych, W. & Thompson, I. 1999, ApJ, 522, L39
Harrison F. A., et al. 1999, ApJ, 1999, 523, L121
Sari, R., Piran, T., & Halpern, J. 1999, ApJ, 519, L17
Galama, T. J. et al. 1999, Nature, 398, 394.
Chevalier, R. A. & Li, Z. Y., 2000, ApJ in press, astro-ph/9908272
Frail, D. A. et al. 1999, astro-ph/9910060
Gruzinov A., & Waxman E., 1999, ApJ, 511, 852
Medvedev, M. V., & Loeb A., 1999, astro-ph/9904363
Gruzinov A. 1999, ApJ, 525, L29
Ghisellini, G., & Lazzati, D., 1999, MNRAS, 309, L7
Sari, R. 1999, ApJ, 524, L43
Greiner et al., these proceedings.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We test several BFKL-like evolution equations for unintegrated gluon distributions against forward-central dijet production at LHC. Our study is based on fitting the evolution scenarios to the LHC data using the high energy factorization approach. Thus, as a by-product, we obtain a set of LHC-motivated unintegrated gluon distributions ready to use. We utilize this application by calculating azimuthal decorrelations for forward-central dijet production and compare with existing data.'
author:
- |
Piotr Kotko$^1$, Wojciech Słomiński$^2$ and Dawid Toton$^3$\
\
$^1$ [*Department of Physics, Penn State University,*]{}\
[*University Park, 16803 PA, USA*]{}\
\
$^2$ [*The M. Smoluchowski Institute of Physics, Jagiellonian University,*]{}\
[*S. Łojasiewicza 11, 30-348 Kraków, Poland*]{}\
\
$^3$ [*The H. Niewodniczański Institute of Nuclear Physics, Polish Academy of Sciences,*]{}\
[*Radzikowskiego 152, 31-342 Kraków, Poland*]{}
bibliography:
- 'library.bib'
title: Unintegrated gluon distributions for forward jets at LHC
---
Introduction
============
A typical procedure in applying QCD to hadronic collisions relies on factorization theorems. They consist in two ingredients: a perturbatively calculable hard part and a nonperturbative piece parametrizing hadrons participating in a collision. The most known and tested is the collinear factorization (see e.g. [@Collins:2011zzd] for a review), which applies for a variety of processes, including jet observables in deep inelastic scattering (DIS) and hadron-hadron collisions. Here, the nonperturbative component is parametrized in terms of parton distribution functions (PDFs) which undergo Dokshitzer-Gribov-Lipatov-Altarelli-Parisi (DGLAP) evolution equations. The key feature of PDFs is the *universality*, i.e. the PDFs that are measured in one process can be used in any other for which the factorization holds. Therefore, for instance one can use PDFs fitted to DIS structure functions and use them to make predictions for jets in hadron-hadron collisions. Although the collinear factorization is powerful and well-tested, it is supposed that for certain observables, e.g. forward jets at high energies, another kind of evolution equations for the PDFs is needed. Namely, the perturbative calculations contain the logarithms of the form $\alpha_{s}\log(1/x)$, where $x$ is the longitudinal fraction of the hadron momentum carried by the parton. At high energies and forward rapidities $x$ is small and these logarithms need to be resumed. This is accomplished by means of various “small $x$” evolution equations, which essentially are various extensions of the pioneering Balitski-Fadin-Kuraev-Lipatov (BFKL) evolution equation (see e.g. [@Lipatov:1996ts]). In the small $x$ domain the transverse momenta of the partons exchanged between the perturbative and nonperturbative parts are not suppressed comparing to the collinear factorization. Therefore, the PDFs have an explicit dependence on the transverse momentum of a parton. Such objects are often referred to, as transverse momentum dependent PDFs (TMDs) or Unintegrated PDFs, although the former are typically used outside the small $x$ physics, and posses unambiguous (though in general process dependent) field theoretic definitions. Actually, at small $x$ one usually deals with initial state gluons only, and thus the object of interest in this paper is an Unintegrated Gluon Distribution (UGD). The UGDs have to be convoluted with a perturbative “hard part” according to so-called $k_{T}$ or High Energy Factorization (HEF). We describe this approach in some more detail in Section \[sec:HEF\]. Here, let us just mention that unlike the collinear factorization, the HEF is not a QCD theorem and actually the universality of UGDs is supposed to be violated for jet production in hadron-hadron collisions. Thus, in principle, the standard procedure of fitting the UGDs to the $F_{2}$ HERA data and using it for jets in hadron-hadron collisions is not correct, but there are no quantitative measures of the factorization violation so far. Actually, HEF is surprisingly quite successful with describing LHC data using UGDs from fits to structure functions, see for instance [@vanHameren:2014ala]. At present, there are several fits to $F_{2}$ data using different small $x$ approaches, see [@Ellis:2008yp; @Ross:2011zzb; @Kutak:2012rf; @Lipatov:2013yra] for more details.
In the present work we undertake another path. We make an attempt to fit various BFKL-like UGDs directly to the LHC data for jet forward jet production. It has a twofold purpose. First, we have an opportunity to explore UGDs using relatively exclusive observables. Second, we want to free ourselves from the aforementioned universality problem when transferring UGDs from DIS to the LHC domain. We consider two separate measurements: jet transverse momentum spectra [@Chatrchyan2012] in forward-central jet production and forward-central dijet decorrelations [@CMS:2014oma]. The first measurement consists of two separate sets of data: for the forward jet and for the central jet. Thus, the mutual description of both spectra imposes a strong constraint on the UGDs and we shall use this measurements to make our fits. The second measurement will be used to test the fits.
The paper is organized as follows. In Section \[sec:HEF\] we describe the approach of HEF. The small $x$ evolution equations with various components incorporating sub-leading effects are discussed in Section \[sec:evolution\_eqs\]. The fitting procedure and the software used are described in Section \[sec:Procedure\]. We give the results in Section \[sec:Results\]. Having the fits, we test them against recent forward-central dijet decorrelations data in Section \[sec:decorrelations\]. Finally, we discuss our research in Section \[sec:Summary\].
High Energy Factorization {#sec:HEF}
=========================
In this introductory section we discuss in more detail issues concerning factorization at small $x$. This task is somewhat complicated, notably because of the various existing approaches and various existing definitions of UGDs.
In the following paper the notion of HEF corresponds to a general class of factorization approaches supposed to be valid at small $x$. Below we list some of the existing realizations:
1. the factorization of Gribov, Levin and Ryskin (GLR) [@Gribov1983] for high-$p_{T}$ inclusive gluon production \[HEFGLR\]
2. the factorization of Catani, Ciafaloni and Hautmann (CCH) [@Catani:1990eg; @Catani:1994sq] for heavy quark production in DIS, photo-production and hadron-hadron collisions \[HEFCCH\]
3. the factorization of Collins and Ellis [@Collins1991] for heavy quark production in hadron-hadron collisions \[HEFCE\]
4. the factorization for inclusive gluon production in the saturation regime for proton-nuclei collisions within the Color Glass Condensate (CGC) approach [@Blaizot:2004wu] and color dipole formalism [@Kovchegov:2001sc; @Nikolaev:2004cu] (the equivalence of both approaches was shown in [@Iancu2004]) \[HEFCGC\]
In these approaches the nonperturbative part is parametrized in terms of UGDs undergoing BFKL evolution (for GRL, CCH, Collins-Ellis factorizations) or nonlinear Balitsky-Kovchegov evolution [@Balitsky:1995ub; @Kovchegov:1999yj] (for CGC). On the other hand, superficially similar objects to UGDs appear in so-called transverse momentum dependent (TMD) factorization and are called TMD PDFs. One should however realize that the enumerated approaches are valid at leading logarithmic approximation, while the TMD factorizations are valid to all orders in the leading twist approximation. Moreover, unlike most of UGDs in the HEF factorizations, the TMD PDFs have precise operator definitions in terms of matrix elements of nonlocal operators. Those definitions require appropriate Wilson lines to be inserted in order to make the definitions gauge invariant and to resum collinear gluons related to final and initial state interactions. These insertions make the TMD PDFs, in general, process dependent and thus non-universal, breaking the principle of factorization (for more details see e.g. [@Bomhof:2006dp; @Mulders:2011zt]). Only for processes with at most two hadrons the TMD factorization is proved to hold to all orders (for example back-to-back single hadron production in DIS or Drell-Yan scattering). The natural question arises whether the non-universality of TMD PDFs transfers to the small $x$ limit. In ref. [@Xiao:2010sp] an explicit arguments were given that this is the case for dilute-dense collisions (actually the arguments hold for so-called “hybrid” factorization – see also below). Moreover it is known from the CGC approach that at really small $x$, i.e. in the saturation regime, the cross sections cannot be described by just dipoles (averages of two Wilson lines), but also higher correlators are needed [@Dominguez:2012ad], what violates the ordinary logic of factorization. However, for the case of back-to-back dijet production in dilute-dense collisions a generalized factorization has been proposed [@Dominguez:2011wm]; that is, the cross section can be given in terms of hard factors and certain universal pieces. Recently, these results were improved to the case of imbalanced dijets [@Kotko:2015ura]. In particular, when the imbalanced transverse momentum is of the order of transverse momenta of the jets the HEF for dijet production can be derived from the dilute limit of the CGC approach.
In the present work we shall constrain ourselves to dijet production in p-p collisions in the linear regime, as the kinematics we are interested in (and where the data exist) do not allow to develop the saturation region. We want to utilize most of the phase space covered by the data, thus we do not constrain ourselves to the back-to-back dijet region analyzed in [@Dominguez:2011wm]. Rather, we shall use the HEF factorization for dijet production. Since this approach is an extension of the CCH formalism, we shall now briefly recall the latter and the required extensions to obtain HEF for dijets. For a direct derivation from CGC approach see [@Kotko:2015ura].
In the CCH high energy factorization, one considers the heavy quark pair produced via the tree-level hard sub-process $g^{*}\left(k_{A}\right)g^{*}\left(k_{B}\right)\rightarrow Q\overline{Q}$ in the axial gauge. The initial state gluons are off-shell and have the momenta of the form $k_{A}=x_A\, p_{A}+k_{TA}$ and $k_{B}=x_B\, p_{B}+k_{TB}$, where $p_{A}$, $p_{B}$ are the momenta of the incoming hadrons and $p_{A}\cdot k_{TA}=p_{B}\cdot k_{TB}=0$. This particular form of the exchanged momenta is a result of the imposed high energy limit. The off-shell gluons have “polarization vectors” that are $p_{A}$ and $p_{B}$ respectively. Thanks to this kinematics the sub-process given by ordinary Feynman diagrams is gauge invariant despite its off-shellness. In CCH approach the factorization formula for heavy quark production reads (see Fig. \[fig:HEF\]A) $$\begin{gathered}
d\sigma_{AB\rightarrow Q\overline{Q}}=\int d^{2}k_{T A}\int\frac{dx_{A}}{x_{A}}\,\int d^{2}k_{T B}\int\frac{dx_{B}}{x_{B}}\,\\
\mathcal{F}_{g^{*}/A}\left(x_{A},k_{T A}\right)\,\mathcal{F}_{g^{*}/B}\left(x_{B},k_{T B}\right)\, d\hat{\sigma}_{g^{*}g^{*}\rightarrow Q\overline{Q}}\left(x_{A},x_{B},k_{T A},k_{T B}\right),\label{eq:HEN_fact_1}\end{gathered}$$ where $d\hat{\sigma}_{g^{*}g^{*}\rightarrow Q\overline{Q}}$ is the partonic cross section build up from the gauge invariant $g^{*}g^{*}\rightarrow Q\overline{Q}$ amplitude and $\mathcal{F}_{g^{*}/A}$, $\mathcal{F}_{g^{*}/B}$ are UGDs for hadrons $A$ and $B$. The contributions with off-shell quarks are suppressed. The UGDs are assumed to undergo the BFKL evolution equations. In Ref. [@Catani:1994sq] it was argued that similar factorization holds to all orders for DIS heavy quark structure function, although the argumentation misses the details comparing to collinear factorization proofs [@Collins:2011zzd], especially the definitions of UGDs and complications arising at higher orders in the axial gauge [@Avsar:2012hj].
In the works [@Deak2010],[@Kutak:2012rf],[@VanHameren2013],[@vanHameren2013a; @vanHameren:2014lna; @vanHameren:2014ala] as well as in this paper the CCH factorization was extended to model the cross section for jet production in hadron-hadron collisions. The first difficulty arises because now one has to consider also gluons in the final state, e.g. $g^{*}g^{*}\rightarrow gg$ sub-process for dijet production. The corresponding amplitude is however not gauge invariant when calculated from ordinary Feynman diagrams. A few approaches have been proposed to calculate a gauge invariant extension of such amplitudes [@vanHameren2012; @vanHameren2013a; @vanHameren:2013csa; @vanHameren:2014iua; @Kotko2014a]. These [*gauge invariant off-shell amplitudes*]{} in fact correspond to a vertex that can be calculated from the well-known Lipatov’s effective action [@Lipatov:1995pn; @Antonov:2004hh] (see Fig. \[fig:HEF\]B). The approaches [@vanHameren2012; @vanHameren2013a; @vanHameren:2013csa; @vanHameren:2014iua; @Kotko2014a] were however oriented on practical and efficient computations of multi-particle off-shell amplitudes using helicity method and computer codes. As stated before, in CCH the UGDs undergo BFKL evolution. In our extensions of CCH approach we allow the UGDs to undergo more complicated evolution equations, which are more suitable for jets. More details will be given in Section \[sec:evolution\_eqs\]. Yet another modification of the CCH formula comes from the fact that the present study concerns the system of dijets where one of the jet is forward, while the second is in the central region. From $2\rightarrow2$ kinematics it follows then, that $x_{A}\ll x_{B}$ (or the opposite), except for the small corner of the phase space. Since $x_{B}$ is typically of the order of $0.5$ the usage of small $x$ evolution for $\mathcal{F}_{g^{*}/B}$ is questionable (this is similar to dilute-dense system considered e.g. in [@Dominguez:2011wm]). Therefore we use collinear approach on the $B$ hadron side [@Deak:2009xt]. Technically, one takes the collinear limit in $d\hat{\sigma}_{g^{*}g^{*}\rightarrow2j}$ by sending $k_{TB}\rightarrow0$ to obtain a sub-process with one off-shell gluon $d\hat{\sigma}_{g^{*}g\rightarrow2j}$ (the off-shell amplitudes have well defined on-shell limit). In this limit one has to take into account also sub-processes with initial state on-shell quarks, $d\hat{\sigma}_{g^{*}q\rightarrow2j}$. The remaining integral over $d^{2}k_{TB}$ gives helicity sum for $B$ partons on one hand, and the integrated (collinear) PDF on the other $\int dk_{B}^{2}\,\mathcal{F}_{a^{*}/B}\left(x_{B},k_{TB}\right)=f_{a}\left(x_{B}\right)$. Thus, the final formula for the factorization model reads $$\begin{gathered}
d\sigma_{AB\rightarrow2j}=\int d^{2}k_{TA}\int\frac{dx_{A}}{x_{A}}\,\int\frac{dx_{B}}{x_{B}}\,\sum_{b}\\
\mathcal{F}_{g^{*}/A}\left(x_{A},k_{T A},\mu\right)\, f_{b}\left(x_{B},\mu\right)\, d\hat{\sigma}_{g^{*}b\rightarrow2j}\left(x_{A},x_{B},k_{T A},\mu\right),\label{eq:HEN_fact_2}\end{gathered}$$ where we have included the hard scale dependence not only in the collinear PDFs $f_{b}$, but in the UGD as well. Such a dependence turns out to be important for certain exclusive observables involving a hard scale (e.g. large $p_{T}$ of jets; see e.g. [@vanHameren:2014ala]). We note, that when the final states become well separated in rapidity, i.e. when the central jet lies in the opposite hemisphere to the forward jet we start to violate our condition $x_{A}\ll x_{B}$ and different approach should be used. The factorization formula (\[eq:HEN\_fact\_2\]) resembles the linearized approach of [@Dominguez:2011wm] but it extends beyond the correlation limit as here the hard sub-processes have injected a nonzero $k_{T}$. As mentioned before, the formula (\[eq:HEN\_fact\_2\]) has been recently derived from the CGC approach in [@Kotko:2015ura].
Small *x* evolution equations {#sec:evolution_eqs}
==============================
Let us now discuss the evolution equations for UGDs which were used in our fits. As described in the preceding section we concentrate on linear evolution equations. Below we list some of them with a short explanation. We consider only gluon UGDs, thus we skip the subscripts in $\mathcal{F}_{g^{*}/A}$.
1. pure BFKL equation\[enu:pure-BFKL-equation\]
The equation in the leading logarithmic approximation reads [@Fadin:1975cb; @Kuraev:1977fs] $$\begin{gathered}
\mathcal{F}\left(x,k_{T}^{2}\right)=\mathcal{F}_{0}\left(x,k_{T}^{2}\right)\\
+\overline{\alpha}_{s}\int_{x}^{1}\frac{dz}{z}\,\int_{0}^{\infty}dq_{T}^{2}\left[\frac{q_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right)-k_{T}^{2}\mathcal{F}\left(\frac{x}{z},k_{T}^{2}\right)}{\left|q_{T}^{2}-k_{T}^{2}\right|}+\frac{k_{T}^{2}\mathcal{F}\left(\frac{x}{z},k_{T}^{2}\right)}{\sqrt{4q_{T}^{4}+k_{T}^{4}}}\right]\label{eq:BFKL}\end{gathered}$$ where $\overline{\alpha}_{s}=N_{c}\alpha_{s}/\pi$ with $N_{c}$ being the number of colors. The initial condition for the evolution is given by $\mathcal{F}_{0}$. The NLO BFKL equation is also known [@Fadin:1998py; @Ciafaloni:1998gs]. One of the drawbacks of the pure BFKL equation comes from the fact that $q_{T}^{2}$ of the gluons emitted along the ladder is unconstrained. Indeed, since in the BFKL regime the virtuality of the exchanged gluons is dominated by the transverse components, the resulting *kinematic constraint* reads [@Andersson:1995jt; @Kwiecinski:1996a] $$q_{T}^{2}<\frac{1-z}{z}\, k_{T}^{2}\approx\frac{1}{z}\, k_{T}^{2}.\label{eq:kinematic_constr}$$ This constraint is also often referred to as the consistency constraint.
2. BFKL with the kinematic constraint (BFKL+C)
To incorporate the consistency constraint one may include the appropriate step function into the real emission part of the BFKL. This operation, actually introduces some higher order corrections into the BFKL equation [@Kwiecinski:1996a]. In addition, one may introduce another class of sub-leading corrections by allowing the strong coupling constant to run with the local scale along the ladder. Finally, one may define the $q_{T}^{2}$ integration region to lie away from the infrared nonperturbative region by separating the $\int_{0}^{k_{T0}^{2}}dq_{T}^{2}$ integration and moving it to the initial condition (the infrared cutoff $k_{T0}^{2}$ is taken to be of the order of $1\,\mathrm{GeV}$). The improved equation reads [@Kwiecinski:1997ee] $$\begin{gathered}
\mathcal{F}\left(x,k_{T}^{2}\right)=\mathcal{F}_{0}\left(x,k_{T}^{2}\right)\\
+\overline{\alpha}_{s}\left(k_{T}^{2}\right)\int_{x}^{1}\frac{dz}{z}\,\int_{k_{T0}^{2}}^{\infty}dq_{T}^{2}\left[\frac{q_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right)\Theta\left(k_{T}^{2}-zq_{T}^{2}\right)-k_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right)}{\left|q_{T}^{2}-k_{T}^{2}\right|}+\frac{k_{T}^{2}\mathcal{F}\left(\frac{x}{z},k_{T}^{2}\right)}{\sqrt{4q_{T}^{4}+k_{T}^{4}}}\right].\label{eq:BFKL_constr}\end{gathered}$$ Recently, it has been studied in the context of the Mueller-Navelet jets, that the energy-momentum conservation violation (which above is cured by a “brute force”) becomes less harmful when full NLO corrections are applied [@Ducloue:2014koa]. The effects of the kinematic constraints in the approximate form (\[eq:kinematic\_constr\]) as well as in the full form have been recently analyzed [@Deak:2015dpa] in the context of the CCFM evolution equation [Ciafaloni:1987ur,Catani:1989yc,Catani:1989sg,CCFMd]{}.
3. BFKL with the kinematic constraint in re-summed form (BFKL+CR)
The equation (\[eq:BFKL\_constr\]) can be casted in yet another form [@Kutak:2011fu] $$\begin{gathered}
\mathcal{F}\left(x,k_{T}^{2}\right)=\tilde{\mathcal{F}}_{0}\left(x,k_{T}^{2}\right)\\
+\overline{\alpha}_{s}\left(k_{T}^{2}\right)\int_{x}^{1}\frac{dz}{z}\,\int_{k_{T0}^{2}}^{\infty}\frac{d^{2}q_{T}}{\pi q_{T}^{2}}\Theta\left(q_{T}^{2}-\mu^{2}\right)\Delta_{R}\left(z,k_{T}^{2},\mu^{2}\right)\mathcal{F}\left(\frac{x}{z},\left|\vec{k}_{T}+\vec{q}_{T}\right|^{2}\right),\label{eq:BFKL+CR}\end{gathered}$$ where $$\Delta_{R}\left(z,k_{T}^{2},\mu^{2}\right)=\exp\left(-\overline{\alpha}_{s}\ln\frac{1}{z}\,\ln\frac{k_{T}^{2}}{\mu^{2}}\right)\label{eq:Regge_ff}$$ is the so-called Regge form factor. This form has been used in Ref. [@Kutak:2011fu] to propose a non-linear extension of the CCFM equation. The scale $\mu$ has been introduced to separate unresolved and resolved emissions in (\[eq:BFKL\_constr\]), i.e. the emissions with $q_{T}^{2}<\mu^{2}$ and $q_{T}^{2}>\mu^{2}$, and further the unresolved part was re-summed to obtain the Regge form factor. Note, that the UGDs undergoing this equation do not explicitly depend on the scale $\mu$ and that the new form of the initial condition has to be used (this is denoted by a tilde sign).
4. BFKL with the kinematic constraint and DGLAP correction (BFKL+CD)
In Ref. [@Kwiecinski:1997ee] yet another improvement of (\[eq:BFKL\]) was proposed. One can make an attempt to account for DGLAP-like behaviour by including the non-singular part of the gluon splitting function (the third term below) $$\begin{gathered}
\mathcal{F}\left(x,k_{T}^{2}\right)=\mathcal{F}_{0}\left(x,k_{T}^{2}\right)\\
+\overline{\alpha}_{s}\left(k_{T}^{2}\right)\int_{x}^{1}\frac{dz}{z}\,\int_{k_{T0}^{2}}^{\infty}dq_{T}^{2}\left[\frac{q_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right)\Theta\left(k_{T}^{2}-zq_{T}^{2}\right)-k_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right)}{\left|q_{T}^{2}-k_{T}^{2}\right|}+\frac{k_{T}^{2}\mathcal{F}\left(\frac{x}{z},k_{T}^{2}\right)}{\sqrt{4q_{T}^{4}+k_{T}^{4}}}\right]\\
+\overline{\alpha}_{s}\left(k_{T}^{2}\right)\int_{x}^{1}\frac{dz}{z}\,\left(\frac{z}{2N_{c}}P_{gg}\left(z\right)-1\right)\int_{k_{T0}^{2}}^{k_{T}^{2}}dq_{T}^{2}\mathcal{F}\left(\frac{x}{z},q_{T}^{2}\right),\label{eq:BFKL+CD}\end{gathered}$$ where $P_{gg}\left(z\right)$ is the standard gluon splitting function. This correction, similar to the kinematic constraint, accounts for certain sub-leading corrections to the BFKL equation.
5. BFKL with DGLAP correction alone\[enu:lastmodel\]
This variant is used to test the significance of the DGLAP term alone.
The above UGDs do not involve any hard scale dependence. For observables involving high-$p_{T}$ jets a presence of large scale $\mu^{2}\sim p_{T}^{2}$ in perturbative calculations would involve additional logarithms of the type $\log\left(\mu^{2}/k_{T}^{2}\right)$ which can spoil the procedure. Therefore a re-summation of those logs is desired and it accounts in hard scale dependence for UGDs, c.f. Eq. (\[eq:HEN\_fact\_2\]). The approach which incorporates both $x$, $k_{T}^{2}$ and $\mu^{2}$ dependence in UGDs is provided for example by the CCFM evolution equation (the code available for a practical use is described for example in [@Hautmann:2014uua]). Another approach, so called KMR (Kimber-Martin-Ryskin) procedure [@Kimber:1999xc; @Kimber:2001sc], takes ordinary PDFs and injects $k_{T}$ dependence via the Sudakov form factor taking care of matching to the BFKL evolution at small $x$. A serious advantage of this procedure is that one can use well known PDF sets, fitted to large data sets. Yet another approach was used in [@vanHameren:2014ala] in therms of so-called “Sudakov resummation model”. This procedure reverts, in a sense, the logic used in the KMR and uses the Sudakov form factor to inject the hard scale dependence instead of $k_{T}$. The procedure is parton-shower-like, i.e. it is applied after the MC events are generated and the cross section is known, and is unitary (i.e. the procedure does not change the total cross section). The advantage is that one may use it on the top of UGDs involving nonlinear effects. The basic idea behind the model is that it assigns the Sudakov probability $P$ for events with given $k_T$ and a hard scale $\mu\sim p_T$. Then, the probability of surviving is $1-P$. For events with small $k_T$ and large $\mu$ the emission probability is $P\sim 1$ and the unitarity of the procedure transfers such events to the region $k_T\sim p_T$. There is one more approach proposed in Ref. [@Kutak:2014wga], similar to the one just described, where analogous procedure is applied at the level of UGDs by fixing its integral over $k_{T}$ (it has an advantage of being independent on any software and one may produce grids for a practical usage). In summary, we may consider the following modifications of UGDs \[enu:pure-BFKL-equation\]-\[enu:lastmodel\]:
1. BFKL with the Sudakov (BFKL+S)
2. BFKL with the kinematic constraint and the Sudakov (BFKL+CS)
3. BFKL with DGLAP correction and the Sudakov (BFKL+DS)
4. BFKL with the kinematic constraint in re-summed form and the Sudakov (BFKL+CRS)
5. BFKL with the kinematic constraint, DGLAP correction and the Sudakov (BFKL+CDS)\[enu:BFKLCDS\]
Unfortunately, as far as fitting of UGDs is considered, the above Sudakov-based models are not suitable. This is because they require the knowledge of an integral (whether it is a cross section or integrated gluon, c.f. [@vanHameren:2014ala] vs [@Kutak:2014wga]) which is unknown at the stage of fitting. In principle, one could try to use the method of successive approximations with the Sudakov model of Ref. [@vanHameren:2014ala]. We shall report on our attempts in Section \[sec:Results\]. There is one more comment in order here. The Sudakov resummation model is very sensitive to the region $k_{T}\lesssim1\,\mathrm{GeV}$ which is not well described by the practical implementations of the equations \[enu:pure-BFKL-equation\]-\[enu:lastmodel\] as they use certain low-$k_{T}$ cut, $k_{T\,0}$. For $k_{T}<k_{T\,0}$ the UGD is typically modelled or extrapolated by a constant value.
Let us now discuss the models for the initial condition $\mathcal{F}_{0}$. In this paper we have tested the following models (in the brackets we give the aliases used below to identify the model):
\[eq:param\]
1. exponential model (EXP)\[enu:Models\_ini\] $$\mathcal{F}_{0}\left(x,k_{T}^{2}\right)=N\, e^{-Ak_{T}^{2}}\left(1-x\right)^{a}\left(1-Dx\right)\label{eq:Model_exp}$$
2. (negative) power-like model with running $\alpha_{s}$ (POW) $$\mathcal{F}_{0}\left(x,k_{T}^{2}\right)=\frac{\overline{\alpha}_{s}\left(k_{T}^{2}\right)}{k_{T}^{2}}\, N
\, x^A
\left(1-x\right)^{a}\label{eq:Model_ask}
(1-Dx)$$
3. DGLAP-based model (Pgg)\[enu:Models\_fini\] $$\mathcal{F}_{0}\left(x,k_{T}^{2}\right)=\frac{\alpha_{s}\left(k_{T}^{2}\right)}{2\pi k_{T}^{2}}\,
\int_{x}^{1}dz
\, P_{gg}\left(z\right)\hat{\mathcal{G}}_{0}\left(x\right),\label{eq:Model_Pgg_1}$$ where $$\hat{\mathcal{G}}_{0}\left(x\right)=N
\, x^A
\left(1-x\right)^{a}\left(1-Dx\right)\label{eq:Model_Pgg_2}$$ is a model for an integrated gluon density.
The parameters $N$, $A$, $a$, $D$ are, in general, free parameters and need to be fitted.
We see that in principle there are quite a few variants to be fitted. Though not all of the combinations make sense, we are still left with several scenarios to be tested.
Fitting procedure {#sec:Procedure}
=================
We have used two data samples measured by CMS detector [@Chatrchyan2012] for inclusive forward-central dijet production at CM energy $\sqrt{s}=7\,\mathrm{TeV}$. The central jet is defined to lie within the pseudo-rapidity interval $\left|\eta_{c}\right|<2.8$ while the forward has to lie within $4.9>\left|\eta_{f}\right|>3.2$. Both jets are high-$p_{T}$ jets with $p_{T}>35\,\mathrm{GeV}$. The jets were reconstructed using anti-$k_{T}$ algorithm with radius $R=0.5$. The data samples consist in jet $p_{T}$ spectra for forward and for central jets, $d\sigma_{S}/dp_{T}\Delta\eta_{S}$ with $S=f,c$. There are in total 12 data bins for both forward and central jets.
We have applied the following fitting procedure. For each existing experimental data bin $B$ we produce a 2-dimensional normalized histogram $\mathcal{H}^{B}$ with bins in $x$ and $k_{T}$, such that the cross section can be calculated as $$\sigma^{B}=\sum_{i,j}\mathcal{H}_{ij}^{B}\mathcal{F}\left(x\left(i\right),k_{T}\left(j\right)\right),\label{eq:Fit1}$$ where $i,j$ enumerate the bins in $\left(x,k_{T}\right)$. To make the histograms $\mathcal{H}^{B}$ we
1. generate Monte Carlo events for the process under consideration with $\mathcal{F}=\mathcal{F}^{*}$, where $\mathcal{F}^{*}$ is a relatively “broad” trial UGD (evolving according to one of the scenarios \[enu:pure-BFKL-equation\]-\[enu:BFKLCDS\]),
2. make histograms $\mathfrak{h}^{B}$ in $\left(x,k_{T}\right)$ of contributions to each data bin $B$,
3. divide by $\mathcal{F}^{*}\left(x,k_{T}\right)$, i.e. $\mathcal{H}_{ij}^{B}=\mathfrak{h}_{ij}^{B}/\mathcal{F}^{*}\left(x\left(i\right),k_{T}\left(j\right)\right)$.
Hence, in principle $\mathcal{H}^{B}$ are independent of $\mathcal{F}^{*}$ used for their generation and are calculated only once. This is advantageous, as the hard cross section calculation is costly in CPU time. The latter is calculated using the Monte Carlo C++ program $\mathtt{LxJet}$ [@Kotko2013a] implementing (\[eq:HEN\_fact\_2\]). The generated events (weighted or unweighted) are stored in a $\mathtt{ROOT}$ [@Brun:1997pa] file for further processing. For the UGD evolution according to scenarios \[enu:pure-BFKL-equation\]-\[enu:lastmodel\] we solve the corresponding integral equations by a straightforward numerical iteration over a grid over $x$ and $k_T$.
In order to make the fitting feasible, we need a fast routine to calculate $\mathcal{F}$ used in (\[eq:Fit1\]) for the cross section calculation. However, since our numerical procedure is too slow for that, we prepare grids over which we can interpolate the fitting parameters. Each such grid corresponds to a particular parametrization model and arguments range. Out of four parameters ($N$, $A$, $a$, $D$) of the initial conditions, we fix $D=0$ (see Sec. \[sec:Results\]). Moreover, we note that the solution for $\mathcal{F}$ is linear in $N$. Thus the actual grids are in $A$ and $a$.
Results {#sec:Results}
=======
We have applied the procedure described in the preceding section to most of the models \[enu:pure-BFKL-equation\]-\[enu:BFKLCDS\] and initial conditions \[enu:Models\_ini\]-\[enu:Models\_fini\]. The best values of [$\chi^{2}\!/\mathrm{NDP}$]{}($\chi^2$ per data point) are listed in Table \[tab:chi2\] for models \[enu:pure-BFKL-equation\]-\[enu:lastmodel\]. Note, that some of the scenarios were unable to describe the data, in particular the pure BFKL and BFKL with the kinematic constraint only. Evidently, the DGLAP correction is essential. The fitted values of the parameters of the initial conditions, $N$, $A$, $a$, for scenarios with ${\ensuremath{\chi^{2}\!/\mathrm{NDP}}\xspace}<2$ are collected in Table \[tab:params\]. The fits are presented in Figs. \[fig:central\]-\[fig:forward\]. For a better comparison we also plot the cross-sections scaled by $p_T^5$. We observe that all the models with the DGLAP correction give excellent description of the central-jet data, while the $p_T$ spectrum of forward jets is reasonably reproduced though less accurately. We also note that the models with lowest $\chi^2$ result in very similar predictions for the $p_T$ spectra.
Our attempts to fit the scenarios with the Sudakov resummation can be summarized as follows. First, we observe that the model has a small overall effect on the $p_T$ spectra, although it slightly shifts the theory points away from the data points. We illustrate this in Fig. \[fig:sudeffect\], where we applied the Sudakov model on the top of the events obtained with one of the fits. When we now try to refit the $\mathcal{F}_0$ parameters, we change the total cross section (used already to apply the resummation) and the fit fails. Although we observe that the successive iterations improve the fit, the procedure turns out to be insufficient to make a reliable fit with the Sudakov resummation.
$\mathcal{F}_{0}$ BFKL BFKL+C BFKL+D BFKL+CD BFKL+CR
------------------- ------ -------- -------- --------- ---------
EXP 2.4 2.2 1.24 1.11 1.52
POW 2.3 1.9 1.02 1.12
Pgg – – 1.13 1.11
: The values of [$\chi^{2}\!/\mathrm{NDP}$]{}for fits of unintegrated gluon density evolving according to various models described in Section \[sec:evolution\_eqs\]. The first column lists the initial condition ansatz, see also Section \[sec:evolution\_eqs\] for details.\[tab:chi2\]
model $N$ $A$ $a$
--------------- --------- --------- ---------
BFKL+CR (EXP) $0.095$ $0.012$ $0^*$
BFKL+D (EXP) $0.37$ $0.18$ $0.5^*$
BFKL+CD (EXP) $0.68$ $0.14$ $2.5^*$
BFKL+C (POW) $320$ $1.4$ $61.0$
BFKL+D (POW) $12.7$ $0.5^*$ $5.7$
BFKL+CD (POW) $562$ $0.96$ $35.7$
BFKL+D (Pgg) $106$ $1.2$ $2.5$
BFKL+CD (Pgg) $628$ $2.9$ $5.7$
: The values of initial condition \[enu:Models\_ini\]-\[enu:Models\_fini\] parameters obtained from the fits to the CMS data. We list only the scenarios with ${\ensuremath{\chi^{2}\!/\mathrm{NDP}}\xspace}< 2$. The values denoted by a star were fixed — see the main text for details.\[tab:params\]
A few comments are in order. The considered jet data are not sufficient to precisely determine all the parameters ($N$, $A$, $a$, $D$) of the initial parametrizations (\[eq:param\]). Thus, first we neglect the $(1-Dx)$ factor, i.e. we take $D=0$. We have checked that we get no improvement when $D$ is a free parameter. Next, in some cases the fits are not sensitive enough to uniquely determine the three remaining free parameters. In these cases we fix $A$ or $a$ at some plausible value (these are marked with a star in Table \[tab:params\]). Actually, besides the initial condition parameters $N$, $A$, $a$, $D$ we have also the boundary values of kinematic parameters $x_{A}$, $k_{T}$ (c.f. (\[eq:HEN\_fact\_2\])), which – to certain extent – are free parameters as well. We set them as follows. First, in order to be in an accordance with the assumptions leading to (\[eq:HEN\_fact\_2\]) we imply the cut $x_{A}<x_{B}$. Next, for all scenarios we set $x_{A\,\mathrm{min}}=0.0001$. For the model with the DGLAP correction we set $x_{A\,\mathrm{max}}=1.0$ while for the others we set $x_{A\,\mathrm{max}}=0.4$. Further we use $k_{T\,\mathrm{min}}=1\,\mathrm{GeV}$ for DGLAP models and $k_{T\,\mathrm{min}}=0.1\,\mathrm{GeV}$ for the others. Finally, we use $k_{T\,\mathrm{max}}=100\,\mathrm{GeV}$ for exponential initial condition and $k_{T\,\mathrm{max}}=400\,\mathrm{GeV}$ for the others. The last comment concerns the hard scale choice: in all fits we have used the average $p_{T}$ of the jets.
![The $p_{T}$ spectra of the central jet calculated using the best fits for individual models versus the CMS data. For the bottom plot the cross sections have been scaled by $p_{T}^{5}$ to better see the differences between the models.\[fig:central\]](Central "fig:"){width="48.00000%"} ![The $p_{T}$ spectra of the central jet calculated using the best fits for individual models versus the CMS data. For the bottom plot the cross sections have been scaled by $p_{T}^{5}$ to better see the differences between the models.\[fig:central\]](Central_scaled "fig:"){width="48.00000%"}
![The $p_{T}$ spectra of the forward jet calculated using the best fits for individual models versus the CMS data. For the bottom plot the cross sections have been scaled by $p_{T}^{5}$ to better see the differences between the models.\[fig:forward\]](Forward "fig:"){width="48.00000%"} ![The $p_{T}$ spectra of the forward jet calculated using the best fits for individual models versus the CMS data. For the bottom plot the cross sections have been scaled by $p_{T}^{5}$ to better see the differences between the models.\[fig:forward\]](Forward_scaled "fig:"){width="48.00000%"}
The influence of the Sudakov resummation model is illustrated in Fig. \[fig:sudeffect\]. Here, we have chosen the best fits to illustrate the effect. We see, that the jet spectra are rather weakly affected by the resummation, although the forward jet spectrum becomes steeper than the data.
![An effect of the Sudakov resummation model (BFKL+CDS) when applied to one of our fits for the model BFKL+CD with exponential initial condition. For comparison we plot also the spectra obtained from the unintegrated gluon density with more involved evolution and fitted to HERA data (KS-HERA), see the main text for more details. \[fig:sudeffect\]](bfklcds_exp_pTc "fig:"){width="48.00000%"} ![An effect of the Sudakov resummation model (BFKL+CDS) when applied to one of our fits for the model BFKL+CD with exponential initial condition. For comparison we plot also the spectra obtained from the unintegrated gluon density with more involved evolution and fitted to HERA data (KS-HERA), see the main text for more details. \[fig:sudeffect\]](bfklcds_exp_pTf "fig:"){width="48.00000%"}
The obtained UGDs are plotted in one-dimensional plots in Fig. \[fig:gluons\] as a function of $x$ and $k_T$. Note, that in order to better reflect the difference between UGDs we plot $k_T^2\, \mathcal{F}(x,k_T)$. We show results of all the models of Table \[tab:params\], hence also those with rather high $\chi^2$ value (see Table \[tab:chi2\]). All the UGDs with the DGLAP contribution are comparable, which shows that the evolution scenario is more important than a particular shape of the initial parametrization. On the other hand, the differences between UGDs are more pronounced than those in the $p_T$ spectra, which means that the currently available data are not sufficient to discriminate among the models. The two most differing UGDs correspond to the BFKL+C (POW) and BFKL+CR (EXP) models which however have significantly higher ${\ensuremath{\chi^{2}\!/\mathrm{NDP}}\xspace}$ (above 1.5).
We compare the new LHC-based UGDs with the one evolving according to a complicated evolution of [@Kwiecinski:1997ee; @Kutak:2004ym] and fitted to HERA data [@Kutak:2012rf] (we abbreviate it as ’KS-HERA’ on the figure). This evolution equation contains the kinematic constraint, full DGLAP correction (including quarks via coupled equations) and a nonlinear term motivated by the Balitsky-Kovchegov equation. The $p_T$ spectra resulting from this gluon density are presented in Fig. \[fig:sudeffect\].
![Unintegrated gluon distributions evolving due to the models \[enu:pure-BFKL-equation\]-\[enu:lastmodel\] with the initial conditions \[enu:Models\_ini\]-\[enu:Models\_fini\] obtained from the fits to the LHC data as a function of $x$ (top) and $k_T$ (bottom). The UGDs are multiplied by $k_T^2$ to better illustrate the differences between the models. The most differing UGDs are those without the DGLAP correction and with significantly higher ${\ensuremath{\chi^{2}\!/\mathrm{NDP}}\xspace}> 1.5$ (BFKL+C and BFKL+RC). \[fig:gluons\]](gluons){width="90.00000%"}
Azimuthal decorrelations {#sec:decorrelations}
========================
In order to apply the fits in practice we have calculated another observable for central-forward dijet production, namely, the differential cross sections in azimuthal angle $\Delta\phi$ between the two jets. At leading order the two jets are produced exactly back-to-back and the distribution is the Dirac delta at $\Delta\phi=\pi$. However, due to QCD emissions of additional partons (either forming additional jets or being soft particles with small $p_{T}$) the two jets are decorrelated. On theory side these decorrelations are well described by QCD-based parton shower algorithms. However, within the HEF there is a natural decorrelation mechanism built-in. Namely, due to the internal transverse momentum $k_{T}$ of a gluon the dijet system with transverse momenta $\vec{p}_{T1}$, $\vec{p}_{T2}$ is unbalanced by the amount $\left|\vec{p}_{T1}+\vec{p}_{T1}\right|=\left|\vec{k}_{T}\right|=k_{T}$. One can think of $k_{T}$ as a cumulative transverse momentum of many gluon emissions. In general, these emissions can be small-$p_{T}$ and large-$p_{T}$ emissions as well. The large-$p_{T}$ emissions may in general contribute a jet, thus we consider an inclusive dijet observables.
Using the new fits and the $\mathtt{LxJet}$ program we have calculated the azimuthal decorrelations for the kinematics described in the beginning of Section \[sec:Procedure\]. The results are presented in Fig. \[fig:decorr\]. The bands represent uncertainty that comes from the scale variation by a factor of two. We compare our calculation with the preliminary CMS data [@CMS:2014oma][^1].
![The results for the azimuthal decorrelations for inclusive forward-central dijet production using our best fits. When the Sudakov resummation model is applied to the generated events we get a better description of the CMS data.\[fig:decorr\]](bfklcds_exp_dphi "fig:"){width="50.00000%"}![The results for the azimuthal decorrelations for inclusive forward-central dijet production using our best fits. When the Sudakov resummation model is applied to the generated events we get a better description of the CMS data.\[fig:decorr\]](bfklcds_pow_dphi "fig:"){width="50.00000%"}
![The results for the azimuthal decorrelations for inclusive forward-central dijet production using our best fits. When the Sudakov resummation model is applied to the generated events we get a better description of the CMS data.\[fig:decorr\]](bfklcd_dphi "fig:"){width="50.00000%"}![The results for the azimuthal decorrelations for inclusive forward-central dijet production using our best fits. When the Sudakov resummation model is applied to the generated events we get a better description of the CMS data.\[fig:decorr\]](bfklcds_dphi "fig:"){width="50.00000%"}
Discussion {#sec:Summary}
==========
In the present paper we went through a thorough study of various small-$x$ evolution equations analyzing an impact of various effects on jet observables. The effects we mean here, are certain sub-leading corrections to the BFKL equation, such as the kinematic constraint or DGLAP corrections. Our study was based on fitting these evolution scenarios to two samples of LHC data for high-$p_{T}$ spectra for dijet production. These samples consist of separate spectra for the central rapidity and forward rapidity jets.
Our findings can be summarized as follows. First observation is that both forward jet and central jet spectra can be simultaneously and reasonably described by the High Energy Factorization approach and BFKL-like evolution. We obtain the best quality fits for BFKL with DGLAP correction and kinematic constraint, with the DGLAP correction being the most important additional ingredient. This matches the fact that the data under consideration can be nicely described by the collinear factorization with a parton shower [@Chatrchyan2012; @CMS:2014oma]. Whereas in the High Energy Factorization the parton shower is – to some extent – simulated by the transverse momentum dependent gluon distribution with the DGLAP correction. For all evolution models we get very good fits to the central jet spectrum, while most of the models have problems with precise reproduction of the shape of the forward jet spectrum. Several models properly describe the dijet data despite some differences in the resulting UGDs. Measurements of some other observables or more differential dijet data could help to discriminate among the models.
Using our fits we have calculated azimuthal decorrelations for the same kinematic domain. This observable was also measured by CMS. The comparison of our calculation with the data is reasonably good, especially when using the Sudakov resummation model on the top of the evolution models. Interestingly, the same resummation procedure spoils the forward jet $p_{T}$ spectrum.
Our final remark is that although the High Energy Factorization with improved BFKL evolution equation catches the main physical aspects of the jet production at small $x$, one definitely needs higher order corrections. Such calculations exist for certain small $x$ processes like Mueller-Navelet jets [@Ducloue:2013hia; @Ducloue:2014koa] or inclusive hadron production p+A collisions within CGC formalism [@Chirilli:2012jd; @Staato2014], but not for the high-$p_{T}$ dijet observables under consideration.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank K. Kutak and A. van Hameren for many fruitful discussions. The work of P.K. and W.S. has been supported by the Polish National Science Center Grant No. . D.T. has been supported by NCBiR Grant No. . P.K. also acknowledges the support of DOE grants No. and .
[^1]: We note that the total cross section obtained from [@CMS:2014oma] does not agree with [@Chatrchyan2012]. The ratio of the two is approx. 1.8. If this is a normalization difference only, our predictions should be shifted up by this factor.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We prove that any weakly acausal curve $\Gamma$ in the boundary of Anti-de Sitter (2+1)-space is the asymptotic boundary of two spacelike $K$-surfaces, one of which is past-convex and the other future-convex, for every $K\in(-\infty,-1)$. The curve $\Gamma$ is the graph of a quasisymmetric homeomorphism of the circle if and only if the $K$-surfaces have bounded principal curvatures. Moreover in this case a uniqueness result holds.
The proofs rely on a well-known correspondence between spacelike surfaces in Anti-de Sitter space and area-preserving diffeomorphisms of the hyperbolic plane. In fact, an important ingredient is a representation formula, which reconstructs a spacelike surface from the associated area-preserving diffeomorphism.
Using this correspondence we then deduce that, for any fixed $\theta\in(0,\pi)$, every quasisymmetric homeomorphism of the circle admits a unique extension which is a $\theta$-landslide of the hyperbolic plane. These extensions are quasiconformal.
address:
- 'Francesco Bonsante: Dipartimento di Matematica “Felice Casorati", Università degli Studi di Pavia, Via Ferrata 5, 27100, Pavia, Italy.'
- 'Andrea Seppi: University of Luxembourg, Mathematics Research Unit, Maison du Nombre, 6 Avenue de la Fonte, Esch-sur-Alzette L-4364 Luxembourg. '
author:
- Francesco Bonsante
- Andrea Seppi
bibliography:
- '../bs-bibliography.bib'
title: 'Area-preserving diffeomorphisms of the hyperbolic plane and $K$-surfaces in Anti-de Sitter space'
---
[^1]
Introduction
============
Since the groundbreaking work of Mess [@Mess], the interest in the study of Anti-de Sitter manifolds has grown, often motivated by the similarities with hyperbolic three-dimensional geometry, and with special emphasis on its relations with Teichmüller theory of hyperbolic surfaces. See for instance [@notes; @bbzads; @barbotzeghib; @bsk_multiblack; @bonschlfixed; @Schlenker-Krasnov; @bonseppitamb]. In fact, as outlined in [@aiyama; @bon_schl; @seppimaximal; @seppiminimal], several constructions can be generalized in the context of universal Teichmüller space. For instance, given a smooth spacelike convex surface $S$ in Anti-de Sitter space, or a maximal surface, one can define two *projections* from $S$ to the hyperbolic plane ${\mathbb{H}}^2$, and their composition provides a diffeomorphism $\Phi$ between domains of ${\mathbb{H}}^2$. In the context of non-smooth surfaces, an example of this phenomenon was already introduced by Mess, who observed that a *pleated surface* provides an *earthquake map* of ${\mathbb{H}}^2$. If $S$ is a smooth *maximal surface* (namely, a surface of zero mean curvature), then the associated map is a *minimal Lagrangian map*. A generalization of minimal Lagragian maps are the so-called $\theta$-*landslides*, which are one of the main objects of this paper. A $\theta$-landslide can be defined as a composition $\Phi=f_2\circ (f_1)^{-1}$, where $f_1$ and $f_2$ are harmonic maps from a fixed Riemann surface, with Hopf differentials satisfying the relation $$\mathrm{Hopf}(f_1)=e^{2i \theta }\mathrm{Hopf}(f_2)\,.$$ The $\theta$-landslides are precisely the maps associated to $K$-surfaces, i.e. surfaces of constant Gaussian curvature.
In this paper, we will frequently jump from one approach to the other: on the one hand, the study of convex surfaces in Anti-de Sitter space, with special interest in $K$-surfaces; on the other hand, the corresponding diffeomorphisms of (subsets of) ${\mathbb{H}}^2$, in particular $\theta$-landslides.
A representation formula for convex surfaces {#a-representation-formula-for-convex-surfaces .unnumbered}
--------------------------------------------
Let us briefly review the definition of the diffeomorphism $\Phi$ associated to a convex spacelike surface in Anti-de Sitter space, and survey the previous literature. Anti-de Sitter space ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ can be identified to ${\mathrm{Isom}}({\mathbb{H}}^2)$, endowed with the Lorentzian metric of constant curvature $-1$ which comes from the Killing form. The group of orientation-preserving, time-preserving isometries of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is identified to ${\mathrm{Isom}}({\mathbb{H}}^2)\times{\mathrm{Isom}}({\mathbb{H}}^2)$, acting by composition on the left and on the right. The essential point of the construction is the fact that the space of timelike geodesics of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is naturally identified to ${\mathbb{H}}^2\times {\mathbb{H}}^2$, where a point $(x,y)$ parameterizes the geodesic $L_{x,y}=\{\gamma\in{\mathrm{Isom}}({\mathbb{H}}^2):\gamma(y)=x\}$, as proved in [@barbotbtz1]. This fact has been used in several directions, see for instance [@fannythesis; @JDK; @guerkassel; @guersurvey; @barbotbtz1; @barbotbtz2; @bbsads]. When $S$ is spacelike surface, for every point $\gamma\in S$ the orthogonal line is timelike and thus it determines two points $(x,y)$ of ${\mathbb{H}}^2$. Thus one can define the *left projection* $\pi_l(\gamma)=x$ and the *right projection* $\pi_r(\gamma)=y$, and therefore construct the associated map $\Phi=\pi_r\circ\pi_l^{-1}$. This fact was first observed in [@Mess] when $S$ is a pleated surface, thus producing earthquake maps. For smooth convex surfaces, as observed in [@Schlenker-Krasnov], it turns out that $\Phi=\pi_r\circ\pi_l^{-1}$ is a diffeomorphism of (subsets of) ${\mathbb{H}}^2$ which preserves the area. In [@aiyama; @bon_schl; @seppimaximal] the case in which $S$ is a maximal surface, and correspondingly $\Phi$ is a minimal Lagrangian diffeomorphism, has been extensively studied. A similar construction has also been applied to surfaces with certain singularities in [@Schlenker-Krasnov; @toulisse]. More recently, progresses have been made on the problem of characterising the area-preserving maps $\Phi$ obtained by means of this construction, satisfying certain equivariance properties, see [@barbotkleinian; @bonsepequivariant; @seppiflux].
The first problem addressed in this paper is to what extent the surface $S$ can be reconstructed from the datum of the area-preserving diffeomorphism $\Phi$. By construction, the surface $S$ has to be orthogonal to the family of timelike geodesics $\{L_{x,\Phi(x)}\}$. Another basic observation is that, given a surface $S$, for every smooth surface obtained from $S$ by following the normal evolution of $S$, the associated map is still the same map $\Phi$. In fact, the orthogonal geodesics of the parallel surfaces of $S$ are the same as those of $S$. Up to this ambiguity, we are able to provide an explicit construction for the inverse of the left projection $\pi_l$, only in terms of $\Phi$. Actually we will give a 1-parameter family of maps into ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, thus reconstructing all the parallel surfaces which have associated map $\Phi$.
Of course there are conditions on the map $\Phi$. In fact, if $\Phi$ is a map which is associated to a surface $S$, then it is not difficult to prove that there exists a smooth $(1,1)$-tensor $b$ such that, if $g_{{\mathbb{H}}^2}$ denotes the hyperbolic metric, $$\label{condition1}
\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b\cdot,b\cdot)\,.$$ and $b$ satisfies the conditions: $$\begin{aligned}
&d^\nabla b=0\,, \label{condition2}\\
&\mathrm{det}\, b=1\,, \label{condition3}\\
&{\mbox{\rm tr}\,}b\in(-2,2) \label{condition4}\,,\end{aligned}$$ where $\nabla$ is the Levi-Civita connection of ${\mathbb{H}}^2$. We prove the following converse statement:
\[rep formula intro\] Let $\Phi:\Omega\to\Omega'$ be a diffeomorphism between two open domains of ${\mathbb{H}}^2$. Suppose that there exists a smooth $(1,1)$-tensor $b$ satisfying Equations ,,,. Consider the map $\sigma_{\Phi,{b}}:\Omega\to{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ defined by the condition that $\sigma_{\Phi,{b}}(x)$ is the unique isometry $\sigma$ satisfying $$\begin{aligned}
&\sigma(\Phi(x))=x\,; \label{definition1}\\
&d\sigma_{\Phi(x)}\circ d\Phi_x=-{b}_x\,. \label{definition2}\end{aligned}$$ Then $\sigma_{\Phi,b}$ is an embedding of $\Omega\subseteq {\mathbb{H}}^2$ onto a convex surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. If $\pi_l$ is the left projection of the image of $\sigma_{\Phi,b}$, then $\pi_l\circ\sigma_{\Phi,b}=\mathrm{id}$, and $\pi_r\circ\sigma_{\Phi, b}=\Phi$.
In fact, given a tensor $b$ satisfying Equations , and , one can still define $\sigma_{\Phi,b}$ by means of Equations and , and the image of the differential of $\sigma_{\Phi,b}$ is orthogonal to the family of timelike lines $\{L_{x,\Phi(x)}\}$, but in general $d\sigma_{\Phi,b}$ will not be injective. Actually, given $b$ which satisfies Equations , and , for every angle $\rho$, also the $(1,1)$-tensor $R_\rho\circ b$ satisfies Equations ,, where $R_\rho$ denotes the counterclockwise rotation of angle $\rho$. Changing $b$ by post-composition with $R_\rho$ corresponds to changing the surface $S$ to a parallel surface, moving along the timelike geodesics $\{L_{x,\Phi(x)}\}$. The condition ${\mbox{\rm tr}\,}b\in(-2,2)$ (which in general is only satisfied for some choices of $b$) ensures that $\sigma_{\Phi,b}$ is an embedding.
We now restrict our attention to $K$-surfaces and $\theta$-landslides. A $\theta$-*landslide* is a diffeomorphism $\Phi$ for which there exists $b$ satisfying the conditions of Equations , and and moreover its trace is constant. More precisely, $${\mbox{\rm tr}\,}b=2\cos\theta \qquad \mathrm{and}\qquad {\mbox{\rm tr}\,}Jb<0$$ for $\theta\in(0,\pi)$. It turns out that $\theta$-landslides are precisely the maps associated to past-convex $K$-surfaces, for $K=-{1}/{\cos^2({\theta}/{2})}$. On the other hand, by means of the map defined in Equations and , one associates a $K$-surface to a $\theta$-landslide. Changing $b$ by $-b$ in the defining Equations and enables to pass from the $K$-surface to its dual surface, which is still a surface of constant curvature $K^*=-K/(K+1)$. Hence a $\theta$-landslide is also the map associated with a future-convex $K^*$-surface.
A special case of $\theta$-landslides are minimal Lagrangian maps, for $\theta=\pi/2$. In this case, we get two $(-2)$-surfaces, dual to one another. It is well-known (see [@bon_schl]) that a minimal Lagrangian map is associated to a maximal surface $S_0$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, and that the two $(-2)$-surfaces are obtained as parallel surfaces at distance $\pi/4$ from $S_0$. Since in this case ${\mbox{\rm tr}\,}b=0$, changing $b$ by $Jb$, one has that $Jb$ is self-adjoint for the hyperbolic metric, and the map $\sigma_{\Phi,Jb}$ recovers the maximal surface with associated minimal Lagrangian map $\Phi$.
Foliations by $K$-surfaces of domains of dependence {#foliations-by-k-surfaces-of-domains-of-dependence .unnumbered}
---------------------------------------------------
We then focus on the case of $K$-surfaces. The first result we prove in this setting concerns the existence of convex $K$-surfaces with prescribed boundary at infinity. The boundary of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is identified to $\partial{\mathbb{H}}^2\times\partial{\mathbb{H}}^2$, where $\partial {\mathbb{H}}^2$ is the visual boundary of hyperbolic space, and thus $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is a torus. It is naturally endowed with a natural conformal Lorentzian structure, for which the null curves have either the first or the second component constant. A weakly acausal curve has the property that in a neighborhood of every point $\xi$, the curve is contained in the complement of the region connected to $\xi$ by timelike segments.
The basic example of a weakly acausal curve is the graph of an orientation-preserving homeomorphism of $\partial{\mathbb{H}}^2$. In fact, in this paper we prove an existence theorem for $K$-surfaces with boundary at infinity the graph of any orientation-preserving homeomorphism of $\partial{\mathbb{H}}^2$:
\[thm:qui\] Given any orientation-preserving homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$, the two connected components of the complement of the convex hull of $\Gamma=graph(\phi)$ in the domain of dependence of $\Gamma$ are both foliated by $K$-surfaces $S_K$, as $K\in(-\infty,-1)$, in such a way that if $K_1<K_2$, then $S_{K_2}$ is in the convex side of $S_{K_1}$.
Analogously to the case of hyperbolic space ${\mathbb{H}}^3$ (as proved in [@MR1293658]), there exists two $K$-surfaces with asymptotic boundary $\Gamma$, one of which is past-convex and the other future-convex. Theorem \[thm:qui\], in the case of $\phi$ a quasisymmetric homeomorphism, gives positive answer to the existence part of Question 8.3 in [@questionsads]. As expressed in [@questionsads], Question 8.3 also conjectured uniqueness and boundedness of principal curvatures. These are proved in Theorem \[thm uniqueness and boundedness\] below.
In general, a weakly acausal curve $\Gamma$ can contain null segments. In particular, if $\Gamma$ contains a *sawtooth*, that is, the union of adjacent “horizontal” and “vertical” segments in $\partial{\mathbb{H}}^2\times\partial{\mathbb{H}}^2$, then the convex hull of the sawtooth is a lightlike totally geodesic triangle, which is contained both in the boundary of the convex hull of $\Gamma$ and in the boundary of the domain of dependence of $\Gamma$. Hence any (future or past) convex surface with boundary $\Gamma$ must necessarily contain such lightlike triangle.
An example is a 1-*step curve*, which is the union of a “horizontal” and a “vertical” segment in $\partial{\mathbb{H}}^2\times\partial{\mathbb{H}}^2$. A 2-*step curve* is the union of four segments, two horizontal and two vertical in an alternate way. It is not possible to have a convex surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with boundary a 1-step or a 2-step curve.
The proof of Theorem \[thm:qui\] actually extends to the case of a general weakly acausal curve $\Gamma$, except the two degenerate cases above.
\[thm Ksurfaces ads intro\] Given any weakly acausal curve $\Gamma$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which is not a 1-step or a 2-step curve, for every $K\in(-\infty,-1)$ there exists a past-convex (resp. future-convex) surface $S_K^+$ (resp. $S_K^-$) with $\partial S_K^\pm=\Gamma$, such that:
- Its lightlike part is union of lightlike triangles associated to sawteeth;
- Its spacelike part is a smooth $K$-surface.
Moreover, the two connected components of the complement of the convex hull of $\Gamma$ in the domain of dependence of $\Gamma$ are both foliated by the spacelike part of surfaces $S_K^\pm$, as $K\in(-\infty,-1)$, in such a way that if $K_1<K_2$, then $S^\pm_{K_2}$ is in the convex side of $S^\pm_{K_1}$.
In [@barbotzeghib], the existence (and uniqueness) of a foliation by $K$-surfaces was proved in the complement of the convex core of any maximal globally hyperbolic Anti-de Sitter spacetime containing a compact Cauchy surface. Using results of [@Mess], this means that the statement of Theorem \[thm Ksurfaces ads intro\] holds for curves $\Gamma$ which are the graph of an orientation-preserving homeomorphism which conjugates two Fuchsian representations of the fundamental group of a closed surface in ${\mathrm{Isom}}({\mathbb{H}}^2)$. Moreover, the $K$-surfaces are invariant for the representation in ${\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)\cong {\mathrm{Isom}}({\mathbb{H}}^2)\times {\mathrm{Isom}}({\mathbb{H}}^2)$ given by the product of the two Fuchsian representations.
The proof of Theorem \[thm:qui\], and more generally Theorem \[thm Ksurfaces ads intro\], relies on an approximation from the case of [@barbotzeghib]. Some technical tools are needed. First, we need to show that it is possible to approximate any weakly spacelike curve $\Gamma$ by curves invariant by a pair of Fuchsian representations. For this purpose, we adapt a technical lemma proved in [@Bonsante:2015vi].
Second, we use a theorem of Schlenker ([@schlenkersurfconv]) which, in this particular case, essentially ensures that a sequence $S_n$ of $K$-surfaces in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ converges $C^\infty$ to a spacelike surface $S_\infty$ (up to subsequences) unless they converge to a totally geodesic lightlike plane (whose boundary at infinity is a 1-step curve) or to the union of two totally geodesic lightlike half-planes, meeting along a spacelike geodesic (in this case the boundary is a 2-step curve).
To apply the theorem of Schlenker, and deduce that the limiting surface $S_\infty$ is a $K$-surface with $\partial S_\infty=\Gamma$ (thus proving Theorem \[thm:qui\]), one has to prove that $S_\infty$ does not intersect the boundary of the domain of dependence of $\Gamma$. More in general, for the proof of Theorem \[thm Ksurfaces ads intro\], one must show that the *spacelike part* of $S_\infty$ does not intersect the boundary of the domain of dependence of $\Gamma$. This is generally the most difficult step in this type of problems, and frequently requires the use of *barriers*. Here this issue is indeed overcome by applying the representation formula of Theorem \[rep formula intro\] in order to construct suitable barriers.
In fact, it is possible to compute a family of $\theta$-landslides from a half-plane in ${\mathbb{H}}^2$ to itself, which commutes with the hyperbolic 1-parameter family of isometries of ${\mathbb{H}}^2$ preserving the half-plane. By using this invariance, the equation which rules the condition of a map $\Phi$ being a $\theta$-landslide is reduced to an ODE. By a qualitative study it is possible to show that there exists a 1-parameter family of smooth, spacelike $K$-surfaces whose boundary coincides with the boundary of a totally geodesic spacelike half-plane in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. In other words, Theorem \[thm Ksurfaces ads intro\] is proved by a *hands-on* approach when the curve $\Gamma$ is the union of two null segments and the boundary of a totally geodesic half-plane, in the boundary at infinity of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Such $K$-surfaces are then fruitfully used as *barriers* to conclude the proof of Theorem \[thm:qui\] - and the proof actually works under the more general hypothesis of Theorem \[thm Ksurfaces ads intro\].
In the case $\Gamma$ is the graph of a quasisymmetric homeomorphism, we then prove that the $K$-surfaces with boundary $\Gamma$ are unique. Moreover, it is not difficult to prove that if $S$ is a convex surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S=\Gamma$ and with bounded principal curvatures, then $\Gamma$ is the graph of a quasisymmetric homeomorphism. We give a converse statement for $K$-surfaces, namely, a $K$-surface with boundary $\Gamma=gr(\phi)$, for $\phi$ quasisymmetric, necessarily has bounded principal curvatures.
\[thm uniqueness and boundedness\] Given any quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$, for every $K\in(-\infty,-1)$ there exists a unique future-convex $K$-surface $S^+_K$ and a unique past-convex $K$-surface $S^-_K$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S_K^\pm=gr(\phi)$. Moreover, the principal curvatures of $S_K^\pm$ are bounded.
To prove uniqueness, the standard arguments for these problems are applications of the *maximum principle*, by using the existence of a foliation $\{S_K\}$ by $K$-surfaces and showing that any other $K$-surface $S_K'$ must coincide with a leaf of the given foliation. However, in this case, due to non-compactness of the surfaces, one would need a form of the maximum principle *at infinity*. This is achieved more easily in this case by applying isometries of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ so as to bring a maximizing (or minimizing) sequence on $S_K'$ to a compact region of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Then one applies two main tools: the first is again the convergence theorem of Schlenker, and the second is a compactness result for quasisymmetric homeomorphisms with uniformly bounded cross-ratio norm. Up to subsequences, both the isometric images of $S_K'$ and the isometric images of the leaves $\{S_K\}$ of the foliation converge to an analogous configuration in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. But now it is possible to apply the classical maximum principle to conclude the argument.
Extensions of quasisymmetric homeomorphisms by $\theta$-landslides {#extensions-of-quasisymmetric-homeomorphisms-by-theta-landslides .unnumbered}
------------------------------------------------------------------
By interpreting Theorems \[thm Ksurfaces ads intro\] and \[thm uniqueness and boundedness\] in the language of diffeomorphism of ${\mathbb{H}}^2$ we can draw a direct consequence.
\[cor intro 2\] Given any quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$ and any $\theta\in(0,\pi)$, there exist a unique $\theta$-landslide $\Phi_\theta:{\mathbb{H}}^2\to{\mathbb{H}}^2$ which extends $\phi$. Moreover, $\Phi_\theta$ is quasiconformal.
Again, uniqueness follows from the uniqueness part Theorem \[thm uniqueness and boundedness\] and from the construction of Theorem \[rep formula intro\], while quasiconformality is a consequence of boundedness of principal curvatures.
For $\theta=\pi/2$, we obtain a new proof of the following result of [@bon_schl]:
\[cor intro 1\] Given any quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$ there exists a unique minimal Lagrangian extension $\Phi:{\mathbb{H}}^2\to{\mathbb{H}}^2$ of $\phi$. Moreover, $\Phi$ is quasiconformal.
In fact, in [@bon_schl] the extension by minimal Lagrangian maps was proved again by means of Anti-de Sitter geometry, by proving existence and uniqueness (and boundedness of principal curvatures, which implies quasiconformality) of a maximal surface with boundary the graph of $\phi$. Here we instead proved the existence of two $(-2)$-surfaces, which coincide with the parallel surfaces at distance $\pi/4$ from the maximal surface - the minimal Lagrangian map $\Phi$ associated to these three surfaces in the same.
Organization of the paper {#organization-of-the-paper .unnumbered}
-------------------------
In Section \[prel ads\] we give an introduction of Anti-de Sitter space as the group ${\mathrm{Isom}}({\mathbb{H}}^2)$, from a Lie-theoretical approach. Section \[sec causal\] discusses some properties of the causal geometry of Anti-de Sitter space, including the definition of domain of dependence. In Section \[sec surfaces\] we introduce the left and right projection from a convex surface, and thus the associated diffeomorphism of ${\mathbb{H}}^2$, and their relation with the differential geometry of smooth surfaces. In Section \[sec representation formula\] we construct the “representation” for the inverse of the left projection of a convex (or maximal) surface, and we prove Theorem \[rep formula intro\]. Section \[sec barrier\] studies $\theta$-landslides which commute with a 1-parameter hyperbolic group, constructs the “barriers” which are necessary to prove Theorem \[thm Ksurfaces ads intro\] and their relevant properties. In Section \[sec existence\] the existence part of Theorem \[thm Ksurfaces ads intro\] is proved, while Section \[sec foliations\] proves that the $K$-surfaces give a foliation of the complement of the convex hull in the domain of dependence. Finally, Section \[sec uniqueness boundedness\] proves Theorem \[thm uniqueness and boundedness\] and discusses Corollaries \[cor intro 1\] and \[cor intro 2\].
Acknowledgements {#acknowledgements .unnumbered}
----------------
We are grateful to an anonymous referee for carefully reading the paper and for several useful comments which improved the exposition. We would also like to thank Thierry Barbot for pointing out several relevant references on the topic.
Anti-de Sitter space and isometries of ${\mathbb{H}}^2$ {#prel ads}
=======================================================
Let ${\mathbb{H}}^2$ denote the hyperbolic plane, which is the unique complete, simply connected Riemannian surface without boundary of constant curvature -1. We denote by $\partial{\mathbb{H}}^2$ its visual boundary, and by ${\mathrm{Isom}}({\mathbb{H}}^2)$ the Lie group of orientation-preserving isometries of ${\mathbb{H}}^2$. Recall that the Killing form $\kappa$ on the Lie algebra of $\mathfrak{isom}({\mathbb{H}}^2)$, namely $$\kappa(v,w)=\mathrm{tr}(\mathrm{ad}(v)\circ\mathrm{ad}(w))\,,$$ is $\mathrm{Ad}$-invariant. Thus it defines a bi-invariant pseudo-Riemannian metric on ${\mathrm{Isom}}({\mathbb{H}}^2)$, still denoted by $\kappa$, which has signature $(2,1)$. We will normalize $\kappa$ to impose that its sectional curvature is $-1$. As this normalization will be relevant in this paper, we will briefly outline the computation of the sectional curvature of $\kappa$.
The Killing form $\kappa$ has constant sectional curvature $-1/8$.
Let us fix $v,w\in\mathfrak{isom}({\mathbb{H}}^2)$. By results in [@milnor_curvature] the sectional curvature of the plane $\Pi$ generated by $v,w$ is given by $$K_{\Pi}=\frac{1}{4}\frac{\kappa([v,w], [v,w])}{\kappa(v,v)\kappa(w,w)-\kappa(v,w)\kappa(v,w)}~.$$ Now suppose that $\kappa(v,v)=1, \kappa(w,w)=1, \kappa(v,w)=0$. Then $[v,w]\neq 0$ as $v$ and $w$ are linearly independent. Denote $u=[v,w]$. Notice that by the properties of the Killing form $\kappa(u, v)=\kappa(u, w)=0$, so $u$ is timelike. As $[v,u]$ is orthogonal to both $v$ and $u$, it turns out that $[v,u]=\lambda w$ and analogously $[w,u]=\mu v$. Imposing that ${\mbox{\rm tr}\,}(\mathrm{ad}(v)^2)={\mbox{\rm tr}\,}(\mathrm{ad}(w)^2)=1$ we deduce that $\lambda=1/2$ and $\mu=-1/2$. So a simple computation shows that $\kappa(u,u)={\mbox{\rm tr}\,}(\mathrm{ad}(u)^2)=-1/2$, so that $K_{\Pi}=-1/8$. By an analogous computation, picking $v$ and $w$ with $\kappa(v,v)=1, \kappa(w,w)=-1, \kappa(v,w)=0$, the sectional curvature of a timelike plane is also $-1/8$.
Anti-de Sitter space of dimension 3 is the Lie group ${\mathrm{Isom}}({\mathbb{H}}^2)$ endowed with the bi-invariant metric $g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}=\frac{1}{8}\kappa$, and will be denoted by ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$.
It turns out that ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ has the topology of a solid torus, is orientable and time-orientable, and the induced pseudo-Riemannian metric has constant sectional curvature $-1$.
We will choose the time orientation of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ in such a way that the timelike vectors which are tangent to the differentiable curve $$t\in[0,\epsilon)\longrightarrow R_t\circ \gamma\,,$$ where $R_t$ is a rotation of positive angle $t$ around any point $x\in{\mathbb{H}}^2$ with respect to the orientation of ${\mathbb{H}}^2$, are future-directed. Moreover, we fix an orientation of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ so that if $v, w$ are linearly independent spacelike elements of $\mathfrak{isom}({\mathbb{H}}^2)$, then $\{v,w,[v,w]\}$ is a positive basis of $\mathfrak{isom}({\mathbb{H}}^2)$.
By construction, the group of orientation-preserving, time-preserving isometries of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is: $${\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)\cong{\mathrm{Isom}}({\mathbb{H}}^2)\times{\mathrm{Isom}}({\mathbb{H}}^2)\,,$$ where the left action on ${\mathrm{Isom}}({\mathbb{H}}^2)$ is given by: $$(\alpha,\beta)\cdot \gamma=\alpha\circ\gamma\circ \beta^{-1}\,.$$
The boundary at infinity of Anti-de Sitter space is defined as $$\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3\cong \partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$$ where a sequence $\gamma_n\in{\mathrm{Isom}}({\mathbb{H}}^2)$ converges to a pair $(p,q)\in \partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$ if there exists a point $x\in{\mathbb{H}}^2$ such that $$\label{defi convergence boundary}
\gamma_n(x)\to p\qquad \gamma_n^{-1}(x)\to q\,,$$ and in this case the condition is true for any point $x$ of ${\mathbb{H}}^2$. The isometric action of ${\mathrm{Isom}}({\mathbb{H}}^2)\times{\mathrm{Isom}}({\mathbb{H}}^2)$ on ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ continuously extends to the product action on $\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$: if $p,q\in\partial {\mathbb{H}}^2$, then $$(\alpha,\beta)\cdot(p,q)=(\alpha(p),\beta(q))\,.$$ The boundary at infinity $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is endowed with a conformal Lorentzian structure, in such a way that ${\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)$ acts on the boundary by conformal transformations. The null lines of $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ are precisely $\partial{\mathbb{H}}^2\times\{\star\}$ and $\{\star\} \times\partial{\mathbb{H}}^2$.
Since the exponential map at the identity for the Levi-Civita connection of the bi-invariant metric coincides with the Lie group exponential map, the geodesics through the identity are precisely the 1-parameter subgroups. In particular elliptic subgroups correspond to timelike geodesics through the identity. Using the action of the isometry group, it follows that timelike geodesics (i.e. those of negative squared length) have the form: $$L_{x,x'}=\{\gamma\in{\mathrm{Isom}}({\mathbb{H}}^2):\gamma(x')=x\}\,.$$ They are closed and have length $\pi$. Observe that with this definition the isometry group acts on timelike geodesics in such a way that $$\label{transf rule timelike geodesic}
(\alpha,\beta)\cdot L_{x,x'}=L_{\alpha(x), \beta(x')}\,.$$
For a similar argument as above, spacelike geodesics (i.e. those on which the metric is positive) are of the form $$L_{\ell,\ell'}=\{\gamma\in{\mathrm{Isom}}({\mathbb{H}}^2):\gamma(\ell')=\ell\}\,,$$ where $\ell$ and $\ell'$ are oriented geodesics of ${\mathbb{H}}^2$. The geodesic $L_{\ell,\ell'}$ has infinite length and its endpoints in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ are $(p_1,q_1)$ and $(p_2,q_2)$, where $p_1$ and $p_2$ are the final and initial endpoints of $\ell$ in $\partial {\mathbb{H}}^2$, while $q_1$ and $q_2$ are the final and initial endpoints of $\ell'$.
The involutional rotations of angle $\pi$ around a point $x\in{\mathbb{H}}^2$, which we denote by $\mathcal{I}_x$, are the antipodal points to the identity in the geodesics $L_{x,x}$, and form a totally geodesic plane $$\mathcal R_\pi=\{\mathcal I_x\,:\,x\in{\mathbb{H}}^2\}\,.$$ See also Figure \[fig:torus\].
![A (topological) picture of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. The totally geodesic plane $\mathcal R_\pi$ is given by the midpoints of timelike geodesics from the identity. Its boundary at infinity is the tangency locus of the lightcone from the identity with the boundary of the solid torus. \[fig:torus\]](torus-eps-converted-to.pdf){height="7cm"}
By using the definition in Equation , it is easy to check that its boundary at infinity $\partial \mathcal R_\pi$ is the diagonal in $\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$: $$\partial \mathcal R_\pi=\{(p,p)\,:\,p\in\partial{\mathbb{H}}^2\}\subset\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2\,.$$ More generally, given any point $\gamma$ of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, the points which are connected to $\gamma$ by a timelike segment of length $\pi/2$ (actually, two timelike segments whose union form a closed timelike geodesic) form a totally geodesic plane, called the *dual* plane $\gamma^*$. From this definition, $\mathcal R_\pi=(\mathrm{id})^*$. If $(\alpha,\beta)$ is an isometry of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which sends $\mathrm{id}$ to $\gamma$, then $\gamma^*=(\alpha,\beta)\cdot \mathcal R_\pi$. In particular, $\gamma^*=(\gamma,1)\cdot \mathcal R_\pi=(1,\gamma^{-1})\cdot \mathcal R_\pi$. The boundary at infinity of a totally geodesic spacelike plane is thus the graph of the trace on $\partial{\mathbb{H}}^2$ of an isometry of ${\mathbb{H}}^2$.
An isometry of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ fixes a point if and only if it preserves its dual plane. In particular, the subgroup which fixes the identity (and therefore preserves $ \mathcal R_\pi=\mathrm{id}^*$) is given by the diagonal: $$\mathrm{Stab}(\mathrm{id})=\mathrm{Stab}(\mathcal R_\pi)=\{(\gamma,\gamma)\,:\,\gamma\in{\mathrm{Isom}}({\mathbb{H}}^2)\}\,.$$ We can give an explicit isometric identification of ${\mathbb{H}}^2$ to the totally geodesic plane $ \mathcal R_\pi$ by means of the following map: $$x\in{\mathbb{H}}^2\mapsto\mathcal{I}_x\in \mathcal R_\pi\,.$$ The identification is a natural isometry, in the sense that the action of ${\mathrm{Isom}}({\mathbb{H}}^2)$ on ${\mathbb{H}}^2$ corresponds to the action of $\mathrm{Stab}(\mathcal R_\pi)$ on $\mathcal R_\pi$, since $$\gamma\mathcal{I}_x\gamma^{-1}=\mathcal{I}_{\gamma(x)}\,.$$
Two useful models {#subsec models}
-----------------
In this subsection we discuss two models of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, arising from the choice of two different models of ${\mathbb{H}}^2$, which will be both useful for different reasons in some computations necessary for this paper.
For the first model, let us consider ${\mathbb{H}}^2$ as a sheet of the two-sheeted hyperboloid in Minkowski space, namely: $${\mathbb{H}}^2=\{x\in {{\mathbb R}}^{2,1}\,:\,\langle x,x\rangle_{{{\mathbb R}}^{2,1}}=-1\,,\,x_3>0\}\,,$$ where the Minkowski product of ${{\mathbb R}}^{2,1}$ is: $$\langle x,x\rangle_{{{\mathbb R}}^{2,1}}=x_1^2+x_2^2-x_3^2\,.$$ In this model, the orientation-preserving isometries of ${\mathbb{H}}^2$ are identified to the connected component of the identity in ${\mathrm{SO}}(2,1)$, namely: $${\mathrm{Isom}}({\mathbb{H}}^2)\cong {\mathrm{SO}}_0(2,1)\,.$$ Hence, the identification of ${\mathbb{H}}^2$ with $\mathcal R_\pi\subset{\mathrm{Isom}}({\mathbb{H}}^2)$ is given by the map which associates to $x\in{\mathbb{H}}^2$ the linear map of ${{\mathbb R}}^{2,1}$ sending $x$ to $x$ and acting by multiplication by $-1$ on the orthogonal complement $x^\perp=T_x{\mathbb{H}}^2$. Moreover, the tangent space at $\mathrm{id}$ is identified to the Lie algebra of ${\mathrm{SO}}(2,1)$: $$T_{\mathrm{id}}{\mathbb{A}\mathrm{d}\mathbb{S}}^3\cong {\mathfrak{so}}(2,1)\,.$$ There is a natural Minkowski cross product, defined by $x\boxtimes y=*(x\wedge y)$, where $*:\Lambda^2(\mathbb R^{2,1})\to\mathbb R^{2,1}$ is the Hodge operator associated to the Minkowski product. Observe that, by means of this cross product, one can write the almost-complex structure of ${\mathbb{H}}^2$. For $v\in T_x{\mathbb{H}}^2$, $$\label{almost-complex cross}
J_x(v)=x\boxtimes v\,.$$ Like the classical case of Euclidean space, one can define an isomorphism $$\Lambda:{{\mathbb R}}^{2,1}\to{\mathfrak{so}}(2,1)\,,$$ which is defined by: $$\Lambda(x)=x\boxtimes(\cdot)\,.$$ See also Figure \[fig:tangent\]. This isomorphism has several remarkable properties:
- It is equivariant for the action of ${\mathrm{SO}}(2,1)$ on ${{\mathbb R}}^{2,1}$ by isometries, and the adjoint action on ${\mathfrak{so}}(2,1)$, which is the natural action of $\mathrm{Stab}(\mathrm{id})\cong {\mathrm{Isom}}({\mathbb{H}}^2)$ on $T_{\mathrm{id}}{\mathbb{A}\mathrm{d}\mathbb{S}}^3$: if $\gamma\in{\mathrm{SO}}(2,1)$, then $$\label{prop lambda 1}
\Lambda(\gamma\cdot x)(v)=(\gamma\cdot x)\boxtimes v=\gamma(x\boxtimes (\gamma^{-1}v))=\gamma\circ \Lambda(x)\circ\gamma^{-1}(v)\,.$$
- It pulls back the Lie bracket of ${\mathfrak{so}}(2,1)$ to the Minkowski cross product: $$\label{prop lambda 3}
[\Lambda(x),\Lambda(y)]=\Lambda(x\boxtimes y)\,.$$
- It is an isometry between the Killing form of ${\mathfrak{so}}(2,1)$ and the Minkowski metric on ${{\mathbb R}}^{2,1}$, up to a factor: $$\label{prop lambda 2}
(g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3})_{\mathrm{id}}(\Lambda(x),\Lambda(y))=\frac{1}{4}\langle x,y\rangle_{{{\mathbb R}}^{2,1}}\,.$$
Let us check the factor in Equation . By for any $x,y\in{{\mathbb R}}^{2,1}$ we have $$\mathrm{ad}(\Lambda(x))(\Lambda(y))=\Lambda(\Lambda(x)y)$$ that is $$\Lambda^{-1}\circ \mathrm{ad}(\Lambda(x))\circ\Lambda=\Lambda(x)~.$$ It follows that for any $x,y\in{{\mathbb R}}^{2,1}$ $$(g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3})_{\mathrm{id}}(\Lambda(x),\Lambda(y))=\frac{1}{8}{\mbox{\rm tr}\,}(\Lambda(x)\Lambda(y))~.$$ Using that $$x\boxtimes (y\boxtimes z)=(\langle x, y\rangle_{{{\mathbb R}}^{2,1}}) z-(\langle x,z\rangle_{{{\mathbb R}}^{2,1}}) y$$ we have that ${\mbox{\rm tr}\,}(\Lambda(x)\Lambda(y))=2\langle x,y\rangle_{{{\mathbb R}}^{2,1}}$, hence obtaining Equation .
![In the identification of $T_{\mathrm{id}}{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with ${{\mathbb R}}^{2,1}$, timelike vectors of Minkowski space (in the interior of the cone) are the generators of elliptic 1-parameter subgroups; lightlike vectors generate parabolic subgroups and spacelike vectors generate hyperbolic subgroups. \[fig:tangent\]](tangent-eps-converted-to.pdf){height="5cm"}
The second model we consider comes from the choice of the upper half-plane model of ${\mathbb{H}}^2$, namely $${\mathbb{H}}^2=\{z\in{{\mathbb C}}:{\mathrm{Im}}(z)>0\}\,,$$ endowed with the Riemannian metric ${|dz|^2}/{{\mathrm{Im}}(z)^2}$ which makes every biholomorphism of the upper-half plane an isometry. In this model ${\mathrm{Isom}}({\mathbb{H}}^2)$ is naturally identified to ${\mathrm{PSL}}(2,{{\mathbb R}})$ and the visual boundary $\partial{\mathbb{H}}^2$ is identified with the extended line ${{{\mathbb R}}\mathrm{P}}^1$. The Lie algebra $\mathfrak{sl}(2,{{\mathbb R}})$ is thus the vector space of traceless 2 by 2 matrices, and in this model the Anti de Sitter metric at the identity is given by: $$\label{eq:minnie}(g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3})_{\mathrm{id}}( m,m')=\frac{1}{2}{\mbox{\rm tr}\,}(mm')\,.$$
Consider the quadratic form $q(M)=-\det M$ on the space of $2$-by-$2$ matrices. Its polarization, say $b$, has signature $(2,2)$. We notice that the restriction of $b$ to ${\mathrm{SL}}(2,\mathbb R)$ is exactly the double cover of $g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}$. Indeed left and right multiplication by elements in ${\mathrm{SL}}(2,{{\mathbb R}})$ preserve $q$, so that the restriction of $b$ to ${\mathrm{SL}}(2,{{\mathbb R}})$ is a bi-invariant metric. Moreover at the identity Equation shows that it coincides with $g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}$.
Hence this model of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, namely $${\mathrm{PSL}}(2,{{\mathbb R}})=\{A:q(A)=-1\}/\{\pm 1\}$$ endowed with the pseudo-Riemannian metric which descends from $b$, is remarkably a *projective model*. In fact, ${\mathrm{PSL}}(2,{{\mathbb R}})$ is a subset of ${{{\mathbb R}}\mathrm{P}}^3$, geodesics for the pseudo-Riemannian metric of ${\mathrm{PSL}}(2,{{\mathbb R}})$ are the intersections of ${\mathrm{PSL}}(2,{{\mathbb R}})$ with a projective line, and totally geodesic planes are the intersections with projective planes. Thus in an affine chart, geodesics are straight lines and totally geodesic planes are affine planes.
![In an affine chart, ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is the interior of a one-sheeted hyperboloid. The null lines of $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ coincide with the rulings of the hyperboloid. The intersection with the horizontal plane $z=0$ is a totally geodesic hyperbolic plane, in the Klein model. \[fig:ruled\]](ruled.png){height="6cm"}
Causal geometry of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ {#sec causal}
=======================================================
Given a continuous curve $\Gamma$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, we say that a $\Gamma$ is *weakly acausal* if for every point $p$ of $\Gamma$, there exists a neighborhood $U$ of $p$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ such that $U\cap\Gamma$ is contained in the complement of the regions of $U$ which are connected to $p$ by timelike curves.
There are two important objects we associate to a weakly spacelike curve.
Given a weakly acausal curve $\Gamma$, the convex hull $\mathcal{C}(\Gamma)$ is the smallest closed convex subset which contains $\Gamma$.
It turns out that $\mathcal{C}(\Gamma)$ is contained in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3\cup\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ and that $\mathcal{C}(\Gamma)\cap\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3=\Gamma$ (see [@bon_schl Lemma 4.8]).
Given a weakly acausal curve $\Gamma$, the domain of dependence $\mathcal{D}(\Gamma)$ is the union of points $p$ of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ such that $p^*$ is disjoint from $\Gamma$.
It turns out that the domain of dependence $\mathcal{D}(\Gamma)$ is always an open subset of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ containing the interior part of $\mathcal C(\Gamma)$. Moreover, $\mathcal{D}(\Gamma)$ is contained in an affine chart for ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, and admits no timelike support planes. As for the convex hull, $\overline{\mathcal{D}(\Gamma)}\cap\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3=\Gamma$. Let us denote by $\mathcal{D}_\pm(\Gamma)$ the connected components of $\mathcal{D}(\Gamma)\setminus \mathcal{C}(\Gamma)$, so that $$\mathcal{D}(\Gamma)\setminus \mathcal{C}(\Gamma)=\mathcal{D}_+(\Gamma)\sqcup\mathcal{D}_-(\Gamma)\,.$$ We choose the notation in such a way that there exists a future-directed timelike arc in $\mathcal{D}(\Gamma)$ going from $\mathcal{D}_-(\Gamma)$ to $\mathcal{D}_+(\Gamma)$ and intersecting $\mathcal{C}(\Gamma)$.
\[defi locally convex\] Let $K$ be any convex subset of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ contained in some affine chart. A *locally convex spacelike (resp. nowhere timelike)* surface $S$ is a connected region of $\partial K$ such that the support planes of $K$ at points of $S$ are all spacelike (resp. non timelike).
\[lemma:bandabassotti\] Let $S$ be a locally convex nowhere timelike surface contained in an affine chart $U$. Then $S$ is contained either in all future half-spaces bounded by support planes at points in $S$ or in all past half-spaces.
In the former case we say that $S$ is *future-convex*, in the latter *past-convex*.
For any point $x\in S$ and any support plane $P$ at $x$, $S$ is contained either in the future or in the past half-space bounded by $P$. It is immediate to verify that the set of $(x,P)$ such that $S$ is contained in the future half-space of $P$ is open and closed in the set of pairs $(x,P)$ with $x\in S$ and $P$ support plane at $x$. As we are assuming that $S$ is connected, we conclude that this set is either the whole set or the empty set.
For instance, the boundary of $\mathcal{C}(\Gamma)$ is composed of two *pleated* nowhere timelike surfaces, which we denote by $\partial_\pm\mathcal{C}(\Gamma)$, so that $\partial_+\mathcal{C}(\Gamma)$ is past-convex and $\partial_-\mathcal{C}(\Gamma)$ is future-convex. They are distinct unless $\Gamma$ is the boundary of a totally geodesic plane. More precisely, the subset of $\partial_\pm\mathcal{C}(\Gamma)$ which admits spacelike support planes is a pleated hyperbolic surface. That is, there exists an isometric map $f_\pm:H_\pm\to \partial_\pm\mathcal{C}(\Gamma)$, where $H_\pm$ are complete simply connected hyperbolic surfaces with geodesic boundary (also called *straight convex sets* in [@bebo]), such that every point $x\in H_\pm$ is in the interior of a geodesic arc which is mapped isometrically to a spacelike geodesic of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$.
Given two spacelike planes $Q_1,Q_2$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, we can define their *hyperbolic angle* $\alpha\geq 0$ by: $$\cosh\alpha=|\langle N_1,N_2\rangle|\,,$$ where $N_1$ and $N_2$ are the unit normal vectors. Using this notion, one can define (similarly to [@epsteinmarden] in the case of hyperbolic geometry) a *bending lamination*, which is a measured geodesic lamination on the straight convex set $H_\pm$ in the sense of [@bebo Section 3.4]. In particular, the *transverse measure* satisfies the usual requirements in the definition of measured geodesic laminations (see also Subsection \[subsec approximation\] for more details), with the additional property that the measure on a transverse arc $I$ is infinite if and only if $I$ has nonempty intersection with $\partial H_\pm$. The data of a straight convex set $H$ and the measured geodesic lamination on $H$ determines the pleated surface up to global isometries of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$.
It is not difficult to show that $\partial_+\mathcal{C}(\Gamma)$ intersects the boundary of the domain of dependence $\mathcal{D}(\Gamma)$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ if and only if the weakly acausal curve $\Gamma$ contains a *past-directed sawtooth*, that is, if it contains two segments of $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, one which is part of a line of the left ruling and the other in a line of the right ruling, having a common past endpoint. In this case, $\partial_+\mathcal{C}(\Gamma)$ contains a totally geodesic lightlike triangle, bounded by the two above segments in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ and by a spacelike complete geodesic of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. The bending lamination of $\partial_+\mathcal{C}(\Gamma)$ has therefore infinite weight on such spacelike geodesic. Of course one can give the analogous definition and characterization for *future-directed sawteeth*, which can possibly be contained in $\partial_-\mathcal{C}(\Gamma)$. See also Figure \[fig:sawtooth\].
![A past-directed and a future-directed sawtooth bounding a lightlike triangle. \[fig:sawtooth\]](pastsawtooth-eps-converted-to.pdf){height="5.5cm"}
![A past-directed and a future-directed sawtooth bounding a lightlike triangle. \[fig:sawtooth\]](sawtooth-eps-converted-to.pdf){height="5.5cm"}
In this paper, we are interested in convex nowhere timelike surfaces having as a boundary at infinity a weakly acausal curve $\Gamma\subset\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Given a future-convex (resp. past-convex) surface $S$ with $\partial S=\Gamma$, $S$ must necessarily be contained in $\overline{\mathcal{D}_-(\Gamma)}$ (resp. $\overline{\mathcal{D}_+(\Gamma)}$), and in the closure of the complement of $\mathcal{C}(\Gamma)$.
If $\partial S=\Gamma$, the lightlike part of a future-convex surface $S$ (i.e. the subsurface of $S$ admitting lightlike support planes) necessarily contains $\partial_-\mathcal C(\Gamma)\cap \partial\mathcal D(\Gamma)$, that is, it contains all future-directed sawteeth. Analogously, the lightlike part of a past-convex surface $S$ contains $\partial_+\mathcal C(\Gamma)\cap \partial\mathcal D(\Gamma)$.
The convex surface $\partial_-\mathcal C$ has the property that its lightlike part coincides precisely with the union of all future-directed sawteeth. The same of course holds for $\partial_+\mathcal C$. We define $\Gamma_-$ as the boundary of the spacelike part of $\partial_-\mathcal C$.
The curve $\Gamma_-$ coincides with $\Gamma$ in the complement of future-directed sawteeth. If $\Gamma$ contains future-directed sawteeth, then $\Gamma_-$ contains the spacelike lines whose endpoints coincide with the endpoints of each sawtooth. The same description, switching future with past sawteeth, can be given for past-convex nowhere timelike surfaces. In particular we define $\Gamma_+$ as the boundary of the spacelike part of $\Gamma_+$. See also Figure \[fig:curvegamma\]. The $K$-surfaces which, in Theorem \[thm Ksurfaces ads intro\], form a foliation of the future (resp. past) connected component of $\mathcal D(\Gamma)\setminus \mathcal C(\Gamma)$ will have as boundary the curve $\Gamma_+$ (resp. $\Gamma_-$).
![If the curve $\Gamma$ contains a past-directed sawtooth, then $\Gamma_+$ coincides with $\Gamma$ in the complement of the sawtooth, whereas $\Gamma_+$ contains the spacelike line which is the third side of the lightlike triangle bounded by the sawtooth. \[fig:curvegamma\]](curvegamma-eps-converted-to.pdf){height="8cm"}
\[ex crucial\] An example of the above constructions, which will be crucial in Section \[sec barrier\] and \[sec existence\], is the following. Consider the curve $\Gamma'$ which is composed of the boundary of a totally geodesic half-plane and of a past-directed sawtooth. See Figure \[fig:example\]. Then the upper boundary of the convex hull is composed of the totally geodesic half-plane and of the lightlike triangle bounded by the past sawtooth. On the other hand, the lower boundary of the convex hull $\partial_-\mathcal{C}(\Gamma')$ is pleated along a lamination which foliates the hyperbolic surface $\partial_-\mathcal{C}(\Gamma')$.
It is not difficult to understand the domain of dependence $\mathcal D(\Gamma')$. In fact, the lower boundary of $\mathcal D(\Gamma')$ consists of the surface obtained by two totally geodesic lightlike half-planes which intersect along a spacelike half-geodesic ($L_{\ell,\ell'}$ in Figure \[fig:example\], left), glued to the cone over the end-point of the half-geodesic. On the other hand, the component $\mathcal D_+(\Gamma')$, in the affine chart of the right side of Figure \[fig:example\], is a region of a vertical cylinder, bounded by the horizontal plane containing $\Gamma'_+$, and by the lightlike plane containing the past-directed sawtooth.
![The two boundaries of the domain of dependence $\mathcal D(\Gamma)$, where $\Gamma$ is the curve of Example \[ex crucial\]. \[fig:example\]](example-eps-converted-to.pdf){height="7cm"}
![The two boundaries of the domain of dependence $\mathcal D(\Gamma)$, where $\Gamma$ is the curve of Example \[ex crucial\]. \[fig:example\]](example2-eps-converted-to.pdf){height="7cm"}
From differentiable surfaces in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ to local diffeomorphisms of ${\mathbb{H}}^2$ {#sec surfaces}
=================================================================================================================
The purpose of this section is to explain the construction which associates to a differentiable embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ a map between subsets of ${\mathbb{H}}^2$. Most of the results of this section have been already known and used before, see also the references mentioned in the introduction.
Left and right projections {#subsec: projections}
--------------------------
Given a differentiable spacelike surface $ S$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, one defines two projections $\pi_l: S\to {\mathbb{H}}^2$ and $\pi_r: S\to {\mathbb{H}}^2$ in the following way. If the timelike geodesic orthogonal to $S$ at $\gamma$ is $L_{x,x'}$, then $$\pi_l(\gamma)=x\qquad\qquad\pi_r(\gamma)=x'\,.$$
An equivalent construction of the projections is the following. Given a point $\gamma\in S$, let us denote by $P_\gamma S$ the totally geodesic plane tangent to $S$ at $\gamma$. Then, there exists a unique isometry $$\Pi^\gamma_l\in\{\mathrm{id}\}\times{\mathrm{Isom}}({\mathbb{H}}^2)$$ which maps $P_\gamma S$ to $\mathcal R_{\pi}$. In fact, if we denote $G(\gamma)=(P_\gamma S)^*$, then the isometry $(\mathrm{id}, G(\gamma))\in{\mathrm{Isom}}({\mathbb{H}}^2)\times {\mathrm{Isom}}({\mathbb{H}}^2)$ maps $G(\gamma)$ to $\mathrm{id}$, and therefore coincides with $\Pi^\gamma_l$. Analogously consider the unique $$\Pi^\gamma_r\in{\mathrm{Isom}}({\mathbb{H}}^2)\times\{\mathrm{id}\}$$ mapping $P_\gamma S$ to $\mathcal R_\pi$, which has the form $(G(\gamma)^{-1},\mathrm{id})$. We then have $$\Pi^\gamma_l(\gamma)=\gamma G(\gamma)^{-1}=\mathcal I_{\pi_l(\gamma)}\,,\qquad\qquad \Pi^\gamma_r(\gamma)=G(\gamma)^{-1}\gamma =\mathcal I_{\pi_r(\gamma)}\,.$$ Indeed, the previous formula is true if $\gamma=\mathcal I_{x}$, and $P_\gamma S=\mathcal R_\pi$, in which case the isometries $\Pi^\gamma_l$ and $\Pi^\gamma_r$ are trivial and the normal geodesic is $L_{x,x}$. The general case then follows by observing that, from Equation , the projection to the point $x$ of $L_{x,x'}$ is invariant by composition with a right isometry, and analogously $x'$ does not change when composing with an isometry on the left.
\[ex pleated earthquake\] The most basic example of left and right projections can be seen when $ S$ is a totally geodesic plane. In this case, the $P_\gamma S=S$ and thus $\pi_l$ (resp. $\pi_r$) is the restriction to $ S$ of the unique left (resp. right) isometry of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ sending $ S$ to $\mathcal R_\pi$ (up to the identification of $\mathcal R_\pi$ with ${\mathbb{H}}^2$). More generally, if $ S$ is a pleated surface, the left and right projections are well-defined for every point $\gamma\in S$ which admits only one support plane. Moreover, the isometries $\Pi_l^\gamma$ and $\Pi_r^\gamma$ are constant on each stratum of the bending lamination. The induced metric on $ S$ is the hyperbolic metric of a straight convex set $H$, and it can be proved (see [@Mess]) that the projections are earthquake maps from $H$ to ${\mathbb{H}}^2$.
Using Example \[ex pleated earthquake\], one easily proves that if $ S$ is convex, then $\pi_l$ and $\pi_r$ are injective.
\[proj injective convex\] Let $ S$ be an embedded differentiable convex spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Then $\pi_l$ and $\pi_r$ are injective.
By Lemma \[lemma:bandabassotti\] we may assume that $S$ is future convex. Pick two points $\gamma,\gamma'\in S$. By the hypothesis on $S$, the totally geodesic planes $P_\gamma S$ and $P_{\gamma'} S$ are spacelike and $S$ is contained in the intersection, say $K_1$, of the future half-spaces bounded by $P_\gamma S$ and $P_{\gamma'} S$, in some affine chart. Let $S_1$ be the boundary of $K_1$. It is either a totally geodesic plane or a pleated surface with pleating locus made by a single geodesic. Notice that $S_1$ is tangent to $S$ at $\gamma$ and $\gamma'$ so that the corresponding projections $\pi_l^1$ and $\pi_r^1$ coincide with $\pi_l$ and $\pi_r$ at $\gamma$ and $\gamma'$. As $\pi_l^1$ and $\pi_r^1$ are both injective, we conclude that $\pi_l(\gamma)\neq\pi_l(\gamma')$ and $\pi_r(\gamma)\neq\pi_r(\gamma')$.
It turns out that, if $S$ is oriented by means of the ambient orientation of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ and the choice of the future-directed normal vector field, then $\pi_l$ and $\pi_r$ are orientation-preserving.
When $ S$ is an embedded differentiable convex spacelike surface, one can define a bijective map from $\Omega_l=\pi_l( S)$ to $\Omega_r=\pi_r( S)$, by $$\Phi:\Omega_l\to\Omega_r\qquad \Phi=\pi_r\circ (\pi_l)^{-1}.$$
In [@bon_schl] it was proved that if $ S$ is a convex surface with $\partial S$ a curve $\Gamma\subset\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3=\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$ which is the graph of an orientation-preserving homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$, then $\Omega_l=\Omega_r={\mathbb{H}}^2$, and $\Phi$ extends to $\phi$ on $\partial{\mathbb{H}}^2$. This is essentially the content of the next lemma.
\[lemma extension\] Let $ S$ be a convex spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S=gr(\phi)$, for $\phi$ an orientation-preserving homeomorphism of $\partial{\mathbb{H}}^2$. If $\gamma_n\in S$ converges to $(p,q)\in\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, then $\pi_l(\gamma_n)$ converges to $p$ and $\pi_r(\gamma_n)$ converges to $q$.
In the case of pleated surfaces $ S$, as already observed by Mess [@Mess], the associated map $\Phi$ is an earthquake map. Hence if $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$ is an orientation-preserving homeomorphism, the boundaries of the convex hull of $gr(\phi)$ are pleated surfaces whose associated maps provide a left and a right earthquake map which extends $\phi$, hence recovering a theorem of Thurston [@thurstonearth].
\[remark normal flow\] The map $\Phi$ is constant under the normal evolution of a surface. More precisely, suppose $ S$ is a differentiable surface such that the parallel surface $ S_\epsilon$ is well-defined for a short time $\epsilon$. Then the normal timelike geodesics of $ S_\epsilon$ are the same as those of $ S$. Hence by the above equivalent definition, using the identification of $ S$ and $ S_\epsilon$ given by the normal flow, the maps $\pi_l$ and $\pi_r$ for $ S$ and $ S_\epsilon$ are the same. In particular, the composition $\Phi=\pi_r\circ (\pi_l)^{-1}$ is invariant for the normal evolution.
\[remark dual surface\] For the same reason, the left and right projections are the “same" for a convex differentiable surface and its dual surface. More precisely, if $S$ is a spacelike convex differentiable surface, then one can define the dual surface $S^*$ as the image of the map from $S$ to ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which associates to $\gamma\in S$ the dual point $(P_\gamma S)^*$. Recalling that $(P_\gamma S)^*$ is the midpoint of all timelike geodesics orthogonal to $P_\gamma S$, it turns out that $S^*$ is the $\pi/2$-parallel surface of $S$ and the identification between $S$ and $S^*$ is precisely the $\pi/2$-normal evolution. Therefore the area-preserving diffeomorphisms $\Phi$ associated with $S$ and $S^*$ coincide.
Relation with the extrinsic geometry of surfaces {#subsec surfaces}
------------------------------------------------
A smooth embedded surface $\sigma:\Sigma{\rightarrow}{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is called spacelike if its tangent plane is a spacelike plane at every point, so that the *first fundamental form* $I(v,w)=\langle d\sigma(v),d\sigma(w)\rangle$ is a Riemannian metric on $\Sigma$. Let $N$ be a unit normal vector field to the embedded surface $S=\sigma(\Sigma)$. We denote by $\nabla^{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}$ and $\nabla^I$ the ambient connection of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ and the Levi-Civita connection of the first fundamental form of the surface $\Sigma$, respectively. If $\sigma$ is $C^2$, the *second fundamental form* of $\Sigma$ is defined by the equation $$\nabla^{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}_{d\sigma(v)}d\sigma(\tilde w)=d\sigma\nabla^I_{v}\tilde w+{I\hspace{-0.1cm}I}(v,w)N$$ if $\tilde w$ is a vector field extending $w$, and $N$ is the future-directed unit normal vector field. The *shape operator* is the $(1,1)$-tensor defined as $$B(v)=(d\sigma)^{-1}\nabla^{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}_{d\sigma(v)} N\,.$$ It is a self-adjoint operator for $I$, such that $${I\hspace{-0.1cm}I}(v,w)=I( B(v),w)\,.$$ It satisfies the Codazzi equation $d^{\nabla^I}\! B=0$, where (for vector fields $\tilde v$ and $\tilde w$ on $S$ extending $v$ and $w$): $$d^{\nabla^I}\!B(v,w)=\nabla^I_{\tilde v} (B(\tilde w))-\nabla^I_{\tilde w} (B(\tilde v))-B[\tilde v,\tilde w]\,.$$ Moreover, $B$ satisfies the Gauss equation $$\label{Gauss}
K_I=-1-\det B\,,$$ where $K_I$ is the curvature of the first fundamental form.
In [@Schlenker-Krasnov], a formula for the pull-back on $S$ of the hyperbolic metric of ${\mathbb{H}}^2$ by the left and right projections was given. We report this formula here:
\[formula KS\] Let $S$ be a smooth embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Then: $$\label{eq pullback left}
\pi_l^*g_{{\mathbb{H}}^2}(v,w)=I((E+J_I B)v,(E+J_I B)w)\,,$$ and $$\label{eq pullback right}
\pi_r^*g_{{\mathbb{H}}^2}(v,w)=I((E-J_I B)v,(E-J_I B)w)\,.$$ Here $E$ denotes the identity operator, and $J_I$ is the almost-complex structure induced by $I$.
Observe that, since $B$ is self-adjoint for $I$, $\mathrm{tr}(J_I B)=0$ and thus $\det(E\pm J_I B)=1+\det B$. Hence if $\det B\neq -1$, then $\pi_l$ and $\pi_r$ are local diffeomorphisms. If moreover $S$ is convex ($\det B\geq 0$), then by Lemma \[proj injective convex\], $\pi_l, \pi_r$ are diffeomorphisms onto their images. Since $\det(E+J_I B)=\det(E-J_I B)$, the composition $\Phi=\pi_r\circ(\pi_l)^{-1}$ is area-preserving. In conclusion, we have:
If $S$ is a smooth convex embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, then $\pi_l$ and $\pi_r$ are diffeomorphisms on their images. Therefore, $\Phi=\pi_r\circ(\pi_l)^{-1}:\Omega_l\to\Omega_r$, where $\Omega_l=\pi_l(S)$ and $\Omega_r=\pi_r(S)$, is an area-preserving diffeomorphism.
Moreover, using Lemma \[lemma extension\], one gets:
If $S$ is a smooth convex embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S=gr(\phi)\subset\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, for $\phi$ an orientation-preserving diffeomorphism of $\partial{\mathbb{H}}^2$, then $\pi_l$ and $\pi_r$ are diffeomorphisms onto ${\mathbb{H}}^2$ and $\Phi=\pi_r\circ(\pi_l)^{-1}$ is an area-preserving diffeomorphism of ${\mathbb{H}}^2$ which extends to $\phi$ on $\partial{\mathbb{H}}^2$.
Let us now discuss some properties of the associated map $\Phi$ to a smooth convex embedded surface $S$. In fact, we will prove the following proposition:
\[prop properties associated map\] Let $S$ be a smooth strictly convex embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, and let $\Phi=\pi_r\circ(\pi_l)^{-1}:\Omega_l\to\Omega_r$ be the associated area-preserving map. Then there exists a smooth tensor $b\in\Gamma(\mathrm{End}(T\Omega_l))$ such that $$\label{eq pullback b}
\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b\cdot,b\cdot)\,,$$ which satisfies:
1. $d^\nabla b=0\,$;
2. $\det b=1\,$;
3. ${\mbox{\rm tr}\,}b\in(-2,2)\,$.
Here $\nabla$ is the Levi-Civita connection of ${\mathbb{H}}^2$.
Let us define $$\label{eq:emy}
b=(E+J_I B)^{-1}(E-J_I B)\,.$$ Then from Equations and , the condition in Equation is satisfied. Let us check the three properties of $b$:
1. Observe that, since $\nabla^I E$ and $\nabla^I J_I$ vanish and $B$ is Codazzi for $I$, then $E-J_I B$ satisfies $$d^{\nabla^I}\!(E-J_I B)=0\,.$$ Using a formula given in [@labourieCP] or [@Schlenker-Krasnov Proposition 3.12], the Levi-Civita connection of the metric $\pi_l^*g_{{\mathbb{H}}^2}$ is $$\nabla_v w=(E+J_I B)^{-1}\nabla^I_v (E+J_I B)w\,,$$ for every $v,w$. Hence: $$\begin{aligned}
d^{\nabla}&\!b(v, w)=\nabla_v b(w)-\nabla_w b(v)-b[v,w] \\
&=(E+J_I B)^{-1}(\nabla^I_v(E-J_I B)(w)-\nabla^I_w(E-J_I B)(v)-(E-J_I B)[v,w]) \\
&=(E+J_I B)^{-1}(d^{\nabla^I}\!(E-J_I B)(v,w))=0\,.\end{aligned}$$ This concludes the first point.
2. Since $B$ is self-adjoint for $I$, then ${\mbox{\rm tr}\,}J_I B=0$, and thus $$\det(E+J_I B)=\det(E-J_I B)=1+\det B\,.$$ Therefore $\det b=1$.
3. Observe that $$\label{eq:ely}
(E+J_I B)^{-1}=\frac{1}{1+\det B}(E-J_I B)\,.$$ In fact, $(E+J_I B)(E-J_I B)=E-(J_I B)^2$ and, since ${\mbox{\rm tr}\,}(J_I B)=0$, by the Cayley-Hamilton Theorem $(J_I B)^2=-\det(J_I B)E=-(\det B)E$. Moreover by a direct computation, $\mathrm{tr}(E-J_I B)^2={\mbox{\rm tr}\,}E+{\mbox{\rm tr}\,}(JB)^2=2(1-\det B)$ and thus:
$$\label{eq: gambadilegno}
{\mbox{\rm tr}\,}b={\mbox{\rm tr}\,}(E+J_I B)^{-1}(E-J_I B)=\frac{1}{1+\det B}{\mbox{\rm tr}\,}(E-J_I B)^2=2\left(\frac{1-\det B}{1+\det B}\right)\,.$$
Since $\det B>0$ by hypothesis, ${\mbox{\rm tr}\,}b$ is always contained in $(-2,2)$ for a strictly convex surface.
\[rmk:picodepaperis\] Using Equation and $J=(E+J_I B)^{-1} J_I (E+J_I B)$, we get $$Jb=J (E+J_I B)^{-1}(E-J_I B)=(E+J_I B)^{-1}J_I(E-J_I B)\,.$$ By applying again Equation , one obtains that $$\mathrm{tr}(Jb)=\frac{\mathrm{tr}(J_I(E-J_I B)^2)}{1+\det B}=\frac{2\mathrm{tr}B}{1+\det B}\,.$$ Since we have chosen the future-directed normal vector field to the surface $S$, this shows that $S$ is future-convex if $\mathrm{tr}(Jb)>0$, and past-convex if $\mathrm{tr}(Jb)<0$.
From $K$-surfaces to $\theta$-landslides
----------------------------------------
Recall that a smooth spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is a *K*-surface if its Gaussian curvature is constantly equal to $K$. In this paper we are interested in strictly convex surfaces, hence we will consider $K$-surfaces with $K\in(-\infty,-1)$, so that by the Gauss equation $$K_I=-1-\det B\,,$$ we will have $\det B>0$.
In [@bms] it is proved that, given a $K$-surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, for $K\in(-\infty,-1)$, the associated map $\Phi$ is a $\theta$-*landslide*. In [@bms] the definition of $\theta$-landslides is the following:
\[defi landslide\] Let $\Phi:\Omega\subseteq {\mathbb{H}}^2\to{\mathbb{H}}^2$ be an orientation-preserving diffeomorphism onto its image and let $\theta\in[0,\pi]$. We say that $\Phi$ is a $\theta$-landslide if there exists a bundle morphism $m\in\Gamma(\mathrm{End}(T\Omega))$ such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}((\cos\theta E+\sin\theta Jm)\cdot,(\cos\theta E+\sin\theta Jm)\cdot)\,,$$ and
- $d^{\nabla}m=0\,$;
- $\det m=1\,;$
- $m$ is positive self-adjoint for $g_{{\mathbb{H}}^2}\,$.
Let us observe that a $\theta$-landslide with $\theta=0,\pi$ is an isometry. In [@bms] it was proved that a $\theta$-landslide can be decomposed as $\Phi=f_2\circ (f_1)^{-1}$, where $f_1$ and $f_2$ are harmonic maps from a fixed Riemann surface, with Hopf differentials satisfying the relation $$\mathrm{Hopf}(f_1)=e^{2i \theta }\mathrm{Hopf}(f_2)\,.$$ We will restrict to the case of $\theta\in (0,\pi)$. In this case, we will mostly use an equivalent characterization of landslides:
\[char landslide\] Let $\Phi:\Omega\subseteq {\mathbb{H}}^2\to{\mathbb{H}}^2$ be an orientation-preserving diffeomorphism onto its image. Then $\Phi$ is a $\theta$-landslide if and only if there exists a bundle morphism $b\in\Gamma(\mathrm{End}(T\Omega))$ such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b\cdot,b\cdot)\,,$$ and
- $d^{\nabla}b=0\,$;
- $\det b=1\,;$
- ${\mbox{\rm tr}\,}b=2\cos\theta\,$;
- ${\mbox{\rm tr}\,}Jb< 0$.
Let us first suppose that $\Phi$ is a $\theta$-landslide and prove the existence of $b$. Let $b=\cos\theta E+\sin\theta Jm$, for $m$ as in Definition \[defi landslide\]. Then ${\mbox{\rm tr}\,}b=2\cos\theta$ since $m$ is self-adjoint for $g_{{\mathbb{H}}^2}$, and thus $Jm$ is traceless. On the other hand, $\mathrm{tr}(Jb)=-\sin\theta\,\mathrm{tr}\, m<0$ since $m$ is positive definite. Moreover, $\det b=\cos^2\theta+\sin^2\theta=1$ since $\det E=\det J=\det m=1$ and again $\mathrm{tr}(Jm)=0$. Finally, $d^{\nabla}b=0$ since $\nabla E=\nabla J=0$.
Conversely, let $b$ satisfy the conditions in the statement. Then $b-\cos\theta E$ is traceless and thus is of the form $a Jm$, with $m$ positive self-adjoint for $g_{{\mathbb{H}}^2}$, $\det m=1$, and $a\in{{\mathbb R}}$. Since $d^{\nabla}b=0$, one has $d^{\nabla}m=0$. Finally, $1=\det b=\cos^2\theta+a^2(\det m)=\cos^2\theta+a^2$ implies that $|a|=|\sin\theta|$. Imposing that ${\mbox{\rm tr}\,}Jb<0$, we conclude that $a=\sin\theta$.
Hence we are now able to reprove the following fact, which is known from [@bms]:
\[prop past-convex\] Let $S$ be a past-convex $K$-surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $K\in(-\infty,-1)$. Then the associated map $\Phi:\Omega_l\to\Omega_r$ is a $\theta$-landslide, with $$K=-\frac{1}{\cos^2(\frac{\theta}{2})}\,.$$
The proof follows, using the characterization of Lemma \[char landslide\], from the choice $$b=(E+J_I B)^{-1}(E-J_I B)$$ in Proposition \[prop properties associated map\]. In fact, the first two points are satisfied, and from Equation one obtains: $$\label{eq:tipetap}
{\mbox{\rm tr}\,}b=2\left(\frac{1-\det B}{1+\det B}\right)=2\cos\theta\,,$$ for some $\theta\in(0,\pi)$. As observed in Remark \[rmk:picodepaperis\], ${\mbox{\rm tr}\,}(Jb)<0$ since $S$ is past-convex by hypothesis. Finally, from Equation one obtains: $$\det B=\frac{1-\cos\theta}{1+\cos\theta}={\tan^2({\theta}/{2})}$$ and thus $$K=-1-\det B=-\frac{1}{\cos^2({\theta}/{2})}\,,$$ thus concluding the claim.
\[remark dual surface2\] There clearly is an analogous statement of Proposition \[prop past-convex\] for future-convex surfaces. In fact, if $S$ is future-convex $K$-surface, consider first the tensor $b'=(E+J_I B)^{-1}(E-J_I B)$ as in the proof of Proposition \[prop past-convex\]. In this case ${\mbox{\rm tr}\,}(Jb')>0$, as observed in Remark \[rmk:picodepaperis\].
Hence it is the choice $b=-(E+J_I B)^{-1}(E-J_I B)$ which makes the conditions in Lemma \[char landslide\] satisfied. This shows that the associated map $\Phi$ is a $\theta$-landslide where $\theta\in(0,\pi)$ is chosen so that $2\cos\theta=-{\mbox{\rm tr}\,}((E+J_I B)^{-1}(E-J_I B))$. The same computation as above shows that the curvature of $S$ is: $$K=-\frac{1}{\sin^2(\theta/2)}\,.$$
Let us remark that the result of this computation is consistent with the fact that the dual surface of the $K$-surface $S$ is a past-convex surface $S^*$, gives the same associated map $\Phi$, and has constant curvature $$K^*=-\frac{1}{\cos^2(\theta/2)}~.$$ In light of this remark, we will always restrict ourselves to consider past-convex $K$-surfaces.
A special case is obtained for $\theta=\pi/2$. Applying Lemma \[char landslide\], a $\pi/2$-landslide is a diffeomorphism $\Phi:\Omega\subseteq {\mathbb{H}}^2\to\Phi(\Omega)\subseteq{\mathbb{H}}^2$ such that $\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b\cdot,b\cdot)$, where:
- $d^{\nabla}b=0\,;$
- $\det b=1\,;$
- ${\mbox{\rm tr}\,}b=0\,;$
- ${\mbox{\rm tr}\,}Jb<0\,.$
Hence, by taking $b_0=-Jb$, we have ${\mbox{\rm tr}\,}(Jb_0)=0$, and thus $b_0$ has the property that $\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b_0\cdot,b_0\cdot)$ with:
- $d^{\nabla}b_0=0\,;$
- $\det b_0=1\,;$
- $b_0$ is positive self-adjoint for $g_{{\mathbb{H}}^2}\,$.
This is known to be equivalent to $\Phi$ being a *minimal Lagrangian map*, namely $\Phi$ is area-preserving and its graph is a minimal surface in ${\mathbb{H}}^2\times{\mathbb{H}}^2$.
In [@bon_schl], the existence of maximal surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ (namely, such that ${\mbox{\rm tr}\,}B=0$ where $B$ is the shape operator) was used to prove that every quasisymmetric homeomorphism of $\partial {\mathbb{H}}^2$ admits a unique minimal Lagrangian extension to ${\mathbb{H}}^2$. We remark that, given a smooth maximal surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with principal curvatures in $(-1,1)$, the two $\pi/4$-parallel surfaces are $(-2)$-surfaces (see [@bon_schl] or [@Schlenker-Krasnov]). As already discussed in Remarks \[remark normal flow\] and \[remark dual surface\], the map associated to the two parallel surfaces (which are dual to one another) is the same as the map associated to the original maximal surface.
A representation formula for convex surfaces {#sec representation formula}
============================================
The purpose of this section is to provide a *representation formula* for convex surfaces in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ in terms of the associated diffeomorphism $\Phi:\Omega\subseteq{\mathbb{H}}^2\to\Phi(\Omega)\subseteq{\mathbb{H}}^2$, for $\Omega$ a domain in ${\mathbb{H}}^2$. A basic observation to start with is the fact that, if $S$ is a smooth strictly convex embedded surface, with associated map $\Phi=\pi_r\circ(\pi_l)^{-1}$, and if $\gamma\in S$ is such that $\pi_l(\gamma)=x$, then $\gamma$ belongs to the timelike geodesic $$L_{x,\Phi(x)}=\{\eta\in{\mathrm{Isom}}({\mathbb{H}}^2):\eta(\Phi(x))=x \}\,.$$ In fact, from Subsection \[subsec: projections\], we know $L_{x,\Phi(x)}$ is precisely the timelike geodesic orthogonal to $S$ at $\gamma$.
Motivated by this observation, we will now define a 1-parameter family of parallel surfaces $S_{\Phi,b}$, all orthogonal to the lines $L_{x,\Phi(x)}$, which depend on the diffeomorphism $\Phi$ and on the choice of $b\in\Gamma(\mathrm{End}(T\Omega))$ satisfying the necessary conditions provided by Proposition \[prop properties associated map\]. The surface $S_{\Phi,b}$ will be parameterized by the inverse map of the left projection $\pi_l:S\to{\mathbb{H}}^2$.
\[defi sigma\] Let $\Omega$ be a domain in ${\mathbb{H}}^2$ and $\Phi:\Omega\subseteq{\mathbb{H}}^2\to\Phi(\Omega)\subseteq{\mathbb{H}}^2$ be an orientation-preserving diffeomorphism such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}\cdot,{b}\cdot)\,,$$ where ${b}\in\Gamma(\mathrm{End}(T\Omega))$ is a smooth bundle morphism such that $d^\nabla\! {b}=0$ and $\det {b}=1$. Then define the map $\sigma_{\Phi,{b}}:{\mathbb{H}}^2\to{\mathrm{Isom}}({\mathbb{H}}^2)$, where $\sigma_{\Phi,{b}}(x)$ is the unique isometry $\sigma$ such that
- $\sigma(\Phi(x))=x\,$;
- $d\sigma_{\Phi(x)}\circ d\Phi_x=-{b}_x\,$.
If $\Phi$ is the identity of ${\mathbb{H}}^2$ and $\cos\theta=1$, then clearly the identity operator ${b}=E$ satisfies the conditions $d^\nabla\!{b}=0$ and $\det{b}=1$. Hence by applying the definition, $\sigma_{\mathrm{id},E}(x)$ is the unique isometry which fixes $x$ and has differential $-E$. In conclusion, $\sigma_{\mathrm{id},E}:{\mathbb{H}}^2\to{\mathrm{Isom}}({\mathbb{H}}^2)$ is our usual identification which maps $x\in{\mathbb{H}}^2$ to $\mathcal{I}_x$.
\[remark parallel immersion\] Clearly the smooth tensor $b\in\Gamma(\mathrm{End}(T\Omega))$ is not uniquely determined. In fact, if ${b}'$ is another tensor satisfying $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}'\cdot,{b}'\cdot)\,,$$ with $\det {b}'=1$, then ${b}'_x=R^x_{\rho(x)}{b}_x$, where $\rho$ is a function and $R^x_{\rho}$ is the rotation of $\rho$ around the basepoint $x\in{\mathbb{H}}^2$. Then we claim that $d^\nabla {b}'=R_\rho (d\rho \wedge Jb+d^\nabla{b})$. Indeed, using that $\nabla R_\rho=d\rho\otimes JR_\rho$ we get
$$\label{eq:etabeta}
\begin{split}
d^\nabla (R_\rho b)(v,w)&=\nabla_v (R_\rho b)(w)-\nabla_w (R_\rho b)(v) \\
&=(\nabla_v R_\rho)\circ b(w)+R_\rho\circ(\nabla_v b)(w)-(\nabla_w R_\rho)\circ b(v)-R_\rho\circ(\nabla_w b)(v) \\
&= d\rho(v)JR_\rho b(w) -d\rho(w)JR_\rho b(v)+R_\rho ((\nabla_v b)(w)-(\nabla_w b)(v)) \\
&=R_\rho (d\rho(v)J b(w) -d\rho(w)J b(v)+d^\nabla b(v,w))\,,
\end{split}$$
Therefore, from the condition $d^\nabla\! {b}=0$, if ${b}'$ is another tensor as in the hypothesis of Definition \[defi sigma\], then (up to replacing $\rho$ with $2\rho$), ${b}'=R^x_{2\rho}{b}$, where $\rho$ is a constant.
In this case, one gets that $\sigma:=\sigma_{\Phi,{b}}(x)$ satisfies the condition $d\sigma_{\Phi(x)}=-{b}_x\circ (d\Phi_{x})^{-1}$, while $\sigma':=\sigma_{\Phi,{b}'}(x)$ satisfies $$d\sigma'_x=- {b}'\circ(d\Phi_{x})^{-1}=-R^x_{2\rho}\circ{b}\circ (d\Phi_x)^{-1}\,.$$ In other words, the map $\sigma_{\Phi,{b}'}$ is obtained by displacing the map $\sigma_{\Phi,{b}}$ along the timelike lines $L_{x,\Phi(x)}$ of a constant length $\rho$.
The next lemma discuss the naturality of the construction of the map $\sigma_{\Phi,{b}}$.
\[lemma composition rule\] Let $b\in\Gamma(\mathrm{End}(T\Omega))$ be a smooth bundle morphism such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}\cdot,{b}\cdot)\,,$$ with $d^\nabla\! {b}=0$ and $\det {b}=1$. Let $\alpha,\beta$ be isometries of ${\mathbb{H}}^2$. Then:
- The smooth tensor ${b}'$ on $\beta(\Omega)$ defined by ${b}'_{\beta x}=d\beta_{x}\circ{b}_{x}\circ (d\beta_x)^{-1}$ is such that $(\alpha\Phi\beta^{-1})^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b'\cdot,b'\cdot)$, $d^\nabla\! {b}'=0$ and $\det b'=1$;
- $\sigma_{\alpha\circ\Phi\circ\beta^{-1},{b}'}(\beta(x))=\beta\circ\sigma_{\Phi,{b}}(x)\circ\alpha^{-1}=(\beta,\alpha)\cdot \sigma_{\Phi,{b}}(x)\,.$
For the first point, since $\beta$ is an isometry, one has: $$\begin{aligned}
(\alpha\Phi\beta^{-1})^*g_{{\mathbb{H}}^2}(v,w)&=g_{{\mathbb{H}}^2}(d(\alpha\Phi\beta^{-1})(v),d(\alpha\Phi\beta^{-1})(w))\\
&=g_{{\mathbb{H}}^2}(d(\Phi\beta^{-1})(v),d(\Phi\beta^{-1})(w))=g_{{\mathbb{H}}^2}({b}'(v),{b}'(w))\,.\end{aligned}$$ Moreover, $$\begin{aligned}
d^\nabla\!{b}'&=\nabla_v (d\beta{b}\,d\beta^{-1}(w))-\nabla_w (d\beta{b}\, d\beta^{-1}(v))-d\beta{b}\, d\beta^{-1}[v,w] \\
&=d\beta \left(\nabla_{d\beta^{-1}(v)}({b}\, d\beta^{-1}(w))-d\beta\nabla_{d\beta^{-1}(w)}({b}\, d\beta^{-1}(v))-{b}[d\beta^{-1}(v),d\beta^{-1}(w)]\right) \\
&=d\beta d^\nabla\!{b}(d\beta^{-1}(v),d\beta^{-1}(w))=0\,.\end{aligned}$$ For the second point, we must check that $\hat\sigma=\beta\circ\sigma_{\Phi,{b}}(x)\circ\alpha^{-1}$ satisfies the two properties defining $\sigma_{\alpha\circ\Phi\circ\beta^{-1},b'}(\beta(x))$, namely $\hat\sigma(\alpha\Phi\beta^{-1}(\beta(x)))=\beta(x)$ and $d\hat\sigma_{\alpha\Phi(x)}\circ d(\alpha\Phi\beta^{-1})_{\beta(x)}=-{b}'_{\beta(x)}$.
For the first defining property, we have that for every $x\in\Omega$: $$(\beta\circ\sigma_{\Phi,{b}}(x)\circ\alpha^{-1})(\alpha\circ\Phi\circ\beta^{-1})(\beta(x))=\beta(\sigma_{\Phi,{b}}(x) \Phi(x))=\beta(x)\,.$$ For the second property, $$\begin{aligned}
d(\beta\sigma_{\Phi}(x)\alpha^{-1})_{\alpha\Phi(x)}\circ d(\alpha\Phi\beta^{-1})_{\beta x}&=d\beta_x \circ d(\sigma_{\Phi}(x))_{\Phi(x)}\circ d\Phi_{x}\circ d\beta^{-1}_{\beta (x)} \\
&=-d\beta_x \circ{b}_{x}\circ (d\beta_x)^{-1}=
-{b}'_{\beta(x)}\,.\end{aligned}$$ This concludes the proof.
Let us now compute the pull-back of the induced metric on the surface $S_{\Phi,{b}}=\sigma_{\Phi,{b}}(\Omega)$ by means of $\sigma_{\Phi,{b}}$.
\[prop first ff\] Let $\Omega$ be a domain in ${\mathbb{H}}^2$ and $\Phi:\Omega\subseteq{\mathbb{H}}^2\to\Phi(\Omega)\subseteq{\mathbb{H}}^2$ be an orientation-preserving diffeomorphism such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}\cdot,{b}\cdot)\,,$$ where ${b}\in\Gamma(\mathrm{End}(T\Omega))$ is a smooth bundle morphism such that $d^\nabla\! {b}=0$ and $\det {b}=1$. Then $$\label{firstff}
(\sigma_{\Phi,{b}})^* g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}=\frac{1}{4} g_{{\mathbb{H}}^2}((E+{b})\cdot,(E+{b})\cdot)\,.$$
In this proof, we will consider ${\mathbb{H}}^2$ in the hyperboloid model (recall Subsection \[subsec models\]), so that ${\mathrm{Isom}}({\mathbb{H}}^2)$ is identified to the connected component of the identity in ${\mathrm{SO}}(2,1)$.
Let us fix a point $x\in{\mathbb{H}}^2$. By post-composing $\Phi$ with an isometry $\alpha$ (where actually $\alpha=\sigma_{\Phi,b}(x)$), we can assume that $\Phi(x)=x$ and that $d\Phi_x=-{b}$. Indeed, by Lemma \[lemma composition rule\], $\sigma_{\Phi,{b}}$ is modified by right multiplication with $\alpha^{-1}$ and thus the embedding data are unchanged. Therefore by Definition \[defi sigma\], $\sigma_{\Phi,{b}}(x)=\mathrm{id}$ since $\sigma_{\Phi,{b}}(x)$ fixes $x$ and its differential at $x$ is the identity.
Consider a smooth path $x(t)$ with $x(0)=x$ and $\dot x(0)=v$. Hence, regarding the isometries of ${\mathbb{H}}^2$ as ${\mathrm{SO}}(2,1)$-matrices and points of ${\mathbb{H}}^2$ as vectors in ${{\mathbb R}}^{2,1}$, by differentiating at $t=0$ the relation $$\sigma_{\Phi,{b}}(x(t))\cdot \Phi(x(t))=x(t)\,,$$ we get $$d\sigma_{\Phi,{b}}(v)\cdot x- {b}(v)=v\,,$$ and therefore $$\label{relation dsigma1}
d\sigma_{\Phi,\rho}(v)\cdot x=(E+{b})v\,.$$ On the other hand, choose a vector $w\in T_x{\mathbb{H}}^2$ and extend $w$ to a parallel vector field $w(t)$ along $x(t)$. We start from the following relation (recall that we consider ${\mathrm{Isom}}({\mathbb{H}}^2)$ in the ${\mathrm{SO}}(2,1)$ model, hence the differential of an isometry $\sigma$ is $\sigma$ itself): $$\sigma_{\Phi,{b}}(x(t))\cdot d\Phi_{x(t)}(w(t))=- {b}_{x(t)}(w(t))\,.$$ By differentiating this identity in ${{\mathbb R}}^{2,1}$, we obtain $$\label{equality differentiate}
-d\sigma_{\Phi,{b}}(v)\cdot ({b}_x w)+{\left.\frac{d}{dt}\right|_{t=0}}( d\Phi_{x(t)}(w(t)))=-{\left.\frac{d}{dt}\right|_{t=0}}({b}_{x(t)}w(t))\,.$$ Now observe that, if $\nabla$ is the Levi-Civita connection of ${\mathbb{H}}^2$ and $\langle\cdot,\cdot\rangle_{{{\mathbb R}}^{2,1}}$ is the Minkowski product of ${{\mathbb R}}^{2,1}$, then $${\left.\frac{d}{dt}\right|_{t=0}}({b}_{x(t)}w(t))=\nabla_v({b}_{x(t)}w(t))+(\langle v,{b}(w)\rangle_{{{\mathbb R}}^{2,1}}) x\,.$$ For the other term, we obtain in a similar fashion that $${\left.\frac{d}{dt}\right|_{t=0}}( d\Phi_{x(t)}(w(t)))=\nabla_{d\Phi_x(v)}(d\Phi_{x(t)}w(t))+(\langle {b}(v),{b}(w)\rangle_{{{\mathbb R}}^{2,1}}) x\,,$$ In the second term, we have used that $d\Phi_x=-{b}$. Since $\Phi$ is an isometry between $g_{{\mathbb{H}}^2}$ and $\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}\cdot,{b}\cdot)$, we have $$\nabla_{d\Phi_x(v)}(d\Phi_{x(t)}w(t))=d\Phi_x\widetilde \nabla_{v}(w(t))=-{b}_x \widetilde \nabla_{v}(w(t))\,,$$ where $\widetilde \nabla$ is the Levi-Civita connection of $\Phi^*g_{{\mathbb{H}}^2}$. Using again the formula given in [@labourieCP] or [@Schlenker-Krasnov Proposition 3.12], it turns out that for any two vector fields $X,Y$ $$\widetilde \nabla_X Y={b}^{-1}\nabla_X({b}(Y))\,.$$ Therefore, from Equation we obtain $$-d\sigma_{\Phi,{b}}(v)\cdot ({b}_x w)-\nabla_v({b}_{x(t)}w(t))+(\langle {b}_x v,{b}_x w\rangle_{{{\mathbb R}}^{2,1}}) x=-\nabla_v({b}_{x(t)}w(t))-(\langle v,{b}_x w\rangle_{{{\mathbb R}}^{2,1}}) x$$ Therefore, replacing $b_x w$ with an arbitrary vector $w$, we obtain the relation $$\label{relation dsigma2}
d\sigma_{\Phi,{b}}(v)\cdot w=(\langle w,(E+{b}) v\rangle _{{{\mathbb R}}^{2,1}})x\,.$$ It is straightforward to check, using Equations and , that $d\sigma_{\Phi,\rho}(v)$ corresponds to the linear map $$\label{eq differential sigma}
d\sigma_{\Phi,b}(v)(w)=J(E+{b})v\boxtimes w =\Lambda(J(E+{b})(v))(w)\,,$$ where $J$ is the almost-complex structure of ${\mathbb{H}}^2$. Recalling from Equation that the identification of ${\mathfrak{so}}(2,1)$ with ${{\mathbb R}}^{2,1}$ sends the AdS metric of ${\mathrm{Isom}}({\mathbb{H}}^2)$ to $1/4$ the Minkowski product, the first fundamental form is: $$\begin{aligned}
g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}(d\sigma_{\Phi,{b}}(v_1),d\sigma_{\Phi,{b}}(v_2))&=\frac{1}{4}\langle J(E+{b})v_1,J(E+{b})v_2\rangle_{{{\mathbb R}}^{2,1}} \\
&=\frac{1}{4}\langle (E+{b})v_1,(E+{b})v_2\rangle_{{{\mathbb R}}^{2,1}}\,,\end{aligned}$$ as in our claim.
\[remark immersion trace condition\] From the proof of Proposition \[prop first ff\], it turns out that, under the usual identification, the differential of $\sigma_{\Phi,{b}}$ is given by $J(E+{b})$. Hence the map $\sigma_{\Phi,{b}}$ is an immersion provided $$\det d \sigma_{\Phi,{b}}=2+{\mbox{\rm tr}\,}{b}\neq 0\,,$$ or equivalently ${\mbox{\rm tr}\,}{b}\neq -2$.
\[cor orthogonal lines\] In the hypothesis of Proposition \[prop first ff\], the image of $d\sigma_{\Phi,b}$ at a point $x\in\Omega$ is orthogonal to the line $L_{x,\Phi(x)}$ at the point $\sigma_{\Phi,b}(x)$. In particular, if $x$ is not a critical point, then the family of lines $L_{\cdot,\Phi(\cdot)}$ foliates a neighborhood of $\sigma_{\Phi,b}(x)$ and the map $\sigma_{\Phi,b}$ gives a local parameterization of an integral surface of the orthogonal distribution.
Moreover, if $\sigma_{\Phi,{b}}$ is an immersion, then $\pi_l\circ\sigma_{\Phi,{b}}$ is the identity, and $\pi_r\circ\sigma_{\Phi,{b}}=\Phi$.
As in the proof of Proposition \[prop first ff\], we can assume that $\Phi(x)=x$ and that $d\Phi_x=-{b}$, which implies that $\sigma_{\Phi,b}(x)=\mathrm{id}$. From Equation , it follows that the image of $d\sigma_{\Phi,b}$ at $x$ is orthogonal to the direction given by $\Lambda(x)$. On the other hand, in this assumption $L_{x,\Phi(x)}=L_{x,x}$ is the 1-parameter subgroup generated by $\Lambda(x)$ itself.
\[rmk unit vector\] From the proof Proposition \[prop first ff\] we have actually shown that, under the assumption $\Phi(x)=x$ and that $d\Phi_x=- {b}$, if $d\sigma_{\Phi,{b}}$ is non-singular at $x$, then under the identification $T_{\mathrm{id}}{\mathrm{Isom}}({\mathbb{H}}^2)\cong{\mathfrak{so}}(2,1)\cong{{\mathbb R}}^{2,1}$, the future unit normal vector at $\sigma_{\Phi,{b}}(x)=\mathrm{id}$ is $N(x)=2\Lambda(x)$. In fact, $N(x)=2\Lambda(x)$ is orthogonal to the surface at $\sigma_{\Phi,{b}}(x)$ by the above observation, and is unit, since the ${\mathbb{A}\mathrm{d}\mathbb{S}}$ metric at $\mathrm{id}$ is identified to $1/4$ the Minkowski product, see Equation .
\[prop sec ff\] Let $\Omega$ be a domain in ${\mathbb{H}}^2$ and $\Phi:\Omega\subseteq{\mathbb{H}}^2\to\Phi(\Omega)\subseteq{\mathbb{H}}^2$ be an orientation-preserving diffeomorphism such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}({b}\cdot,{b}\cdot)\,,$$ where ${b}\in\Gamma(\mathrm{End}(T\Omega))$ is a smooth bundle morphism such that $d^\nabla\! {b}=0$ and $\det {b}=1$. Then the shape operator of the immersion $\sigma_{\Phi,{b}}$ is: $$\label{formula shape operator sigma}
B=-J_I(E+{b})^{-1}(E-{b})\,,$$ where $J_I$ denotes the almost-complex structure associated to the induced metric $I$.
To compute the covariant derivative of the normal vector field $N$ along a tangent vector $d\sigma_\Phi(v)$, we use the following formula (see [@milnor_curvature]): $$\nabla^{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}_{d\sigma_{\Phi,{b}}(v)} N=\nabla^l_{d\sigma_{\Phi,{b}}(v)} N+\frac{1}{2}[d\sigma_{\Phi,{b}}(v),N]=\nabla^r_{d\sigma_{\Phi,{b}}(v)} N-\frac{1}{2}[d\sigma_{\Phi,{b}}(v),N]\,,$$ where $\nabla^l$ (resp. $\nabla^r$) is the left-invariant (resp. right-invariant) flat connection which makes the left-invariant (resp. right-invariant) vector fields parallel, and $[\cdot,\cdot]$ denotes the bi-invariant extension of Lie bracket on $\mathfrak{isom}({\mathbb{H}}^2)$. Assuming that $\sigma_{\Phi,{b}}(x)=\mathrm{id}$ as before, we can compute $\nabla^r_{d\sigma_{\Phi,{b}}(v)} N$ as the differential at the identity of the vector field $N\sigma_{\Phi,{b}}(x)^{-1}$. Observe that in the argument of Proposition \[prop first ff\] and Remark \[rmk unit vector\], while doing the assumption that $\Phi(x)=x$ and that $d\Phi_x=-{b}$ (so that $\sigma_{\Phi,{b}}(x)=\mathrm{id}$), one is actually composing with an isometry of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ of the form $(\mathrm{id},\alpha)$, where $\alpha=\sigma_{\Phi,{b}}(x)$. Thus the formula $N(x)\sigma_{\Phi,{b}}(x)^{-1}=2\Lambda(x)$ is still true for every point $x$. Recall that, under the identification of the Lie algebra ${\mathfrak{so}}(2,1)$ with ${{\mathbb R}}^{2,1}$, the Lie bracket $[v,w]$ is identified to the Minkowski product $v\boxtimes w$ (see Equation ). Therefore we get, using Equation , $$\nabla^{{\mathbb{A}\mathrm{d}\mathbb{S}}^3}_{d\sigma_{\Phi,{b}}(v)} N=\Lambda\left(2v-\frac{1}{2}J(E+{b})v\boxtimes 2x\right)=\Lambda\left(2v-v-{b}(v)\right)=\Lambda\left((E-{b})v\right)\,,$$ where we have used Equation . Therefore, using Equation , that is, $d\sigma_{\Phi,{b}}(v)=\Lambda(J(E+{b})v)$, one concludes that $$B=-(E+{b})^{-1}J(E-{b})\,.$$ Using that, from Equation for the metric, $J_I=(E+{b})^{-1}J(E+{b})$, one concludes the proof.
Observe that, if $\Phi$ is a $\theta$-landslide, then by Lemma \[char landslide\] we can choose ${b}$ as in the hypothesis, with moreover ${\mbox{\rm tr}\,}{b}=2\cos\theta$. Hence by Equation , the curvature of $\sigma_{\Phi,{b}}(\Omega)$ is $$\det B=\frac{2-{\mbox{\rm tr}\,}{b}}{2+{\mbox{\rm tr}\,}{b}}=\frac{1-\cos\theta}{1+\cos\theta}=\tan^2\left(\frac{\theta}{2}\right)\,,$$ and thus $\sigma_{\Phi,{b}}$ is a $K$-surface for $K=-1/\cos^2(\theta/2)$.
More generally, we can now prove:
Let $S$ be a smooth strictly convex embedded spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, and let $\Phi=\pi_r\circ(\pi_l)^{-1}:\Omega_l\to\Omega_r$ be the associated map. By identifying $g_{{\mathbb{H}}^2}$ with the metric $I((E+J_I B)\cdot,(E+J_I B)\cdot)$ by means of the left projection $\pi_l$, and choosing $${b}=(E+J_I B)^{-1}(E-J_I B)\,,$$ then the map $\sigma_{\Phi,{b}}:\Omega_l\to{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ defined by $\sigma_{\Phi,{b}}(x)=\sigma$, with:
- $\sigma(\Phi(x))=x\,$;
- $d\sigma_{\Phi(x)}\circ d\Phi_{x}=-{b}_x\,$.
is an embedding whose image is the original surface $S$.
Recall from Proposition \[prop properties associated map\] that ${b}$ satisfies the hypothesis $d^{\nabla}{b}=0$ and $\det{b}=1$. Moreover, since $S$ is strictly convex, then ${\mbox{\rm tr}\,}{b}\in(-2,2)$ and thus $\sigma_{\Phi,{b}}$ is an immersion by Remark \[remark immersion trace condition\]. We will check that the embedding data of $\sigma_{\Phi,{b}}\circ \pi_l$ coincide with those of $S$, and thus in principle $\sigma_{\Phi,{b}}\circ \pi_l$ extends to a global isometry $(\gamma_l,\gamma_r)$ of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. By Corollary \[cor orthogonal lines\], if $\pi_l(\gamma)=x$, then the point $\sigma_{\Phi,{b}}(x)$ belongs to the geodesic $L_{x,\Phi(x)}$ which is orthogonal to $S$ at $\gamma$, and is also orthogonal to the image of $\sigma_{\Phi,{b}}$. So we deduce that $(\gamma_l,\gamma_r)$ preserves every geodesic of the form $L_{x,\Phi(x)}$. Hence $\gamma_l$ fixes every point $x\in\Omega_l$, and analogously $\gamma_r$ fixes every point $\Phi(x)$. Therefore $\gamma_l=\gamma_r=\mathrm{id}$ and this will conclude the proof.
To check the claim of the embedding data, using Equation , $$\pi_l^* \sigma_{\Phi,{b}}^*(g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3})=\frac{1}{4}\pi_l^*g_{{\mathbb{H}}^2}((E+{b})\cdot,(E+{b})\cdot)=\frac{1}{4}I((E+J_I B)(E+{b})\cdot,(E+J_I B)(E+{b})\cdot)\,.$$ On the other hand, $$\begin{aligned}
(E+J_I B)(E+{b})&=(E+J_I B)(E+(E+J_I B)^{-1}(E-J_I B)) \\
&=(E+J_I B)(E+J_I B)^{-1}(E+J_I B+E-J_I B)=2E\,.\end{aligned}$$ In a similar way, according to Equation , the shape operator of $\sigma_{\Phi,{b}}\circ\pi_l$ is: $$\begin{aligned}
-J_I(E+{b})^{-1}(E-{b})&=J_I(E+(E+J_I B)^{-1}(E-J_I B))^{-1}(E-(E+J_I B)^{-1}(E-J_I B)) \\
&=J_I(2(E+J_I B)^{-1})^{-1}(-2(E+J_I B)^{-1}J_I B)=-J_I^{2}B\,,\end{aligned}$$ and thus it coincides with $B$, the shape operator of the original surface $S$. This concludes the proof.
As already observed, the condition that ${\mbox{\rm tr}\,}{b}$ is a constant in $(-2,2)$ coincides with the condition that $S$ has constant curvature in $(-\infty,-1)$. Hence we obtain the following corollary:
\[cor reconstruct ksurf\] Let $\Omega$ be a domain in ${\mathbb{H}}^2$ and $\Phi:\Omega\subseteq{\mathbb{H}}^2\to{\mathbb{H}}^2$ be a $\theta$-landslide with $\theta\in(0,\pi)$. If $b\in\Gamma(\mathrm{End}(T\Omega))$ satisfies $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b\cdot,b\cdot)\,,$$ with $d^\nabla b=0$, $\det b=1$, ${\mbox{\rm tr}\,}b=2\cos\theta$ and ${\mbox{\rm tr}\,}Jb<0$, then the map $\sigma_{\Phi,b}$ defined by:
- $\sigma(\Phi(x))=x\,$;
- $d\sigma_{\Phi(x)}\circ d\Phi_{x}=-{b}_x\,$.
is the embedding of the past-convex $K$-surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ (for $K=-1/\cos^2(\theta/2)$) such that $\pi_l\circ \sigma_{\Phi,b}=\mathrm{id}$ and $\pi_r\circ \sigma_{\Phi,b}=\Phi$.
Applying Remark \[remark parallel immersion\], the map $\sigma_{\Phi,-b}$, which is defined by:
- $\sigma(\Phi(x))=x\,$;
- $d\sigma_{\Phi(x)}\circ d\Phi_{x}={b}_x\,$.
corresponds to the dual embedding, which is a $K^*$-surface, with $K^*=-K/(K+1)$, whose associated landslide map $\Phi$ is the same.
Although this is not the main point of this paper, we observe that this representation formula holds also for maximal surfaces. In fact, if $\Phi$ is a minimal Lagrangian map (which is the same as a $(\pi/2)$-landslide, as already observed), then there exists a smooth $b_0\in\Gamma(\mathrm{End}(T\Omega))$ such that $\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b_0\cdot,b_0\cdot)$ with
- $d^{\nabla}b_0=0\,;$
- $\det b_0=1\,;$
- $b_0$ is self-adjoint for $g_{{\mathbb{H}}^2}\,$.
Moreover, one can assume $b_0$ is positive definite (up to replacing $b_0$ with $-b_0$). The condition that $b_0$ is positive definite ensures that ${\mbox{\rm tr}\,}b_0\neq -2$, and thus $\sigma_{\Phi,b_0}$ is an immersion. Moreover, by a direct computation (see [@bon_schl §3]), $B$ is traceless, that is, $\sigma_{\Phi,b_0}(\Omega)$ is a maximal surface. In fact, in [@bon_schl] the existence of maximal surfaces was proved in order to obtain results on the existence of minimal Lagrangian extensions. Our next corollary goes in the opposite direction.
Let $\Omega$ be a open domain in ${\mathbb{H}}^2$ and $\Phi:\Omega\subseteq{\mathbb{H}}^2\to{\mathbb{H}}^2$ be a minimal Lagrangian map such that $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b_0\cdot,b_0\cdot)\,,$$ so that $b_0$ is positive definite, $g_{{\mathbb{H}}^2}$-self-adjoint, $d^\nabla b_0=0$ and $\det b_0=1$. Then the map $\sigma_{\Phi,b_0}$ defined by:
- $\sigma(\Phi(x))=x\,$;
- $d\sigma_{\Phi(x)}\circ d\Phi_{x}=-({b}_0)_x\,$.
is the embedding of the maximal surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ such that $\pi_l\circ \sigma_{\Phi,b_0}=\mathrm{id}$ and $\pi_r\circ \sigma_{\Phi,b_0}=\Phi$.
In fact, as it is to be expected, considering $b=\pm Jb_0$ one recovers the $K$-surfaces which are $(\pi/4)$-parallel surfaces to the maximal surface.
The explicit construction of a barrier {#sec barrier}
======================================
In this section we will construct an explicit example of $\theta$-landslides between hyperbolic surfaces, which commute with a 1-parameter hyperbolic group of isometries.
We will use the upper half-plane model of ${\mathbb{H}}^2$, and will denote by $z=x+iy$ the standard coordinates of the upper half-plane. Let us introduce a new coordinate $w=s+it$, with $w\in {{\mathbb R}}\times (-\pi/2,\pi/2)$, defined by $$z=i\exp(w)\,.$$ (See also [@hubbardbook Chapter 2].) Clearly $w$ is a conformal coordinate, and the hyperbolic metric takes the form $$\label{eq:edera}
h_0=\frac{|dz|^2}{y^2}=\frac{|dw|^2}{\cos^2(t)}\,.$$ In this coordinates, the line $l=\{t=0\}$ is a geodesic with endpoints $0,\infty\in {{{\mathbb R}}\mathrm{P}}^1$. The isometries preserving the geodesic $l$ have the form $\gamma_a(s,t)=(s+a,t)$. Moreover, the lines $\{s=s_0\}$ are geodesics orthogonal to $l$. See Figure \[fig:coordinateshp\].
![In the $w=s+it$ on the upper half-plane, the curves $\{t=t_0\}$ are equidistant curves from the geodesic $\{t=0\}$. The curves $\{s=s_0\}$ are geodesics orthogonal to $\{t=0\}$. \[fig:coordinateshp\]](coordinateshp-eps-converted-to.pdf){height="4cm"}
We will look for a $\theta$-landslide $\Phi_\theta$ invariant for the 1-parameter group $\gamma_a$. More precisely, we require that $\Phi_\theta\circ\gamma_a=\gamma_a\circ\Phi_\theta$. Hence $\Phi_\theta$ will necessarily have the form $$\Phi_\theta(s,t)=(s-{\varphi}(t),\psi(t))\,.$$ We want to impose that $\Phi_\theta$ is an orientation-preserving diffeomorphism, hence we will assume $\psi$ increasing. Thus we will always consider $\psi'>0$.
We will make use of the following remark:
\[rmk varpi\] Given an orientation-preserving local diffeomorphism $\Phi:\Omega\subseteq {\mathbb{H}}^2\to{\mathbb{H}}^2$, there exists a unique bundle morphism $b_0\in\Gamma(\mathrm{End}(T{\mathbb{H}}^2))$ with $$\Phi^*g_{{\mathbb{H}}^2}=g_{{\mathbb{H}}^2}(b_0\cdot,b_0\cdot)\,,$$ such that $b_0$ is positive definite and self-adjoint for $g_{{\mathbb{H}}^2}$ (which is equivalent to $\mathrm{tr}Jb_0=0$). Indeed, $b_0$ is the square root of $(g_{{\mathbb{H}}^2})^{-1}\Phi^*g_{{\mathbb{H}}^2}$. In general, $\Phi^*g_{{\mathbb{H}}^2}$ is equal to $g_{{\mathbb{H}}^2}(b\cdot,b\cdot)$, for a (smooth) bundle morphism $b$, if and only if $b$ is of the form $R_{\rho}b_0$ for some smooth function $\rho$.
Hence, let us begin by computing $$\begin{aligned}
\Phi_\theta^*h_0=&\frac{1}{\cos^2(\psi(t))}\left((ds-{\varphi}'(t)dt)^2+\psi'(t)^2dt^2\right) \\
=&\frac{ds^2-2{\varphi}'(t)dsdt+({\varphi}'(t)^2+\psi'(t)^2)dt^2}{\cos^2(\psi(t))}\,.\end{aligned}$$ Therefore we have $\Phi_\theta^*h_0=h_0(b_0\cdot,b_0\cdot)=h_0(\cdot,b_0^2)$, where $b_0$ is self-adjoint for $h_0$, positive definite, and in the $(s,t)$-coordinates $$b_0^2=h_0^{-1}\Phi_\theta^*h_0=\frac{\cos^2(t)}{\cos^2(\psi(t))}\begin{pmatrix} 1 & -{\varphi}'(t) \\ -{\varphi}'(t) & {\varphi}'(t)^2+\psi'(t)^2 \end{pmatrix}\,.$$ Let us observe that $$\det b_0^2=\frac{\cos^4(t)}{\cos^4(\psi(t))}\psi'(t)^2\,.$$ Since we are imposing that $\Phi_\theta$ is area-preserving, we must require $\det b=1$, which is equivalent to $\det b_0=\det b_0^2=1$. We obtain (assuming $\psi'(t)>0$): $$\frac{\psi'(t)}{\cos^2(\psi(t))}=\frac{1}{\cos^2(t)}\,,$$ which leads to the condition $\tan(\psi(t))=\tan t+C$ (after a choice of the sign).
In the following, we choose $C=0$, thus $\psi(t)=t$ and we look for a map of the form $\Phi(s,t)=(s-{\varphi}(t),t)$.
Observe that, for the Cayley-Hamilton theorem, $b_0^2-({\mbox{\rm tr}\,}b_0)b_0+(\det b_0)E=0$. Hence we have $b_0^2+E=({\mbox{\rm tr}\,}b_0)b_0$ and, taking the trace, $({\mbox{\rm tr}\,}b_0)^2=2+{\mbox{\rm tr}\,}(b_0^2)$. In our case, ${\mbox{\rm tr}\,}(b_0^2)=2+{\varphi}'(t)^2$ and therefore we obtain the following positive definite root of $b_0^2$: $$b_0=\frac{1}{\sqrt{4+{\varphi}'(t)^2}}(b_0^2+E)=\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\begin{pmatrix} 2 & -{\varphi}'(t) \\ -{\varphi}'(t) & 2+{\varphi}'(t)^2 \end{pmatrix}\,.$$
From Remark \[rmk varpi\], it remains to impose that $b=R_{\rho}b_0$ satisfies ${\mbox{\rm tr}\,}b=2\cos\theta$, ${\mbox{\rm tr}\,}Jb<0$ and the Codazzi condition for $g_{{\mathbb{H}}^2}$.
The condition ${\mbox{\rm tr}\,}b=2\cos\theta$ is equivalent to ${\mbox{\rm tr}\,}(R_{\rho}b_0)=2\cos\theta$. Observe that $$R_{\rho}b_0=(\cos\rho E+\sin\rho J)b_0=(\cos\rho)b_0+(\sin\rho)Jb_0~.$$ Since $b_0$ is self-adjoint, we have ${\mbox{\rm tr}\,}(Jb_0)=0$, hence the following formula holds: $$\mathrm{tr}(R_{\rho}b_0)=\cos\rho\,{\mbox{\rm tr}\,}b_0~.$$ Using this formula, the condition ${\mbox{\rm tr}\,}(R_{\rho}b_0)=2\cos\theta$ can be rewritten as $$\label{equation varpi trace}
\cos\rho=\frac{2\cos\theta}{{\mbox{\rm tr}\,}b_0}\,.$$ Note that ${\mbox{\rm tr}\,}b_0\geq 2$, since $\det b_0=1$ and $b_0$ is positive definite. Thus there are precisely two possible continuous choices of the angle $\rho$, namely $$\label{choice varpi}
\rho=\pm\arccos\left(\frac{2\cos\theta}{\mathrm{tr}(b_0)}\right)\,.$$ Since $\cos\theta\neq 1$, neither of these two functions can vanish at any point.
In the case being considered, ${\mbox{\rm tr}\,}(b_0)=\sqrt{4+\varphi'(t)^2}$. Hence, by imposing that ${\mbox{\rm tr}\,}(Jb)<0$, we get $\sin\rho>0$ and thus $$\rho=\arccos\left(\frac{2\cos\theta}{\sqrt{4+\varphi'(t)^2}}\right)\,.$$ It thus remains to impose the Codazzi condition to $b=R_\rho b_0$ to determine the function ${\varphi}(t)$. Using Equation , it suffices to impose $$d\rho(\partial_s)J b_0(\partial_t) -d\rho(\partial_t)J b_0(\partial_s)+d^\nabla b_0(\partial_s,\partial_t)=0\,,$$ and since $d\rho(\partial_s)=0$ and $[\partial_s,\partial_t]=0$, we must impose $$\label{eq:pino}
\nabla_{\partial_s}b_0(\partial_t)-\nabla_{\partial_t}b_0(\partial_s)-d\rho(\partial_t)J b_0(\partial_s)=0\,.$$ For this purpose, let us compute $$\begin{aligned}
\nabla_{\partial_s}b_0(\partial_t)&=\nabla_{\partial_s}\left(\frac{1}{\sqrt{4+{\varphi}'(t)^2}}(-{\varphi}'(t)\partial_s+(2+{\varphi}'(t)^2)\partial_t)\right) \\
&=\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\left(-{\varphi}'(t)\nabla_{\partial_s}\partial_s+(2+{\varphi}'(t)^2)\nabla_{\partial_s}\partial_t\right)\end{aligned}$$ Observing that $\nabla_{\partial_s}\partial_s=-\tan(t)\partial_t$ and $\nabla_{\partial_s}\partial_t=\tan(t)\partial_s$, we obtain $$\nabla_{\partial_s}b_0(\partial_t)=\frac{\tan(t)}{\sqrt{4+{\varphi}'(t)^2}}\left((2+{\varphi}'(t)^2){\partial_s}+{\varphi}'(t)\partial_t\right)\,.$$ On the other hand, $$\begin{aligned}
\nabla&_{\partial_t}b_0(\partial_s)=\nabla_{\partial_t}\left(\frac{1}{\sqrt{4+{\varphi}'(t)^2}}(2\partial_s-{\varphi}'(t)\partial_t)\right) \\
&=-\frac{{\varphi}'(t){\varphi}''(t)}{(4+{\varphi}'(t)^2)^{3/2}}(2\partial_s-{\varphi}'(t)\partial_t)
+ \frac{1}{\sqrt{4+{\varphi}'(t)^2}}\left(2\nabla_{\partial_t}\partial_s-{\varphi}''(t)\partial_t-{\varphi}'(t)\nabla_{\partial_t}\partial_t\right) \,,\end{aligned}$$ Using moreover $\nabla_{\partial_t}\partial_t=\tan(t)\partial_t$, $$\begin{aligned}
\nabla_{\partial_t}b_0(\partial_s)=&\left(-\frac{2{\varphi}'(t){\varphi}''(t)}{(4+{\varphi}'(t)^2)^{3/2}}+\frac{2\tan(t)}{\sqrt{4+{\varphi}'(t)^2}} \right)\partial_s \\
&+ \left(\frac{{\varphi}'(t)^2{\varphi}''(t)}{(4+{\varphi}'(t)^2)^{3/2}}-\frac{{\varphi}''(t)+{\varphi}'(t)\tan(t)}{\sqrt{4+{\varphi}'(t)^2}} \right)\partial_t\,.\end{aligned}$$ We also compute: $$\begin{aligned}
\partial_t\rho&= \frac{1}{\sqrt{1-\frac{4\cos^2\theta}{4+\varphi'(t)^2}}} \frac{2\cos\theta}{(4+\varphi'(t)^2)^{3/2}}\varphi'(t)\varphi''(t) \\
&=\frac{2\cos\theta\,\varphi'(t)\varphi''(t)}{(4+\varphi'(t)^2)\sqrt{4\sin^2\theta+\varphi'(t)^2}}\,.\end{aligned}$$
Finally $Jb_0(\partial_s)=(\varphi'(t)\partial_s+2\partial_t)/\sqrt{4+\varphi'(t)^2}$. Therefore the Codazzi condition in Equation gives, for the $\partial_s$ components, $$\frac{\tan(t){\varphi}'(t)^2}{\sqrt{4+{\varphi}'(t)^2}}=-\frac{2{\varphi}'(t){\varphi}''(t)}{(4+{\varphi}'(t)^2)^{3/2}}+\frac{2\cos\theta\,\varphi'(t)^2\varphi''(t)}{(4+\varphi'(t)^2)^{3/2}\sqrt{4\sin^2\theta+\varphi'(t)^2}}\,,$$ and thus $$\label{eq min lag}
-\tan(t)=\frac{2}{4+{\varphi}'(t)^2}\left(\frac{1}{{\varphi}'(t)}-\frac{\cos\theta}{\sqrt{4\sin^2\theta+\varphi'(t)^2}}\right){\varphi}''(t)\,.$$ On the other hand, equating the $\partial_t$-components in gives $$\frac{{\varphi}'(t)^2{\varphi}''(t)}{(4+{\varphi}'(t)^2)^{3/2}}=\frac{{\varphi}''(t)+2{\varphi}'(t)\tan(t)}{\sqrt{4+{\varphi}'(t)^2}}
-\frac{4\cos\theta\,\varphi'(t)\varphi''(t)}{(4+\varphi'(t)^2)^{3/2}\sqrt{4\sin^2\theta+\varphi'(t)^2}}$$ and thus $$2{\varphi}'(t)\tan(t)=\frac{{\varphi}''(t)({\varphi}'(t)^2-4-{\varphi}'(t)^2)}{(4+{\varphi}'(t)^2)}+
\frac{2\cos\theta\,\varphi'(t)\varphi''(t)}{(4+\varphi'(t)^2)\sqrt{4\sin^2\theta+\varphi'(t)^2}}\,,$$ hence leading again to Equation .
Observe that, for the first term in the RHS of Equation , $$\int \frac{2{\varphi}''(t)}{{\varphi}'(t)(4+{\varphi}'(t)^2)}dt=\frac{1}{2}\log({\varphi}'(t))-\frac{1}{4}\log({\varphi}'(t)^2+4)+\text{cost.}$$ Hence if $\cos\theta=0$, by straightforward algebra one obtains $$\frac{{\varphi}'(t)^2}{{\varphi}'(t)^2+4}=e^{-4C}\cos^4(t)\,,$$ for some constant $C$. In conclusion, $${\varphi}'(t)^2=4\left(\frac{1}{1-e^{-4C}\cos^4(t)}-1\right)\,,$$ or equivalently $$\label{derivative phi}
{\varphi}'(t)=2\frac{e^{-2C}\cos^2(t)}{\sqrt{1-e^{-4C}\cos^4(t)}}\,.$$
Observe that, if $C>0$, then ${\varphi}'(t)$ is defined for all $t\in[-\pi/2,\pi/2]$, and therefore we obtain minimal Lagrangian maps from ${\mathbb{H}}^2$ to ${\mathbb{H}}^2$. Up to composing with a hyperbolic translation still of the form $\gamma_a(s,t)=(s+a,t)$, we can assume ${\varphi}$ is a primitive of the RHS of Equation such that ${\varphi}(-\pi/2)=0$, so that the points in $\partial{\mathbb{H}}^2$ with coordinates $w=(s,-\pi/2)$ (namely, those points in the boundary on one side of the geodesic connecting $0$ and $\infty$) are fixed. It is also easy to check that the map obtained by choosing the opposite sign in Equation provides the inverse map. See Figure \[fig:dynamics1\]. However, in this paper we are interested in the solutions of Equation which are *not* defined on the whole interval $[-\pi/2,\pi/2]$.
![The dynamics of the map $\Phi$ when the constant $C$ is positive, in the $w=s+it$ coordinate. This is a conformal model of ${\mathbb{H}}^2$. \[fig:dynamics1\]](dynamics1-eps-converted-to.pdf){width="80.00000%"}
Landslides of the hyperbolic half-plane {#subsec half-plane}
---------------------------------------
Let us now study the ODE in Equation . Let us define $$F:[0,+\infty)\to{{\mathbb R}}\qquad F(r)=-2\cos\theta\int_{0}^r \frac{du}{(4+u^2)\sqrt{4\sin^2\theta+u^2}}\,.$$ Clearly $F(0)=0$ and $$|F(+\infty)|:=\lim_{r\to+\infty} |F(r)|<+\infty\,.$$ By integrating both sides of Equation , one obtains
$$\label{integration 1}
\frac{1}{2}\log({\varphi}'(t))-\frac{1}{4}\log({\varphi}'(t)^2+4)+F(\varphi'(t))=\log(\cos(t))+C\,.$$
Observe that the the function $$r\mapsto \frac{1}{2}\log r-\frac{1}{4}\log(r^2+4)+F(r)\,,$$ which corresponds to the LHS of Equation with ${\varphi}'(t)=r$, is an increasing function of $r$, since by construction its derivative is the expression $$r\mapsto \frac{2}{4+r^2}\left(\frac{1}{r}-\frac{\cos\theta}{\sqrt{4\sin^2\theta+r^2}}\right)$$ which is positive (even if $\cos\theta>0$). (Compare Equation .) An equivalent form of Equation is: $$\label{eq ph F cos}
\frac{{\varphi}'(t)^{1/2}}{({\varphi}'(t)^2+4)^{1/4}}e^{F(\varphi'(t))}=e^C \cos(t)\,.$$ Let us put $C=F(+\infty)$. With this choice, the real function $$G(r)=-\arccos\left(\frac{r^{1/2}}{(r^2+4)^{1/4}}e^{F(r)-C}\right)$$ is strictly increasing and sends $[0,+\infty)$ to $[-\pi/2,0)$. Hence Equation defines ${\varphi}'(t)=G^{-1}(t)$, and in particular $$\label{eq:ciccio}
\lim_{t\to 0^-}{\varphi}'(t)=+\infty \qquad \mathrm{and}\qquad \lim_{t\to -\pi/2}{\varphi}'(t)=0\,.$$
We will need a more precise analysis of the behavior of ${\varphi}(t)$ close to $0$.
\[lemma:rockerduck\] Any solution ${\varphi}$ of Equation , with $C=F(+\infty)$, satisfies $$\label{eq:zorro}
\lim_{t\to 0^-} t{\varphi}'(t)=-\sqrt{2(1-\cos\theta)}\,.$$ Therefore, $\lim_{t\to 0}{\varphi}(t)=+\infty$ and $$\label{eq:zorro2}
\lim_{t\to 0^-} t e^{{\varphi}(t)/2}=0\,.$$
Recalling that ${\varphi}'$ is defined by ${\varphi}'(t)=G^{-1}(t)$, to prove we need to show that $$\lim_{t\to 0^-}tG^{-1}(t)=-\sqrt{2(1-\cos\theta)}\,.$$ Being $G^{-1}$ a diffeomorphism between $[-\pi/2,0)$ and $[0,+\infty)$, observe that $$\lim_{t\to 0^-}tG^{-1}(t)=\lim_{v\to+\infty} vG(v)=\lim_{u\to 0}\frac{G(1/u)}{u}\,.$$ A simple computation shows that $$F(1/u)-C=2\cos\theta\int_{1/u}^{+\infty}\frac{dr}{(4+r^2)\sqrt{4\sin^2\theta+r^2}}=2\cos\theta\int_0^u\frac{sds}{(4s^2+1)\sqrt{4s^2\sin^2\theta+1}}$$ so $F(1/u)-C=\cos\theta u^2+o(u^2)$. Thus $$\cos G(1/u)= \frac{e^{F(1/u)-C}}{(1+4u^2)^{1/4}} =(1-u^2+o(u^2))(1+\cos\theta u^2+o(u^2))=1-(1-\cos\theta)u^2+o(u^2)$$ so $$G^2(1/u)/2=(1-\cos\theta)u^2+o(u^2)$$ and we conclude that $G(1/u)=-\sqrt{2(1-\cos\theta)}u +o(u)$, which implies .
For , notice that $\sqrt{2(1-\cos\theta)}<1$, and choose a number $\alpha\in(\sqrt{2(1-\cos\theta)},1)$. Now fixing $t_0<0$ so that ${\varphi}'(t)<\alpha/|t|$ for $t\in[t_0, 0)$ (this exists by ). So for $t>t_0$ we have $${\varphi}(t)-{\varphi}(t_0)=\int_{t_0}^t{\varphi}'(r)dr<\alpha\ln(t_0/t)$$ so $$e^{{\varphi}(t)}\leq e^{{\varphi}(t_0)}\left(\frac{t_0}{t}\right)^\alpha=c|t|^{-\alpha}~,$$ and this proves .
The associated map $$\Phi_\theta:{\mathbb{H}}^2_+\to{\mathbb{H}}^2_+\,,$$ where ${\mathbb{H}}^2_+=\{z\in{\mathbb{H}}^2\,|\,\Re(z)>0\}=\{t\leq 0\}$ is a half-plane in ${\mathbb{H}}^2$, of the form $$\Phi_\theta(s,t)=(s-{\varphi}(t),t)\,,$$ is a $\theta$-landslide of ${\mathbb{H}}^2_+$. Here ${\varphi}$ is a primitive of ${\varphi}'$ chosen in such a way that ${\varphi}(-\pi/2)=0$, so that $\Phi(s,-\pi/2)=(s,-\pi/2)$ for every $s\in{{\mathbb R}}$. Since from Lemma \[lemma:rockerduck\] we know ${\varphi}(t)$ diverges as $t\to 0$, the map does not extend to the geodesic boundary of ${\mathbb{H}}^2_+$, which is the geodesic invariant for the 1-parameter hyperbolic group.
We additionally observe that $\Phi_\theta$ maps the curve $q_{s_0}:(0,\pi/2)\to {\mathbb{H}}^2_+$, parameterized by $$q_{s_0}(t)=\left(s_0+\frac{1}{2}{\varphi}(t),t\right)\,,$$ to the curve $p_{s_0}$ with parameterization $$p_{s_0}(t)=\left(s_0-\frac{1}{2}{\varphi}(t),t\right)\,,$$ which is the image of $q_{s_0}$ by means of the reflection $(s,t)\mapsto(2s_0-s,t)$. See Figure \[fig:dynamics2\].
![A sketch of the dynamics of the map $\Phi_\theta$, in the $w=s+it$ coordinate. \[fig:dynamics2\]](dynamics2-eps-converted-to.pdf){width="80.00000%"}
Let us remark that, since $\Phi_\theta$ commutes with the 1-parameter hyperbolic group $\gamma_a$, then the images of the constant curvature embeddings $\sigma_{\Phi_\theta,R_\rho b_0}$ are invariant for the group of isometries of the form $(\gamma_a,\gamma_a)$, which preserve the plane $\mathcal R_\pi$. We want to prove that the image of $\sigma_{\Phi_\theta,R_\rho b_0}$ is the spacelike part of a past-convex $K$-surface $S_K$ (with as usual $K=-1/\cos^2(\theta/2)$) whose boundary in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is the union of the boundary at infinity of an invariant half-plane in $\mathcal R_\pi$ and a sawtooth, i.e. two null segments in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Moreover, the lightlike part of $S_\theta$ is precisely the lightlike triangle bounded by the two null segments. This last condition will be the key property we will need to use $S_K$ as a *barrier* in the next sections.
Defining $\sigma(t)=\sigma_{\Phi_\theta,R_\rho b_0}(q_{s_0}(t))$, by applying the definition, the isometry $\sigma(t)$ maps the point $p_{s_0}(t)$ to $q_{s_0}(t)$. From Equation , the distance of a point from the geodesic $\{t=0\}$ only depends on the $t$-coordinate, and thus $p_{s_0}$ and $q_{s_0}$ have the same distance from the geodesic $\{t=0\}$. Hence $\sigma(t)$ can be an elliptic isometry with fixed point on the geodesic $\{s=s_0\}$, a parabolic isometry fixing an endpoint of $\{s=s_0\}$, or a hyperbolic isometry whose axis is orthogonal to the geodesic $\{s=s_0\}$.
The goal of this part is to prove that for $t\to 0$ the family of isometries $\sigma_{\Phi_\theta,R_\rho b_0}(q_{s_0}(t))$ actually converges to $\mathcal{I}_{(s_0,0)}$, that is, to the elliptic isometry of order two fixing the point of intersection of $\{s=s_0\}$ and $\{t=0\}$. In this way we will get that the surface associated to $\Phi_\theta$ contains the spacelike line in the totally geodesic plane $\mathcal R_\pi$ invariant by $(\gamma_a, \gamma_a)$.
\[lemma:dinamitebla\] If in the $w=s+it$ coordinate we denote $p(t)=p_{s_0}(t)=(s_0-({1}/{2}){\varphi}(t),t)$, $q(t)=q_{s_0}(t)=(s_0+({1}/{2}){\varphi}(t),t)$ and $\sigma(t)=\sigma_{\Phi_\theta,R_\rho b_0}(q_{s_0}(t))$, then $$\label{eq:archimede}
\begin{split}
d(\sigma(t))\left(\left.\frac{\partial\,}{\partial s}\right |_{p(t)}\right)&=\cos\xi(t) \left(\left.\frac{\partial\,}{\partial s}\right |_{q(t)}\right)+\sin\xi(t)\left(\left.\frac{\partial\,}{\partial t}\right |_{q(t)}\right)\\
d(\sigma(t))\left(\left.\frac{\partial\,}{\partial t}\right |_{p(t)}\right)&=-\sin\xi(t) \left(\left.\frac{\partial\,}{\partial s}\right |_{q(t)}\right)+\cos\xi(t)\left(\left.\frac{\partial\,}{\partial t}\right |_{q(t)}\right)\,.
\end{split}$$ where $\xi$ is a function of $t\in(-\pi/2,0)$ with $$\begin{aligned}
&\lim_{t\to 0}{\varphi}'(t)\sin\xi(t)=2(1-\cos\theta)\,, \label{eq:orazio} \\
&\lim_{t\to 0} \cos\xi(t)=-1\,.\end{aligned}$$ In particular, $\xi(t)\to \pi \mod 2\pi$ as $t\to 0$
First observe that $\{\cos t\frac{\partial\,}{\partial s}|_{p(t)}, \cos t\frac{\partial\,}{\partial t}|_{p(t)}\}$ and $\{\cos t\frac{\partial\,}{\partial s}|_{q(t)}, \cos t\frac{\partial\,}{\partial t}|_{q(t)}\}$ are orthonormal basis respectively at $p(t)$ and $q(t)$, for the form of the metric. Thus $d(\sigma(t))$ at $p(t)$ must have the form of Equation for some angle $\xi(t)$.
Let us now compute $\xi(t)$. Observe that, with respect to the $(\partial_s,\partial_t)$-frame, $$(d\Phi_0)_{(s,t)}=\begin{pmatrix} 1 & -{\varphi}'(t) \\ 0 & 1 \end{pmatrix}\,.$$ Recall that, $$b_{(s,t)}=R_\rho\circ (b_0)_{(s,t)}=\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\begin{pmatrix} \cos\rho(t) & -\sin\rho(t) \\ \sin\rho(t) & \cos\rho(t) \end{pmatrix}\begin{pmatrix} 2 & -{\varphi}'(t) \\ -{\varphi}'(t) & 2+{\varphi}'(t)^2 \end{pmatrix}\,,$$ and thus $$d\sigma=-b\circ (d\Phi_0)^{-1}=-\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\begin{pmatrix} \cos\rho(t) & -\sin\rho(t) \\ \sin\rho(t) & \cos\rho(t) \end{pmatrix}\begin{pmatrix} 2 & {\varphi}'(t) \\ -{\varphi}'(t) & 2 \end{pmatrix} \,.
$$ It follows that $$\sin\xi(t)=-\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\left(2\sin\rho(t)-{\varphi}'(t)\cos\rho(t)\right)$$
By Equation , ${\varphi}'(t)\to+\infty$ as $t\to 0$, so $\mathrm{tr}(b_0)=\sqrt{4+{\varphi}'(t)^2}$ tends to infinity as well. By Equation , $${\varphi}'(t)\cos\rho(t) ={\varphi}'(t)\frac{2\cos\theta}{\mathrm{tr}(b_0)}\to 2\cos\theta\,.$$ On the other hand, as $\rho=\arccos(2\cos\theta/\mathrm{tr}(b_0))$, $\sin\rho(t)\to 1$ as $t\to 0$. Therefore, as $t\to 0$: $${\varphi}'(t)\sin\xi(t)=-\frac{{\varphi}'(t)}{\sqrt{4+{\varphi}'(t)^2}}\left(2\sin\rho(t)-{\varphi}'(t)\cos\rho(t)\right)\to -2(1-\cos\theta)~.$$ Finally an explicit computation shows that $$\cos\xi(t)=-\frac{1}{\sqrt{4+{\varphi}'(t)^2}}\left(2\cos\rho(t)+{\varphi}'(t)\sin\rho(t)\right)\,.$$ Hence $\lim \cos\xi(t)=-\lim\sin\rho(t)=-1$.
\[prop:nonnapapera\] Denoting $p(t)=(s_0-({1}/{2}){\varphi}(t),t)$, and $\sigma(t)=\sigma_{\Phi_\theta,R_\rho b_0}(q(t))$ (in the $w=s+it$ coordinate, as in Lemma \[lemma:dinamitebla\]), then as $t\to 0$ the family of isometries $\{\sigma(t)\}$ converges to $\mathcal{I}_{m}$, the elliptic rotation of angle $\pi$ around the point $m=(s_0,0)$.
The isometry $\sigma(t)$ is defined by the properties that it sends $p(t)$ to $q(t)$ and its differential sends the vector $\frac{\partial\,}{\partial s}|_{p(t)}$ to a vector forming the positive angle $\xi(t)$ with $\frac{\partial\,}{\partial s}|_{q(t)}$, by Equation .
Recall that in the $z=x+iy$ coordinate of the upper half-plane model, $z=i\exp w=\exp(s+i(\pi/2+t))$, so $$p(t)=\exp(s_0-{\varphi}(t)/2+i(\pi/2+t))\qquad q(t)=\exp(s_0+{\varphi}(t)/2+i(\pi/2+t))\,.$$
The transformation $\delta(t): z\mapsto e^{{\varphi}(t)}z$ sends $p(t)$ to $q(t)$ and $\frac{\partial\,}{\partial s}|_{p(t)}$ to $\frac{\partial\,}{\partial s}|_{q(t)}$. So $\sigma(t)$ is obtained as the rotation of angle $\xi(t)$ around $p(t)$ post-composed with $\delta(t)$.
An explicit computation shows that the rotation around a point $p=|p|e^{i\eta}$ of angle $\xi$ is represented by the ${\mathrm{SL}}(2,{{\mathbb R}})$ matrix $$\begin{pmatrix}
\frac{\sin(\eta+\xi/2)}{\sin\eta} & -\frac{\sin(\xi/2)|p|}{\sin\eta}\\
\frac{\sin(\xi/2)}{|p|\sin\eta} & \frac{\sin(\eta-\xi/2)}{\sin\eta}
\end{pmatrix}~.$$ (Notice that changing $\xi$ by $\xi+2\pi$ the matrix changes by sign.)
Applying this formula to $p(t)$, $\eta=\eta(t)=\pi/2+t$ and $|p|=e^{s_0-{\varphi}(t)/2}$ and multiplying the rotation matrix by $\delta(t)=\mathrm{diag}(e^{{\varphi}(t)/2}, e^{-{\varphi}(t)/2})$ we get that
$$\begin{aligned}
( \sigma(t))_{21}&=-e^{s_0}\frac{\sin(\xi(t)/2)}{\cos t}\\
(\sigma(t))_{12}&= e^{-s_0}\frac{\sin(\xi(t)/2)}{\cos t}\\
(\sigma(t))_{22}&=e^{-{\varphi}(t)/2}\frac{\sin(\pi/2+t-\xi(t)/2)}{\cos t} \,. \\
\end{aligned}$$
As $\xi(t)\to\pi\ (\mathrm{mod} 2\pi)$, we obtain $\lim(\sigma(t))_{21}=\pm e^{-s_0}$ and $\lim(\sigma(t))_{12}=\mp e^{s_0}$, whereas $(\sigma(t))_{22}\to 0$.
On the other hand, the first entry of $\sigma(t)$ is $$(\sigma(t))_{11}=e^{{\varphi}(t)/2}\frac{\sin(\pi/2+t+\xi(t)/2)}{\cos t}= e^{{\varphi}(t)/2}\frac{\cos(t+\xi(t)/2)}{\cos t}\,.$$ Now, $$\begin{aligned}
\cos(t+\xi(t)/2)&=\cos t\cos(\xi(t)/2)-\sin t\sin(\xi(t)/2)\\
&=\cos t\sqrt{\frac{\cos\xi(t)+1}{2}}-\sin t\sin(\xi(t)/2)\\
&= \frac{\cos t\sin\xi(t)}{\sqrt{2(1-\cos\xi(t))}} -\sin t\sin(\xi(t)/2)~.\end{aligned}$$
By Equations and it follows that the ratio $$\frac{\cos(t+\xi(t)/2)}{t}$$ is bounded. Thus using Equation we deduce that $\lim_{t\to 0}\sigma_{11}(t)=0$.
In conclusion $$\sigma(t)\to \begin{bmatrix} 0 & e^{-s_0}\\ -e^{s_0} &0\end{bmatrix}=\mathcal{I}_{m}~,$$ which proves the claim.
\[lemma barrier\] The constant curvature embedding $\sigma_{\Phi_\theta,R_\rho b_0}:{\mathbb{H}}^2_+\to{\mathrm{Isom}}({\mathbb{H}}^2)$ defined by
- ($\sigma_{\Phi_\theta,R_\rho b_0}(z))(\Phi_\theta(z))=z\,$;
- $(d\sigma_{\Phi_\theta,R_\rho b_0}(z))_z\circ (d\Phi_\theta)_{z}=-R_\rho b_0\,$;
with respect to the map $\Phi_\theta$ and the bundle morphism $b=R_\rho b_0$ constructed above, has image a constant curvature surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ whose boundary coincides with the boundary of the half-plane ${\mathbb{H}}^2_+$.
From Proposition \[prop:nonnapapera\], for every $s_0$ we found a family of points $p_{s_0}(t)$ whose images for the embedding $\sigma_{\Phi_\theta,R_\rho b_0}$ converges to $(s_0,0)$. That is, all points on the geodesic boundary of ${\mathbb{H}}^2_+$ (identified to a half-plane of the totally geodesic plane $\mathcal R_\pi$) are in the frontier of $\sigma_{\Phi_\theta,R_\rho b_0}({\mathbb{H}}^2_+)$.
To conclude the proof, it remains to show that, when $t\to-\pi/2$, $\sigma_{\Phi_\theta,R_\rho b_0}$ tends to the boundary at infinity of the half-plane ${\mathbb{H}}^2_+$ of $\mathcal R_\pi$. Denoting again by $\sigma(t)$ the image of the point $p_{s_0}(t)$ by the embedding $\sigma_{\Phi_\theta,R_\rho b_0}$, we claim that both sequences $\sigma(t)(m)$ and $\sigma(t)^{-1}(m)$, where $m$ is the point $ie^{s_0}$, converge to $e^{s_0}$, which is in the boundary of the upper half-plane model. Since $s_0$ is chosen arbitrarily, and recalling the definition of Equation , this will conclude the proof.
Here the computation is similar, though much simpler. Indeed, recalling that ${\varphi}'(t)\to 0$ as $t\to -\pi/2$, we have that $\mathrm{tr}(b_0)=\sqrt{4+{\varphi}'(t)^2}\to 2$ as $t\to-\pi/2$. Therefore, from the choice of Equation , we get $\rho\to\theta$. Moreover, recall that we have chosen ${\varphi}$ as the primitive of ${\varphi}'$ such that ${\varphi}(-\pi/2)=0$. Using the above, a straigthforward computation shows that $\sigma(t)(ie^{s_0})\to e^{s_0}$. Analogously, $\sigma(t)^{-1}(ie^{s_0})\to e^{s_0}$.
Existence of constant-curvature surfaces {#sec existence}
========================================
The idea of the proof of the existence of $K$-surfaces with prescribed boundary is to obtain a $K$-surface by approximation from surfaces which are invariant by a cocompact action. In that case, the existence of constant curvature surfaces is guaranteed by the following theorem:
\[bbz\] Let $\phi:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$ be an orientation-preserving homeomorphism which is equivariant for a pair of Fuchsian surface group representation $$\rho_l,\rho_r:\pi_1(\Sigma_g)\to{\mathrm{Isom}}({\mathbb{H}}^2)\,,$$ where $\Sigma_g$ is a closed surface. Let $\Gamma$ the curve in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which is the graph of $\phi$. Then there exists a foliation of $\mathcal{D}_+(\Gamma)$ and $\mathcal{D}_-(\Gamma)$ by surfaces of constant curvature $K\in(-\infty,-1)$, which are invariant for the representations $$(\rho_l,\rho_r):\pi_1(\Sigma_g)\to{\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)\cong {\mathrm{Isom}}({\mathbb{H}}^2)\times{\mathrm{Isom}}({\mathbb{H}}^2)\,.$$ The foliations are such that, if $|K_1|<|K_2|$, then the $K_1$-surface is in the convex side of the $K_2$-surface.
For this purpose, we will use three main technical tools. The first is the following lemma, which we prove in Subsection \[subsec approximation\] below.
\[lemma approx acausal\] Given any weakly acausal curve $\Gamma$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, there exists a sequence of curves $\Gamma_n$ invariant for pairs of Fuchsian representations of the fundamental group of closed surfaces $$((\rho_l)_n,(\rho_r)_n):\pi_1(\Sigma_{g_n})\to{\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)\cong{\mathrm{Isom}}({\mathbb{H}}^2)\times{\mathrm{Isom}}({\mathbb{H}}^2)$$ such that $\Gamma_n$ converges to $\Gamma$ in the Hausdorff convergence.
The second is the following theorem of Schlenker on the smooth convergence of surfaces, specialized to the case of the ambient manifold $(M,g)={\mathbb{A}\mathrm{d}\mathbb{S}}^3$:
\[schl\_degeneration\] Let $\sigma_n:{\mathbb{D}}\to M$ be a sequence of uniformly elliptic spacelike immersions (i.e. with uniformly positive determinant of the shape operator) in a Lorentzian manifold $(M,g)$. Assume $\sigma_n^*g$ converges $C^\infty$ to a metric $g_\infty$. If $x_n\in{\mathbb{D}}$ is a sequence converging to a point $x_\infty$, such that the 1-jet $j^1\sigma_n(x_n)$ is converging, but $\sigma_n$ does not converge $C^\infty$ in a neighborhood of $x_\infty$, then there exists a maximal geodesic $\gamma$ of $({\mathbb{D}},g_\infty)$ containing the point $x_\infty$ such that $\sigma_n$, restricted to $\gamma$, converges to an isometry onto a geodesic $\gamma'$ of $M$.
Moreover, in the conditions of Theorem \[schl\_degeneration\], the degeneration of the immersions $\sigma_n$ is well-understood in [@schlenkersurfconv]: if $\sigma_n$ does not converge $C^\infty$ in a neighborhood of $x_\infty$, then the surfaces $\sigma_n({\mathbb{D}})$ converge to a surface which contains all the (future-directed or past-directed) light rays starting from points on the geodesic $\gamma'$ of $M$.
The third tool is the use of the barrier we constructed in Section \[sec barrier\], in particular Corollary \[lemma barrier\].
Approximation of measured geodesic laminations {#subsec approximation}
----------------------------------------------
We recall here the definition of measured geodesic lamination on ${\mathbb{H}}^2$. Let $\mathcal{G}$ be the set of (unoriented) geodesics of ${\mathbb{H}}^2$. The space $\mathcal{G}$ is identified to $((\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2)\setminus \Delta)/\sim$ where $\Delta$ is the diagonal and the equivalence relation is defined by $(p,q)\sim(q,p)$. Note that $\mathcal{G}$ has the topology of an open Möbius strip.
\[defi mgl\] A geodesic lamination on ${\mathbb{H}}^2$ is a closed subset of $\mathcal{G}$ such that its elements are pairwise disjoint geodesics of ${\mathbb{H}}^2$. A measured geodesic lamination is a locally finite Borel measure on $\mathcal{G}$ such that its support is a geodesic lamination.
For the approximation procedure, we will use the following Lemma \[fuchsian weak approximation\] on the approximation of measured geodesic laminations, which is proved in [@Bonsante:2015vi Lemma 3.4]. We first need to recall some definitions.
A sequence $\{\mu_n\}_n$ of measured geodesics laminations converges in the weak\* topology to a measured geodesic lamination, $\mu_n{\rightharpoonup}\mu$, if $$\lim_{n{\rightarrow}\infty}\int_{\mathcal{G}}f d\mu_n=\int_{\mathcal{G}}f d\mu$$ for every $f\in C_0^0(\mathcal{G})$.
\[fuchsian weak approximation\] Given a measured geodesic lamination $\mu$ on ${\mathbb{H}}^2$, there exists a sequence of measured geodesic laminations $\mu_n$ such that $\mu_n$ is invariant under a torsion-free cocompact Fuchsian group $G_n<{\mathrm{Isom}}({\mathbb{H}}^2)$ and $\mu_n{\rightharpoonup}\mu$.
The second technical lemma ensures that the weak\*-convergence of the bending laminations implies the convergence of the curves in the boundary of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, which is basically the same as the uniform convergence of the corresponding left earthquake maps.
\[convergence earthquakes\] Let $\mu_n,\mu$ be measured geodesic laminations on ${\mathbb{H}}^2$ such that $\mu_n{\rightharpoonup}\mu$ in the weak\* topology. Assume $\mu_n$ and $\mu$ induce earthquakes of ${\mathbb{H}}^2$ and $p_0$ is a point of ${\mathbb{H}}^2$ which is not on any weighted leaf of $\mu$. Then the homeomorphism $\phi_n$ of $\partial{\mathbb{H}}^2$ obtained by earthquake along $\mu_n$, normalized in such a way that the stratum containing $p_0$ is fixed, converges uniformly to the homeomorphism $\phi$ obtained by earthquake along $\mu$, normalized in the analogous way.
Let $C_n=gr(\phi_n)$ be the 1-dimensional submanifold of $\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$ which is the graph of $\phi_n:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$. Up to a subsequence, $C_n$ converges to a subset $C$ of $\partial{\mathbb{H}}^2\times \partial{\mathbb{H}}^2$ in the Hausdorff convergence. As $C_n$ are acausal curves, it turns out that $C$ is a weakly acausal curve. Hence, in order to prove that $C=gr(\phi)$, it suffices to show that $gr(\phi)\subseteq C$. In fact we will prove that for every point $x\in \partial{\mathbb{H}}^2$ there exists a sequence $x_k$ converging to $x$ such that $(x_k,\phi(x_k))\in C$.
Let $p\in{\mathbb{H}}^2$ which is not on any weighted leaf of $\mu$. Denote by $F_n(p)$ the stratum of $\mu_n$ which contains $p$, and analogously $F(p)$ is the stratum of $\mu$ contaning $p$. We claim that for any sequence $y_n\in \partial{\mathbb{H}}^2$ in the boundary of $F_n(p)$ and any $y_0$ in the boundary of $F(p)$, if $y_n\to y_0$, then $\phi_n(y_n)\to \phi(y_0)$. This follows by [@epsteinmarden Theorem 3.11.5], where in the language of cocycles, $E_{\mu_n}(p_0,p)\to E_{\mu}(p_0,p)$ in ${\mathrm{Isom}}({\mathbb{H}}^2)$, and therefore $\phi_n(y_n)=E_{\mu_n}(p_0,p)(y_n)$ converges to $\phi(y_0)=E_\mu (p_0,p)(y_0)$. In this case, one then has $(y_0,\phi(y_0))\in C$.
Next, we claim that for every $x\in \partial{\mathbb{H}}^2$ there exists a sequence of points $x_k$ such that $x_k\to x$ and $(x_k,\phi(x_k))\in C$. This will conclude the proof, by the above observations. Let $p_k$ be a sequence of points in ${\mathbb{H}}^2$ converging to $x$, such that every $p_k$ is not on any weighted leaf of $\mu$. For every fixed $k$, consider the stratum $F_n(p_k)$ and let $y_n(p_k)$ be the point of $\partial{\mathbb{H}}^2$, in the boundary of $F_n(p_k)$, closest to $x$ (in the round metric of $\partial{\mathbb{H}}^2\sim S^1$, in the Poincaré disc model, for instance). If $x$ itself is in the boundary of $F_n(p_k)$, then obviously $y_n(p_k)=x$.
Observe that the distance of $y_n(p_k)$ from $x$ is bounded in terms of the distance of $p_k$ from $x$ (in the Euclidean metric on the disc model, for instance). Let $x_k=\lim_{n} y_n(p_k)$, up to a subsequence. By our first claim, $(x_k,\phi(x_k))\in C$. Moreover, $d(y_n(p_k),x)\to 0$ as $k\to+\infty$ uniformly in $n$ (as it is estimated by $d(p_k,x)\to 0$). Therefore also $d(x_k,x)$ tends to zero as $k\to +\infty$. This implies that every $x\in \partial{\mathbb{H}}^2$ is approximated by points $x_k$ such that $(x_k,\phi(x_k))\in C$.
![Possible positions of the points $y_n(p_k)$, as the position of $p_k$ changes. The Euclidean distance $d(y_n(p_k),x)$ is uniformly bounded in terms of $d(p_k,x)$, and it goes to zero as $d(p_k,x)\to 0$. \[fig:positions\]](geodesics1-eps-converted-to.pdf){height="4.5cm"}
![Possible positions of the points $y_n(p_k)$, as the position of $p_k$ changes. The Euclidean distance $d(y_n(p_k),x)$ is uniformly bounded in terms of $d(p_k,x)$, and it goes to zero as $d(p_k,x)\to 0$. \[fig:positions\]](geodesics2-eps-converted-to.pdf){height="4.5cm"}
One might consider the definition of measured geodesic lamination on a straight convex set of ${\mathbb{H}}^2$, in the sense of [@bebo], and prove more general statements of Lemma \[fuchsian weak approximation\] and Lemma \[convergence earthquakes\]. However, to avoid technicalities, we will circumvent this issue by approximating in two steps.
Let $\Gamma$ be a weakly acausal curve in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$. It is possible to find a sequence $\phi_n$ of orientation-preserving homeomorphisms of $\partial{\mathbb{H}}^2$ such that the graphs $\Gamma_n=gr(\phi_n)$ converge to $\Gamma$ in the Hausdorff convergence. Moreover, by Lemma \[fuchsian weak approximation\] for every $n$ there exists a sequence $\Gamma_{n,k}$ of graphs invariant for some pairs $((\rho_l)_{n,k},(\rho_r)_{n,k})$ of torsion-free cocompact Fuchsian representations. By Lemma \[convergence earthquakes\], the $\Gamma_{n,k}$ converge to $\Gamma_n$ in the Hausdorff convergence. By a standard diagonal argument, one finds a sequence $\Gamma_{n,{k(n)}}$ which converges to $\Gamma$.
Proof of the existence part
---------------------------
The purpose of this section is to prove the existence part of $K$-surfaces of Theorem \[thm foliation part ads\]. A 1-step curve is the boundary of a totally geodesic lightlike plane, while a 2-step curve is the union of four null segments in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Those are the two cases in which the convex hull of the curve $\Gamma$ coincides with the domain of dependence $\mathcal{D}(\Gamma)$. See Figure \[fig:step\].
![A 1-step curve (left) bounds a lightlike plane. On the right, a 2-step curve, which is the boundary of the surface obtained as the union of two lightlike half-planes which intersect along a spacelike geodesic. \[fig:step\]](1step-eps-converted-to.pdf){height="6cm"}
![A 1-step curve (left) bounds a lightlike plane. On the right, a 2-step curve, which is the boundary of the surface obtained as the union of two lightlike half-planes which intersect along a spacelike geodesic. \[fig:step\]](2step-eps-converted-to.pdf){height="6cm"}
\[prop existence\] Given any weakly acausal curve $\Gamma$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which is not a 1-step or a 2-step curve, for every $K\in(-\infty,-1)$ there exist two convex surfaces $S^+_{K},S^-_{K}$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S^\pm_{K}=\Gamma$, such that:
- The lightlike part of $S_K^\pm$ is $\overline{\mathcal D_\pm(\Gamma)}\cap \mathcal C(\Gamma)$, that is, the union of all lightlike triangles bounded by past-directed (resp. future-directed) sawteeth of $\Gamma$;
- The spacelike part of $S_K^\pm$ is a smooth $K$-surface.
The first property of the surface $S_K^\pm$ in Theorem \[prop existence\] means that its lightlike part is the smallest possible, under the condition that the boundary is the curve $\Gamma$. In particular, if it is possible to construct a past-convex spacelike surface with boundary $\Gamma$, then the lightlike part of $S_K^+$ is empty and therefore $S_K^+$ is a smooth spacelike $K$-surface with boundary $\Gamma$. The same holds for future-convex surfaces. In general, the boundary of the spacelike part of $S_K^\pm$ coincides with the boundary of the spacelike part of $\mathcal C_\pm(\Gamma)$.
For definiteness, we will provide the argument for the past-convex surface $S_K^+$ contained in $\mathcal{D}_+(\Gamma)$, the other argument being completely analogous. Let $\Gamma_n$ be a sequence of curves in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ invariant for a pair of torsion-free cocompact Fuchsian representations $((\rho_l)_n,(\rho_r)_n)$, converging to $\Gamma$, as in Lemma \[lemma approx acausal\]. Fix $K<-1$. By Theorem \[bbz\], there exists a past-convex $K$-surface $(S_K)_n$ in $\mathcal{D}_+(\Gamma_n)$, with $\partial (S_K)_n=\Gamma_n$. *Step 1: There exists a $C^0$-limit of the surfaces $(S_K)_n$.* Introducing the model of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ given by (see [@bon_schl]): $$\label{eq:battista}
\left({\mathbb{D}}^2 \times S^1,\frac{4|dz|^2-(1+|z|^2)dt^2}{(1-|z|^2)^2}\right)\,,$$ we see that, since $(S_K)_n$ is spacelike, $\overline{(S_K)_n}$ is the graph of a $4$-Lipschitz function from $\overline{{\mathbb{D}}^2}$ to $S^1$, with respect to the Euclidean metric of the disc. By an application of the Ascoli-Arzelà Theorem, there exists a subsequence (which we still denote by $(S_K)_{n}$) whose closure converges uniformly to the closure of a locally convex nowhere timelike surface $(S_K)_\infty$. As a consequence $\partial(S_K)_\infty=\Gamma$. It remains to show the two properties in the statement. *Step 2: The lightlike part of the limit $(S_K)_\infty$ is the union of all maximal lightlike triangles bounded by past-directed sawteeth.* We already know that, for any past-directed sawtooth contained in $\Gamma$, the surface $(S_K)_\infty$ contains entirely the totally geodesic lightlike triangle corresponding to the sawtooth. We will need to show the other inclusion. Namely, we will show that any point in the complement of the past-directed lightlike triangles is not in the lightlike part of $(S_K)_\infty$. (Observe that the existence of a point $x$ in the lightlike part of $(S_K)_\infty$ lying in the complement of the lightlike triangles implies that $\Gamma$ is not the boundary of a lightlike plane or the boundary of two lightlike half-planes intersecting in a spacelike geodesic. See Figure \[fig:step\].)
For this purpose, suppose by contradiction that $x\in (S_K)_\infty$ is a point in the complement of such lightlike triangles, but still in the lightlike part of $(S_K)_\infty$. This implies that $(S_K)_\infty$ has a lightlike support plane $P$ which contains $x$. Let us show that the dual point $y$ of $P$ is necessarily on the curve $\Gamma$. Indeed, let $\ell$ be the past-directed light-ray joining $x$ to $y$ (which is contained in $P$). On the one hand, by convexity of $(S_K)_\infty$, $\ell$ does not intersect the open convex region bounded by $(S_K)_\infty$ (in an affine chart containing $\Gamma$). On the other hand, as $\ell$ is the limit of timelike rays though $x$, which are contained in this open convex region, it turns out that $\ell$ is in $(S_K)_\infty$. Hence the endpoint $y$ of $\ell$ is in the curve at infinity $\Gamma$.
Now, we can find a curve $\Gamma'$, as in Example \[ex crucial\] (compare also Figure \[fig:comparison\]), with the following properties:
- $\Gamma'$ is composed of a past-directed sawtooth contained inside the plane $P$ and of the boundary of a totally geodesic spacelike half-plane;
- $\Gamma'$ does not intersect $\Gamma$ transversely;
- In an affine chart which entirely contains $\mathcal D(\Gamma)$, the totally geodesic spacelike plane which contains a portion of $\Gamma'$ disconnects $x$ from $\Gamma$.
Indeed, if $P\cap \Gamma$ is composed of a single point $y$, one can insert a small enough past-directed sawtooth with vertex in $y$, such that the corresponding lightlike triangle is contained inside $P$ but does not contain $x$. Moreover $\Gamma'$ can be arranged to intersect $\Gamma$ only at $y$. The situation is similar if $P\cap\Gamma$ contains a light-like segment, but not a past-directed sawtooth. Similarly, if $P\cap \Gamma$ contains a past-directed sawtooth, it suffices to choose a slightly larger sawtooth, in such a way that $\Gamma\cap\Gamma'$ coincides with the sawtooth $P\cap\Gamma$. See Figure \[fig:comparison\]. Of course this is possible provided $\Gamma$ is not the boundary of a totally geodesic lightlike plane.
![The construction of the curve $\Gamma'$, when $\Gamma$ is the graph of an orientation-preserving homeomorphism (left) and when $\Gamma$ contains a past-directed sawtooth (right). \[fig:comparison\]](comparison1-eps-converted-to.pdf){height="6.5cm"}
![The construction of the curve $\Gamma'$, when $\Gamma$ is the graph of an orientation-preserving homeomorphism (left) and when $\Gamma$ contains a past-directed sawtooth (right). \[fig:comparison\]](comparison2-eps-converted-to.pdf){height="6.5cm"}
Let us consider the surface $S'_K$, which was constructed in Corollary \[lemma barrier\], with the following properties:
- The lightlike part of $S'_K$ is the maximal lightlike triangle with boundary in $\Gamma'$,
- The spacelike part of $S'_K$ is a smooth $K$-surface.
To conclude Step 2, it suffices to show that $S'_K$ and $(S_K)_\infty$ do not intersect transversely. In fact, from Corollary \[lemma barrier\] the surface $S'_K$ intersects $P$ exactly in the maximal lightlike triangle with boundary in $\Gamma'$. But by construction $x$ is not in such maximal lightlike triangle. Hence if we show that $S'_K$ and $(S_K)_\infty$ do not intersect transversely, it follows that $(S_K)_\infty$ cannot intersect $P$ at $x$, thus giving a contradiction.
To show the claim, observe that since $\Gamma_n$ converges to $\Gamma$ as $n\to\infty$, one can choose a sequence of isometries $\chi_n\in{\mathrm{Isom}}({\mathbb{A}\mathrm{d}\mathbb{S}}^3)$ such that:
- $\chi_n(\Gamma_n)$ is disjoint from $\Gamma'$;
- $\chi_n\to\mathrm{id}$ as $n\to\infty$.
We claim that the surfaces $S'_K$ and $\chi_n (S_K)_n$ do not intersect transversely. Since their boundaries $\Gamma'$ and $\chi_n(\Gamma_n)$ are disjoint, and $S'_K$ contains a lightlike triangle, the intersection must be contained in a compact region of the spacelike part of $S'_K$. Observe that $S'_K$ lies outside of the convex hull of $\chi_n(\Gamma_n)$. By Theorem \[bbz\], there is a function $$\kappa:\mathcal D_+(\chi_n(\Gamma_n))\to(-\infty,-1)$$ such that the level sets $\{\kappa=K\}$ are $K$-surfaces. Observe that $\kappa\to -\infty$ as $x\to\partial\mathcal D(\Gamma)$. Let $K_0$ be the maximum of $\kappa$ on $S'_K\cap \mathcal D_+(\chi_n(\Gamma_n)$. Then the $K_0$-surface $\chi_n (S_{K_0})_n$ is tangent to the spacelike part of $S'_K$ at an interior point and $\chi_n (S_{K_0})_n$ is contained in the convex side of $S_K'$. By an application of the maximum principle at the intersection point, $K_0<K$. But again by Theorem \[bbz\], $\chi_n (S_{K})_n$ is in the convex side of $\chi_n (S_{K_0})_n$ and so $S_K'$ cannot intersect $\chi_n (S_{K})_n$.
Since $\chi_n (S_K)_n$ converges uniformly on compact sets to $(S_K)_\infty$, this shows that $S'_K$ and $(S_K)_\infty$ do not intersect transversely. This concludes Step 2. In fact, we have proved that for any point $x\in (S_K)_\infty$ which is in the complement of the lightlike triangles corresponding to sawteeth of $\Gamma$, $x$ does not lie in the boundary of the domain of dependence $\mathcal D(\Gamma)$. Therefore, the lightlike part of $(S_K)_\infty$ coincides precisely with the union of all maximal past-directed lightlike triangles.
In particular, the spacelike part of $(S_K)_\infty$ in nonempty unless $\Gamma$ is a 1-step or a 2-step curve.
*Step 3: The spacelike part of $(S_K)_\infty$ is a smooth $K$-surface.* Let $x$ be any point of $(S_K)_\infty$, in the complement of the union of all maximal lightlike triangles, and let $x_n\in (S_K)_n$ such that $x_n\to x$. Regarding $(S_K)_n$ as the graph of a Lipschitz function in the conformal model of , we argue that there is a cylindrical neighborhood $U$ of $x$ such that the restriction of the projection $\pi:{\mathbb{A}\mathrm{d}\mathbb{S}}^3\to {\mathbb{D}}^2$ is uniformly bi-Lipschitz on $U\cap (S_K)_n$. We deduce that there exists $\epsilon$ such that the metric ball $B_{(S_K)_n}(x,\epsilon)$ is compact in $(S_K)_n$, contained in $U$, and thus its projection is uniformly bounded in ${\mathbb{D}}^2$. Recall that the induced metric on $(S_K)_n$ has constant curvature $K$, and thus, up to rescaling the metric by a factor $|K|$, it is isometric to a subset of ${\mathbb{H}}^2$. Let $o\in{\mathbb{H}}^2$ and let $$\sigma_n=:B_{{\mathbb{H}}^2}(o,\epsilon)\to S_n$$ be an isometric embedding (after rescaling the metric by $|K|$) with $\sigma_n(o)=x_n$. Composing with the projection $\pi$, it then turns out that $\pi\circ\sigma_n$ are uniformly bounded and uniformly bi-Lipschitz. Hence the maps $\sigma_n$ converge $C^0$ up to a subsequence. The limit map $\sigma_\infty$ has image $B_{(S_K)_\infty}(x,\epsilon)$.
Observe that the 1-jet of $\sigma_n$ at $0$ is bounded. Indeed, if the 1-jet were not bounded, there would be a subsequence $n_k$ such that the tangent planes $T_{x_{n_k}}S_{n_k}$ converge to a lightlike plane, which is a support plane for $(S_K)_\infty$ at $x$. Hence $x$ is a point in the boundary of the domain of dependence $\mathcal{D}(\Gamma)$, contradicting the previous claim.
We can thus extract a subsequence in such a way that $j^1\sigma_n(0)$ converges. Since $\sigma_n$ are isometric embeddings of $B_{{\mathbb{H}}^2}(o,\epsilon)$, (up to rescaling by the constant factor $|K|$), the pull-back $\sigma_n^*(g_{{\mathbb{A}\mathrm{d}\mathbb{S}}^3})$ of the Anti-de Sitter metric is constant, and we are in the right assumptions to apply Theorem \[schl\_degeneration\]. It follows that $\sigma_n$ converges $C^\infty$ in a neighborhood of $0$. Indeed, if this were not the case, by Theorem \[schl\_degeneration\] the surfaces $(S_K)_n$ would converge to the union of two half-planes of lightlike type in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, intersecting along a spacelike geodesic, and this possibility is ruled out again by the fact that $x$ is in the spacelike part of $(S_K)_\infty$. This concludes the proof that $\sigma_n$ converges $C^\infty$, and therefore the spacelike part of $(S_K)_\infty$ is smooth and has curvature $K$ at every point $x$.
Foliations of the complement of the convex hull {#sec foliations}
===============================================
The purpose of this section is to prove that the $K$-surfaces obtained in Theorem \[prop existence\] provide foliations of $\mathcal D_+(\Gamma)$ and $\mathcal D_-(\Gamma)$, as $K\in(-\infty,-1)$. We first compute the curvature of a spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, expressed as a graph over the horizontal plane in an affine chart for the projective model of Subsection \[subsec models\].
\[prop formula affine curvature\] Let $V$ be an open subset of a smooth spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Suppose that, in an affine chart of the projective model for which ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is the interior of the standard one-sheeted hyperboloid, $V$ is expressed as the graph of a function $u:{{\mathbb R}}^2\to{{\mathbb R}}$, namely: $$V=\{(x,y,t):(x,y)\in\Omega,\,t=u(x,y)\}\,,$$ where $\Omega\subset{{\mathbb R}}^2$ is some domain. Then $u$ satisfies the equation: $$\label{monge ampere graph}
\det D^2 u=\frac{(1-|Du|^2+((x,y)\cdot Du-u)^2)^2}{(1+u^2-x^2-y^2)^2}\det B\,,$$ where $B$ is the shape operator of $S$.
Recall that in the projective description of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ of Subsection \[subsec models\], the double cover ${\mathrm{SL}}(2,{{\mathbb R}})$ of ${\mathrm{PSL}}(2,{{\mathbb R}})$ is identified to the subset of ${{\mathbb R}}^4$ where the quadratic form $x_1^2+x_2^2-x_3^2-x_4^2$ takes the value $-1$. Hence we will denote by ${{\mathbb R}}^{2,2}$ the ambient ${{\mathbb R}}^4$ with the standard bilinear form of signature $(2,2)$, and we will suppose that the affine chart is given by $x_4\neq 0$. We use the affine coordinates: $$[x_1:x_2:x_3:x_4]\longrightarrow (x,y,t)=\left(\frac{x_1}{x_4},\frac{x_2}{x_4},\frac{x_3}{x_4}\right)\,,$$ so that the horizontal hyperplane $H_0$ corresponds to $x_3=0$. Denote by $H$ the hyperplane defined by $x_4=1$, and by $H^{-}$ the region of $H$ where the quadratic form is negative, that is the image of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ in the affine chart. We denote by $\xi: H^{-}\to {\mathrm{SL}}(2,{{\mathbb R}})$ the radial map $\xi(x_1,x_2,x_3,1)=f(x_1,x_2,x_3)\cdot (x_1,x_2,x_3,1)$ where $f(x_1, x_2,x_3)=(1+x_3^2-x_1^2-x_2^2)^{-1/2}$. The local parameterization of $S$ in the affine chart is of the form $$\bar\sigma(x,y)=(x,y,u(x,y),1)\,,$$ while the parameterization of the lifting of the surface in ${\mathrm{SL}}(2,{{\mathbb R}})$ is $\sigma=\xi\circ\bar\sigma=f(\bar\sigma)\cdot\bar\sigma$. We want to compute the determinant of the shape operator $B$ of $\sigma$. If $I$ e ${I\hspace{-0.1cm}I}$ denote the first and second fundamental form matrices of $\sigma$ in the $\{\partial_x,\partial_y\}$ frame, it turns out that $$\det B=(\det{I\hspace{-0.1cm}I})/(\det I)~.$$ First let us compute $\det I$. If $\boxtimes$ denotes the vector product on ${\mathrm{SL}}(2,{{\mathbb R}})$, then $\sigma_x\boxtimes\sigma_y$ is a tangent vector at $\sigma$ normal to immersion. In particulat it is timelike and by standard facts $$\det I=-\langle\sigma_x\boxtimes\sigma_y, \sigma_x\boxtimes\sigma_y\rangle~.$$ Considering ${\mathrm{SL}}(2,{{\mathbb R}})$ as a submanifold of ${{\mathbb R}}^{2,2}$ we can orient ${{\mathbb R}}^{2,2}$ so that the positive normal vector at $\eta\in{\mathrm{SL}}(2,{{\mathbb R}})$ is $N(\eta)=\eta$. Now it is immediate to check that the following formula holds for any $v,w\in T_\eta{\mathrm{SL}}(2,{{\mathbb R}})$ $$v\boxtimes w=*(\eta\wedge v\wedge w)\,,$$ where $*:\Lambda^3({{\mathbb R}}^4)\to{{\mathbb R}}^4$ is the Hodge-operator on ${{\mathbb R}}^{2,2}$. So a direct computation shows that $$\sigma_x\boxtimes\sigma_y=*(f\bar\sigma\wedge d\xi(\bar\sigma_x)\wedge d\xi(\bar\sigma_y))=f^3\,*(\bar\sigma\wedge\bar\sigma_x\wedge\bar\sigma_y)$$ Now $\bar\sigma=\bar\sigma_0+e_4$, where $e_4$ is the positive normal of the hyperplane $H$, and $$\bar\sigma_0(x,y)=(x,y,u(x,y),0)\,.$$ Denoting with $\boxtimes$ also the vector product on $H$ (that is intrinsically a copy of Minkowski space), $$*(e_4\wedge\bar\sigma_x\wedge\bar\sigma_y)=(\bar\sigma_x\boxtimes\bar\sigma_y)=\nu\,,$$ where it turns out that $\nu=(u_x,u_y,1,0)$. Writing $\bar\sigma_0=\bar\sigma-e_4$, one obtains: $$*(\bar\sigma_0\wedge\bar\sigma_x\wedge\bar\sigma_y)=\langle \sigma_0, \nu\rangle e_4=\langle \sigma, \nu\rangle e_4\,.$$ We thus conclude that $$\label{eq:hinwil}
\sigma_x\boxtimes\sigma_y=f^3(\nu+\langle\bar\sigma,\nu\rangle e_4)$$ and thus $\det I=f^6g$ where $g(x,y)=\langle\bar\sigma,\nu\rangle^2-\langle\nu,\nu\rangle=1-|Du|^2+((x,y)\cdot Du-u)^2$.
To compute ${I\hspace{-0.1cm}I}$, notice that the normal vector of the immersion $\sigma$ is $$N=\frac{\sigma_x\boxtimes\sigma_y}{\sqrt{\det I}}\in T_\sigma{\mathrm{SL}}(2,{{\mathbb R}})~.$$ Now ${I\hspace{-0.1cm}I}_{11}=\langle N, \sigma_{xx}\rangle$. Observe that $\sigma_{xx}=f_{xx}\bar\sigma+2f_x\bar\sigma_x+f\bar\sigma_{xx}$. Using that $\sigma_x=f_x\bar\sigma+f\bar\sigma_x$ and that $N$ is orthogonal to $\sigma_x$ and to $\bar\sigma$, we get $${I\hspace{-0.1cm}I}_{11}=f \langle N, \bar\sigma_{xx}\rangle=\frac{f}{\sqrt{\det I}}\langle \sigma_x\boxtimes\sigma_y, \bar\sigma_{xx}\rangle=\frac{f^4}{\sqrt{\det I}}\langle \bar\sigma_x\boxtimes\bar\sigma_y, \bar\sigma_{xx}\rangle\,,$$ where in the last equality we have used Equation and the fact that $\langle e_4,\bar\sigma_{xx}\rangle=0$, since $\bar\sigma_{xx}=(0,0,u_{xx},0)$. Hence, observing that $\langle\nu,\bar\sigma_{xx}\rangle=-u_{xx}$, one finally has: $${I\hspace{-0.1cm}I}_{11}
=-\frac{f^4}{\sqrt{\det I}}u_{xx}$$ With analogous computations we get $${I\hspace{-0.1cm}I}=-\frac{f^4}{\sqrt{\det I}}D^2u\,,$$ and therefore $\det{I\hspace{-0.1cm}I}=\frac{f^8}{\det I}\det D^2 u$. This concludes the proof that $$\det B=\frac{f^8}{(\det I)^2}\det D^2 u=f^{-4}g^{-2}\det D^2 u~.$$ which corresponds to Equation .
\[remark degeneration\] From the proof of Proposition \[prop formula affine curvature\], the factor $(1-|Du|^2+((x,y)\cdot Du-u)^2)$ does not vanish as long as $S$ is a spacelike surface, whereas the factor $(1+u^2-x^2-y^2)$ vanishes precisely when $(x,y,u(x,y))$ is in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$.
We will now introduce a notion of *locally pleated* surface. Recall that the definition of locally convex nowhere timelike surface was given in Definition \[defi locally convex\].
A locally convex nowhere timelike surface $S$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ is *locally pleated* if for every $\gamma\in S$ there exists a neighborhood $U$ of $\gamma$ in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ such that $S\cap U$ is contained in the boundary of the convex hull of $S\cap \partial U$.
The condition that $S$ is a locally pleated surface can be expressed by saying that no point of $S$ is a vertex (or extremal point) in the sense of convex geometry ([@rockafellar]). Using this characterization, we prove the following lemma.
\[lemma boundary convex hull\] Let $S$ be a locally pleated surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$, with $\partial S=\Gamma$ a weakly acausal curve in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$. Then $S$ is a pleated surface, that is, $S$ coincides with a boundary component of the convex hull of $\Gamma$.
Let $K$ be the convex hull of $S$. By convexity, $S$ lies in $\partial K$. By Krein-Milman Theorem ([@kreinmilman]), $K$ is the convex hull of extreme points of $S\cup \partial S$. As $S$ contains no extreme points from the above definition, we conclude that $K$ is the convex hull of $\partial S$.
Recall that, given a convex function $u:\Omega\subset{{\mathbb R}}^2\to{{\mathbb R}}$, the *Monge-Ampère measure* $M\!A_u$ of $u$ is defined in such a way that, if $\omega\subseteq \Omega$ is a Borel subset, then $M\!A_u(\omega)$ is the Lebesgue measure of the union of the subdifferentials of $u$ over points in $\omega$. In particular, if $u$ is $C^2$, then $$M\!A_u(\omega)=\int_\omega (\det D^2 u)d\mathcal L\,.$$
We will use the following property of Monge-Ampère measures:
\[convergence of solutions\] Given a sequence of convex functions $u_n$ which converges uniformly on compact sets to $u$, the Monge-Ampère measure $M\!A_{u_n}$ converges weakly to $M\!A_{u}$.
Moreover, the following is a characterization of functions with vanishing Monge-Ampère measure:
\[thm gutierrez\] Given a convex function $u:\Omega\subset{{\mathbb R}}^2\to{{\mathbb R}}$, $u\in C^0(\overline\Omega)$, where $\Omega$ is a convex bounded domain, if $M\!A_u\equiv 0$, then $u$ is the convex envelope of $u|_{\partial\Omega}$.
We are now ready to prove that the $K$-surfaces with prescribed boundary a weakly spacelike curve $\Gamma\subset\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$, whose existence was proved in Theorem \[prop existence\], provide a foliation of $\mathcal{D}_+(\Gamma)$ and $\mathcal{D}_-(\Gamma)$.
\[thm foliation part ads\] Given any weakly acausal curve $\Gamma$ in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ which is not a 1-step or a 2-step curve, $\mathcal{D}_+(\Gamma)$ and $\mathcal{D}_-(\Gamma)$ are foliated by $K$-surfaces $S^\pm_K$, as $K\in(-\infty,-1)$, in such a way that if $K_1<K_2$, then $S^\pm_{K_2}$ is in the convex side of $S^\pm_{K_1}$.
For definiteness, we give the proof for the region $\mathcal{D}_+(\Gamma)$, as the other case is completely analogous.
As in the proof of Theorem \[prop existence\], let $\Gamma_n$ be a sequence of curves in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ invariant for a pair $((\rho_l)_n,(\rho_r)_n)$ of torsion-free cocompact Fuchsian representations of $\pi_1(\Sigma_{g_n})\to{\mathrm{Isom}}({\mathbb{H}}^2)\times {\mathrm{Isom}}({\mathbb{H}}^2)$. Recall that in Theorem \[prop existence\] we showed that for every $K\in(-\infty,-1)$ there exists a subsequence on which the $K$-surfaces $(S_K)_{n}$ invariant for $((\rho_l)_n,(\rho_r)_n)$ converge to a convex nowhere timelike surface $(S_K)_\infty$ with $\partial (S_K)_\infty=\Gamma$, whose lightlike part is the union of all maximal past-directed lightlike triangles bounded by $\Gamma$, and whose spacelike part is a smooth $K$-surface. By choosing a numeration of ${{\mathbb Q}}\cap(-\infty,-1)$, by a classical diagonal argument we can extract a subsequence $n_k$ such that for every $q\in{{\mathbb Q}}$, as $k\to\infty$, $(S_q)_{n_k}$ converges uniformly on compact sets to a convex surface, whose spacelike part is a smooth $q$-surface, which we denote (with a slight abuse of notation) by $S_q$. Let us denote again by $(S_q)_n$ the converging subsequence, omitting the subindex $k$. We will prove that the family of surfaces $\{S_q\}$ extend to a foliation of $\mathcal D_+(\Gamma)$ by $K$-surfaces. *Step 1: The spacelike parts of the surfaces $S_q$ are pairwise disjoint.* Indeed, for every $q_1<q_2$, the surface $(S_{q_2})_n$ lies in the convex side of $(S_{q_1})_n$. Therefore the same holds for $S_{q_2}$ and $S_{q_1}$, and in particular they do not intersect transversely. Moreover, it is not possible that $S_{q_2}$ and $S_{q_1}$ are tangent at one point in the spacelike part, since the determinant of the shape operator of $S_{q_2}$ is strictly smaller than the determinant of the shape operator of $S_{q_1}$. Therefore if they were tangent, $S_{q_2}$ could not be in the convex side of $S_{q_1}$, since this would contradict the maximum principle.
*Step 2: For every irrational $K<-1$ the sequences $\{S_{q'}:q'>K\}$ and $\{S_{q''}:q''<K\}$ both converge to the same surface.* Let us denote $$(S_K)'=\lim_{q'>K} S_{q'}\qquad\text{and}\qquad (S_K)''=\lim_{q''<K} S_{q''}\,,$$ where the limit clearly exists as the sequences are monotone. Moreover, since every $S_{q'}$ is in the convex side of every $S_{q''}$, the limit $(S_K)'$ is in the convex side of $(S_K)''$, and in particular they do not intersect transversely. Suppose by contradiction that there exists an open set $U$ which is contained in the region between $(S_K)'$ and $(S_K)''$. Now, since the surfaces $(S_K)_n$ foliate $\mathcal D_+(\Gamma_n)$ by Theorem \[bbz\], for large $n$ there is a surface $(S_{q_n})_n$ in the approximating sequence, for $q_n\in{{\mathbb Q}}\cap(-\infty,-1)$, which intersects $U$. By the uniform convergence on compact sets, for $n$ larger than some $n_0$, $q_n<q'$ for every $q'>K$. But for the same argument, for some possibly larger $n_0$, if $n\geq n_0$ then $q_n>q''$ for every $q''<K$. This gives a contradiction.
*Step 3: The spacelike part of the limit $S_K:=(S_K)'=(S_K)''$ is a smooth $K$-surface.* By the usual application of Theorem \[schl\_degeneration\], for every $\gamma\in S_K$, pick a sequence $\gamma_{q'}\in S_{q'}$ converging to $\gamma$ and pick the homothetic embeddings $$\sigma_{q'}:B_{{\mathbb{H}}^2}(o,\epsilon)\to S_{q'}\,.$$ As in the proof of Theorem \[prop existence\], the $\sigma_{q'}$ converge $C^\infty$ and therefore $S_K$ is a smooth $K$-surface.
So far we have proved that $\{S_K:K\in(-\infty,-1)\}$ is a foliation by smooth $K$-surfaces of a subset of $\mathcal{D}_+(\Gamma)$. We finally need to prove that $\bigcup\{S_K:K\in(-\infty,-1)\}$ coincides with $\mathcal{D}_+(\Gamma)$.
*Step 4: The surfaces $S_K$ approach the spacelike part of $\partial_+\mathcal{C}(\Gamma)$ as $K\to-1$.* As $S_K$ is a monotone sequence in $K$, consider the $C^0$-limit $$S_{-1}:=\lim_{K\to -1}S_K\,.$$
We claim that $S_{-1}$ is locally pleated. As the lightlike part of $S_{-1}$ is the union of lightlike triangles corresponding to sawteeth of $\Gamma$, we only have to prove the claim for the spacelike part of $S_{-1}$. Let $\gamma\in S_{-1}$ and let us choose an affine chart so that a support plane of $S_{-1}$ at $\gamma$ is the horizontal plane as in Proposition \[prop formula affine curvature\].
Hence, in a neighborhood of $\gamma$, the surfaces $S_K$ are the graph of a convex function $u_K:\Omega_K\to[0,+\infty)$. Using Remark \[remark degeneration\], as $K\to -1$ then $(1+u_K(x,y)^2-x^2-y^2)$ remains bounded for every $(x,y)\in\Omega_{-1}$ (since the point $(x,y,u_{-1}(x,y))$ is not in the boundary of ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$). Also $(1-|Du_K(x,y)|^2+((x,y)\cdot Du_K(x,y)-u_K(x,y))^2)$ remains bounded for $(x,y)\in\Omega_{-1}$, as otherwise the limiting surface would have a lightlike support plane over $(x,y)$, and this is not possible for the usual argument, as $\gamma$ is in the spacelike part of $S_{-1}$. Hence, applying Lemma \[convergence of solutions\] to Equation , we get that $u_{-1}$ satisfies $\det D^2 u_{-1}\equiv 0$ on $\Omega_{-1}$. Since $\gamma$ was arbitrary, by Theorem \[thm gutierrez\], $S_{-1}$ is locally pleated. By Lemma \[lemma boundary convex hull\], $S_{-1}$ coincides with $\partial_+ \mathcal C(\Gamma)$ as claimed.
*Step 5: The surfaces $S_K$ approach the boundary $\partial_+\mathcal{D}(\Gamma)$ as $K\to-\infty$.* As before, we can take the limit $$S_{-\infty}=\lim_{K\to-\infty}S_K\,.$$ Suppose that $S_{-\infty}$ does not coincide with the boundary of $\mathcal{D}(\Gamma)$. Since $\partial_-\mathcal C(\Gamma)$ is the dual surface to $\partial_+\mathcal{D}(\Gamma)$, the dual surfaces $(S_K)^*$ would converge to $(S_{-\infty})^*$. But the shape operator of $(S_K)^*$ is identified to the inverse of the shape operator of $S_K$, and thus the $(S_K)^*$ are again surfaces of constant Gaussian curvature $K^*=-K/(K+1)$. In particular $K^*$ tends to $-1$ as $K\to-\infty$. Thus applying the same argument as in Step 4, one gets a contradiction.
Uniqueness and boundedness of principal curvatures {#sec uniqueness boundedness}
==================================================
In this section we consider the problem of the uniqueness of the $K$-surfaces with fixed boundary curve $\Gamma$ (which we will be able to prove under the assumption that $\Gamma$ is the graph of a quasisymmetric homeomorphism) and the boundedness of principal curvatures. The two problems are tackled with similar techniques, and the key tool is the following compactness theorem for quasisymmetric homeomorphisms:
\[Compactness property of quasisymm homeo\] Let $\phi_n:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$ be a family of orientation-preserving quasisymmetric homeomorphisms of $\partial {\mathbb{H}}^2$. Suppose there exists a constant $M$ such that, for every symmetric quadruple $Q$ (i.e. such that the cross-ratio of $Q$ is $cr(Q)=-1$) and for every $n$, $$\label{cross-ratio boundedness}
|\log|cr\phi_n(Q)||\leq M\,.$$ Then there exists a subsequence $\phi_{n_k}$ for which one of the following holds:
- The homeomorphisms $\phi_{n_k}$ converge uniformly to a quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$;
- The homeomorphisms $\phi_{n_k}$ converge uniformly on the complement of any open neighborhood of a point of $\partial{\mathbb{H}}^2$ to a constant map $c:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$.
We start with a remark, which shows us that the condition that $\Gamma=gr(\phi)$, with $\phi$ quasisymmetric, is not restrictive for the study of surfaces with bounded principal curvatures. That is, if $S$ is a surface with bounded principal curvatures and $\partial S=\Gamma$, then $\Gamma$ is necessarily the graph of a quasisymmetric homeomorphism.
Suppose $S$ is a convex spacelike surface in ${\mathbb{A}\mathrm{d}\mathbb{S}}^3$ with $\partial S=\Gamma$, such that the principal curvatures of $S$ are bounded. We claim that the associated map $\Phi$ is a bi-Lipschitz diffeomorphism of ${\mathbb{H}}^2$. In particular $\Phi$ is quasiconformal and $\Gamma$ is the graph of the quasisymmetric extension of $\Phi$. By the same argument as in the proof of [@bebo Proposition 6.21], the bound on the second fundamental form implies that $S$ is geodesically complete. Let us show that the projections $\pi_l,\pi_r$ are bi-Lipschitz. Using the completeness of $S$, this implies that they are coverings, hence diffeomorphisms. Therefore $\Phi$ is a bi-Lipschitz diffeomorphism of ${\mathbb{H}}^2$.
The minimal and maximal stretch of $\pi_l$ (resp. $\pi_r$) are the eigenvalues of $E+JB$ (resp. $E-JB$). Since $B$ is bounded, the maximal stretch is bounded from above by some constant $M$. To obtain the bound from below, it suffices to notice that $\det(E\pm JB)=1+\det B>1$. Hence the minimal stretch is bounded from below by $1/M$. This proves that $\pi_l$ and $\pi_r$ are bi-Lipschitz.
Another basic observation is the fact that the condition $\Gamma=gr(\phi)$, $\phi$ being quasisymmetric, is equivalent to the condition that the measured geodesic lamination on the upper (or on the lower) boundary of the convex hull of $\Gamma$ is bounded. Indeed, the latter condition is equivalent to saying that the Thurston norm of the left (or right) earthquake lamination is bounded, which is known to be equivalent to quasisymmetry of $\phi$, as proved independently in [@garhulakic] and [@Saric:2006ds] (see also [@thurstonearth; @Saric:2008ek]). Boundedness of the measured geodesic lamination can be seen as a condition of boundedness of curvatures for the non-smooth surface $\partial_\pm\mathcal C(\Gamma)$.
Theorem \[thm uniqueness and boundedness\] gives a similar statement for $K$-surfaces, that is, a $K$-surface $S$ with $\partial S=\Gamma$ has bounded principal curvatures if and only if $\phi$ is quasisymmetric - and moreover it is unique. The proof of Theorem \[thm uniqueness and boundedness\] is split in two propositions, which are proved with similar techniques. Proposition \[prop uniqueness quasisym\] first proves the uniqueness part of Theorem \[thm uniqueness and boundedness\], while Proposition \[prop boundedness quasisym\] proves boundedness of principal curvatures.
\[prop uniqueness quasisym\] Given any quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$, for every $K\in(-\infty,-1)$ there exists a unique (spacelike, smooth) future-convex $K$-surface $S^-_K$ and a unique past-convex $K$-surface $S_K^+$ with $\partial S^\pm_K=gr(\phi)$.
Let us give the proof for past-convex surfaces, for definiteness. Let $\Gamma=gr(\phi)$. We have already proved the existence of a foliation by $K$-surfaces $S_K^+$ of $\mathcal D_+(\Gamma)$. Hence we have a function $\kappa:\mathcal D_+(\Gamma)\to(-\infty,-1)$, such that $\kappa(\gamma)=\kappa_0$ if $\gamma\in S_{\kappa_0}^+$. Let $S$ be another $K$-surface, for fixed $K$. We must show that $S$ coincides with $S_K^+=\kappa^{-1}(K)$.
Let $$\kappa_{\mathrm{max}}=\sup_{\gamma\in S} \kappa(\gamma)\qquad\text{and}\qquad \kappa_{\mathrm{min}}=\inf_{\gamma\in S} \kappa(\gamma)\,.$$ Let us show that $\kappa_{\mathrm{max}}\leq K$. Then by an analogous argument, we will have $\kappa_{\mathrm{min}}\geq K$, and thus $\kappa_{\mathrm{max}}=\kappa_{\mathrm{min}}=K$.
Suppose the supremum $\kappa_{\mathrm{max}}$ is achieved at some point $\gamma_{\mathrm{max}}$, such that $\kappa_{\mathrm{max}}=\kappa(\gamma_{\mathrm{max}})$. First, it follows that $\kappa_{\mathrm{max}}>-\infty$, for otherwise $S$ would touch the boundary of the domain of dependence. Then in this case, $S$ and $S_{\kappa_{\mathrm{max}}}$ are tangent at $\gamma$, and $S_{\kappa_{\mathrm{max}}}$ is on the convex side of $S$. Hence by a standard application of the maximum principle, the determinant of the shape operator of $S_{\kappa_{\mathrm{max}}}$ at $\gamma$ is larger than the determinant of the shape operator of $S$ (which equals $-1-K$). Therefore $K\geq \kappa_{\mathrm{max}}$.
On the other hand, suppose that $\kappa_{\mathrm{max}}=\lim_n \kappa(\gamma_n)$, where $\gamma_n\in S$ is a diverging sequence. Let $\chi_n\in {\mathrm{Isom}}({\mathbb{H}}^2)$ such that $\chi_n(\gamma_n)=\mathcal I_x$ (for $x\in{\mathbb{H}}^2$ a fixed point) and $\chi_n(P_{\gamma_n}S)=\mathcal R_\pi$, where $P_{\gamma_n}S$ is the totally geodesic plane tangent to $S$ at $\gamma_n$. Observe that, if $\Gamma=gr(\phi)$ is the curve in $\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ of $S$, then $\chi_n(\Gamma)=gr(\phi_n)$, where $\phi_n$ is obtained by pre-composing and post-composing $\phi$ with isometries of ${\mathbb{H}}^2$. Hence the condition of Equation in Lemma \[Compactness property of quasisymm homeo\] is satisfied.
We claim that there can be no subsequence $\phi_{n_k}$ which converges to a constant map. In terms of Anti-de Sitter geometry, this means that the curve $\chi_{n_k}(\Gamma)$ would converge (in the Hausdorff convergence) to the boundary of a totally geodesic lightlike plane $P$. Let $\xi\in\partial{\mathbb{A}\mathrm{d}\mathbb{S}}^3$ be the point which determines the lightlike plane, namely the self-intersection point of $\partial P$ (see Figure \[fig:step\], left). Since $P_{\gamma_n}S$ is a support plane for $S$, $\partial P_{\gamma_n}S$ does not intersect $\Gamma$ for every $n$. By applying $\chi_n$, this means that $\partial \mathcal R_\pi$ does not intersect $\chi_n(\Gamma)$, and thus $\xi$ must necessarily be in $\partial \mathcal R_\pi$. See Figure \[fig:lightlikeplane\]. But in this case, $(\mathcal I_x)^*$ does not contain $\xi$, so $(\mathcal I_x)^*$ meets $\chi_n(\Gamma)$ for $n$ large, which contradicts the fact that $\mathcal I_x$ is in the domain of dependence of $\chi_n(\Gamma)$ for every $n$.
![The position of the lightlike plane $P$ of the claim in the proof of Proposition \[prop uniqueness quasisym\]. \[fig:lightlikeplane\]](lightlikeplane-eps-converted-to.pdf){height="6.5cm"}
Therefore, by Lemma \[Compactness property of quasisymm homeo\], there exists a subsequence $\phi_{n_k}$ which converges to a quasisymmetric homeomorphism $\phi_\infty$ of $\partial{\mathbb{H}}^2$. Let $\Gamma_\infty=gr(\phi_\infty)$. By the same argument as the proofs of Theorem \[prop existence\] and Theorem \[thm foliation part ads\], there exists a foliation of $\mathcal D_+(\Gamma_\infty)$ by $K$-surfaces which are obtained as limits (up to taking subsequences) of the surfaces $\chi_n(S^+_K)$. Moreover, taking a further subsequence, we can suppose $\chi_n(S)$ converges to a smooth $K$-surface $S_\infty$ contained in $\mathcal D_+(\Gamma_\infty)$ (as in the proof of Theorem \[prop existence\]). Now we are again in the situation of the beginning of the proof, where $\kappa_{\mathrm{max}}=\sup_{\gamma\in S_\infty} \kappa_\infty(\gamma)$, and this supremum is achieved at $\mathcal I_x$. Thus applying again the above argument, the proof is concluded.
We will now apply a similar argument to show that, under the assumption that $\Gamma$ is the graph of a quasisymmetric homeomorphism, the principal curvatures of the $K$-surfaces $S_K$ with $\partial S_K=\Gamma$ are uniformly bounded.
\[prop boundedness quasisym\] Let $S$ be a $K$-surface, for $K\in(-\infty,-1)$, with $\partial S=gr(\phi)$, where $\phi:\partial{\mathbb{H}}^2\to \partial{\mathbb{H}}^2$ is a quasisymmetric homeomorphism. Then the principal curvatures of $S$ are bounded.
By contradiction, assume that there exists a sequence of points $\gamma_n\in S$ where one principal curvature $\lambda_1(\gamma_n)$ tends to infinity (and thus the other tends to zero). As in the proof of Proposition \[prop uniqueness quasisym\], let $\chi_n\in {\mathrm{Isom}}({\mathbb{H}}^2)$ such that $\chi_n(\gamma_n)=\mathcal I_x$ and $\chi_n(T_{\gamma_n}S)=\mathcal R_\pi$. By the same argument, the curves $\chi_n(\Gamma)$ converge to the graph of a quasisymmetric homeomorphism $\Gamma_\infty=gr(\phi_\infty)$, and as in Theorem \[prop existence\], the surfaces $\chi_n(S)$ converge to a smooth $K$-surface $S_\infty$ with $\partial S_\infty=\Gamma_\infty$. But then for the $C^\infty$ convergence, the principal curvatures $\lambda_1(\gamma_n)$, which are equal to the largest principal curvature of $\chi_n(S)$ at $\chi_n(\gamma_n)=\mathcal I_x$, converge to the largest principal curvature of $S_\infty$ at $\mathcal I_x$. Thus they cannot go to infinity and this gives a contradiction.
We remark that the key points in the proofs of Proposition \[prop uniqueness quasisym\] and Proposition \[prop boundedness quasisym\] are the use of the compactness result for quasisymmetric homeomorphisms (Lemma \[Compactness property of quasisymm homeo\]), and on the other hand, the convergence result of Theorem \[schl\_degeneration\], which essentially can be fruitfully applied (as for Theorem \[prop existence\]) provided the curve $\Gamma$ is not a 1-step or a 2-step curve.
Proposition \[prop uniqueness quasisym\] and Proposition \[prop boundedness quasisym\] together give the statement of Theorem \[thm uniqueness and boundedness\].
Recall that the map $\Phi:\pi_r\circ(\pi_l)^{-1}$ associated to a past-convex $K$-surface $S$ is a $\theta$-landslide, for $K=-{1}/{\cos^2({\theta}/{2})}$, and that if $\partial S=gr(\phi)$ then $\Phi$ extends to $\phi$ on $\partial{\mathbb{H}}^2$. Then we conclude by a corollary about the extensions by $\theta$-landslides of quasisymmetric homeomorphisms.
Given any quasisymmetric homeomorphism $\phi:\partial{\mathbb{H}}^2\to\partial{\mathbb{H}}^2$ and any $\theta\in(0,\pi)$, there exist a unique $\theta$-landslides $\Phi_\theta:{\mathbb{H}}^2\to{\mathbb{H}}^2$ which extend $\phi$. Moreover, $\Phi_\theta$ is quasiconformal.
In fact, Theorem \[prop existence\] and Proposition \[prop uniqueness quasisym\] prove existence and uniqueness of a past-convex $K$-surface for every $K\in(-\infty,-1)$. Therefore the associated map is a $\theta$-landslide. On the other hand, given any $\theta$-landslide $\Phi_\theta$, by the procedure described in Section \[sec representation formula\], in particular Corollary \[cor reconstruct ksurf\], $\Phi_\theta$ must necessarily come from one of the two $K$-surfaces with boundary $gr(\phi)$.
[^1]: The authors were partially supported by FIRB 2010 project “Low dimensional geometry and topology” (RBFR10GHHH003). The first author was partially supported by PRIN 2012 project “Moduli strutture algebriche e loro applicazioni”. The authors are members of the national research group GNSAGA
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The use of Convolutional Neural Networks (CNN) to estimate the galaxy photometric redshift probability distribution by analysing the images in different wavelength bands has been developed in the recent years thanks to the rapid development of the Machine Learning (ML) ecosystem. Authors have set-up CNN architectures and studied their performances and some sources of systematics using standard methods of training and testing to ensure the generalisation power of their models. So far so good, but one piece was missing : does the model generalisation power is well measured? The present article shows clearly that very small image perturbations can fool the model completely and opens the Pandora’s box of *adversarial* attack. Among the different techniques and scenarios, we have chosen to use the Fast Sign Gradient one-step Method and its Projected Gradient Descent iterative extension as adversarial generator tool kit. However, as unlikely as it may seem these adversarial samples which fool not only a single model, reveal a weakness both of the model and the classical training. A revisited algorithm is shown and applied by injecting a fraction of adversarial samples during the training phase. Numerical experiments have been conducted using a specific CNN model for illustration although our study could be applied to other models - not only CNN ones - and in other contexts - not only redshift measurements - as it deals with the complexity of the boundary decision surface.'
author:
- |
Jean-Eric Campagne$^{1}$[^1]\
$^{1}$Université Paris-Saclay, CNRS/IN2P3, IJCLab, 91405 Orsay, France\
bibliography:
- 'papier.bib'
title: Adversarial training applied to Convolutional Neural Network for photometric redshift predictions
---
\[firstpage\]
galaxies: distances and redshifts - surveys - techniques:photometric redshifts - methods: data analysis - statistical
Introduction
============
The outcomes of forthcoming very large optical surveys as LSST [@2008arXiv0805.2366I; @Ivezi__2019] or red-optical and infrared surveys as Euclid [@2011arXiv1110.3193L] will depend crucially up on the galaxy [^2] positions on the sky and especially along the light of sight, i.e. the redshift. As the accurate spectroscopic redshift determinations are costly and time consuming such that they are limited to sub-samples of galaxies as in BOSS (SDSS-III) [@2015ApJS..219...12A] and eBOSS (SDSS-IV) [@2016AJ....151...44D] and the future DESI [@2013arXiv1308.0847L] or Euclid (spectro sub-sample) surveys, the majority of the galaxy redshifts of past surveys as DES Y1 [@2019MNRAS.489.5453S] and future as LSST and Euclid rely on the photometric measurements. The reconstruction performances of photometric redshift (hereafter named photo-z or $z_{phot}$) has been extensively studied in the literature since the work of [@1962IAUS...15..390B] with different kinds of methods such as *template-fitting*, *feature based machine learning* and *image based machine learning*.
The *template-fitting* methods, first developped by [@1986ApJ...303..154L], have been used for instance in the references [@1999MNRAS.310..540A; @2000ApJ...536..571B; @2006MNRAS.372..565F; @2008ApJ...686.1503B] and recently in . For a given galaxy, the algorithm matches the magnitude distributions in the different filter wide bands (e.g. *ugriz*) to a suite of interpolate spectral energy distributions from few known galaxies and it results a photo-z value and error in a fully probabilistic fashion with explicit priors.
The *feature based machine learning* methods use different types of tools as Decision Trees (DT; [@Quinlan:1986:IDT:637962.637969]), Random Forests (RF; [@Breiman:2001:RF:570181.570182]), Support Vector Machine (SVM; [@Boser:1992:TAO:130385.130401]), k-nearest neighbors (KNN; [@doi:10.1080/00031305.1992.10475879]) as well as Muti-Perceptron Layers (nicknamed either MLP or ANN; [@Werbos:74; @Rumelhart1986LearningRB]). They have been used for instance in the references . These methods use as input for a given galaxy the different magnitudes (or colors) measured in the different filters augmented eventually by other user driven information, and give as output a single photo-z value or a probability density distribution (p.d.f). But contrary to the *template-fitting* methods, the prediction is based on the determination of the internal parameters of each tool during a supervised preliminary phase, where a sub-sample of galaxies with known redshifts are processed. The later redshifts considered as the *true* values are mostly known thanks to the accurate spectroscopic measurements. The *feature based machine learning* methods have shown better prediction accuracy than the *template-fitting* methods although a combination of the two class of methods has been investigated for instance in reference [@2017MNRAS.466.2039C]. The *feature based machine learning* methods rely explicitly on manual feature extractions that may not capture all the information present in the images: for example the point spread function (PSF) variations are problematic .
The last type of methods, i.e. *image based machine learning*, have been investigated to overcome these drawbacks, and their popularity has grown in the recent years thanks to the Machine Learning ecosystem developments. Notably, the success of the deep convolutional neural networks (CNN) since the pioneer work at the end of the 80’s and the 90’s [@NIPS1989_293; @44e2afaa580a48bc8b13633b22ff10b4; @cccfa4f7238441b0a9021bb9f917e8ed] has dramatically changed the paradigm in many fields among the computer vision (image and video classification, segmentation), the speech recognition, the natural language treatment [@LecBen15]. In brief, a CNN is a feed-forward neural network with two main parts: the first one is composed of several blocks of convolutional layers followed by sub-sampling layers, while the second one is composed of fully connected (or dense) layers from which the last layer gives the desired output (e.g. redshift p.d.f). As it is common to say, during the training phase, the first part somewhat learns the optimal feature representation which replaces the manual features extraction of the previous methods, while the second part is the classifier. For photometric redshift predictions, one can read for intance the following references . Notice that in astronomy CNNs have been also used for other tasks as image processing task (e.g. [@2016arXiv161204526F]) among which the deblending (e.g. [@2019MNRAS.485.2617R; @2019MNRAS.tmp.2430B]), and also object classifications (e.g. ).
The *image based machine learning* methods for photometric redshift predictions do suffer from systematics among which some are shared with other supervised methods as the representativeness quality and size of the training samples and the quality of the input information. The later source of systematics are closely related to the images themselves which may influence the prediction: for instance the presence or not of neighbouring objects even if there do not blend the central object, the possible orientation of the object, its possible non central position, the non uniformity of the PSF, the different filter response calibrations, and so on. All these source of systematics are of course relevant, but this article is motivated by the existence of intrinsic perturbations of the images that can fool the classifier prediction, i.e the photometric redshift p.d.f as discussed in the present article. The subject known as *adversarial examples* in the ML literature has been first studied by [@Szegedy2013IntriguingPO; @2014arXiv1412.6572G], it was one of the challenges proposed during the NIPS ’17 Competition [@10.1007/978-3-319-94042-7_11; @2018arXiv180400097K] and it gave rise to communications in later NISP conferences as it is an important subject on ML robustness and security[^3].
Contrary to the expected good generalisation power of trained machine learning models with good practices and with good representativeness of the different training/testing/validation sets, the *adversarial examples* show dramatic prediction failures while the perturbations of the inputs are so tiny that one expects very marginal impacts. Unfortunately, this is not the case and it has been realised that one should envisage countermeasures. The different scenarios studied in the ML literature are: *target* or *non-target* attack which distinguishes between image perturbations that fool the model predictions either by forcing the model towards a predefined redshift value either just by forcing the model to not-predict the correct redshift; and depending on the knowledge of the architecture of the model, one qualifies the attack as *white box*-like (full knowledge) or *black box*-like (no real knowledge of the model). Put in the context of photometric redshift predictions, we think that *target* attacks are not to be envisaged in our case as if there exist input perturbations that fool the model, we do not think that they are so malicious. Concerning white versus black box scenarios, we think that white box ones are valuable as they give us some chance to propose a new training method that leads to better immunity against image perturbations. However, it will be shown a black box like effect using adversarial examples of a model given as inputs to a different model architecture.
The present article is organised as followed. Section \[sec:data-cnn\] summarises the data and CNN models used. Essentially, we are using the *Inception* CNN proposed in reference which has shown good redshift prediction performances and serves as reference for our studies. We have also chosen this CNN model because the present article is rather a proof of concept which needs a CNN of known architecture details with good performances, well trained in the classic sense and the possibility to use the same set of input data. Section \[sec:adv-gene\] introduces the adversarial image generation and shows their dramatic effects, and explains what are the effects of these samples telling us about? Then, in Section \[sec:adv-training\] a new training algorithm is detailed as well as the performances of *Inception* retrained model depending on the adversarial generator even if the outcomes are not perfect. Finally, in Section \[sec:conclusion\] we summarize our main results.
Data and CNN model {#sec:data-cnn}
==================
The numerical experiments undertaken in this article rely mostly on the model and data used in and we only briefly introduce these key ingredients. The data are originated from the SDSS multi-band imaging and spectroscopic redshift survey using a dedicated 2.5-meter telescope at Apache Point Observatory in New Mexico. The authors used the data release 12 (DR12; [@2015ApJS..219...12A]) from which has been extracted the sources classified as galaxy with spectroscopic redshifts mostly in the $[0, 0.3]$ range. The ’corrected frames’ image of the SDSS *ugriz* filters of data release 8 have been treated by the authors using the `SWARP` code[^4] [@2002ASPC..281..228B] to resample to a common pixelgrid and stack all the available image data. It results of a $5\times 64\times 64$ pixel cube centred on the galaxy coordinates ([Fig. \[fig-imgs-sdss\]]{}). The pixel values are float values possibly negative, this is in contrast to usual RGB images used in general by ML designers (eg. CIFAR images for data challenges [^5]). An example is shown in [Fig. \[fig-pixel-val\]]{}. Besides the images, the $E(B-V)$ (’ebv’ variable) redenning correction in magnitudes at the position of each object is also used as extra variable given to the fully connected part of the network.
![Examples of the $5\times 64\times 64$ pixel cube as input of the CNN. For each imaged galaxy we show the *ugriz* filter responses and on top the $z_{spec}$ value as well as the $E(B-V)$ redenning correction. Each series of filter frames has its own LUT shown on the right side.[]{data-label="fig-imgs-sdss"}](img5.png "fig:"){width="\columnwidth"}\
![Examples of the $5\times 64\times 64$ pixel cube as input of the CNN. For each imaged galaxy we show the *ugriz* filter responses and on top the $z_{spec}$ value as well as the $E(B-V)$ redenning correction. Each series of filter frames has its own LUT shown on the right side.[]{data-label="fig-imgs-sdss"}](img2.png "fig:"){width="\columnwidth"}\
![Examples of the $5\times 64\times 64$ pixel cube as input of the CNN. For each imaged galaxy we show the *ugriz* filter responses and on top the $z_{spec}$ value as well as the $E(B-V)$ redenning correction. Each series of filter frames has its own LUT shown on the right side.[]{data-label="fig-imgs-sdss"}](img3.png "fig:"){width="\columnwidth"}
![Example of the distribution of the pixel values of the five filter patches of a single image with in orange the values of the central galaxy pixels located at $z=0.27$.[]{data-label="fig-pixel-val"}](pixel-value.png "fig:"){width="0.75\columnwidth"}\
Concerning the CNN architecture, the authors have used an *Inception* model inspired from GoogleLeNet [@2014arXiv1409.4842S]. Originally, GoogleLeNet has been used for the ImageNet Large-Scale Visual Recognition Challenge 2014 and the architecture was motivated by mainly the following arguments: first, to deal with size and location variations of the information in the input image, one needs small kernel size for localized information and larger kernel for more distributed information; second, very deep networks are more prone to overfitting, so one needs to keep the network as short as possible; third, stacking a lot of convolution layers tends to be computationally intensive, this also tends to keep the network as short as possible. These arguments remains true concerning the information variations, and also if the hardware and algorithmic codes have been improved since 2014 and allows very deep architecture as the Resnet series [@2015arXiv151203385H], for this study keeping architecture simple is suited to feed a single GPU and keep the training phase in reasonable time-scale (i.e. not larger than 48 hours). But, anticipating to the result of the Section \[sec:adv-training\], one has to keep in mind that the classification power of the network also has to be taken into account to design the architecture.
So, the model of is composed of a $5\times 5$ kernel convolutional layer followed by a series of 5 inception cells with $1\times 1$, $3\times 3$ and $5\times 5$ kernels, and finally a pair of fully connected layers to transform the learned features into a prediction. Instead of using a max pooling layer after a convolution due to possible large single pixel intensity variations, one prefers to use average pooling. The activation functions for the fully connected layers are ReLU, while for the convolutional layer one applies a PReLU activation with a learnable parameter. Some details upon the size of the different layers are given in Appendix \[app-inception\]. In Section \[sec:adv-training\], we examine some modifications of this model. The inputs are typically batches of 64 samples[^6] and the output is a vector of 180 bins to sample photometric redshift p.d.f prediction taking into account that $z_{spec} \in [0, 1]$ (i.e. the bin width is $5.5\times 10^{-3}$ small enougth to sample the photometric redshift p.d.f[^7]).
For comparison purpose, a simpler CNN has been set-up consisting of 1 convolutional $5\times 5$ kernel layer followed by 2 convolutional $3\times 3$ kernel layers and 2 fully connected layers. Details can be seen on Appendix \[app-inception\]. Notice that the ’ebv’ variable is also used at the entrance of fully connected part of the network. This simpler CNN model gives satisfactory results compared to the *Inception* model, but it was not intended to outperform the later.
As in the following sections we need to train the different models in different configurations, so we could not use the pre-trained *Inception* model available on github[^8], and we have used the latest `pyTorch` library[^9] v1.3.0 [@2019arXiv191010775O] from which we have changed the default weight initialisations into the Xavier’s uniform method and the bias are initialised to 0.1, to follow initialisation conditions used by .
For the minimisation process we use the cross-entropy loss between the CNN output (i.e. the 180 bins vector) and the spectroscopic redshift value converted to a hot vector of same size[^10]. Concerning the optimiser, we have performed numerical studies with the traditional SGD [@Sutskever:2013:IIM:3042817.3043064] and Adam codes [@2014arXiv1412.6980K], as well as the recent Adam improvement code [@2017arXiv171105101L] called AdamW which is available in `pyTorch` since version 1.2.0. Notice that both for Adam and AdamW, we have set the $\epsilon$-parameter to $0.1-1.0$ values much higher than the default value (i.e. $10^{-8}$), first because it is recommended for Inception architecture in the `TensorFlow` library, and second by our own experience below $0.01$ the training manifests large instabilities. The initial learning rate is typically set to $0.01$, and we have used a reduce learning rate scheduler to decrease by 20% the learning rate if after 5 or 10 epochs the test accuracy has not been improved.
From the initial $659\, 857$ input samples, after random shuffling, we have set-up different sets of $100\, 000$ samples, one common for the test set and several for different training sets, and also a different set of $10\, 000$ samples for the adversarial test experiments. After been gathered in batches, we randomly apply a combination of geometrical transformations such as horizontal and vertical flips and $90\degr$, $180\degr$ and $270\degr$ rotations. The number of training epochs is typically 150, but first sometimes the minimization has required early stopping and relaunching with a different optimiser settings, and second we have pushed up to 300 epochs for some special training tests. All the numerical experiments have been conducted at the CC-IN2P3[^11] GPU-Nvidia farm from which a single Tesla V100 is used at a time.
Adversarial images generation {#sec:adv-gene}
=============================
Classical training of *Inception* model
---------------------------------------
First of all, an *Inception* model, nickname hereafter *Iref* has been trained “classically”, that is to say using a pair of train/test sets and tuning the optimizer hyper-parameters to get the training and testing losses as close as possible up to an epoch where the network starts to overfit, that is to say when the test loss increases while the train loss still decreases. The results are presented in terms of the distribution of the variable $\delta z = (z_{phot}-z_{spec})/(1+z_{spec})$ ([Fig. \[fig-dz-ref\]]{} top) and the $z_{phot}$ versus $z_{spec}$ comparison ([Fig. \[fig-dz-ref\]]{} bottom). For the sake of comparison between different numerical experiments, we have used the following variables:
- the bias defined as the mean of the $\delta z$ distribution;
- the $\sigma_\mathrm{MAD} = 1.4826 \times |\delta z - \mathrm{Median}(\delta z)|$;
- and the fraction $\eta$ of outliers such that $|\delta z| > 0.05$.
![Results obtained with the the reference *Inception* model trained and tested with unperturbed images: (top) distribution of the $\delta z$ variable; (bottom) $z_{phot}$ versus $z_{spec}$.[]{data-label="fig-dz-ref"}](dz-reference.png "fig:"){width="0.8\columnwidth"}\
![Results obtained with the the reference *Inception* model trained and tested with unperturbed images: (top) distribution of the $\delta z$ variable; (bottom) $z_{phot}$ versus $z_{spec}$.[]{data-label="fig-dz-ref"}](zspecVszphot-reference.png "fig:"){width="0.8\columnwidth"}
The results concerning *Iref* are presented on [Tab. \[tab-non-robust\]]{} and for the “classical” training with non perturbed images (i.e. non adversarial images) the reader should have a look at the line ’non perturbed’. The obtained results are satisfactory enough and comparable with the ones obtained in , so it can serve as a rather good reference starting point.
Adversarial samples
-------------------
Let us now get a brief introduction to adversarial samples generation. In a classical supervised training, the parameters $\theta$ of a model $f$ are adjusted to minimise an empirical risk, thanks to the set of $\{x_i,z_i\}_{i\leq N}\in D_{train}$ samples, where $x_i$ is an input image[^12] and $z_i$ is the spectroscopic redshift considered as the true value. The expression of the empirical risk is $$\begin{aligned}
R_e(f_\theta,D_{train}) &= \frac{1}{|D_{train}|} \sum_{(x,z)\sim D_{train}}\ell(f_\theta(x),z) \nonumber \\
&= \frac{1}{N}\sum_{i=1}^N \ell(f_\theta(x_i),z_i).
\label{eq-normal-train}\end{aligned}$$ with $\ell$ the cross-entropy loss function. Depending on the optimizer (SGD, Adam, and so on), the algorithm is different but for the sake of definiteness and illustration, we use here a Batched Gradient Descent method where $(x,z)$ samples are gathered on batches named $B_{train}$. After an initialisation phase, the values of the $\theta$ parameters are updated step-by-step as: $$\theta_{t+1} = \theta_t - \alpha\ \frac{1}{|B_{train}|}\sum_{(x,z)\sim B_{train}} \left[ \nabla_\theta \ell(f_\theta(x),z)\right]_{\theta=\theta_t}
\label{eq-normal-gd}$$ with $\alpha$ the *learning rate*. Now, the adversarial samples are images to which is added a small amount, noted $\delta$ according to $$\begin{gathered}
R_{adv,e}(f_\theta, D_{train})=\\ \frac{1}{|D_{train}|}\sum_{(x,z)\sim D_{train}} \left[\underset{\|\delta \|\leq \varepsilon}{\mathrm{max}}\ \ell(f_\theta(x+\delta ),z)\right].
\label{eq-Radv-e}\end{gathered}$$ The minimisation of $R_{adv,e}$ to get the $\theta$ parameters needs to solve a *minimax problem* also known as a *saddle point problem*, with an inner maximization part included in the more familiar outer minimisation part.
The maximisation problem leads to several kinds of methods that generate adversarial examples. Such kind of generating method is called an *attack* in the literature and we occasionally use this terminology later on. The first method, called the Fast Sign Gradient Method (FSGM), has been proposed in [@2014arXiv1412.6572G] and more elaborated attacks are detailed for instance in the following references [@2016arXiv161101236K; @2017arXiv171207107Y; @2017arXiv170606083M; @2018arXiv180400097K; @10.1007/978-3-319-94042-7_11; @2019arXiv190502175I]. In the following, we briefly summarize the FSGM attack which is simple and powerful, as well as one of its generalisation, i.e. the Projected Gradient Descent (PGD)[^13] iterative method which is more aggressive.
The FSGM one-step generator
---------------------------
The solution of the maximisation problem (Eq. \[eq-Radv-e\]) is noted $\delta^\ast$, it generally depends on $x$ and satisfies: $$\delta^\ast(x) = \underset{\| \delta\| \leq \varepsilon }{\mathrm{argmax}}\ \ell(f_\theta(x+\delta ),z).
\label{eq-deltaast-general}$$ The maximisation ensures that perturbations smaller than $\delta^\ast$ in terms of the norm used, have smaller impact on the model predictions. In the following, the infinity norm (L$_\infty$) is used as it gives good results, but for the sake of completeness a numerical experiment has been conducted with the L$_2$ norm[^14] giving more sparse perturbations[^15]. In case of the infinity norm $\| \delta \|_\infty\leq \varepsilon$ and under some hypothesis mentioned later, the solution reads: $$\delta^\ast(x) = \varepsilon\times \mathrm{sign}\left( \nabla_\delta \ell(f_\theta(x+\delta),z) \right)
\label{eq-deltaast-linear}$$ from which the FSGM name is originated, i.e. Fast Sign Gradient Method.
[lccc]{}
Models (images) & bias ($\times 10^{-4}$) & $\sigma_\mathrm{mad}$ ($\times 10^{-3}$) & $\eta$ (%)\
*Iref* (non perturbed) & $0.3$ & $11$ & $1$\
\
*Iref* & – & $66$ & $42$\
*I0*-*I3* (*Iref* adv) & – & $[63,68]$ & $[40,43]$\
CNN (*Iref* adv) & – & $76$ & $49$\
\
*Iref* & – & $82$ & $59$\
*I0*-*I3* (*Iref* adv) & – & $[76,78]$ & $[52,55]$\
\
*Iref* & – & $79$ & $57$\
*I0*-*I3* (*Iref* adv) & – & $[77,79]$ & $[53,55]$\
To see the impact of such perturbation added to image $x$, we have first conducted a numerical study to determined $\varepsilon$. When adding a random uniform noise to an original image $x$, i.e. $\delta(x) \sim \mathcal{U}[-\epsilon, +\epsilon]$, if $\varepsilon = 0.01$ there is absolutely no effect on the generalisation power of *Iref* (i.e. the bias, $\sigma_{MAD}$ and $\eta$ variables remain unchanged), while if $\varepsilon = 0.1$ then the $\sigma_{MAD}$ is increased by 50% and the number of outliers is multiplied by 3. So, we have fixed $\varepsilon = 0.01$ to perform an FSGM attack on *Iref*. The results of such attack are displayed both in [Fig. \[fig-dz-fgsm\]]{} that can be compared to [Fig. \[fig-dz-ref\]]{} and in [Tab. \[tab-non-robust\]]{} (line “$\varepsilon=10^{-2}$, Single FSGM”).
![Same kind of histograms (same binning) of [Fig. \[fig-dz-ref\]]{} but after a FSGM attack with $\varepsilon=0.01$.[]{data-label="fig-dz-fgsm"}](dz-fgsm-0d01-perturbe.png "fig:"){width="0.8\columnwidth"}\
![Same kind of histograms (same binning) of [Fig. \[fig-dz-ref\]]{} but after a FSGM attack with $\varepsilon=0.01$.[]{data-label="fig-dz-fgsm"}](zspecVszphot-fgsm-0d01-perturbe.png "fig:"){width="0.7\columnwidth"}
On a single image, [Fig. \[fig-advex-fgsm\]]{} shows the kind of perturbation generated by such attack.
![Example of the impact of a FGSM attack ($\varepsilon = 10^{-2}$). From top to bottom: on the 1st line are displayed the 5 filter frames of an non perturbed original/reference image ($x$); on the 2nd line the corresponding frames of the perturbation $\delta^\ast(x)$ magnified by a factor 100 to be visible; on the 3rd line for the sake of completeness are shown the results $x+\delta^\ast(x)$; on the 4th line are shown the horizontal slices at the vertical middles of the 5 frames from $x$ in blue and $x+\delta^\ast(x)$ in red, notice in the left most panel that the perturbation as expected is far smaller than the background level; on the last line are shown in blue and red curves the predicted redshift p.d.f for the non perturbed and the perturbed image as well as the blue and red vertical dashed lines corresponding to the $z_{phot}$ estimations as weighted means of the bin centre locations, and the black vertical dashed line corresponding to the $z_{spec}$ “true” redshift value. The color scale is common to all the filter frames.[]{data-label="fig-advex-fgsm"}](advex-fgsm-0d01.png){width="\columnwidth"}
As one clearly can appreciate, the very small perturbation generated by the attack has dramatic impacts on the photo-z predictions. But more than that, in [Tab. \[tab-non-robust\]]{}, one can see that the adversarial sample of the reference *Inception* model impact not only other *Inception* models trained with different seeds as well as different training sets, noted *I0*-*I3*, but also the different CNN model. Some remarks can be drawn from this attack experiment: first, adversarial samples of one model also affect other models which implies by consequence that one cannot use several models to combine their results to cancel this underlying systematic error; second, the adversarial samples of one type of model architecture also affect other model architectures. The later remark has been identified in reference [@Szegedy2013IntriguingPO] and allows one for some applications to put a corner in the defence of a ML model (i.e. *black box* attack). So, all this tends to suggest that adversarial samples are quite special samples as they clearly break the traditional estimation of the generalisation power of a model using test and validation samples.
The first reaction which was probably a common sense when these adversarial samples were discovered in the ML world, relied on the fact that the FSGM generated perturbations are so special that the probability of occurrence should be close to 0. But, this is not so simple even if we exclude malicious intent in this work. First, the level of the FSGM perturbation is so tiny that one cannot totally exclude that such perturbation could not be made during the processing of the different codes involved in a image treatment pipeline (i.e. CCD pixel correction, deflating, de-biasing, calibration, stacking, resampling, cosmic tracks removal and so on[^16]) even if these codes are very sophisticated, especially if the experimental set-up evolves in time or if the observation conditions are not 100% of perfect photometric nights. Although, one can argue that these artefacts can be well mastered thanks to the long experience accumulate in image treatments. Second, intrinsically these adversarial samples reveal a weakness of the training and also probably of the model design. However, in the Section \[sec:adv-training\] we show how one can take advantage of the adversarial generation to perform a new training.
The PGD multi-steps generator
-----------------------------
Before exploring the adversarial samples benefit, let us explore an other attack method which generalises in a simple manner the FSGM one. In fact, the $\delta^\ast$ expression [Eq. \[eq-deltaast-linear\]]{} is only exact in the linear case where $f_{\theta=(w,b)}(x) = w^T x + b$ [@2014arXiv1412.6572G], even if this solution is nicely very effective as we have experienced. In the non-linear general case, one needs to use an algorithm to get an approximated solution of [Eq. \[eq-deltaast-general\]]{}. To do so, one can use a hill climbing algorithm. After an initialisation phase $\delta=\delta_0$, one performs a certain number of iterations ($n_{iter}$) to update $\delta$ as followed: $$\delta \leftarrow \delta + \underset{\|u\|\leq \alpha}{\mathrm{argmax}} \left[ u^T. \nabla_\delta \ell(f_\theta(x+\delta),z)\right]$$ which in case of the infinity norm leads to $$\delta \leftarrow \mathcal{P}^\infty_\varepsilon\left[\delta + \alpha\ \mathrm{sign}(\nabla_\delta \ell(f_\theta(x+\delta),z))\right]
\label{eq-delta-iter-linfty}$$ with $\alpha$ a *learning rate* different from the one used for the $\theta$ minimisation part and with $ \mathcal{P}^\infty_\varepsilon$ the projection on the $\varepsilon$-L$_\infty$ ball such that $\| \delta \|_\infty \leq \varepsilon$. As one recognises, in some sense, the method applies iteratively an FSGM attack with a small step size $\alpha$ decoupled from the $\varepsilon$ value. In the literature, this method is part of a series known as Projected Descend Gradient attacks (PGD). Notice that a refinement of the above method consists of randomly initialise $\delta_0$ in the $\varepsilon$-L$_\infty$ ball, however it has shown very little impact in our numerical experimental cases. There exists other methods to solve [Eq. \[eq-deltaast-general\]]{}, one is using a neural network which learns how to generate adversarial samples [@Jang_2019_ICCV].
Considering $n_{iter} = 10$, [Tab. \[tab-non-robust\]]{} presents the results of the PGD attacks conditioned by $\alpha=\varepsilon/10$ both on the reference *Inception* model as well as its four twins *I0-I3*. See [Fig. \[fig-advex-pgd-linf\]]{} to have an idea of the perturbation pattern (i.e. $\delta$) generated by the PGD generator method.
![Pattern of a PGD adversarial iterative attack with $\varepsilon=10^{-2}$ operating on the same original image of [Fig. \[fig-advex-fgsm\]]{} (with the same LUT). By comparison, one can see that such iterative method leads to smoother pattern.[]{data-label="fig-advex-pgd-linf"}](advex-pattern-pgd_ninf_norand-0d01.png){width="0.8\columnwidth"}
As anticipated, greater is the number of iterations, greater is the negative impact on the photo-z prediction. Although the effect is not proportional to the number of iterations as using $n_{iter}=20$ yields to an increase by 10% of the $\sigma_{MAD}$ value and the number of outliers. One notices that the FSGM method, in a single step, already points towards the weakness of the network and the training.
Before diving into the robust training algorithm, for the sake of completeness, let us investigate what happens if we use the L$_2$ norm instead of the infinity norm. The $\delta^\ast$ solution [Eq. \[eq-delta-iter-linfty\]]{} is replaced by: $$\delta \leftarrow \mathcal{P}^2_\varepsilon\left[\delta + \alpha\ \frac{\nabla_\delta \ell(f_\theta(x+\delta),z)}{\| \ell(f_\theta(x+\delta),z)\|_2}
\right]
\label{eq-delta-iter-l2}$$ where the projection $\mathcal{P}^2_\varepsilon$ on the $\varepsilon$-L$_2$ ball is defined according to $$\mathcal{P}^2_\varepsilon(x) = \varepsilon\ \frac{x}{\mathrm{max}(\varepsilon, \|x\|_2)}.
\label{eq-P2-projection}$$ The results on the reference *Inception* and its four twins are written on the last lines of [Tab. \[tab-non-robust\]]{}. Notice that to conduct such numerical experiment with L$_2$ norm in a comparable manner to the infinity norm, one needs to scale $\varepsilon$ according to the change of $\varepsilon$-ball volume between the two norms, and to get an similar impact on the photo-z prediction, we were forced to push the number of iterations up to $n_{iter}=40$. At this price, L$_2$ norm can also clearly lead to negative impact on the photo-z predictions and the pattern of the perturbation is little sparser than with the infinity norm. But, this norm is significantly more time consuming with a not so different behaviour compared to the FSGM or the PGD methods. So, it will not be used hereafter in the rest of this article.
Adversarial training {#sec:adv-training}
====================
The algorithm
-------------
In the previous section, we have shown that a simple adversarial sample generator, i.e. the FSGM attack with very small $\varepsilon$ value far below the image background, can lead to dramatic impact on the photo-z predictions, even if the *Inception* model has been correctly trained, in a classical sense. At this stage, one may just forget this adversarial samples advocating a very unlucky probability to occur. But as we have argued, one may have a more proactive attitude taking as fact that the model training as a certain amount of weakness due to our inability to understand the exact essence of the features that CNNs are really capturing [@2019arXiv190502175I]. Even if in photo-z prediction domain, the security is not engaged as it can be the case in cryptography or autonomous vehicle for instance, we envisage the second attitude, especially because we can strengthen the training.
Choose adversarial samples fraction and the attack generator (FSGM/PGD, $\varepsilon$, $\alpha$, number of iterations) Do $\theta$ (model weights) initialisation $g \leftarrow 0$ , according to initial generator choice, find $\delta^\ast$: $\delta^\ast \leftarrow \varepsilon\ \mathrm{sign}\left( \nabla_\delta \ell(f_\theta(x+\delta),z) \right)]$ $\delta \leftarrow \mathcal{P}^\infty_\varepsilon\left[\delta + \alpha\ \mathrm{sign}(\nabla_\delta \ell(f_\theta(x+\delta),z))\right]$ $\delta^\ast \leftarrow 0$ $g \leftarrow g + \nabla_\theta \ell(f_\theta(x+\delta^\ast),z)$ $\theta \leftarrow \theta - \alpha \frac{g}{|B_{train}|} $
Then, let us elaborate a new training algorithm. In fact to minimize the empirical adversarial risk ([Eq. \[eq-Radv-e\]]{}), one can use a gradient descent method similar to [Eq. \[eq-normal-gd\]]{} but adapted to the adversarial empirical risk. This would lead to the following natural modified version: $$\begin{aligned}
\theta_{t+1} &= \theta_t \nonumber \\
&- \alpha\ \frac{1}{|B_{train}|}\sum_{(x,z)\sim B_{train}} \nabla_\theta\left[\underset{\| \delta\|\leq \varepsilon}{\mathrm{max}}\ \ell(f_\theta(x+\delta ),z)\right]_{\theta=\theta_t}.\end{aligned}$$ However, there are two remarks that can be raised: the first one concerns the method to compute the gradient regarding the maximisation inner problem, the other one pinpoints the fact that only adversarial samples are considered for the loss computation. Concerning the gradient computation, the difficulty is bypassed as followed: in a convex problem under some mathematical hypothesis, one can use the J. Danksin’s theorem [@danskin1966theory] which yields the simple result: $$\nabla_\theta\left[\underset{\| \delta\|\leq \varepsilon}{\mathrm{max}}\ \ell(f_\theta(x+\delta ),z)\right] =
\nabla_\theta\ \ell(f_\theta(x+\delta^\ast),z)$$ with $\delta^\ast$ the solution of the maximisation problem[^17], but it is also used even if the convexity is not guaranteed, as experimentally it gives satisfactory numerical results.The second point has been discussed in [@2016arXiv161101236K], and it results that a mixture of unperturbed samples and adversarial samples is more effective. The introduction of adversarial samples acts like a regularisation term [@2018arXiv181000953F; @2018arXiv181000363B].
So, with the tools developped in the previous section, we can set-up a new training algorithm (alg. \[algo-1\]) where after choosing the fraction of adversarial samples that populate each training mini-batch, one computes $\delta^\ast$ either by using FSGM one-step method or by using the PGD iterative method. Then, one updates the gradients involved in the gradient descent step to find the loss minimum with respect to the model weights ($\theta$). This schema is rather simple and for instance the last step (12) is usually taken in charge by more sophisticated algorithms already used in classical training, i.e. SGD, Adam, and so on, with for instance momentum memory and weight decay. Moreover, the attack generators listed are only those presented in the previous section, and for other applications they can be modified on purpose.
fraction of adv.
------------------ -------- -------- ------ ------ ----- ------
$0\%$ $-0.3$ $-105$ $11$ $66$ $1$ $42$
$5\%$ $-20$ $-40$ $11$ $9$ $1$ $4$
$10\%$ $40$ $-25$ $11$ $8$ $1$ $2$
$20\%$ $-6$ $23$ $11$ $8$ $1$ $1$
: Results after an adversarial training (alg. \[algo-1\]) of an *Inception* model using 100k images (non perturbed) among which a certain fraction ($f_a$) are perturbed using a FSGM method with $\varepsilon = 10^{-2}$. For each $f_a$ value, are presented the values of the bias, the $\sigma_\mathrm{MAD}$ and $\eta$ as for [Tab. \[tab-non-robust\]]{} for the cases of ’non perturbed/perturbed’ images obtained with the same attack that has been used for the training. For the special case where $f_a=0\%$, are reminded the values presented in [Tab. \[tab-non-robust\]]{} to ease the comparison and the bias value obtained with perturbed images are noted for completeness.[]{data-label="tab-robust"}
Training with FSGM adversarial samples
--------------------------------------
The results obtained after an adversarial training of an *Inception* model using (step (7) of alg. \[algo-1\]) a FSGM attack generator with $\varepsilon=10^{-2}$ are presented in [Tab. \[tab-robust\]]{}. We have varied the fraction ($f_a$) of adversarial samples (i.e. perturbed images) among each batch training sets. It is quite striking that a rather small fraction of adversarial samples as $5\%$, already decreases the sensibility of the model to adversarial samples. Increasing the fraction $f_a$ leads to better results with similar performances for non perturbed or perturbed images. We note however a worse bias compared to a classical training using non perturbed images. Investigating different optimiser codes and tweaking their own settings does not change drastically this behaviour. In passing, we notice that if the absolute values of the variables, notably the bias, are varying with the optimiser settings, the relative behaviour after training between non perturbed and perturbed images remains unchanged.
The PGD adversarial samples
---------------------------
The results obtained in the previous section using the FSGM adversarial sample generator seem encouraging, but the *Inception* model trained with $f_a=20\%$ or greater value of FSGM adversarial samples has no increase of robustness against the PGD attack using $\varepsilon = 10^{-2}$ and $n_{iter}=10$. This weakness is even still present when we use during the training a fraction of adversarial samples generated by the PGD iterative method. This problem has been identified in reference [@2016arXiv161101236K] and an explanation has been elaborated by the authors of [@2017arXiv170606083M]. In summary, they show that around a given sample $x$, the decision boundary should be modified in a complex manner in such a way that the perturbed samples $(x+\delta^\ast)$ are correctly classified. This is a sort of local underfitting problem. To do so, obtaining a stronger robustness against PGD iterative perturbations requires at least a model with a larger capacity.
Considering the *Inception* architecture ([Tab. \[tab-inception-model\]]{}), to get a larger capacity is not just a matter of increasing the number of model parameters (i.e. the length of the $\theta$ vector). For instance, by increasing the output size of the first fully connected layer, that is to say the ’fc0’ layer and accordingly the input size of the ’fc1’ layer ([Tab. \[tab-inception-model\]]{}), from 1096 to 2000 neurons, leads to a new CNN network with $48\ 722\ 852$ parameters compared to the original *Inception* model with $27\ 596\ 372$ parameters. However, this larger network has no better robustness against PGD attack. It is not neither a matter of increasing the depth of the convolution part. For instance, we have double the number of ’i3’ inception cells without any success of increasing the robustness of the modified model. The only way we have found to manage a better robustness is by increasing the number of features entering the classifier, that is to say the input size of the ’fc0’ layer. For instance, we have increased the output size of the ’i4.s2\_2’ layer from 128 to 256, then the ’fc0’ input length has been increased from 22273 to 38657, leading a model with $45\ 720\ 660$ parameters. Pushing further the output size of the ’i4.s2\_2’ layer does not change the conclusions.
Model bias ($\times 10^{-4}$) $\sigma_\mathrm{mad}$ ($\times 10^{-3}$) $\eta$ (%)
--------------- ------------------------- ------------------------------------------ ------------
*Iref* $0.3$/–/– $11/66/82$ $1/42/59$
*I(modified)* $-21/-32/-32$ $15/24/25$ $2/6/6$
: Results of modified version (see text) of the *Inception* baseline model ([Tab. \[tab-inception-model\]]{}) trained with 50% of PDG adversarial samples ($\varepsilon = 10^{-2}$ and $n_{iter}=10$) and 50% of non perturbed samples. To ease the comparison, we have reported from [Tab. \[tab-non-robust\]]{} the result obtained with the *Iref* model. For each variables (bias, $\sigma_\mathrm{mad}$ and $\eta$) are given the results considering non perturbed samples, samples generated by FSGM method ($\varepsilon = 10^{-2}$) and samples generated by the PGD method used for training.
\[tab-PGD-training\]
The results are shown in [Tab. \[tab-PGD-training\]]{} with $f_a=50\%$. Clearly we have gained in robustness with comparable resistance against FSGM and PGD adversarial sample generations. But, in the same times we have degraded the results obtained for normal samples which is not satisfactory. Actually, using $f_a\leq10\%$ leads to results very similar the case of *Iref* and increasing $f_a$ tends to increase the robustness against adversarial PGD attack and simultaneously worsen the results for normal samples. Notice that training with the PGD generator $n_{iter}$ parameter set to 30 do not change the results. The problem is probably due to the architecture of the *Inception* model. Although, testing over architecture is beyond the purpose of this article and is postponed to further studies. As suggested by reference [@2018arXiv181000144Y], the inspection of the loss landscape in the input space rather than in the parameters space could be a diagnose of the susceptibility of a model to adversarial samples and a guideline to propose an alternative.
Summary {#sec:conclusion}
=======
In this article we have conducted some numerical experiments with a specific convolutional neural network (i.e. the *Inception* model) specially trained to predict galaxy redshift from datacubes composed of SDSS multi-band photometric imaging. We have shown that despite a good performance using “classical” training on normal images (i.e. non perturbed images), the model has a very poor generalisation power when considering very tiny, although special, perturbations of these images far below the background noise. This is the quintessence of adversarial images to find the best perturbation that fool the network prediction. These perturbed images reveal weaknesses that we think one should take care about, especially to study the robustness of the model. Notice that we would have drawn the same conclusions with an other CNN network, this is not specific to the *Inception* model.
Following ML developments on this subject, we have presented a simple algorithm for adversarial training (alg. \[algo-1\]) that injects during the training phase a fraction of adversarial images. This tends to strengthen the model against those adversarial images without degrading to much the results on normal images. This is especially efficient considering Fast Sign Gradient Method to generate adversarial images and we have shown that the present *Inception* model can be well retrained. But, still the model does not master more aggressive adversarial generators such as the Projected Gradient Descent iterative methods. We have tried different modifications of the model to enlarge its capacity and found that increasing the number of features entering the classifier part of the network is the only way to get a better robustness to PGD generated samples, but it is at the price of the degradation of the redshift predictions with non perturbed images which is not satisfactory. This might be the sign that the *Inception* model architecture is not well adapted to fight against such attack. However, we note that the adversarial training alg. \[algo-1\] may also be improved to perform better inner maximisation, for instance using a network specially trained to generate adversarial samples [@Jang_2019_ICCV]. This is postponed to further studies.
As final remarks, we want to point that the sensitivity to adversarial samples is neither restricted to photometric redshift predictions and the other usages of neural networks in the different fields of image based analysis should be addressed, nor a specificity of convolutional neural networks as tree-based models like Decision Trees, Gradient Boosted DT, Random Forests may also be affected as described in reference [@2019arXiv190210660C].
Acknowledgements {#acknowledgements .unnumbered}
================
The author wants to warmly thank J. Pasquet-Itam and E. Bertin for the fruitful discussions and the access to the SDSS data set, as well as the details of the *Inception* model design they have provided to me.
*Inception* and CNN architectures {#app-inception}
=================================
In this appendix, we give a more advanced description of the different layers of the two models used in this article: the *Inception* CNN used by ([Tab. \[tab-inception-model\]]{}), and a simpler CNN architecture ([Tab. \[tab-cnn-model\]]{}).
[[lllrr]{}]{} name & input\_shape & output\_shape & nb\_params & Comments\
conv0 & (5, 64, 64) & (64, 64, 64) & 8128 & kernel=(5, 5), padding=(2, 2)\
pool0 & (64, 64, 64) & (64, 32, 32) & 0 & average\
i0.s1\_0 & (64, 32, 32) & (48, 32, 32) & 3168 & kernel=(1, 1)\
i0.s2\_0 & (48, 32, 32) & (64, 32, 32) & 27776 & kernel=(3, 3)\
i0.s1\_2 & (64, 32, 32) & (48, 32, 32) & 3168 & kernel=(1, 1)\
i0.pool0 & (48, 33, 33) & (48, 32, 32) & 0 &\
i0.s1\_1 & (64, 32, 32) & (48, 32, 32) & 3168 & kernel=(1, 1)\
i0.s2\_1 & (48, 32, 32) & (64, 32, 32) & 76928 & kernel=(5, 5), padding=(2, 2)\
i0.s2\_2 & (64, 32, 32) & (64, 32, 32) & 4224 & kernel=(1, 1)\
i0 & (64, 32, 32) & (240, 32, 32) & 118432 & concat i0.(s2\_2, s2\_1, s2\_0, pool0)\
i1.s1\_0 & (240, 32, 32) & (64, 32, 32) & 15488 &\
i1.s2\_0 & (64, 32, 32) & (92, 32, 32) & 53176 &\
i1.s1\_2 & (240, 32, 32) & (64, 32, 32) & 15488 &\
i1.pool0 & (64, 33, 33) & (64, 32, 32) & 0 &\
i1.s1\_1 & (240, 32, 32) & (64, 32, 32) & 15488 &\
i1.s2\_1 & (64, 32, 32) & (92, 32, 32) & 147384 &\
i1.s2\_2 & (240, 32, 32) & (92, 32, 32) & 22264 &\
i1 & (240, 32, 32) & (340, 32, 32) & 269288 & concat i1.(s2\_2, s2\_1, s2\_0, pool0)\
pool0 & (340, 32, 32) & (340, 16, 16) & 0 &\
i2.s1\_0 & (340, 16, 16) & (92, 16, 16) & 31464 &\
i2.s2\_0 & (92, 16, 16) & (128, 16, 16) & 106240 &\
i2.s1\_2 & (340, 16, 16) & (92, 16, 16) & 31464 &\
i2.pool0 & (92, 17, 17) & (92, 16, 16) & 0 &\
i2.s1\_1 & (340, 16, 16) & (92, 16, 16) & 31464 &\
i2.s2\_1 & (92, 16, 16) & (128, 16, 16) & 294656 &\
i2.s2\_2 & (340, 16, 16) & (128, 16, 16) & 43776 &\
i2 & (340, 16, 16) & (476, 16, 16) & 539064 & concat i2.(s2\_2, s2\_1, s2\_0, pool0)\
i3.s1\_0 & (476, 16, 16) & (92, 16, 16) & 43976 &\
i3.s2\_0 & (92, 16, 16) & (128, 16, 16) & 106240 &\
i3.s1\_2 & (476, 16, 16) & (92, 16, 16) & 43976 &\
i3.pool0 & (92, 17, 17) & (92, 16, 16) & 0 &\
i3.s1\_1 & (476, 16, 16) & (92, 16, 16) & 43976 &\
i3.s2\_1 & (92, 16, 16) & (128, 16, 16) & 294656 &\
i3.s2\_2 & (476, 16, 16) & (128, 16, 16) & 61184 &\
i3 & (476, 16, 16) & (476, 16, 16) & 594008 & concat i3.(s2\_2, s2\_1, s2\_0, pool0)\
pool0 & (476, 16, 16) & (476, 8, 8) & 0 &\
i4.s1\_0 & (476, 8, 8) & (92, 8, 8) & 43976 &\
i4.s2\_0 & (92, 8, 8) & (128, 8, 8) & 106240 &\
i4.s1\_2 & (476, 8, 8) & (92, 8, 8) & 43976 &\
i4.pool0 & (92, 9, 9) & (92, 8, 8) & 0 &\
i4.s2\_2 & (476, 8, 8) & (128, 8, 8) & 61184 &\
i4 & (476, 8, 8) & (348, 8, 8) & 255376 & concat i4.(s2\_2, s2\_0, pool0)\
fc0 & (22273) & (1096) & 24412304 & concat (i0,i1,i2,i3,i4, ’ebv’)\
fc1 & (1096) & (1096) & 1202312 &\
fc2 & (1096) & (180) & 197460 &\
name input\_shape output\_shape nb\_params Comments
------- --------------- --------------- ------------ -------------------------------
conv0 (5, 64, 64) (64, 64, 64) 8064 kernel=(5, 5), padding=(2, 2)
pool0 (64, 64, 64) (64, 32, 32) 0 average
conv1 (64, 32, 32) (92, 34, 34) 53084 kernel=(3, 3), padding=(2, 2)
pool1 (92, 34, 34) (92, 17, 17) 0 average
conv2 (92, 17, 17) (128, 19, 19) 106112 kernel=(3, 3), padding=(2, 2)
pool2 (128, 19, 19) (128, 10, 10) 0 average
fc0 (12801) (1024) 13109248 concat (pool2, ’ebv’)
fc1 (1024) (180) 184500
: *CNN* scheme with a total number of trainable parameters $13\ 461\ 008$. Kernel sizes of convolution layers are detailed as well as padding if not 1x1. ’pool0’ are average pooling layers with possibly associated padding layers. The last convolutional-polling associated layers (’pool2’)’ output is concatenated with the ’ebv’ information added to feed the first fully connected layer. The tensor conventions follow those describes in [Tab. \[tab-inception-model\]]{}.[]{data-label="tab-cnn-model"}
\[lastpage\]
[^1]: Email: [email protected]
[^2]: The galaxies are not the only objects that will be catalogued and the conclusion to our study can nicely be generalized.
[^3]: See <https://nicholas.carlini.com/writing/2019/all-adversarial-example-papers.html> to get an idea of the literature growth about this subject.
[^4]: <http://astromatic.net/software/swarp>
[^5]: https://www.cs.toronto.edu/ kriz/cifar.html
[^6]: It is implicit that a sample is composed of the collection of the *ugriz* filter datacubes associated with the ’ebv’ variables.
[^7]: use a bin width of $2.2\times 10^{-3}$ but the redshift p.d.f sampling is not affected by a larger bin width as it is shown on [Fig. \[fig-advex-fgsm\]]{}.
[^8]: <https://github.com/jpasquet/photoz>
[^9]: <https://pytorch.org/>
[^10]: That is to say a vector of length 180, with all the components set to 0, except the one corresponding to the $z_{spec}$ value which is set to 1.
[^11]: <https://cc.in2p3.fr/en/>
[^12]: Notice that the ’ebv’ variable is implicitly added but it remains unperturbed in the following numerical experiments.
[^13]: This method is also called Basic Iterative Method in the literature.
[^14]: As a remainder, consdering a vector $x=(x_1,\dots,x_n)\in\mathbb{R}^n$, $\|x\|_2 = (x_1^2\dots+x_n^2)^{1/2}$, while $\| x \|_\infty = \mathrm{sup}(|x_1|,\dots,|x_n|)$.
[^15]: Notice that models trained with adversarial samples generated by the infinity norm give better robustness [@2017arXiv170606083M].
[^16]: See for instance the different steps of the pipeline detailled in reference [@2018PASJ...70S...5B].
[^17]: Such theorem is used rigorously in reference [@2019arXiv190502175I].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This study presents a fabrication process for lithium-drifted silicon (Si(Li)) detectors that, compared to previous methods, allows for mass production at a higher yield, while providing a large sensitive area and low leakage currents at relatively high temperatures. This design, developed for the unique requirements of the General Antiparticle Spectrometer (GAPS) experiment, has an overall diameter of 10 cm, with ${\sim}$9 cm of active area segmented into 8 readout strips, and an overall thickness of 2.5 mm, with ${\gtrsim}2.2$ mm (${\gtrsim}90\%$) sensitive thickness. An energy resolution ${\lesssim}4$ keV full-width at half-maximum (FWHM) for 20$-$100 keV X-rays is required at the operating temperature ${\sim}-40^\circ$C, which is far above the liquid nitrogen temperatures conventionally used to achieve fine energy resolution. High-yield production is also required for GAPS, which consists of ${\gtrsim}1000$ detectors. Our specially-developed Si crystal and custom methods of Li evaporation, diffusion and drifting allow for a thick, large-area and uniform sensitive layer. We find that retaining a thin undrifted layer on the $p$-side of the detector drastically reduces the leakage current, which is a dominant component of the energy resolution at these temperatures. A guard-ring structure and optimal etching of the detector surface are also confirmed to suppress the leakage current. We report on the mass production of these detectors that is ongoing now, and demonstrate it is capable of delivering a high yield of ${\sim}90\%$.'
address:
- 'Institute of Space and Astronautical Science, Japan Aerospace Exploration Agency (ISAS/JAXA), Sagamihara, Kanagawa 252-5210, Japan'
- 'Sensor Device Business Unit, Device Department, Shimadzu Corporation, Atsugi, Kanagawa 243-0213, Japan'
- 'Massachusetts Institute of Technology, Cambridge, MA 02139, USA'
- 'Columbia University, New York, NY 10027, USA'
- 'Kanagawa University, Yokohama, Kanagawa 221-8686, Japan'
author:
- 'M. Kozai'
- 'H. Fuke'
- 'M. Yamada'
- 'K. Perez'
- 'T. Erjavec'
- 'C. J. Hailey'
- 'N. Madden'
- 'F. Rogers'
- 'N. Saffold'
- 'D. Seyler'
- 'Y. Shimizu'
- 'K. Tokuda'
- 'M. Xiao'
bibliography:
- 'bibfile.bib'
title: 'Developing a mass-production model of large-area Si(Li) detectors with high operating temperatures'
---
Si(Li) detector ,Large-area detector ,Leakage current ,Cosmic-rays ,Antinuclei ,Dark matter
Introduction
============
We present here a high-yield mass production process for lithium-drifted silicon (Si(Li)) detectors that meet the unique requirements of the General Antiparticle Spectrometer (GAPS) experiment. GAPS is a balloon-borne experiment that aims to survey low-energy (${<}0.25$ GeV/n) cosmic-ray antinuclei for the first time, by adopting a novel detection concept based on the physics of exotic atoms [@Aramaki16a; @Hailey09; @Hailey13; @Fuke17]. Low-energy cosmic-ray antinuclei, especially antideuterons, are predicted to be distinctive probes for the dark matter annihilation or decay occurring in the Galactic halo [@Donato00; @Donato04; @Duperray05; @Aramaki16a; @Cuoco17; @Korsmeier18]. Precise measurement of the low-energy antiproton spectra will also provide crucial information on the source and propagation mechanisms of cosmic rays [@Moskalenko02; @Lin17; @Bindi17; @Aramaki14]. GAPS sensitivities to antideuterons and antiprotons are discussed in [@Aramaki16b] and [@Aramaki14], and capabilities for antihelium detection are being evaluated. The first flight of GAPS via a NASA Antarctic long duration balloon is planned for late 2021.
GAPS is comprised of a $1.6\;\rm{m}^W \times 1.6\;\rm{m}^D \times 1.0\;\rm{m}^H$ tracker made of Si(Li) detectors surrounded by a time-of-flight (TOF) system made of plastic scintillator paddles. A low-energy antinucleus triggered by the TOF is slowed and captured by the Si(Li) detector array, forming an excited exotic atom with a silicon nucleus. It immediately decays, radiating de-excitation X-rays of characteristic energies. The antinucleus then annihilates with the silicon nucleus, producing pions and protons with a multiplicity that scales with the incident antinucleus mass. The surrounding Si(Li) detectors measure the energies of the characteristic X-rays, which are specific to the incident antinucleus species. The ${\rm d}E/{\rm d}x$, trajectories, and total kinetic energies of the incoming antinucleus and outgoing annihilation products, as reconstructed by the Si(Li) and TOF systems, provide additional antiparticle identification power.
Therefore, the Si(Li) detector array plays an essential role in the GAPS detection scheme, providing the absorption depth, active area, tracking efficiency, and X-ray energy resolution necessary for this exotic atom particle identification technique. Our optimized flight detector design has an overall diameter of 10 cm, with ${\sim}$9 cm of active area segmented into 8 readout strips, and an overall thickness of 2.5 mm, with ${\gtrsim}2.2$ mm (${\gtrsim}90\%$) sensitive thickness. Both 4- and 8-strip designs have been developed and proved to meet the GAPS requirement. However, the 8-strip design has been adopted for the GAPS flight instrument, as it improves tracking performance and allows use of a custom ASIC readout, while still achieving the required energy resolution [@Rogers19]. This low-power ASIC reduces inactive material in the tracker and permits for lower temperature operation, and thus improved noise performance, compared to a discrete-component preamplifier readout. The detectors are arranged in 10 layers each with ${\gtrsim}100$ detectors to achieve the absorption depth necessary for incident antinuclei with energies ${<}0.25$ GeV/n [@Aramaki16b; @Aramaki14]. Hence the mass production of ${\gtrsim}1000$ detectors is required.
The leakage current and capacitance of each Si(Li) strip should be lower than 5 nA and ${\sim}$40 pF, respectively, to achieve the required energy resolution ${\lesssim}4$ keV (FWHM) for 20$-$100 keV characteristic X-rays [@Rogers19]. To suppress the power consumption, GAPS adopts a newly-developed cooling system for the Si(Li) detectors [@Fuke16; @Okazaki18]. Considering all thermal and mechanical restrictions for a balloon-craft, the cooling system is designed to cool Si(Li) detectors down to ${\sim}-40^\circ$C ($-35^\circ$C to $-45^\circ$C). The leakage current, which depends exponentially on temperature, must meet the requirement in this operating range. The capacitance requirement can be simultaneously achieved by realizing a thick sensitive layer. A relatively low bias voltage of 250 V is also required, because it allows for operation at the ambient flight pressure without suffering breakdown.
We have adopted Si(Li) detectors for the GAPS design, as they provide a sensitive layer thicker than a few millimeters with modest bias voltages, by compensating $p$-type Si crystal with Li ions [@Pell60; @Goulding66]. However, a high-yield, low-cost fabrication method for large-area Si(Li) detectors, with operating temperatures of at minimum ${\sim}-35$ to $-45^\circ$C, currently has no experimental precedent.
Si(Li) detectors have been commercially-produced previously, mainly in the energy dispersive X-ray (EDX) spectroscopy field. However, all of these EDX Si(Li) detectors are small with a diameter of ${\sim}1$ cm, and operated mainly at liquid nitrogen temperature, which is significantly lower than that required for GAPS. Several previous studies have reported on large-area Si(Li) detectors [@Miyachi88; @Miyachi94; @Kashiwagi90]; however, all these methods required removing the un-drifted region after the Li drift, because achieving a uniform drift across a large-area was impossible. As an approach to address this difficulty, it was demonstrated that uniform Li-drifting can be conducted into a large-area Si wafer by drifting towards a boron-implanted $p$-side layer [@Onabe02; @Tindall04; @Protic02; @Protic03; @Protic05]. However, boron implantation requires annealing at ${\sim}500^\circ$C, which has the potential to damage the Si crystals and increases the fabrication cost.
Two key techniques in the GAPS Si(Li) development are a uniform Li drift in large-area Si(Li) and a suppression of leakage current at the relatively high temperature. Based on previous research on prototype detectors [@Aramaki10; @Aramaki12; @Perez13; @Perez18], we have established a mass-production method for the GAPS Si(Li) detectors. Figure \[fig:det\_pic\] shows an example of a GAPS Si(Li) detector.
![A GAPS Si(Li) detector with a thickness of 2.5 mm, diameter of 10 cm, and 8 readout strips.[]{data-label="fig:det_pic"}](FIG1.pdf){width="50.00000%"}
In this paper, Sec. \[sec:overview\] briefly explains the overall fabrication flow of the GAPS Si(Li) detector. Detailed descriptions of the fabrication process are given in Sec. \[sec:dev\], where we highlight the specific developments investigated in over 50 prototype detectors. Sec. \[sec:performance\] demonstrates the performance and high-yield rate of the mass-produced detectors. Conclusions are presented in Sec. \[sec:conclusion\].
Overview of fabrication flow {#sec:overview}
============================
![Fabrication flow of the GAPS Si(Li) detector. The image on the right shows the cross-section of a Si(Li) detector in each process. Dimensions of the cross-section diagram are not to scale.[]{data-label="fig:flow"}](FIG2.pdf){width="\textwidth"}
Figure \[fig:flow\] provides an overview of the fabrication process. First, lithium is evaporated on one surface of a $p$-type Si wafer and diffused through a shallow depth by heating the wafer. This process supplies Li ions for compensating impurities in the $p$-type Si, while forming a Li-diffused $n^+$-layer. A metal contact is evaporated on $n$-side after the diffusion. The wafer is machined into the top-hat geometry by ultrasonic impact grinding (UIG) to confine the Li-drifted region. Li ions in the Li-diffused layer are drifted toward the $p$-side by applying a bias voltage to the heated wafer. During drifting, Li ions compensate impurities in the $p$-type Si, forming a well-compensated intrinsic layer ($i$-layer) that functions as the sensitive volume in the final detector. Grooves are machined on the $n$-side via UIG to isolate the guard-ring structure and readout strips. A metal contact is evaporated on the $p$-side. The GAPS Si(Li) detector is then completed by performing wet etchings and cleanings of its exposed Si surfaces.
![Half-section (upper figure) and its machined dimensions (lower table) of the GAPS Si(Li) detector. The cross-section figure is not to scale.[]{data-label="fig:dim"}](FIG3.pdf){width="\textwidth"}
We summarize the dimensions of the mass-production model of the GAPS Si(Li) detector in Fig. \[fig:dim\]. Displayed dimensions represent the machined dimensions by UIG. The final detectors deviate somewhat from these values mainly due to material removed during etching.
Our developments prioritize cost effectiveness for mass production while maximizing detector quality. Shimadzu Corporation, Japan, is an industry-leading company of Si(Li) EDX detectors. The fabrication apparatuses and methods herein are developed by extending the work of our previous studies [@Aramaki10; @Aramaki12; @Perez13; @Perez18], while utilizing Shimadzu’s techniques for quality control and cost reduction.
Development {#sec:dev}
===========
Procurement of p-type Si crystal
--------------------------------
--------------------------- ---------------------------------------
Fabrication method Floating zone
Type p
Dopant Boron
Crystal orientation $\langle111\rangle$
Oxygen concentration $<1 \times 10^{16} \rm{\;atoms/cm^3}$
Carbon concentration $<2 \times 10^{16} \rm{\;atoms/cm^3}$
Resistivity ${\sim}1000 \rm{\;\Omega \cdot cm}$
Minority carrier lifetime ${\sim}1$ ms
Diameter ${\sim}100$ mm (4 inches)
--------------------------- ---------------------------------------
: Specifications of Si crystal used for GAPS Si(Li) detector.
\[tbl:crystal\]
Several studies have reported that uniformly drifting Li ions into a thick, large-area, $p$-type Si wafer is difficult mainly due to defects and contaminants, such as oxygen and carbon, in the Si crystal [@Miyachi88; @Miyachi94; @Kashiwagi90; @Onabe02; @Murray66; @Fong82; @Litovchenko03]. These behave as traps for Li ions and hence decrease Li ion mobility in the crystal, making it hard to uniformly drift. Previous works also demonstrate that these impurities and defects concentrate especially in the central region of Si wafer, resulting in an uncompensated region at the center of the wafer’s $p$-side after the Li drift [@Kashiwagi90; @Miyachi94; @Onabe02].
We have successfully developed a high-purity $p$-type Si crystal specifically for the GAPS Si(Li) detectors in collaboration with SUMCO Corporation, Japan. Table \[tbl:crystal\] lists the specifications of the crystal used for our fabrications. For the raw material of our crystal growth, we employ polycrystalline silicon made from mono-silane. The crystal is grown to be oxygen free using the floating zone method with an axis of $\langle111\rangle$. Both Si crystals with $\langle111\rangle$ and $\langle100\rangle$ orientations are used in previous studies, but $\langle111\rangle$ is more proven for Si(Li) detector fabrication, including for Shimadzu’s commercial detectors. There is also one report indicating that it is empirically preferable for Li drift [@Kashiwagi90].
Resistivity of ${\sim}1000 \rm{\;\Omega \cdot cm}$ corresponds to an acceptor concentration of $N_A \approx 10^{13} \rm{\;atoms/cm^3}$ [@Goulding66], which is an order of magnitude lower density than that used in some of previous studies of large-area Si(Li) detectors [@Miyachi88; @Miyachi94; @Kashiwagi90; @Onabe02]. Substrate with a lower $p$-type acceptor concentration requires fewer Li ions for compensation, thus reducing the temperature and time required in the diffusion process. Reduction of the heating treatment prevents the Si crystal from forming defects. The in-plane non-uniformity of the resistivity of our Si wafer is ${\sim}10 \%$ based on the measurement of sample wafers.
The lifetime of minority carriers is an indicator of crystal defects and contaminants. The lifetime of ${\sim}1$ ms is enough to make a high-quality compensated region, as proven by our in-house development [@Perez18].
After procuring 2.5 mm-thick wafers cut from the 10 cm-diameter Si crystal, we remove foreign matter on the surface by organic-solvent cleaning with methanol, xylene, and acetone. We then etch the wafer surface with our etchant (a solution of hydrofluoric acid, nitric acid and acetic acid) for 2 min to remove surface contaminants and mechanical defects. Si oxide on the surface is then removed by immersing the wafer in a solution of hydrofluoric acid for 1 min.
Li evaporation and diffusion
----------------------------
Li is evaporated and diffused through a shallow depth, forming the $n^+$-layer. The Si wafer is set in a chamber of a custom vacuum-based thermal Li evaporator by sandwiching it between a heater plate and a lower mask. The mask has a large round aperture that prevents Li from being deposited onto the other side of the wafer. Our custom heater plate realizes in-plane uniformity for heating the large-area wafer, which is a key issue for obtaining a uniform Li-diffused layer. A molybdenum evaporation boat filled with Li pieces is set under the wafer. The shutter between the wafer and the boat is closed at the beginning of the process to prevent Li oxide, on the surface of the Li pieces, from evaporating onto the wafer. The chamber is then pumped to ${<}10^{-4}$ Pa, and the heater is set to 280$^\circ$C. After confirming pressure and thermal stability, the shutter is opened and the current through the boat is increased. Li is evaporated onto the wafer surface by maintaining the current for 1 min, and the pressure and temperature are kept for another minute to diffuse Li through the shallow depth. The pressure valve is then opened. The wafer is naturally cooled to ${\lesssim}50^\circ$C, and extracted from the chamber. We obtain a $n^+$-layer with a thickness of ${\sim}100\;\upmu\rm{m}$ by this diffusion process, as confirmed by our visual inspection of the cross-section of the Si(Li) detector via copper staining (see Sec. \[sec:drift\]).
The depth of the $n^+$-$p$ junction formed by Li diffusion for time $t$ is expressed as [@Goulding66] $$x_j = 2\sqrt{D t} \cdot \rm{erfc}^{-1} \left( \frac{N_A}{N_0} \right) \label{eq:xj},$$ where $D$ is the diffusion constant of Li in the Si crystal and $N_0$ is the Li surface concentration ($N_0 \approx 10^{16} \rm{\;atoms/cm^3}$ based on measurement of commercial Si(Li) [@Shimadzu16]). The diffusion constant is obtained by [@Pell60] with ${\sim}1000 \rm{\;\Omega \cdot cm}$ $p$-type Si crystal as $$D = 6 \times 10^{-4} \exp \left(\frac{-0.61q}{k_{\rm B} T} \right) \rm{\;[cm^2/s]}, \label{eq:D}$$ where $q$ is the elementary charge, $k_{\rm B}$ is the Boltzmann constant, and $T$ is temperature in K. Our diffusion time $t$ (1${\sim}$2 min at 280$^\circ$C and ${\sim}$150 min of the natural cooling time) leads to $x_j$ which is consistent with the practical depth ${\sim}100\;\upmu\rm{m}$, though the empirical equation (\[eq:D\]) of the diffusion constant is derived or referred with different coefficients in each previous work [@Pell60; @Kashiwagi90; @Fuller54; @Maita58] and provides only an approximation.
Reference [@Pell60] indicates that the diffusion constant should decrease with lower resistivity, i.e., highly-doped silicon crystal, by the ion-pairing effect. This implies our high-purity Si crystal allows us to obtain a Li-diffused layer thick enough with a lower temperature and a shorter heating time, preventing defects from being generated in the crystal by our heating procedure. Indeed, our heating temperature and time during the diffusion are cooler and shorter than previous research on large-area Si(Li) using ${\sim}100 \rm{\;\Omega \cdot cm}$ crystal. For example, [@Miyachi88] and [@Miyachi94] maintained the wafer at 400$^\circ$C for 15 min to form a ${>}100 \;\upmu \rm{m}$ diffused layer. Reference [@Kashiwagi90] requires the wafer to be maintained at a temperature ranging 300$^\circ$C to 400$^\circ$C for 5$-$20 min to obtain a diffused depth of 100$-$200 $\;\upmu \rm{m}$.
The Li oxide layer formed on the wafer surface is removed by chemical etching for 1 min with our etchant. This etching also removes mechanical defects and contaminants from the surface.
Evaporation of [*n*]{}-electrode and top-hat machining {#sec:ncontact}
------------------------------------------------------
In a thermal evaporator set at room temperature and pressure of ${<}10^{-4}$ Pa, an 18 nm layer of nickel followed by a 120 nm layer of gold are evaporated onto the $n$-side of the detector to form a metal contact. Nickel has good adhesion with the underlying Si, and gold has a high tolerance to oxidation.
The circumference of the [*n*]{}-side is then ground by UIG to make a top-hat geometry (see Fig. \[fig:flow\]) and to confine the Li from drifting to the sides of the wafer during the Li drifting procedure. UIG is generally less expensive than other methods such as diamond-saw cutting, and is thus more suitable for mass production.
The geometry of the top-hat brim is designed to be as narrow as possible to maximize the detector’s active area, while still being wide enough to be used for handling the detector. The top-hat brim is preferred for handling the detector since it is $p$-type even after the Li drift and thus does not need careful treatment (unlike the intrinsic region ($i$-region) surface, see Sec. \[sec:groove\]). As a result of this optimization, the inner diameter of the top-hat brim is set to 97 mm in the UIG machining, and the thickness is set to 1 mm, to prevent breakage during handling.
After the top-hat machining, an etch-resisting wax, Apiezon$^{\scriptsize\textregistered}$, is dissolved in xylene and is painted on the $n$- and $p$-side surfaces by hand. Damaged layers and metal contaminants generated by UIG are then removed by etching on the side surface of the top-hat for 12 min. Organic-solvent cleaning with methanol, xylene, and acetone is then performed to remove the wax.
Li drift {#sec:drift}
--------
![Profiles of the applied bias voltage (top), heater output (middle), and leakage current (bottom) during the drift of a sample detector. The inset plot in each panel displays each parameter’s variation in the first 3 hours of the drift, as the voltage is increased step-wise to the 600 V set-point.[]{data-label="fig:drift"}](FIG4.pdf){width="130.00000%"}
A uniform Li-drift is achieved by our drift apparatus, which has been custom-designed for large-area Si(Li) detectors. We found that retaining a thin undrifted layer on the $p$-side effectively suppresses the leakage current, as demonstrated in this section. The radial uniformity of the growth of the drifted region during Li drift is key to realizing this thin, uniform undrifted layer.
Li drifting is performed in a custom drift apparatus consisting of an electrically grounded heater plate, a pressure contact for applying a bias voltage, a resistance temperature detector (RTD), and a computer-based controller. The Si wafer is set on the heater plate with the $n$-side up and, the $p$-side connected to the grounded heater plate. The pressure contact and RTD are connected to the $n$-electrode to apply a positive bias voltage and monitor the wafer’s temperature. The drift sequence described below is automatically managed via the computer-based controller.
Figure \[fig:drift\] depicts an example of the bias voltage, heater output, and leakage current during the ${\sim}$110 hours of the drifting routine. Our sample detector is a 10-cm Si(Li) detector. The voltage is increased step-wise, in 100 V intervals every 30 min, to prevent rapid increase of the leakage current. The insets in each panel of Fig. \[fig:drift\] display the variations over the first 3 hours of the drift, as the voltage is ramped up to the set-point of 600 V. In the first panel, open arrows indicate the timing of the voltage steps. In the final panel, the leakage current shows a step-like increase of ${\sim}0.5$ mA corresponding to each voltage increase.
As Li drifts toward the $p$-side and the depletion layer expands from the $n$-side, the leakage current gradually increases. The Joule heat generated by the leakage current also increases. Displayed by the middle panel, heater output is automatically decreased to compensate for the Joule heat and keep the wafer at 100$^\circ$C.
At the end of the drift, the depletion layer approaches the $p$-side. At this point, the leakage current rapidly increases, the wafer temperature exceeds 100$^\circ$C due to the Joule heat, and the heater output decreases to zero. The bias voltage is automatically turned off, i.e., the Li drift is terminated, either when the leakage current reaches 25 mA or when the heater output becomes zero. The wafer is then allowed to naturally cool to room temperature.
Under a bias voltage as high as 600 V, the depletion layer expands slightly toward the $p$-side beyond the “$i$-$p$ junction” formed between the drifted and undrifted layer. Thus, despite the steep increase in leakage current, a thin undrifted layer is retained on the $p$-side of our Si(Li) detectors after the drifting process, as demonstrated in Fig. \[fig:copper\].
The necessary drift time ($t$) to obtain a drifted depth ($W$) under bias voltage ($V$) is given by [@Goulding66] $$t = \frac{W^2}{2V\mu_L}. \label{eq:tdrift}$$ The Li mobility ($\mu_L$) is related to the diffusion constant ($D$) by the Einstein relation $$\mu_L = \frac{q}{k_{\rm B} T} D \;[\rm{cm^2/(V \cdot s)}]. \label{eq:mu}$$ For our drift parameters, we derive $t \approx 100$ h to obtain a drifted depth $W \approx 2.2$ mm (${\sim}90\%$ of the overall 2.5 mm thickness). This calculated result is comparable to our actual drift time displayed in Fig. \[fig:drift\]. In the case that the Li drift is unexpectedly terminated before ${\sim}90$ h, we resume the drift sequence.
The temperature and voltage allow us to achieve a high quality drift in an acceptable drift time (${\sim}4$ days) for mass production. Higher bias voltage and temperature would reduce the required drift time. However, setting the wafer at a higher temperature by higher Joule heat or higher heater output will generate more hole-electron pairs, which attract Li-compensation, hence disturbing the ideal Li distribution, which should only compensate for acceptors [@Goulding66]. Our drift parameters, 600 V and 100$^\circ$C, are found to be sufficiently low to fabricate uniform and thick Li-drifted.
The Li-diffused and drifted regions can be visually inspected by copper staining on a cross-section of the Li-drifted Si wafer [@Whoriskey58; @Iles60; @Kume01]. In the copper-staining process, we first remove the $n$-electrode via etching and cut the wafer into two cross-sectional pieces via diamond-saw dicing. The cross-section is then polished and immersed in a saturated $\rm{CuSO_4}$ solution containing a few drops of concentrated hydrofluoric acid under the illumination of white light. Electrons generated by the light are more attracted to $n^+$- and $i$-regions than $p$-regions due to the photovoltage effect and cause the deionization of copper ions, resulting in copper deposition on the $n^+$- and $i$-regions.
![Six cross-sections of a sample Si(Li) detector after Li-drifting and copper staining. Sections (a)-(f) are indicated in the top figure. Section (g) displays an enlarged view of the indicated region in section (d) highlighting the undrifted layer.[]{data-label="fig:copper"}](FIG5.jpg){width="\textwidth"}
The copper-stained area is visually identified by a yellow-ish color. Figure \[fig:copper\] displays a cross-section of our sample detector processed by this copper-staining method. The light yellow-ish region corresponds to the Li-drifted region, whereas the darker grey-ish region filling the brim of the top-hat in Figs. \[fig:copper\](a) and \[fig:copper\](f) corresponds to the undrifted $p$-type bulk. The thin but intense yellow-ish layer on the $n$-side of the wafer indicates the $n^+$-layer, i.e., the Li-diffused layer. We confirm via this staining that the $n^+$-layer has a thickness of ${\sim}100 \;\upmu\rm{m}$.
The undrifted layer is also found in Fig. \[fig:copper\]; the dark grey-ish region seen at the brim spreads toward the center and extends to other side of the brim, keeping a thin undrifted region on the $p$-side. Figure \[fig:copper\](g) displays an enlarged view of the $p$-side region in picture (d), highlighting the undrifted layer. This undrifted layer appears uniform, with a thickness of ${\sim}100 \;\upmu\rm{m}$ across what will become the active region of the detector, indicating a successful uniform Li-drift.
![Leakage current of a sample detector as a function of bias voltage (I$-$V curve) at $-35^\circ$C before and after polishing undrifted layer on $p$-side. Circles display the I$-$V curve before polishing undrifted layer on $p$-side, whereas rectangles and triangles display the I$-$V curves after polishing $p$-side for depths of 40 and $100 \;\upmu\rm{m}$, respectively, from the unpolished $p$-side surface.[]{data-label="fig:polish"}](FIG6.pdf){width="75.00000%"}
In conventional Si(Li) detector fabrication techniques, the undrifted layer is removed to expose the Li-drifted layer on the $p$-side. Then a metal contact such as gold is evaporated as a Schottky barrier. However, we observed excessively high leakage currents at $-35^\circ$C with a detector using this conventional contact.
Figure \[fig:polish\] displays the leakage current of our 10-cm diameter Si(Li) detector at $-35^\circ$C as a function of bias voltage (I$-$V curve). In this prototype detector, the guard-ring structure described in Sec. \[sec:groove\], which is necessary to suppress the dominant surface leakage current and without which the effect of the undrifted layer is difficult to observe, had already been machined. On the contrary, machining of the strip grooves and the second etching, described in the Sections \[sec:groove\] and \[sec:etching\] respectively, were not performed in this detector. The circles in Fig. \[fig:polish\] display the I$-$V curve of the detector without polishing the $p$-side, i.e., with the ${\sim}100 \;\upmu\rm{m}$ undrifted layer remaining underneath the $p$-side nickel and gold contact. This $p$-side contact is formed after etching the $p$-side surface, in the same manner as Sec. \[sec:pcontact\]. After measuring this I$-$V curve, the $p$-side was polished to remove 40 $\upmu$m, leaving ${\sim}60 \;\upmu\rm{m}$ of undrifted layer, and the nickel and gold $p$-side contacts were re-applied. In this manner the thickness of the undrifted layer can be reduced easily and precisely in comparison to other options such as modifying the undrifted thickness by modifying the drifting parameters.
The rectangles in Fig. \[fig:polish\] show the I$-$V curve of this detector after removing 40 $\upmu$m of the undrifted layer. The leakage current is significantly higher than that of the unpolished detector by an order of ${\sim}10^2$. We next removed an additional 60 $\upmu$m from the $p$-side of the detector via polishing, thus removing the entire undrifted $p$-side layer, and re-applied the nickel and gold contacts. The resulting I$-$V curve is indicated by the triangles in Fig. \[fig:polish\]. The metal contact should form a Schottky barrier on the $p$-side but the leakage current is higher than that of the unpolished detector (circles) by an order of ${\sim}10^3$.
This result implies that a large-area Schottky barrier contact easily breaks down at our operating temperature (${\sim}-35$ to $-45^\circ$C), which is significantly higher than that of conventional Si(Li) detectors used for X-ray spectrometry. Based on this presumption, we decided to retain the undrifted layer of ${\sim}100 \;\upmu\rm{m}$ on the $p$-side to suppress the bulk leakage current, which we posit is generated in or flowing through the depletion layer via a junction between the $i$- and $p$-layers.
In the conventional detectors, the $p$-contacts are used as windows for low-energy X-rays [@Lyman89; @Rossington91; @Cox05]. Therefore, such detectors adopted Schottky barrier contacts to minimize the $p$-side insensitive layers while suppressing bulk leakage currents. Since GAPS aims to detect X-rays with energies higher than 20 keV, the ${\sim}100 \;\upmu\rm{m}$ insensitive layer is acceptable for the GAPS detectors.
Machining grooves for the guard-ring and strips {#sec:groove}
-----------------------------------------------
![I$-$V characteristics of a sample detector at $-35^\circ$C before machining the guard-ring groove (triangles); after machining the groove, with a floated guard-ring (rectangles); and after grounding the guard-ring (circles).[]{data-label="fig:GR"}](FIG7.pdf){width="75.00000%"}
The exposed $i$-region can be easily contaminated and contribute to the leakage current [@Goulding66; @Goulding61]. The side surface of the top-hat has the largest area of the exposed $i$-layer and is a major source of the leakage current. The guard-ring groove [@Goulding61] is machined to suppress this surface leakage current, preventing it from flowing into the readout electronics.
After the Li-drifting process, a circular groove, with a ${\sim}300 \;\upmu\rm{m}$ depth and 1-mm width, is cut into the $n$-side by UIG so that the central area, or the readout electrode of the detector, is electrically isolated from the perimeter, or guard-ring electrode (see Fig. \[fig:det\_pic\]). The groove is deep enough to cut through the $n^+$-layer (${\sim}100 \;\upmu\rm{m}$ depth). During detector operation with a high bias voltage, the $i$-layer between the guard ring and the active area is fully depleted and its resistivity increases dramatically to isolate the readout electrode from the guard-ring electrode.
Figure \[fig:GR\] displays the I$-$V curves of a sample detector. The undrifted layer described in Sec. \[sec:drift\] had been formed in this prototype detector, but the machining of the strip grooves (described below) and the second etching (described in Sec. \[sec:etching\]) were not performed. The I$-$V curves in Fig. \[fig:GR\] are measured before machining the guard-ring groove (triangles); after machining the groove, with a guard-ring not connecting to any external electrode (rectangles); and after grounding the guard-ring (circles). As described in Sec. \[sec:performance\], negative bias voltage is applied to the $p$-electrode, and the strips that are not being measured are connected to the ground. As shown in Fig. \[fig:GR\], the grounded guard-ring drastically reduces the leakage current flowing into the readout electrode by an order of ${<}10^{-2}$ in comparison to the detector without a guard-ring. This indicates that the readout electrode is successfully isolated from both the guard-ring electrode and the side surface of the detector.
Grooves that divide the active area into 8 readout strips of equal area are cut at the same time as machining the guard-ring groove, using the same groove depth and width. Under the operating bias voltage, the strips are electrically isolated from each other by the same principle as the guard-ring.
A narrower groove width is preferable because the exposed $i$-layer in the groove can generate leakage current. However, it is important to note that UIG is limited by the width of its grinding tool. An extremely thin tool is difficult to make and is easily damaged in the UIG process. A narrower groove also has disadvantages for groove etchings. If the groove is too narrow, bubbles generated via chemical reactions in the etching process stick to the groove surface. The bubbles make it difficult to etch uniformly and sufficiently. We optimized the UIG tool for the mass-production model of the GAPS Si(Li) detector and deduced the optimal groove width as 1 mm.
The guard-ring width, i.e., width of the perimeter electrode in Fig. \[fig:det\_pic\], is also optimized. A narrow guard-ring electrode is preferable to make the active area as wide as possible. However, it is difficult to precisely paint etch-resisting wax, which is necessary during the groove etching process, onto a narrow guard ring. We deduced that the optimal guard-ring width is 2.5 mm.
Forming $p$-side contact {#sec:pcontact}
------------------------
Wax is painted on the $n$-side including the grooves. Then, the side surface of the top-hat and the $p$-side surface are etched for 1 min. After the wax is removed by an organic solvent, the metal contact on the $p$-side is evaporated in the same manner as Sec. \[sec:ncontact\].
Etching on side of the top-hat and grooves {#sec:etching}
------------------------------------------
![Leakage current of each strip at a bias voltage of 200 V and a temperature of $-35^\circ$C after the 1st, 2nd, 3rd, and 4th $n$-side groove etchings. The 1st etching is performed for 10 min, whereas the 2nd$-$4th etchings are performed for 5 min. The numbers below the horizontal axis show the cumulative etching time.[]{data-label="fig:etching"}](FIG8.pdf){width="75.00000%"}
The side of the top-hat and the $n$-side grooves are etched after painting wax on the $n$- and $p$-electrodes. This last etching not only removes the damaged layer in the grooves formed by UIG in Sec. \[sec:groove\] but also smooths the surfaces and removes contaminants from all areas of exposed silicon. Organic-solvent cleaning is performed after this etching to remove the wax. It is assumed that this cleaning with methanol produces a lightly $n$-type surface on the exposed $i$-region, which ensures no electric breakdown of the detector under the high bias voltage [@Goulding66; @Llacer64].
An optimized etching time is required because etching for too long not only decreases the active area but also expands the area in which the $i$-layer is exposed. This area is easily contaminated, which increases leakage current. The number of separate etchings is also a key parameter. The wax is painted by hand, creating a non-uniform line at the edge of the painted area and thus introducing non-uniformities to the edge of the electrode after etching. Bubbles generated in the etchant also make the surface state non-uniform. Repeating the wax painting and the etching reduces these irregularities.
We surveyed the optimal etching time and number of discrete wax painting and etching processes. Figure \[fig:etching\] displays the leakage current of each strip of a prototype detector with 4 strips (A$-$D) at a bias voltage of 200 V and a temperature of $-35^\circ$C. The leakage current is displayed as a function of the cumulative etching time or the total number of etchings. The first etching is performed for 10 min, whereas the second, third, and fourth etchings are 5 min each. Leakage currents in all strips are clearly reduced after the second etching, i.e., cumulative etching time of 15 min, compared to that after the first etching. It was also confirmed by a visual inspection that the groove surfaces are clearly smoothed and glassy after the second etching compared to after the first etching. The leakage currents are not significantly affected after the additional third and fourth etchings. This result indicates that the 15 min of etching in two discrete etching procedures is sufficient to minimize the surface leakage current.
Performance of the mass-produced detectors {#sec:performance}
==========================================
![Diagram of the I$-$V curve measurement.[]{data-label="fig:measure"}](FIG9.pdf){width="\textwidth"}
Ten prototype detectors (test detectors) were fabricated to validate and fix the mass-production process. Then, the mass production of the flight detectors had been started. In this section, we evaluate leakage currents and capacitances of these mass-produced detectors (10 test detectors and 10 flight detectors), which were fabricated by identical procedures. Thanks to the sufficiently low leakage currents, it has been confirmed that energy resolution ${\lesssim}4$ keV (FWHM) for 20$-$100 keV X-rays are achieved [@Kozai18]. Details of the energy resolution measurements are described in a separate publication [@Rogers19].
Figure \[fig:measure\] illustrates the setup for the I$-$V curve measurement. The detector is mounted in a vacuum chamber set at $10^{-2}$ Pa. A Teflon plate mounted on the lower plate of the detector holder electrically isolates the $p$-electrode of the detector from the lower plate, which is fixed to the chamber and connected to the ground. An aluminum plate between the Teflon plate and the detector applies negative bias voltage to the $p$-electrode. The upper plate of the holder is bolted to the lower plate, holding in place a Teflon ring pressed against the guard-ring. The detector is fixed by this pressure. Copper foils are pasted on the Teflon ring to electrically connect the guard-ring and the upper plate; therefore, the guard-ring electrode is grounded via the chamber.
A cold head bolted to the detector holder is cooled by a Stirling cooler. The detector is cooled by the thermal connections to the holder through the copper foils, Teflon ring, alminum plate, and Teflon plate. Calibration of the cooler was performed using a dummy wafer on which an RTD was used to monitor the detector’s temperature.
All strips of the detector are connected to the hermetic connector mounted on the side of the chamber via probes made of gold wires. The gold wire probe is selected to prevent damage to the strip electrodes on the detector. Using a multimeter, we verify the electrical connections between the hermetic connector and strips and reasonable resistivities between the strips, guard-ring, and $p$-electrode, each time the detector is set. The strip to be measured is connected to a picoammeter or capacitance meter while the other strips are connected to the ground outside the chamber.
For the picoammeter, capacitance meter, and bias supply, HP 4140B, HP 4280A, and ORTEC 428 were used, respectively. Recently the picoammeter and bias supply were replaced with a Keithley 6487 picoammeter, which has an internal bias supply. We constructed an automatic I$-$V measurement system controlled by LabVIEW software with this Keithley picoammeter. It will improve the speed of the mass production of the GAPS flight detectors in terms of detector evaluation.
![Leakage currents (upper panel) and capacitances (lower panel) of all strips (A$-$H) of a sample detector at $-35^\circ$C, as functions of the bias voltage.[]{data-label="fig:IC"}](FIG10.pdf){width="80.00000%"}
The upper panel of Fig. \[fig:IC\] displays I$-$V curves of 8 strips of a sample detector at $-35^\circ$C. The leakage current of each strip is stable and is ${\sim}1.5$ nA at our operating voltage (200$-$300 V), significantly lower than the requirement, ${\sim}5$ nA [@Rogers19]. The capacitance as a function of the bias voltage in the lower panel of Fig. \[fig:IC\] reaches the minimum by ${\sim}100$ V, indicating the detector is fully depleted around this voltage. The capacitance of $C \approx 38$ pF also meets the requirement and corresponds to the thickness of the depletion layer [@Goulding66], $$W_d \approx 1.05 A/C \approx 2.2 \;\rm{mm},$$ for a strip area of $A=8.1 \;\rm{cm^2}$. This is ${\sim}90\%$ of the 2.5 mm total detector thickness. The uniformity of the capacitance, with a fluctuation of ${\lesssim}{\pm}1.5\%$ between strips, indicates a uniform Li-drifted layer.
![Histograms of the leakage current measured for all strips of the 10 test detectors (black line) and 10 flight detectors (red dashed line). The bias voltage and the temperature are set at 250 V and $-35^\circ$C, respectively.[]{data-label="fig:hist"}](FIG11.pdf){width="80.00000%"}
Figure \[fig:hist\] displays histograms of the leakage current measured for all strips of the 10 test detectors and 10 flight detectors. The bias voltage and the temperature are set at 250 V and $-35^\circ$C, respectively. In 80 strips of the 10 test (10 flight) detectors, 73 (75) strips have the leakage currents below 5 nA. Our acceptance criteria of the leakage current of each detector, that at least 7 strips out of 8 must have leakage currents ${\leq}$5 nA, was satisfied by 9/10 test detectors and 9/10 flight detectors. All strips of the test and flight detectors pass our acceptance criteria of the capacitance, ${\leq}$42 pF at 250 V at room temperature, and we confirmed that the capacitances at $-35^\circ$C stays within 97-99% of the room temperature value for all 80 strips of the test detectors. Capacitances of the flight detectors at cold temperature are not measured in our mass-production scheme to cut the lead-time of the mass production.
Conclusions {#sec:conclusion}
===========
We have developed and established, for the first time, a high-yield fabrication method for large-area Si(Li) detectors operated at relatively high temperature. Our 10 cm-diameter $p$-type crystal is verified to have sufficiently low impurity concentrations and radial non-uniformities to fabricate a uniform Li-drifted layer. While a large-area Schottky barrier contact on the $p$-side can easily break down at $-35^\circ$C, our optimized drift sequence retains a uniform undrifted layer on the $p$-side with a thickness of ${\sim}100 \;\upmu\rm{m}$, which drastically reduces bulk leakage current. The guard-ring groove and optimized etching process are also confirmed to effectively reduce the leakage current.
Our Si(Li) detector design has a sensitive layer with a ${\gtrsim}2.2$ mm depth, ${\gtrsim}90\%$ depth of the overall detector thickness of 2.5 mm, and an overall diameter of 10 cm, with ${\sim}$9 cm of active area. The detector is segmented into 8 readout strips and the operating temperature is ${\sim}-40\pm5^\circ$C. Based on the 20 mass-produced detectors we have produced, the production yield is sufficiently high (${\sim}$90%) at this point. The detectors’ leakage currents and capacitances are low enough to achieve ${\lesssim}4$ keV energy resolution for 20$-$100 keV X-rays. Good uniformity of the capacitance between strips indicates that the detector volume is uniformly compensated by our Li drift. This is also confirmed by copper staining on cross-sections of sample Si(Li) detectors.
The mass production of ${\gtrsim}1000$ GAPS Si(Li) detectors was started in late 2018 and will extend through early 2020.
Acknowledgements {#acknowledgements .unnumbered}
================
We thank SUMCO Corporation and Shimadzu Corporation for their co-operation in our detector development.
M. Kozai is supported by the JSPS KAKENHI under Grant No. JP17K14313. H. Fuke is supported by the JSPS KAKENHI under Grants No. JP2670715 and JP17H01136. K. Perez receives supports from the Heising-Simons Foundation, the Alfred P. Sloan Foundation, and the ULVAC-Hayashi MISTI Seed Fund. F. Rogers is supported through the National Science Foundation Graduate Research Fellowship under Grant No. 1122374.
This work was partially supported by the JAXA/ISAS Small Science Program in Japan and the NASA APRA program through Grant NNX17AB44G in the US.
We are grateful to all GAPS collaborators.
References {#references .unnumbered}
==========
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'High-resolution field localization in three dimensions is one of the main challenges in optics and has immense importance in fields such as chemistry, biology, and medicine. Time-reversal symmetry of waves has been a fertile ground for applications such as generating a subwavelength focal spot and coherent-perfect absorption. However, in order to generate the time reversed signal of a monochromatic source *discrete* sources that are modulated according to the wave amplitude on a spherical envelope are required, rendering it applicable only in acoustics. Here we approach these challenges by introducing a spherical layer with a resonant permittivity, which naturally generates the spatially *continuous* time-reversed signal of an atomic and molecular multipole transition at the origin. We start by utilizing a spherical layer with a resonant TM $l=1$ permittivity situated in a uniform medium to generate a free-space-subwavelength focal spot at the origin. We remove the degeneracy of the eigenfunctions of the composite medium by situating a point current source (or polarization) directed parallel to the spherical layer, which generates a focal spot at the origin *independently* of its location. The free-space focal spot has a full width at half maximum of $0.4\lambda$ in the lateral axes and $0.58\lambda$ in the axial axis, which is tighter by a factor of $\sqrt{2}$ in each dimension in excitation-collection mode, overcoming the $\lambda/2$ far-field resolution limit in three dimensions. We then explore two directions to localize electric field with deep-subwavelength resolution in three dimensions using this setup. Since the imaginary part of the eigenvalue is also realized in the physical parameter and the setup can be in an exact resonance, it can also open avenues in fields such as cavity QED, entanglement, and quantum information. In addition, we show that spherical structures exhibit a new type of degeneracy in which an infinite number of eigenvalues asymptotically coalesce. This high degeneracy results in a variety of optical phenomena such as strong scattering and enhancement of absorption and emission from an atom or molecule by orders of magnitude compared with a standard resonance.'
author:
- Asaf Farhi
title: 'Three-dimensional-subwavelength field localization, time reversal of sources, and infinite-asymptotic degeneracy in spherical structures'
---
The focal spot size that can be achieved by uniformly illuminating a circular aperture in the scalar approximation is given by an Airy disc, which is the Fourier Transform of a circular window [@jackson2012classical]. The full width at half maximum (FWHM) of this function is $1.02\lambda/\mathrm{NA},$ where $\lambda$ is the wavelength and $\mathrm{NA}$ is the numerical aperture $\left(\mathrm{NA}\lesssim1\right).$ This size is associated with the lateral axes and in the axial axis the FWHM is $~2.5-3$ times larger due to the fact that a smaller range of $k_z\mathrm{s}$ is involved. For gaussian beams, however, the focal spot is larger and depends on the width of the beam. The optimal lens resolution enables to image most biological cells but not viruses, proteins, and smaller molecules. Techniques such as confocal microscopy, structured illumination, beam shaping, and hyperlens imaging have been used to increase the lateral resolution [@minsky1988memoir; @muller2010image; @rho2010spherical]. In a $4\pi$ microscope the sample is illuminated from both sides and better resolution in the axial axis can be achieved [@cremer1974considerations]. However, in this setup side lobes are generated and the optical system needs to be realigned before every measurement in order for the focal spots to merge. Techniques based on fluorescence such as STED [@hell1994breaking], and PALM and STORM [@betzig2006imaging; @moerner1989optical] enable subwavelength resolution by stimulating emission at another frequency using an additional torus-like illumination and by activating subsets of fluorescent molecules, which enables to accurately calculate the molecule locations, respectively. Maxwell fisheye is a spherical lens with a radius-dependent refraction index in which all light rays emitted from a point meet at the antipodal point. The possibility of obtaining subwavelength resolution inside this setup has been the subject of recent works [@bitton2018two; @alonso2015maxwell]. Time reversal of waves has also been applied for generating a subwavelength focal spot [@ScienceFink; @ma2018towards]. Finally, methods based on evanescent waves to enhance resolution such as near-field imaging and negative-refractive index lens enable subwavelength focusing usually for two-dimensional imaging [@pendry2000negative]. Here, we utilize a resonant spherical layer to localize far-field light in several settings. We first situate the spherical layer in a uniform medium and excite it with a point current. This setup generates a three-dimensional free-space subwavelength focal that has very minor side lobes. Since it is composed of one “lens” it may not need to be aligned. In excitation-collection mode the effective focal spot is further minimized and there are almost no side lobes. We then explore two directions to localize far-field with deep-subwavelength resolution using this setup.
Time reversal of waves has been utilized for various interesting applications such as wave localization [@fink1992time; @ScienceFink; @lerosey2004time; @ma2018towards] and coherent-perfect absorption [@CPA2012]. Recently, it was shown that the time reversal of a source, in the presence of a near-perfect absorber, results in a subwavelength focal spot [@ma2018towards]. In order to generate the time reversal of a wave generated by a source, one would have to let the wave propagate from the source, “freeze” time, and generate discrete sources on a spherical envelope modulated according to the wave amplitude. Here, we will utilize the resonant-spherical layer setup to generate the spatially-continuous time reversal wave of sources, enabling its use in electrodynamics.
Degeneracies of eigenvalues can arise from a symmetry of the system or from a special feature of the system. While the first type of degeneracy is widely known (e.g., $m$ degeneracy in spherical multipoles), the second, called accidental degeneracy, is more exotic and includes phenomena such as Landau levels [@landau1977quantum], exceptional points [@bender1998real; @el2007theory; @ruter2010observation; @lin2016enhanced; @sweeney2019perfectly] and the accumulation point of the eigenpermittivities of evanescent modes [@bergman1979]. Degeneracies are associated with a strong response of the system as several modes are excited. In exceptional points for example, the degeneracy is usually second or third order and can lead to enhancement of emission from a molecule by two orders of magnitude due to enhancement of the density of states [@pick2017general]. In addition, the accumulation point of the eigenpermittivities of the evanescent modes can enhance the field (and emission) for a source that is very close to a metal-dielectric interface. Here we will show analytically and numerically that a spherical structure with a radius larger than $20\lambda$ exhibits infinite asymptotic all-even and all-odd TE/TM degeneracies of the second type. These degeneracies are associated with *far field* and *dielectric* spherical structures, in some cases with gain.
In a homogeneous medium the continuous-wave source-free solutions of Maxwell’s equation are plane waves, vector spherical harmonics, and vector cylindrical harmonics. It was recently shown that similarly to the situation in phased arrays in which plane currents proportional to a homogeneous medium source-free solution with a planar geometry generate the same function, currents proportional to a vector spherical harmonic (VSH) on a spherical surface generate the same VSH. Interestingly, a TM $l=1$ VSH near the origin has a subwavelength far-field focal spot [@farhi2017generating], which is smaller in volume by a factor of $\sim27$ compared with the focal spot that can be achieved by uniformly illuminating a lens. For a medium with a refractive index larger than 1, the TM $l=1$ field will have even a smaller focal spot. Generating this mode by oscillating currents can be thought of as a continuous-wave time reversal of the field of an oscillating dipole at the origin. Importantly, generating these VSH propagating towards the origin are the time reversal of the atomic and molecular multipole transitions. It is thus of interest to generate these modes near the origin. However, the spatial distributions of these VSH are complex and a setup of currents modulated accordingly is infeasible.
Eigenfunctions of Maxwell’s equations are fields, which exist without a source for certain physical parameters that correspond to resonances of the system [@bergman1980theory; @ge2010steady]. Here, we utilize resonances in a setup of a spherical layer in a host medium to naturally generate a the VSHs. This setup requires only a point source or emission in order to generate these field patterns. The permittivity value of the spherical layer $\epsilon_{1}$ will be close to a resonant TM $l=1$ permittivity value in order to generate this VSH (a resonant permittivity enables the existence of a field without a source as in a gain medium in laser). Similarly, all the other modes can be excited for the permittivity values close the eigenpermittivities, generating the time reversal of all the multipole radiation patterns, which correspond to all the emission and absorption transitions of atoms and molecules [@condon1934absolute]. Alternatively, a frequency which is close to an eigenfrequency can be used. Using an eigenpermittivity is advantageous in this context since the resonance can be fully reached by introducing a gain. While these eigenvalues are usually associated with a gain that is needed to generate the field, there are some cases when they are real valued [@farhi2016electromagnetic] or have epsilon near zero [@alu2007epsilon; @farhi2016electromagnetic].
The electromagnetic field expansion for a physical electric field $\boldsymbol{E}$ at a given angular frequency $\omega$ can be written as follows [@bergman1980theory] $${\bf E}={\bf E}_{0}+\sum_{n}\frac{s_{n}}{s-s_{n}}\frac{\langle\tilde{{\bf E}}_{n}|{\bf E}_{0}\rangle}{\langle\tilde{{\bf E}}_{n}|{\bf E}_{n}\rangle}\left|{\bf E}_{n}\right\rangle ,$$ where $s_{n}\equiv\epsilon_{2}/\left(\epsilon_{2}-\epsilon_{1n}\right)$ is the eigenvalue, $\epsilon_{2}$ is the host-medium permittivity, $s\equiv\epsilon_{2}/\left(\epsilon_{2}-\epsilon_{1}\right),$ $\boldsymbol{E}_{n}$ and $\tilde{\boldsymbol{E}}_{n}$ are the eigenfunction and its dual, and $\boldsymbol{E}_{0}$ is the incoming field. $\left\langle \boldsymbol{E}_{1}|\boldsymbol{E}_{2}\right\rangle =\int d\boldsymbol{r}\theta_{1}\left(\boldsymbol{r}\right)\boldsymbol{E}_{1}\cdot\boldsymbol{E}_{2}$ and $\theta_{1}\left(\boldsymbol{r}\right)$ is a window function which equals 1 inside the inclusion volume. Thus, when $\epsilon_{1}$ is close to $\epsilon_{1n},$ $1/(s-s_n)\gg 1$ and the corresponding eigenfunction has a large contribution in the electric field expansion (see for example Ref. [@farhi2017eigenstate], Fig. 2). Clearly, other modes and the incoming field exist in the expansion. Fortunately, close to a resonance, the TM $l=1$ eigenfunction will have the dominant contribution inside the spherical volume.
Still, VSHs have a degeneracy in the $m$ index, which usually results in the generation of all the $m$ modes as a response to an incoming electric field. We therefore employ the current formulation of the field expansion in order to remove this degeneracy. In this formulation we express the incoming field in terms of Green’s tensor ${\bf E}_{0}({\bf r})=\int dV\overleftrightarrow{G}\left(\mathbf{r},\mathbf{r}'\right)\cdot\mathbf{J}\left(\mathbf{r'}\right)$ and substitute it in $\langle\tilde{{\bf E}}_{n}|{\bf E}_{0}\rangle.$ Then, we change the order of integration and use the definition of the eigenfunction to obtain [@farhi2016electromagnetic] $$\begin{aligned}
\langle{\bf \tilde{E}}_{n}|{\bf E}_{0}\rangle=-\frac{4\pi i}{\epsilon_{2}\omega}\int dV'\theta_{1}\left(\mathbf{r}'\right){\bf E}_{n}\left(\mathbf{r}'\right)\cdot\int dV\overleftrightarrow{G}\left(\mathbf{r}',\mathbf{r}\right)\cdot\mathbf{J}\nonumber\\
=-\frac{4\pi i}{\epsilon_{2}\omega}s_{n}\int dV{\bf E}_{n}\left(\mathbf{r}\right)\cdot\mathbf{J}_{{\rm dip}}\left(\mathbf{r}\right)=-\frac{4\pi s_{n}}{\epsilon_{2}}\mathbf{p}\cdot\mathbf{E}_{n}({\bf r}_{0}),\nonumber\end{aligned}$$ where $\mathbf{J}_{{\rm dip}}\left(\mathbf{r}\right)$ is an oscillating point electric dipole, $\mathbf{p}$ is the dipole moment, and $\omega$ is the oscillation frequency.
Now the expansion of the electric field reads $$\begin{aligned}
{\bf E}={\bf E}_{0}-\frac{4\pi}{\epsilon_{2}}\sum_{n}\frac{s_{n}^{2}}{s-s_{n}}\frac{\mathbf{p}\cdot\tilde{{\bf E}}_{n}\left(\mathbf{r}_{0}\right)}{\langle\tilde{{\bf E}}_{n}|{\bf E}_{n}\rangle}\left|{\bf E}_{n}\right\rangle .\end{aligned}$$ Thus, situating an oscillating dipole may result in the generation of one TM $l=1$ mode (see Fig. 1).
The general form of a TM VSH is [@jackson2012classical] $$\begin{aligned}
\boldsymbol{E}_{lm}^{^{\left(E\right)}}\propto&\frac{1}{\epsilon\left(r\right)}\nabla\times f_{l}\left(kr\right)\boldsymbol{X}_{lm},\,\,\,\boldsymbol{X}_{lm}=\frac{1}{\sqrt{l\left(l+1\right)}}\boldsymbol{L}Y_{lm},\nonumber\\
&f_{l}\left(kr\right)=A_{l}^{^{\left(1\right)}}h_{l}^{^{\left(1\right)}}\left(kr\right)+A_{l}^{^{\left(2\right)}}h_{l}^{^{\left(2\right)}}\left(kr\right),\nonumber\end{aligned}$$ where $f_{l}\left(r\right)$ is a linear combination of spherical Hankel functions, $h_{l}\left(r\right)$ is a spherical Hankel function, $k$ is the wavevector, $\boldsymbol{L}=\frac{1}{i}\left(\boldsymbol{r}\times\nabla\right),$ and $Y_{lm}$ is a spherical harmonic.
For a spherical layer in $r_{1}<r<r_{2},$ $f_{l}\left(r\right)$ that satisfies boundary conditions is of the form $$f_{l}\left(r\right)=\left\{ \begin{array}{cc}
C_{l}h_{l}^{^{\left(1\right)}}\left(k_2 r\right) & r>r_{2}\\
B^{(1)}_{l}h_{l}^{^{\left(1\right)}}\left(k_{1n}r\right)+B_{l}^{^{\left(2\right)}}h_{l}^{^{\left(2\right)}}\left(k_{1n}r\right) & r_{1}<r<r_{2}\\
A_{l}j_{l}\left(k_2r\right) & r<r_{1}
\end{array}\right.,$$ where $j_{l}\left(r\right)$ is a spherical Bessel function and $k_{1n},k_2$ correspond to $\epsilon_{1n},\epsilon_2,$ respectively. These eigenfunctions are standing waves for $r<r_1$ and propagating waves for $r>r_2$ at a given frequency. The eigenpermittivity $\epsilon_{1n}$ in $r_{1}<r<r_{2}$ is calculated using an eigenvalue equation as we now explain.
An eigenpermittivity enables the existence of the field without a source and we therefore only need to impose boundary conditions. From continuity of tangential $\boldsymbol{E}$ and $\boldsymbol{H}$ we have for a TM eigenfunction (assuming $\epsilon_2=1$) $$\begin{aligned}
f_{\text{l}}\left(r_{1}^{-}\right)=f_{\text{l}}\left(r_{1}^{+}\right),\,\,\,f_{\text{l}}\left(r_{2}^{-}\right)=f_{\text{l}}\left(r_{2}^{+}\right),\nonumber\\
\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{1}^{-}}=\frac{1}{\epsilon_{1n}}\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{1}^{+}},\nonumber\\
\frac{1}{\epsilon_{1n}}\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{2}^{-}}=\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{2}^{+}},\nonumber\end{aligned}$$ from which we obtain an eigenvalue equation and $\epsilon_{1n}.$ Similarly for a TE eigenfunction we write $$\boldsymbol{E}_{lm}^{^{\left(M\right)}}\propto f_{l}\left(kr\right)\boldsymbol{X}_{lm},$$ with the boundary conditions $$\begin{aligned}
f_{\text{l}}\left(r_{1}^{-}\right)=f_{\text{l}}\left(r_{1}^{+}\right),\,\,\,f_{\text{l}}\left(r_{2}^{-}\right)=f_{\text{l}}\left(r_{2}^{+}\right),\nonumber\\
\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{1}^{-}}=\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{1}^{+}},\nonumber\\
\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{2}^{-}}=\left.\frac{\partial\left(rf_{\text{l}}\left(r\right)\right)}{\partial r}\right|_{r=r_{2}^{+}}.\nonumber\end{aligned}$$ Clearly, the eigenpermittivities of the TE and TM modes depend on the radius and the thickness of the spherical layer. The eigenfunctions in the radiation zone (far field) can be expressed as [@jackson2012classical] $$\boldsymbol{E}_{lm}^\mathrm{TM}\rightarrow Z_{0}\boldsymbol{H}_{lm}^{\mathrm{TM}}\times\boldsymbol{n},$$ where $\boldsymbol{n}=\boldsymbol{r}/r.$ Hence, since $\boldsymbol{H}_{lm}^{\mathrm{TM}}\propto \boldsymbol{E}_{lm}^{\mathrm{TE}} $ is parallel to the sphere surface [@jackson2012classical], $\boldsymbol{E}_{lm}^{\mathrm{TM}}$ is also parallel to the sphere surface. Thus, due to the inner product in Eq. (2), when an oscillating dipole is placed in the radiation zone it may excite a mode if it is oriented parallel to the spherical-layer surface.
For concreteness, we situate an oscillating dipole outside the spherical layer on the positive $x$ axis. The $y,z$ components of $H_{lm}^{\mathrm{TM}}$ can be found from [@jackson2012classical] $$H_{lm}^{\mathrm{TM}} \propto \boldsymbol{L}Y_{lm},\,\,\,L_{y}=\frac{1}{2i}\left(L_{+}-L_{-}\right),\,\,L_zY_{lm}=mY_{lm}.$$ The $z$ components of the TM $l=1$ eigenfunctions in the radiation zone readily follow from the two relations above $$E_{l=1,m=0\,z}^{\mathrm{TM}}\neq0,\,\,\,E_{l=1,m=\pm1\,z}^{\mathrm{TM}}=0.$$ Thus, by placing an oscillating dipole on the $x$ axis directed along the $z$ axis we have removed the $m$ degeneracy of the TM modes. It can be seen that objects *at all locations* will generate a focal spot at the origin. In addition, the $\theta$ dependency of the field can be written as $\boldsymbol{E}_{\mathrm{TM}\,l=1}\left(\theta\right)\propto\sin\theta\left(-\hat{x}\cos\theta+\hat{z}\sin\theta\right),$ which equals the $\theta$ dependency of the far field of an oscillating dipole and shows that the mode is indeed its time reversal. Oscillating dipoles on the $xy$ plane directed along $z$ will generate fields in the $z$ axis at the focal spot. From symmetry, situating several current sources will result in a superposition of the TM $l=1,m=0$ mode according to their locations and orientations. In addition, other forms of illumination (which correspond to current distributions) such as a laser illumination may also be used to generate a subwavelength focal spot (the current source may be associated with the gain medium). Also, since [@jackson2012classical] $$\sum_{m=-l}^{l}\left|\boldsymbol{X}_{lm}\left(\theta,\phi\right)\right|^{2}=\frac{2l+1}{4\pi}$$ combining two spherical structures (e.g., a sphere and a spherical layer), each corresponding to a TM $l=1$ resonance at a given frequency, and using oscillating dipoles such that all the $m$ modes are excited, will result in isotropic radiation.
In order to have a dominant contribution of the TM $l=1$ modes, the physical permittivity has to be much closer to the corresponding eigenpermittivity compared with its distances from the eigenpermittivities of the other modes. The high-order modes have a minor contribution to the expansion and we can focus on a certain $l$ range when comparing these distances [@bohren2008absorption]. The resonant permittivity usually has an imaginary part that corresponds to gain. While incorporating gain in the spherical layer will bring the system to a resonance, if a real-valued permittivity will be close enough to a resonance, a similar effect is expected. The spacing between resonances and the imaginary part of the permittivity depend on the thickness of the spherical layer. A thin spherical layer will result in a large eigenpermittivity gain and widely-spaced resonances. A thick spherical layer will result in a a small imaginary part of the eigenpermittivities and more closely spaced resonances.
When the system is close to a resonance and there is a polarizable/absorbing medium at the origin, the dominant contribution to the electric field can be from the emission at the origin. From Eq. (2) it can be seen that the source location near the origin translates into $\mathbf{\tilde{E}}_n(\mathbf{r}_0)$ and the source magnitude is proportional to $\mathbf{E}_n(\mathbf{r}_0)$. We thus get that when the field is generated by the medium at the origin there is an additional factor of $\sqrt{2}$ in the effective FWHM in each dimension. This is demonstrated in Supplementary Material, Fig. 2.
An additional degeneracy arises when $r_{1},r_{2}\apprge10\lambda$ since at the $r\gg\lambda$ limit $j_{l},h_{l}^{\left(1\right)}$ have the form $$j_{l}\left(r\right)\rightarrow\frac{1}{r}\sin\left(r-\frac{l\pi}{2}\right),\,\,\,h_{l}^{\left(1\right)}\left(r\right)\rightarrow\left(-i\right)^{l+1}\frac{e^{ir}}{r}.$$ As a result the even and odd eigenvalues will be almost identical. A possible way to remove this degeneracy is to slightly change the structure so that the eigenfunctions and the eigenvalues will change. For example, the spherical layer can be capped from above (or in several places), which will also enable to easily place objects inside. Alternatively, this high degeneracy can be utilized for a strong optical response of the system (e.g., strong scattering, enhancement of spontaneous emission etc.). This degeneracy is an asymptotic degeneracy and is in addition to the $m$ degeneracy so it includes a very large number of modes. In practice, an excitation at a given frequency can excite all the even/odd TE/TM modes. Similarly, such a degeneracy is also expected for a sphere inclusion and possibly cylindrical structures. Combining spherical structures may result in an all-mode degeneracy and further enhance the response of the system. Note that the total radiated power is a sum of the contributions of all the multipoles [@jackson2012classical].
To cross validate our analysis we calculated for setups of spherical layers in vacuum the eigenmodes and $|\mathbf{E}|^2$ as a response to an excitation of a dipole and a current loop using Comsol. In Fig. 2 we present a TM $l=1$ mode for $\ensuremath{r_{1}=0.7,\,\,r_{2}=1.4\mu\mathrm{m}},\,\epsilon_1=1.5,\,\epsilon_2=1,\,\mathrm{and}\,\,\ensuremath{\omega_{\mathrm{TM}\,l=1}/2\pi=7.92781\cdot 10^{14}+7.397\cdot 10^7i},$ where $\omega_{\mathrm{TM}\,l=1}$ is an eigenfrequency. It can be seen that the focal-spot size (normalized by $\lambda$) matches the one in the analytical calculation presented in Fig. 1. Eigenmodes exist without a source, which in the eigenpermittivity formulation arises from gain in the spherical layer, similarly to a laser. In addition, $\omega_{l=1}$ is almost real and we therefore expect that at $\omega=\mathrm{Re(\omega_{l=1})},$ $\epsilon_{1l}\approx 1.5$ will be almost real. The eigenfrequencies in this case are closely spaced, which requires high precision in $\epsilon_1$ to obtain a resonance. We then considered a setup of $r_1=0.7,\,\,\,r_2=0.9 \mu \mathrm{m},\,\epsilon_2=1,\,$ $\lambda=430 \mathrm{nm}$ and an oscillating point dipole parallel to the spherical layer. We calculated $\epsilon_{\mathrm{TM}\,l=1}$ using the TM eigenvalue equation around $\epsilon_{1l}=1.5$ and substituted the result rounded to two digits after the decimal point as the physical permittivity $\epsilon_1$ in a Comsol simulation. In Fig. 3 we present $|\mathbf{E}|^2$ and $\mathbf{E}$ (arrows) for $\epsilon_1=1.75-0.7i,\,\,\mathbf{r}_0=1\hat{x} \mu \mathrm{m}, \mathbf{p}=1\hat{z} \mathrm{mA}$ in axial cross section. It can be seen that the focal-spot normalized size matches the ones in Figs. 1 and 2. Situating the dipole at any other distance will also result a focal spot at the origin, unlike imaging using a lens. In Fig. 4 we present $|\mathbf{E}|^2$ for a setup with a current loop with $\epsilon_1=1.45 -0.57i, \epsilon_2=1, r_1=1.7, r_2=2\mu \mathrm{m},\lambda=430 \mathrm{nm}, \mathbf{J}=1\hat{z}\mathrm{A/m^2},\, \mathrm{and}\, r_0=2.2\mu \mathrm{m}.$ Interestingly, the field intensity is much stronger at the origin compared to the one around the current loop. In addition, the current distribution reminds a gain medium distribution in a laser, which may mean that a laser can also be used to generate this TM $l=1$ mode. In the Supplementary Material we demonstrate focusing using a capped spherical layer with $r_2\approx 10\lambda$. This structure has a full azimuthal-angle coverage unlike focusing light using two lenses. In all the simulations we used a perfectly matched layer to account for boundary conditions (external layer).
Now we analyze the TM eigenpermittivities for a sphere inclusion in vacuum as we increase the sphere radius. Similarly to the spherical-layer setup all the odd/even eigenvalue equations coalesce when increasing the sphere radius $r_1.$ In Fig. 5 we present $\epsilon_{1l\,\mathrm{TM}}$ as a function of $r_1$ for $\lambda=430\mathrm{nm.}$ The eigenpermittivities have a negligible imaginary part (smaller than $10^{-8}$) and we therefore present only the real part. It can be seen that for $r_1>8\mu \mathrm{m}$ all the even/odd eigenvalues are practically the same. Thus, using a physical permittivity $\epsilon_1$ that is close to the odd or even eigenpermittivity, will excite all (or most) of these eigenstates, leading to a very strong response of the system (without requiring gain in this case). Note that at large sphere radii the eigenvalues are more robust to changes in the radius.
We now evaluate the enhancement of various optical phenomena due to the infinite-asymptotic degeneracy. We investigate the enhancement of spontaneous emission [@spontaneous2015] of a dipole in a sphere/spherical-layer setup when $r_1,r_2,r_{\mathrm{dipole}}\gg\lambda$ due to the infinite degeneracy. To that end, we write the expression for the density of states [@Taflove2013; @Wijnands1997; @Lagendijk1996], which is dominant in Fermi-Golden-Rule calculation [@cohen1977quantum] $$\rho_\mu=-\frac {2\omega}{\pi}\mathrm{Im}\left[G_{\mu\mu}(\mathbf{r},\mathbf{r}',\omega)\right].$$ We then evaluate the sum in the eigenfunction expansion in Eq. (2). We consider a spherical layer with a physical permittivity that is slightly above the first or second eigenpermittivity, namely $\epsilon_1>\epsilon_{11}$ or $\epsilon_1>\epsilon_{12}$ (see Fig. 5). In this situation $s_l^2/(s-s_l)$ have the same sign and approximately *the same value* for a very large number of modes (e.g., at least 20 modes for $r_1,r_2\geq40\lambda).$ We now analyze ${\tilde{E}}_{l\mu}(\mathbf{r}_{\mathrm{dipole}}){{E}}_{l\mu}(\mathbf{r}_{\mathrm{dipole}}).$ Since the dual eigenfunctions [@bergman1980theory] $$\boldsymbol{\tilde{E}}_{lm}^{\left(E\right)}\propto\nabla\times f_{l}\left(kr\right)\boldsymbol{X}^*_{lm},\,\,\, \boldsymbol{\tilde{E}}_{lm}^{\left(M\right)}\propto f_{l}\left(kr\right)\boldsymbol{X}^*_{lm},$$ we get that the phase of ${\tilde{E}}_{l\mu}(\mathbf{r}_{\mathrm{dipole}}){{E}}_{l\mu}(\mathbf{r}_{\mathrm{dipole}})$ is determined by $f_l.$ Since $ f_{l}\approx f_{l+2}$ we get approximately the same phase for all the modes whose eigenvalues coalesce. Similar arguments apply for the inner product $\langle\tilde{{\bf E}}_{n}|{\bf E}_{n}\rangle$, see Appendix A in Ref. [@bergman1980theory]. For example, the integral in the inner product of the TE modes can be performed analytically and can be shown to be invariant to $l\rightarrow l+2.$ This leads to a constructive interference in the field summation. Thus, if we have $n$ modes that have effectively the same eigenvalue, their resonance constributions will add constructively and we get approximately $n$ times enhancement in the density of states compared with a standard resonance of the same structure. Clearly, the larger the spherical-structure radius, the more eigenvalues will be effectively the same (see Fig. 5). This should be multiplied by the enhancement factor that arises from the proximity of the physical permittivity to the resonant permittivity $(\propto1/(s-s_n)$ from Eq. (2)). See Ref. [@farhi2017eigenstate] in which the modes also interfere constructively. In this reference when the physical permittivity is close to first eigenpermittivity $s_l^2/(s-s_l)$ decays upon increasing $l$ and when the physical permittivity is close to the accumulation point the field is enhanced very close to the metal-dielectric interface since the high-order modes, which have approximately the same $s_l^2/(s-s_l),$ decay spatially rapidly. Here the modes have approximately the same $s_l^2/(s-s_l)$ contribution and they all scale as $1/r$ at large distances, leading to a strong response that extends relatively far from the dielectric sphere.
We proceed to analyze the enhancement of absorption and stimulated emission induced by a dipole on itself due to the infinite degeneracy. Clearly, the enhancement of the density of states in Fermi-Golden-rule calculation [@cohen1977quantum] will be the same. Assuming that the multiple expansion for light-matter interaction holds and the dipole interaction is the dominant interaction for all the modes in the field expansion, we get that if $n$ modes are effectively on resonance and the field that is generated by the dipole is enhanced by a factor of $n,$ $\left|\left\langle \psi_{j}|H_{\mathrm{int}}|\psi_{i}\right\rangle \right|^{2}$ will scale as $n^2$ and the overall enhancement will scale as $n^3$ compared with a standard resonance. This is a very large enhancement and for $n=20$ we get an enhancement factor of 8000 (needs to be multiplied by $\propto1/(s-s_n)^3$). Note that a sphere with $r_1=20\lambda$ is of the order of a human cell for visible and infrared light and hence this phenomenon has potential use in biomedical applications such as targeting cells with light. Another potential application is omnidirectional antenna/detector, which directs its its field pattern according to the source location. The enhancement of the scattering arises from the fact that the total radiated power is a sum of the contributions of all the multipoles [@jackson2012classical]. Thus, we deduce that the total power is enhanced by a factor of $n,$ compared with a standard resonance, where $n$ is the number of modes that are effectively on resonance. This should be multiplied by the enhancement factor that arises from the proximity of the physical permittivity to the resonant permittivity $\propto 1/(s-s_n)^2.$ Similar analysis follows for the enhancement of absorption by a sphere as the absorption power is given by $\omega\cdot\mathrm{Im}(\epsilon_1)|\mathbf{E}|^2/2$ [@jackson2012classical] and many modes can be excited inside the sphere.
{width="8cm"}
{width="9cm"}
We now explore two directions to localize electric field with deep subwavelength resolution. We first present a three-body-resonance mechanism in which we slightly change the permittivity value of the spherical layer to move the system away from resonance and introduce a spherical particle that will bring the system back to resonance when located at the origin. We consider a spherical layer in a host medium that is off-resonance and close to a resonance, possibly having a dielectric material with gain. We then introduce a spherical particle that when situated at the origin results in a TM $l=1$ resonance of the three-body system for a given permittivity value of the particle, possibly a dielectric material that is different from the host-medium permittivity, and we set the physical permittivity value of the particle to be equal to this eigenpermittivity. For a different location of the particle the system will be on resonance for a different permittivity value of the particle. This setup translates location changes of the particle to changes in the eigenpermittivity, utilizing the $1/(s-s_n)$ factor to localization of the particle. We thus may achieve strong localization capability of the system - for a slight change in the location of the particle the field intensity everywhere will change significantly. To translate this idea into practical applications one can use frequencies for which the host medium that can have in general spatially-varying permittivity, is relatively uniform/transparent.
We also consider the possibility that time reversal of the field emitted in a multipole transition at the origin of an atom or molecule will spatially match the quantum transition current. It was suggested based on a classical wave equation analysis that when the time reversal of the field emitted by a point source impinges on a perfect absorber at the origin, the field pattern will have a $1/r$ scaling near the origin [@ma2018towards]. Now we turn to the quantum analysis. We first note that in the semi-classical quantum treatment in Ref. [@condon1934absolute] there is a $1/r$ scaling in the transition-rate calculation. One can think that the time reversal of an emission process is absorption, having a field with a $1/r$ dependency near the origin. In practice, emission and absorption are related to the transition between electronic or nuclear eigenstates. We can thus expect that the field will not diverge and think of a classical analogue of a dipole with a characteristic size of the average distance of the probability density function from the center of mass. Let us analyze the emission process and its time reversal. We consider a hydrogen atom for simplicity and assume that there is a transition from an eigenstate $\psi_1$ to an eigenstate $\psi_2.$ As a result of the spatial change in the probability density function electric field is emitted. We express the quantum current $\mathbf{j}=\mathbf{v}\rho=\mathbf{v}|\psi|^2=\frac{1}{2m}\left(\psi^{*}\boldsymbol{p}\psi-\psi\boldsymbol{p}\psi^{*}\right),$ where $\psi$ is the wavefunction that can transition between states and $\mathbf{v}$ is the group velocity of the particle [@cohen1977quantum]. The electric field $\mathbf{E}$ then propagates in space occupying a spherical shell. Now we time reverse the process. We assume that the field is generated on the spherical shell. The field then propagates toward the atom/molecule. We assume that when the field reaches the atom or molecule they are in the same state as when they emitted the field up to a $\pi$ phase difference in $\mathbf{v}$. Using reciprocity and treating the quantum current as classical the field near the origin will then be in the same form of the quantum current $\mathbf{j}$ that generated the field. Thus, the field pattern matches the form of the transition current and can be optimal for driving the transition. The field is thus deep subwavelength with typical size of the average distance of the density function from the center of mass. In this situation the spatial variations of the electric field are comparable to the electron/nuclear wavefunction and the spatial variations of $\mathbf{E}$ or $\mathbf{A}$ will have to be taken into account explicitly in light-atom/molecule interaction calculations. In standard light-atom/molecule interaction, the term $-\frac{q}{m}\mathbf{P}\cdot\mathbf{A}$ for the value of $\mathbf{A}$ at the atom/molecule location drives the dipole transition. However, $\mathbf{A}$ is constant [@cohen1977quantum] and not necessarily spatially overlaps optimally with the current that drives the transition. This absorption process can be complemented by stimulated emission for a field that oscillates at a frequency $\omega.$ Note that the resonant spherical layer should be tuned to this $\omega.$ This process can have unique characteristics such as strong absorption and emission, high-order multipole transitions involved etc. See for example Refs. [@Agostini79; @Corkum2007; @Ghimire2019; @Dahan2019] in which phase matching of the electric field to the electron wave function results in a stronger interaction. If this is indeed the situation it would make sense that a slight change in the position of the atom/molecule from the origin will bring the interaction to the standard multipole-expansion interaction. Hence, if this can be realized experimentally using a resonant spherical layer, it may enable to localize atoms/molecules with deep-subwavelength resolution (for scattering medium with ballistic photons). In addition, we note that this description is applicable to all the transition types (dipole, quadrupole etc.). While it is true that the spontaneous-emission rate of high-order radiation multipoles is usually slow, when it will occur for an atom/molecule at the origin the incoming time-reversed field can spatially match the transition current and drive the transition. Alternatively, transitions can be driven by an external current source. In order for the spherical layer to respond to several transitions one can utilize the infinite-asymptotic degeneracy. Note that when radiation is emitted by an atom/molecule at the origin, the spherical-layer setup generates the time-reversed field also according to the orientation, which maximizes the spatial overlap when interacting with the atom/molecule. In addition, close to a resonance the density of states given by $\rho_\mu=-\frac {2\omega}{\pi}\mathrm{Im}\left[G_{\mu\mu}(\mathbf{r},\mathbf{r}',\omega)\right]$ [@Taflove2013; @Wijnands1997; @Lagendijk1996], where $G_{\mu\mu}(\mathbf{r},\mathbf{r}',\omega)$ can be expressed as the electric field due to a dipole at the dipole location and direction in Eq. (2). Thus, when approaching a resonance the density of states and the field increase and as a result the transitions are enhanced. Hence, quantum mechanically we have enhancement in two aspects: field overlap with the transition current and increase in the density of states and electric field.
We introduced a setup of a spherical layer, that close to a resonance generates the time reversal of the atomic and molecular multipole transitions. The time reversed signal in our setup is spatially continuous and is naturally generated by a medium with a uniform permittivity.
We started by situating the spherical layer in a uniform medium, which generates a subwavelength free-space focal spot in three dimensions. The degeneracy of the excited mode is removed by incorporating currents on a plane which is perpendicular to the spherical layer. Such currents can be realized by a medium, which is polarized due to an impinging electric field or even a laser source. Interestingly, when situating an object at the origin the field emitted by the polarized medium at the focal spot excites the TM $l=1$ spherical layer mode, which reexcites the medium at the focal spot etc. This coupling can enhance the emission from the medium at the focal spot. Also, near a resonance the field becomes very strong and may enable larger penetration of ballistic photons and enhancement of the signal generated at the focal spot by the spherical layer. To image from the focal spot, one can think of collecting light from the other side of the spherical layer by means of a lens or another optical element. This signal is mostly composed of the sum of the excitation of the TM $l=1$ mode due to the sources and the polarized medium at the focal spot, which may enable to acquire also the phase in the measurement. To further minimize the effective focal-spot size techniques such as nonlinear optics, PALM or STORM [@betzig2006imaging; @moerner1989optical], and quantum imaging [@tenne2018super] can be used. In addition, the TM $l=2$ and TE $l=1$ modes have a torus shape [@farhi2017generating] and may be used to stimulate fluorescence emission at another wavelength similarly to STED [@hell1994breaking].
We then explored two directions to localize field with deep subwavelength resolution. We presented a three-body-resonance mechanism in which we slightly change the permittivity value of the spherical layer to move the system away from resonance and introduce a spherical particle that will bring the system back to resonance when located at the origin. We then situated an atom or molecule at the origin and considered the possibility that the time reversed field of a multipole transition will generate field near the origin that spatially correlates with the quantum-transition current, resulting in a much stronger interaction at the origin.
The resonant spherical shell setup differs from a spherical cavity in several aspects: 1. It enables light from outside of the spherical shell to generate field inside and vice versa. 2. There is a strong amplification of the signal. Thus, even spontaneous emission can generate substantial field at the focal spot. When the system is on resonance, the mode is generated without a source. 3. It couples to a single multipole or equivalently an atomic/molecular transition spatially and temporally.
This analysis is applicable to all wavelengths and due to its wave nature it may also apply to acoustics, in which gain materials were recently introduced [@willatzen2014acoustic], and matter waves. In addition, each spherical-layer mode has several eigenvalues and therefore there is flexibility in choosing the spherical-layer material, which may have importance for frequencies where it is more challenging to find materials that can focus waves [@graydon2018tight]. Importantly, it was shown that spherical waves (VSHs) can be generated by a single source, which may enable their practical generation, also at high frequencies where current modulation is impractical. Potential applications are high-resolution 3D imaging and precise tissue ablation. In addition, the fact that this setup has a very high Q factor may be utilized to cavity QED, entanglement, and quantum information [@raimond2001manipulating]. Finally, for spherical structures with $r_{1}\apprge10\lambda$ there are all-odd and all-even TM/TE eigenvalue degeneracies, which results in a variety of optical phenomena of the system close to one of these eigenvalues. Combining spherical structures e.g., a sphere and spherical layer(s), each with a permittivity close to one of these resonances, may even result in an all-mode resonance.
Acknowledgments {#acknowledgments .unnumbered}
---------------
M. Segev is acknowledged for the funding. G. Bartal, D. Oron, and I. Kaminer are acknowledged for the useful comments.
[10]{}
J. D. Jackson. . John Wiley & Sons, 2012.
M. Minsky, Scanning [10]{} 1988
C. B M[ü]{}ller and J. Enderlein
J. Rho, Z. Ye, Y. Xiong, X. Yin, Z. Liu, H. Choi, G. Bartal, and X. Zhang
Christoph Cremer and Thomas Cremer. , pages 31–44, 1974.
S. W. Hell and J. Wichmann, , 19(11):780–782, 1994; V. A. Okhonin. , 1374992, 1986.
E. Betzig, G. H. Patterson, R. Sougrat, O. W. Lindwasser, S. Olenych, J. S. Bonifacino, M. W. Davidson, J. Lippincott-Schwartz, and H. F. Hess, , 313(5793):1642–1645, 2006.
W. E. Moerner and L. Kador. , 62(21):2535, 1989.
O. Bitton, R Bruch, and U Leonhardt,
M. A Alonso,
G. Lerosey, J. De Rosny, A. Tourin, M. Fink Science *315* 5815 (2007).
G. Ma, X. Fan, F. Ma, J. de Rosny, P. Sheng, and M. Fink, Nature Physics *14* 6 608, 2018
J. B. Pendry,
M. Fink, IEEE transactions on ultrasonics, ferroelectrics, and frequency control *39* 5, 555-566, 1992.
G. Lerosey, J De Rosny, A Tourin, A Derode, G Montaldo, and M Fink, PRL *92* 19 193904, 2004
H. Noh, Y. Chong, A. D. Stone, and H. Cao,
L. D. Landau and E. M. Lifshitz, Quantum Mechanics: Non-relativistic theory, volume 3.
C. M Bender and S. Boettcher, PRL *80* 24, 1998.
R. El-Ganainy, K. G. Makris, D. N. Christodoulides, and Z. H. Musslimani,
C. E. Ruter, K. G. Makris, R. El-Ganainy, D. N Christodoulides, M. Segev, and D. Kip,
Z. Lin, A. Pick, M. Lon[č]{}ar, and A. W Rodriguez, PRL *117* 10, 2016
W. R. Sweeney, C. W. Hsu, S. Rotter, and A. D. Stone, Phys. Rev. Lett. *122* 093901, 2019
D.J. Bergman J. Phys. C: Solid State Phys. *12*, 1979.
A. Pick, B. Zhen, O. D. Miller, C. Hsu, F. Hernandez, A. W. Rodriguez, M. Soljacic, and S. Johnson, Optics express, *25* 11, 2017.
A. Farhi and D. J. Bergman, , 96(2):023857, 2017.
D. J. Bergman and D. Stroud, , 22(8):3527, 1980.
L. Ge, Y. D. Chong, and A. D. Stone,
E. U. Condon,
A. Farhi and D. J. Bergman, , 93(6):063844, 2016.
A. Alu, M. G. Silveirinha, A. Salandrino, and N. Engheta, Physical Review B [*75*]{}, 15 155410, 2007.
A. Farhi and D. J Bergman, Physical Review A [*96*]{} 4 043806, 2017.
C. F. Bohren and D. R. Huffman, [*Absorption and scattering of light by small particles*]{}. John Wiley & Sons, 2008.
M. S. Egglestona, K. Messera, L. Zhangb, E. Yablonovitcha, and M. C. Wua PNAS *115* (2005).
A. Taflove, A. Oskooi, and S. G. Johnson, Advances in FDTD Computational Electrodynamics: Photonics and Nanotechnology (Artech House, 2013), arXiv:1301.5366.
F. Wijnands, J. B. Pendry, F. J. Garcia-Vidal, P. M. Bell, L. M. N. Moreno, and P. J. Roberts, Opt. Quant. Elect. 29, 199-216 (1997) A. Lagendijk and B. A. V. Tiggelen, Phys. Rep. 270, 143-215 (1996)
C. Cohen-Tannoudji, B. Diu, and F. Laloe. Quantum Mechanics. Wiley (1977)
P. Agostini et al., Phys. Rev. Lett. 42, 1127 (1979),
P. B. Corkum and F. Krausz, Nat. Phys. 3, 381-387 (2007).
S. Ghimire and D. A. Reis, Nat. Phys. 15, 10-16 (2019).
R. Dahan, S. Nehemia, M. Shentcis, O. Reinhardt, Y. Adiv, K. Wang, O. Beer, Y. Kurman, X. Shi, M. H. Lynch, and I. Kaminer, arXiv:1909.00757 (2019).
R. Tenne, U. Rossman, B. Rephael, Y. Israel, A. Krupinski-Ptaszek, R. Lapkiewicz, Y. Silberberg, and D. Oron, Nat. Phot., 41566 018 0324, 2018.
W. Morten and C. Johan, Physical Review B, [*89*]{} 4 041201, 2014.
O. Graydon,
J. M. Raimond, M Brune, and S. Haroche,
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The possibility of deformation of two body quantum Calogero-Moser-Sutherland models is studied. Obtained are some necessary conditions for the singular locus of the potential function. Such locus is determined if it consists of two, three or four lines. Furthermore, a new deformation of elliptic $B_{2}$ type Calogero-Moser-Sutherland model is explicitly constructed.'
author:
- 'Kenji Taniguchi [^1]'
title: 'Deformation of two body quantum Calogero-Moser-Sutherland models'
---
Introduction {#section:introduction}
============
A Schrödinger operator $$L :=
-\sum_{i=1}^{n} \frac{\partial^{2}}{\partial x_{i}^{2}} + R(x)$$is called completely integrable if there exist $n$ algebraically independent differential operators $P_{1} = L, P_{2}, \dots, P_{n}$ which commute each other. Let $(\Sigma, W)$ be a pair of a root system and its Weyl group. The $n$-body Calogero-Moser-Sutherland (CMS) operator $$\begin{aligned}
L &=
-\sum_{i=1}^{n} \frac{\partial^{2}}{\partial x_{i}^{2}}
+ \sum_{\alpha \in \Sigma^{+}}
m_{\alpha} (m_{\alpha} + 1) |\alpha|^{2}
u(\langle \alpha, x \rangle),
\label{eq:intro-2}\end{aligned}$$ with $$\begin{aligned}
u_{\alpha}(t)
&=
\begin{cases}
1/t^{2} \quad \mbox{(rational case)},
\\
\omega^{2}/\sin^{2} \omega t, \enskip
\omega^{2}/\sinh^{2} \omega t
\quad \mbox{(trigonometric case)},
\\
\wp(t) \quad \mbox{(elliptic case)},
\end{cases}
\\
m_{w \alpha} &= m_{\alpha} \quad (\alpha \in \Sigma, w \in W), \end{aligned}$$ is an example of completely integrable operator. Here, $\wp(t)$ is the Weierstrass $\wp$ function. The constants $m_{\alpha}$ are called the *coupling constants*.
Obviously, these potential functions possess inverse square singularities along the walls of Weyl chambers. As a generalisation of CMS operator, let us consider a Schrödinger operator $$\label{eq:generalization}
L =
-\sum_{i=1}^{n} \frac{\partial^{2}}{\partial x_{i}^{2}} + R(x),
\quad \mbox{with} \quad
R(x)
=
\sum_{\alpha \in \mathcal{H}}
\frac{C_{\alpha}}
{\langle \alpha, x \rangle^{2}}
+
\widetilde{R}(x).$$ Here, $\mathcal{H}$ is a finite set of mutually non-parallel vectors in $\boldsymbol{R}^{n}$, $C_{\alpha}$ are non-zero constants and $\widetilde{R}(x)$ is real analytic at $x = 0$. We call $\mathcal{H}$ the [*singular locus*]{} of $L$ or the singular locus of $R(x)$. Note that we do not assume the symmetry of either $R(x)$ or $P$, nor do we assume $\mathcal{H}$ to be a subset of a root system.
In [@T], the author investigated what kind of differential operator $P$ commutes with $L$ in . One of the main results of [@T] is that, if $C_{\alpha} \not\in \{m(m+1) |\alpha|^{2} ;
m \in \boldsymbol{Z}\}$ for any $\alpha \in \mathcal{H}$, then the principal symbol of $P$ is invariant under the action of the group $W$ generated by reflections $r_{\alpha}$ with respect to the hyperplanes $\langle \alpha, x \rangle = 0$ ($\alpha \in \mathcal{H}$). Therefore, if $L$ possesses a non-trivial commutant, then $W$ must be a finite reflection group and $\mathcal{H}$ must be a subset of the root system of this reflection group ([@T Theorem 4.4]).
On the other hand, if some of the coupling constants are one, i.e. $C_{\alpha} = 1 \cdot 2 |\alpha|^{2}$ for some $\alpha \in \mathcal{H}$, it is known that there exist completely integrable Schrödinger operators like , but whose singular loci are not root systems but deformed ones [@CFV; @VFC].
The final objective of this research is to classify such deformed completely integrable CMS type operators and to construct such operators explicitly. But, in this paper, we do not consider general cases, but restrict our interest to the rank two rational cases. Namely, we consider what kind of operator $P$ commutes with $$\begin{aligned}
L & =
-\left(
\frac{\partial^{2}}{\partial x_{1}^{2}}
+
\frac{\partial^{2}}{\partial x_{2}^{2}}
\right)
+
\sum_{\alpha \in \mathcal{H}}
\frac{C_{\alpha}}{\langle \alpha, x \rangle^{2}},
\qquad
(\mathcal{H} \subset \boldsymbol{R}^{2},
C_{\alpha} \not=0).
\label{eq:rank two rational}\end{aligned}$$ The reason to do so is as follows: If the operator $L$ in and a differential operator $P$ commute, then, by “restricting” them to a two dimensional subspace, we obtain a two body completely integrable CMS type operator $L'$, whose potential function is a rational function. For details, see §\[section:rank two reduction\]. Therefore, two body rational integrable models are building blocks of general integrable models, and it is important to classify and construct them.
The first result of this paper is the relation between the order of $P$ and the cardinality of $\mathcal{H}$.
\[theorem:intro-1\] Assume that the Schrödinger operator $L$ in has a non-trivial commutant $P$, whose principal symbol is constant with respect to $x$. Then the order of $P$ is not less than the cardinality of $\mathcal{H}$.
Next results are the conditions for the singular locus $\mathcal{H}$ and the constants $C_{\alpha}$. For $\alpha = (\alpha_{1}, \alpha_{2}) \in \mathcal{H}$, let $\alpha^{\bot} = (-\alpha_{2}, \alpha_{1})$.
\[theorem:intro-2\] Assume that $L$ and $P$ satisfy the same condition as in Theorem \[theorem:intro-1\]. Then, for each $\alpha_{0} \in \mathcal{H}$, $$\sum_{\beta \in \mathcal{H}, \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
C_{\beta} = 0
\qquad \mbox{and} \qquad
(C_{\alpha_{0}} - 2 |\alpha_{0}|^{2})
\sum_{\beta \in \mathcal{H}, \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle |\beta|^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
C_{\beta} = 0
\label{eq:main}$$ are satisfied.
In §§\[section:construction of P2,3,4\], \[section:possibility\], we investigate what kind of $\mathcal{H}$ and $C_{\alpha}$ satisfy , when $\# \mathcal{H}= 2, 3$ or $4$. Since $\mathcal{H}$ describes the singular locus of the potential function in , the norm of each vector in $\mathcal{H}$ is not essential. Actually, if you replace $\alpha \in \mathcal{H}$ and $C_{\alpha}$ with $k \alpha$ and $k^{2} C_{\alpha}$ ($k \in \boldsymbol{R}^{\times}$) respectively, the operator $L$ and the conditions are unchanged. Therefore, we consider two singular loci $\mathcal{H}$ and $\mathcal{H}'$ to be equivalent if each vector in $\mathcal{H}'$ is a non-zero multiple of a vector in $\mathcal{H}$. Moreover, we also consider $\mathcal{H}$ and $\mathcal{H}'$ to be equivalent if $\mathcal{H}' = \{ g \alpha ; \alpha \in \mathcal{H}\}$ for some $g \in O(2)$.
\[theorem:intro-3\]
1. If $\# \mathcal{H} = 2$, then the two vectors in $\mathcal{H}$ cross at right angles. Therefore, the singular locus is of type $A_{1} \times A_{1}$.
2. If $\# \mathcal{H} = 3$, then $\mathcal{H} = \{e_{1}, \pm a e_{1} + e_{2}\}$ for some $a \not= 0$. Moreover, if $\mathcal{H}$ is not a positive system of $A_{2}$ type root system, then the coupling constants for $\pm a e_{1} + e_{2}$ must be one and there is no other completely integrable model than the one constructed in [@CFV].
3. If $\# \mathcal{H} = 4$, then $\mathcal{H} = \{e_{1}, e_{2}, \pm a e_{1} + e_{2}\}$ for some $a \not= 0$.
As stated above, deformation of CMS operators is known if some of the coupling constants are one. On the other hand, Theorem \[theorem:intro-2\] implies that there may be other deformation of a CMS operator even if no coupling constant is one. In §\[section:new deformation\], we present an example of a new deformation of the $B_{2}$ type CMS operator. The result is as follows.
Let $L$ be the Schrödinger operator defined by $$\begin{aligned}
L = &
- \left(
\frac{\partial^{2}}{\partial x_{1}^{2}}
+ \frac{\partial^{2}}{\partial x_{2}^{2}}
\right)
+ u_{1}(2 a x_{1}) + u_{2} (2 x_{2}) + u_{+}(a x_{1} + x_{2})
+ u_{-}(-a x_{1} + x_{2}),
\\
& u_{1}(t) = \frac{3}{4} (a^{2} + 1) (3 a^{-2} - 1) \wp(t),
\qquad
u_{2}(t) = \frac{3}{4} (a^{2} + 1) (3 a^{2} - 1) \wp(t),
\\
& u_{+}(t) = u_{-}(t) = 2 \cdot 3 (a^{2} + 1) \wp(t). \end{aligned}$$ Then, there exists a sixth order commutant $P$ of $L$, whose principal symbol is $$a(4 - a^{2}) \xi_{1}^{6}
+ 5a \xi_{1}^{4} \xi_{2}^{2}
+ 5 a^{-1} \xi_{1}^{2} \xi_{2}^{4}
+ a^{-1}(4 - a^{-2}) \xi_{2}^{6}.$$
**Acknowledgements.** This research was supported in part by Grant-in-Aid for Scientific Research (C)(2) No.15540183, Japan Society for the Promotion of Science.
Rank two reduction {#section:rank two reduction}
==================
To begin, we introduce some notation. Let $\{e_{1}, \dots, e_{n}\}$ be the standard basis of $\boldsymbol{R}^{n}$ and $x = (x_{1}, \dots, x_{n})$ be the corresponding coordinates. For simplicity, denote by $\partial_{x_{i}}$ the partial differential $\partial/\partial x_{i}$ and define $\partial_{x} = (\partial_{x_{1}}, \dots, \partial_{x_{n}})$. An $m_{0}$-th order differential operator $P$ is expressed as $$P=
\sum_{k = 0}^{m_{0}} P_{k},
\qquad
P_{k} =
\sum_{|p| = m_{0} - k} a_{p}(x) \partial_{x}^{p},$$ where $p = (p_{1}, \dots, p_{n}) \in \boldsymbol{N}^{n}$ is a multi-index, and $|p|$ is the length $\sum_{i} p_{i}$ of $p$. Corresponding to this operator, we introduce $$\widetilde{P}_{k}
=
\sum_{|p| = m_{0} - k} a_{p}(x) \xi^{p}
\qquad
(\xi
= (\xi_{1}, \dots, \xi_{n})),$$ and call it the *symbol* of $P_{k}$. In particular, $\widetilde{P}_{0}$ is called the [*principal symbol*]{} of $P$.
Let $\langle u, v \rangle$ be the standard inner product on $\boldsymbol{R}^{n}$, and let $|v|$ be the norm of $v$. We also use the symbol $\langle \enskip, \enskip \rangle$ to other couplings. For example, $\langle \partial_{x}, \partial_{\xi} \rangle
=
\sum_{i=1}^{n} \partial_{x_{i}} \partial_{\xi_{i}}$. For notational convenience, let $$\begin{aligned}
x_{\alpha} &:= \langle \alpha, x \rangle,
&
\xi_{\alpha} &:= \langle \alpha, \xi \rangle,
&
\partial_{x, \alpha} &:= \langle \alpha, \partial_{x} \rangle,
&
\partial_{\xi, \alpha} &:= \langle \alpha, \partial_{\xi} \rangle. \end{aligned}$$
Assume that the operator $L$ in commutes with $P$, whose principal symbol $\tilde{P}_{0}$ is constant with respect to $x$. Then, by rank one reduction, we have the following results.
[([@T Lemma 2.1])]{}\[lemma:review-1\] For any $\alpha \in \mathcal{H}$, $P$ is regular singular along the hyperplane $x_{\alpha} = 0$, i.e. $x_{\alpha}^{k} \widetilde{P}_{k}$ is analytic at $x_{\alpha} = 0$.
Put $x' = (x_{1}', x_{2}', x_{3}', \dots, x_{n}')
=
(\varepsilon^{-1} x_{1}, \varepsilon^{-1} x_{2}, x_{3}, \dots, x_{n})$ and consider the Laurent expansion of $L$ and $P$ as meromorphic functions of $\varepsilon$. By Lemma \[lemma:review-1\] and $\partial_{x} =
(\varepsilon^{-1} \partial_{x_{1}'},
\varepsilon^{-1} \partial_{x_{2}'},
\partial_{x_{3}'}, \dots, \partial_{x_{n}'},)$, we have $$\begin{aligned}
L &=
\sum_{i=-2}^{\infty} \varepsilon^{i} L(i),
&
& L(-2) =
-(\partial_{x_{1}'}^{2} + \partial_{x_{2}'}^{2})
+ \sum_{\alpha \in \mathcal{H} \cap
(\boldsymbol{R} e_{1} + \boldsymbol{R} e_{2})}
\frac{C_{\alpha}}{x_{\alpha}'{}^{2}}
\\
P &=
\sum_{i=-m_{0}}^{\infty} \varepsilon^{i} P(i),
&
& \mbox{$P(-m_{0})$ is a differential operator
on $\boldsymbol{R} e_{1} + \boldsymbol{R} e_{2}$}. \end{aligned}$$ Here, $L(i)$ and $P(i)$ are differential operators. Moreover, by Lemma \[lemma:review-1\], the principal symbol of $P(-m_{0})$ is constant with respect to $x$. This expansion implies that if $[L, P] = 0$, we have $[L(-2), P(-m_{0})] = 0$. In other words, we obtain a two body rational CMS type completely integrable system.
Therefore, we restrict our interest to this case. Namely, let $n = 2$ and define afresh $L$ and $P$ by $$\begin{aligned}
L =& -(\partial_{x_{1}}^{2} + \partial_{x_{2}}^{2}) + R(x),
\quad
R(x) =
\sum_{\alpha \in \mathcal{H}} u_{\alpha}(x_{\alpha}),
\quad
u_{\alpha}(t) := \frac{C_{\alpha}}{t^{2}},
\\
P =& \sum_{k=0}^{m_{0}} P_{k},
\quad
P_{k} = \sum_{p \in \boldsymbol{N}^{2}, |p| = m_{0} - k}
a_{p}(x) \partial_{x}^{p}, \end{aligned}$$ where $\mathcal{H} \subset \boldsymbol{R}^{2}$, $C_{\alpha} \not= 0$ for any $\alpha \in \mathcal{H}$, and $a_{p}(x)$ is a homogeneous rational function of degree $-k$ if $|p| = k$. Hereafter, we will seek conditions for $\mathcal{H}$, $C_{\alpha}$ or $P$ so that $L$ and $P$ commute. For notational convenience, we will abbreviate $u_{\alpha}^{(k)}(x_{\alpha})$ to $u_{\alpha}^{(k)}$.
Order condition for $P$ {#section:conditions for P}
=======================
By Leibniz rule, we have $$[P_{k}, R(x)]\tilde{\ }
=
\sum_{j=1}^{m_{0}-k}
\frac{1}{j!}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}^{(j)}
\partial_{\xi, \alpha}^{j}
\widetilde{P}_{k}.$$ Therefore, we have the following lemma.
\[lemma:order-1\] The condition $[L, P] = 0$ is equivalent to $$2 \langle \xi, \partial_{x} \rangle \widetilde{P}_{k}
+
\Delta \widetilde{P}_{k-1}
+
\sum_{j=1}^{k-1}
\frac{1}{j!}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}^{(j)}
\partial_{\xi, \alpha}^{j}
\widetilde{P}_{k-j-1}
= 0$$for any $k = 0, \dots, m_{0}$. Here, we set $\widetilde{P}_{-1} = \widetilde{P}_{m_{0}+1} = 0$, and we defined $\langle \xi, \partial_{x} \rangle
:=
\xi_{1} \partial_{x_{1}} + \xi_{2} \partial_{x_{2}}$, $\Delta := \partial_{x_{1}}^{2} + \partial_{x_{2}}^{2}$.
By Lemma 2.2 and Lemma 3.2 in [@T], we can easily show the following proposition.
\[proposition:review-2\] Choose $\alpha \in \mathcal{H}$ and express $\widetilde{P}_{0}$ as a polynomial in $\xi_{\alpha}$, $\xi_{\alpha^{\bot}}$; $$\widetilde{P}_{0}
= \sum_{k=0}^{m_{0}} c_{k} \xi_{\alpha}^{k}
\xi_{\alpha^{\bot}}^{m_{0}-k}.$$
1. If $C_{\alpha} \not= m(m+1)|\alpha|^{2}$ for any $m \in \boldsymbol{Z}$, then $c_{k} = 0$ for all odd $k$.
2. If $C_{\alpha} = m(m+1)|\alpha|^{2}$ for some $m \in \boldsymbol{Z}_{> 0}$, then $c_{1} = c_{3} = \dots = c_{2m-1} = 0$.
Especially, $\partial_{\xi, \alpha}
\widetilde{P}_{0}|_{\xi_{\alpha} \to 0} = 0$ for any $\alpha \in \mathcal{H}$ since $C_{\alpha} \not= 0$.
\[proposition:rotation\] Let $D_{\theta}$ be the differential operator $\xi_{2} \partial_{\xi_{1}} - \xi_{1} \partial_{\xi_{2}}$. If $c_{1}$ in the above proposition is $0$, $D_{\theta} \widetilde{P}_{0}$ is divisible by $\xi_{\alpha}$.
For any $v = (v_{1}, v_{2})$, $w = (w_{1}, w_{2}) \in \boldsymbol{R}^{2}$, we have $$\begin{aligned}
\xi_{w} \partial_{\xi, v}
-
\xi_{v} \partial_{\xi, w}
&=
(w_{1} \xi_{1} + w_{2} \xi_{2})
(v_{1} \partial_{\xi_{1}} + v_{2} \partial_{\xi_{2}})
-
(v_{1} \xi_{1} + v_{2} \xi_{2})
(w_{1} \partial_{\xi_{1}} + w_{2} \partial_{\xi_{2}})
\notag\\
&= (w_{2} v_{1} - w_{1} v_{2})
(\xi_{2} \partial_{\xi_{1}} - \xi_{1} \partial_{\xi_{2}})
\notag\\
&= \langle v^{\bot}, w \rangle D_{\theta}.
\label{eq:D_theta} \end{aligned}$$ Therefore, $$\begin{aligned}
D_{\theta} \widetilde{P}_{0}
&=
\frac{1}{|\alpha|^{2}}
(\xi_{\alpha^{\bot}} \partial_{\xi, \alpha}
-
\xi_{\alpha} \partial_{\xi, \alpha^{\bot}})
\left(
\sum_{k=0}^{m_{0}} c_{k} \xi_{\alpha}^{k}
\xi_{\alpha^{\bot}}^{m_{0}-k}
\right)
\\
&= \sum_{k=2}^{m_{0}} k c_{k}
\xi_{\alpha}^{k-1}
\xi_{\alpha^{\bot}}^{m_{0}-k+1}
-
\sum_{k=0}^{m_{0}} (m_{0} - k) c_{k}
\xi_{\alpha}^{k+1}
\xi_{\alpha^{\bot}}^{m_{0}-k-1},\end{aligned}$$ since $c_{1} = 0$. The right hand side is divisible by $\xi_{\alpha}$.
\[corollary:order condition\] If $L$ and $P$ commute, then $D_{\theta} \widetilde{P}_{0}$ is divisible by $\prod_{\alpha \in \mathcal{H}} \xi_{\alpha}$. Therefore, if $\widetilde{P}_{0}$ is not a polynomial in $\xi_{1}^{2} + \xi_{2}^{2}$, the order of $P$ is not less than the cardinality of $\mathcal{H}$.
The first part is a direct consequence of Proposition \[proposition:review-2\], \[proposition:rotation\]. Since $D_{\theta} \widetilde{P}_{0} = 0$ is equivalent to $\widetilde{P}_{0} \in \boldsymbol{C} [\xi_{1}^{2} + \xi_{2}^{2}]$, the second assertion follows from the first one.
Construction of $\widetilde{P}_{2}$, $\widetilde{P}_{3}$ and $\widetilde{P}_{4}$ {#section:construction of P2,3,4}
================================================================================
For a differential operator $Q = \sum_{p} a_{p}(x) \partial_{x}^{p}$, let ${}^{t} Q$ be the formal adjoint operator $\sum_{p}(-\partial_{x})^{p} \circ a_{p}(x)$ of $Q$. Since $L$ is formally self-adjoint, if $P$ commutes with $L$, so does ${}^{t}P$. Therefore, we may assume that $P$ is formally (skew-)self-adjoint, that is, ${}^{t}P = (-1)^{\mathrm{ord} P} P$.
\[lemma:order-2\] If $P$ is formally (skew-)self-adjoint, then $$\widetilde{P}_{2k+1}
=
\frac{1}{2}
\sum_{j=1}^{2k+1}
\frac{(-1)^{j+1}}{j!}
\langle \partial_{x}, \partial_{\xi} \rangle^{j}
\widetilde{P}_{2k+1-j}.$$
By the Leibniz rule, we have $$(-1)^{\mathrm{ord} P} \times {}^{t} P
=
\sum_{l=0}^{\mathrm{ord} P}
(-1)^{k}
\sum_{j=0}^{k}
\frac{(-1)^{j}}{j!}
\langle \partial_{x}, \partial_{\xi} \rangle^{j}
\widetilde{P}_{l-j}.$$ The lemma is easily deduced from this equation.
Since $\widetilde{P}_{0}$ is constant with respect to $x$, this lemma implies $\widetilde{P}_{1} = 0$. By Lemma \[lemma:order-1\], $\widetilde{P}_{2}$ satisfies $$2 \langle \xi, \partial_{x} \rangle \widetilde{P}_{2}
+
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}'
\partial_{\xi, \alpha} \widetilde{P}_{0}
= 0
\quad \Leftrightarrow \quad
\langle \xi, \partial_{x} \rangle
\left(
\widetilde{P}_{2}
+ \frac{1}{2} \sum_{\alpha \in \mathcal{H}}
u_{\alpha} \frac{\partial_{\xi, \alpha} \widetilde{P}_{0}}
{\xi_{\alpha}}
\right)
= 0.$$ Note that Proposition \[proposition:review-2\] implies that $\partial_{\xi, \alpha} \widetilde{P}_{0} / \xi_{\alpha}$ is a polynomial.
Let $\widetilde{Q} =
\widetilde{P}_{2}
+ (1/2) \sum_{\alpha \in \mathcal{H}}
u_{\alpha} \partial_{\xi, \alpha} \widetilde{P}_{0} / \xi_{\alpha}$. It is a polynomial in $\xi$ of degree $m_{0} - 2$ and its coefficients are homogeneous rational functions of degree $-2$. On the other hand, since $\widetilde{Q}$ satisfies $\langle \xi, \partial_{x} \rangle \widetilde{Q} = 0$, it is a function in $\xi_{1}, \xi_{2}$ and $x_{2} \xi_{1} - x_{1} \xi_{2}$. From these conditions, we can conclude $\widetilde{Q} = 0$. Therefore, $$\label{eq:P_2}
\widetilde{P}_{2}
=
- \frac{1}{2}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha} \widetilde{P}_{2}^{\alpha},
\quad
\widetilde{P}_{2}^{\alpha}
:=
\frac{\partial_{\xi, \alpha} \widetilde{P}_{0}}
{\xi_{\alpha}}
\quad \mbox{and} \quad
\widetilde{P}_{3}
= \frac{1}{2}
\langle \partial_{x}, \partial_{\xi} \rangle \widetilde{P}_{2}
=
- \frac{1}{4}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}' \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\alpha}.$$ By these formulae and Lemma \[lemma:order-1\], $\widetilde{P}_{4}$ satisfies the following equation: $$\begin{aligned}
\langle \xi, \partial_{x} \rangle
\widetilde{P}_{4}
&=
\frac{1}{8}
\sum_{\alpha \in \mathcal{H}}
|\alpha|^{2} u_{\alpha}^{(3)} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\alpha}
+ \frac{1}{4}
\sum_{\alpha, \beta \in \mathcal{H}}
u_{\alpha}' u_{\beta}
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
-\frac{1}{12}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}^{(3)} \partial_{\xi, \alpha}^{3}
\widetilde{P}_{0}
\notag\\
& =
\langle \xi, \partial_{x} \rangle
\sum_{\alpha \in \mathcal{H}}
\left(
(u_{\alpha}^{2} + |\alpha|^{2} u_{\alpha}'')
\frac{\partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha}}
{8 \xi_{\alpha}}
-
u_{\alpha}''
\frac{\partial_{\xi, \alpha}^{3} \widetilde{P}_{0}}
{12 \xi_{\alpha}}
\right)
+ \frac{1}{4}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}.
\label{eq:P4 cross}\end{aligned}$$ Now, since $u_{\alpha}(t) = C_{\alpha}/t^{2}$, we have $u_{\alpha}(t)^{2} = C_{\alpha} u_{\alpha}''(t)/6$ and $$\begin{aligned}
\sum_{\alpha \in \mathcal{H}}
& \left(
(u_{\alpha}^{2} + |\alpha|^{2} u_{\alpha}'')
\frac{\partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha}}
{8 \xi_{\alpha}}
-
u_{\alpha}''
\frac{\partial_{\xi, \alpha}^{3} \widetilde{P}_{0}}
{12 \xi_{\alpha}}
\right)
=
\frac{1}{48}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}''
\widetilde{P}_{4}^{\alpha}
\notag\\
& \mbox{where }
\widetilde{P}_{4}^{\alpha}
= \frac{(C_{\alpha} + 6 |\alpha|^{2})
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha}
- 4 \partial_{\xi, \alpha}^{3} \widetilde{P}_{0}}
{\xi_{\alpha}}.
\notag$$ By Proposition \[proposition:review-2\], the coefficients $c_{1}, c_{3}$ in the expression $\widetilde{P}_{0}
= \sum_{k=0}^{m_{0}}
c_{k} \xi_{\alpha}^{k}
\xi_{\alpha^{\bot}}^{m_{0}-k}$ satisfy $c_{1} = c_{3} (C_{\alpha} - 2 |\alpha|^{2}) = 0$. Therefore, $\widetilde{P}_{4}^{\alpha}$ is a polynomial in $\xi$, since $$\begin{aligned}
(C_{\alpha} + 6 |\alpha|^{2})
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha}
- 4 \partial_{\xi, \alpha}^{3} \widetilde{P}_{0}
&=
|\alpha|^{4}
\sum_{k=0}^{m_{0}}
k(k-2)
\{C_{\alpha} + (10 - 4k) |\alpha|^{2}\} c_{k} \xi_{\alpha}^{k-3}
\xi_{\alpha^{\bot}}^{m_{0}-k}
\\
&=
|\alpha|^{4}
\sum_{k=4}^{m_{0}}
k(k-2)
\{C_{\alpha} + (10 - 4k) |\alpha|^{2}\} c_{k} \xi_{\alpha}^{k-3}
\xi_{\alpha^{\bot}}^{m_{0}-k}, \end{aligned}$$ is divisible by $\xi_{\alpha}$. Moreover, the last term in is expressed as $$\label{eq:compatibility for P2}
\langle \xi, \partial_{x} \rangle
F(x, \xi)
=
\frac{1}{4}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta},
\quad
F(x, \xi)
:= \widetilde{P}_{4}
- \frac{1}{48}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}'' \widetilde{P}_{4}^{\alpha}.$$ Here, $F(x, \xi)$ is a polynomial in $\xi$ and a meromorphic function in $x$ with poles along $x_{\alpha} = 0$ of order at most two for each $\alpha \in \mathcal{H}$. Therefore, we have $$\begin{aligned}
& \lim_{x_{\alpha_{0}} \to 0}
(x_{\alpha_{0}} \langle \xi, \partial_{x} \rangle
+ 2 \xi_{\alpha_{0}})
(x_{\alpha_{0}} \langle \xi, \partial_{x} \rangle
+ \xi_{\alpha_{0}})
(x_{\alpha} \langle \xi, \partial_{x} \rangle)
F(x, \xi)
= 0
\notag \\
& \quad \Leftrightarrow \quad
\lim_{x_{\alpha_{0}} \to 0}
\langle \xi, \partial_{x} \rangle^{2}
x_{\alpha_{0}}^{3}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}
= 0
\label{eq:pole condition for P4}\end{aligned}$$ for each $\alpha_{0} \in \mathcal{H}$. Here, $\lim_{x_{\alpha_{0}} \to 0} G(x)$ is the limit $\lim_{x_{\alpha_{0}} \to 0}
\hat{G}(x_{\alpha_{0}}, x_{\alpha_{0}^{\bot}})$, where $\hat{G}(x_{\alpha_{0}}, x_{\alpha_{0}^{\bot}})$ is the expression of $G(x)$ as a function of $x_{\alpha_{0}}$, $x_{\alpha_{0}^{\bot}}$.
\[theorem:linear relation-1\] If $\widetilde{P}_{0}$ is not a polynomial in $\xi_{1}^{2} + \xi_{2}^{2}$, then, for each $\alpha_{0} \in \mathcal{H}$, $$\label{eq:first relation}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle }
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
C_{\beta}
= 0$$ holds.
Since $u_{\alpha}(t) = C_{\alpha}/t^{2}$ and $$\begin{aligned}
\lim_{x_{\alpha_{0}} \to 0}
u_{\beta}''
&=
\lim_{x_{\alpha_{0}} \to 0}
(-2)(-3) C_{\beta}
\left(
\frac{\langle \alpha_{0}, \beta \rangle x_{\alpha_{0}}
+
\langle \alpha_{0}^{\bot}, \beta \rangle x_{\alpha_{0}^{\bot}}}
{|\alpha_{0}|^{2}}
\right)^{-4}
=
\frac{6 C_{\beta} |\alpha_{0}|^{8}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
x_{\alpha_{0}^{\bot}}^{4}},\end{aligned}$$ we have $$\begin{aligned}
&
\lim_{x_{\alpha_{0}} \to 0}
\langle \xi, \partial_{x} \rangle^{2}
x_{\alpha_{0}}^{3}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}
= 0
\\
& \quad \Leftrightarrow \quad
\lim_{x_{\alpha_{0}} \to 0}
\langle \xi, \partial_{x} \rangle^{2}
x_{\alpha_{0}}^{3}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
C_{\alpha_{0}}
\left(-2 x_{\alpha_{0}}^{-3} u_{\beta}
\partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\beta}
+ x_{\alpha_{0}}^{-2} u_{\beta}'
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha_{0}}
\right)
= 0
\\
& \quad \Leftrightarrow \quad
\lim_{x_{\alpha_{0}} \to 0}
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\left(
-2 u_{\beta}'' \xi_{\beta}^{2}
\partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\beta}
+ 2 u_{\beta}'' \xi_{\alpha_{0}} \xi_{\beta}
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha_{0}}
\right)
= 0
\\
& \quad \Leftrightarrow \quad
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\beta}
(\xi_{\beta} \partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\beta}
-
\xi_{\alpha_{0}} \partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha_{0}})
= 0
\\
& \quad \Leftrightarrow \quad
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\beta}
(\widetilde{P}_{2}^{\alpha_{0}} - \widetilde{P}_{2}^{\beta})
= 0
\\
& \quad \Leftrightarrow \quad
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\frac{\xi_{\beta} \partial_{\xi, \alpha_{0}} \widetilde{P}_{0}
-
\xi_{\alpha_{0}} \partial_{\xi, \beta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}}}
= 0
\\
& \quad \Leftrightarrow \quad
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}}}
= 0. \end{aligned}$$ Here, we used $\xi_{\beta} \partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
=
\xi_{\beta}
\partial_{\xi, \alpha}
(\partial_{\xi, \beta} \widetilde{P}_{0} / \xi_{\beta})
=
\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
- \langle \alpha, \beta \rangle \widetilde{P}_{2}^{\beta}$ and .
\[cor:2-sheets\] Under the assumption of Theorem \[theorem:linear relation-1\], if $\mathcal{H}$ consists of two vectors, then they cross at right angles and the singular locus is of type $A_{1} \times A_{1}$.
Let $\mathcal{H} = \{\alpha, \beta\}$. Then by Theorem \[theorem:linear relation-1\], we have $\langle \alpha, \beta \rangle C_{\beta} /
\langle \alpha^{\bot}, \beta \rangle^{3}
= 0$. Since $C_{\beta} \not= 0$, this implies $\langle \alpha, \beta \rangle = 0$.
Next, let us consider the last term in . Let $N := \# \mathcal{H}$ and $$\begin{aligned}
d_{\alpha, \beta, \gamma}
& :=
\frac{\langle \beta, \gamma \rangle}
{\langle \beta^{\bot}, \gamma \rangle^{3} C_{\alpha}}
+
\frac{\langle \gamma, \alpha\rangle}
{\langle \gamma^{\bot}, \alpha \rangle^{3} C_{\beta}}
+
\frac{\langle \alpha, \beta \rangle}
{\langle \alpha^{\bot}, \beta \rangle^{3} C_{\gamma}}. \end{aligned}$$
\[lemma:d\_abc\] The constant $d_{\alpha, \beta, \gamma}$ is skew-symmetric with respect to $\alpha$, $\beta$, $\gamma$ and satisfies $$\label{eq:d_abc}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha, \beta}
C_{\gamma} d_{\alpha, \beta, \gamma}
= N
\frac{\langle \alpha, \beta \rangle}{
\langle \alpha^{\bot}, \beta \rangle^{3}}.$$
The first statement follows from $\langle \alpha^{\bot}, \beta \rangle
= - \langle \beta^{\bot}, \alpha \rangle$.
The second statement is a consequence of : $$\begin{aligned}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha, \beta}
C_{\gamma} d_{\alpha, \beta, \gamma}
&=
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha, \beta}
\left(
\frac{1}{C_{\alpha}}
\frac{\langle \beta, \gamma \rangle C_{\gamma}}
{\langle \beta^{\bot}, \gamma \rangle^{3}}
+
\frac{1}{C_{\beta}}
\frac{\langle \gamma, \alpha\rangle C_{\gamma}}
{\langle \gamma^{\bot}, \alpha \rangle^{3}}
+
\frac{\langle \alpha, \beta \rangle}
{\langle \alpha^{\bot}, \beta \rangle^{3}}
\right)
\\&=
-\frac{1}{C_{\alpha}}
\frac{\langle \beta, \alpha \rangle C_{\alpha}}
{\langle \beta^{\bot}, \alpha \rangle^{3}}
-
\frac{1}{C_{\beta}}
\frac{\langle \beta, \alpha \rangle C_{\beta}}
{\langle \beta^{\bot}, \alpha \rangle^{3}}
+ (N-2)
\frac{\langle \alpha, \beta \rangle}
{\langle \alpha^{\bot}, \beta \rangle^{3}}
\\
&=
N
\frac{\langle \alpha, \beta \rangle}
{\langle \alpha^{\bot}, \beta \rangle^{3}}. \end{aligned}$$
Since $u_{\alpha}(t) = C_{\alpha}/t^{2}$ and $$\label{eq:triangle}
\langle \beta^{\bot}, \gamma \rangle \alpha
+
\langle \gamma^{\bot}, \alpha \rangle \beta
+
\langle \alpha^{\bot}, \beta \rangle \gamma
= 0$$ for any $\alpha$, $\beta$, $\gamma \in \mathcal{H}$, we have $$\label{eq:pe-relation-1}
\begin{vmatrix}
\langle \beta^{\bot}, \gamma \rangle^{3} C_{\alpha}
&
\langle \gamma^{\bot}, \alpha \rangle^{3} C_{\beta}
&
\langle \alpha^{\bot}, \beta \rangle^{3} C_{\gamma}
\\
\langle \beta^{\bot}, \gamma \rangle u_{\alpha}
&
\langle \gamma^{\bot}, \alpha \rangle u_{\beta}
&
\langle \alpha^{\bot}, \beta \rangle u_{\gamma}
\\
u_{\alpha}'
&
u_{\beta}'
&
u_{\gamma}'
\end{vmatrix}
= 0.$$ By Corollary \[corollary:order condition\], $D_{\theta} \widetilde{P}_{0} / \xi_{\alpha} \xi_{\beta} \xi_{\gamma}$ is a polynomial in $\xi$ and it is symmetric with respect to $\alpha,
\beta, \gamma$. Then by , we have $$\label{eq:pe-relation-2}
\sum_{\alpha, \beta, \gamma \in \mathcal{H} \atop
\alpha \not= \beta \not= \gamma \not= \alpha}
\langle \alpha^{\bot}, \beta \rangle^{3}
\langle \gamma^{\bot}, \alpha \rangle
C_{\gamma} d_{\alpha, \beta, \gamma}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}
u_{\alpha}' u_{\beta}
= 0.$$
\[lemma:P4ab another\] For $\alpha, \beta \in \mathcal{H}$, let $$\begin{aligned}
\widetilde{P}_{4}^{\alpha, \beta}
&=
-\frac{
\langle \gamma^{\bot}, \alpha \rangle
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}}
{\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}}
+
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha, \beta, \gamma}
\langle \delta^{\bot}, \gamma \rangle C_{\delta}
d_{\alpha, \beta, \delta}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma} \xi_{\delta}},
\label{eq:P4ab} \end{aligned}$$ where $\gamma$ is any vector in $\mathcal{H}$ other than $\alpha, \beta$. Then, $\widetilde{P}_{4}^{\alpha, \beta}$ is a polynomial in $\xi$ and satisfies $$\label{eq:P4ab another}
\xi_{\alpha} \widetilde{P}_{4}^{\alpha, \beta}
=
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
+
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha, \beta}
\langle \delta^{\bot}, \alpha \rangle C_{\delta}
d_{\alpha, \beta, \delta}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\delta}}.$$
By Corollary \[corollary:order condition\], $D_{\theta} \widetilde{P}_{0}/
\xi_{\alpha} \xi_{\beta} \xi_{\gamma} \xi_{\delta}$ is a polynomial. Therefore, to prove $\widetilde{P}_{4}^{\alpha, \beta}$ being a polynomial, we have only to show that $\langle \gamma^{\bot}, \alpha \rangle
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}$ is divisible by $\xi_{\gamma}$. Since $$\begin{aligned}
&
\xi_{\alpha} \xi_{\beta}
(\langle \gamma^{\bot}, \alpha \rangle
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta})
+ \langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}
\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
\\
&=
\langle \gamma^{\bot}, \alpha \rangle \xi_{\beta}
(\partial_{\xi, \beta} \partial_{\xi, \alpha} \widetilde{P}_{0}
-
\langle \alpha, \beta \rangle \widetilde{P}_{2}^{\alpha})
+
\langle \beta^{\bot}, \gamma \rangle \xi_{\alpha}
(\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
-
\langle \alpha, \beta \rangle \widetilde{P}_{2}^{\beta})
+ \langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}
\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
\\
&=
(\langle \gamma^{\bot}, \alpha \rangle \xi_{\beta}
+ \langle \beta^{\bot}, \gamma \rangle \xi_{\alpha}
+ \langle \alpha^{\bot}, \beta \rangle \xi_{\gamma})
\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
- \langle \alpha, \beta \rangle
(\langle \gamma^{\bot}, \alpha \rangle \xi_{\beta}
\widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle \xi_{\alpha}
\widetilde{P}_{2}^{\beta})
\\
&=
\langle \alpha, \beta \rangle
\{
(\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}
+ \langle \beta^{\bot}, \gamma \rangle \xi_{\alpha})
\widetilde{P}_{2}^{\alpha}
+
(\langle \gamma^{\bot}, \alpha \rangle \xi_{\beta}
+ \langle \alpha^{\bot}, \beta \rangle \xi_{\gamma})
\widetilde{P}_{2}^{\beta}
\}
\\
&=
\langle \alpha, \beta \rangle
\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}
(\widetilde{P}_{2}^{\alpha} + \widetilde{P}_{2}^{\beta})
+
\langle \alpha, \beta \rangle
(\langle \beta^{\bot}, \gamma \rangle \partial_{\xi, \alpha}
+ \langle \gamma^{\bot}, \alpha \rangle \partial_{\xi, \beta})
\widetilde{P}_{0}
\\
&=
\langle \alpha, \beta \rangle
\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}
(\widetilde{P}_{2}^{\alpha} + \widetilde{P}_{2}^{\beta}
- \widetilde{P}_{2}^{\gamma}), \end{aligned}$$ we have $$\langle \gamma^{\bot}, \alpha \rangle
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
= \frac{\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}}
{\xi_{\alpha} \xi_{\beta}}
\{-\partial_{\xi, \alpha} \partial_{\xi, \beta} \widetilde{P}_{0}
+ \langle \alpha, \beta \rangle
(\widetilde{P}_{2}^{\alpha} + \widetilde{P}_{2}^{\beta}
- \widetilde{P}_{2}^{\gamma})\}.$$ By the uniqueness of factorization, this is divisible by $\xi_{\gamma}$.
Let us prove . Firstly, we have $$\begin{aligned}
- \xi_{\alpha}
&
\frac{
\langle \gamma^{\bot}, \alpha \rangle
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
\langle \beta^{\bot}, \gamma \rangle
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}}
{\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}}
- \partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
\\
&=
-\frac{
\langle \gamma^{\bot}, \alpha \rangle \xi_{\alpha}
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+
(\langle \beta^{\bot}, \gamma \rangle \xi_{\alpha}
+
\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma})
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}}
{\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}}
\\
&=
-\frac{
\langle \gamma^{\bot}, \alpha \rangle
(\xi_{\alpha}
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
- \xi_{\beta}
\partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta})}
{\langle \alpha^{\bot}, \beta \rangle \xi_{\gamma}}
\\
&=
\langle \gamma^{\bot}, \alpha \rangle
\langle \alpha, \beta \rangle
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}. \end{aligned}$$ Here, we used and calculated as in the proof of Theorem \[theorem:linear relation-1\].
Secondly, since $$\frac{\langle \delta^{\bot}, \gamma \rangle}
{\xi_{\beta} \xi_{\gamma} \xi_{\delta}}
-
\frac{\langle \delta^{\bot}, \alpha \rangle}
{\xi_{\alpha} \xi_{\beta} \xi_{\delta}}
=
\frac{\langle \delta^{\bot}, \gamma \rangle \xi_{\alpha}
+
\langle \alpha^{\bot}, \delta \rangle \xi_{\gamma}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma} \xi_{\delta}}
=
- \frac{\langle \gamma^{\bot}, \alpha \rangle}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}},$$ we have $$\begin{aligned}
&
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\left(
\xi_{\alpha}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha, \beta, \gamma}
\langle \delta^{\bot}, \gamma \rangle C_{\delta}
d_{\alpha, \beta, \delta}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma} \xi_{\delta}}
-
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha, \beta}
\langle \delta^{\bot}, \alpha \rangle C_{\delta}
d_{\alpha, \beta, \delta}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\delta}}
\right)
\\
&=
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\left(
- \frac{\langle \gamma^{\bot}, \alpha \rangle
D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha, \beta, \gamma}
C_{\delta} d_{\alpha, \beta, \delta}
-
\langle \gamma^{\bot}, \alpha \rangle C_{\gamma}
d_{\alpha, \beta, \gamma}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}
\right)
\\
&=
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\left(
- \frac{\langle \gamma^{\bot}, \alpha \rangle
D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}
\right)
N
\frac{\langle \alpha, \beta \rangle}{
\langle \alpha^{\bot}, \beta \rangle^{3}}
\\
&=
-\langle \gamma^{\bot}, \alpha \rangle
\langle \alpha, \beta \rangle
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\gamma}}, \end{aligned}$$ by . Therefore, is proved.
\[proposition:P4ab\] $\widetilde{P}_{4}^{\alpha, \beta}$ satisfies $$\label{eq:determination of P4ab}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}
=
\langle \xi, \partial_{x} \rangle
\left(
\sum_{\{\alpha, \beta\} \subset \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha} u_{\beta} \widetilde{P}_{4}^{\alpha, \beta}
\right).$$
By Lemma \[lemma:P4ab another\] and , we have $$\begin{aligned}
\langle \xi, \partial_{x} \rangle
&
\left(
\sum_{\{\alpha, \beta\} \subset \mathcal{H}
\atop \alpha \not= \beta}
u_{\alpha} u_{\beta} \widetilde{P}_{4}^{\alpha, \beta}
\right)
-
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}
\\
&=
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta}
\left(\xi_{\alpha} \widetilde{P}_{4}^{\alpha, \beta}
- \partial_{\xi, \alpha} \widetilde{P}_{2}^{\beta}
\right)
\\
&=
\sum_{\alpha, \beta, \delta \in \mathcal{H}
\atop \alpha \not= \beta \not= \delta \not= \alpha}
u_{\alpha}' u_{\beta}
\frac{\langle \alpha^{\bot}, \beta \rangle^{3}}{N}
\langle \delta^{\bot}, \alpha \rangle C_{\delta}
d_{\alpha, \beta, \delta}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha} \xi_{\beta} \xi_{\delta}}
\\
&= 0. \end{aligned}$$
Putting together these results, we obtain the explicit expression of $\widetilde{P}_{4}$.
\[proposition:explicit form of P4\] $\widetilde{P}_{4}$ is expressed as $$\widetilde{P}_{4}
=
\frac{1}{48}
\sum_{\alpha \in \mathcal{H}}
u_{\alpha}'' \widetilde{P}_{4}^{\alpha}
+
\frac{1}{4}
\sum_{\{\alpha, \beta \} \subset \mathcal{H}
\atop \alpha \not= \beta}
u_{\alpha} u_{\beta} \widetilde{P}_{4}^{\alpha, \beta}.$$
Moreover, we can write down $\widetilde{P}_{5}$ explicitly by using Lemma \[lemma:order-2\].
Second condition for $\mathcal{H}$ and $C_{\alpha}$ {#section:second condition}
===================================================
In the previous section, we obtained a condition for $\mathcal{H}$ and $C_{\alpha}$ by investigating the pole of $\widetilde{P}_{4}$ at $x_{\alpha_{0}} = 0$. In this section, we investigate the pole of $\widetilde{P}_{6}$ at $x_{\alpha_{0}} = 0$ and obtain another condition for $\mathcal{H}$ and $C_{\alpha}$.
\[lemma:basic equation for P6\] Let $$\widetilde{P}_{6}^{\alpha}
:=
\frac{1}{\xi_{\alpha}}
\{
24 \partial_{\xi, \alpha}^{5} \widetilde{P}_{0}
+ (C_{\alpha} - 90 |\alpha|^{2})
\partial_{\xi, \alpha}^{3} \widetilde{P}_{2}^{\alpha}
+ C_{\alpha} \partial_{\xi, \alpha}
\widetilde{P}_{4}^{\alpha}
\}.$$Then, $\widetilde{P}_{6}^{\alpha}$ is a polynomial in $\xi$ and $\widetilde{P}_{6}$ satisfies the following equation: $$\begin{aligned}
\langle \xi, \partial_{x} \rangle \widetilde{P}_{6}
&=
\langle \xi, \partial_{x} \rangle
\left(
- \frac{1}{8} \langle \partial_{\xi}, \partial_{x} \rangle^{2}
\widetilde{P}_{4}
+ \frac{1}{2} \langle \partial_{\xi}, \partial_{x} \rangle
\widetilde{P}_{5}
- \frac{1}{5760}
\sum_{\alpha \in \mathcal{H}} u_{\alpha}^{(4)}
\widetilde{P}_{6}^{\alpha}
\right)
\label{eq:compatibility for P4}
\\
& \quad
+ \frac{1}{96}
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
\{
u_{\alpha}^{(3)} u_{\beta}
(\partial_{\xi, \alpha}^{3} \widetilde{P}_{2}^{\beta}
- C_{\alpha} \partial_{\xi, \alpha}
\widetilde{P}_{4}^{\alpha, \beta})
-
u_{\alpha}'' u_{\beta}'
(3 \partial_{\xi, \alpha}^{2} \partial_{\xi, \beta}
\widetilde{P}_{2}^{\alpha}
+
\partial_{\xi, \beta} \widetilde{P}_{4}^{\alpha})
\}
\notag\\
& \quad
-\frac{1}{8}
\sum_{\alpha \in \mathcal{H}}
\sum_{\{\beta, \gamma \} \subset \mathcal{H} \atop
\alpha \not= \beta \not= \gamma \not= \alpha}
u_{\alpha}' u_{\beta} u_{\gamma}
\partial_{\xi, \alpha} \widetilde{P}_{4}^{\beta, \gamma}.
\notag\end{aligned}$$
Let us consider the expression $\widetilde{P}_{0}
= \sum_{k=0}^{m_{0}}
c_{k} \xi_{\alpha}^{k}
\xi_{\alpha^{\bot}}^{m_{0}-k}$. For this expression, we have $$\widetilde{P}_{6}^{\alpha}
=
15 |\alpha|^{6}
c_{5} (C_{\alpha} - 2 |\alpha|^{2}) (C_{\alpha} - 6 |\alpha|^{2})
\frac{\xi_{\alpha^{\bot}}^{m_{0} - 5}}{\xi_{\alpha}}
+ \mbox{(a polynomial in $\xi$)}.$$ Since $c_{5}$ satisfies $c_{5} (C_{\alpha} - 2 |\alpha|^{2})
(C_{\alpha} - 6 |\alpha|^{2}) = 0$ by Proposition \[proposition:review-2\], $\widetilde{P}_{6}^{\alpha}$ is a polynomial in $\xi$. The second assertion is a consequence of Lemma \[lemma:order-1\], , Proposition \[proposition:P4ab\], $\Delta =
[
\langle \partial_{x}, \partial_{\xi} \rangle,
\langle \xi, \partial_{x} \rangle]$ and $\langle \partial_{x}, \partial_{\xi} \rangle \Delta
=
(1/2)
[
\langle \partial_{x}, \partial_{\xi} \rangle^{2},
\langle \xi, \partial_{x} \rangle]$.
We obtained by investigating the pole of $\widetilde{P}_{4}$ at $x_{\alpha_{0}} = 0$. In the same way, we obtain $$\begin{aligned}
\lim_{x_{\alpha_{0}} \to 0}
&
\langle \xi, \partial_{x} \rangle^{4} x_{\alpha_{0}}^{5}
\left(
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
\left\{
u_{\alpha}^{(3)} u_{\beta}
(\partial_{\xi, \alpha}^{3} \widetilde{P}_{2}^{\beta}
- C_{\alpha} \partial_{\xi, \alpha}
\widetilde{P}_{4}^{\alpha, \beta})
-
u_{\alpha}'' u_{\beta}'
(3 \partial_{\xi, \alpha}^{2} \partial_{\xi, \beta}
\widetilde{P}_{2}^{\alpha}
+
\partial_{\xi, \beta} \widetilde{P}_{4}^{\alpha})
\right\}
\right.
\\
& \hspace{40mm}
\left.
-12
\sum_{\alpha \in \mathcal{H}}
\sum_{\{\beta, \gamma \} \subset \mathcal{H} \atop
\alpha \not= \beta \not= \gamma \not= \alpha}
u_{\alpha}' u_{\beta} u_{\gamma}
\partial_{\xi, \alpha} \widetilde{P}_{4}^{\beta, \gamma}
\right)
\\
= 0& \end{aligned}$$ for each $\alpha_{0} \in \mathcal{H}$. Since $u_{\alpha} = C_{\alpha}/x_{\alpha}^{2}$, this is equivalent to $$S_{1} + S_{2} + S_{3} = 0,$$ where $$\begin{aligned}
& S_{1} :=
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\beta}^{3}
\left\{
\xi_{\beta}
(\partial_{\xi, \alpha_{0}}^{3} \widetilde{P}_{2}^{\beta}
- C_{\alpha_{0}}
\partial_{\xi, \alpha_{0}} \widetilde{P}_{4}^{\alpha_{0}, \beta})
+ \xi_{\alpha_{0}}
(3 \partial_{\xi, \alpha_{0}}^{2} \partial_{\xi, \beta}
\widetilde{P}_{2}^{\alpha_{0}}
+
\partial_{\xi, \beta} \widetilde{P}_{4}^{\alpha_{0}})
\right\},
\\
& S_{2} :=
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\left\{
\xi_{\alpha_{0}}
(\partial_{\xi, \beta}^{3} \widetilde{P}_{2}^{\alpha_{0}}
- C_{\beta}
\partial_{\xi, \beta} \widetilde{P}_{4}^{\alpha_{0}, \beta})
+ \xi_{\beta}
(3 \partial_{\xi, \beta}^{2} \partial_{\xi, \alpha_{0}}
\widetilde{P}_{2}^{\beta}
+
\partial_{\xi, \alpha_{0}} \widetilde{P}_{4}^{\beta})
\right\}
\\ \intertext{and}
& S_{3} :=
\frac{1}{5}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \xi_{\gamma})
(\xi_{\beta} \partial_{\xi, \alpha_{0}}
\widetilde{P}_{4}^{\beta, \gamma}
-
\xi_{\alpha_{0}} \partial_{\xi, \beta}
\widetilde{P}_{4}^{\alpha_{0}, \gamma}). \end{aligned}$$
\[theorem:linear relation-2\] If $\widetilde{P}_{0}$ is not a polynomial in $\xi_{1}^{2} + \xi_{2}^{2}$, then, for each $\alpha_{0} \in \mathcal{H}$, $$\label{eq:second relation}
(C_{\alpha_{0}} - 2 |\alpha_{0}|^{2})
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle |\beta|^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
C_{\beta}
= 0$$ holds.
<span style="font-variant:small-caps;">Proof.</span> The proof is divided into many lemmas. Let $$\widetilde{Q}_{\alpha, \beta, \gamma}
:= D_{\theta} \widetilde{P}_{0}/\xi_{\alpha} \xi_{\beta} \xi_{\gamma}
\quad \mbox{and} \quad
\widetilde{Q}_{4}^{\alpha, \beta}
:= \sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha, \beta}
\langle \gamma^{\bot}, \alpha \rangle
C_{\gamma} d_{\alpha, \beta, \gamma}
\widetilde{Q}_{\alpha, \beta, \gamma}.$$ By , we have $$\begin{aligned}
&
\xi_{\beta} \partial_{\xi, \alpha_{0}}
\widetilde{P}_{4}^{\beta, \gamma}
-
\xi_{\alpha_{0}} \partial_{\xi, \beta}
\widetilde{P}_{4}^{\alpha_{0}, \gamma}
\\
&=
\partial_{\xi, \alpha_{0}}
(\xi_{\beta}
\widetilde{P}_{4}^{\beta, \gamma})
-
\partial_{\xi, \beta}
(\xi_{\alpha_{0}}
\widetilde{P}_{4}^{\alpha_{0}, \gamma})
- \langle \alpha_{0}, \beta \rangle
(\widetilde{P}_{4}^{\beta, \gamma}
- \widetilde{P}_{4}^{\alpha_{0}, \gamma})
\\
&=
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}}{N}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \beta, \gamma}
\langle \delta^{\bot}, \beta \rangle
C_{\delta} d_{\beta, \gamma, \delta}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}
-
\frac{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}{N}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
+ \langle \alpha_{0}, \beta \rangle
(\widetilde{P}_{4}^{\alpha_{0}, \gamma}
- \widetilde{P}_{4}^{\beta, \gamma}). \end{aligned}$$ Therefore, we put $$S_{3} = S_{4} + S_{5} + S_{6} + S_{7} + S_{8},$$ where $$\begin{aligned}
S_{4} :&=
\frac{3}{5}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
(\widetilde{P}_{4}^{\alpha_{0}, \gamma}
- \widetilde{P}_{4}^{\beta, \gamma}),
\\
S_{5} :&=
\frac{2}{5}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\gamma}
(\widetilde{P}_{4}^{\alpha_{0}, \gamma}
- \widetilde{P}_{4}^{\beta, \gamma})
\\
&=
\frac{2}{5}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
(\widetilde{P}_{4}^{\alpha_{0}, \beta}
- \widetilde{P}_{4}^{\beta, \gamma}),
\\
S_{6} :&=
\frac{1}{5N}
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \xi_{\gamma})
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}
\\
S_{7} :&=
\frac{1}{5N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \alpha_{0}^{\bot}, \beta \rangle
C_{\beta} C_{\gamma} C_{\alpha_{0}} d_{\beta, \gamma, \alpha_{0}}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \xi_{\gamma})
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \alpha_{0}}
\\
&=
\frac{1}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
C_{\alpha_{0}} C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma},
\\
S_{8} :&=
-\frac{1}{5N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}}^{2}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \xi_{\gamma})
\partial_{\xi, \beta} \widetilde{Q}_{4}^{\alpha_{0}, \gamma}. \end{aligned}$$
\[lemma:S6\] $S_{6} = 0$.
For $\eta = (\eta_{1}, \eta_{2})$, let $$\bar{S}_{6}(\eta)
=
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \eta_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \eta_{\gamma})
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}.$$ We prove $\bar{S}_{6}(\eta) = 0$. If so, we have $S_{6} = 0$, since $S_{6} = \bar{S}_{6}(\xi)/5N$.
For an ordered triple $\{\beta, \gamma, \delta\} \subset
\mathcal{H} \setminus \{\alpha_{0}\}$, let $$\begin{aligned}
A_{\beta}
&:=
\langle \gamma^{\bot}, \delta \rangle
\langle \alpha_{0}^{\bot}, \beta \rangle,
&
A_{\gamma}
&:=
\langle \delta^{\bot}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle,
&
A_{\delta}
&:=
\langle \beta^{\bot}, \gamma \rangle
\langle \alpha_{0}^{\bot}, \delta \rangle. \end{aligned}$$ Note that they satisfy $\sigma(A_{\varepsilon})
= (\mathrm{sgn} \sigma) A_{\sigma(\varepsilon)}$ ($\varepsilon \in \{\beta, \gamma, \delta\}$) for a permutation $\sigma$ of $\{\beta, \gamma, \delta\}$, and $$A_{\beta} + A_{\gamma} + A_{\delta} = 0$$ by . Hereafter, we denote by ${\sum_{\beta, \gamma, \delta}}' F(b, c, d)$ the sum $F(\beta, \gamma, \delta) + F(\gamma, \delta, \beta)
+ F(\delta, \beta, \gamma)$.
Firstly, we have $$\begin{aligned}
\bar{S}_{6}(\eta)
& =
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \alpha_{0}^{\bot}, \delta \rangle^{4}
C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{4}
\langle \alpha_{0}^{\bot}, \delta \rangle^{4}
}
\xi_{\alpha_{0}}^{2}
\\
& \hspace{40mm}
\times
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \eta_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \eta_{\gamma})
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}
\\
& =
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \alpha_{0}^{\bot}, \delta \rangle
A_{\gamma} A_{\delta}^{3} \xi_{\alpha_{0}}^{2}
C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{4}
\langle \alpha_{0}^{\bot}, \delta \rangle^{4}
}
(3\langle \alpha_{0}^{\bot}, \gamma \rangle \eta_{\beta}
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle \eta_{\gamma})
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}
\\
& =
\sum_{
\mbox{\tiny
$\begin{matrix}
\{\beta, \gamma, \delta\} \subset \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{4}
\langle \alpha_{0}^{\bot}, \delta \rangle^{4}}
(3 S_{9} + 2 S_{10})
\xi_{\alpha_{0}}^{2}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\beta, \gamma, \delta}, \end{aligned}$$ where $$\begin{aligned}
S_{9}
:= &
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, d \rangle A_{d}^{3}
(\langle \alpha_{0}^{\bot}, c \rangle A_{c} \eta_{b}
- \langle \alpha_{0}^{\bot}, b \rangle A_{b} \eta_{c}),
\\
S_{10}
:= &
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, d \rangle A_{d}^{3}
(\langle \alpha_{0}^{\bot}, b \rangle A_{c} \eta_{c}
- \langle \alpha_{0}^{\bot}, c \rangle A_{b} \eta_{b}). \end{aligned}$$ Since $$\begin{aligned}
S_{9}
= &
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, b \rangle
\langle \alpha_{0}^{\bot}, c \rangle
A_{b} A_{c} \eta_{d} (A_{c}^{2} - A_{b}^{2})
&
&
=
A_{\beta} A_{\gamma} A_{\delta}
{\sum_{\beta, \gamma, \delta}}'
(A_{b} - A_{c})
\langle \alpha_{0}^{\bot}, b \rangle
\langle \alpha_{0}^{\bot}, c \rangle
\eta_{d}
\\
= &
A_{\beta} A_{\gamma} A_{\delta}
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, b \rangle A_{b}
(\langle \alpha_{0}^{\bot}, c \rangle \eta_{d}
-
\langle \alpha_{0}^{\bot}, d \rangle \eta_{c})
&
&=
- A_{\beta} A_{\gamma} A_{\delta}
\eta_{\alpha_{0}}
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, b \rangle
\langle c^{\bot}, d \rangle
A_{b}
\\
=&
- A_{\beta} A_{\gamma} A_{\delta}
(A_{\beta}^{2} + A_{\gamma}^{2} + A_{\delta}^{2}) \eta_{\alpha_{0}}
&
&
=
2 A_{\beta} A_{\gamma} A_{\delta}
(A_{\beta} A_{\gamma} + A_{\gamma} A_{\delta} + A_{\delta} A_{\beta})
\eta_{\alpha_{0}} \end{aligned}$$ and $$\begin{aligned}
S_{10} =&
{\sum_{\beta, \gamma, \delta}}'
\langle \alpha_{0}^{\bot}, d \rangle A_{d}^{3}
\{
\langle \alpha_{0}^{\bot}, b \rangle
(A_{c} - A_{b}) \eta_{c}
+
A_{b}
(\langle \alpha_{0}^{\bot}, b \rangle \eta_{c}
-
\langle \alpha_{0}^{\bot}, c \rangle \eta_{b})
\}
\\
=&
{\sum_{\beta, \gamma, \delta}}'
\{
\langle \alpha_{0}^{\bot}, d \rangle
\langle \alpha_{0}^{\bot}, b \rangle
A_{d}^{2} (A_{b}^{2} - A_{c}^{2}) \eta_{c}
-
\langle \alpha_{0}^{\bot}, d \rangle
\langle b^{\bot}, c \rangle A_{d}^{3} A_{b} \eta_{\alpha_{0}}
\}
\\
=&
{\sum_{\beta, \gamma, \delta}}'
\{
A_{b}^{2} A_{c}^{2}
\langle \alpha_{0}^{\bot}, c \rangle
(\langle \alpha_{0}^{\bot}, b \rangle \eta_{d}
-
\langle \alpha_{0}^{\bot}, d \rangle \eta_{b})
- A_{d}^{4} A_{b} \eta_{\alpha_{0}}
\}
\\
=&
\eta_{\alpha_{0}}
{\sum_{\beta, \gamma, \delta}}'
(A_{b}^{2} A_{c}^{3} - A_{b}^{4} A_{c})
\\
=&
\eta_{\alpha_{0}}
{\sum_{\beta, \gamma, \delta}}'
A_{b}^{2} A_{c} (A_{c}^{2} - A_{b}^{2})
\\
=&
\eta_{\alpha_{0}}
{\sum_{\beta, \gamma, \delta}}'
A_{b}^{2} A_{c} A_{d} (A_{b} - A_{c})
\\
=&
A_{\beta} A_{\gamma} A_{\delta} \eta_{\alpha_{0}}
{\sum_{\beta, \gamma, \delta}}'
(A_{b}^{2} - A_{b} A_{c})
\\
=&
-3 A_{\beta} A_{\gamma} A_{\delta}
(A_{\beta} A_{\gamma} + A_{\gamma} A_{\delta} + A_{\delta} A_{\beta})
\eta_{\alpha_{0}}, \end{aligned}$$ we have $3 S_{9} + 2 S_{10} = 0$ and $\bar{S}_{6}(\eta) = 0$.
\[lemma:S4\] $\displaystyle S_{4} = -\frac{3}{5} (S_{11} + S_{12})$, where $$\begin{aligned}
S_{11}
&:=
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\alpha_{0}}^{2}
(\xi_{\beta} \widetilde{P}_{4}^{\alpha_{0}, \beta}
- \partial_{\xi, \beta} \widetilde{P}_{2}^{\beta}),
&
S_{12} &:=
\frac{1}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2} \widetilde{Q}_{4}^{\beta, \gamma}. \end{aligned}$$
We use , and . Firstly, $$\begin{aligned}
\frac{5}{3}
S_{4} &=
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}
\\
& \hspace{20mm}
\times
\left\{
\xi_{\beta}
\left(
\partial_{\xi, \alpha_{0}}
\widetilde{P}_{2}^{\gamma}
+
\frac{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}{N}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\right)
-
\xi_{\alpha_{0}}
\left(
\partial_{\xi, \beta}
\widetilde{P}_{2}^{\gamma}
+
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}}{N}
\widetilde{Q}_{4}^{\beta, \gamma}
\right)
\right\}
\\
&=
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\gamma}
+
\frac{1}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}} \xi_{\beta}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
- S_{12}.\end{aligned}$$ The lemma is a consequence of the following calculations: $$\begin{aligned}
\mbox{(i)} \quad &
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\gamma}
\\
&
=
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
\frac{C_{\gamma}}{\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\gamma}
\sum_{\beta \in \mathcal{H} \atop
\beta \not= \alpha_{0}, \gamma}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
=
-
\sum_{\gamma \in \mathcal{H} \atop \gamma \not \alpha_{0}}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\gamma}^{2}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{5}}
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\gamma}
\\
&=
-
\sum_{\gamma \in \mathcal{H} \atop \beta \not \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\beta}.
\\
\mbox{(ii)} \quad &
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}} \xi_{\beta}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\\
&=
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}} \xi_{\beta}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\sum_{\delta \in \mathcal{H} \atop \delta \not= \alpha_{0}, \gamma}
\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \delta^{\bot}, \alpha_{0} \rangle
C_{\gamma} C_{\delta}
d_{\alpha_{0}, \gamma, \delta}
\widetilde{Q}_{\alpha_{0}, \gamma, \delta}
\\
&=
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}} \xi_{\beta}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \beta^{\bot}, \alpha_{0} \rangle
C_{\gamma} C_{\beta}
d_{\alpha_{0}, \gamma, \beta}
\widetilde{Q}_{\alpha_{0}, \gamma, \beta}
\\
&=
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}} \xi_{\beta}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
&=
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}} \xi_{\beta}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}.
\\
\mbox{(iii)} \quad &
\langle \alpha_{0}^{\bot}, \beta \rangle
\xi_{\alpha_{0}} D_{\theta} \widetilde{P}_{2}^{\beta}
+ \frac{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}{N}
\xi_{\alpha_{0}} \xi_{\beta} \widetilde{Q}_{4}^{\alpha_{0}, \beta}
\\
& =
\xi_{\alpha_{0}} \xi_{\beta}
\left( \partial_{\alpha_{0}} \widetilde{P}_{2}^{\beta}
+ \frac{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}{N}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
\right)
- \xi_{\alpha_{0}}^{2} \partial_{\beta} \widetilde{P}_{2}^{\beta}
=
\xi_{\alpha_{0}}^{2}
(\xi_{\beta} \widetilde{P}_{4}^{\alpha_{0}, \beta}
- \partial_{\beta} \widetilde{P}_{2}^{\beta}). \end{aligned}$$
Let us rewrite $S_{5}$ analogously. By , there exists a constant $K_{\alpha_{0}}$ such that $$\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \beta}
= K_{\alpha_{0}} \partial_{\xi, \alpha_{0}^{\bot}}.$$
\[lemma:S5\] $\displaystyle
S_{5} = - \frac{2}{5} (S_{11} + S_{12}) - S_{13} + S_{14}$, where $$\begin{aligned}
S_{13}
=&
\frac{2 K_{\alpha_{0}}}{5}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\partial_{\xi, \alpha_{0}^{\bot}} \widetilde{P}_{2}^{\beta},
\\
S_{14}
=&
\frac{1}{N}
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma} C_{\delta}
d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2}
\widetilde{Q}_{\beta, \gamma, \delta}. \end{aligned}$$
Let us calculate $S_{5} + 2 S_{12}/5 - S_{14}$: $$\begin{aligned}
S_{5} &+ \frac{2}{5} S_{12} - S_{14}
\notag
\\
&=
\frac{2}{5}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\widetilde{P}_{4}^{\alpha_{0}, \beta}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\notag
\\
& \qquad
-
\frac{2}{5}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\left(\partial_{\xi, \beta} \widetilde{P}_{2}^{\gamma}
+
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}}{N}
\widetilde{Q}_{4}^{\beta, \gamma}
\right)
+ \frac{2}{5} S_{12} - S_{14}
\notag
\\
&=
-\frac{2}{5}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\widetilde{P}_{4}^{\alpha_{0}, \beta}
-\frac{2}{5}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\left(
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}, \gamma}
\frac{C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \beta}
\right)
\widetilde{P}_{2}^{\gamma}
\notag
\\
& \qquad
- \frac{2}{5N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \alpha_{0}, \gamma \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\widetilde{Q}_{4}^{\beta, \gamma}
+ \frac{2}{5} S_{12} - S_{14}
\notag
\\
&=
-\frac{2}{5}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\widetilde{P}_{4}^{\alpha_{0}, \beta}
-\frac{2}{5}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \gamma \rangle C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\left(
K_{\alpha_{0}} \partial_{\xi, \alpha_{0}^{\bot}}
-
\frac{C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\partial_{\xi, \gamma}
\right)
\widetilde{P}_{2}^{\gamma}
\label{eq:S_5-1}
\\
& \qquad
-\frac{1}{5 N}
\sum_{\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
C_{\beta} C_{\gamma} C_{\delta} d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\xi_{\alpha_{0}}^{2}
\notag
\\
& \hspace{40mm}
\times
(3 \langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \alpha_{0}, \beta \rangle
+ 2 \langle \alpha_{0}^{\bot}, \beta \rangle
\langle \alpha_{0}, \gamma \rangle)
\widetilde{Q}_{\beta, \gamma, \delta}
\label{eq:S_5-2}
\\
& \qquad
- \frac{2}{5N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
(\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \alpha_{0}, \beta \rangle
- \langle \alpha_{0}^{\bot}, \beta \rangle
\langle \alpha_{0}, \gamma \rangle)
\xi_{\alpha_{0}}^{2}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}.
\label{eq:S_5-3}\end{aligned}$$ The terms in are equal to $-2 S_{11}/5 - S_{13}$. The remaining terms vanish, since is $-\bar{S}_{6}(\alpha_{0})/5N$ and the summand in is skew-symmetric with respect to $\beta, \gamma$.
\[lemma:S8\] $\displaystyle
S_{8}
= S_{13} + S_{15}$, where $\displaystyle
S_{15} :=
\frac{1}{N}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}$.
This lemma is a consequence of the following calculations: $$\begin{aligned}
\mbox{(i)} \quad &
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{
\langle \alpha_{0}^{\bot}, \gamma \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\\
& \quad =
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\sum_{
{\mbox{\tiny
$\begin{matrix}
\gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}}
\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \delta^{\bot}, \alpha_{0} \rangle
C_{\gamma} C_{\delta}
d_{\alpha_{0}, \gamma, \delta}
\partial_{\xi, \beta} \widetilde{Q}_{\alpha_{0}, \gamma, \delta}
\\
& \qquad \qquad
+
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\langle \alpha_{0}^{\bot}, \gamma \rangle
\langle \beta^{\bot}, \alpha_{0} \rangle
C_{\gamma} C_{\beta}
d_{\alpha_{0}, \gamma, \beta}
\partial_{\xi, \beta} \widetilde{Q}_{\alpha_{0}, \gamma, \beta}
\\
& \quad =
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}^{2}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \beta}
\left(
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\right)
\\
& \quad =
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}^{2}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
\\
& \quad = - N S_{15}.
\\
\mbox{(ii)} \quad &
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\gamma}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\\
& \quad
=
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
C_{\gamma} \xi_{\alpha_{0}}^{2} \xi_{\gamma}
\left(
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}, \gamma}
\frac{C_{\beta}}{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \beta}
\right)
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\\
& \quad
=
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
C_{\gamma} \xi_{\alpha_{0}}^{2} \xi_{\gamma}
\left(
K_{\alpha_{0}} \partial_{\xi, \alpha_{0}^{\bot}}
-
\frac{C_{\gamma}}{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\partial_{\xi, \gamma}
\right)
\widetilde{Q}_{4}^{\alpha_{0}, \gamma}
\\
& \quad
=
K_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
C_{\beta} \xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \alpha_{0}^{\bot}}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
- N S_{15}.
\\
\mbox{(iii)} \quad &
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
C_{\beta} \xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \alpha_{0}^{\bot}}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
\\
& \quad =
\xi_{\alpha_{0}}^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
C_{\beta} \xi_{\beta}
\partial_{\xi, \alpha_{0}^{\bot}}
\left(
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}, \beta}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}} \xi_{\beta} \xi_{\gamma}}
\right)
\\
& \quad =
\xi_{\alpha_{0}}^{2}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
\left\{
\partial_{\xi, \alpha_{0}^{\bot}}
\left(
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}} \xi_{\gamma}}
\right)
-
\langle \alpha_{0}^{\bot}, \beta \rangle
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}} \xi_{\beta} \xi_{\gamma}}
\right\}
\\
& \quad =
\xi_{\alpha_{0}}^{2}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\gamma}
\partial_{\xi, \alpha_{0}^{\bot}}
\left(
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}} \xi_{\gamma}}
\right)
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}, \gamma}
C_{\beta}
d_{\alpha_{0}, \beta, \gamma}
\\
& \quad =
-\xi_{\alpha_{0}}^{2}
\sum_{\gamma \in \mathcal{H} \atop \gamma \not= \alpha_{0}}
\langle \gamma^{\bot}, \alpha_{0} \rangle
C_{\gamma}
\partial_{\xi, \alpha_{0}^{\bot}}
\left(
\frac{D_{\theta} \widetilde{P}_{0}}
{\xi_{\alpha_{0}} \xi_{\gamma}}
\right)
\times N
\frac{\langle \alpha_{0}, \gamma \rangle}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{3}}
\\
& \quad =
- N \xi_{\alpha_{0}}^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \alpha_{0}^{\bot}}
(\widetilde{P}_{2}^{\beta} - \widetilde{P}_{2}^{\alpha_{0}})
\\
& \quad =
- N \xi_{\alpha_{0}}^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \alpha_{0}^{\bot}}
\widetilde{P}_{2}^{\beta}
+
N \xi_{\alpha_{0}}^{2}
\partial_{\xi, \alpha_{0}^{\bot}}
\widetilde{P}_{2}^{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\\
& \quad =
- N \xi_{\alpha_{0}}^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\partial_{\xi, \alpha_{0}^{\bot}}
\widetilde{P}_{2}^{\beta}
\\
& \quad = -\frac{5 N S_{13}}{2 K_{\alpha_{0}}}. \end{aligned}$$
\[lemma:S3 second\] $- S_{12} + S_{14} = S_{16} + S_{17}$, where $$\begin{aligned}
S_{16} :=&
\frac{1}{N}
\sum_{\beta, \gamma \in \mathcal{H}
\atop \beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
(\langle \gamma^{\bot}, \alpha_{0} \rangle
\langle \alpha_{0}, \beta \rangle
- \langle \alpha_{0}^{\bot}, \beta \rangle
\langle \alpha_{0}, \gamma \rangle)
\xi_{\beta}^{3}
\frac{\widetilde{Q}_{\alpha_{0}, \beta, \gamma}}{\xi_{\alpha_{0}}},
\\
S_{17} :=&
\frac{3}{N}
\sum_{\beta, \gamma \in \mathcal{H}
\atop \beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle
\langle \alpha_{0}, \beta \rangle
C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}
\langle \alpha_{0}^{\bot}, \gamma \rangle}
\xi_{\beta} \xi_{\gamma}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}. \end{aligned}$$
$$\begin{aligned}
N(-S_{12} + S_{14})
=&
-\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2}
\sum_{\delta \in \mathcal{H}
\atop
\delta \not= \beta, \gamma}
\langle \delta^{\bot}, \beta \rangle C_{\delta}
d_{\beta, \gamma, \delta}
\widetilde{Q}_{\beta, \gamma, \delta}
\\
& \qquad \qquad +
\sum_{
\mbox{\tiny
$\begin{matrix}
\beta, \gamma, \delta \in \mathcal{H}
\\
\alpha_{0}, \beta, \gamma, \delta
\\
\mbox{ are all different}
\end{matrix}$
}}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
\langle \delta^{\bot}, \beta \rangle
\langle \alpha_{0}, \beta \rangle
C_{\beta} C_{\gamma} C_{\delta}
d_{\beta, \gamma, \delta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{4}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2}
\widetilde{Q}_{\beta, \gamma, \delta}
\\
=&
-\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \alpha_{0}, \beta \rangle
C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\beta, \gamma, \alpha_{0}}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
(\langle \beta^{\bot}, \gamma \rangle
\xi_{\alpha_{0}})^{3}
\frac{\widetilde{Q}_{\beta, \gamma, \alpha_{0}}}
{\xi_{\alpha}}
\\
=&
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\langle \alpha_{0}, \beta \rangle
C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
\frac{\widetilde{Q}_{\alpha_{0}, \beta, \gamma}}
{\xi_{\alpha}}
\\
& \qquad
\times
\left(
\frac{\langle \gamma^{\bot}, \alpha_{0} \rangle \xi_{\beta}^{3}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
+ 3
\frac{\langle \beta^{\bot}, \gamma \rangle
\xi_{\alpha_{0}} \xi_{\beta} \xi_{\gamma}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}
\langle \gamma^{\bot}, \alpha_{0} \rangle}
+
\frac{\xi_{\gamma}^{3}}
{\langle \gamma^{\bot}, \alpha_{0} \rangle^{2}}
\right)
\\
=& N (S_{16} + S_{17}). \end{aligned}$$
We summarise the above calculations once.
\[corollary:S3\] $S_{3} = S_{7} - S_{11} + S_{15} + S_{16} + S_{17}$.
Next, let us calculate the terms $S_{1}$, $S_{2}$.
\[lemma:S1+S2\] $S_{1} + S_{2}
=
S_{11} - S_{15} + S_{18} + S_{19} + S_{20}$, where $$\begin{aligned}
S_{18}
&=
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{1}{\xi_{\alpha_{0}}}
(\xi_{\beta}^{3} \partial_{\xi, \alpha_{0}}^{3} \widetilde{P}_{0}
- 3 \xi_{\alpha_{0}} \xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}}^{2} \widetilde{P}_{2}^{\beta}
+ 3 \xi_{\alpha_{0}}^{3} \xi_{\beta} \xi_{\beta}
\partial_{\xi, \beta}^{2} \widetilde{P}_{2}^{\alpha_{0}}
- \xi_{\alpha_{0}}^{3}
\partial_{\xi, \beta}^{3} \widetilde{P}_{0}),
\\
S_{19}
&=
- C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{\xi_{\beta}^{3}}{\xi_{\alpha_{0}}}
(\xi_{\alpha_{0}} \widetilde{P}_{4}^{\alpha_{0}, \beta}
-
\partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\alpha_{0}}),
\\
S_{20}
&=
- \frac{C_{\alpha_{0}}}{N}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}} \widetilde{Q}_{4}^{\beta, \alpha_{0}}. \end{aligned}$$
Since $$\begin{aligned}
\xi_{\beta} &
(\partial_{\xi, \alpha}^{3} \widetilde{P}_{2}^{\beta}
- C_{\alpha} \partial_{\xi, \alpha}
\widetilde{P}_{4}^{\alpha, \beta})
+
\xi_{\alpha}
(3 \partial_{\xi, \alpha}^{2} \partial_{\xi, \beta}
\widetilde{P}_{2}^{\alpha}
+
\partial_{\xi, \beta}
\widetilde{P}_{4}^{\alpha})
\\
&=
\langle \alpha, \beta \rangle
(-3 \partial_{\xi, \alpha}^{2} \widetilde{P}_{2}^{\beta}
+ C_{\alpha} \widetilde{P}_{4}^{\alpha, \beta}
-3 \partial_{\xi, \alpha}^{2} \widetilde{P}_{2}^{\alpha}
- \widetilde{P}_{4}^{\alpha})
- 6 |\alpha|^{2} \partial_{\xi, \alpha} \partial_{\xi, \beta}
\widetilde{P}_{2}^{\alpha}
+ \partial_{\xi, \alpha}^{3} \partial_{\xi, \beta}
\widetilde{P}_{0}
\\
& \qquad
- C_{\alpha} \partial_{\xi, \alpha}
\left(
\partial_{\xi, \beta} \widetilde{P}_{2}^{\alpha}
+ \frac{\langle \beta^{\bot}, \alpha \rangle^{3}}{N}
\widetilde{Q}_{4}^{\beta, \alpha}
\right)
+ 3 \partial_{\xi, \alpha}^{3} \partial_{\xi, \beta}
\widetilde{P}_{0}
+ \partial_{\xi, \beta}
\{
(C_{\alpha} + 6 |\alpha|^{2}) \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\alpha}
- 4 \partial_{\xi, \alpha}^{3} \widetilde{P}_{0}
\}
\\
&=
\frac{\langle \alpha, \beta \rangle}{\xi_{\alpha}}
\{
-3 \xi_{\alpha} \partial_{\xi, \alpha}^{2} \widetilde{P}_{2}^{\beta}
-3 \xi_{\alpha} \partial_{\xi, \alpha}^{2} \widetilde{P}_{2}^{\alpha}
- 6 |\alpha|^{2} \partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha}
+ 4 \partial_{\xi, \alpha}^{3} \widetilde{P}_{0}
+ C_{\alpha}
(\xi_{\alpha} \widetilde{P}_{4}^{\alpha, \beta}
- \partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha})
\}
\\
& \qquad
+
\frac{1}{N} C_{\alpha} \langle \alpha^{\bot}, \beta \rangle^{3}
\partial_{\xi, \alpha} \widetilde{Q}_{4}^{\beta, \alpha}
\\
& =
\frac{\langle \alpha, \beta \rangle}{\xi_{\alpha}}
\left\{
-3 \xi_{\alpha} \partial_{\xi, \alpha}^{2} \widetilde{P}_{2}^{\beta}
+ \partial_{\xi, \alpha}^{3} \widetilde{P}_{0}
+ C_{\alpha}
(\xi_{\alpha} \widetilde{P}_{4}^{\alpha, \beta}
- \partial_{\xi, \alpha} \widetilde{P}_{2}^{\alpha})
\right\}
+
\frac{1}{N} C_{\alpha} \langle \alpha^{\bot}, \beta \rangle^{3}
\partial_{\xi, \alpha} \widetilde{Q}_{4}^{\beta, \alpha}, \end{aligned}$$ we have $$\begin{aligned}
S_{1} &+ S_{2}
\\
=&
-
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{1}{\xi_{\alpha_{0}}}
(\xi_{\beta}^{3} \partial_{\xi, \alpha_{0}}^{3} \widetilde{P}_{0}
- 3 \xi_{\alpha_{0}} \xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}}^{2} \widetilde{P}_{2}^{\beta}
+ 3 \xi_{\alpha_{0}}^{3} \xi_{\beta} \xi_{\beta}
\partial_{\xi, \beta}^{2} \widetilde{P}_{2}^{\alpha_{0}}
- \xi_{\alpha_{0}}^{3}
\partial_{\xi, \beta}^{3} \widetilde{P}_{0})
\\
&
- C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{\xi_{\beta}^{3}}{\xi_{\alpha_{0}}}
(
\xi_{\alpha_{0}} \widetilde{P}_{4}^{\alpha_{0}, \beta}
-
\partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\alpha_{0}}
)
+ \sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\xi_{\alpha_{0}}^{2}
(\xi_{\beta} \widetilde{P}_{4}^{\alpha_{0}, \beta}
- \partial_{\xi, \beta} \widetilde{P}_{2}^{\beta})
\\
&
- \frac{C_{\alpha_{0}}}{N}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}} \widetilde{Q}_{4}^{\beta, \alpha_{0}}
-
\frac{1}{N}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \beta}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
\\
=& S_{18} + S_{19} + S_{11} + S_{20} - S_{15}. \end{aligned}$$
\[lemma:S7+S11+S17+S18\] $S_{7} + S_{16} + S_{19} + S_{20} = 0$.
$$\begin{aligned}
S_{7}& + S_{20}
\\
=&
\frac{1}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta^{\bot}, \gamma \rangle^{3}
C_{\alpha_{0}} C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\alpha_{0}}^{2} \xi_{\beta}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
- \frac{C_{\alpha_{0}}}{N}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}} \widetilde{Q}_{4}^{\beta, \alpha_{0}}
\\
=&
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\frac{\xi_{\beta}}{\xi_{\alpha_{0}}}
\left(
\frac{(\langle \beta^{\bot}, \gamma \rangle \xi_{\alpha_{0}})^{3}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
-
\langle \beta^{\bot}, \gamma \rangle \xi_{\alpha_{0}} \xi_{\beta}^{2}
\right)
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\frac{\xi_{\beta}}{\xi_{\alpha_{0}}}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
& \hspace{8mm}
\times
\left(
\langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}^{3}
- 3 \langle \alpha_{0}^{\bot}, \beta \rangle
\xi_{\beta}^{2} \xi_{\gamma}
+ 3
\frac{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle}
\xi_{\beta} \xi_{\gamma}^{2}
-
\frac{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\xi_{\gamma}^{3}
- \langle \alpha_{0}^{\bot}, \gamma \rangle \xi_{\beta}^{3}
+ \langle \alpha_{0}^{\bot}, \beta \rangle
\xi_{\beta}^{2} \xi_{\gamma}
\right)
\\
=&
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}
\frac{\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}}
{\xi_{\alpha_{0}}}
\left(
-
\frac{2 \xi_{\beta}^{3} \xi_{\gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
+
\frac{3 \xi_{\beta}^{2} \xi_{\gamma}^{2}}
{\langle \alpha_{0}^{\bot}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle}
-
\frac{\xi_{\beta} \xi_{\gamma}^{3}}
{\langle \alpha_{0}^{\bot}, \gamma \rangle^{2}}
\right)
\\
=&
-
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
\frac{\xi_{\beta}^{3} \xi_{\gamma}}
{\xi_{\alpha_{0}}}
\partial_{\xi, \alpha_{0}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
-
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
\frac{\xi_{\beta}^{3}}
{\xi_{\alpha_{0}}}
\left\{
\partial_{\xi, \alpha_{0}}
\left(
\xi_{\gamma}
\frac{\widetilde{P}_{2}^{\alpha_{0}} - \widetilde{P}_{2}^{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle \xi_{\gamma}}
\right)
- \langle \gamma, \alpha_{0} \rangle
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\right\}
\\
=&
-
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta \in \mathcal{H} \atop
\beta \not= \alpha_{0}}
\frac{C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\left(
\sum_{\gamma \in \mathcal{H} \atop
\gamma \not= \alpha_{0}, \beta}
C_{\gamma} d_{\alpha_{0}, \beta, \gamma}
\right)
\frac{\xi_{\beta}^{3}}
{\xi_{\alpha_{0}}}
\partial_{\xi, \alpha_{0}}
(\widetilde{P}_{2}^{\alpha_{0}} - \widetilde{P}_{2}^{\beta})
\\
& \qquad
+
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \gamma, \alpha_{0} \rangle
C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
\frac{\xi_{\beta}^{3}}
{\xi_{\alpha_{0}}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
-
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop
\beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{\xi_{\beta}^{3}}
{\xi_{\alpha_{0}}}
\partial_{\xi, \alpha_{0}}
(\widetilde{P}_{2}^{\alpha_{0}} - \widetilde{P}_{2}^{\beta})
+
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \gamma, \alpha_{0} \rangle
C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
\frac{\xi_{\beta}^{3}}
{\xi_{\alpha_{0}}}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{6}}
\frac{\xi_{\beta}^{3}}{\xi_{\alpha_{0}}}
\left(\partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\beta}
+ \frac{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}{N}
\widetilde{Q}_{4}^{\alpha_{0}, \beta}
- \partial_{\xi, \alpha_{0}} \widetilde{P}_{2}^{\alpha_{0}}
\right)
\\
& \qquad
-
\frac{C_{\alpha_{0}}}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\beta} C_{\gamma} d_{\alpha_{0}, \beta, \gamma}
}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\frac{\xi_{\beta}^{3}}{\xi_{\alpha_{0}}}
(\langle \alpha_{0}, \beta \rangle
\langle \gamma^{\bot}, \alpha_{0} \rangle
- \langle \gamma, \alpha_{0} \rangle
\langle \alpha_{0}^{\bot}, \beta \rangle)
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=& - S_{19} - S_{16}. \end{aligned}$$
\[lemma:S12\] $\displaystyle
S_{17}
=
3 C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle
|\beta|^{2} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}$.
$$\begin{aligned}
S_{17} =&
\frac{3}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}
\langle \alpha_{0}^{\bot}, \gamma \rangle}
(-\langle \gamma^{\bot}, \alpha_{0} \rangle |\beta|^{2}
- \langle \alpha_{0}^{\bot}, \beta \rangle
\langle \beta, \gamma \rangle)
\xi_{\beta} \xi_{\gamma}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
\frac{3}{N}
\sum_{\beta \in \mathcal{H} \atop
\beta \not= \alpha_{0}}
\frac{|\beta|^{2} C_{\alpha_{0}} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{2}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}
\sum_{\gamma \in \mathcal{H} \atop
\gamma \not= \alpha_{0}, \beta}
C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}
-
\frac{3}{N}
\sum_{\beta, \gamma \in \mathcal{H} \atop
\beta \not= \gamma \not= \alpha_{0} \not= \beta}
\frac{\langle \beta, \gamma \rangle
C_{\alpha_{0}} C_{\beta} C_{\gamma}
d_{\alpha_{0}, \beta, \gamma}}
{\langle \alpha_{0}^{\bot}, \beta \rangle
\langle \alpha_{0}^{\bot}, \gamma \rangle}
\xi_{\beta} \xi_{\gamma}
\widetilde{Q}_{\alpha_{0}, \beta, \gamma}
\\
=&
3 C_{\alpha_{0}}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle
|\beta|^{2} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}. \end{aligned}$$
\[lemma:D\_theta\^3\] $\displaystyle
S_{18} =
-6 |\alpha_{0}|^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle |\beta|^{2} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}$.
By using and $|\alpha_{0}|^{2} \xi_{\beta} \partial_{\xi, \beta}
+
|\beta|^{2} \xi_{\alpha_{0}} \partial_{\xi, \alpha_{0}}
- \langle \alpha_{0}, \beta \rangle
(\xi_{\beta} \partial_{\xi, \alpha_{0}}
+ \xi_{\alpha_{0}} \partial_{\xi, \beta})
=
\langle \alpha_{0}^{\bot}, \beta \rangle^{2}
\langle \xi, \partial_{\xi} \rangle$, we can easily shown $$\begin{aligned}
\xi_{\beta}^{3} \partial_{\xi, \alpha_{0}}^{3} \widetilde{P}_{0}
&
- 3 \xi_{\alpha_{0}} \xi_{\beta}^{3}
\partial_{\xi, \alpha_{0}}^{2} \widetilde{P}_{2}^{\beta}
+ 3 \xi_{\alpha_{0}}^{3} \xi_{\beta}
\partial_{\xi, \beta}^{2} \widetilde{P}_{2}^{\alpha_{0}}
- \xi_{\alpha_{0}}^{3}
\partial_{\xi, \beta}^{3} \widetilde{P}_{0}
\\
=&
\langle \alpha_{0}^{\bot}, \beta \rangle^{3}
(D_{\theta}^{3}
+ 3 \langle \xi, \partial_{\xi} \rangle D_{\theta} - 8 D_{\theta})
\widetilde{P}_{0}
+
6 \langle \alpha_{0}^{\bot}, \beta \rangle |\alpha|^{2} |\beta|^{2}
D_{\theta}
\widetilde{P}_{0}. \end{aligned}$$ Therefore, we have $$\begin{aligned}
S_{18} =& -
\left(
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{3}}
\right)
\frac{
(D_{\theta}^{3}
+ 3 \langle \xi, \partial_{\xi} \rangle D_{\theta} - 8 D_{\theta})
\widetilde{P}_{0}}
{\xi_{\alpha_{0}}}
-6
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
|\alpha_{0}|^{2} |\beta|^{2}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}
\\
=&
-6 |\alpha_{0}|^{2}
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle |\beta|^{2} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}.\end{aligned}$$
<span style="font-variant:small-caps;">Proof</span> of Theorem \[theorem:linear relation-2\].
By the above long discussion, the equality $S_{1} + S_{2} + S_{3} = 0$ reduces to the equality $S_{17} + S_{18} = 0$. By Lemma \[lemma:S12\] and Lemma \[lemma:D\_theta\^3\], we have $$0
= S_{17} + S_{18}
=
3 (C_{\alpha_{0}} - 2 |\alpha_{0}|^{2})
\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\frac{\langle \alpha_{0}, \beta \rangle
|\beta|^{2} C_{\beta}}
{\langle \alpha_{0}^{\bot}, \beta \rangle^{5}}
\frac{D_{\theta} \widetilde{P}_{0}}{\xi_{\alpha_{0}}}.$$ Since $\widetilde{P}_{0}$ is not a polynomial in $\xi_{1}^{2} + \xi_{2}^{2}$, $D_{\theta} \widetilde{P}_{0}$ is not zero, and the theorem is proved. $\square$
Possible deformation of root systems {#section:possibility}
====================================
In this section, we investigate what kind of $\mathcal{H}$ and $C_{\alpha}$ satisfy and .
Let $\# \mathcal{H} = N$ and $\mathcal{H} = \{\alpha_{1}, \dots, \alpha_{N}\}$. For notational convenience, we define $$\begin{aligned}
A_{ij} :=&
\frac{\langle \alpha_{i}, \alpha_{j} \rangle}
{\langle \alpha_{i}^{\bot}, \alpha_{j} \rangle^{3}}
\qquad \mbox{(if $i \not= j$)},
& A_{ii} :=& 0 \qquad \mbox{(if $i = j$)},
& \mathcal{A} :=& (A_{ij})_{1 \leq i,j \leq N},
\\
B_{ij} :=&
\frac{\langle \alpha_{i}, \alpha_{j} \rangle
|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
{\langle \alpha_{i}^{\bot}, \alpha_{j} \rangle^{5}}
\qquad \mbox{(if $i \not= j$)},
&
B_{ii} :=&
0 \qquad \mbox{(if $i = j$)},
& \mathcal{B} :=& (B_{ij})_{1 \leq i,j \leq N},
\\
C_{i} :=& C_{\alpha_{i}},
&
\boldsymbol{v} :=&
{}^{t} (C_{1}, \dots, C_{N}). \end{aligned}$$ Then, and are equivalent to $$\begin{aligned}
&
\mathcal{A} \boldsymbol{v} = \boldsymbol{0},
\label{eq:Av=0}
\\
&
\mathrm{diag}
(C_{1} - 2 |\alpha_{1}|^{2}, \dots, C_{N} - 2 |\alpha_{N}|^{2})
\mathcal{B} \boldsymbol{v} = \boldsymbol{0},
\label{eq:Bv=0}\end{aligned}$$ respectively.
If $N$ is odd, then $\det \mathcal{A} = 0$, since $\mathcal{A}$ is an alternative matrix. Therefore, the solution space of is at least one dimensional. If the rank of $\mathcal{A}$ is $N-1$, the non-trivial solution of is given by $$C_{i} = C (-1)^{i} \mathrm{Pf}_{i}(\mathcal{A})
\qquad
(i = 1, 2, \dots, N),$$ where $C$ is a non-zero constant and $\mathrm{Pf}_{i}(\mathcal{A})$ is the Pfaffian of the $(N-1) \times (N-1)$ alternative matrix obtained by deleting the $i$ th row and column of $\mathcal{A}$. Especially, if $N = 3$, then $$\begin{aligned}
& C_{1} = C A_{23},
&
& C_{2} = C A_{31},
&
& C_{3} = C A_{12}.
\label{eq:C_i when H=3}\end{aligned}$$
\[lemma:C\_1 not=2\] If $N = 3$ and $C_{1} \not= 2 |\alpha_{1}|^{2}$, $\alpha_{2}$ and $\alpha_{3}$ are symmetric with respect to the reflection $r_{\alpha_{1}}$.
Since $C_{\alpha_{1}} \not= 2 |\alpha_{1}|^{2}$, the equations and imply $$\begin{aligned}
B_{12} C_{2} + B_{13} C_{3} = 0
& \quad \Leftrightarrow \quad
-
C
\frac{
\langle \alpha_{1}, \alpha_{3} \rangle
\langle \alpha_{1}, \alpha_{2} \rangle
|\alpha_{2}|^{2}}
{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle^{3}
\langle \alpha_{1}^{\bot}, \alpha_{2} \rangle^{5}}
+
C
\frac{
\langle \alpha_{1}, \alpha_{2} \rangle
\langle \alpha_{1}, \alpha_{3} \rangle
|\alpha_{3}|^{2}}
{\langle \alpha_{1}^{\bot}, \alpha_{2} \rangle^{3}
\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle^{5}}
= 0
\\
& \quad \Leftrightarrow \quad
\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle^{2} |\alpha_{2}|^{2}
=
\langle \alpha_{1}^{\bot}, \alpha_{2} \rangle^{3} |\alpha_{3}|^{2}
\\
& \quad \Leftrightarrow \quad
\left(
\frac{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle}
{|\alpha_{1}| |\alpha_{3}|}
\right)^{2}
=
\left(
\frac{\langle \alpha_{1}^{\bot}, \alpha_{2} \rangle}
{|\alpha_{1}| |\alpha_{2}|}
\right)^{2}
\\
& \quad \Leftrightarrow \quad
\sin^{2} \theta_{2} = \sin^{2} \theta_{3}, \end{aligned}$$ where $\theta_{i}$ are the angles from $\alpha_{1}$ to $\alpha_{i}$ ($i = 2, 3$). Since $\alpha_{2}$ and $\alpha_{3}$ are not parallel, this implies the lemma.
\[corollary:A2, all not 2\] If $\# \mathcal{H} = 3$ and (i) more than or equal to two of $C_{i}$’s are not equal to $2 |\alpha_{i}|^{2}$ or (ii) all $C_{i}$’s are equal to $2 |\alpha_{i}|^{2}$, then $\mathcal{H}$ is a positive system of the $A_{2}$ type root system and $L$ is the $A_{2}$ type CMS operator.
The first assertion follows directly from the last lemma.
If $C_{i} = 2 |\alpha_{i}|^{2}$ for $i = 1, 2, 3$, implies $$\begin{aligned}
&
\frac{\langle \alpha_{2}, \alpha_{3} \rangle}
{\langle \alpha_{2}^{\bot}, \alpha_{3} \rangle^{3} |\alpha_{1}|^{2}}
=
\frac{\langle \alpha_{3}, \alpha_{1} \rangle}
{\langle \alpha_{3}^{\bot}, \alpha_{1} \rangle^{3} |\alpha_{2}|^{2}}
=
\frac{\langle \alpha_{1}, \alpha_{2} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{2} \rangle^{3} |\alpha_{3}|^{2}}
\\
& \Leftrightarrow \quad
\cot(\theta_{3} - \theta_{2})
\{1 + \cot^{2} (\theta_{3} - \theta_{2})\}
=
- \cot \theta_{3} (1 + \cot^{2} \theta_{3})
= \cot \theta_{2} (1 + \cot^{2} \theta_{2})
\\
& \Leftrightarrow \quad
\cot (\theta_{3} - \theta_{2})
= - \cot \theta_{3}
= \cot \theta_{2}
\\
& \Leftrightarrow \quad
\cot \theta_{2} = - \cot \theta_{3} = \pm 1/\sqrt{3}. \end{aligned}$$ By changing the norm of vectors if necessary, we may regard $\mathcal{H}$ to be a positive system of $A_{2}$ type root system.
As a result of Lemma \[lemma:C\_1 not=2\] and Corollary \[corollary:A2, all not 2\], we obtain the following theorem.
\[theorem:A2\] When $\# \mathcal{H} = 3$, the possible hypeplane arrangement $\mathcal{H}$ is $\mathcal{H} = \{e_{1}, \pm a e_{1} + e_{2}\}$ ($a \not= 0$). If $a \not= \pm 1/\sqrt{3}$, in other words if $\mathcal{H}$ is not a positive system of $A_{2}$ type, the coupling constants for $\pm a e_{1} + e_{2}$ must be one.
Next, let us consider the case $N=4$.
\[proposition:possibility, N=4\] If $\# \mathcal{H} = 4$, at least two vectors in $\mathcal{H}$ cross at right angles.
Put $\mathcal{H}
= \{\alpha_{1}, \alpha_{2}, \alpha_{3}, \alpha_{4}\}$ and let $x_{i} = \cot \theta_{i}$ ($i = 2, 3, 4$), where $\theta_{i}$ is the angle from $\alpha_{1}$ to $\alpha_{i}$. Note that $x_{i}$ ($i = 2, 3, 4$) are all different since any two vectors in $\mathcal{H}$ are not parallel. Since $$\begin{aligned}
& A_{ij} =
\frac{\langle \alpha_{i}, \alpha_{j} \rangle}
{\langle \alpha_{i}^{\bot}, \alpha_{j} \rangle^{3}}
= \frac{1}{|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
\frac{\cos (\theta_{j} - \theta_{i})}
{\sin^{3} (\theta_{j} - \theta_{i})}
=
\frac{1}{|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
\cot (\theta_{j} - \theta_{i})
(1 + \cot^{2} (\theta_{j} - \theta_{i})),
\\
& B_{ij} =
\frac{\langle \alpha_{i}, \alpha_{j} \rangle
|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
{\langle \alpha_{i}^{\bot}, \alpha_{j} \rangle^{5}}
= \frac{1}{|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
\frac{\cos (\theta_{j} - \theta_{i})}
{\sin^{5} (\theta_{j} - \theta_{i})}
=
\frac{1}{|\alpha_{i}|^{2} |\alpha_{j}|^{2}}
\cot (\theta_{j} - \theta_{i})
(1 + \cot^{2} (\theta_{j} - \theta_{i}))^{2}, \end{aligned}$$ we have $$\begin{aligned}
A_{1i}
&=
\frac{x_{i} (1 + x_{i}^{2})}{|\alpha_{1}|^{2} |\alpha_{i}|^{2}},
&
B_{1i}
&=
\frac{x_{i} (1 + x_{i}^{2})^{2}}{|\alpha_{1}|^{2} |\alpha_{i}|^{2}}
&
& (i = 2, 3, 4),
\\
A_{ij}
&=
\frac{(1 + x_{i}^{2}) (1 + x_{j}^{2}) (1 + x_{i} x_{j})}
{|\alpha_{i}|^{2} |\alpha_{j}|^{2} (x_{i} - x_{j})^{3}},
&
B_{ij}
&=
\frac{(1 + x_{i}^{2})^{2} (1 + x_{j}^{2})^{2} (1 + x_{i} x_{j})}
{|\alpha_{i}|^{2} |\alpha_{j}|^{2} (x_{i} - x_{j})^{5}}
&
& (2 \leq i \not= j \leq 4). \end{aligned}$$
Assume that any two vectors in $\mathcal{H}$ do not cross at right angles and deduce contradiction. Note that $x_{i} \not= 0$ for $i = 2, 3, 4$ and $1 + x_{i} x_{j} \not= 0$ for $2 \leq i \not= j \leq 4$ since $\langle \alpha_{i}, \alpha_{j} \rangle \not= 0$ for any $i, j$.
We divide the proof of this theorem into three parts, since the method of poof is different for the following cases:
1. More than or equal to three coupling constants are one.
2. More than or equal to three coupling constants are not one.
3. Just two coupling constants are one.
\(1) In this case, we may assume $C_{i} = 2 |\alpha_{i}|^{2}$ for $i = 2, 3, 4$. Let $p_{k} := \sum_{i = 2}^{4} x_{i}^{k}$ be the $k$-th power sum of $x_{2}, x_{3}, x_{4}$. By and , we have $$\begin{cases}
2(A_{12} |\alpha_{2}|^{2}
+ A_{13} |\alpha_{3}|^{2}
+ A_{14} |\alpha_{4}|^{2})
= 0
\\
2(
B_{12} |\alpha_{2}|^{2}
+ B_{13} |\alpha_{3}|^{2}
+ B_{14} |\alpha_{4}|^{2}
)
= 0
\end{cases}
\quad \Leftrightarrow \quad
\begin{cases}
p_{1} + p_{3} = 0
\\
p_{1} + 2 p_{3} + p_{5} = 0
\end{cases}
\quad \Leftrightarrow \quad
p_{1} = - p_{3} = p_{5},$$ Since $6 p_{5}
= p_{1}^{5} - 5 p_{1}^{3} p_{2} + 5 p_{1}^{2} p_{3} + 5 p_{2} p_{3}$, we have $$\begin{aligned}
p_{1}
\{5(p_{1}^{2} + 1) p_{2} - (p_{1}^{4} - 5 p_{1}^{2} - 6)\}
= 0
& \quad \Leftrightarrow \quad
p_{1} (p_{1}^{2} + 1) (5 p_{2} - p_{1}^{2} + 6) = 0
\\
& \quad \Leftrightarrow \quad
p_{1} (p_{1}^{2} + 1)
\left(3 \sum_{i=2}^{4} x_{i}^{2}
+ \sum_{2 \leq i < j \leq 4} (x_{i} - x_{j})^{2} + 6
\right)
= 0
\\
& \quad \Rightarrow \quad
p_{1} = 0. \end{aligned}$$ If $p_{1} = 0$, then $p_{3} = x_{2}^{3} + x_{3}^{3} - (x_{2} + x_{3})^{3}
= 3 x_{2} x_{3} x_{4} = 0$, which contradicts our assumption.
\(2) Assume that the coupling constants for $\alpha_{1}, \alpha_{2}, \alpha_{3}$ are not one. Then $C_{1}, \dots, C_{4}$ satisfy $$\label{eq:B'v=0}
\begin{pmatrix}
0 & B_{12} & B_{13} & B_{14}
\\
B_{21} & 0 & B_{23} & B_{24}
\\
B_{31} & B_{32} & 0 & B_{34}
\\
A_{41} & A_{42} & A_{43} & 0
\end{pmatrix}
\begin{pmatrix}
C_{1} \\ C_{2} \\ C_{3} \\ C_{4}
\end{pmatrix}
=
\boldsymbol{0}.$$ Since all $C_{i}$’s are not zero, the determinant of the coefficient matrix is zero; $$\label{eq:detB'=0}
(B_{12} B_{34} - B_{13} B_{24} + B_{14} B_{23})
(B_{12} A_{34} - B_{13} A_{24} + B_{23} A_{14}) = 0.$$
On the other hand, since $C_{1}, \dots, C_{4}$ satisfy , we have $$\mathrm{Pf}(\mathcal{A})
= A_{12} A_{34} - A_{13} A_{24} + A_{14} A_{23}
= 0.$$By our assumption that any two vectors in $\mathcal{H}$ do not cross at right angles, each $A_{ij}$ is not zero. Therefore, the solution of is expressed as $$\begin{aligned}
& C_{1} = s A_{34},
&
& C_{2} = t A_{34},
&
& C_{3} = -(s A_{14} + t A_{24}),
&
& C_{4} = s A_{13} + t A_{23},
&
& \quad (s, t \not= 0). \end{aligned}$$ Since this satisfies , we have $$s(A_{13} B_{14} - A_{14} B_{13})
+ t(B_{12} A_{34} - B_{13} A_{24} + B_{23} A_{14}) = 0.$$ Assume that $B_{12} A_{34} - B_{13} A_{24} + B_{23} A_{14} = 0$. In this case, $$A_{13} B_{14} = A_{14} B_{13}
\quad \Leftrightarrow \quad
x_{3} (1 + x_{3}^{2}) x_{4} (1 + x_{4}^{2})^{2}
= x_{4} (1 + x_{4}^{2}) x_{3} (1 + x_{3}^{2})^{2}$$ and this implies $x_{3} = - x_{4}$, since $x_{3}, x_{4} \not= 0$ and $x_{3} \not= x_{4}$. Therefore, $A_{14} |\alpha_{4}|^{2} = - A_{13} |\alpha_{3}|^{2}$, $B_{14} |\alpha_{4}|^{2} = - B_{13} |\alpha_{3}|^{2}$, and we have $$\begin{cases}
A_{12} C_{2} + A_{13} C_{3} + A_{14} C_{4} = 0
\\
B_{12} C_{2} + B_{13} C_{3} + B_{14} C_{4} = 0
\end{cases}
\quad \Leftrightarrow \quad
\begin{cases}
A_{12} C_{2}
+ A_{13} (C_{3} -|\alpha_{3}|^{2} C_{4} / |\alpha_{4}|^{2})
= 0
\\
B_{12} C_{2}
+ B_{13} (C_{3} -|\alpha_{3}|^{2} C_{4} / |\alpha_{4}|^{2})
= 0.
\end{cases}$$ By these equations, we have $(B_{13} A_{12} - B_{12} A_{13}) C_{2} = 0$, which implies $x_{2}^{2} = x_{3}^{2}$. Since $x_{2} \not= x_{3}$, we have $x_{2} = - x_{3} = x_{4}$. But this contradicts the condition $x_{2} \not= x_{4}$.
Therefore, $B_{12} A_{34} - B_{13} A_{24} + B_{23} A_{14} \not= 0$, and $A_{ij}$, $B_{ij}$ satisfy $$\begin{cases}
A_{12} A_{34} - A_{13} A_{24} + A_{14} A_{23} = 0
\\
B_{12} B_{34} - B_{13} B_{24} + B_{14} B_{23} = 0
\end{cases}
\quad \Leftrightarrow \quad
\begin{cases}
\frac{x_{2} (1 + x_{3} x_{4})}{(x_{3} - x_{4})^{3}}
+
\frac{x_{3} (1 + x_{4} x_{2})}{(x_{4} - x_{2})^{3}}
+
\frac{x_{4} (1 + x_{2} x_{3})}{(x_{2} - x_{3})^{3}}
= 0
\\
\frac{x_{2} (1 + x_{3} x_{4})}{(x_{3} - x_{4})^{5}}
+
\frac{x_{3} (1 + x_{4} x_{2})}{(x_{4} - x_{2})^{5}}
+
\frac{x_{4} (1 + x_{2} x_{3})}{(x_{2} - x_{3})^{5}}
= 0,
\end{cases}$$ by . From these equations, we have $$\begin{aligned}
&
\frac{x_{2} (1 + x_{3} x_{4}) (2 x_{3} - x_{2} - x_{4})}
{(x_{3} - x_{4})^{6}}
=
\frac{x_{3} (1 + x_{2} x_{4}) (2 x_{2} - x_{3} - x_{4})}
{(x_{2} - x_{4})^{6}}
\label{eq:x2 and x3}
\\
& \Leftrightarrow \quad
x_{2} (1 + x_{3} x_{4})
(2 x_{3} - x_{2} - x_{4}) (2 x_{4} - x_{2} - x_{3})
\notag
\\
& \qquad \qquad \qquad
\times
\{(x_{2} - x_{4})^{4} + (x_{2} - x_{4})^{2} (x_{3} - x_{4})^{2}
+ (x_{3} - x_{4})^{4}\}
\notag
\\
& \qquad \qquad
+ (x_{2} + x_{3} + x_{4} + 3 x_{2} x_{3} x_{4}) (x_{3} - x_{4})^{6}
= 0.
\label{eq:x2 and x3, no.2} \end{aligned}$$ Here, we used $x_{2} \not= x_{3}$. Similarly, we have $$\begin{aligned}
&
x_{2} (1 + x_{3} x_{4})
(2 x_{4} - x_{2} - x_{3}) (2 x_{3} - x_{2} - x_{4})
\notag
\\
& \qquad \qquad \times
\{(x_{2} - x_{3})^{4} + (x_{2} - x_{3})^{2} (x_{3} - x_{4})^{2}
+ (x_{3} - x_{4})^{4}\}
\notag
\\
& \qquad + (x_{2} + x_{3} + x_{4} + 3 x_{2} x_{3} x_{4}) (x_{3} - x_{4})^{6}
= 0.
\label{eq:x2 and x4, no.2} \end{aligned}$$ By and , we have $$\begin{aligned}
&
x_{2} (1 + x_{3} x_{4})
(2 x_{2} - x_{3} - x_{4}) (2 x_{3} - x_{4} - x_{2})
(2 x_{4} - x_{2} - x_{3})
\\
& \qquad \qquad
\times
\{(x_{2} - x_{4})^{2} + (x_{2} - x_{3})^{2} + (x_{3} - x_{4})^{2}\}
= 0.\end{aligned}$$ Since $x_{2} \not= 0$, $1 + x_{3} x_{4} \not= 0$ and $x_{2}, x_{3}, x_{4}$ are all different, $2x_{2} = x_{3} + x_{4}$, $2x_{3} = x_{4} + x_{2}$ or $2x_{4} = x_{2} + x_{3}$. But this is impossible since we obtain $x_{2} = x_{3}$ or $x_{4}(1 + x_{2} x_{3}) = 0$ from .
\(3) Let us assume the coupling constants for $\alpha_{3}$ and $\alpha_{4}$ are $1$ and those for $\alpha_{1}$ and $\alpha_{2}$ are not $1$. In this case, all the conditions for $A_{ij}$, $B_{ij}$ and $C_{i}$ are $$\begin{aligned}
&
A_{12} C_{2} + 2 |\alpha_{3}|^{2} A_{13} + 2 |\alpha_{4}|^{2} A_{14}
= 0,
&
&
B_{12} C_{2} + 2 |\alpha_{3}|^{2} B_{13} + 2 |\alpha_{4}|^{2} B_{14}
= 0,
\\
&
A_{21} C_{1} + 2 |\alpha_{3}|^{2} A_{23} + 2 |\alpha_{4}|^{2} A_{24}
= 0,
&
&
B_{21} C_{1} + 2 |\alpha_{3}|^{2} B_{23} + 2 |\alpha_{4}|^{2} B_{24}
= 0,
\\
&
\mathrm{Pf}(A) = A_{12} A_{34} - A_{13} A_{24} + A_{14} A_{23} = 0.
&
& \end{aligned}$$ By eliminating $C_{1}$ and $C_{2}$, we have $$\begin{aligned}
&
x_{2}^{2} =
\frac{x_{3}^{2} - x_{3} x_{4} + x_{4}^{2}
+ x_{3}^{4} - x_{3}^{3} x_{4} + x_{3}^{2} x_{4}^{2}
- x_{3} x_{4}^{3} + x_{4}^{4}}
{1 + x_{3}^{2} - x_{3} x_{4} + x_{4}^{2}},
\label{eq:x2^2}
\\
&
\frac{(1 + x_{3}^{2}) (1 + x_{2} x_{3})
(1 + 2 x_{2} x_{3} - x_{2}^{2})}
{(x_{2} - x_{3})^{5}}
+
\frac{(1 + x_{4}^{2}) (1 + x_{2} x_{4})
(1 + 2 x_{2} x_{4} - x_{2}^{2})}
{(x_{2} - x_{4})^{5}}
= 0,
\label{eq:x2-x3x4}
\\
&
\frac{x_{2} (1 + x_{3} x_{4})}{(x_{3} - x_{4})^{3}}
+
\frac{x_{3} (1 + x_{4} x_{2})}{(x_{4} - x_{2})^{3}}
+
\frac{x_{4} (1 + x_{2} x_{3})}{(x_{2} - x_{3})^{3}}
= 0.
\label{eq:PfA, explicit}\end{aligned}$$ Let $F(x_{2})$ and $G(x_{2})$ be the left hand sides of and , respectively. Since $\{F(x_{2}) G(-x_{2}) - F(-x_{2}) G(x_{2})\}/x_{2}$ is an even polynomial in $x_{2}$, it is a polynomial in $x_{2}^{2}$. By substituting for it and putting $x_{3} = s + t$, $x_{4} = s - t$, we have $$\begin{aligned}
t^{6} & (s^{2} - t^{2})
\\
& \times
(3 s^{2} + 24 s^{4} + 75 s^{6} + 120 s^{8} + 105 s^{10} + 48 s^{12}
+ 9 s^{14} + 3 t^{2} + 92 s^{2} t^{2} + 559 s^{4} t^{2}
+ 1408 s^{6} t^{2}
\\
& \qquad
+ 1745 s^{8} t^{2} + 1060 s^{10} t^{2} + 253 s^{12} t^{2}
+ 44 t^{4} + 889 s^{2} t^{4} + 4368 s^{4} t^{4} + 8658 s^{6} t^{4}
+ 7580 s^{8} t^{4}
\\
& \qquad
+ 2445 s^{10} t^{4} + 237 t^{6} + 3744 s^{2} t^{6} + 14538 s^{4} t^{6}
+ 20616 s^{6} t^{6} + 9777 s^{8} t^{6} + 600 t^{8} + 7461 s^{2} t^{8}
\\
& \qquad
+ 20616 s^{4} t^{8} + 15675 s^{6} t^{8} + 773 t^{10}
+ 6932 s^{2} t^{10} + 10263 s^{4} t^{10} + 492 t^{12}
+ 2415 s^{2} t^{12} + 123 t^{14})
\\
& \times
(5 s^{4} + 30 s^{6} + 75 s^{8} + 100 s^{10} + 75 s^{12} + 30 s^{14}
+ 5 s^{16} + 50 s^{2} t^{2} + 478 s^{4} t^{2} + 1744 s^{6} t^{2}
+ 3196 s^{8} t^{2}
\\
& \qquad
+ 3154 s^{10} t^{2} + 1606 s^{12} t^{2} + 332 s^{14} t^{2}
+ 121 t^{4} + 1818 s^{2} t^{4} + 9358 s^{4} t^{4} + 22792 s^{6} t^{4}
+ 28609 s^{8} t^{4}
\\
& \qquad
+ 17910 s^{10} t^{4} + 4432 s^{12} t^{4} + 1386 t^{6}
+ 14280 s^{2} t^{6} + 52248 s^{4} t^{6} + 87916 s^{6} t^{6}
+ 69246 s^{8} t^{6}
\\
& \qquad
+ 20684 s^{10} t^{6} + 6543 t^{8} + 48852 s^{2} t^{8}
+ 123037 s^{4} t^{8} + 126170 s^{6} t^{8} + 45458 s^{8} t^{8}
+ 16172 t^{10}
\\
& \qquad
+ 83490 s^{2} t^{10} + 126546 s^{4} t^{10} + 54532 s^{6} t^{10}
+ 21879 t^{12} + 69266 s^{2} t^{12} + 45080 s^{4} t^{12}
+ 15210 t^{14}
\\
& \qquad
+ 21860 s^{2} t^{14} + 4225 t^{16})
\\
& = 0. \end{aligned}$$ This equation implies $t = 0$, $s = \pm t$ or $(s, t) = (0, 0)$, but these are impossible since $x_{3}, x_{4} \not= 0$ and $x_{3} \not= x_{4}$.
\[theorem:4 lines arrangement\] When $\# \mathcal{H} = 4$, the possible singular locus $\mathcal{H}$ is $\mathcal{H} = \{e_{1}, e_{2}, \pm a e_{1} + e_{2}\}$ ($a \not =0$).
By Proposition \[proposition:possibility, N=4\], at least two vectors in $\mathcal{H}$, say $\alpha_{1}$ and $\alpha_{2}$, cross at right angles. In this case, we have $A_{12} = 0$ and we may assume $\alpha_{2} = \alpha_{1}^{\bot}$. Note that $\alpha_{2}^{\bot} = \alpha_{1}^{\bot \bot} = -\alpha_{1}$. Then, $$\begin{aligned}
\mathrm{Pf}(\mathcal{A}) = -A_{13} A_{24} + A_{14} A_{23} =0
\quad & \Leftrightarrow \quad
\frac{\langle \alpha_{1}, \alpha_{3} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle^{3}}
\frac{\langle \alpha_{2}, \alpha_{4} \rangle}
{\langle \alpha_{2}^{\bot}, \alpha_{4} \rangle^{3}}
=
\frac{\langle \alpha_{1}, \alpha_{4} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{4} \rangle^{3}}
\frac{\langle \alpha_{2}, \alpha_{3} \rangle}
{\langle \alpha_{2}^{\bot}, \alpha_{3} \rangle^{3}}
\\
& \Leftrightarrow \quad
\frac{\langle \alpha_{1}, \alpha_{3} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle^{3}}
(-1)
\frac{\langle \alpha_{1}^{\bot}, \alpha_{4} \rangle}
{\langle \alpha_{1}, \alpha_{4} \rangle^{3}}
=
\frac{\langle \alpha_{1}, \alpha_{4} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{4} \rangle^{3}}
(-1)
\frac{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle}
{\langle \alpha_{1}, \alpha_{3} \rangle^{3}}
\\
& \Leftrightarrow \quad
\left(
\frac{\langle \alpha_{1}, \alpha_{3} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{3} \rangle}
\right)^{4}
=
\left(
\frac{\langle \alpha_{1}, \alpha_{4} \rangle}
{\langle \alpha_{1}^{\bot}, \alpha_{4} \rangle}
\right)^{4}
\\
& \Leftrightarrow \quad
x_{3}^{4} = x_{4}^{4}
\\
& \Leftrightarrow \quad
x_{3} = - x_{4},\end{aligned}$$ since $x_{3} \not= x_{4}$. By an appropriate coordinate change, we may assume $\alpha_{1} = e_{1}$. Then $\alpha_{2} = e_{1}^{\bot} = e_{2}$ and, by changing the norm of $\alpha_{3}, \alpha_{4}$ if necessary, we have $\alpha_{3} = x_{3} e_{1} + e_{2}$ and $\alpha_{4} = - x_{3} e_{1} + e_{2}$.
New deformation of $B_{2}$ type commutative pair {#section:new deformation}
================================================
In this section, we construct a pair of commuting differential operators with the hyperplane arrangement $\mathcal{H} = \{\alpha_{1} := e_{1},
\alpha_{2} := e_{2},
\alpha_{\pm} := \pm a e_{1} + e_{2}\}$. If $a = \pm 1$, then $\mathcal{H}$ is the positive system of $B_{2}$ type root system and the commuting differential operators are known. Therefore, we assume $a \not= \pm 1$ in this section.
If the coupling constants for $\alpha_{\pm}$ are one, the existence of such commuting operators is proved by Veselov-Feigin-Chalykh for rational or trigonometric potential cases ([@CFV]). Here, we consider the case where the coupling constants for $\alpha_{\pm}$ are two. In this case, there exists a pair of commuting differential operators $L$ and $P$ for rational, trigonometric and even elliptic cases. Remember the lowest order of the commutant $P$ for the original CMS model is four. But in our case, we can not find a fourth order commutant $P$ because no fourth order operator satisfies Proposition \[proposition:review-2\]. The lowest order of a commutant $P$ is six.
Let the coupling constants for $\alpha_{\pm}$ be two and let $C_{\pm} : = C_{\alpha_{\pm}} = 2 \cdot (2 + 1) |\alpha_{\pm}|^{2}
= 6 (a^{2} + 1)$. Since $C_{1}, C_{2}$ satisfy , , we have $$\begin{aligned}
& a C_{1} + \frac{1}{a^{3}} C_{2}
- \frac{1 - a^{2}}{8 a^{3}} 6(a^{2} + 1)
= 0,
\qquad
a C_{1} + \frac{1}{a^{5}} C_{2}
- \frac{(1 - a^{2})(a^{2} + 1)}{32 a^{5}} 6(a^{2} + 1)
= 0
\notag
\\
& \Leftrightarrow \qquad
C_{1} =
\frac{3}{16 a^{2}} (a^{2} + 1) (3 a^{-2} - 1),
\qquad
C_{2} = \frac{3}{16} (a^{2} + 1) (3 a^{2} - 1).
\notag$$ By these equations, $a \not = \pm \sqrt{3}, \pm 1/\sqrt{3}$, since $C_{1}, C_{2} \not= 0$.
For $\alpha \in \mathcal{H}$, let $u_{\alpha}$ be a function of the form $$\begin{aligned}
& u_{\alpha} = u_{\alpha}(x_{\alpha}),
&
& u_{\alpha}(t)
= \frac{C_{\alpha}}{t^{2}}
+ \mbox{(real analytic at $t=0$)}, \end{aligned}$$ and consider the equation $[L, P] = 0$ for $$L =
- (\partial_{x_{1}}^{2} + \partial_{x_{2}}^{2})
+
\sum_{\alpha \in \mathcal{H}} u_{\alpha}.$$
If $a$ is generic, $C_{1}, C_{2}$ are not of the form $k(k+1) |\alpha_{i}|^{2}$ ($k \in \boldsymbol{Z}$) for $i = 1, 2$. Therefore, the principal symbol $\widetilde{P}_{0}$ of $P$ is an even polynomial in $\xi_{1}$ and $\xi_{2}$ because of Proposition \[proposition:review-2\] (1). Moreover, since the coupling constants for $\alpha_{\pm}$ are two, $\widetilde{P}_{0}$ must satisfy $\lim_{\xi_{\alpha_{\pm} \to 0}}
\partial_{\xi, \alpha_{\pm}} \widetilde{P}_{0}
=
\lim_{\xi_{\alpha_{\pm} \to 0}}
\partial_{\xi, \alpha_{\pm}}^{3} \widetilde{P}_{0}
= 0$ because of Proposition \[proposition:review-2\] (2). Such a polynomial of degree six is unique up to constant multiple and modulo $(\xi_{1}^{2} + \xi_{2}^{2})^{3}$. We choose $$\widetilde{P}_{0}
=
a(4 - a^{2}) \xi_{1}^{6}
+ 5a \xi_{1}^{4} \xi_{2}^{2}
+ 5 a^{-1} \xi_{1}^{2} \xi_{2}^{4}
+ a^{-1}(4 - a^{-2}) \xi_{2}^{6}.$$
\[proposition:new potential functions\] If $a^{2} \not= 7/3, 3/7, (13 \pm 4 \sqrt{10})/3$, then we have $$\begin{aligned}
&
u_{1}(t)
= \frac{3(3 a^{-2} - 1) (a^{2} + 1)}{4} \wp(2 a t),
&
&
u_{2}(t)
= \frac{3 (3 a^{2} - 1) (a^{2} + 1)}{4} \wp(2 t),
&
&
u_{\pm}(t)
= 6 (a^{2} + 1) \wp(t), \end{aligned}$$ modulo constant factors.
<span style="font-variant:small-caps;">Sketch of proof.</span> Let us consider the equation . Even if the potential functions $u_{\alpha}$ are not rational, we can show that $F(x, \xi)$ in is a polynomial in $\xi$ of degree two. Since $[\partial_{x_{2}} \partial_{\xi_{1}}
- \partial_{x_{1}} \partial_{\xi_{2}},
\langle \xi, \partial_{x} \rangle]
= 0$, we have $$\label{eq:compatibility condition for P4}
(\partial_{x_{2}} \partial_{\xi_{1}}
- \partial_{x_{1}} \partial_{\xi_{2}})^{3}
\left(
\sum_{\alpha, \beta \in \mathcal{H} \atop \alpha \not= \beta}
u_{\alpha}' u_{\beta} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\beta}
\right)
= 0.$$ Here, we used $\partial_{\xi, \alpha^{\bot}}^{3} \partial_{\xi, \alpha}
\widetilde{P}_{2}^{\alpha} = 0$ for any $\alpha \in \mathcal{H}$, which is a direct consequence of $\lim_{\xi_{\alpha} \to 0} \partial_{\xi, \alpha}^{3}
\widetilde{P}_{0}
= 0$.
Choose $\alpha_{0} \in \mathcal{H}$. By taking the limit $\lim_{x_{\alpha_{0}} \to 0} x_{\alpha_{0}}^{6}
\times \eqref{eq:compatibility condition for P4}$, we obtain $$\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
(\partial_{\xi, \alpha_{0}^{\bot}}^{3} \partial_{\xi, \alpha_{0}}
\widetilde{P}_{2}^{\beta})
u_{\beta} (\langle \alpha_{0}^{\bot}, \beta \rangle t) = 0,$$ Similarly, we obtain $$\sum_{\beta \in \mathcal{H} \atop \beta \not= \alpha_{0}}
\partial_{\xi, \alpha_{0}^{\bot}}
(\partial_{\xi, \alpha_{0}}^{3} \widetilde{P}_{2}^{\beta}
-
C_{\alpha_{0}}
\partial_{\xi, \alpha_{0}} \widetilde{P}_{4}^{\alpha_{0}, \beta})
u_{\beta} (\langle \alpha_{0}^{\bot}, \beta \rangle t) = 0$$ from . These equations imply $$\begin{aligned}
& (3 a^{-2} - 7) \{u_{+}(t) - u_{-}(t)\} = 0,
\\
&
(3 a^{2} - 7) \{u_{+}(-at) - u_{-}(at)\} = 0,
\\
& 2 a^{2} u_{1}(-t) - 2 u_{2}(a t) + (a^{2} - 1) u_{-}(2 a t) = 0,
\\
&
2 a^{2} u_{1}(-t) - 2 u_{2}(-a t) + (a^{2} - 1) u_{+}(- 2 a t) = 0,
\\
& 3 a (a^{2} - 1) (a^{2} + 1)^{2}
(3 a^{4} - 26 a^{2} + 3)
\{8 a^{2} u_{1}(t) + (a^{2} - 3) u_{+}(2 a t)\}
= 0. \end{aligned}$$ Here, we have abbreviated $u_{\alpha_{1}}$ to $u_{1}$ etc. If $a^{2} \not= 7/3, 3/7, (13 \pm 4 \sqrt{10})/3$, we obtain from these that $u_{+}(t)$ is an even function, $u_{1}(t) = (3 a^{-2} - 1) u_{+}(2 a t) / 8$, $u_{2}(t) = (3 a^{2} - 1) u_{+}(2 t) / 8$ and $u_{-}(t) = u_{+}(t)$.
Finally, we can show $u_{+}(t) = 6(a^{2} + 1) \wp(t)$ by the same method as in §7 of [@OS], namely, by studying the coefficients in the Laurent expansion of as a function of $x_{\alpha_{+}}$. $\square$
For such potential function, we can construct a commutant $P$ of $L$. Since we can check the commutativity by a direct method, we omit the proof and write the conclusion only.
Before the statement of theorem, we introduce some notation. Let $g_{2}$, $g_{3}$ be the invariants of $\wp$ appearing in the differential equation $(\wp')^{2} = 4 \wp^{3} - g_{2} \wp - g_{3}$. As above, we abbreviate $u_{\alpha_{1}}$ to $u_{1}$ etc. We put $$\begin{aligned}
& L_{1} := \partial_{x_{1}}^{2} - u_{1},
&
& L_{2} := \partial_{x_{2}}^{2} - u_{2},
&
& L_{\pm} := \partial_{x, \alpha_{\pm}}^{2} - (a^{2} + 1) u_{\pm} \end{aligned}$$ and $$A_{\pm}(5)
:= \partial_{x, \alpha_{\pm}}^{5}
- \frac{5}{2} u_{\pm} \partial_{x, \alpha_{\pm}}^{3}
- \frac{15}{4} u_{\pm}' \partial_{x, \alpha_{\pm}}^{2}
+ \frac{1}{8} (a^{2} + 1)^{2}
\{15 u_{\pm}^{2} - 25 (a^{2} + 1) u_{\pm}''\}
\partial_{x, \alpha_{\pm}}.$$ Note that $- (\partial_{x_{1}}^{2} + \partial_{x_{2}}^{2}) + u_{\pm}$ commutes with $L_{\pm}$, since $\partial_{x_{1}}^{2} + \partial_{x_{2}}^{2}
= (a^{2} + 1)^{-1}
(\partial_{x, \alpha_{\pm}}^{2}
+ \partial_{x, \alpha_{\pm}^{\bot}}^{2})$. Moreover, since the coupling constants for $\alpha_{\pm}$ are two, $L_{\pm}$ has a commutant of order five ([@BC]). The operator $A_{\pm}(5) - (21/8) (a^{2} + 1)^{4} g_{2}
\partial_{x, \alpha_{\pm}}$ is such a commutant.
\[theorem:new commutative pair\] Let $$L =
- (\partial_{x_{1}}^{2} + \partial_{x_{2}}^{2})
+ u_{1} + u_{2} + u_{+} + u_{-}.$$
Then the following operator commutes with $L$. $$\begin{aligned}
P =&
a(4-a^{2}) L_{1}^{3} + 5 a L_{1}^{2} L_{2} + 5 a^{-1} L_{1} L_{2}^{2}
+ a^{-1} (4 - a^{-2}) L_{2}^{3} + P_{2}
\\
& \quad
+
P_{4}
+
\frac{1}{2} \langle \partial_{x}, \partial_{\xi} \rangle
\left.
\widetilde{P}_{4}
\right|_{\xi \to \partial_{x}}
+ \frac{1}{8}
\langle \partial_{x}, \partial_{\xi} \rangle^{2}
\widetilde{P}_{4} + Q_{4}
+ P_{6}, \end{aligned}$$ where $$\begin{aligned}
P_{2} :=& \frac{1}{(a^{2} + 1)^{4}}
[
- 20 a (a^{2} + 1)
(u_{+} \partial_{x, \alpha_{+}^{\bot}}^{4}
+ u_{-} \partial_{x, \alpha_{-}^{\bot}}^{4})
\\
& \hspace{17mm} - 10 a^{-1} (a^{-2} - 4 + a^{2})
\{
(L_{+}^{2} - \partial_{x, \alpha_{+}}^{4})
\partial_{x, \alpha_{+}^{\bot}}^{2}
+
(L_{-}^{2} - \partial_{x, \alpha_{-}}^{4})
\partial_{x, \alpha_{-}^{\bot}}^{2}
\}
\\
& \hspace{17mm}
- 2 (a^{2} - 1) (3 a^{-2} - 1) (3 a^{2} - 1)
\{
(A_{+}(5) - \partial_{x, \alpha_{+}}^{5})
\partial_{x, \alpha_{+}^{\bot}}
- (A_{-}(5) - \partial_{x, \alpha_{-}}^{5})
\partial_{x, \alpha_{-}^{\bot}}
\}
\\
& \hspace{17mm}
- a (a^{4} - 6 a^{2} + 6 - 6 a^{-2} + a^{-4})
\{
(L_{+}^{3} - \partial_{x, \alpha_{+}}^{6})
+
(L_{-}^{3} - \partial_{x, \alpha_{-}}^{6})
\}
],
\\
P_{4}
= &
u_{1} (u_{+} + u_{-})
\{
6a (4 - a^{2}) \partial_{x_{1}}^{2}
+ 7 (a + a^{-1}) \partial_{x_{2}}^{2}
\}
+ u_{1} (u_{+} - u_{-}) (3 a^{2} - 7)
\partial_{x_{1}} \partial_{x_{2}}
\\
& +
u_{2} (u_{+} + u_{-})
\{
7 (a + a^{-1}) \partial_{x_{1}}^{2}
+ 6 a^{-1} (4 - a^{-2}) \partial_{x_{2}}^{2}
\}
+ u_{2} (u_{+} - u_{-}) (3 a^{-2} - 7)
\partial_{x_{1}} \partial_{x_{2}}
\\
& +
\frac{1}{8}
(u_{+} + u_{-})
\{
(35 a^{-1} + 156 a - 39 a^{3}) \partial_{x_{1}}^{2}
+
(35 a + 156 a^{-1} - 39 a^{-3}) \partial_{x_{2}}^{2}
\},
\\
Q_{4}
= &
- \frac{21}{16}
(a^{2} + 1)^{2} (a - a^{-1}) (3 a^{2} - 1) (3 a^{-2} - 1) g_{2}
\left(
L_{1} - L_{2}
- \frac{a^{2} - 1}{a^{2} + 1} (u_{+} + u_{-})
\right) \end{aligned}$$ and $$\begin{aligned}
P_{6} = &
3 a (4 - a^{2}) (u_{1}'' - u_{1}^{2}) (u_{+} + u_{-})
+
3 a^{-1} (4 - a^{-2}) (u_{2}'' - u_{2}^{2}) (u_{+} + u_{-})
\\
& + \frac{1}{8}
\{
(15 a^{3} - 72 a - 7 a^{-1}) u_{1}
+
(15 a^{-3} - 72 a^{-1} - 7 a) u_{2}
\}(u_{+}^{2} + u_{-}^{2})
\\
& +
\frac{a}{4}
\{
(20 a^{4} - 83 a^{2} + 24 + 7 a^{-2}) u_{1}
+
(20 a^{-4} - 83 a^{-2} + 24 + 7 a^{2}) u_{2}
\}
(u_{+}'' + u_{-}'')
\\
& +
\frac{3}{16}
(a + a^{-1}) (7 a^{2} - 38 + 7 a^{-2})
(u_{+}^{2} u_{-} + u_{+} u_{-}^{2})
- \frac{5}{32} a (5 a^{4} - 202 + 5 a^{-4})
(u_{+}'' u_{-} + u_{+} u_{-}'')
\\
& + \frac{3}{16} a (a^{2} - a^{-2})
(19 a^{2} - 122 + 19 a^{-2}) u_{+}' u_{-}'
- 7 (a + a^{-1}) u_{1} u_{2} (u_{+} + u_{-})
\\
& + \frac{1}{8}
\{
a (57 a^{2} - 216 + 7 a^{-2}) u_{1}
+
a^{-1} (57 a^{-2} - 216 + 7 a^{2}) u_{2}
\} u_{+} u_{-}. \end{aligned}$$
[99]{}
Burchnall, J.L. and Chaundy, T.W.: Commutative ordinary differential operators. Proc. London Math. Soc. (2) [**21**]{} (1922), 420–440.
Chalykh, O., Feigin, M. and Veselov, A.: New integrable generalization of Calogero-Moser quantum problem. J. Math. Phys. [**39**]{} (1998), 695–703.
Oshima, T. and Sekiguchi, H.: Commuting families of differential operators invariant under the action of a Weyl group. J. Math. Sci. Univ. Tokyo, [**2**]{} (1995), 1–75.
Taniguchi, K.: On the symmetry of commuting differential operators with singularities along hyperplanes, Int. Math. Res. Notices 2004, no. 36, 1845-1867.
Veselov, A.P., Feigin, M. and Chalykh, O.A.: New integrable deformations of the quantum Calogero-Moser problem. Russian Math. Surveys [**51**]{} (1996), No. 3, 573-574.
[^1]: Department of Physics and Mathematics, Aoyama Gakuin University, 5-10-1, Fuchinobe, Sagamihara, Kanagawa 229-8558, Japan. ([email protected])
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a novel method for active continuous-tuning of a band gap which has the great potential to revolutionalize current photonic technologies. We study a periodic structure of x and y-aligned nanorod dimers. Refractive index of an y-polarized probe pulse can be continuously-tuned by the intensity of an x-polarized auxiliary (pump) pulse. Order of magnitude index-tuning can be achieved with a [*vanishing loss*]{} using the plasmon-analog of refractive index enhancement \[PRB 100, 075427 (2019)\]. Thus, a large band gap can be created from a non-existing gap via the auxiliary pulse. The new method, working for any crystal dimensions, can also be utilized as a [*linear*]{} photonic switch operating at tens of femtoseconds.'
author:
- Emre Yüce
- Zafer Artvin
- Ramazan Sahin
- Alpan Bek
- Mehmet Emre Tasgin
bibliography:
- 'bibliography.bib'
title: 'Ultra-large actively tunable photonic band gaps via plasmon-analog of index enhancement'
---
[^1]
Ability of crafting materials at dimensions comparable to wavelength of light not only allowed control of electromagnetic radiation at nanoscale, but also made development of metamaterials [@liu2011metamaterials], such as photonic crystals (PCs) [@yablonovitch1987inhibited; @yablonovitch1991photonic; @Wijnhoven1998], possible. PCs, periodically altering refractive index materials, reshape the propagation (dispersion) of the incident light and even can forbid it at certain propagation directions within some frequency ranges, named as photonic band gaps (PBGs) [@Leistikow2011]. PBGs, whose structure depends on the periodicity and index contrast, can be utilized as waveguides [@knight1996all], optical cavities [@akahane2003high], optical memories [@kuramochi.2014] and switches [@euser2007ultrafast; @yuce2013all]. This way PBGs enable on-chip photonic signal processing [@wang2018chip]. These structures can even enable control over quantum phenomena [@haroche1989cavity; @lodahl2004controlling; @madsen2011observation] such as spontaneous emission [@scully2015single; @scully2009super] and entanglement [@tasgin2017many] at nanoscale.
A conventional PC —whose PBG structure remains fixed after manufacturing— cannot operate, e.g., as a switch, although it is useful for photonic waveguide and cavity applications. External control of the constituent material’s refractive index, however, demonstrated to tune the PBG width about a few percent which allows the utilization of PCs as photonic switches. Refractive index of semiconductors (materials commonly employed in current PC technologies owing to their relatively-high index, $n$=3.5-4, and small absorption) can be tuned externally (actively) via optical heating [@reed2004silicon], free-carrier excitation [@almeida2004all] and electronic Kerr-effect [@ctistis2011ultimate; @yuce2013all]. While technique of free-carrier excitation provides the largest index change ($\sim$3%), its response time (tens of picoseconds) is much longer compared to the modulation time of the Kerr effect (a few femtosecond). Using the Kerr effect, in contrast, achievable index change is only $\sim$1%.
Metal nanostructures, which can localize the incident light into nm-sized hotspots, are also utilized for photonic technologies. Hybrid metastructures, created with semiconductors and metallic nanostructures, increase the versatility integrated photonic structures [@fan2015dynamic; @he2019tunable; @kang2019recent]. Yet, regarding the modulation depth, response time and metal-induced losses their functionalities are limited [@fan2015dynamic; @he2019tunable; @kang2019recent]. Therefore, a mechanism providing; (i) a large refractive index change (ii) minimal loss, (iii) short response time, and (iv) a large modulation depth will revolutionalize the dynamic control of light on-a-chip. Faithfully, in this paper, we propose the utilization of a recently-discovered extraordinary index enhancement (control) scheme for achieving a game-changing control over the PGBs of a PC.
While plasmon analogs of electromagnetically induced transparency (EIT) [@Atac2001controlling; @boller1991observation; @ScullyZubairyBook], in the linear (called as Fano resonances [@luk2010fano; @limonov2017fano; @peng2014and]) and nonlinear response [@butet2014fano; @singh2016enhancement; @TasginFanoBook2018; @postaci2018silent] have been demonstrated before; demonstration of the plasmon analog of index enhancement [@scully1991enhancement] could be achieved just recently [@LavrinenkoPRB2019]. Lavrinenko and colleagues [@LavrinenkoPRB2019] demonstrate that polarization response of a y-aligned silver nanorod to a y-polarized probe pulse can be controlled by the presence of an x-polarized auxiliary pulse, see Fig. \[fig1\]. Owing to the shape resonance (selective coupling [@andrews2019comprehensive]) of nanorods; x,y-polarized pulse can excite “only" the x,y-aligned nanorods, respectively. Beyond studying the enhancement using a basic analytical model, which has widely been demonstrated to work very well for plasmonic path interference effects [@Pelton2010OptExp; @lovera2013mechanisms; @TasginFanoBook2018], Lavrinenko and colleagues also demonstrated the phenomenon via exact solutions of 3D Maxwell equations. This scheme not only enhances the refractive index more than 2-orders of magnitude, but also results a canceled absorption at the enhancement frequency —just as the same in its EIT counterpart [@fleischhauer2005electromagnetically]. This phenomenon allows us to adjust the “loss-free" refractive index of metamaterials without sacrificing the probe signal to metallic losses.
In this paper, we show that plasmonic index enhancement is the dream-spouse for a photonic crystal. While other methods [@reed2004silicon; @almeida2004all; @ctistis2011ultimate; @yuce2013all] can tune the band gap size only about 2-3%, the presented scheme can *create* a large (e.g. 0.3 and 0.56 eV in Fig. \[fig3\]) from a nonexisting gap only with a 6% filling ratio [^2]. One can achieve “any particular value" of the band gap width between 0-0.3 eV for any y-polarized probe pulse ($E_1$) by merely arranging the [*intensity*]{} ($\sim |E_2|^2$) of the x-polarized auxiliary (pump) pulse between $E_2$=0-10$E_1$, working in the low intensity regime. $|E_1|^2$ is the intensity of the weak probe field. Moreover, response time is limited within the plasmon lifetime ($\sim$ fs [@zoric2011gold]) and relation time (referred as $1/g$) since the scheme utilizes metal nanoparticles. In the case of silver nanorods, Ref. [@LavrinenkoPRB2019] originally studies, a 100 fs linear photonic switching can be achieved using 30 fs auxiliary pulse, see Fig. \[fig5\], where silver’s plasmon lifetime we use is already 40 fs long. It is apparent that using a small plasmon lifetime material, such as platinum [@zoric2011gold; @yildiz2020plasmon], should result much faster switching times [^3].
Taking also the advantage of lossless medium, modulation depths of 110 dB ($10^{11}$) can be achieved using only 5 layers of silver dimers, one dimer in the 10$\times$100$\times$100 volume of each index-enhanced region. This is because, supreme index enhancement can be achieved. The sample setup, plotted in Fig. \[fig1\], presents a 750 nm long enhanced photonic crystal on a PC-waveguide chip employing CMOS-compatible noble metal (silver) materials.
Though we content this prominent work with propagation through a single direction [^4], since we do not aim an exact simulation of an already conducted experiment, the method is applicable to crystal dimension obviously. (See also the discussion in the following text.) Index enhancement scheme, appearing via the coupling of the two orthogonal nanorods, can be obtained also on easier-to-manufacture structures.
Utilizing index-enhanced materials for periodic structures, actually, is not a new idea [@tasgin2007photonic; @mustecapliouglu2005photonic], employed cold atom systems. Implementation of the plasmon analog of the phenomenon, however, provides a feasible “loss-free" backbone for PC technologies. Plasmon index enhancement has been exploited in Ref. [@Cherenkov_Nanophotonics_2020] for achieving a continuously-tunable Cherenkov-radiation-based particle detector, ignoring the periodicity effects.
[**Index-enhancement scheme.**]{}— Fig. \[fig1\] demonstrates a sample setup for implementing index-enhancement phenomenon for on-chip photonic operations. 5 layers of silver nanorod dimers (see the inset) switches the transmission of a $\rm y$-polarized probe pulse ($E_1$) of frequency $\omega$ via an x-polarized auxiliary (pump) pulse ($E_2$). When the auxiliary pulse is off, the refractive index of $V_{\rm rod}$=10$\times$10$\times$30 sized nanorods becomes exceedingly small for a density of 1 dimer in volume $V_{\rm site}$=150$\times$100$\times$100 ${\rm nm}^3$. Here 150 nm is the periodicity and 100 nm corresponds to width and depth of the cavity in which silver dimers are placed. Filling ratio of the index-enhanced layer, a single dimer in each one, is 10/150 $\simeq$ 6%. Thus, probe is transmitted with a relatively small absorption, i.e., 35%. (Probe can be transmitted also without any absorption, see below.) When the auxiliary pulse is turned on, with amplitudes $E_2=10E_1$ and $100E_1$, however, effective refractive index of the thin layers (10 nm length) is enhanced dramatically, see Fig. \[fig2\]a, thus turning ON a PBG of width $\sim$0.3 eV and $\sim$0.56 eV, respectively. See Fig \[fig3\]a. The x-polarized auxiliary pulse couples only to the x-aligned silver nanorod and x-aligned nanorod affects the response of the y-aligned nanorod via a coupling through the hotspot appearing at the intersection of the two orthogonal rods. When the auxiliary pulse is present, y-polarized nanorod gives different polarization responses ($\chi_1=P_1/E_1$) for different values of the probe field $E_1$.
Dynamics of such a coupled system can be described by a basic model [@LavrinenkoPRB2019; @Cherenkov_Nanophotonics_2020] $$\begin{aligned}
\ddot{x}_1+\gamma_1 \dot{x}_1 + \omega_1^2 x_1 - gx_2 = \tilde{E}_1(t), \label{x1}
\\
\ddot{x}_2+\gamma_2 \dot{x}_2 + \omega_2^2 x_1 - gx_1 = \tilde{E}_2(t), \label{x2}\end{aligned}$$ which is very successful in demonstrating the Fano-like path-interference effects in plasmonic nanostructures [@Pelton2010OptExp; @lovera2013mechanisms; @TasginFanoBook2018]. FDTD simulations in Ref. [@LavrinenkoPRB2019], also demonstrate the phenomenon. $\gamma_1=\gamma_2=0.026\omega_0^2$, $\omega_1=\omega_2=\omega_0$ and $g=0.06\omega_0$ [@LavrinenkoPRB2019] are the damping rates, resonances and the coupling between the two plasmon oscillations of the two nanorod, respectively. $\omega_0$ is the resonance of both nanorods. Here, only $\gamma_{1,2}$ is different from the one of Ref. [@LavrinenkoPRB2019] since we determine the damping rates of nanorods from experimental dielectric function of silver. Actually, such a tidy choice of parameters is not necessary for this initial work which aims to present the implementation of the phenomenon for photonic technologies.
For a monochromatic auxiliary pulse $E_2 \sim e^{-i\omega t}$, of the same frequency with the probe, the exact solution for the susceptibility of the medium $$\chi(\omega)=f\omega_0^2\frac{\delta_2+ge^{-i\phi}E_2/E_1}{\delta_1\delta_2-g^2}
\label{chi}$$ can be obtained [@LavrinenkoPRB2019; @Cherenkov_Nanophotonics_2020]. The dimensionless oscillator strength $f$ is determined by setting nonenhanced susceptibility $\chi(\omega=\omega_0,g=0)=f/(\gamma_1/\omega_0)$ [@Cherenkov_Nanophotonics_2020] to $P_1/E_1$, with $P_1$ is the polarization density calculated for ellipsoid nanorods [@LavrinenkoPRB2019; @bohren2008absorption; @Cherenkov_Nanophotonics_2020], for 1 dimer in index-enhanced layer (10 nm), with volume $V_{\rm layer}$=10$\times$100$\times$100 ${\rm nm}^3$ in Fig. \[fig1\].
Since the auxiliary field, in our case an ultrashort pulse, has a broad frequency band, we take $\tilde{E}_2=\sum_\omega E_2 e^{-i\phi} \exp[{-(\omega-\Omega)^2/(\Delta\omega)^2}] e^{-i\omega t}$ pulse unlike Ref. [@LavrinenkoPRB2019]. We solve Eqs. (\[x1\]-\[x2\]) in the frequency domain for the Gaussian auxiliary pulse. Fig. \[fig2\] plots the real and imaginary parts of the enhanced index [^5] for $E_2=100E_1$, $10E_1$ and $0$. We note that a double resonance for $E_2=0$ appears, because coupling between the two nanorods exists when the auxiliary pulse is turned off.
In Fig. \[fig2\]b, we observe that imaginary part of the index becomes zero almost at the same frequency (vertical dashed-line) for $E_2=100E_1$ and $10E_1$, where the index becomes superiorly enhanced.
![(a) Real and (b) imaginary parts of the refractive index. Index of the y-polarized probe pulse is controlled by the intensity of the auxiliary pulse ($E_2$). Index is both enhanced and real at $\omega_{\rm enh}$=0.37$\times2\pi c/a$. ($a$=150 nm) Without the periodicity, probe pulse is amplified, e.g., to the left of $\omega_{\rm enh}$. []{data-label="fig2"}](fig2){width="47.00000%"}
[**The phase plate.**]{}— Setup in Fig. \[fig1\] includes a critical apparatus; a phase plate. Eq. (\[chi\]) shows that the phase-difference between the auxiliary and probe pulse, $e^{-i\phi}$, plays an important role for the index enhancement to appear. The aux-probe phase-differences for the first and second dimer are not the same, since the probe does not have the same phase at different dimers (x-positions). Hence, when the first dimer is set to enhanced index (for y-polarized probe), the second dimer can possess a non-enhanced index; destroying the periodicity. Thus, we rephase the front of the auxiliary pulse with a phase plate either in the entrance of the aux port of the waveguide, or in front of (or inside) he auxiliary source. [**Such a necessity for phase-tuning, most probably, is the reason large PBG changes cannot be observe in experiments with periodic nanoparticles [@fan2015dynamic; @he2019tunable; @kang2019recent]**]{}.
[**Photonic bands.**]{}— In Fig. \[fig3\]a, we present the transmission and reflection for the 5-layer index-enhanced PC of dimers of periodicity $a$=150 nm and filling ratio of 10/150$\sim$6%. It is striking that: while there exists a large amount of gain in the left of $\omega_{\rm enh}\simeq 0.37\times 2\pi c/a$ in Fig. \[fig2\]b, Fig. \[fig3\]a shows that PGB, the 5-layer enhanced PC creates, can turn-OFF the transmission in a wide frequency region both for $E_2=100E_1$ and $10E_1$. We note that, in the figures we scale the frequencies by $\omega_a=2\pi c/a$ and momentum-vector by $k_a=2\pi /a$ with $a$=150 nm. In determining the natural band gap region, we first calculate the band diagram for the constant index $n=n(\omega_{\rm enh})$, then we choose periodicity $a$ accordingly as performed in Ref. [@tasgin2007photonic].
![(a) Transmission and (b) reflection from only 5 layers of silver nanorods crystal (Fig. \[fig1\]). $E_2=100E_1$ and $10E_1$ display photonic band gaps even in the left of $\omega_{\rm enh}$=0.37 (vertical line) where there is a strong gain (Fig. \[fig2\]b). The width and existence of the band gap can be continuously-tuned by the auxiliary pulse intensity. []{data-label="fig3"}](fig3){width="47.00000%"}
We also calculate the photonic band structure in Fig. \[fig4\], which appears quite similar to the one for index enhancement in multilevel atomic ensembles \cite{}. The modulation depth for the 5$\times$150 =750 nm enhanced PC, at $\omega_{\rm enh}\simeq$0.37 calculated by transfer matrix method [@li2005principles] (Fig. \[fig3\]) and via band structure damping $(e^{ik_I a})^5$, are similar, where $k_I$ is the imaginary part of the wavevector in Fig. \[fig4\] at $\omega_{\rm enh}$. Since the enhanced PC dimensions are quite small, modulation depth can be increased by setting more layers into the enhanced PC.
![ Photonic band diagrams for a periodic structure of silver nanorod dimers (Fig. \[fig1\]) with a filling ration of 6.2%. For $\omega=\omega_{\rm enh}$=0.37 (vertical line), index is real (Fig. \[fig2\]b) and $E_2=100E_1$ and $10E_1$ display a band gap. Band diagrams are similar to the one obtained for index enhancement in atomic ensembles [@tasgin2007photonic; @ScullyZubairyBook]. []{data-label="fig4"}](fig4){width="47.00000%"}
[**PC operation regimes.**]{}— There can be several options for the operation regions in utilizing the system as a photonic switch. (a) PC can be operated at $\omega=\omega_{\rm enh}$, but transmission in the switch on mode drops to 0.15 for this regime, see Fig. \[fig3\]a. (b) One can choose a better switch-on transmission 0.55 for $\omega\simeq 0.38$, since absorption in the switch-off mode is not important for this application. (c) Index-enhancement scheme is an active medium. It displays a substantial gain, e.g., between $\omega$=0.35-0.37, when the auxiliary (pump) field is set to $E_2=10E_1$. Thus, working, for instance, $\omega=0.36$ enhances the transmitted pulse several orders of magnitude for $E_2=10E_1$ via gain, but, suppresses the transmission when auxiliary is turned to $E_2=100E_1$ due to the band gap. Thus, the modulation depth between switch-on (former case) and switch-off (latter case) can be increased further, actually, leading to unimaginable values. While we demonstrate the switch mechanism for quite different values of the auxiliary field, i.e., $E_2=100E_1$ and $10E_1$, an optimization can be carried out for smaller differences.
[**Response time.**]{}— In Fig. \[fig5\], we present the time development of the transmitted pulse when a 30 fs, auxiliary pulse passes through the enhanced PC. We observe that, although the temporal width of the auxiliary pulse is 30 fs PBG remains off for about $\sim$200 fs. This limitation, actually, occurs due to two transient times. The decay time of the surface plasmon oscillations on silver, used in this work [@LavrinenkoPRB2019; @Cherenkov_Nanophotonics_2020], is already about 40 fs. The switch time is also determined by the relaxation time, appearing due to the interaction between the two nanorods [^6]. After $t\simeq$700 fs, we observe the gain amplification of the transmission (it reaches ....) during the PNG off regime. This transient also lasts $\sim$200 fs, but, increases the modulation depth between $t=$600 fs and 700 fs dramatically. When, for instance, platinum nanorods, of decay time $\sim$1 fs are used in the enhanced PC, much faster response times can be achieved. We remark that, a larger damping rate necessitates also a larger coupling $g$ for enhancement scheme to appear.
![When the applied auxiliary field $E_2$ is an ultrashort pulse ($\Delta t$=30 fs), the crystal of silver dimers behaves as a linear photonic switch. The transmission (switch) along the x-direction is shutdown when the auxiliary pulse is on the y-aligned nanorods in Fig. \[fig1\]. The switch remains off for $\simeq 200$ fs. Length of this interval, limited by the damping of the silver rod ($\tau \sim$ 38 fs), can be refines (shortened) substantially using a bad-quality nanostructure. For instance it is 1.3 fs in platinum [@yildiz2020plasmon]. []{data-label="fig5"}](fig5){width="47.00000%"}
[**Periodicity on 2D and 3D.**]{}— Index enhancement scheme depends critically on the phase difference between the probe and the auxiliary pulses. When there is periodicity of nanodimers along the z-axis in Fig. \[fig1\], there would appear no problem. Because, the phase of the auxiliary pulse is constant in the x-z plane. On working the periodicity along the propagation direction of the auxiliary pulse, y, however, one needs to take care of the different positions (so phases) of adjacent dimer arrays along the y-direction. Traveling along the y-directions, the first and the second dimer arrays show different phase-differences for the auxiliary and probe pulses. Index enhancement scheme can be different, even be demolished, in this case. This problem, however, can be circumvented in two ways. (i) One can choose the y-periodicity of the PC such that at a desired operation wavelength altering phases, $e^{i\phi_1}$ and $e^{i\phi_2}$, overlap. (ii) A second approach: once could induce an appropriate phase dilation also in the probe field along the y-axis either in the waveguide input or just before PC input. One important point to stress is: in the enhancement scheme, index can be superiorly strengthened such that a single dimer array becomes sufficient for switching operations.
[**In summary,**]{} we show that recently-explored index enhancement for plasmonic nanoparticles [@LavrinenkoPRB2019] can be utilized in photonic crystal technologies for achieving far-off-records photonic band gap tuning and modulation depths. This is because, available adjustment of refractive index, using this scheme, is incomparably larger than the currently applied methods [@reed2004silicon; @almeida2004all; @ctistis2011ultimate; @yuce2013all]. Index enhancement, achieved with a “vanishing loss", is such a large value (as in its EIT-counterpart [@tasgin2007photonic; @mustecapliouglu2005photonic; @fleischhauer1992resonantly]) that using very small filling ratio for nanoparticles becomes sufficient. Switching time of the PBG is determined by the decay rate of the metal nanoparticles, so, it is in the femtosecond regime. Vanishing loss at the enhancement frequency can be utilized as undamped, but can be turned on and off, cavities at specific frequencies important for quantum optics (information) applications [@reiserer2015cavity; @scully2003extracting; @hardal2015superradiant; @zou2017quantum].
Therefore, implementation of the plasmon-analog of refractive index enhancement (control) to photonic crystals provides a revolutionary instrument for photonic applications.
[^1]: Correspondence: [email protected]
[^2]: Filling ratio of a silver dimer in Fig. \[fig1\] is 0.1%.
[^3]: Using a material of larger damping also necessitates a larger coupling strength $g$ between the two nanorods, thus a smaller relaxation time, for the enhancement scheme to be observable.
[^4]: Structure need not be one-dimensional. We only consider a single propagation direction by calculating an average polarization density on the other periodicity directions, if exist. In Fig. \[fig1\], there are only 5 silver nanodimers and no periodicity assumed in other directions.
[^5]: We note that sign of the index is determined by choosing the real part of index positive, similar to performed in EIT counter-part of the index enhancement [@fleischhauer1992resonantly].
[^6]: In a coupled system, transient time is determined by the minimum of the damping rates.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Zero-Knowledge Proof-of-Identity from trusted public certificates (e.g., national identity cards and/or ePassports; eSIM) is introduced here to permissionless blockchains in order to remove the inefficiencies of Sybil-resistant mechanisms such as Proof-of-Work (i.e., high energy and environmental costs) and Proof-of-Stake (i.e., capital hoarding and lower transaction volume). The proposed solution effectively limits the number of mining nodes a single individual would be able to run while keeping membership open to everyone, circumventing the impossibility of full decentralization and the blockchain scalability trilemma when instantiated on a blockchain with a consensus protocol based on the cryptographic random selection of nodes. Resistance to collusion is also considered.
Solving one of the most pressing problems in blockchains, a zk-PoI cryptocurrency is proved to have the following advantageous properties:
- an incentive-compatible protocol for the issuing of cryptocurrency rewards based on a unique Nash equilibrium
- strict domination of mining over all other PoW/PoS cryptocurrencies, thus the zk-PoI cryptocurrency becoming the preferred choice by miners is proved to be a Nash equilibrium and the Evolutionarily Stable Strategy
- PoW/PoS cryptocurrencies are condemned to pay the Price of Crypto-Anarchy, redeemed by the optimal efficiency of zk-PoI as it implements the social optimum
- the circulation of a zk-PoI cryptocurrency Pareto dominates other PoW/PoS cryptocurrencies
- the network effects arising from the social networks inherent to national identity cards and ePassports dominate PoW/PoS cryptocurrencies
- the lower costs of its infrastructure imply the existence of a unique equilibrium where it dominates other forms of payment
**Keywords**: zero-knowledge, remote attestation, anonymous credentials, incentive compatibility, dominant strategy equilibria, Nash equilibria, Price of Crypto-Anarchy, Pareto dominance, blockchain, cryptocurrencies
author:
- |
David Cerezo Sánchez^^\
[[email protected]]{}
bibliography:
- '15\_home\_amnesia\_Persistent\_articleHaMafteach\_bib.bib'
title: '**Zero-Knowledge Proof-of-Identity**: [Sybil-Resistant, Anonymous Authentication on Permissionless Blockchains and Incentive Compatible, Strictly Dominant Cryptocurrencies]{}'
---
Introduction
============
Sybil-resistance for permissionless consensus comes at a big price since it needs to waste computation using Proof-of-Work (PoW), in addition to assuming that a majority of the participants must be honest. In contrast, permissioned consensus is able to overcome these issues assuming the existence of a Public-Key Infrastructure[@aaba; @cpps; @pbft] otherwise it would be vulnerable to Sybil attacks[@the-sybil-attack]: indeed, it has been recently proved[@cryptoeprint:2018:302] that consensus without authentication is impossible without using Proof-of-Work. Proof-of-Stake, the alternative to PoW, is economically inefficient because participants must keep capital at stake which incentivise coin hoarding and ultimately leads to lower transaction volume.
Another major challenge in permissionless blockchains is scalability, both in number of participants and total transaction volume. Blockchains based on Proof-of-Work are impossible to scale because they impose a winner-take-all contest between rent-seeking miners who waste enormous amounts of resources, and their proposed replacements based on Proof-of-Stake don’t exhibit the high decentralization desired for permissionless blockchains.
The solution proposed in this paper prevents Sybil attacks without resorting to Proof-of-Work and/or Proof-of-Stake on permissionless blockchains while additionally guaranteeing anonymous identity verification: towards this goal, zero-knowledge proofs of trusted PKI certificates (i.e., national identity cards and/or ePassports) are used to limit the number of mining nodes that a single individual could run; alternatively, a more efficient solution based on mutual attestation is proposed and demonstrated practical \[sub:Performance-Evaluation\]. Counterintuitively, the blockchain would still be permissionless even though using government IDs because the term “permissionless” literally means “without requiring permission” (i.e., to access, to join, ...) and governments would not be authorizing access to the blockchain; moreover, the goal is to be open to all countries of the world \[sub:Worldwide-Coverage-and\], thus its openness is indistinguishable from PoW/PoS blockchains (i.e., the union of all possible national blockchains equals a permissionless, open and global blockchain). Coincidently, the latest regulations [@chinaIdentityRegulation; @chinaRules] point to the obligation to verify and use real-world identities on blockchains, and the banning of contaminant cryptocurrency mining[@chinaBanMining; @NDRCbanlist].
Blockchain research has focused on better consensus algorithms obviating that incentives are a central aspect of permissionless blockchains and that better incentive mechanisms would improve the adoption of blockchains much more that scalability improvements. To bridge this gap, new proofs are introduced to demonstrate that mining a new cryptocurrency based on Zero-Knowledge Proof-of-Identity would strictly dominate previous PoW/PoS cryptocurrencies, thus replacing them is proved to be a Nash equilibrium; additionally, the circulation of the proposed cryptocurrency would Pareto dominate other cryptocurrencies. Furthermore, thanks to the network effects arising from the network of users of trusted public certificates, the proposed cryptocurrency could become dominant over previous cryptocurrencies and the lower costs of its infrastructure imply the existence of a unique equilibrium where it dominates other forms of payment.
Contributions
-------------
The main and novel contributions are:
- The use of anonymous credentials in permissionless blockchains in order to prevent Sybil attacks \[sec:Authentication-Protocols\]: previous works[@DBLP:journals/corr/KulynychITD17; @blockchainCA] considered the use of PKI infrastructures in blockchains (i.e., permissioned ledgers) but without transforming them into anonymous credentials in order to obtain the equivalent of a permissionless blockchain. Other works have considered anonymous credentials on blockchains[@cryptoeprint:2013:622; @coconut; @DBLP:conf/ccs/CamenischDD17; @quisquis; @indyAnonCreds], but requiring the issuance of new credentials and not reusing previously existing ones: verifying real-world identities and issuing their corresponding digital certificates is the most expensive part of any real-world deployment.
- The practical implementation and its perfomance evaluation \[sub:Performance-Evaluation\] for national identity cards and ePassports.
- Circumventing the impossibility of full decentralization \[sub:Circumventing-the-Impossibility\] and the blockchain scalability trilemma.
- A protocol for an incentive-compatible cryptocurrency \[alg:Incentive-Compatible-Protocol\]: previous blockchains mint cryptocurrencies tied to the process of reaching a consensus on the order of the transactions, but the game-theoretic properties of this mechanism is neither clear nor explicit.
- A proof that mining the proposed cryptocurrency is a dominant strategy over other PoW/PoS blockchains and a Nash equilibrium over previous cryptocurrencies \[sub:Strictly-Dominant-Cryptocurrenci\], in addition to an Evolutionary Stable Strategy \[sub:Evolutionary-Stable-Strategies\].
- The insight that the optimal efficiency of zk-PoI resides in that it’s implementing the social optimum, unlike PoW/PoS cryptocurrencies that have to pay the Price of (Crypto-)Anarchy \[sub:Price-of-Crypto-Anarchy\].
- A proof that the circulation of the proposed zk-PoI cryptocurrency Pareto dominates other PoW/PoS cryptocurrencies \[sub:Pareto-Dominance-on\].
- A proof that the proposed cryptocurrency could become dominant over previous ones due to stronger network effects and the lack of acceptance of previous cryptocurrencies as a medium of payment \[sub:On-Network-Effects\].
- Finally, the lower costs of its infrastructure imply the existence of a unique equilibrium where it dominates other forms of payment \[sub:Dominance-over-Cash\].
Related Literature
==================
This section discusses how the present paper is significantly better and more innovative than previous approaches in order to fulfill the objective of providing a Sybil-resistant and permissionless blockchain with anonymous transaction processing nodes (i.e., miners). Moreover, it’s considerably cheaper than other approaches[@coconut; @pop] that would require the re-identification and issuing of new identities to the global population because the current proposal relies on the previously issued credentials of electronic national identity cards (3.5 billion issued at the time of publication) and electronic passports (1 billion issued at the time of publication).
Proof of Space[@cryptoeprint:2013:796; @cryptoeprint:2013:805] reduces the energy costs of Proof-of-Work but it’s not economically efficient. Proof of Authority[@poa](PoA) maintains a public list of previously authorised nodes: the identities are not anonymised and the blockchain is not open to everyone (i.e., the blockchain is permissioned). Proof of Personhood[@pop](PoP) can be understood as an improvement over Proof of Authority in that identities are anonymised, but the parties/gatherings used to anonymise and incorporate identities into the blockchain don’t scale to national/international populations and could compromise Sybil resistance because it’s trivial to get multiple identities by using different disguises on different parties/gatherings (i.e., they need to be validated simultaneously and without disguises): however, the present paper produces Sybil-resistant, anonymised identities on a global scale for a permissionless blockchain. Moreover, Proof of Personhood[@pop] is endogenizing all the costly process of credential verification and issuing: by contrast, Zero-Knowledge Proof of Identity is exogenizing/outsourcing this costly process to governments, thus making the entire blockchain system cheaper. More recently, Private Proof-of-Stake protocols[@cryptoeprint:2018:1105; @cryptoeprint:2018:1132](PPoS) achieve anonymity, but the economic inefficiencies of staking capital still remain and the identities have no relation to the real world.
A conceptually close work (“*Decentralized Multi-authority Anonymous Authentication for Global Identities with Non-interactive Proofs*”, [@cryptoeprint:2019:701]), concurrently developed, doesn’t reuse real-world certificates and therefore it would require that governments re-issue the cryptographic credentials of their citizens: therefore, it doesn’t consider neither Sybil-resistance nor blockchain integration. Pseudo-anonymous signatures[@pseudoanonymousSignatures] for identity documents provide an interesting technical solution to the problem of anonymous authentication using identity documents. However, the proposed schemes present a number of shortcomings that discourage their use in the present setting: some schemes are closely tied to particular countries (i.e., the German Identity Card[@cryptoeprint:2012:558; @bsitr03110; @cryptoeprint:2018:1148]), thus non-general purpose enough to include any country in the world, or flexible to adapt to future changes; they require interaction with an issuer during card initialization; they feature protocols for deanonymisation and revocation, not desired in the setting considered in this paper; the initial German scheme[@bsitr03110] could easily be subverted[@10.1007/978-3-319-40367-0_31] because the formalization of pseudo-anonymous signatures is still incipient[@10.1007/978-3-319-49151-6_17], and improvements are being worked out[@cryptoeprint:2014:067; @cryptoeprint:2016:070; @cryptoeprint:2018:1148].
Anonymous credentials, first envisioned by David Chaum[@10.1007/978-1-4757-0602-4_18], and first fully realised by Camenisch and Lysyanskaya[@cryptoeprint:2001:019] with follow-up work improving its security/performance[@cryptoeprint:2009:107; @cryptoeprint:2005:060; @cryptoeprint:2010:496; @10.1007/978-3-540-28628-8_4; @cryptoeprint:2012:298], are a centrally important building block in e-cash. The use of anonymous credentials to protect against Sybil attacks[@the-sybil-attack] has already been proposed in previous works[@DBLP:conf/wistp/AnderssonKMP08; @cryptoeprint:2007:384] although with different cryptographic techniques and for different goals. The main problem with anonymous credentials is that they require a first identification step to an issuing party[@indyAnonCreds] and that would compromise anonymity. This problem is shared with other schemes for pseudoanymization: for example, Bitnym[@sybilPseudonym] requires that a Trusted Third Party must check the real identity of a user before allowing the creation of a bounded number of valid genesis pseudonyms. Decentralized Anonymous Credentials[@cryptoeprint:2013:622] was first to show how to decentralise the issuance of anonymous credentials and integrate them within a blockchain (i.e., Bitcoin), but they do not re-use previously existing credentials and they still rely on Proof-of-Work for Sybil-resistance. Decentralized Blacklistable Anonymous Credentials with Reputation[@cryptoeprint:2017:389] introduce blacklistable reputation on blockchains, but users must also publish their real-world identity (i.e., non-anonymous). QuisQuis[@quisquis] introduces the novel primitive of updatable public keys in order to provide anonymous transactions in cryptocurrencies, but it doesn’t consider their Sybil-resistance. DarkID[@darkID] is a practical implementation of an anonymous decentralised identification system, but requires non-anonymous pre-authentication and doesn’t consider Sybil-resistance. A previous work[@whoami] on secure identity registration on distributed ledgers achieved anonymity from a credential issuer, but the pre-authentication is non-anonymous, it doesn’t consider Sybil-resistance and it doesn’t re-use real-world cryptographic credentials. Recently, anonymous credentials on standard smart cards have been proved practical[@cryptoeprint:2019:460], but in a different setting where the credential issuer and the verifier are the same entity.
Previous works have also considered anonymous PKIs: for example, generating pseudonyms[@anonymousPKI] using a Certificate Authority and a separate Private Certificate Authority; however, this architecture is not coherent for a permissionless blockchain because both certificate authorities would be open to everyone and that would allow the easy linking of anonymous identities. Another recent proposal for a decentralised PKI based on a blockchain[@cryptoeprint:2018:853] does not provide anonymity, although it improves the work on cryptographic accumulators on blockchains started by Certcoin[@certcoin; @cryptoeprint:2015:718]; another proposals introduce privacy-aware PKIs on blockchains[@pbpki; @cryptoeprint:2019:527], but they are not Sybil-resistant and do not re-use certificates from other CAs. Previously, BitNym[@sybilPseudonym] introduced Sybil-resistant pseudonyms to Bitcoin, but a Trusted-Third Party must check the real identities of users before allowing the creation of a bounded number of valid genesis pseudonyms. ChainAnchor[@chainAnchor] wasn’t Sybil-resistant and used Direct Anonymous Attestation just for anonymous authentication, but not for mutual authentication: it worked on the permissioned model, explicitly not permissionless, and the GroupOwner initially knew the true identity of members; moreover, the Permissions Issuer is supposed not to collude with the Verifier, although it has reading access to the identity database. ClaimChain[@DBLP:journals/corr/KulynychITD17] improves the decentralised distribution of public keys in a privacy-preserving way with non-equivocation properties, but it doesn’t consider their Sybil-resistance because it’s more focused on e-mail communications. Blind Certificate Authorities[@WAPRS18; @cryptoeprint:2016:925] can simultaneously validate an identity and prove a certificate binding a public key to it, without ever learning the identity, which sounds perfect for the required scenario except that it requires 3 parties and it’s impossible to achieve in the 2-party setting; moreover, it doesn’t consider Sybil-resistance.
Other approaches to anonymous identity include: Lightweight Anonymous Subscription with Efficient Revocation[@cryptoeprint:2018:290], although it doesn’t consider the real-world identity of users because it’s focused on the host and its Trusted Platform Module; One Time Anonymous Certificates[@onetimeanonymouscertificate] extends the X.509 standard to support anonymity through group signatures and anonymous credentials, although it doesn’t consider Sybil-resistance and their group signatures require that users hold two group secret keys, a requisite that is not allowed in the current scenario because the user is not trusted to store them on the national identity card (for the very same reasons, Linkable Ring Signatures[@cryptoeprint:2004:027] and Linkable Message Tagging[@cryptoeprint:2014:014] are not allowed as cryptographic tools whilst group signatures and Deniable Anonymous Group Authentication[@daga] would require a non-allowed setup phase). Opaak[@opaak] provides anonymous identities with Sybil-resistance based on the scarcity of mobile phone numbers: however, users must register by receiving an SMS message (i.e., the Anonymous Identity Provider knows the real identity of participants). Oblivious PRFs[@cryptoeprint:2018:733] are not useful in the permissionless blockchains because the secret key of the OPRF would be known by everyone, and the forward secrecy of the scheme that would provide security even if the secret key is known would not be of any use because the object identifiers ObjID would be easily predictable (i.e., derived from national identifiers). SPARTA[@sparta] provides pseudonyms through a distributed third-party-based infrastructure; however, it requires non-anonymous pre-registration. UnlimitID[@unlimitid] provides anonymity to OAuth/OpenID protocols, although users must create keypairs and keep state between and within sessions, a requisite that is not allowed in the current scenario. Another proposal for anonymous pseudonyms with one Trusted-Third Party[@Yilek_traceableanonymous] requires a division of roles between the TTP and the server that is not coherent in a permissionless blockchain. With Self-Certified Sybil-free Pseudonyms[@MKAP08; @DBLP:conf/wistp/AnderssonKMP08], the user must keep state (i.e., dispenser D) generated by the issuer during enrollment and the Sybil-free identification is based on unique featurs of the devices, not on the user identity. Another anonymous authentication using smart cards[@smartcardAnonAuth] is only anonymous from an eavesdropping adversary, not from the authentication server itself. TATA provides a novel way to achieve Sybil-resistant anonymous authentication: members of an induction group must interact and keep a list of who has already been given a pseudonym; therefore, a list of participants could be collected, but they can’t be linked to their real-world identities; it’s not clear how to bootstrap the initial set of trusted users to get them to blindly sign each other’s certificates.
Self-sovereign identity solutions usually rely on identities from social networks, but their Sybil-resistance is very questionable because almost half of their accounts could be fake[@facebookNYT]: in spite of this, SybilQuorum[@sybilQuorum; @sybilQuorumArxiv] proposes the use of social network analysis techniques to improve their Sybil-resistance; other research projects consider privacy-preserving cryptographic credentials from federated online identities[@2014arXiv1406.4053M].
Regarding the game-theoretic aspects, most papers focus on attacking only one cryptocurrency (e.g., selfish mining[@DBLP:journals/corr/EyalS13], miner’s dilemma[@DBLP:journals/corr/Eyal14], fork after withholding[@DBLP:journals/corr/abs-1708-09790]). For a recent survey of these topics, see[@sokGameTheoryCryptocurrencies]. Exceptionally, “*Game of Coins*”[@DBLP:journals/corr/abs-1805-08979] considers the competition between multiple cryptocurrencies: a manipulative miner alters coin rewards in order to move miners to other cryptocurrencies of his own interest (with a fixed cost and a finite number of steps). However, in this paper, it’s the cryptocurrency issuer who changes the rewards in order to attract miners from other cryptocurrencies by producing the most efficient cryptocurrency to mine.
****
PoW PoSpace PoS PPoS PoA PoP **zk-PoI**
------------------------ ---------- ---------- ---------- ---------- ---------- ---------- ------------
(Pseudo)-Anonymity $\times$ $\times$ ****
Energy-Efficient $\times$ ****
Economically Efficient $\times$ $\times$ $\times$ $\times$ $\times$ ****
Permissionless $\times$ ([\*]{}) ****
Proof-of-Personhood Considered Harmful (and Illegal)
----------------------------------------------------
To be considered lawful in the real world, Proof-of-Personhood (PoP, [@pop]) requires the concurrence of multiple unrestricted freedoms: assembly, association, and wearing of masks. However, in most countries these freedoms are limited:
- freedom of assembly[@freedomAssembly] and association[@freedomAssociation]: most countries usually require previous notification and permission from the governing authorities, that may reject for multiple grounds including but not limited a breach of public order. Thus, PoP cannot be considered permissionless in these countries.
- it’s forbidden to wear a mask in most countries[@antiMaskLaw], as required for the anonimity of PoP (“All parties are recorded for transparency, but attendees are free to hide their identities by dressing as they wish, including hiding their faces for anonymity.”, [@pop]). Thus, PoP won’t be anonymous in countries that outlaw the covering of faces.
- promoters and organizers of PoP parties may themselves be committing a crime, due to incitement, conspiracy and complicity.
The solution proposed in this paper it’s the only possible lawful one according to current regulations that require the use of national IDs to register on blockchains (AMLD5[@AMLD5], FATF[@fatfGuidance], Cyberspace Administration of China[@chinaRules; @chinaIdentityRegulation]).
Building Blocks
===============
Consensus based on the Cryptographic Random Selection of Transaction Processing Nodes\[sub:Modern-Consensus-based-on\]
----------------------------------------------------------------------------------------------------------------------
The new family of consensus algorithms based on the cryptographically random selection of transaction processing nodes[@cryptoeprint:2016:918; @cryptoeprint:2016:919; @cryptoeprint:2017:406; @cryptoeprint:2017:454; @dfinityConsensus] is characterised by:
[|c|>p[5cm]{}|c|]{} Consensus algorithm & Random selection method & Sybil resistance[\
]{} OmniLedger & PVSS + collective BLS/BDN signatures [@cryptoeprint:2016:1067; @cryptoeprint:2017:406; @cryptoeprint:2018:483; @cryptoeprint:2019:676] & PoW/PoS[\
]{} RapidChain & Performance improvements over OmniLedger[@cryptoeprint:2018:460] & PoS[\
]{} Algorand & Cryptographic sortition by a unique digital signature & PoS[\
]{} Dfinity & BLS threshold signature scheme[@Boneh01shortsignatures] & PoS[\
]{} Snow White & Extract public keys based on the amount of currency owned & PoS[\
]{}
- Transaction processing workers/nodes are randomly selected from a larger group: in the case of Dfinity[@dfinityConsensus], an unbiasable, unpredictable verifiable random function (VRF) based on the BLS threshold signature scheme[@Boneh01shortsignatures] with the properties of uniqueness and non-interactivity; in the case of OmniLedger[@cryptoeprint:2017:406], the original proposal used a collective Schnorr threshold signature scheme[@DBLP:journals/corr/SytaTVWF15; @cryptoeprint:2016:1067; @cryptoeprint:2017:406], although it has been updated to collective BLS/BDN signatures[@cryptoeprint:2018:483] and now it uses MOTOR[@cryptoeprint:2019:676] instead of ByzCoin[@DBLP:journals/corr/Kokoris-KogiasJ16] with improvements for open and public settings; in the case of Algorand, secure cryptographic sortition is generated using an elliptic curve-based verifiable random function (ECVRF-ED25519-SHA512-Elligator2[@algorandVRF]); in the case of Snow White, cryptographic committee reconfiguration is done by extracting public keys from the blockchain based on the amount of currency owned. For a detailed comparison of random beacon protocols, see [@cryptoeprint:2018:319].
- Regular time intervals (also named epochs or rounds) on which randomly selected workers/nodes process the transactions.
- Faster transaction confirmation and finality.
- High scalability.
- Decoupling Sybil-resistance from the consensus mechanism (PoW/PoS is about membership, not consensus).
- PoW/PoS to protect against Sybil attacks: however, the present paper proposes the use of Zero-Knowledge Proof-of-Identity (i.e., more economically[@stakedPoS] and environmentally efficient[@natureEnergyCarbonCosts; @bitcoinCarbonFootprint]).
X.509 Public Key Infrastructure\[sub:X.509-Public-Key\]
-------------------------------------------------------
X.509 is an ITU-T standard[@x509] defining the format of public key certificates, itself based on the ASN.1 standard[@asn1]: these certificates underpin most implementations of public key cryptography, including SSL/TLS and smartcards. An X.509v3 certificate has the following structure:
Certificates are signed creating a certificate chain: the root certificate of an organization is a self-signed certificate that signs intermediate certificates that themselves are used to sign end-entities certificates. To obtain a signed certificate, the entity creates a key pair and signs a Certificate Signing Request (CSR) with the private key: the CSR contains the applicant’s public key that is used to verify the signature of the CSR and a unique Distinguished Name within the organization. Then, one of the intermediate certificate authorities issues a certificate binding a public key to the requested Distinguished Name and that also contains information identifying the certificate authority that vouches for this binding.
The certificate validation chain algorithm checks the validity of an end-entity certificate following the next steps:
1. The certificates are correct regarding the ASN.1 grammar of X.509 certificates.
2. The certificates are within their validity periods (i.e., non-expired).
3. If access to a Certificate Revocation List is granted, the algorithm checks that none of the certificates is included (i.e., the certificate has not been revoked).
4. The certificate chain is traversed checking that:
1. The issuer matches the subject of the next certificate in the chain.
2. The signature is valid with the public key of the next certificate in the chain.
5. The last certificate is a valid self-signed certificated trusted by the end-entity checker.
Additionally, the algorithm could also check complex application policies (i.e., the certificate can be used for web server authentication and/or web client authentication).
Electronic Passports\[sub:Electronic-Passport\]
-----------------------------------------------
****
**Data Group** **Data Elements**
---------------- ------------------------------------
Document Types
Issuing State or Organizaton
Name (of Holder)
Document Number
Check Digit - Doc Number
Nationality
Date of Birth
Check Digit - DOB
Sex
Date of Expiry or Valid Until Date
Check Digit DOE/VUD
Optional Data
Check Digit - Optional Data Field
Composite Check Digit
Data Group 11 Personal Number
Data Group 15 User’s Public Key
Modern electronic passports feature NFC chips[@icaoDoc9303part11] that contain all their printed information in digital form, using a proprietary format set by International Civil Aviation Organization[@icaoDoc9303part10] and not X.509 certificates \[sub:X.509-Public-Key\] like the ones used in national identity cards: the relevant fields are contained within its Data Group 1 \[tab:Data-Group-1\] (i.e., the same information available within the Machine Readable Zone), and the Document Security Object contains a hash of all the Data Groups signed by a Document Signing Certificate issued every three months (also stored on the passports), itself signed by a Country Signing Certificate Authority (all the certificates are available online[@icaoPKD]). Additionally, the data within the NFC chips are cryptographically protected and it’s necessary to derive the cryptographic keys by combining the passport number, date of birth and expiry date (i.e., BAC authentication).
Finally, note that the electronic identity cards of some countries can also work as ePassports (e.g., Spanish Identity Card -Documento Nacional de Identidad-).
Verifiable Computation\[sub:Verifiable-Computation\]
----------------------------------------------------
A public verifiable computation scheme allows a computationally limited client to outsource to a worker the evaluation of a function $F\left(u,w\right)$ on inputs $u$ and $w$: other alternative uses of these schemes allow a verifier $V$ to efficiently check computations performed by an untrusted prover $P$. More formally, the following three algorithms are needed:
(Public Verifiable Computation). A public verifiable computation scheme $VC$ consists of three polynomial-time algorithms $\left(\mbox{Keygen},\mbox{ Compute, Verify}\right)$ defined as follows:
- $\left(EK_{F},VK_{F}\right)\leftarrow\mbox{Keygen}\left(F,1^{\lambda}\right)$: the key generation algorithm takes the function $F$ to be computed and security parameter $\lambda$; it outputs a public evaluation key $EK_{F}$ and a public verification key $VK_{F}$.
- $\left(y,\pi_{y}\right)\leftarrow\mbox{Compute}\left(EK_{F},u,w\right)$: the prover runs the deterministic worker algorithm taking the public evaluation key $EK_{F}$, an input $u$ supplied by the verifier and an input $w$ supplied by the prover. It outputs $y\leftarrow F\left(u,w\right)$ and a proof $\pi_{y}$ of $y$’s correctness (as well as of prover’s knowledge of $w$).
- $\left\{ 0,1\right\} \leftarrow\mbox{Verify}\left(VK_{F},u,w,y,\pi_{y}\right)$: the deterministic verification algorithm outputs $1$ if $F\left(u,w\right)=y$, and $0$ otherwise.
A public verification computation scheme $VC$ must comply with the following properties of correctness, security, and efficiency:
- Correctness: for any function $F$ and any inputs to $F$, if we run $\left(EK_{F},VK_{F}\right)\leftarrow\mbox{Keygen}\left(F,1^{\lambda}\right)$ and $\left(y,\pi_{y}\right)\leftarrow\mbox{Compute}\left(EK_{F},u,w\right)$ then we always get $\mbox{Verify}\left(VK_{F},u,w,y,\pi_{y}\right)=1$.
- Efficiency: $\mbox{Keygen}$ is a one-time setup operation amortised over many calculations and $Verify$ is computationally cheaper than evaluating $F$.
- Security: for any function $F$ and any probabilistic polynomial-time adversary $A$, we require that $$\mbox{Pr}\left[\left(\hat{u},\hat{w},\hat{y},\hat{\pi_{y}}\right)\leftarrow A\left(EK_{F},VK_{F}\right):F\left(\hat{u},\hat{w}\right)\neq\hat{y}\right]\leq\mbox{negl}\left(\lambda\right)$$ and $$1=\mbox{Verify}\left(VK_{F},\hat{u},\hat{w},\hat{y},\hat{\pi_{y}}\right)\leq\mbox{negl}\left(\lambda\right)$$ where $\mbox{negl}\left(\lambda\right)$ denotes a negligible function of inputs $\lambda$.
Additionally, we require the public verification computation scheme $VC$ to be succinct and zero-knowledge:
- Succinctness: the generated proofs $\pi_{y}$ are of constant size, that is, irrespective of the size of the function $F$ and inputs $u$ and $w$.
- Zero-knowledge: the verifier learns nothing about the prover’s input $w$ beyond the output of the computation.
Practical implementations are Pinocchio[@cryptoeprint:2013:279] and Geppeto[@cryptoeprint:2014:976], or Buffet[@cryptoeprint:2014:674] and Pequin[@pequin](a simplified version of Pepper[@Setty12makingargument]).
### Verifiable Validation of X.509 Certificates as Anonymous Credentials
The algorithm for certificate chain validation chain in section \[sub:X.509-Public-Key\] can be implemented with the public verifiable computation scheme of section \[sub:Verifiable-Computation\] using zk-SNARKS to obtain a verifiable computation protocol so that a certificate holder is able to prove that he holds a valid X.509 certificate chain with a unique Distinguished Name, without actually sending the public key to the verifier and selectively disclosing the contents of the certificate: in other words, we re-use existing certificate chains and PKI infrastructure without requiring any modifications, turning X.509 certificates into anonymous credentials. A previous work already demonstrated the technical and practical viability of this approach[@cinderella-turning-shabby-x-509-certificates-into-elegant-anonymous-credentials-with-the-magic-of-verifiable-computation]: the only handicap was that the proof generation could take a long time (e.g., more than 10 minutes) and large keys (e.g., 1 Gbyte)..
Recent research advances have improved[@cryptoeprint:2017:602] the initial setup of the zk-SNARK protocol used to generate the Common-Reference String (CRS) with an MPC protocol, such that it’s secure even if all participants are malicious (except one). And faster proving times could be obtained by efficiently composing the non-interactive proving of algebraic and arithmetic statements[@cryptoeprint:2018:557] since QAP-based zk-SNARKs are only efficient for arithmetic representations and not algebraic statements, but at the cost of increasing the proof size.
In this paper, a practical implementation has been completed to check a certificate chain with an additional validation policy and written as C code for Pequin[@pequin], then compiled into a public evaluation and verification keys: unfortunately, it isn’t scalable to millions of users and/or the large circuits/constraints required to cover all the typologies of national identity cards/ePassports, thus an implementation based on TEE and mutual attestation is the preferred implementation \[sub:Detailed-Authentication-Protocol-Remote-Attestation\]. The only zero-knowledge proof system that could be scalable enough[@cryptoeprint:2018:691] works on a computer cluster, thus it doesn’t fit the setting of a single user authenticating on his own device, and a libsnark backend can’t handle more than 4 million gates requiring more than an hour of computation. Therefore, an implementation only using software means is still Work-In-Progress.
Cryptographic Accumulators
--------------------------
Firstly devised by Benaloh and de Mare[@10.1007/3-540-48285-7_24], a cryptographic accumulator [@cryptoeprint:2015:087] is a compact binding set of elements supporting proofs of membership and more space-efficient than storing all of the elements of the set; given an accumulator, an element, and a membership witness, the element’s presence in the accumulated set can be verified. Generally speaking, an accumulator consists of four polynomial-time algorithms:
- $Generate\left(1^{k}\right)$: given the security parameter $k$, it instantiates the initial value of the empty accumulator.
- $Add\left(a,y\right)\rightarrow\left(a',w\right)$: adds the element $y$ to the current state of the accumulator $a$ producing the updated accumulator value $a'$ and the membership witness $w$ for $y$.
- $WitnessAdd\left(w,y\right)\rightarrow w'$: on the basis of the current state of a witness $w$ and the newly added value $y$, it returns an updated witness $w'$.
- $Verify\left(a,y,w\right)\rightarrow\left\{ true,false\right\} $: verifies the membership of $y$ using its witness $w$ on the current state of accumulator $a$.
The following are interesting security properties of accumulators:
- Dynamic accumulators[@10.1007/3-540-45708-9_5]: accumulators supporting the removal of elements from the accumulator by means of a deletion algorithm $Removal()$ and a witness update algorithm $WitnessRemoval\left(\right)$.
- Universality[@10.1007/978-3-540-72738-5_17]: accumulators supporting non-membership proofs, $NonWitnessAdd\left(\right)$, $NonWitnessRemoval\left(\right)$ and $NonVerify\left(\right)$.
- Strong accumulators[@Camacho2012]: deterministic and publicly executable, meaning that it does not rely on a trusted accumulator manager.
- Public checkable accumulators, the correctness of every operation can be publicly verified.
Recent constructions of cryptographic accumulators specifically tailored for blockchains are: a dynamic, universal, strong and publicly checkable accumulator [@certcoin]; an asynchronous accumulator[@cryptoeprint:2015:718] with low frequency update and old-accumulator compatibility (i.e., up-to-date witnesses can be verified even against an outdated accumulator); a constant-sized, fair, public-state, additive, universal accumulator[@cryptoeprint:2018:853], and an accumulator optimised for batch and aggregation operations[@cryptoeprint:2018:1188].
Remote Attestation
------------------
In the terminology of Intel SGX, remote attestation is used to prove that an enclave has been established without alterations of any kind: in other words, remote parties can verify that an application is running inside an SGX enclave. Concretely, remote attestation is used to verify three properties: the identity of the application, that it has not been tampered with, and that it is running securely within an SGX enclave. Remote attestation is carried out in several stages: requesting a remote attestation from the challenger; performing a local attestation of the enclave; converting said local attestation to a remote attestation; returning the remote attestation to the challenger, and the challenger verifying the remote attestation to the Intel Attestation Service.
A detailed technical description is outside of the scope of this paper: detailed descriptions can be found in the standard technical documentation[@cryptoeprint:2016:086; @officialAttestation; @sampleAttestation]. Recent attacks[@vanbulck2018foreshadow] can be used to extract the secret attestation keys used to verify the identity of an SGX enclave, and microcode updates must be installed[@intelL1TF] to prevent their exploitation: that is, it’s essential to check that parties to a remote attestation are using a safe and updated version. However, our protocols are inherently resistant to deniability attacks[@cryptoeprint:2018:424] because they are based on mutual attestation.
As it would be shown in the next section \[sub:Detailed-Authentication-Protocol-Remote-Attestation\], remote attestation can be used as a more efficient substitute of verifiable computation.
Authentication Protocols\[sec:Authentication-Protocols\]
========================================================
In this section, we describe authentication protocols for Sybil-resistant, anonymous authentication using Zero-Knowledge protocols \[sub:Detailed-Authentication-Protocol\] and remote attestation \[sub:Detailed-Authentication-Protocol-Remote-Attestation\].
Authentication Protocols using Zero-Knowledge\[sub:Detailed-Authentication-Protocol\]
-------------------------------------------------------------------------------------
### Security Goals
The following security goals must be met for the system to be considered secure:
1. The registered miner’s key to the blockchain *opens, but no one can shut; he can shut, but no one can open* (*Isaiah 22:22*, [@isaiah2222]). For the security of the system to be considered equivalent to the currently available permissionless blockchains, anyone holding a valid public certificate should be able to register a pseudo-anonymous identity on the blockchain but no one should be able to remove it (i.e., uncensorable free entry is guaranteed).
2. Protection against malicious issuers: some certification authorities may turn against some citizens and try to cancel access to the permissionless blockchain or stole their funds.
1. Mandatory passphrase. An issuer may counterfeit a certificate with the same unique identifiers, thus possessing a valid certificate isn’t secure enough and a passphrase is deemed mandatory.
2. Non-bruteforceable. Operations must be computationally costly on the client side to prevent brute-forcing.
3. No OCSP checking. Prevention against malicious blacklisting.
3. Privacy: miner’s real identity can’t be learned by anyone.
4. Unique pseudonyms: from each identity card/ePassport, only one unique identifier can be generated.
5. Publicly verifiable: anyone should be able to verify the validity of the miner’s public key and its pseudonym.
### Zero-Knowledge Protocols (X.509)\[sub:Zero-Knowledge-Protocols-(X.509)\]
**Anonymous miner registration of a new public key on a permissionless blockchain.** This protocol generates a unique pseudonym for each miner, and attaches a verifiable proof that its new public key to be stored on-chain is signed with a valid public certificate included on a recognised certification authorities list, and that the new public key is linked to the blockchain-specific pseudonym that is in turn uniquely linked to the citizen’s public key certificate.
Miners holding a public key certificate must execute the following steps:
1. Create a deterministic public/secret key pair based on a secret passphrase (no need for verifiable computation): $$pk,sk=\mbox{Det\_KeyPairGen}\left(KDF\left(passphrase,hash(publicCert)\right)\right)$$ The generation algorithm must be determistic because the smartcard may be unable to store them and/or the miner may loose them (i.e., as in deterministic wallets). KDF is a password-based key derivation function (e.g., PBKDF2).
2. Obtain a signature of the previously generated public key $pk$ with the miner’s public key certificate (no need for verifiable computation, this operation could be executed on a smartcard): $$sign_{PK}=\mbox{PKCS\_Sign}\left(secretKey_{publicCert},pk\right)$$
3. Check the validity of the certificate chain of the miner’s public key certificate as extracted from the smartcard:
1. Load the public key of the root certificate.
2. Hash and verify all intermediates, based on their certificate templates, and the public key of their parent certificate starting from the root certificate and following with the verified public key from the previous intermediate certificate template.
3. Hash and verify the miner’s public key certificate using the last verified public key returned from the previous step.
4. Check the time validity of the miner’s public key certificate.
5. Check that the miner’s public key certificate is contained on a list of trusted certification authorities.
4. Obtain the unique identifier from the miner’s public key: $$uniqueID=getID(publicCert)$$ Note that the unique identifier is usually contained on Serial Number of the certificate, or the Subject Alternative Name extension under different OIDs, depending on the country.
5. Generate a deterministic pseudonym using the blockchain identifier: $$\begin{aligned}
signatureSecret & = & \mbox{PKCS\_Sign}\left(secretKey_{publicCert},\right.\\
& & \left."\mbox{PREFIXED\_COMMON\_STRING}"\right)\end{aligned}$$ $$\begin{aligned}
pseudonym & = & Hash\left(signatureSecret||BlockchainIdentifier||uniqueID\right)\\
& & ||"\mbox{REG}"\end{aligned}$$ PKCS\_Sign is the deterministic PKCS\#1.5 signing algorithm executed on a prefixed string to obtain a unique, non-predictable secret based on the certificate’s owner. The obtained signature is appended to the blockchain identifer and the unique identifier, and then hashed to derive a unique pseudonym. Finally, the string “REG” is appended to differentiate this pseudonym from the one generated during a remove protocol and prevent replay attacks for removal reusing the generated zero-knowledge proof.
6. Verify the signature $sign_{PK}$ on the miner’s public key certificate $pk$: $$\mbox{PKCS\_Verify}\left(publicCert,sign_{PK}\right)$$
7. As the $signatureSecret$ is calculated offline by the smartcard, it’s also necessary to verify it using the miner’s public key certificate $publicCert$: $$\mbox{PKCS\_Verify}\left(publicCert,signatureSecret\right)$$
8. Generate the zero-knowledge proof $\pi$ (e.g., zk-SNARK, zk-STARK or zk-SNARG) of the miner’s public key certificate $pk$, the generated pseudonym and, signature $sign_{PK}$ such that all the previous conditions 3-7 hold.
9. Anonymously contact the permissionless blockchain:
1. optionally, check the miner’s real identity on a cryptographic accumulator:
1. establish a shared secret running a Diffie-Hellman key exchange between the prospective miner and the permissionless blockchain
2. send attributes of the miner’s real identity encrypted with the shared secret
3. execute the non-membership proof $NonWitnessAdd\left(w,y\right)$ on the cryptographic accumulator
2. register the generated pseudonym, the new public key $pk$, the signature $sign_{PK}$ and $\pi$: note that they don’t reveal the miner’s real identity ($publicCert$, $uniqueID$ and $signatureSecret$ are all keep as a secret).
The registering node of the permissionless blockchain verifies $\pi$ before adding the new public key, the associated pseudonym, the signature $sign_{PK}$ and the succinct proof $\pi$: note that the miner is unable to register multiple pseudonyms, and he can only use one running node that would be signing messages with the generated secret key $sk$. Other nodes would be able to efficiently verify $\pi$ to confirm that the public key $pk$ is a signed by someone from an allowed certificate authority, and that the pseudonym is the miner’s unique alias for the blockchain.\
**Taking offline registrations from a permissionless blockchain.** This protocol takes offline a pseudonym and its associated public key $pk$ and signature $sign_{PK}$ from a permissionless blockchain. Miners must execute the following steps to take offline an identity from a permissionless blockchain:
1. Generate a zero-knowledge proof $\pi$ (e.g., zk-SNARK, zk-STARK or zk-SNARG) of the steps 3-7 of the previous protocol to prove secret knowledge of $sk$ and that he’s able to re-generate the pseudonym, but this time appending the string “OFF” to the pseudonym.
2. Anonymously contact the permissionless blockchain to take offline the generated pseudonym and all its associated data (including the cryptographic accumulator), attaching $\pi$.
The registering node of the permissionless blockchain verifies $\pi$ before taking offline the pseudonym without learning the real identity of the miner (publicCert, uniqueID and signatureSecret remain secret).
### Zero-Knowledge Protocols (ePassports)
Analogous to the zero-knowledge protocols for X.509 \[sub:Zero-Knowledge-Protocols-(X.509)\], but now considering the specific details of ePassports \[sub:Electronic-Passport\], which usually contain a unique keypair with the public key on Data Group 15 and the private key hidden within the chip: the Active Authentication protocol can be used to sign random challenges that can be verified with the corresponding public key. Some ePassports don’t feature Active Authentication, nonetheless a modified version of the following protocols could still be executed (see subsection \[sub:zkAbsence-of-AA\]).
**Anonymous miner registration of a new public key on a permissionless blockchain.** This protocol generates a unique pseudonym for each miner, and attaches a verifiable proof that its new public key to be stored on-chain is signed with a valid public certificate included on the list of Country Signing Certificate Authorities, and that the new public key is linked to the blockchain-specific pseudonym that is in turn uniquely linked to the public key certificate of the passport holder.
Miners holding a public key certificate must execute the following steps:
1. Create a deterministic public/secret key pair based on a secret passphrase (no need for verifiable computation): $$pk,sk=\mbox{Det\_KeyPairGen}\left(KDF\left(passphrase,hash(publicCert)\right)\right)$$ The *publicCert* is taken from the Data Group 15. KDF is a password-based key derivation function (e.g., PBKDF2).
2. Obtain a signature of the previously generated public key $pk$ with the miner’s public key certificate (no need for verifiable computation, this operation is executed within the ePassport’s chip using the Active Authentication protocol): $$sign_{PK}=\mbox{Sign}\left(secretKey_{publicCert},pk\right)$$
3. Check the validity of the Data Security Object of the miner’s ePassport:
1. Load the public key of the Country Signing Certificate from a trusted source [@icaoPKD] and the Document Signing Certificate from the ePassport.
2. Hash all the Data Groups and check their equivalence to the Data Security Object.
3. Verify the signature of the Data Security Object using the Document Signing Certificate.
4. Verify the signature of the Document Signing Certificate using the Country Signing Certificate.
5. Check the time validity of the certificates.
4. Obtain the unique identifier of the ePassport: $$uniqueID=getID(DataGroups)$$ Note that the unique identifier is usually contained on the Data Element “Document Number” of the Data Group 1: as it’s legally valid for the same person to own multiple passports with different Document Numbers, some countries include a unique “Personal Number” on the Data Group 11.
5. Generate a deterministic pseudonym using the blockchain identifier: $$\begin{aligned}
signatureSecret & = & \mbox{Sign}\left(secretKey_{publicCert},\right.\\
& & \left."\mbox{PREFIXED\_COMMON\_STRING}"\right)\end{aligned}$$ $$\begin{aligned}
pseudonym & = & Hash\left(signatureSecret||BlockchainIdentifier||uniqueID\right)\\
& & ||"\mbox{REG}"\end{aligned}$$ Sign is the Active Authentication protocol executed within the ePassport’s chip, a deterministic signing algorithm executed on a prefixed string to obtain a unique, non-predictable secret based on the certificate’s owner. The obtained signature is appended to the blockchain identifer and the unique identifier, and then hashed to derive a unique pseudonym. Finally, the string “REG” is appended to differentiate this pseudonym from the one generated during a remove protocol and prevent replay attacks for removal reusing the generated zero-knowledge proof (e.g., zk-SNARK, zk-STARK or zk-SNARG).
6. Verify the signature $sign_{PK}$ on the miner’s public key certificate $pk$: $$\mbox{PKCS\_Verify}\left(publicCert,sign_{PK}\right)$$ The *publicCert* is taken from the Data Group 15.
7. As the $signatureSecret$ is calculated offline by the ePassport’s chip, it’s also necessary to verify it using the miner’s public key certificate $publicCert$: $$\mbox{PKCS\_Verify}\left(publicCert,signatureSecret\right)$$ The *publicCert* is taken from the Data Group 15.
8. Generate the zero-knowledge proof $\pi$ (e.g., zk-SNARK, zk-STARK or zk-SNARG) of the miner’s public key certificate $pk$, the generated pseudonym, and signature $sign_{PK}$ such that all the previous conditions 3-7 hold.
9. Anonymously contact the permissionless blockchain
1. optionally, check the miner’s real identity on a cryptographic accumulator:
1. establish a shared secret running a Diffie-Hellman key exchange between the prospective miner and the permissionless blockchain
2. send attributes of the miner’s real identity encrypted with the shared secret
3. execute the non-membership proof $NonWitnessAdd\left(w,y\right)$ on the cryptographic accumulator
2. register the generated pseudonym, the new public key $pk$, the signature $sign_{PK}$ and $\pi$: note that they don’t reveal the miner’s real identity ($publicCert$, $uniqueID$ and $signatureSecret$ are all keep as a secret).
The registering node of the permissionless blockchain verifies $\pi$ before adding the new public key, the associated pseudonym, the signature $sign_{PK}$ and the succinct proof $\pi$: note that the miner is unable to register multiple pseudonyms, and he can only use one running node that would be signing messages with the generated secret key $sk$. Other nodes would be able to efficiently verify $\pi$ to confirm that the public key $pk$ is a signed by someone from an allowed certificate authority and that the pseudonym is the miner’s unique alias for the blockchain.\
**Taking offline registrations from a permissionless blockchain.** This protocol takes offline a pseudonym and its associated public key $pk$ and signature $sign_{PK}$ from a permissionless blockchain. Miners must execute the following steps to take offline an identity from a permissionless blockchain:
1. Generate a zero-knowledge proof $\pi$ (e.g., zk-SNARK, zk-STARK or zk-SNARG) of the steps 3-7 of the previous protocol to prove secret knowledge of $sk$ and that he’s able to re-generate the pseudonym, but this time appending the string “OFF” to the pseudonym.
2. Anonymously contact the permissionless blockchain to take offline the generated pseudonym and all its associated data (including the cryptographic accumulator), attaching $\pi$.
The registering node of the permissionless blockchain verifies $\pi$ before taking offline the pseudonym without learning the real identity of the miner (publicCert, uniqueID and signatureSecret remain secret).
### Mapping to goals
The previous protocols achieve the security goals:
1. The registered miner’s key to the blockchain *opens, but no one can shut; he can shut, but no one can open*. Only someone in possession of a valid public certificate can create a unique miner identity on the open blockchain and destroy it. Please note that the signing and verification of steps 2, 5, 6 and 7 are only needed if it’s required to check that the miner is the real owner of the smartcard/ePassport.
2. Protection against malicious issuers: the passphrase is mandatory, there’s no OCSP checking and the protocol is non-bruteforceable because it requires the generation of a proof $\pi$ for every passphrase that is going to be tried (>60 secs per $\pi$).
3. Privacy: miner’s real identity can’t be learned by anyone because publicCert and uniqueID are keep secret.
4. Unique pseudonyms: from each identity card/ePassport, only one unique identifier can be generated because there’s only one uniqueID per citizen.
5. Publicly verifiable: using the proof $\pi$, anyone is able to validate the miner’s public key and its pseudonym.
Additionally, cryptographic accumulators could be added to the protocols in order to prevent multiple registrations whenever an expired certificate is renovated.
### Absence of Active Authentication\[sub:zkAbsence-of-AA\]
Signing using the secret key of the Active Authentication protocol provides an extra layer of security: it guarantess that the remote party executing the protocol owns a physical copy of the ePassport (i.e., it hasn’t stolen a copy of the public certificates from others). However, some ePassports don’t feature Active Authentication, requiring a simplified version of the previous protocols:
- Steps 2,6 and 7 are removed.
- Step 5 doesn’t calculate the signature.
- The zero-knowledge$\pi$ is extended to Step 1, with a password-based key derivation function using less steps.
Detailed Authentication Protocols using Mutual Attestation\[sub:Detailed-Authentication-Protocol-Remote-Attestation\]
---------------------------------------------------------------------------------------------------------------------
![\[fig:Legend:-(1)-Attestation\]Simplified overview of mutual attestation.](mutualAttestation.png)
### Security Goals
The following security goals must be met for the system to be considered secure:
1. The registered miner’s key to the blockchain *opens, but no one can shut; he can shut, but no one can open* (*Isaiah 22:22*, [@isaiah2222]). For the security of the system to be considered equivalent to the currently available permissionless blockchains, anyone holding a valid public certificate should be able to register a pseudo-anonymous identity on the blockchain but no one should be able to remove it (i.e., uncensorable free entry is guaranteed).
2. Protection against malicious issuers: some certification authorities may turn against some citizens and try to cancel access to the permissionless blockchain or stole their funds.
1. Mandatory passphrase. An issuer may counterfeit a certificate with the same unique identifiers, thus possessing a valid certificate isn’t secure enough and a passphrase is deemed mandatory.
2. Non-bruteforceable. Operations must be computationally costly on the client side to prevent brute-forcing.
3. No OCSP checking. Prevent against malicious blacklisting.
3. Privacy: miner’s real identity can’t be learned by anyone.
4. Unique pseudonyms: from each identity card/ePassport, only one unique identifier can be generated.
### Mutual Attestation for X.509 Certificates\[sub:Mutual-Attestation-X509\]
**Anonymous miner registration of a new public key on a permissionless blockchain.** This protocol generates a unique pseudonym for each miner, with a new public key linked to the blockchain-specific pseudonym that is in turn uniquely linked to the citizen’s public key certificate: the mutual attestation between the parties guarantees the correctness of the execution of both parties.
The following are the steps to the protocol:
1. The client locally generates a signature secret using its secret key: $$\begin{aligned}
signatureSecret & = & \mbox{PKCS\_Sign}\left(secretKey_{publicCert},\right.\\
& & \left."\mbox{PREFIXED\_COMMON\_STRING}"\right)\end{aligned}$$
2. Mutual attestation between the authenticating client and the blockchain: the attestation is anonymous thanks to the use of unlinkable signatures (Enhanced Privacy ID -EPID-), and both parties obtain a temporary secret key to encrypt their communications.
3. Client’s attested code checks the validity of the certificate chain of the miner’s public key certificate as extracted from the smartcard:
1. Load the public key of the root certificate.
2. Hash and verify all intermediates, based on their certificate templates, and the public key of their parent certificate starting from the root certificate and following with the verified public key from the previous intermediate certificate template.
3. Hash and verify the miner’s public key certificate using the last verified public key returned from the previous step.
4. Check the time validity of the miner’s public key certificate.
5. Check that the miner’s public key certificate is contained on a list of trusted certification authorities.
4. If the previous step concluded satisfactorily, then the client’s attested code verifies the $signatureSecret$ using the miner’s public key certificate $publicCert$ because the $signatureSecret$ is calculated offline by the smartcard: $$\mbox{PKCS\_Verify}\left(publicCert,signatureSecret\right)$$
5. If the previous step concluded satisfactorily, then the client’s attested code creates a deterministic public/secret key pair based on a secret passphrase: $$pk,sk=\mbox{Det\_KeyPairGen}\left(KDF\left(passphrase,hash(publicCert)\right)\right)$$ The generation algorithm must be deterministic because the smartcard may be unable to store them and/or the miner may lose them (i.e., as in deterministic wallets). KDF is a password-based key derivation function (e.g., PBKDF2).
6. The client’s attested code generates a deterministic pseudonym using the blockchain identifier: $$\begin{aligned}
pseudonym & = & Hash\left(signatureSecret||BlockchainIdentifier||uniqueID\right)\\
& & ||"\mbox{REG}"\end{aligned}$$ and it obtains the unique identifier from the miner’s public key: $$uniqueID=getID(publicCert)$$ Note that the unique identifier is usually contained on Serial Number of the certificate, or the Subject Alternative Name extension under different OIDs, depending on the country.
7. Anonymously contact the attested encrypted database of the permissionless blockchain to register the generated pseudonym and the new public key $pk$: the uniqueID is also included using the temporary encrypted key, but it won’t be revealed to the host computer of the blockchain node because it will only be decrypted within the attested enclave.
8. The blockchain’s attested code checks within its encrypted database that the uniqueID has never been included: then, it proceeds to store the encrypted uniqueID (i.e., this time with a database secret key that only resides within the enclaves), the generated pseudonym and the new public key $pk$.
9. Then, the encrypted database’s attested code contacts the permissionless blockchain to register the generated pseudonym and its new public key $pk$.
**Taking offline registrations from a permissionless blockchain.** This protocol takes offline a pseudonym and its associated public key $pk$ from a permissionless blockchain. To take offline an identity from a permissionless blockchain, miners must re-run the previous protocol to prove that the client is able to re-generate the pseudonym with the same certificate, but this time appending the string “OFF” to the pseudonym.
The registering encrypted database of the permissionless blockchain verifies that the encrypted uniqueID is included in the database before taking offline the pseudonym from the permissionless blockchain without it learning the real identity of the miner.
### Mutual Attestation for ePassports\[sub:Mutual-Attestation-ePassports\]
Analogous to the zero-knowledge protocols for X.509 \[sub:Mutual-Attestation-X509\], but now considering the specific details of ePassports \[sub:Electronic-Passport\], which usually contain a unique keypair with the public key on Data Group 15 and the private key hidden within the chip: the Active Authentication protocol can be used to sign random challenges that can be verified with the corresponding public key. Some ePassports don’t feature Active Authentication, nonetheless a modified version of the following protocols could still be executed (see subsection \[sub:attAbsence-of-AA\]).
**Anonymous miner registration of a new public key on a permissionless blockchain.** This protocol generates a unique pseudonym for each miner, with a new public key linked to the blockchain-specific pseudonym that is in turn uniquely linked to the citizen’s ePassport: the mutual attestation between the parties guarantees the correctness of the execution of both parties.
The following are the steps to the protocol:
1. The client locally generates a signature secret using its secret key: $$\begin{aligned}
signatureSecret & = & \mbox{Sign}\left(secretKey_{publicCert},\right.\\
& & \left."\mbox{PREFIXED\_COMMON\_STRING}"\right)\end{aligned}$$
2. Mutual attestation between the authenticating client and the blockchain: the attestation is anonymous thanks to the use of unlinkable signatures (Enhanced Privacy ID -EPID-), and both parties obtain a temporary secret key to encrypt their communications.
3. Client’s attested code checks the validity of the Data Security Object of the miner’s ePassport:
1. Load the public key of the Country Signing Certificate from a trusted source[@icaoPKD] and the Document Signing Certificate from the ePassport.
2. Hash all the Data Groups and check their equivalence to the Data Security Object.
3. Verify the signature of the Data Security Object using the Document Signing Certificate.
4. Verify the signature of the Document Signing Certificate using the Country Signing Certificate.
5. Check the time validity of the certificates.
4. If the previous step concluded satisfactorily, then the client’s attested code verifies the $signatureSecret$ using the miner’s public key certificate $publicCert$ because the $signatureSecret$ is calculated offline by the ePassport’s chip: $$\mbox{PKCS\_Verify}\left(publicCert,signatureSecret\right)$$ The publicCert is taken from the Data Group 15.
5. If the previous step concluded satisfactorily, then the client’s attested code creates a deterministic public/secret key pair based on a secret passphrase: $$pk,sk=\mbox{Det\_KeyPairGen}\left(KDF\left(passphrase,hash(publicCert)\right)\right)$$ The generation algorithm must be deterministic because the ePassport is unable to store them and/or the miner may lose them (i.e., as in deterministic wallets). KDF is a password-based key derivation function (e.g., PBKDF2).
6. The client’s attested code generates a deterministic pseudonym using the blockchain identifier: $$\begin{aligned}
pseudonym & = & Hash\left(signatureSecret||BlockchainIdentifier||uniqueID\right)\\
& & ||"\mbox{REG}"\end{aligned}$$ and it obtains the unique identifier from the ePassport: $$uniqueID=getID(DataGroups)$$ Note that the unique identifier is usually contained on the Data Element “Document Number” of the Data Group 1: as it’s legally valid for the same person to own multiple passports with different Document Numbers, some countries include a unique “Personal Number” on the Data Group 11. Sign is the Active Authentication protocol executed within the ePassport’s chip, a deterministic signing algorithm executed on a prefixed string to obtain a unique, non-predictable secret based on the certificate’s owner.
7. Anonymously contact the attested encrypted database of the permissionless blockchain to register the generated pseudonym and the new public key $pk$: the uniqueID is also included using the temporary encrypted key, but it won’t be revealed to the host computer of the blockchain node because it will only be decrypted within the attested enclave.
8. The blockchain’s attested code checks within its encrypted database that the uniqueID has never been included: then, it proceeds to store the encrypted uniqueID (i.e., this time with a database secret key that only resides within the enclaves), the generated pseudonym and the new public key $pk$.
9. Then, the encrypted database’s attested code contacts the permissionless blockchain to register the generated pseudonym and its new public key $pk$.
**Taking offline registrations from a permissionless blockchain.** This protocol takes offline a pseudonym and its associated public key $pk$ from a permissionless blockchain. To take offline an identity from a permissionless blockchain, miners must re-run the previous protocol to prove that the client is able to re-generate the pseudonym with the same certificate, but this time appending the string “OFF” to the pseudonym.
The registering encrypted database of the permissionless blockchain verifies that the encrypted uniqueID is included in the database before taking offline the pseudonym from the permissionless blockchain without it learning the real identity of the miner.
### Mapping to goals
The previous protocols achieve the security goals:
1. The registered miner’s key to the blockchain *opens, but no one can shut; he can shut, but no one can open*[@isaiah2222]. Only someone in possession of a valid public certificate can create a unique miner identity on the open blockchain and destroy it. The signing and verification operations of steps 1 and 4 are only needed if it’s required to check that the miner is the real owner of the smartcard/ePassport.
2. Protection against malicious issuers: the passphrase is mandatory, there’s no OCSP checking and the protocol is non-bruteforceable because it can be rate-limited.
3. Privacy: miner’s real identity can’t be learned by anyone because publicCert and uniqueID are keep secret.
4. Unique pseudonyms: from each identity card/ePassport, only one unique identifier can be generated because there’s only one uniqueID per citizen.
The proposed solution depends on the security of Intel SGX (enclave and remote attestation protocols): in order to limit the impact of side-channels attacks on Intel SGX, mining nodes featuring the role of the Attested Encrypted DB will be restricted to trustworthy nodes.
### Performance Evaluation\[sub:Performance-Evaluation\]
\# VMs Mean Time/Req. \#Req./Sec Time/Connections Total time
-------- ---------------- ------------ ------------------ ------------
1 VM 416 ms 4.76 210 ms 21 secs
4 VM 112 ms 16.5 59 ms 5.9 secs
A load testing scenario featuring an Intel Xeon E3-1240 3.5 GHz and running 1 or 4 virtual machines was performed (with 5 users executing 100 requests per user). Operations like reading and/or signing from the smartcard were not included in the performance evaluation. The implementation will be open-sourced.
### Absence of Active Authentication\[sub:attAbsence-of-AA\]
Signing using the secret key of the Active Authentication protocol provides an extra layer of security: it guarantess that the remote party executing the protocol owns a physical copy of the ePassport (i.e., it hasn’t stolen a copy of the public certificates from others). However, some ePassports don’t feature Active Authentication, requiring a simplified version of the previous protocols by removing steps 1 and 4.
### Removing Single-Points of Failure
One of the shortcomings of relying on Intel’s Attestation Service (IAS) is that it becomes a single-point of failure: in practice, Intel would learn who is performing the attestation. For a public, permissionless blockchain it would be preferable to remove this trusted third party: to solve this problem, OPERA[@operaSGX] provides the first open, privacy-preserving attestation service to substitute Intel’s Attestation Service.
Worldwide Coverage and Distribution\[sub:Worldwide-Coverage-and\]
-----------------------------------------------------------------
![\[fig:Legend:-(1)-National\]Legend: (1) National identity card is a mandatory smartcard; (2) National identity card is a voluntary smartcard; (3) No national identity card, but cryptographic identification is possible using an ePassport, driving license and/or health card; (4) Non-digital identity card.](identityCards.png)
Fortunately, there is a unique cryptographic identifier for most people in the world: figure \[fig:Legend:-(1)-National\] shows a worldmap of the distribution of national identity cards. For some countries, there is no national identity card -code 3-, but some other unique cryptographic identifier is available (e.g., ePassport[@cryptoeprint:2009:200] and/or biometric passports as in figure \[fig:Availability-of-biometric\], social security card, driving license and/or health card). Transforming these unique identifiers into anonymous credentials enables the unique identification of individuals in a permissionless blockchain without revealing their true identities, making them indistinguishable: that is, authentication is not only anonymous but permissionless since there is no need to be pre-invited. Please note the enormous cost savings resulting from this approach compared to other anonymous credential[@cryptoeprint:2013:622; @coconut; @DBLP:conf/ccs/CamenischDD17] proposals that would require re-issuing new credentials: for example, consider that the UK’s national identity scheme was estimated at £5.4bn[@homeOfficeCostReport].
In some cases, an individual could obtain multiple cryptographic identifiers (e.g., multiple nationalities), but their number would still be limited and certainly less than the number of mining nodes that could be spawned on PoW permissionless blockchains. Additionally, the true identities provided by national identity cards could be used for other purposes, such as non-anonymous accounts identified by their legal identities.
![\[fig:Availability-of-biometric\]Availability of biometric passports. Source (ICAO, 2019)](icaopkd.jpg)
### eSIM’s Public Key Infrastructure
Latest specifications of SIM cards determine that SIM’s identity and data can be downloaded and remotely provisioned to devices[@eSIMwhitepaper]: instead of the traditional SIM card, there is an embedded SIM (i.e., eUICC[@eUICCtechnical]) that can store multiple SIM profiles containing the operator and subscriber data that would be stored on a traditional SIM card (e.g., IMSI, ICCID, ...).
A novel public key infrastructure has been created in order to protect the distribution of these new eSIM profiles[@gsmaPKI]: every certified eSIM is signed by its certified manufacturer, with a certificate that is itself signed by the GSMA root certificate issuer[@gsmaRootCertificateIssuer]. Network operators must also get certified and obtain certificates for their Subscription Management roles.
The eSIM’s PKI provides an alternative identification system for users where national identity cards and/or ePassports are difficult to obtain, as they must be unique and non-anonymous (4.13 and 4.1.5[@gsmaPKI]), but only when the mobile operator’s KYC processes can be considered trustworthy.
### Combining with Non-Zero-Knowledge Authentication
![\[fig:Mobile Authentication\]Mobile App using Non-Zero-Knowledge Authentication](readPassport.png "fig:")![\[fig:Mobile Authentication\]Mobile App using Non-Zero-Knowledge Authentication](transfer.png "fig:")
The use of zero-knowledge techniques to authenticate miners doesn’t preclude non-zero-knowledge authentication on the same blockchain: figure \[fig:Mobile Authentication\] shows a mobile app[@calctopiaApp] using BAC authentication to read a National Identity Card and/or ePassport (left), and then transferring to another ePassport (right). All national identifiers are publicly addressable.
Circumventing the Impossibility of Full Decentralization\[sub:Circumventing-the-Impossibility\]
-----------------------------------------------------------------------------------------------
Most blockchains using PKIs are consortium blockchains, thus it has become widespread that they always are permissioned and centralised. However, the term “permissionless” literally means “without requiring permission” (i.e., to access, to join, ...), thus a blockchain with a PKI could be permissionless if it accepts any self-signed certificate (i.e., a behaviour conceptually equivalent to Bitcoin), or any certificate from any government in the world as described in the previous subsection \[sub:Worldwide-Coverage-and\]. In the same way, a blockchain using PKIs doesn’t imply that its control has become centralised, it means that it accepts identities from said PKIs as described in this paper: actually, decentralization in the blockchain context strictly means that the network and the mining are distributed in a large number of nodes, thus unrelated to authentication.
A recent publication[@impossibleDecentralization] proves that it’s impossible for blockchains to be fully decentralised without real identity management (e.g., PoW, PoS and DPoS) because they cannot have a positive Sybil cost, defined as the additional cost that a player should pay to run multiple time nodes compared to the total cost of when those nodes are run by different players. To reflect the level of decentralization, they introduce the following definition:
($\left(m,\epsilon,\delta\right)$-Decentralization)[@impossibleDecentralization]. For $1\leq m$, $0\leq\epsilon$ and $0\leq\delta\leq100$, a system is $\left(m,\epsilon,\delta\right)$-decentralized if it satisfies the following:
1. The size of the set of players running nodes in the consensus protocol, $P$, is not less than $m$ (i.e., $\left|P\right|\geq m$).
2. Define $EP_{p_{i}}$ as the effective power of player $p_{i}$ as $\sum_{n_{i}\in N_{p_{i}}}\alpha_{n_{i}}$ where $N$ is the set of all nodes in the consensus protocol and $\alpha_{p_{i}}$ is the resource power of player $p_{i}$. The ratio between the effective power of the richest player, $EP_{max}$, and the $\delta-th$ percentile player, $EP_{\delta}$, is less than or equal to $1+\epsilon$ (i.e., $\left(EP_{max}/EP_{delta}\right)\leq1+\epsilon$).
Ideally, the number $m$ should be as high as possible (i.e., too many players do not combine into one node); and for the most resourceful and the $\delta$-th percentile player running nodes, the gap between their effective power is small. Therefore, full decentralization is represented by $\left(m,0,0\right)$ for sufficiently large $m$.
(Sufficient Conditions for Fully Decentralized Systems)[@impossibleDecentralization]. The four following conditions are sufficient to reach $\left(m,\epsilon,\delta\right)$-decentralization with probability 1.
1. (Give Rewards (GR-$m$)). Nodes with any resource power earn rewards.
2. (Non-delegation (ND-$m$)). It is not more profitable for too many players to delegate their resource power to fewer participants than to directly run their own nodes.
3. (No Sybil nodes (NS-$\delta$)). It is not more profitable for a participant with above the $\delta$-th percentile effective power to run multiple nodes than to run one node.
4. (Even Distribution (ED-($\epsilon,\delta$)). The ratio between the resource power of the richest and the $\delta$-th percentile nodes converges in probability to a value less than $\text{1+\ensuremath{\epsilon}}$.
For any initial state, a system satisfying GR-$m$, ND-$m$, NS-$\delta$, and ED-$\left(\epsilon,\delta\right)$ converges in probability to $\left(m,\epsilon,\delta\right)$-decentralization. [@impossibleDecentralization]
As it should be obvious by now, a blockchain using zk-PoI with strong identities from trusted public certificates (e.g., national identity cards and/or ePassports \[sub:Worldwide-Coverage-and\]) as described in this paper is the perfect candidate to achieve a fully decentralized blockchain.
A blockchain using zk-PoI with strong identities from trusted public certificates (e.g., national identity cards and/or ePassports \[sub:Worldwide-Coverage-and\]) reaches $\left(m,\epsilon,\delta\right)$-decentralization with probability 1.
A blockchain using zk-PoI with strong identities from trusted public certificates effectively limits the number of mining nodes to one per individual (ND-$m$), independently of how resourceful they are (NS-$\delta$, ED-($\epsilon,\delta$)), while keeping membership open to everyone (i.e., achieves a large number of participants (GR-$m$)). The presence of strong identities allows positive Sybil costs, thus the fulfillment of the sufficient conditions for fully decentralized systems[@impossibleDecentralization].
Preventing delegation (ND-$m$) is the most difficult condition to meet:
- market-enforced: richest participants could buy rights-of-use of others’ identities, but the market value of said identities (e.g., the Net Present Value of future profits obtained from the exploitation of said identities by their real owners) should wipe away almost all the profits from these exchanges.
- strictly-enforced: miners’ software could frequently check for the presence of the physical trusted public certificate (e.g., national identity cards and/or ePassports) and/or require them when transferring funds out of their accounts.
A posterior revision of the paper[@impossibleDecentralizationv2] introduces new definitions that try to emphasize that Trusted Third Parties (i.e., Certificate Authorities) shouldn’t be used in decentralized blockchains: as described in this paper\[sub:Worldwide-Coverage-and\], using 400 CAs of national identity cards and ePassports from over the world is still being decentralized according to the original definition of decentralization, and certainly much more decentralized than Bitcoin/Ethereum that concentrate >50% of their hashrate in 4-3 entities[@decentralizationBitCoinEthereum]. It’s very important not to mix concepts and misattribute qualities to different concepts:
- permissionless doesn’t imply without identification
- a decentralized consensus protocol doesn’t imply that it can’t use identification from TTPs: recent consensus protocols decouple Sybil-resistance from the consensus mechanism\[sub:Modern-Consensus-based-on\]
Thus, permissionless, decentralized and identification(-less) are different qualities that shouldn’t be intermixed. The results of this paper hold in the *trusted* decentralized setting, while the results of [@impossibleDecentralizationv2] hold in the *trustless* decentralised setting.
Resistance against Dark DAOs
----------------------------
Dark DAOs[@darkDaos] appear as a consequence of permissionless blockchains where users can create their own multiple identities and there’s no attributability of the actions.
1. When using real-world identities, it’s possible to establish the identity of the parties running the Dark DAO that are committing frauds (attributability) or at least, their pseudonyms: therefore, it’s possible to punish them.
2. To prevent Dark DAOs buying real-world identities, a smart contract can be setup that pays a reward for denouncing the promoters of the fraud: the whistleblowers would be paid a multiple of what they would get from the defrauders, thus making denunciation the preferred option. Then defrauders would be banned as in step 1.
Resistance against Collusion and other Attacks
----------------------------------------------
In this sub-section, we consider different avenues for attack and provide detailed defense mechanisms:
1. Corrupt root certificate authorities
2. Attacks against consensus protocols
3. Resistance against collusion
4. On achieving collusion-freeness
### Corrupt Root Certificate Authorities
Corrupt countries may be tempted to create fake identities or frequently renovate existing ones: these countries can be easily banned out by removing them from the list of valid authorities (i.e., root X.509 certificate and/or Country Signing Certificate). Bounties in cryptocurrency could be offered for whistleblowing any corrupt attack against the long-term existence of the blockchain.
### Attacks against Consensus Protocols
Modern consensus protocols based on the cryptographically secure random choice of the leader (e.g., [@dfinityConsensus; @cryptoeprint:2017:406]) detect cheating by monitoring changes to the chain quality. The following table gathers cheating events for different consensus algorithms that could be detected and punished:
[|c|>p[7.5cm]{}|]{} Protocol & Cheater detection[\
]{} & Equivocation: multiple blocks for same round with same rank.[\
]{} & Equivocation: multiple blocks with the highest priority.[\
]{} & All the blocks must be timely published.[\
]{} & All the notarizations must be timely published within one round.[\
]{} & Core validators can detect rogue validators.[\
]{} & Withholders can be detected after multiple consecutive rounds.[\
]{} & 5>= failed RandHound views from a rogue validator.[\
]{}
### Resistance against Collusion
Consensus protocols already provide collusion-tolerance by design: an adversary controlling a high number of nodes, or equivalently all said nodes colluding for the same attack, must confront the difficulties introduced by shard re-assignment at the beginning of every new epoch. For the case of OmniLedger[@cryptoeprint:2017:406], the security of the validator assignment mechanism can be modeled as a random sampling problem using the binomial distribution, $$P\left[X\leq\left\lfloor \frac{n}{3}\right\rfloor \right]=\sum_{k=0}^{n}\left(\begin{array}{c}
n\\
k
\end{array}\right)m^{k}\left(1-m\right)^{n-k}$$ assuming that each shard has less than $\left\lfloor \frac{n}{3}\right\rfloor $ malicious validators. Then, the failure rate of an epoch is the union bound of the failures rates of individual shards, each one calculated as the cumulative distribution over the shard size $n$, with $X$ being the random variable that represents the number of times we pick a malicious node. An upper bound of the epoch failure event, $X_{E}$, is calculated as: $$P\left[X_{E}\right]\leq\sum_{k=0}^{l}\frac{1}{4^{k}}\cdot n\cdot P\left[X_{S}\right]$$ where $l$ is the number of consecutive views the adversary controls, $n$ is the number of shards and $P\left[X_{S}\right]$ is the failure rate of one individual shard. Finally, for $l\rightarrow\infty$, we obtain $$P\left[X_{E}\right]\leq\frac{4}{3}\cdot n\cdot P\left[X_{S}\right]$$
### On Achieving Collusion-Freeness
Start noticing that collusion-freeness is not about preventing malicious behaviour, only preventing that malicious players act as independently of each other as possible. Following a previous seminal work[@collusionFreeProtocols], collusion-freeness can only be obtained under very stringent conditions: (a) the game must be finite; (b) the game must be publicly observable; and (c) the use of private channels at the beginning of the game is essential, but forbidden during the execution of the protocol. Although blockchains are publicly observable, they are also an infinite game where parties can freely communicate between them using private channels at any time: therefore, collusion-freeness is impossible in the sense of [@collusionFreeProtocols].
Fortunately, there is a way to get around this impossibility result: forbid malicious/Byzantine behaviours requiring the use of mutual attestation for all the nodes, thus precluding any deviation from the original protocol.
If mutual attestation is required for all the nodes, any infinite, partial-information blockchain game with publicly observable actions has a collusion-free protocol.
As mutual attestation is already required for zk-PoI \[sub:Detailed-Authentication-Protocol-Remote-Attestation\], we would only be extending its use for the rest of the blockchain protocol.
Incentive Compatible and Strictly Dominant Cryptocurrencies
===========================================================
The success of cryptocurrencies is better explained by their incentive mechanisms rather than their consensus algorithms: a cryptocurrency with poor incentives (e.g., a cryptocurrency not awarding coins to miners) will not achieve any success; conversely, better incentives and much more inefficient consensus algorithm could still find some success.
Much research has been focused on conceiving better consensus algorithms for decentralised systems and cryptocurrencies[@cryptoeprint:2016:918; @cryptoeprint:2016:919; @cryptoeprint:2017:406; @cryptoeprint:2017:454; @DBLP:journals/corr/Kokoris-KogiasJ16; @dfinityConsensus]: unfortunately, obtaining consensus mechanisms with better incentives and economic properties is an area that is lacking much research, and the combination of all the game-theoretic results contained in this section fills this gap for the sake of achieving a *focal point* (i.e., Schelling point[@schellingConflict]) in the multi-equilibria market of cryptocurrencies. Thus, a selective advantage is introduced by design over all the other cryptocurrencies, in explicit violation of the neutral model of evolution[@2017arXiv170505334E] in order to obtain an incentive compatible and strictly dominant cryptocurrency.
Incentive-Compatible Cryptocurrency\[sub:Incentive-Compatible-Cryptocurre\]
---------------------------------------------------------------------------
Shard-based consensus protocols have been recently introduced in order to increase the scalability and transaction throughput of public permissionless blockchains: however, the study of the strategic behaviour of rational miners within shard-based blockchains is very recent. Unlike Bitcoin, that grants all rewards to the most recent miner, block rewards and transactions fees must be proportionally shared between all the members of the sharding committee[@DBLP:journals/corr/Kokoris-KogiasJ16], and this includes incentives to remain live during all the lifecycle of the consensus protocol. Even so, existing sharding proposals[@cryptoeprint:2017:406; @cryptoeprint:2018:460] remain silent on how miners will be rewarded to enforce their good behaviour: as it’s evident, if all miners are equally rewarded without detailed consideration of their efforts, rational players will *free-ride* on the efforts of others.
One significant difference introduced in this paper with respect to other shard-based consensus protocols is the use of Zero-Knowledge Proof-of-Identity as the Sybil-resistance mechanism: as we will see in the following sections, it’s a significant novelty because solving Proof-of-Work puzzles is the most computationally expensive activity of consensus protocols, thus it’s no longer dominated by computational costs. This makes the necessity for an incentive-compatible protocol even more acute: the preferred rational miner’s strategy is to execute the Proof-of-Work of the initial phase of the protocol for each epoch and to refrain from the transaction verification and consensus of subsequent phases of the protocol, but still selfishly claim the rewards as if they had participated. The substitution of costly PoW for cheap Zero-Knowledge Proof-of-Identity only increases the attractiveness of this rational strategy, that can only be counteracted by using an incentive-compatible protocol.
### A Nash Equilibrium for a Cryptocurrency on a Shard-Based Blockchain
This section is based on a stylised version of a recent game-theoretic model[@2018arXiv180907307M], taking into consideration that there is no cost associated with committee formation to enter each shard since we are using Zero-Knowledge Proof-of-Identity, and not costly Proof-of-Work: instead, a penalty $p$ is imposed to defective and/or cheating miners. The following is a list of symbols:
Symbol Definition
------------------- -------------------------------------------------------------
$k$ Number of shards
$N$ Number of miners
$x_{i}^{j}$ Vector of received transactions by miner $i$ in shard $j$
$y^{j}$ Vector of transactions submitted by shard $j$ to blockchain
$c$ Minimum number of miners in each committee
$\tau$ Required number of miners in shard for consensus
$r$ Benefit for each transaction
$b_{i}$ Benefit of miner $i$ after adding the block
$c_{i}^{t}$ Total cost of computation for miner $i$
$c^{o}$ Total optional costs in each epoch
$c^{v}$ Cost of transaction verification
$c^{f}$ Fixed costs in optional cost
$p$ Penalty cost
$BR$ Block Reward
$l_{j}$ Number of cooperative miners in each shard
$L$ Total number of cooperative miners in all shards
$C_{j}^{l_{j}}$ Set of all cooperative miners in shard $j$
$D_{j}^{n-l_{j}}$ Set of all defective miners in shard $j$
$C^{L}$ Set of all cooperative miners
$D^{N-L}$ Set of all defective miners
$s^{r}$ Signed receipt of a transaction
\
Let $\mathbb{G}$ denote the shard-based blockchain game, defined as a triplet $\left(P,S,U\right)$ where $P=\left\{ P_{i}\right\} _{i=1}^{N}$ is the set of players, $S=\left\{ C,D\right\} $ is the set of strategies (Cooperate $C$, or Defect $D$) and $U$ is the set of payoff values. Each miner receives a reward if and only it has already cooperated with other miners within the shard, the payoff of cooperative miners in set $C^{l_{j}}$ is
$$u_{i}\left(C\right)=\frac{BR}{kl_{j}}+\frac{r\left|y^{j}\right|}{l_{j}}-\left(c^{f}+\left|x_{i}^{j}\right|c^{v}\right)\label{eq:5.1}$$
We assume that the block reward $BR$ is uniformly distributed among shards and each cooperative miner can receive a share of it. A miner might be cooperative but all other miners may agree on a vector of transactions $y^{j}$ that is different from his own vector of transactions $x_{i}^{j}$ (i.e., $\left|x_{i}^{j}\right|\neq\left|y^{j}\right|$): nonetheless, transaction rewards are uniformly distributed among all cooperative miners in each shard, proportional to all the transactions submitted to the blockchain by each shard.
The defective miners’ payoff can be calculated as $$u_{i}^{D}=-p^{m}$$ because the defective miners will have to pay a penalty and they will not receive any benefit (and it doesn’t incur in any mandatory cost such as solving PoW puzzles because we use cheap Zero-Knowledge Proof-of-Identity).
There exists a cooperative Nash equilibrium profile in game $\mathbb{G}$ under the following conditions:
\[thm:UniqueNashEqulibriumCryptocurrency\]Let $C_{j}^{l_{j}}$ and $D_{j}^{m-l_{j}}$ denote the sets of $l_{j}$ cooperating miners and $n-l_{j}$ defecting miners inside each shard $j$ with $n$ miners, respectively. $\left(C^{L},D^{N-L}\right)$ represents a Nash equilibrium profile in each epoch of game $\mathbb{G}$, if the following conditions are satisfied:
1. In all shards $j$, $l_{j}\geq\tau$.
2. If for a given miner $P_{i}$ in shard $j$, with $x_{i}^{j}=y^{j}$, then the number of transactions $\left|x_{i}^{j}\right|$ must be greater than $$\theta_{c}^{1}=\frac{c^{f}-\frac{BR}{kl_{j}}+p}{\nicefrac{r}{l_{j}}-c^{v}}$$
3. If for a given miner $P_{i}$ in shard $j$, with $x_{i}^{j}\neq y^{j}$, then the number of transactions $\left|x_{i}^{j}\right|$must be smaller than $$\theta_{c}^{2}=\frac{\frac{BR}{kl_{j}}+\frac{r\left|y^{j}\right|}{l_{j}}-c^{f}-p}{c^{v}}$$
The first condition $l_{j}\geq\tau$ (i.e., the number of cooperative miners must be greater than $\tau$) must hold so that cooperative miners will receive benefits for transactions and block rewards.
Let $l_{j}^{*}$ be the largest set of cooperative miners in each shard, where no miner in $D_{j}^{n-l_{j}}$ can join $C_{j}^{l_{j}}$ to increase its payoff: if miner $P_{i}^{j}$ is among the set of cooperative miners where $x_{i}^{j}=y^{j}$, then it would not unilaterally deviate from cooperation if: $$\frac{BR}{kl_{j}}+\frac{r\left|x_{i}^{j}\right|}{l_{j}}-\left(c^{f}+\left|x_{i}^{j}\right|c^{v}\right)\geq-p$$ which shows that $x_{i}^{j}\geq\theta_{c}^{1}$, whereas in the second condition, $$\theta_{c}^{1}=\frac{c^{f}-\frac{BR}{kl_{j}}-p}{\nicefrac{r}{l_{j}}-c^{v}}$$ But if $P_{i}^{j}$ is among the cooperators whose vector of transactions is different from the output of the shard, $x_{i}^{j}\neq y^{j}$, then it would not deviate from cooperation if: $$\frac{BR}{kl_{j}}+\frac{r\left|y^{j}\right|}{l_{j}}-\left(c^{f}+\left|x_{i}^{j}\right|c^{v}\right)\geq-p$$ which shows that $x_{i}^{j}<\theta_{c}^{2}$, whereas in the third condition, $$\theta_{c}^{2}=\frac{\frac{BR}{kl_{j}}+\frac{r\left|y^{j}\right|}{l_{j}}-c^{f}-p}{c^{v}}$$ Then if $l_{j}^{\text{*}}$ represents the largest set of cooperative miners in each shard, then $\left(C^{L},D^{N-L}\right)$ would be the unique cooperative Nash equilibrium of the game $\mathbb{G}$.
As can be understood from the proof, cooperative miners have less incentive to cooperate when: 1) the number of participants $N$ increases; 2) the optional costs of computation increase ($c^{f}$ is in the numerator or $c^{v}$ in denominator of $\theta_{C}$); 3) or in general, when the number of transactions is not large enough compared to a fixed threshold.
### Incentive-Compatible Cryptocurrency on a Shard-Based Coordinated Blockchain
In order to increase the incentives to cooperate rather than defect, an incentive-compatible protocol enforcing cooperation based on the previously presented Nash equilibrium is introduced here \[alg:Incentive-Compatible-Protocol\]: all miners should disclose their list of transactions to a coordinator, who then announces to each miner whether their cooperation would be in their interests based on being within the maximum subset of miners with a similar list of transactions (i.e., $x_{i}^{j}=y^{j}$), and then enforces their cooperation by checking their compliance and rewarding them properly.
The protocol proceeds as follows: for the first function (i.e., *ShardTransactionsAssignment*), each miner receives a list of transactions $x_{i}^{j}$ to verify based on the epochRandomness and his pseudonymous identity and public key obtained by the Zero-Knowledge Proof-of-Identity.
For the second function (i.e., *NodeSelection*), all miners calculate a hash $H\left(x_{i}^{j}\right)$ over their transaction list and send it to the coordinator. The coordinator finds the subset with the maximum number of miners with a common transaction list, thus calculating $\theta_{c}^{1}$, $\theta_{c}^{2}$, $l_{j}$ and $C_{j}^{l_{j}}$: in each epoch, the coordinator publicly defines the list of cooperative miners $C_{j}^{l_{j}}$ and defective miners $D_{j}^{n-l_{j}}$ using on .
For the third function (i.e., *ShardParticipation*), all the transactions of each miner are verified and a signed consensus is reached.
For the fourth function (i.e., *VerificationAndRewards*), the rewards are shared between the cooperative miners and denied to those miners in $C_{j}^{l_{j}}$ that didn’t cooperate.
### Improved Incentive-Compatible Cryptocurrency on a Shard-Based Blockchain
Although the role of a coordinator is essential to BFT protocols[@DBLP:journals/corr/Kokoris-KogiasJ16], its expanded functionality in the previous incentive-compatible protocol is problematic: it introduces latency and network costs due to the new obligations to report to the coordinator; moreover, it creates new opportunities for malicious miners which may report false $H\left(x_{i}^{j}\right)$ or not follow coordinator’s instruction to cooperate or defect. The next incentive-compatible protocol significantly improves over the state of the art: the role of the coordinator is minimised, strengthing the protocol by removing the previous vulnerabilities and making it resistant to malicious miners.
Information propagation[@bitcoinInformationPropagation] is an essential part of any blockchain, and gossiping transactions to neighbouring miners is one of its key features. In the new incentive-protocol protocol, we require that any broadcasted/gossiped/propagated transaction gets acknowledged with a signed receipt to its sender: then, senders must attach these receipts to the consensus leaders and verification nodes in order to ease detection of defective and/or cheating miners. . In other words, the signed receipts serve as snitches that denounce non-cooperative miners thus preventing that any reward gets assigned to them: at the same time, all miners are incentivised to participate in the denunciation in order to gain the rewards of non-cooperative miners and other *free-riders*.
In order to save bandwidth, note that it’s not obligatory to send the full list of all signed transaction receipts to consensus leaders and/or verification nodes: only a random subset per each miner should be enough to catch defective miners.
Additionally, the absence of signed receipts could be used to detect the need of a change of a consensus leader (i.e., “view-change”) in BFT protocols[@DBLP:journals/corr/Kokoris-KogiasJ16; @cryptoeprint:2017:406].
On Strictly Dominant Cryptocurrencies
-------------------------------------
A cryptocurrency using Zero-Knowledge Proof-of-Identity as the Sybil-resistance mechanism strictly dominates PoW/PoS cryptocurrencies: a miner having to choose between mining different cryptocurrencies, one with no costs associated with its Sybil-resistance mechanism and distributing equally the rewards, and the others using costly PoW/PoS and thus featuring mining concentration, will always choose the first one. That is, mining equally distributed cryptocurrencies using Zero-Knowledge Proof-of-Identity is a dominant strategy; in other words, the strategy of mining Bitcoin and other similar cryptocurrencies is strictly dominated by the hereby described cryptocurrency. *Ceteris paribus*, this cryptocurrency will have better network effects, thus better long-term valuation.
### Strictly Dominant Cryptocurrencies and a Nash Equilibrium\[sub:Strictly-Dominant-Cryptocurrenci\]
The intuition behind the preference to mine fully decentralised cryptocurrencies with the lowest expenditure (i.e., lowest CAPEX/OPEX implies higher profitability), thus the search for better hash functions[@cryptoeprint:2017:1168; @cryptoeprint:2016:989; @cryptoeprint:2015:430; @cryptoeprint:2016:027], is formally proved here and then applied to the specific case of the proposed cryptocurrency.
(Power-Law Fee-Concentrated (PLFC) cryptocurrency)\[def:(Power-law-fee-concentrated-cryp\]. A cryptocurrency whose distribution of mining and/or transaction fees follows a power-law (i.e., a few entities earn most of the fees/rewards), usually due to the high costs of its Sybil-resistance mechanism.
Proof-of-Work cryptocurrencies are Power-Law Fee-Concentrated: 90% of the mining power is concentrated in 16 miners in Bitcoin and 11 in Ethereum[@decentralizationBitCoinEthereum].
Proof-of-Stake cryptocurrencies are Power-Law Fee-Concentrated: miners earn fees proportional to the amount of money at stake, and wealth is Pareto-concentrated[@Pareto2014-PARMOP-2].
(Uniformly-Distributed Capital-Efficient (UDCE) cryptocurrency)\[def:(Uniformly-distributed-fee-uncon\]. A cryptocurrency whose distribution of mining and/or transaction fees is uniformly distributed among all the transaction processing nodes, and doesn’t require significant investments from the participating miners.
The proposed cryptocurrency using Zero-Knowledge Proof-of-Identity is a Uniformly-Distributed Capital-Efficient cryptocurrency.
(Game of Rational Mining of Cryptocurrencies)\[def:(Game-of-Rational\]. A rational miner ranks the cryptocurrencies according to their expected mining difficulty, and chooses to mine those with lowest expected difficulty.
Awesome Miner[@awesomeMiner], MinerGate[@minerGate], MultiMiner[@multiMiner], MultiPoolMiner[@multipoolMiner], Smart-Miner[@smartMiner; @smartMinerPaper] and NiceHash Miner[@niceHash] are practical implementations of the Game of Rational Mining of Cryptocurrencies (although also considering their prices in addition to their difficulties). Specific calculators for mining profitability[@whatToMine; @2CryptoCalc; @CoinWarz; @CryptoCompare; @CryptoZone; @Crypto-CoinZ] could also be used for the similar purposes. Additionally, other papers[@modelMinerHashRate] provide models regarding optimal hash rate allocation.
Let $u_{i}$ be the payoff or utility function for each miner, expressing his payoff in terms of the decisions or strategies $s_{i}$ of all the miners, $$u_{i}\left(s_{1},s_{2},\ldots,s_{n}\right)=u_{i}\left(s_{i},s_{-i}\right)$$ whese $s_{-i}$ are set of the strategies of the rest of miners, $$s_{-i}=\left(s_{1},s_{2},\ldots,s_{i-1},s_{i+1},\ldots,s_{n}\right)$$
A strategy $s_{1}$ *strictly dominates a strategy $s_{2}$ for miner $i$* if and only if, for any $s_{-i}$ that miner $i$’s opponents might use, $$u_{i}\left(s_{1},s_{-i}\right)>u_{i}\left(s_{2},s_{-i}\right)$$
That is, no matter what the other miners do, playing $s_{1}$ is strictly better than playing $s_{2}$ for miner $i$. Conversely, we say that the strategy *$s_{2}$ is strictly dominated by $s_{1}$*: a rational miner $i$ would never play a strictly dominated strategy.
A strategy $s_{i}^{*}$ is a *strictly dominant strategy for miner $i$* if and only if, for any profile of opponent strategies $s_{-i}$ and any other strategy $s_{i}^{'}$ that miner $i$ could choose, $$u_{i}\left(s_{i}^{*},s_{-i}\right)>u_{i}\left(s_{i}^{'},s_{-i}\right)$$
We now demonstrate that mining *UDCE crypto-cryptocurrencies* \[def:(Uniformly-distributed-fee-uncon\] is a strictly dominant strategy with regard to *PLFC cryptocurrencies \[def:(Power-law-fee-concentrated-cryp\]* in the *Game of Rational Mining of Cryptocurrencies* \[def:(Game-of-Rational\] by showing that every miner’s expected profitability is higher in UDCE cryptocurrencies.
\[thm:Uniformly-distributed-crypto-cur\]UDCE cryptocurrencies yield a strictly higher miner’s expected profitability compared to PLFC cryptocurrencies in the Game of Rational Mining of Cryptocurrencies.
Let $N$ be the number of miners and $R$ the average daily minted reward per day: UDCE cryptocurrencies award an average of $\nicefrac{N}{R}$ units of cryptocurrency to every participant miner. For every miner on the long tail of the power distribution, the amount earned with UDCE cryptocurrencies is obviously higher than with PLFC cryptocurrencies. For the few miners that dominate PLFC cryptocurrencies, their profitability is lower because they have to account for the energy[@natureEnergyCarbonCosts; @bitcoinCarbonFootprint] and equipment costs in the case of PoW cryptocurrencies or the opportunity cost of staking capital in volatile PoS cryptocurrencies[@stakedPoS], meanwhile in UDCE cryptocurrencies their cost of mining is so negligible compared to PLFC cryptocurrencies that the balance of profitability is always tipped in their favour.
\[def:The-process-to\]The process to solve games called *Iterated Deletion of Strictly Dominated Strategies (IDSDS)* is defined by the next steps:
1. For each player, eliminate all strictly dominated strategies.
2. If any strategy was deleted during Step 1, repeat Step 1. Otherwise, stop.
If the process eliminates all but one unique strategy profile $s^{*}$, we say it is the *outcome of iterated deletion of strictly dominated strategies* or a *dominant strategy equilibrium*.
\[def:pure-nash\]A strategy profile $s^{*}=\left(s_{1}^{*},s_{2}^{*},\ldots,s_{n}^{*}\right)$ is a *Pure-Strategy Nash Equilibrium* (PSNE) if, for every player $i$ and any other strategy $s_{i}^{'}$ that player $i$ could choose, $$u_{i}\left(s_{i}^{*},s_{-i}^{*}\right)\geq u_{i}\left(s_{i}^{'},s_{-i}^{*}\right)$$
\[def:strict-nash\]A strategy profile $s^{*}=\left(s_{1}^{*},s_{2}^{*},\ldots,s_{n}^{*}\right)$ is a *Strict Nash Equilibrium* (SNE) if, for every player $i$ and any other strategy $s_{i}^{'}$ that player $i$ could choose, $$u_{i}\left(s_{i}^{*},s_{-i}^{*}\right)>u_{i}\left(s_{i}^{'},s_{-i}^{*}\right)$$
Additionally, if a game is solvable by Iterated Deletion of Strictly Dominated Strategies, the outcome is a Nash equilibrium.
\[thm:UDCE-cryptocurrency-dominates\]A UDCE cryptocurrency dominating PLFC cryptocurrencies is a Nash equilibrium.
Mining a UDCE cryptocurrency is a strictly dominant strategy with regard to other miners mining PLFC cryptocurrencies because PLFC cryptocurrencies are strictly dominated by UCDE cryptocurrencies by Theorem \[thm:Uniformly-distributed-crypto-cur\], thus a rational miner will always prefer to miner the latter.
Thus, by the application of Iterated Deletion of Strictly Dominated Strategies (IDSDS) to the Game of Rational Mining of Cryptocurrencies \[def:(Game-of-Rational\], each miner will eliminate mining PLFC cryptocurrencies in favor of mining an UDCE cryptocurrency, leaving this as the unique outcome: therefore, mining a UDCE cryptocurrency is a *dominant strategy equilibrium* by Definition \[def:The-process-to\] and a *Nash equilibrium* by Definition \[def:pure-nash\] or by Definition \[def:strict-nash\].
\[claim:(Uniqueness-of-Technical-Solution)\](Uniqueness of Technical Solution). The proposed technical solution using Zero-Knowledge Proof of Identity from trusted public certificates (i.e., national identity cards and/or ePassports) is the only practical and unique solution for a UCDE cryptocurrency.
As demonstrated in the paper describing “The Sybil Attack”[@the-sybil-attack], Sybil attacks are always possible unless a trusted identification agency certifies identities.
As National Identity Cards and ePassports are the only globally available source of trusted cryptographic identity (3.5B for National Identity Cards and 1B for ePassports), the only way to bootstrap a UCDE cryptocurrency is by using the proposed Zero-Knowledge Proof-of-Identity from trusted public certificates (National Identity Cards and/or ePassports).
### Strictly Dominant Cryptocurrencies and Evolutionary Stable Strategies\[sub:Evolutionary-Stable-Strategies\]
Another interesting viewpoint to consider in the analysis of the cryptocurrency market is the one offered by behavioural ecology and its Evolutionary Stable Strategies \[def:ESS\]: each cryptocurrency can be considered a unique individual in a population, genetically programmed to play a pre-defined strategy. New cryptocurrencies are introduced into the population as individuals with different mutations that define their technical features (e.g., forking the code of a cryptocurrency to change the hashing algorithm, or a zk-PoI cryptocurrency). An Evolutionary Stable Strategy \[def:ESS\] is a strategy that cannot be invaded by any alternative strategy, that is, it can resist to the invasion of a mutant and it’s impenetrable to them: once it’s introduced and becomes dominant in a population, natural selection is sufficient to prevent invasions from new mutant strategies.
\[def:ESS\]The pure strategy $s^{*}$ is an Evolutionary Stable Strategy[@logicAnimalConflict] if there exists $\epsilon_{0}>0$ such that: $$\left(1-\epsilon\right)\left(u\left(s^{*},s^{*}\right)\right)+\epsilon\left(u\left(s^{*},s^{'}\right)\right)>\left(1-\epsilon\right)\left(u\left(s^{'},s^{*}\right)\right)+\epsilon\left(u\left(s^{'},s^{'}\right)\right)$$ for all possible deviations $s^{'}$and for all mutation sizes $\epsilon<\epsilon_{0}$. There are two conditions for a strategy $s^{*}$ to be an Evolutionary Stable Strategy: for all $s^{*}\neq s^{'}$ either
1. $u\left(s^{*},s^{*}\right)>u\left(s^{'},s^{*}\right)$, that is, it’s a Strict Nash Equilibrium \[def:strict-nash\], **or**,
2. if $u\left(s^{*},s^{*}\right)=u\left(s^{'},s^{*}\right)$ then $u\left(s^{*},s^{'}\right)>u\left(s^{'},s^{'}\right)$
Mining a UDCE cryptocurrency is an Evolutionary Stable Strategy.
Since mining a UCDE cryptocurrency is a strictly-dominant strategy and a Strict Nash Equilibrium \[thm:UDCE-cryptocurrency-dominates\], then it is an Evolutionary Stable Strategy because it fulfills its first condition \[def:ESS\].
Additionally, mining a UCDE cryptocurrency based on the global network of National Identity Cards and ePassports is an Evolutionary Stable Strategy over national variants/mutants due to Claim \[claim:(Uniqueness-of-Technical-Solution)\].
Thus, the Game of Rational Mining of Cryptocurrencies \[def:(Game-of-Rational\] is a “survival of the fittest” ecology, where the cheapest cryptocurrency to mine offering the higher profits rises above the others.
### Obviating the Price of Crypto-Anarchy\[sub:Price-of-Crypto-Anarchy\]
The most cost efficient Sybil-resistant mechanism is the one provided by a trusted PKI infrastructure[@the-sybil-attack] and a centralised social planner would prefer the use of National Identity Cards and/or ePassports in order to minimise costs: instead, permissionless blockchains are paying very high costs by using PoW/PoS as Sybil-resistant mechanisms. In this paper, Zero-Knowledge Proof-of-Identity is introduced as a compromise solution between both approaches, thus obtaining a very efficient Sybil-resistant mechanism with the best of both worlds.
In order to measure how the efficiency of a Sybil-resistant mechanism degrades due to the selfish behaviour of its agents (i.e., a fixed amount of block reward to be distributed among a growing and unbounded number of miners paying high energy costs, as in Bitcoin), we compare the ratio between the worst Nash equilibrium and the optimal centralised solution, a concept known as Price of Anarchy in game theory because it bounds and quantifies the costs of the selfish behavior of the agents.
*The Price of Anarchy*[@DBLP:conf/stacs/KoutsoupiasP99]. Consider a game $G=\left(N,S,u\right)$ defined as a set of players $N$, strategy sets $S_{i}$ for each player and utilities $u_{i}:S\rightarrow\mathbb{R}$ (where $S=S_{1}x\ldots xS_{n}$ are also called the set of outcomes). Define a measure of efficiency of each outcome that we want to minimise, $Cost:S\rightarrow\mathbb{R}$, and let $Equil\subseteq S$ be the set of strategies in Nash equilibria. The *Price of Anarchy* is given by the following ratio: $$\mbox{Price of Anarchy}=\frac{\max_{s\in Equil}Cost\left(s\right)}{\min_{s\in S}Cost\left(s\right)}$$
The competition game between several blockchains and their cryptocurrencies can be reformulated[@altman:hal-01906954] as a congestion game[@congestionGames; @potentialGames] (hereby included for completeness), more amenable to the formulations commonly used for analyzing the Price of Anarchy (the necessity for the following definitions is already intuited in the Discussion of [@sokGameTheoryCryptocurrencies]): as the number of miners increases, it also exponentially decreases the chance that a given miner wins the block reward by being the first to solve the mining puzzle (i.e., the system becomes increasingly congested); it has been proved that free entry is solely responsible for determining the resource usage[@bitcoinMarketStructure], and that the difficulty is not an instrument that can regulate it.
#### Miners, mining servers and crypto-currencies
Denote by $\mathcal{N}\coloneqq\left\{ 1,2,\ldots,N\right\} $ the finite set of miners that alter the utilities of other miners if any of them change strategies and let $\mathcal{K}\coloneqq\left\{ 1,2,\ldots,K\right\} $ be the set of cryptocurrencies, each associated to exactly one puzzle that miners are trying to solve. Let $\mathcal{M}\coloneqq\left\{ 1,2,\ldots,M\right\} $denote the set of Edge computing Service Providers (ESPs), or mining servers used to offload the costly computational processing.
#### Strategies
Let $\mathcal{S}_{i}\subset\mathcal{K}x\mathcal{M}$ denote the set of ordered pairs (cryptocurrency, ESP) corresponding to ESPs that miner $i$ can rely on to solve the puzzles of a given cryptocurrency. A strategy for miner $i$ is denoted by $s_{i}\in\mathcal{S}_{i}$ corresponding to the cryptocurrency (puzzle) which a miner intends to solve using a given infrastructure. A strategy vector $s\coloneqq\left(s_{i}\right)_{i\in\mathcal{N}}$ produces a load vector $l\coloneqq\left(l_{k,m}\right)_{k,m}$, where $l_{k,m}$ denotes the number of users mining blockchain $k$ at ESP $m$.
#### Rewards, costs, and utilities
Let $\eta_{k}$ be the load of miners across all ESPs towards cryptocurrency $k.$ Then, $$\eta_{k}\coloneqq\sum_{m^{'}\in\mathcal{M}}l_{k,m'}\mu_{k,m'}$$ For a given load vector $l$, the time to solve the puzzle of the $k^{th}$ cryptocurrency is exponentially distributed with expectation $\nicefrac{1}{\eta_{k}}$. Let $q_{k}$ be the probability that puzzle $k$ is solved by time $T$, $$q_{k}=1-\mbox{exp}\left(-T\eta_{k}\right)$$ The probability that a given miner using ESP $m$ is the first to solve puzzle $k$ is $$p_{k,m}=1_{l_{k,m>0}}q_{k}\mu_{k,m}/\eta_{k}$$ where $1_{c}$ equals 1 if condition $c$ holds and 0 otherwise. For simplification, subscript $m$ can be dropped and we consider a single ESP. Then, the probability that a miner is the first to solve the puzzle is $$p_{k}\left(l_{k}\right)=\left(1-\mbox{exp}\left(-T\mu_{k}l_{k}\right)\right)/l_{k}$$
Let $U_{k,m}\left(l\right)$ denote the utility to a miner who tries to find the solution of the current puzzle associated to cryptocurrency $k$ using ESP $m$ and $\gamma_{k,m}$ denote the cost of mining blockchain $k$ at ESP $m$: $$U_{k,m}\left(l\right)=\begin{cases}
p_{k,m}-\gamma_{k,m} & \mbox{if }p_{k,m}>\gamma_{k,m},\\
0 & \mbox{otherwise}
\end{cases}$$ and the utility of a tagged miner to mine a cryptocurrency $k$ when there are $l_{k}$ miners associated with the same cryptocurrency is $$U_{k}\left(l_{k}\right)=p_{k}-\gamma_{k},\mbox{ if }p_{k}-\gamma_{k}\geq0$$
\[thm:altmanTheo\][@altman:hal-01906954]If for all $i$ and $j$, $S_{i}=S_{j}$ and $s_{i}$ does not depend on $i$, then the Nash equilibrium is given by the solution of the following optimization problem, $$\begin{aligned}
\mbox{argmin}_{s}\Phi\left(s\right) & \coloneqq & \sum_{k\in\mathcal{K}}\sum_{l=1}^{l_{k}}p_{k}\left(l\right)-\gamma_{k}\\
\mbox{subject to:} & & \sum_{k\in\mathcal{K}}l_{k}\leq N,l_{k}\geq0\end{aligned}$$
Let $NashCongestedEquil\subseteq S$ be the set of strategies given as solution of the optimization problem of Theorem \[thm:altmanTheo\], then the *Price of Crypto-Anarchy* is given by the following ratio: $$\mbox{Price of Crypto-Anarchy=\ensuremath{\frac{\mbox{max}_{\mbox{\ensuremath{s}}\in\mbox{\ensuremath{NashCongestedEquil}}}\mbox{\ensuremath{Cost\left(s\right)}}}{\mbox{\ensuremath{Cost\left(\mbox{zk-PoI}\right)}}}}}$$
In practice, the real-world costs of the Zero-Knowledge Proof of Identity can be considered almost zero because it’s subsidised by governments and thus exogenous to any blockchain system. Quite the opposite, the energy costs of PoW cryptocurrencies are notoriously high[@natureEnergyCarbonCosts; @bitcoinCarbonFootprint]: it is estimated that mining Bitcoin, Ethereum, Litecoin and Monero consumed an average of 17, 7, 7 and 14 MJ to generate one US\$, respectively; and that Bitcoin causes around 22 megatons in CO2 emissions annually[@bitcoinCarbonFootprint].
The trivial extension to Proof-of-Stake is left as an exercise to the reader, although it’s not as affordable as it may be seen: as of March 2019, an average of 40% of the cryptocurrency supply is staked at a total of \$4Bn between all PoS blockchains[@stakedPoS]. Actually, Proof-of-Stake is not strictly better than Proof-of-Work as the distribution of the market shares between both technologies has been shown to be indistinguishable (Appendix 3, [@2017arXiv170505334E]).
### Pareto Dominance on Currency Circulation\[sub:Pareto-Dominance-on\]
For completeness, a stylised version of a model of competing currencies[@RePEc:nbr:nberwo:22157] is introduced here to prove that UDCE cryptocurrencies also dominate PLFC cryptocurrencies on their circulation (i.e., trading, speculating) due to their stronger network effects, and not only mining as previously proved. The key observation here is that by definition \[def:(Power-law-fee-concentrated-cryp\], the returns of mining PLFC cryptocurrencies is concentrated on a very limited number of miners and the newly minted cryptocurrency could be held for long periods of times: otherwise, if they didn’t expect that the held cryptocurrencies would appreciate in time, they would be mining another set of cryptocurrencies with better expectations. In direct contrast, the distribution of mining and/or transaction fees of UDCE cryptocurrencies is uniformly distributed by definition \[def:(Uniformly-distributed-fee-uncon\]: therefore, the returns of the holding strategy after minting them would be lower and their subsequent circulation much less restricted.
Suppose an economy divided into periods, each period divided into two subperiods: in the first subperiod, a perishable good demanded by everyone is produced and consumed in a Centralised Market; in the second subperiod, buyers who only consume are randomly matched with sellers who only produce with probability $\sigma\in\left(0,1\right)$ in a Decentralised Market. Let $\beta\in\left(0,1\right)$ denote the discount factor, $\phi_{t}^{i}\in\mathbb{R}_{+}$ denote the value of a unit of currency $i\in\left\{ 1,\ldots,N\right\} $ in terms of the CM food and $\phi_{t}=\left(\phi_{t}^{1},\ldots,\phi_{t}^{N}\right)\in\mathbb{R}_{+}^{N}$ denote the vector of real prices.
(Buyers). In a $\left[0,1\right]$-continuum of buyers, $x_{t}^{b}\in\mathbb{R}$ denotes the buyer’s net consumption of the CM good and $q_{t}\in\mathbb{R}_{+}$denotes the consumption of the DM good. The utility function of the buyer’s preferences is given by $$U^{b}\left(x_{t}^{b},q_{t}\right)=x_{t}^{b}+u\left(q_{t}\right)$$ with $u:\mathbb{R}_{+}\rightarrow\mathbb{R}$ continuously differentiable, increasing and strictly concave with $u'\left(0\right)=\infty$ and $u\left(0\right)=0$.
Let $W^{b}\left(M_{t-1}^{b},t\right)$ denote the value function for a buyer who starts period $t$ holding a portfolio $M_{t-1}^{b}\in\mathbb{R}_{+}^{N}$ of cryptocurrencies in the CM and let $V^{b}\left(M_{t}^{b},t\right)$ denote the value function in the DM: the dynamic programming equation is $$W^{b}\left(M_{t-1}^{b},t\right)=\underset{\left(x_{t}^{b},M_{t}^{b}\right)\in\mathbb{R}\times\mathbb{R}_{+}^{N}}{\mbox{max}}\left[x_{t}^{b}+V^{b}\left(M_{t}^{b},t\right)\right]$$ subject to the budget constraint $$\phi_{t}\cdot M_{t}^{b}+x_{t}^{b}=\phi_{t}\cdot M_{t-1}^{b}.$$ The value for a buyer holding a portfolio $M_{t}^{b}$ in the DM is $$\begin{aligned}
V^{b}\left(M_{t}^{b},t\right) & = & \sigma\left[u\left(q\left(M_{t}^{b},t\right)\right)+\beta W^{b}\left(M_{t}^{b}-d\left(M_{t}^{b},t\right),t+1\right)\right]\\
& & +\left(1-\sigma\right)\beta W^{b}\left(M_{t}^{b},t+1\right)\end{aligned}$$ and let $q^{*}\in\mathbb{R}$ denote the quantity satisfying $u'\left(q^{*}\right)=w'\left(q^{*}\right)$ so that $q^{*}$ gives the surplus-maximizing quantity that determines the efficient level of production in the DM. The solution to the bargaining problem is given by $$q\left(M_{t}^{b},t\right)=\begin{cases}
m^{-1}\left(\beta\times\phi_{t+1}\cdot M_{t}^{b}\right) & \mbox{if}\,\phi_{t+1}\cdot M_{t}^{b}<\beta^{-1}\left[\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)\right]\\
q^{*} & \mbox{if}\,\phi_{t+1}\cdot M_{t}^{b}\geq\beta^{-1}\left[\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)\right]
\end{cases}$$ and $$\phi_{t+1}\cdot d\left(M_{t}^{b},t\right)=\begin{cases}
\phi_{t+1}\cdot M_{t}^{b} & \mbox{if}\,\phi_{t+1}\cdot M_{t}^{b}<\beta^{-1}\left[\theta w\left(q^{*}\right)\right.\\
& \left.+\left(1-\theta\right)u\left(q^{*}\right)\right]\\
\beta^{-1}\left[\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)\right] & \mbox{if}\,\phi_{t+1}\cdot M_{t}^{b}\geq\beta^{-1}\left[\theta w\left(q^{*}\right)\right.\\
& \left.+\left(1-\theta\right)u\left(q^{*}\right)\right]
\end{cases}$$ with the function $m:\mathbb{\mathbb{R}}_{+}\rightarrow\mathbb{R}$ defined as $$m\left(q\right)\equiv\frac{\left(1-\theta\right)u\left(q\right)w'\left(q\right)+\theta w\left(q\right)u'\left(q\right)}{\theta u'\left(q\right)+\left(1-\theta\right)w'\left(q\right)}.$$ The optimal portfolio problem can be defined as $$\underset{M_{t}^{b}\in\mathbb{R}_{+}^{N}}{\mbox{max}}\left\{ -\phi_{t}\cdot M_{t}^{b}+\sigma\left[u\left(q\left(M_{t}^{b},t\right)\right)-\beta\times\phi_{t+1}\cdot d\left(M_{t}^{b},t\right)\right]+\beta\times\phi_{t+1}\cdot M_{t}^{b}\right\}$$ thus the optimal choice satisfies $$\phi_{t}^{i}=\beta\phi_{t+1}^{i}L_{\theta}\left(\phi_{t+1}\cdot M_{t}^{b}\right)\label{eq:1}$$ for every type $i\in\left\{ 1,\ldots,N\right\} $ together with the transversality condition $$\underset{t\rightarrow\infty}{\mbox{lim}}\beta^{t}\times\phi_{t}\cdot M_{t}^{b}=0\label{eq:2}$$ where $L_{\theta}:\mathbb{R}_{+}\rightarrow\mathbb{R}_{+}$ is given by $$L_{\theta}\left(A\right)=\begin{cases}
\sigma\frac{u'\left(m^{-1}\left(\beta A\right)\right)}{w'\left(m^{-1}\left(\beta A\right)\right)}+1-\sigma & \mbox{if}\, A<\beta^{-1}\left[\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)\right]\\
1 & \mbox{if}\, A\geq\beta^{-1}\left[\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)\right]
\end{cases}$$
(Sellers). In a $\left[0,1\right]$-continuum of sellers, $x_{t}^{s}\in\mathbb{R}$ denotes the seller’s net consumption of the CM good and $n_{t}\in\mathbb{R}_{+}$denotes the seller’s effort level to produce the DM good. The utility function of the seller’s preferences is given by $$U^{s}\left(x_{t}^{s},n_{t}\right)=x_{t}^{s}-w\left(n_{t}\right)$$ with $w:\mathbb{R}_{+}\rightarrow\mathbb{R}_{+}$ continuously differentiable, increasing and weakly convex with $w\left(0\right)=0$.
Let $W^{s}\left(M_{t-1}^{s},t\right)$ denote the value function for a seller who enters period $t$ holding a portfolio $M_{t-1}^{s}\in\mathbb{R}_{+}^{N}$ of cryptocurrencies in the CM and let $V^{s}\left(M_{t}^{s},t\right)$ denote the value function in the DM: the dynamic programming equation is $$W^{s}\left(M_{t-1}^{s}\right)=\underset{\left(x_{t}^{s},M_{t}^{s}\right)\in\mathbb{R}\times\mathbb{R}_{+}^{N}}{\mbox{max}}\left[x_{t}^{s}+V^{s}\left(M_{t}^{s},t\right)\right]$$ subject to the budget constraint $$\phi_{t}\cdot M_{t}^{s}+x_{t}^{s}=\phi_{t}\cdot M_{t-1}^{s}.$$ The value for a seller holding a portfolio $M_{t}^{s}$ in the DM is $$\begin{aligned}
V^{s}\left(M_{t}^{s},t\right) & = & \sigma\left[-w\left(q\left(M_{t}^{b},t\right)\right)+\beta W^{s}\left(M_{t}^{s}+d\left(M_{t}^{b},t\right),t+1\right)\right]\\
& & +\left(1-\sigma\right)\beta W^{s}\left(M_{t}^{s},t+1\right)\end{aligned}$$
(Miners). In a $\left[0,1\right]$-continuum of miners of each type-$i\in\left\{ 1,\ldots,N\right\} $ token, $x_{t}^{i}\in\mathbb{R}_{+}$ denotes the miner’s consumption of the CM good and $\triangle_{t}^{i}\in\mathbb{R}_{+}$denotes the production of the type-$i$ token. The utility function of the miner’s preferences is given by $$U^{e}\left(x_{t}^{i},\Delta_{t}\right)=x_{t}^{i}-c\left(\Delta_{t}^{i}\right)$$ with the cost function $c:\mathbb{R}_{+}\rightarrow\mathbb{R}_{+}$ continuously differentiable, strictly increasing and weakly convex with $c\left(0\right)=0$.
Let $M_{t}^{i}\in\mathbb{R}_{+}$ denote the per-capita supply of cryptocurrency $i$ in period $t$ and $\Delta_{t}^{i}\in\mathbb{R}$ denote the miner $i$’s net circulation of newly minted tokens in period $t$. To describe the miner’s problem to determine the money supply in the economy, we start assuming that all miners solve the same decision problem, thus the law of motion of type-$i$ tokens at all date $t\geq0$ is given by
$$M_{t}^{i}=\Delta_{t}^{i}+M_{t-1}^{i}\label{eq:4}$$
where $M_{-1}^{i}\in\mathbb{R}_{+}$ denotes the initial stock. The budget constraint is $$x_{j}^{i}=\phi_{t}^{i}\Delta_{t}^{i},$$ and given that the miner takes prices $\left\{ \phi_{t}\right\} _{t=0}^{\infty}$ as given, the profit maximization of the cryptocurrency emission problem is solved by $$\Delta_{t}^{*,i}\in\underset{\Delta\in\mathbb{R}_{+}}{\mbox{arg max}}\left[\phi_{t}^{i}\Delta-c\left(\Delta\right)\right]\label{eq:3}$$
(Equilibrium). A perfect-foresight monetary equilibrium is an array $\left\{ M_{t},M_{t}^{b},\Delta_{t}^{*},\phi_{t}\right\} _{t=0}^{\infty}$ satisfying \[eq:1\], \[eq:2\], \[eq:3\] and \[eq:4\] for each $i\in\left\{ 1,\ldots,N\right\} $ at all dates $t\geq0$ and satisfying the following market-clearing condition $$M_{t}=M_{t}^{b}+M_{t}^{s}$$
Suppose that each miner $j$ starts with $M^{i}>0$ units of currency $i\in\left\{ 1,\ldots,N\right\} $: let $\delta$ denote the fraction $1-\delta$ of randomly selected miners in each location $j$ at each date $t\geq0$ who doesn’t offer their tokens to sellers because they are holding them in expectation of their appreciation (i.e., PLFC cryptocurrencies), so these tokens don’t circulate to other $j$ positions whenever sellers are relocated.
Conversely, an equilibrium with the property that miners don’t restrict the circulation of recently mined tokens (i.e., UDCE cryptocurrencies) is as follows: the optimal portfolio choice implies the first-order condition $$\frac{u'\left(q\left(M_{t},t\right)\right)}{w'\left(q\left(M_{t},t\right)\right)}=\frac{1}{\beta\gamma_{t+1}^{i}}$$ for each currency $i$, where $\gamma_{t+1}\in\mathbb{R}_{+}$ represents the common return across all valued currencies between dates $t$ and $t+1$. The demand for real balances in each location is given by $$z\left(\gamma_{t+1};1\right)\equiv\frac{1}{\gamma_{t+1}}L_{1}^{-1}\left(\frac{1}{\beta\gamma_{t+1}}\right)$$ because $$\beta\gamma_{t+1}\sum_{i=1}^{N}b_{t}^{i}<\theta w\left(q^{*}\right)+\left(1-\theta\right)u\left(q^{*}\right)$$ and with $$L_{\delta}\left(A\right)=\begin{cases}
\delta\frac{u'\left(m^{-1}\left(\beta A\right)\right)}{w'\left(m^{-1}\left(\beta A\right)\right)}+1-\delta & \mbox{if}\, A<\beta^{-1}w\left(q^{*}\right)\\
1 & \mbox{if}\, A\geq\beta^{-1}w\left(q^{*}\right)
\end{cases}$$ Because the market-clearing condition implies $$\sum_{i=1}^{N}\phi_{t}^{i}M^{i}=z\left(\gamma_{t+1};1\right)$$ the equilibrium sequence $\left\{ \gamma_{t}\right\} _{t=0}^{\infty}$ satisfies the law of motion $$z\left(\gamma_{t+1};1\right)=\gamma_{t}z\left(\gamma_{t};1\right)$$ because $$M_{t}^{i}=M_{t-1}^{i}=\Delta^{i}$$ for each $i$ and provided that $\gamma_{t}\leq t$, and the boundary condition $$\beta\gamma_{t}z\left(\gamma_{t};1\right)\leq w\left(q^{*}\right)$$ Suppose $u\left(q\right)=\left(1-\eta\right)^{-1}q^{1-\eta}$, with $0<\eta<1$, and $w\left(q\right)=\left(1+\alpha\right)^{-1}q^{1+\alpha}$ with $\alpha\geq0$. The dynamic system describing the equilibrium evolution of $\gamma_{t}$ is $$\gamma_{t+1}^{\frac{1+\alpha}{\eta+\alpha}-1}=\gamma_{t}^{\frac{1+\alpha}{\eta+\alpha}}\label{eq:28}$$
The allocation associated with the circulation of UDCE cryptocurrencies Pareto dominates the allocation with the associated the circulation of PLFC cryptocurrencies, on a stationary equilibrium with the property that the quantity traded in the Decentralised Market is given by $\hat{q}\left(1\right)\in\left(\hat{q}\left(\delta\right),q^{*}\right)$ satisfying $$\frac{u'\left(\hat{q}\left(1\right)\right)}{w'\left(\hat{q}\left(1\right)\right)}=\beta^{-1}\label{eq:29}$$
The sequence $\gamma_{t}=1$ for all $t\geq0$ satifies \[eq:28\]. Then, the solution to the optimal portfolio problem implies that the DM output must satisfy \[eq:29\]. The quantity $\hat{q}$ satisfies $$\frac{u'\left(\hat{q}\left(1\right)\right)}{w'\left(\hat{q}\left(1\right)\right)}=\delta\frac{u'\left(\hat{q}\left(\delta\right)\right)}{w'\left(\hat{q}\left(\delta\right)\right)}+1-\delta$$ Because $\delta\in\left(0,1\right)$, we have $\hat{q}\left(1\right)>\hat{q}\left(\delta\right)$, that is, the allocation associated with the circulation of UCDE cryptocurrencies -$\hat{q}\left(1\right)$- Pareto dominates the allocation associated with the circulation of PLFC cryptocurrencies -$\hat{q}\left(\delta\right)$-.
### On Network Effects\[sub:On-Network-Effects\]
At the time of the release of this paper, cryptocurrencies have failed to provide an alternative to traditional payment networks due to a combination of high transaction fees, high finalization time and high volatility. The failure to find the favor of merchants is also their biggest weakness: they aren’t part of two-sided networks, and thus easily replaceable by any newer cryptocurrency better able to create them. Actually, the first-mover advantage of the most valued cryptocurrencies is lower than expected if any competing cryptocurrency leverages network effects from other different sources (e.g., Zero-Knowledge Proof-of-Identity from trusted PKI certificates).
A simple model is introduced here to analyze the evolution of competing payment networks: consider the two-sided and incompatible payment networks of two cryptocurrencies, BTC and zk-PoI, each with their corresponding groups of merchants $m$ and customers $c$; let $m_{BTC}^{t},m_{zkPOI}^{t}$ denote the number of merchants at time $t$ and $c_{BTC}^{t},c_{zkPOI}^{t}$ the number of customers. A user joins the payment networks at each time step $t$, with $\lambda$ being the probability of being a customer and $1-\lambda$ of being a merchant: each merchant prefers to join BTC or zk-PoI depending on the number of customers in the same network, thus the probabilities to join one of the networks are given by $$\frac{c_{BTC}^{\beta}}{c_{BTC}^{\beta}+c_{zkPOI}^{\beta}},\frac{c_{zkPOI}^{\beta}}{c_{BTC}^{\beta}+c_{zkPOI}^{\beta}}$$ and conversely, for customers the probabilities are given by $$\frac{m_{BTC}^{\alpha}}{m_{BTC}^{\alpha}+m_{zkPOI}^{\alpha}},\frac{m_{zkPOI}^{\alpha}}{m_{BTC}^{\alpha}+m_{zkPOI}^{\alpha}}.$$ Note that some categories of users would prefer to use the expected number of users and not their current tally: forward-looking merchants that need to invest on equipment to access the payment network are within this group, thus they would prefer to use expected numbers, $$\frac{E\left(c_{BTC}^{\beta}\right)}{E\left(c_{BTC}^{\beta}\right)+E\left(c_{zkPOI}^{\beta}\right)},\frac{E\left(c_{zkPOI}^{\beta}\right)}{E\left(c_{BTC}^{\beta}\right)+E\left(c_{zkPOI}^{\beta}\right)}.$$ Each user can only join one payment network, modelling the fact that single-homing is preferred to multi-homing in the real-world, and the particular network is determined by the distribution of users on the other side at each time $t$. The parameters $\alpha,\beta>0$ are elasticities of demand with regard to the numbers of users on the other side of the payment network, effectively acting as measures of indirect network effects: $\alpha$ can be empirically estimated by observing joining customers over a small period of time and then calculating $$\alpha=\frac{\mbox{ln }\left(\nicefrac{m_{BTC}^{\alpha}}{\left(m_{BTC}^{\alpha}+m_{zkPOI}^{\alpha}\right)}\right)-\mbox{ln }\left(1-\left(\nicefrac{m_{BTC}^{\alpha}}{\left(m_{BTC}^{\alpha}+m_{zkPOI}^{\alpha}\right)}\right)\right)}{\mbox{ln }m_{BTC}-\mbox{ln }m_{zkPOI}}$$ and conversely$\beta$ can be empirically estimated by observing joining merchants and then calculating $$\beta=\frac{\mbox{ln }\left(\nicefrac{c_{BTC}^{\beta}}{\left(c_{BTC}^{\beta}+c_{zkPOI}^{\beta}\right)}\right)-\mbox{ln }\left(1-\left(\nicefrac{c_{BTC}^{\beta}}{\left(c_{BTC}^{\beta}+c_{zkPOI}^{\beta}\right)}\right)\right)}{\mbox{ln }c_{BTC}-\mbox{ln }c_{zkPOI}}.$$
(Dominance of the Zero-Knowledge Proof-of-Identity cryptocurrency). A new cryptocurrency could achieve dominance over previous cryptocurrencies, overcoming first-mover advantages, if the expected number of accepting customers would be much higher and the number of merchants using the previous cryptocurrencies is low.
Note that the number of steps needed for a new cryptocurrency, $m_{zkPOI}$, to overtake the previous one, $m_{BTC}$, on the number of merchants, $m_{zkPOI}>m_{BTC}$, is given by $$\left(m_{BTC}+1\right)\cdot\left(1-\lambda\right)^{-1}$$ It’s possible for a new cryptocurrency to overtake a previous one on the number of merchants whenever $$E\left(c_{zkPOI}\right)-E\left(c_{BTC}\right)>\left(m_{BTC}+1\right)\cdot\left(1-\lambda\right)^{-1}$$ and since $m_{BTC}$ is a low number and $E\left(c_{zkPOI}\right)\gg E\left(c_{BTC}\right)$, it’s conceivable that the previous condition could hold.
Now let’s consider the results of strong network effects on the final market shares of both payment networks by examining the following differential equations, $$\frac{d\left(\nicefrac{m_{BTC}}{m_{zkPOI}}\right)}{dt}=\left(1-\lambda\right)\frac{\left(\nicefrac{c_{BTC}}{c_{zkPOI}}\right)^{\beta}-\left(\nicefrac{m_{BTC}}{m_{zkPOI}}\right)}{\left(1+\left(\nicefrac{c_{BTC}}{c_{zkPOI}}\right)^{\beta}\right)m_{zkPOI}}$$ and $$\frac{d\left(\nicefrac{c_{BTC}}{c_{zkPOI}}\right)}{dt}=\lambda\frac{\left(\nicefrac{m_{BTC}}{m_{zkPOI}}\right)^{\alpha}-\left(\nicefrac{c_{BTC}}{c_{zkPOI}}\right)}{\left(1+\left(\nicefrac{m_{BTC}}{m_{zkPOI}}\right)^{\alpha}\right)c_{zkPOI}}$$
According to the signs of the previous derivatives, when $\alpha\cdot\beta>1$ and $t\rightarrow\infty$, the payment network with even a slight advantage over the other will end acquiring all the merchants and customers, for example $$\underset{t\rightarrow\infty}{\lim}m_{zkPOI}=\infty,\underset{t\rightarrow\infty}{\lim}c_{zkPOI}=\infty$$ $$\underset{t\rightarrow\alpha}{\lim}m_{BTC}=0,\underset{t\rightarrow\alpha}{\lim}c_{BTC}=0$$ but with $\alpha\cdot\beta<1$, the number of merchants and customers will equalize $$m_{zkPOI}=m_{BTC},c_{zkPOI}=c_{BTC}$$ thus highlighting the importance of network effects.
### Dominance over Cash and other Cryptocurrencies\[sub:Dominance-over-Cash\]
The dominance of subsection \[sub:Strictly-Dominant-Cryptocurrenci\] is based on mining and subsection \[sub:Pareto-Dominance-on\] extends said dominance to the circulation of currencies: in this subsection, the dominance will be based on the lower costs of a payment network of the cryptocurrency using Zero-Knowledge Proof-of-Identity; therefore, there exists a unique equilibrium in which this payment system dominates.
A recent paper[@coordinationElectronicPaymentInstrument] offers a model based on a version of Lagos-Wright[@lagosWright] to explain the substitution of cash by debit cards or any other non-deferred electronic payment system incurring a fixed cost $\Omega\left(z\right)\tau$ per each period $\tau$, the cost $\Omega\left(z\right)$ being financed by imposing fee $\omega$ on each payment where $\omega$ should satisfy $$\Omega\left(z\right)=S\left[\theta\omega+\left(1-\theta\right)\omega\right]$$ and where $S$ denotes the instantaneous measure of electronic payment transactions, $z$ is the state of development of the economy, $\theta\in\left[0,1\right]$ is the share of cost allocated to a buyer and $\left(1-\theta\right)$ is the share of cost allocated to a seller. In this model, an electronic payment system can achieve dominance over cash using the solution concept of iterative elimination of conditionally dominated strategies whenever the state of development of the economy $z$ is sufficiently high, and there exists a unique equilibrium in the model such that agents choose electronic payment transactions when $z$ is strictly higher than the limiting cut-off function $Z_{\infty}$ of the sequence of boundaries $Z_{0,}Z_{1,}\ldots$ of regions where an agent chooses electronic payment transactions regardless of the choices of other agents. In other words, it’s strictly dominant to choose electronic payments in an economy having sufficiently advanced information technology so that $\Omega$ is negligible.
Since the cost function $\Omega\left(z\right)$ of a UCDE cryptocurrency based on Zero-Knowledge Proof-of-Identity is much cheaper than of PoW/PoS cryptocurrencies and other forms of electronic payment because its cards are already distributed (i.e., de facto subsidised by governments), there exists a unique equilibrium in the model [@coordinationElectronicPaymentInstrument] such that the agents choose the UCDE cryptocurrency using zk-PoI and it dominates the other forms of payment.
Conclusion
==========
Although all permissionless blockchains critically depend on Proof-of-Work or Proof-of-Stake to prevent Sybil attacks, their high resource consumption corroborates their non-scalability and act as a limiting factor to the general diffusion of blockchains. This paper proposed an alternative approach that not only doesn’t waste resources, it could also help in the real-world identity challenges faced by permissionless blockchains: the derivation of anonymous credentials from widely trusted public PKI certificates.
Additionally, we study the better incentives offered by the proposed cryptocurrency based on our anonymous authentication scheme: mining is proved to be incentive-compatible and a strictly dominant strategy over previous cryptocurrencies, thus a Nash equilibrium over previous cryptocurrencies and an Evolutionary Stable Strategy; furthermore, zk-PoI is proved to be optimal because it implements the social optimum, unlike PoW/PoS cryptocurrencies that are paying the Price of (Crypto-)Anarchy. The circulation of the proposed cryptocurrency is proved to Pareto dominate other cryptocurrencies based on its negligible mining costs and it could also become dominant thanks to stronger network effects; finally, the lower costs of its infrastructure imply the existence of a unique equilibrium where it dominates other forms of payment.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we contribute to the literature on energy market co-movement by studying its dynamics in the time-frequency domain. The novelty of our approach lies in the application of wavelet tools to commodity market data. A major part of economic time series analysis is done in the time or frequency domain separately. Wavelet analysis combines these two fundamental approaches allowing study of the time series in the time-frequency domain. Using this framework, we propose a new, model-free way of estimating time-varying correlations. In the empirical analysis, we connect our approach to the dynamic conditional correlation approach of [@engle2002] on the main components of the energy sector. Namely, we use crude oil, gasoline, heating oil, and natural gas on a nearest-future basis over a period of approximately 16 and $1/2$ years beginning on November 1, 1993 and ending on July 21, 2010. Using wavelet coherence, we uncover interesting dynamics of correlations between energy commodities in the time-frequency space.'
address:
- 'Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic, Pod Vodarenskou Vezi 4, 182 00, Prague, Czech Republic'
- 'Institute of Economic Studies, Charles University, Opletalova 21, 110 00, Prague, Czech Republic'
author:
- Lukas Vacha
- Jozef Barunik
bibliography:
- 'Energy.bib'
title: 'Co-movement of energy commodities revisited: Evidence from wavelet coherence analysis'
---
correlation ,co-movement ,wavelet analysis ,wavelet coherence
Introduction
============
Energy commodities play an important role in economic research, as they affect a wide range of markets and a variety of market participants operating on these markets. Study of the dynamics and statistical properties of energy commodities has become an important part of financial analysis since commodities became an additional tool for international diversification between stocks, bonds and currencies. Energy-related costs play a crucial role in the decision making of industrial companies and entrepreneurs. A proper understanding of the price dynamics and interconnections between commodities has become a fundamental concern. Moreover, energy commodity prices tend to have different and more extreme statistical properties than prices of other financial assets such as stocks, bonds, exchange rates, and corresponding derivatives. Prices of energy commodities are also influenced by various types of investors. Apart from standard financial investors, demand for energy commodities consists of industrial entities and power stations using the commodities to produce electricity.
The commodity markets are complex systems of interacting agents with different term objectives. Hence, time series resulting from this process are formed by a combination of different components operating at different frequencies. Standard time series econometric methods usually consider the frequency and time components separately. In this paper we introduce the wavelet approach, which allows us to study the frequency components of time series without losing the time information. The introduction of wavelets helps us uncover interactions which are hard to see using any other modern econometric method and which would otherwise stay hidden. Moreover, the wavelet analysis approach is model-free. This property makes it a very powerful tool in comparison with other methods that rely on parameters as well as the estimation method. We are not the first to apply wavelets in the analysis of commodities. [@davidson] proposed a form of semi-nonparametric regression based on wavelet analysis to study commodity price behavior. [@yousefi] used wavelets for prediction of oil prices. [@ConnorRossiter] were the first to estimate price correlations by basic scale decomposition of time series using wavelets on commodity markets. They use the discrete type of wavelet transformation, while in our analysis we follow the continuous transform approach. Contrary to [@ConnorRossiter], we study the dynamics of correlations in the time-frequency space; our results are thus even more general. Recently, [@Naccache2011] used wavelets for the correlation analysis of oil prices and economic activity. However, that study was restricted to monthly data, with its main focus on longer cycles.
In recent years, applied research on the dynamics of energy commodities has grown significantly. Literature on the topic has been presented in three main categories – research on the dynamics of specific commodity time series, research on the macroeconomic relationships of energy commodities, and research on co-movements between commodities. Several studies have examined the topic of co-movement of commodities. For example, [@Lanza2006] examined the dynamic conditional correlations in daily returns of WTI oil futures prices and found these to vary significantly. [@Grine2010] used cross-correlations to successfully construct a jump-diffusion model for pricing of cross-commodity derivatives of energy assets. Finally, [@Ghoshray2010] concluded that trends in energy prices change frequently and are hard to predict.
In our research, we contribute to the discussion of co-movements and use a novel approach which is much easier to interpret. The application of wavelets enables us to study the interdependence of energy time series in the time as well as frequency domains, providing a deeper understanding of possible dependencies. The main question of our analysis is whether the interconnection between studied commodity markets changes significantly in time and varies across different investment horizons.
More specifically, we focus on dynamic correlations based on wavelet coherence between crude oil, gasoline, heating oil, and natural gas. By doing so, we analyze the evolution of these correlations in time as well as for different frequencies. This approach distinguishes between different types of investors with different investment horizons. Moreover, cyclical components, which are an important part of energy commodity dynamics, are also easily distinguishable and separated between various time periods. In addition to wavelet coherence in the time-frequency space, we propose a new way of estimating model-free time-varying correlations. To relate our results to the standard econometric literature, we connect our approach with the standard econometric approach of [@engle2002] – dynamic conditional correlations from a multivariate GARCH model.
This paper is structured as follows. After a brief introduction to wavelets, Section 2 presents the wavelet coherence used for the estimation of local correlation in the time-frequency domain, as well as the phase differences used to characterize the relationship between the two time series. The comparison model – DCC GARCH – is also briefly introduced in this part. Section 3 introduces the data and provides results of the empirical analysis. Section 4 concludes by stating our main result, namely, a very interesting evolution of correlations between energy commodities.
Methodology
===========
Wavelet analysis
----------------
The wavelet transform offers localized frequency decomposition, providing information about frequency components. As a result, wavelets have significant advantages over basic Fourier analysis when the object under study is locally stationary and inhomogeneous – see [@Gencay2002; @PercivalWalden2000; @Ramsay2002]. In our work we use continuous wavelet analysis tools, mainly wavelet coherence, measuring the degree of local correlation between two time series in the time-frequency domain, and wavelet coherence phase differences. Before introducing wavelet coherence, let us provide basic definitions of wavelet and wavelet transform.
### Wavelet
A wavelet is a real-valued square integrable function, $\psi\in L^2(\mathbb{R})$[^1], defined as: $$\psi_{u,s}(t)=\frac{1}{\sqrt s}\psi \left( \frac{t-u}{s}\right),$$ where the term $1/\sqrt{s}$ denotes a normalization factor ensuring unit variance of the wavelet, $\| \psi_{u,s} \|^2=1$. A wavelet has two control parameters, $u$ and $s$. The location parameter $u$ determines the exact position of the wavelet and the scale parameter $s$ defines how the wavelet is stretched or dilated. Scale has an inverse relation to frequency; thus lower (higher) scale means a more (less) compressed wavelet, which is able to detect higher (lower) frequencies of a time series. In addition, there are several conditions that a wavelet needs to satisfy. The most important is the admissibility condition, ensuring reconstruction of a time series from its wavelet transform. The admissibility condition is defined as $$C_{\psi}=\int_{0}^{\infty }\frac{\mid\Psi(f)\mid^2}{f}df<\infty,$$ where $\Psi(f)$ is the Fourier transform of a wavelet $\psi(.)$. The admissibility condition implies that the wavelet does not have a zero frequency component and so the wavelet has zero mean, $\int_{-\infty }^{\infty }\psi (t)dt=0$. Further, the wavelet is usually normalized to have unit energy, i.e., $\int_{-\infty }^{\infty }\psi^2 (t)dt=1$, implying that the wavelet makes some excursion away from zero.
There is a large number of different wavelets. Each wavelet has its specific characteristics and is used for different purposes – see [@PercivalWalden2000; @Adisson2002]. In our analysis we use the Morlet wavelet, defined as: $$\psi^M(t)=\frac{1}{\pi^{1/4}}e^{i\omega_0 t}e^{-t^2/2}.$$ Parameter $\omega_0$ denotes the central frequency of the wavelet. We set $\omega_0=6$, which is often used in economic applications – see for example [@Conraria2008; @RuaNunes2009]. The Morlet wavelet belongs to the family of complex or analytic wavelets, hence this wavelet has both real and imaginary parts, allowing us to study both amplitude and phase.
### The continuous wavelet transform
The continuous wavelet transform $W_x(u,s)$ is obtained by projecting a specific wavelet $\psi(.)$ onto the examined time series $x(t)\in L^2(\mathbb{R})$, i.e., $$W_x (u,s)=\int_{-\infty}^\infty x(t)\frac{1}{\sqrt s}\overline{\psi \left( \frac{t-u}{s}\right)} dt.
\label{eq7}$$ An important feature of the continuous wavelet transform is the ability to decompose and then subsequently perfectly reconstruct a time series $x(t)\in L^2(\mathbb{R})$: $$x(t)=\frac{1}{C_\psi}\int_0^\infty \left[\int_{-\infty}^\infty W_x (u,s) \psi_{u,s}(t) du \right] \frac{ds}{s^2} ,\hspace{5 mm}s>0.$$ Furthermore, the continuous wavelet transform preserves the energy of the examined time series, $$\| x \|^2 =\frac{1}{C_\psi}\int_0^\infty \left[\int_{-\infty}^\infty \left| W_x (u,s)\right|^2 du \right] \frac{ds}{s^2}.$$ We use this property for the definition of wavelet coherence, which measures the size of the local correlation between two time series.
### Wavelet coherence
To be able to study the interaction between two time series, we need to introduce a bivariate framework called wavelet coherence. For the proper definition of the wavelet coherence, we need to introduce the cross wavelet transform and cross wavelet power first. [@TorenceCompo98] defined the cross wavelet transform of two time series $x(t)$ and $y(t)$ as $$W_{xy} (u,s) = W_x (u,s) W_y^* (u,s),$$ where $W_x (u,s)$ and $W_y (u,s)$ are continuous wavelet transforms of $x(t)$ and $y(t)$, respectively, $u$ is a position index, and $s$ denotes the scale, while the symbol $^*$ denotes a complex conjugate. The cross wavelet power can easily be computed using the cross wavelet transform as $|W_{xy} (u,s)|$. The cross wavelet power reveals areas in the time-frequency space where the time series show a high common power, i.e., it represents the local covariance between the time series at each scale.
The wavelet coherence can detect regions in the time-frequency space where the examined time series co-move, but do not necessarily have a high common power. Following the approach of [@TorrenceWebster99], we define the squared wavelet coherence coefficient as: $$R^2 (u,s)=\frac{|S(s^{-1}W_{xy} (u,s))|^2}{S(s^{-1}|W_x (u,s)|^2) S(s^{-1}|W_y (u,s)|^2)},$$ where $S$ is a smoothing operator[^2]. The squared wavelet coherence coefficient is in the range $0\le R^2 (u,s) \le1$. Values close to zero indicate weak correlation, while values close to one provide evidence of strong correlation. Hence, the squared wavelet coherence measures the local linear correlation between two stationary time series at each scale and is analogous to the squared correlation coefficient in linear regression.
Since the theoretical distribution for the wavelet coherence is not known, we test the statistical significance using Monte Carlo methods. In the testing procedure, we follow the approach of [@Grinsted2004] and [@TorenceCompo98].
The use of wavelets brings with it the difficulty of dealing with boundary conditions on a dataset with finite length. This is a common problem with any transformation relying on filters. In our paper, we deal with this problem by padding the time series with a sufficient number of zeroes. The area where the errors caused by discontinuities in the wavelet transform cannot be ignored, i.e., where edge effects become important, is called the cone of influence[^3] [@Grinsted2004]. The cone of influence lies under a cone which is bordered by a thin black line.
### Phase
To complete our analysis, we also use wavelet coherence phase differences showing us details about the delays in the oscillation (cycles) between the two time series under study. Following [@TorrenceWebster99] we define the wavelet coherence phase difference as: $$\phi_{xy} (u,s)=\tan ^{-1}\left( \frac{\Im \{S(s^{-1}W_{xy} (u,s))\}}{\Re \{S(s^{-1}W_{xy} (u,s))\} }\right).$$ Phase is indicated by arrows on the wavelet coherence plots. A zero phase difference means that the examined time series move together. The arrows point to the right (left) when the time series are in-phase (anti-phase) or are positively (negatively) correlated. Arrows pointing up means that the first time series leads the second one by $90^{\circ}$, whereas arrows pointing down indicates that the second time series leads the first one by $90^{\circ}$. Usually we have a mixture of positions, for example, an arrow pointing up and right means that the time series are in phase, with the first times series leading the second one.
DCC GARCH analysis
------------------
We would like to provide a connection of this non-traditional analysis to standard econometric tools. In fact, we step back from the time-frequency domain into the time domain to provide this connection. For this purpose, we compute local time-varying correlations from the wavelet coherence and compare them with the well-established multivariate concept of Dynamic Conditional Correlation Generalized Autoregressive Conditional Heteroscedasticity (DCC GARCH) of [@engle2002]. In this part, we will provide a very basic overview of the model.
The DCC estimator is a logical extension of the constant conditional correlation (CCC) model introduced by [@bollerslev90]. In Bollerslev’s model, correlation matrix $R$ is constant: $H_t=D_t R D_t,$ where $D_t=diag\{\sqrt{h_{i,t}}\}$ and $h_{i,t}$ represents the $i$-th univariate (G)ARCH$(p,q)$ process, and $i=1,\dots,n$ at time $t=1,\dots,T$. [@engle2002] allowed $R$ to vary in time $t$, thus $$H_t=D_t R_t D_t.$$ The correlation matrix is then given by the transformation $$R_t=diag(\sqrt{q_{11,t}},\dots,\sqrt{q_{nn,t}}) Q_t diag(\sqrt{q_{11,t}},\dots,\sqrt{q_{nn,t}}),$$ where $Q_t=(q_{ij,t})$ is $$Q_t=(1-\alpha-\beta) \overline{Q}+\alpha\eta_{t-1}\eta'_{t-1}+\beta Q_{t-1},
\label{eqQ}$$ where $\eta_t=\epsilon_{i,t}/\sqrt{h_{i,t}}$ are the standardized residuals from the (G)ARCH model, $\overline{Q}=T^{-1}\sum{\eta_t \eta'_t}$ is a $n\times n$ unconditional variance matrix of $\eta_t$, and $\alpha$ and $\beta$ are non-negative scalars such that $\alpha+\beta<1$.
To estimate DCC GARCH, we use the standard procedure proposed by [@engle2002].
Empirical results
=================
Data description
----------------
To study the dependence between energy markets, we use the main components of the energy sector according to the Continuous Commodity Index constructed by the Commodity Research Bureau. Namely, we use crude oil, gasoline, heating oil, and natural gas on a nearest-future basis. The energy prices were collected on a daily basis over a period of approximately 16 and $1/2$ years beginning on November 1, 1993 and ending on July 21, 2010. Altogether, the sample includes 3573 daily prices for each commodity. The data were obtained from the Pinnacle Data Corporation. Figure \[fig:WTC\] shows the normalized plots of the prices and Table \[stats\] in Appendix A provides descriptive statistics for the logarithmic returns used in the analysis.
In Figure \[fig:WTC\] we can directly observe that heating oil and crude oil are closely related. Gasoline departs from this relationship but still has some common trends during several specific periods. Natural gas seems to have the weakest relation to the other commodities. Let us explore these dependencies in more detail. We will first use the simple unconditional correlation coefficient of the returns for the whole data set, and then we will look at the evolution of the local correlations using wavelet coherence and connect the wavelet coherence results using the standard DCC GARCH approach.
![Normalized prices of heating oil, gasoline, natural gas, and crude oil.[]{data-label="fig:WTC"}](prices)
Unconditional correlations in Table \[correlations\] show that heating oil, crude oil, and gasoline are closely related, providing high correlations. Heating oil has the strongest relationship with crude oil. Gasoline is related to crude oil more strongly than to heating oil, but all three pairs show positive correlations of around 0.6. The dependence of natural gas on the rest of the group is the weakest, from the point of view of the correlations.
Evidence from the wavelet coherence
-----------------------------------
Unconditional correlations provide evidence of strong dependencies in the energy commodities. However, the time span of our study is quite long, so it may be interesting to see how the correlations develop in time. It may also be of interest to learn whether the dependencies vary across different frequencies, i.e., if there are stronger dependencies in the longer or shorter investment horizons. The wavelet coherence approach will be used as a tool allowing us to study the dependence in time as well as frequency domains.
To assess the statistical significance of the local correlations in the time-frequency space, we use Monte Carlo simulations. Figure \[fig:WTC2\] shows the estimated wavelet coherence and the phase difference for all examined pairs of indices from scale 1 (one day) up to a scale of 256 (approximately one market year). Time is shown on the horizontal axis, while the vertical axis refers to frequency; the lower the frequency, the higher the scale. The wavelet coherence finds the regions in time-frequency space where the two time series co-vary. Regions inside the black lines plotted in warmer colors represent regions with significant dependence. The colder the color is, the less dependent the series are. Cold regions outside the significant areas represent time and frequencies with no dependence in the commodities. Thus, we can clearly see both the frequency and the time intervals where the commodities move together significantly. A continuous wavelet transform at any given point uses the information of neighboring data points, so areas at the beginning and end of the time interval should be interpreted with caution, as discussed in previous sections. This is also the reason we include only scales up to 256.
![Wavelet coherence of heating oil, gasoline, natural gas and crude oil pairs. The horizontal axis shows time, while the vertical axis shows the period in days. The warmer the color of a region, the higher the degree of dependence between the pair.[]{data-label="fig:WTC2"}](Energy_WTC)
From analysis of the wavelet coherence, we can observe very interesting results. A first glance confirms our findings from the previous analysis. The three commodities of heating oil, gasoline, and crude oil show the strongest dependence. Heating oil is strongly related to crude oil over many periods and at many frequencies. Gasoline is also quite strongly related to crude oil. The dependence of heating oil and gasoline turns out to be restricted to only some periods. For example, around 1998, and 2001, these two were very closely related at almost all frequencies, but around the year 2000, there seems to be no dependence at all. This interesting finding may have its origins in the period of recessions with falling prices. The result will be discussed in the following text in greater detail.
We can thus see that the dependence is highly dynamic as it varies in time. Phases, represented by arrows, do not provide any additional value to the analysis, as they point to the right most of the time, meaning that the significant local correlations are positive, but no commodity is leading (affecting the other one). There are short periods with changing phases and where one commodity seems to be leading the other, but these are not consistent so we cannot conclude that there is any directional influence. When we compare these three indices to natural gas, we can see very small significant areas of dependence.
Connection to the standard analysis in time domain
--------------------------------------------------
The wavelet coherence measures co-movement of two time series in the time-frequency space. Its main advantage is the ability to decompose the time-varying co-movement into different investment horizons. Since the wavelet coherence can be interpreted, in some sense, as a measure of local correlation, we would like to connect the obtained results to standard econometric analysis, usually performed in the time domain only. In this section, we connect the time-varying co-movement (estimated with the wavelet coherence) with the DCC GARCH model, which we borrow as an representative of standard methods[^4].
--- --------------- ------------------- ------------------------------- ---------------
Ê Heating oil Gasoline Natural gas
Unconditional 0.568 Ê Ê
DCC 0.638 (0.137) Ê Ê Ê Ê
WTC 0.541 (0.180) Ê Ê Ê Ê
Unconditional 0.147 0.083 Ê
Ê DCC 0.172 (0.0235) 0.101 (9.188$\times10^{-12}$) Ê
Ê WTC 0.268 (0.130) Ê 0.181 (0.123) Ê
Ê Unconditional 0.643 0.612 0.077 Ê
DCC 0.749 (0.141) 0.645 (0.123) 0.109 (0.025)
WTC 0.708 (0.151) 0.626 (0.131) 0.186 (0.129)
--- --------------- ------------------- ------------------------------- ---------------
: Correlation matrix - comparison of unconditional correlation coefficient with mean value of correlations obtained from the DCC GARCH and the wavelet coherence (WTC), with their standard deviations in the parentheses.[]{data-label="correlations"}
To connect the two approaches, we have to reduce the wavelet coherence to the time dimension only. More precisely, we compute the correlation coefficient based on the squared wavelet coherence coefficient for each time $t$ by simply averaging wavelet coherence (those not significant at a 95 % level of significance are taken to be zero) at all scales for the particular time $t$ and taking the square root of this average. This reduces the wavelet coherence analysis to the time-varying correlations which are computed without the use of models.
The squared wavelet coherence coefficients are in absolute value, hence negative correlations cannot be seen directly. While the wavelet coherence is performed with a complex wavelet, we obtain phase positions of the examined time series which might be used for distinguishing between positive and negative correlations. In our particular datasets, there are no negative correlations (anti-phase positions of time series); hence we do not need to compensate for the negative correlation.
Let us see how the time-varying correlations from the wavelet coherence are related to the correlations from the DCC GARCH. Table \[correlations\] provides comparison of both methods to unconditional correlation. More precisely, Table \[correlations\] compares the mean values of the correlation using different methods for the whole sample. For the three most correlated pairs, the estimates of correlation from the wavelet coherence are closer to the unconditional correlations than the estimates from the DCC GARCH. The opposite is true for the other three, least correlated, pairs. In the cases with very low unconditional correlation, DCC GARCH seems to converge to a constant correlation, which is close to an unconditional one. Still, except for the heating oil - crude oil pair, all unconditional correlations are within the standard deviation of arithmetic mean of correlations from the wavelet coherence as well as the DCC GARCH. These results point to the correctness of both approaches[^5].
![Comparison of correlations between the DCC GARCH (dashed line) and time-varying correlations from the wavelet coherence (black line) for heating oil, gasoline, natural gas, and crude oil pairs. The results are smoothed using 50-day moving averages[]{data-label="fig:Comparison"}](DCC_WTC_energy){width="5.5in"}
Figure \[fig:Comparison\] provides comparison of time-varying correlations using both approaches[^6]. Since we use the comparison to DCC GARCH only as an illustration of how wavelet coherence is related to standard methods, we do not concentrate on the differences between approaches; rather, we discuss the findings in the following section.
Time-frequency dynamics of the co-movement
------------------------------------------
The time-varying dynamics of correlations is confirmed using both approaches. But the wavelet coherence provides a rather complex insight into this dynamics. Turning back to Figure \[fig:WTC2\], we can see that co-movement changes not only in time, but is different also at various investment horizons. The study of the pair of heating oil and gasoline is a good example:
In 1998, there was a large increase in the correlation of this pair, up to 0.8, confirmed by both DCC GARCH as well as wavelet coherence methods. Wavelet coherence plotted in Figure \[fig:WTC2\] shows that this strong dependence was present for all investment horizons from the longer ones of several months, up to the shorter, daily ones. In the following two years, the correlation dropped quickly. The wavelet coherence plot again brings the very interesting insight that the only dependence present in this pair around the year 2000 was a short-term one – up to one week. After the year 2000, the dependence again increased rapidly to the 0.8 level during the following year. The latest development of dependence among the three most correlated commodities – heating oil, gasoline, and crude oil – is very similar. Correlations between them increased rapidly to the 0.8 levels at the beginning of 2009.
For these three most correlated commodities, the periods of high coherence around the years 1998 and 2001 are closely related to periods of recession with falling prices. Specifically, they relate to the Asian financial crisis in 1998-2000, the 9/11 terrorist attacks, and the resulting fear in the markets in 2001-2002. Finally, the current financial crisis of the period 2008-2010 has shown a similar behavior. All three commodities again have significant coherences through all investment horizons.
While we can see that the dynamics of correlation is changing rapidly in time, we can also see that the dynamics change in frequency as well. For example, the gasoline - crude oil pair shows a large dependence for investment horizons of several months during the years 1997-2001, while the short-term dependence seems to be very low during this period. The most correlated pair, heating oil - crude oil, shows the dynamics of the dependence in both time and frequency as well. The longer-term (up to six months) investors believe that this dependence is very high, while in the short-term investment horizons, the dependence varies over time much more quickly.
We can compare wavelet coherence with the findings of entropy analysis published recently by [@Martina2011]. The authors show that for crude oil, the periods with economic downturns exhibit a reduced market complexity in terms of reduced entropy levels. We observe similar behavior, because these periods are characterized by falling prices, i.e., movement in one direction. As a consequence, the complexity is reduced, and in the multivariate setting the co-movement is high and significant for a large number of investment horizons.
Concluding remarks
==================
In this paper, we contribute to the literature on co-movement on the energy markets by researching the interconnections between the main components of the energy sector in the time-frequency space. The novelty of our approach lies in studying the co-movement of energy markets in the time-frequency space for the first time and comparing the results to the standard econometric tools for studying relationships between markets.
The main finding of this paper is that some energy pairs show strong dynamics in co-movement in time during various investment horizons. The results suggest that when looking at the dependence of energy markets, one should always keep in mind its time-varying nature and look at it for various investment horizons. While the strongest dependence occurs during the periods of sharp price drops, it seems that the periods of recession creating fear in the markets imply a much higher downside risk to a portfolio based on these commodities. This inefficiency of the energy market is muted after recovery from the recession.
Wavelet coherence also uncovered long cycles (64 to 128 days) in heating oil - crude oil pair that were also present in the periods outside of recession, or more precisely, periods of stable growth. Dependence on higher frequency cycles changes considerably in time. Still, the three commodities, heating oil, gasoline and crude oil strongly co-move, thus for the manager willing to keep a well diversified portfolio, the trio will imply great exposure to risk. On the other hand, natural gas seems to be unrelated to all three commodities for all investment horizons as well as the studied time periods.
In conclusion, we uncover some interesting dynamics of the co-movement between energy markets in time as well as various investment horizons. Our findings are model-free and provide the possibility of new research on financial risk modeling, as they show that dynamic diversification is required in order to preserve a higher profit.
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank Aslak Grinsted for providing us with the MATLAB wavelet coherence package. The authors would like to thank the anonymous referees whose suggestions helped to improve this paper substantially. The support from the Czech Science Foundation under Grants 402/09/0965, 402/09/H045, and 402/10/1610 and from the Department of Education MSMT 0021620841 is gratefully acknowledged.
Appendix A {#Appendix A .unnumbered}
==========
Heating oil Gasoline Natural gas Crude oil
---------- ------------- ------------- ------------- -------------
Mean 0.000368314 0.000400265 0.000211919 0.000412954
St.dev 0.029718 0.0320496 0.049457 0.0277684
Skewness -1.28498 -0.284786 0.178664 -0.242562
Kurtosis 34.5392 6.79143 22.049 8.01979
Min -0.520032 -0.222545 -0.5187 -0.200321
Max 0.244974 0.230015 0.659246 0.181293
: Descriptive statistics of the daily logarithmic returns for heating oil, gasoline, natural gas, and crude oil in the period of November 1, 1993 to July 21, 2010.[]{data-label="stats"}
Coefficient Standard Deviation
---------------- ------------- --------------------
$\alpha_{DCC}$ 0.04435448 0.01306025
$\beta_{DCC}$ 0.90056051 0.03265073
Log.Likelihood 82605.52
: Coefficient estimates of multivariate DCC GARCH for heating oil, gasoline, natural gas, and crude oil in the period of November 1, 1993 to July 21, 2010. $\alpha_{DCC}$ and $\beta_{DCC}$ are coefficients from Equation \[eqQ\][]{data-label="DCCresults"}
[^1]: A function $x(t)$ is called a square integrable if $\int_{-\infty }^{\infty } x(t)^2 dt<\infty$.
[^2]: Without smoothing, the wavelet coherence equals one at all scales. Smoothing is achieved by convolution in both time and scale. The time convolution is performed with a Gaussian window, while the scale convolution is done with a rectangular window – see [@Grinsted2004].
[^3]: The cone of influence is highly dependent on the type of wavelet used – see [@TorenceCompo98].
[^4]: We perform only a simple-estimate comparison: this paper is not meant to provide a detailed discussion of the differences and capabilities of these models.
[^5]: It should be noted that forecasting with the aid of wavelet coherences is difficult in the present setting due to the nature of wavelets.
[^6]: We use 50-day moving averages for illustration.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We show that for any finite group $G$ and for any $d$ there exists a word $w\in F_{d}$ such that a $d$-tuple in $G$ satisfies $w$ if and only if it generates a solvable subgroup. In particular, if $G$ itself is not solvable, then it cannot be obtained as a quotient of the one relator group $F_{d}/\left\langle w\right\rangle $.
As a corollary, the probability that a word is satisfied in a fixed non-solvable group can be made arbitrarily small, answering a question of Alon Amit.
author:
- Miklós Abért
title: On the probability of satisfying a word in a group
---
Introduction
============
Let $F_{n}$ denote the free group on $n$ letters and let $G$ be a group. For $w\in F_{n}$ we say that the $n$-tuple $(g_{1},g_{2},\ldots ,g_{n})\in G^{n}$ *satisfies* $w$ if the substitution $w(g_{1},g_{2},\ldots ,g_{n})=1$. Our first result is the following.
\[nagy\]Let $G$ be a finite group. Then for all $n$ there exists a word $w\in F_{n}$ such that for all $g_{1},g_{2},\ldots ,g_{n}\in G$, the tuple $(g_{1},g_{2},\ldots ,g_{n})$ satisfies $w$ if and only if the subgroup $\left\langle g_{1},g_{2},\ldots ,g_{n}\right\rangle \leq G$ is solvable.
Note that if $G$ itself is not solvable, then a word as in Theorem \[nagy\] has to use at least $n-2$ letters of $F_{n}$. Indeed, if $w$ omits at least two letters, then any two elements of $G$ generates a solvable subgroup, which, using a theorem of Thompson [@thom] (see also [@flav]) implies that $G$ itself is solvable.
For $w\in F_{n}$ let $F_{n}/\left\langle w\right\rangle $ denote the one-relator group defined by $w$. As an immediate corollary of Theorem \[nagy\], we get the following.
Let $G$ be a finite non-solvable group. Then for all $n$ there exists a word $w\in F_{n}$ such that $G$ is not a quotient of $F_{n}/\left\langle
w\right\rangle $.
We suspect that this property holds exactly when $G$ is not solvable.
**Question 1.** *Let* $G$* be a finite solvable group. Does there exist* $k\in \mathbb{N}$* such that for all* $n\geq k$* and every* $w\in F_{n}$* the one-relator group* $F_{n}/\left\langle w\right\rangle $* has a surjective homomorphism to* $G$*?*
For $w\in F_{n}$ let $P(G,w)$ denote the probability that for $n$ independent uniform random elements $g_{1},\ldots ,g_{n}\in G$ we have $w(g_{1},\ldots ,g_{n})=1$. Note that $P(G,w)$ only depends on the word $w$ and not on $n$ so we can assume $w\in F_{\infty }$.
The probabilities $P(G,w)$ have been investigated in the literature mainly for a fixed word and running $G$. The strongest result in this direction is of Dixon, Pyber, Seress and Shalev [@dixetal] who proved that for any fixed word $1\neq w$ the probability $P(G,w)$ tends to $0$ in the size of $G$ assuming that $G$ is non-abelian simple. In this paper we will fix the finite group $G$ and let $w$ run through $F_{\infty }$.
Alon Amit [@amit] has shown that if $G$ is nilpotent then there exists a constant $c>0$ depending on $G$ only such that for all $w\in $ $F_{\infty }$ we have $P(G,w)>c$. Note that this answers Question 1 affirmatively for nilpotent groups. He conjectures that the same holds if $G$ is solvable and that if $G$ is nilpotent then actually $$P(G,w)\geq \frac{1}{\left| G\right| }\text{ for all }w\in F_{\infty }\text{.}$$ He also asked if in turn for a non-solvable finite group $G$ the probability $P(G,w)$ can be made arbitrarily small with a suitable $w\in F_{\infty }$.
It is easy to see that Theorem \[nagy\] already answers Amit’s question affirmatively, but the following stronger result also holds. A group $G$ is *just non-solvable* if every proper quotient of $G$ is solvable, but $G$ itself is not.
\[suru\]Let $G$ be a finite just non-solvable group. Then the set $$\left\{ P(G,w)\mid w\in F_{\infty }\right\}$$ is dense in $[0,1]$.
**Acknowledgement.** The author is grateful to Alon Amit for communicating his results and questions to him and to Laci Pyber for helpful advices on how to present the paper.
Proofs
======
Let us introduce some notation. Let $G$ be a just non-solvable group and let $N$ be a minimal normal subgroup of $G$. Then $N\cong S^{m}$ for some simple group $S$. By the minimality of $G$ the quotient $G/N$ is solvable, so $S$ is non-abelian and $Z(N)=1$, which implies that $G/Z_{G}(N)$ is non-solvable so $Z_{G}(N)=1$. Then $G$ embeds into the wreath product $$\mathrm{Aut}(N)\cong \mathrm{Aut}(S)\,\mathrm{wr}\,\mathrm{Sym}(m)$$ where $\mathrm{Sym}(m)$ denotes the symmetric group on $m$ letters and by the minimality of $N$, $G$ has a transitive image in $\mathrm{Sym}(m)$. Since $G/N$ is solvable, $N$ is a characteristic subgroup of $G$. Also, every nontrivial normal subgroup $K\vartriangleleft G$ contains $N$ (using the minimality of $N$ and that $G$ is just non-solvable). Finally, if in addition $K=K^{\prime }$ (the commutator subgroup of $K$), then $K=N$.
From now on $G$, $N$, $S$ and $m$ will be as above. Let $G_{j}\cong G$ ($1\leq j\leq n$), let $$P=G_{1}\times \cdots \times G_{n}$$ and let $$\pi _{j}:P\rightarrow G_{j}\text{ \ }(1\leq j\leq n)$$ denote the projection to the $j$-th coordinate. Let $$N\cong N_{j}\vartriangleleft G_{j}(1\leq j\leq n)\text{,}$$ let $N_{j}=S_{j,1}\times \cdots \times S_{j,m}$ where $S_{j,i}\cong S$ and let $$M=N_{1}\times \cdots \times N_{n}\vartriangleleft P\text{.}$$
The first lemma is folklore.
\[trivi\]If $S_{1},\ldots ,S_{n}$ are nonabelian finite simple groups, then every normal subgroup $K\vartriangleleft S_{1}\times \cdots \times
S_{n} $ is of the form $$K=K_{1}\times \cdots \times K_{n}$$ where $K_{i}=S_{i}$ or $1$ ($1\leq i\leq n$).
The next lemma tells us about the normal subgroup structure of subgroups of $P$ which project onto each $G_{j}$.
\[subdir\]Let $H\leq P$ be a subgroup containing $M$ such that $$\pi _{j}(H)=G_{j}\text{ \ }(1\leq j\leq n)$$ Let $K$ be a normal subgroup of $H$. Then $$K\cap M=\bigotimes_{\pi _{j}(K)\neq 1}N_{j}$$
$K\cap M$ is normal in $$M\cong \bigotimes_{1\leq j\leq n\text{, }1\leq i\leq m}S_{j,i}$$ so by Lemma \[trivi\] it is the direct product of some of the $S_{j,i}$, that is, $$K\cap M=K_{1}\times \cdots \times K_{n}$$ where $K_{j}\vartriangleleft N_{j}$ ($1\leq j\leq n$).
If $\pi _{j}(K)=1$ then $K\cap N_{j}=1$ so $K_{j}=1$.
If $\pi _{j}(K)\neq 1$ then $\pi _{j}(K)\vartriangleleft \pi _{j}(H)=G_{j}$ so $N_{j}\subseteq \pi _{j}(K)$, since $N_{j}$ is a minimal normal subgroup in $G_{j}$. In this case $$K\cap M\supseteq \lbrack K,M]\supseteq \lbrack K,N_{j}]=[\pi
_{j}(K),N_{j}]=N_{j}$$ so $K_{j}=N_{j}$ (here we use the direct product form and that the commutator $[N_{j},N_{j}]=N_{j}$).
The lemma holds.
Let $$\left( a_{1},\ldots ,a_{k}\right) ,\text{ }\left( b_{1},\ldots ,b_{k}\right)
\in G^{k}$$ be $k$-tuples from $G$. We say that $\left( a_{1},\ldots ,a_{k}\right) $ and $\left( b_{1},\ldots ,b_{k}\right) $ are *automorphism independent over* $G$ if there exists no $\alpha \in \mathrm{Aut}(G)$ such that $a_{i}^{\alpha }=b_{i}$ for all $1\leq i\leq k$.
Our next lemma shows that subgroups of $G_{1}\times \cdots \times G_{n}$ satisfying some natural conditions contain $N_{1}\times \cdots \times N_{n}$.
\[trukk\]Let $a_{i,j}\in G_{j}$ ($1\leq i\leq k$, $1\leq j\leq n$), such that we have $$\left\langle a_{1,j},\ldots ,a_{k,j}\right\rangle =G_{j}\text{ }(1\leq j\leq
n)$$ and that for all $1\leq j<l\leq n$ the $k$-tuples $\left( a_{1,j},\ldots
,a_{k,j}\right) $ and $\left( a_{1,l},\ldots ,a_{k,l}\right) $ are automorphism independent over $G$. For $1\leq i\leq k$ let $$h_{i}=\left( a_{i,1},\ldots ,a_{i,n}\right) \in G_{1}\times \cdots \times
G_{n}$$ and let $$H=\left\langle h_{1},\ldots ,h_{k}\right\rangle \leq G_{1}\times \cdots
\times G_{n}$$ Then $$M=N_{1}\times \cdots \times N_{n}\leq H.$$
Let $$f:G_{1}\times \cdots \times G_{n}\rightarrow G_{1}\times \cdots \times
G_{n-1}$$ denote the projection to the first $n-1$ coordinates. Let $H_{1}=f(H)$ and let $$R=\pi _{n}(\mathrm{Ker}(f))\leq G_{n}\text{.}$$ By induction on $n$, we have $N_{1}\times \cdots \times N_{n-1}\leq H_{1}$. Also $R$ is normal in $G_{n}$ so by the minimality of $N_{n}$ in $G_{n}$ either $N_{n}\leq R$ or $R=1$.
We claim that $N_{n}\leq R$. Assume $R=1$. Let us define the function $\varphi :H_{1}\rightarrow G_{n}$ by $$\varphi (g_{1},\ldots ,g_{n-1})=g_{n}\text{ if }(g_{1},\ldots
,g_{n-1,}g_{n})\in H\text{. }$$ Then $\varphi $ is well-defined, since $$(g_{1},\ldots ,g_{n-1,}g_{n}),(g_{1},\ldots ,g_{n-1,}g_{n}^{\prime })\in H$$ implies $g_{n}^{-1}g_{n}^{\prime }\in R$. So $\varphi $ is a homomorphism. Using $\left\langle a_{1,n},\ldots ,a_{k,n}\right\rangle =G_{n}$ we also see that $\varphi $ is surjective.
Let $K=\mathrm{Ker}(\varphi )$. Then $K$ is normal in $H_{1}$ and $$H_{1}/K\cong G_{n}\cong G$$ which is not solvable. Since $N_{1}\times \cdots \times N_{n-1}\subseteq
H_{1}$, the use of Lemma \[subdir\] for $H_{1}$ and $K$ gives us $$K\cap M=\bigotimes_{\pi _{j}(K)\neq 1}N_{j}\text{. }$$
Now $M\subseteq K$ would imply that $H_{1}/K$ is solvable, a contradiction. So there exists a coordinate $1\leq l<n$ such that $\pi _{l}(K)=1$, that is, $K\subseteq \mathrm{Ker}(\pi _{l})$. Moreover $$H_{1}/\mathrm{Ker}(\pi _{l})\cong G_{l}\cong G$$ which implies $K=\mathrm{Ker}(\pi _{l})$. This shows that the function $\alpha :G_{l}\rightarrow G_{n}$ defined by $$\alpha (g_{l})=g_{n}\text{ if }(g_{1},\ldots ,g_{l},\ldots ,g_{n})\in H$$ is an isomorphism. In particular, $\alpha (a_{i,l})=a_{i,n}$ ($1\leq i\leq k$), so the $k$-tuples $\left( a_{1,l},\ldots ,a_{k,l}\right) $ and $\left(
a_{1,n},\ldots ,a_{k,n}\right) $ are not automorphism independent over $G$ which contradicts the assumptions of the lemma. So the claim $N_{n}\leq R$ holds and so $1\times \cdots \times 1\times N_{n}\leq H$.
Now let $L=f^{-1}(N_{1}\times \cdots \times N_{n-1})\leq H$. Let $L^{(i)}$ denote the $i$-th element of the derived series of $L$ and let $r$ be a number such that $L^{(r)}=L^{(r+1)}$. Then $f(L^{(r)})=N_{1}\times \cdots
\times N_{n-1}$ and since $1\times \cdots \times 1\times N_{n}\leq L$ also $1\times \cdots \times 1\times N_{n}\leq L^{(r)}$. Now $J=\pi _{n}(L^{(r)})$ is normal in $G_{n}$, $N_{n}\leq J$ and $J^{\prime }=J$, so $J=N_{n}$. This implies $$L^{(r)}=N_{1}\times \cdots \times N_{n}\leq H$$ what we wanted to prove.
**Remark.** This lemma is well-known in the subcase when $G$ is a nonabelian finite simple group (see [@canlub], or [@wie]). We will state a light corollary of that which we will use in the proof of Theorem \[nagy\].
\[easy\]Let $G_{i}$ ($1\leq i\leq n$) be finite nonabelian simple groups and let $$H\leq G_{1}\times \cdots \times G_{n}$$ such that the projections $\pi _{i}(H)=G_{i}$ ($1\leq i\leq n$). Then there exists $g\in H$ such that $\pi _{i}(g)\neq 1$ ($1\leq i\leq n$).
We proceed by induction on $n$. For $n=1$ the lemma is trivial. By induction we have an element $g\in H$ such that $\pi _{i}(g)\neq 1$ ($1\leq i<n$). If the last coordinate is automorphism dependent on some previous coordinate $k$ then $\pi _{k}(g)\neq 1$ implies $\pi _{n}(g)\neq 1$. If it is not, then $1\times \cdots \times 1\times G_{n}\leq H$ and we can set the last coordinate of $g$ as we wish.
Let $S$ be a set of finite simple groups. We say that a finite group $G$ is in $\mathrm{Comp}(S)$ if all nonabelian composition factors of $G$ are in $S$. An affirmative answer for the following question would be a far-reaching generalization of Theorem \[nagy\].
**Question 2.** *Let* $S$* be a finite set of finite simple groups. Is it true that for all* $n$* there exists* $w\in F_{n}$* such that every quotient of the one-relator group* $F_{n}/\left\langle w\right\rangle $* which lies in* $\mathrm{Comp}(S) $* is solvable?*
Note that we dont know the answer even in the case when $S$ consists of one simple group.
Now we prove Theorem \[suru\].
Let $m$ be the number of maximal subgroups of $G$. Let $d>\log _{2}m$ be an integer to be chosen later. The probability that $d$ independent random elements all fall into a fixed maximal subgroup $M$ is at most $\left|
G:M\right| ^{-d}\leq 2^{-d}$ so the probability that $d$ random elements do not generate $G$ is at most $m2^{-d}<1$. In particular, $G$ can be generated by $d$ elements. Let $$S=\left\{ (g_{1},\ldots ,g_{d})\in G^{d}\mid g_{1},\ldots ,g_{d}\text{
generate }G\right\}$$ be the set of generating $d$-tuples.
Now $\mathrm{Aut}(G)$ acts on $S$ by $(g_{1},\ldots ,g_{d})^{\alpha
}=(g_{1}^{\alpha },\ldots ,g_{d}^{\alpha })$ where $\alpha \in \mathrm{Aut}(G)$. This action is fixed-point free, as if $\alpha $ fixes all the elements of a generating set then it fixes the whole $G$. Let $r$ be the number of $\mathrm{Aut}(G)$-orbits and let $t_{1},\ldots ,t_{r}\in S$ be an orbit representative system.
It is easy to see that the conditions of Lemma \[trukk\] hold for $a_{i,j}=t_{j}(i)$. This implies that the $r$-tuples $$h_{i}=\left( t_{1}(i),\ldots ,t_{n}(i)\right)$$ generate a group $H$ which contains $N_{1}\times \cdots \times N_{r}$.
Let $1\neq g\in N$, let $k\leq r$ be a natural number to be chosen later and let the $r$-tuple $h$ be defined by $$h(i)=1\text{ (}1\leq i\leq k\text{) and }h(i)=g\text{ (}k<i\leq r\text{)}$$ Then $h\in N_{1}\times \cdots \times N_{r}\subseteq H$, so there exists a word $w\in F_{d}$ such that $w(h_{1},\ldots ,h_{d})=h$.
Now let us evaluate $w$ on the set of possible $d$-tuples from $G$. We completely control the evaluation on generating tuples; since $$w(g_{1}^{\alpha },\ldots ,g_{d}^{\alpha })=(w(g_{1},\ldots ,g_{d}))^{\alpha }\text{ (}\alpha \in \mathrm{Aut}(G)\text{)}$$ we have $$\left| \left\{ (g_{1},\ldots ,g_{d})\in S\mid w(g_{1},\ldots
,g_{d})=1\right\} \right| =k\left| \mathrm{Aut}(G)\right| \text{.}$$
On $d$-tuples $(g_{1},\ldots ,g_{d})$ not generating $G$ we do not control $w(g_{1},\ldots ,g_{d})$. This gives the estimate $$k\frac{\left| \mathrm{Aut}(G)\right| }{\left| G\right| ^{d}}\leq P(G,w)\leq k\frac{\left| \mathrm{Aut}(G)\right| }{\left| G\right| ^{d}}+m2^{-d}$$ and for $k=r$ we get $$k\frac{\left| \mathrm{Aut}(G)\right| }{\left| G\right| ^{d}}=\frac{\left|
S\right| }{\left| G\right| ^{d}}\geq 1-m2^{-d}\text{.}$$
Since $d$ can be chosen to be arbitrarily large, both $\left| \mathrm{Aut}(G)\right| /\left| G\right| ^{d}$ and $m2^{-d}$ get arbitrarily small. Now $k\leq r$ is arbitrary which shows that the set $$\left\{ P(G,w)\mid w\in F_{\infty }\right\}$$ is dense in $[0,1]$.
The answer to Amit’s question follows as an easy corollary of Theorem \[suru\].
Let $G$ be a finite non-solvable group. Then the set $$\left\{ P(G,w)\mid w\in F_{\infty }\right\}$$ accumulates in $0$.
Let $K$ be a normal subgroup in $G$ such that $G/K$ is just non-solvable and let $g_{1},\ldots ,g_{n}$ be independent uniform random elements of $G$. Then $g_{1}K,\ldots ,g_{n}K$ are independent uniform random elements of $G/K$ which yields $$P(G/K,w)=P(w(g_{1},\ldots ,g_{n})\in K)\geq P(w(g_{1},\ldots
,g_{n})=1)=P(G,w)$$ for $w\in F_{\infty }$. Using Theorem \[suru\] we get that for every $\epsilon >0$ we have $w\in F_{\infty }$ such that $$P(G,w)\leq P(G/K,w)<\epsilon$$ and so the corollary holds.
We are ready to prove Theorem \[nagy\].
For each subgroup $H\leq G$ let us choose a homomorphism $$\varphi _{H}:H\rightarrow .$$ as follows. If $H$ is solvable then let $\varphi _{H}=\mathrm{Id}$ be the identity, otherwise let $\varphi _{H}$ be a homomorphism to a just non-solvable quotient of $H$.
Let us enumerate all the $n$-tuples from $G$ as $t_{1},t_{2},\ldots ,t_{k}$ where $k=\left| G\right| ^{n}$. Let $t_{i,j}$ denote the $j$-th element of $t_{i}$ ($1\leq j\leq n$). For $1\leq i\leq k$ let $$H_{i}=\left\langle t_{i,1},t_{i,2},\ldots ,t_{i,n}\right\rangle$$ Let $\varphi _{i}=\varphi _{H_{i}}$ and let $G_{i}=\varphi _{i}(H_{i})$. Let $N_{i}$ be the minimal normal subgroup of $G_{i}$ if $G_{i}$ is just non-solvable, otherwise let $N_{i}=1$. Also let $$u_{i,j}=\varphi _{i}(t_{i,j})\text{ }(1\leq i\leq k,1\leq j\leq n)$$ and let $$p_{j}=(u_{1,j},u_{2,j},\ldots ,u_{k,j})\in G_{1}\times \cdots \times G_{k}\text{ }(1\leq j\leq n)$$ Let $$L=\left\langle p_{1},p_{2},\ldots ,p_{n}\right\rangle \leq G_{1}\times
\cdots \times G_{k}$$ and let $$\pi _{i}:G_{1}\times \cdots \times G_{k}\rightarrow G_{i}\text{ }(1\leq
i\leq k)$$ denote the projection to the $i$-th coordinate. Then $\pi _{i}(L)=G_{i}$ ($1\leq i\leq k$). Let $L^{(i)}$ denote the $i$-th derived subgroup of $L$ and let $r$ be an integer such that $M=L^{(r)}=L^{(r+1)}$. Then $\pi
_{i}(M)\vartriangleleft G_{i}$ and $\pi _{i}(M)^{\prime }=\pi _{i}(M)$ so $\pi _{i}(M)=N_{i}$. Now all the $N_{i}\neq 1$ are isomorphic to some direct power of a nonabelian simple group so $M$ lies in a direct product of nonabelian simple groups and projects to each factor of the product. By Corollary \[easy\] there exists an element $g\in M\leq L$ such that $\pi
_{i}(g)\neq 1$ if and only if $N_{i}\neq 1$. Let $w\in F_{n}$ be a word such that $w(p_{1},p_{2},\ldots ,p_{n})=g$.
We claim that this $w$ will be good for our purposes. Indeed, we have $$\pi _{i}(g)=w(u_{i,1},\ldots ,u_{i,n})=w(\varphi _{i}(t_{i,1}),\ldots
,\varphi _{i}(t_{i,n}))=\varphi _{i}(w(t_{i,1},\ldots ,t_{i,n}))$$ Now if $H_{i}$ is solvable then $\varphi _{i}$ is the identity map and $\pi
_{i}(g)=1$, so we get $w(t_{i,1},\ldots ,t_{i,n})=1$. If $H_{i}$ is not solvable, then $\pi _{i}(g)\neq 1$ and since $\varphi _{i}$ is a homomorphism we have $w(t_{i,1},\ldots ,t_{i,n})\neq 1$. The theorem holds.
**Remark on Question 1.** Let $G$ be a finite group for which there exists a constant $c>0$ such that for all $w\in $ $F_{\infty }$ we have $P(G,w)>c$. Then as we saw for large enough $d$ most of the $d$-tuples generate $G$ and so for every word $w\in F_{d}$ there exists a generating set $\left\langle g_{1},\ldots ,g_{d}\right\rangle =G$ such that $w(g_{1},\ldots ,g_{d})=1$, that is, $G$ is a quotient of the one-relator group $F_{d}/\left\langle w\right\rangle $. In particular, Amit’s result [@amit] implies an affirmative answer for Question 1 for finite nilpotent groups.
[DPSS]{} A. Amit, On equations in nilpotent groups, in preparation
P. Flavell, Finite groups in which every two elements generate a soluble subgroup, Invent. Math. 121 (1995), no. 2, 279–285.
W.M. Kantor and A. Lubotzky, The probability of generating a finite classical group, Geom. Dedicata 36 (1990), no. 1, 67–87.
J.D. Dixon, L. Pyber, Á. Seress and A. Shalev, Residual properties of free groups and probabilistic methods, J. Reine Angew. Math. 556 (2003), 159–172.
J. G. Thompson, Nonsolvable finite groups all of whose local subgroups are solvable, Bull. Amer. Math. Soc. 74, 1968, 383–437.
J. Wiegold, Growth sequences of finite groups III., J. Austral. Math. Soc. Ser. A 25 (1978), no. 2, 142–144.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
This paper contains a rich collection of results related to weight structures and $t$-structures. For any weight structure $w$ we study [*pure*]{} (co)homological functors; these “ignore all weights except weight zero” and have already found several applications (in particular, to Picard groups of triangulated categories). Moreover, we study [*virtual $t$-truncations*]{} of cohomological functors. The resulting functors are defined in terms of $w$ but are closely related to $t$-structures; so we prove in several cases that a weight structure $w$ “gives” a $t$-structure (that is [*adjacent*]{} or [*$\Phi$-orthogonal*]{} to it).
We also study in detail [*well generated*]{} weight structures (and prove that any [ *perfect*]{} set of objects generates a certain weight structure). We prove the existence of weight structures right adjacent to [*compactly generated*]{} $t$-structures (using Brown-Comenetz duality); this implies that the hearts of the latter have injective cogenerators and satisfy the AB3\* axiom. Actually, “most of” these hearts are Grothendieck abelian (due to the existence of “bicontinuously orthogonal” weight structures).
It is convenient for us to use the notion of [*torsion pairs*]{}; these essentially generalize both weight structures and $t$-structures. We prove several new properties of torsion pairs; in particular, we generalize a theorem of D. Pospisil and J. Šťovíček to obtain a complete classification of compactly generated torsion pairs.
author:
- 'Mikhail V. Bondarko [^1]'
title: 'On torsion pairs, (well generated) weight structures, adjacent $t$-structures, and related (co)homological functors'
---
[***2010 Mathematics Subject Classification.*** Primary 18E30 18E40 18G25; Secondary 14F05 55P42.]{}
[***Key words and phrases*** Triangulated category, torsion pair, weight structure, $t$-structure, adjacent structure, cohomological functor, pure functor, virtual $t$-truncation, compact object, perfect class, symmetric classes, Brown-Comenetz duality, well generated category, derived category of coherent sheaves, duality, pro-objects, projective class.]{}
Introduction {#introduction .unnumbered}
============
The main goal of the current paper is to demonstrate the utility of weight structures to the construction and study of $t$-structures and of (co)homological functors from triangulated categories (into abelian ones). In particular, for a weight structure $w$ we study [*$w$-pure functors*]{} (i.e., those that “only see $w$-weight zero”).[^2] Functors of this type have already found interesting applications in several papers (note in particular that the results of our §\[sdetect\] are important for the study of Picard groups of triangulated categories in [@bontabu]). So the author believes that the reader not interested in the construction weight structures and $t$-structures (that we will start discussing very soon) may still benefit from §\[sws\] of the paper where a rich collection of properties of pure functors and [*virtual $t$-truncations*]{} of (co)homological functors (with respect to $w$) is proved.
Now, virtual $t$-truncations are defined in terms of weight structures; still they are closely related to $t$-structures (whence the name). Respectively, our results yield the existence of some vast new families of $t$-structures. To describe one of the main results of this sort here we recall that a $t$-structure $t=({\underline{C}}^{t\le 0}, {\underline{C}}^{t\ge 0})$ (for a triangulated category ${\underline{C}}$) is said to be (right) adjacent to $w$ if ${\underline{C}}^{t\le 0}={\underline{C}}_{w\ge 0}$.[^3] For a triangulated category ${\underline{C}}$ that is closed with respect to (small) coproducts and a weight structure $w$ on it we will say that $w$ is [*smashing*]{} whenever ${\underline{C}}_{w\ge 0}$ is closed with respect to ${\underline{C}}$-coproducts (note that ${\underline{C}}_{w\le 0}$ is $\coprod$-closed automatically).
\[tadjti\] Let ${\underline{C}}$ be a triangulated category that is closed with respect to coproducts and satisfies the following Brown representability property: any functor ${\underline{C}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ that respects (${\underline{C}}{{^{op}}}$)-coproducts is representable.[^4]
Then for a weight structure $w$ on ${\underline{C}}$ there exists a $t$-structure right adjacent to it if and only if $w$ is smashing. Moreover, the heart of $t$ (if $t$ exists) is equivalent to the category of all those additive functors ${{\underline{Hw}}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ that respect products.[^5]
Note here that triangulated categories closed with respect to coproducts have recently become very popular in homological algebra and found applications in various areas of mathematics; so a significant part of this paper is dedicated to categories of this sort. Still we prove certain alternative versions of Theorem \[tadjti\]; some of them can be applied to “quite small” triangulated categories.
So, if instead of the Brown representability condition for ${\underline{C}}$ we demand it to satisfy the [*$R$-saturatedness one*]{} (see Definition \[dsatur\] below; this is an “$R$-linear finite” version of the Brown representability) then for any [*bounded*]{} $w$ on ${\underline{C}}$ there will exist a $t$-structure right adjacent to it. Note that this version of the result can be applied to the bounded derived category $D^b(X)$ of coherent sheaves on regular separated finite-dimensional scheme that is proper over the spectrum of a Noetherian ring $R$ (see Proposition \[psatur\](II) and Remark \[rsatur\](1)). We also prove two generalizations of this existence result (see Proposition \[psaturdu\] and Remark \[roq\]); they “produce” certain $t$-structures from weight structures on the bounded derived category of coherent sheaves on any proper $R$-scheme $X$ and also on the triangulated category of perfect complexes on $X$.[^6] So, one may (roughly) say that any “reasonable” weight structure (on a triangulated category satisfying some Brown representability condition) can be used to construct certain $t$-structures. This demonstrates the importance of constructing weight structures.
However, the $t$-structures constructed using Theorem \[tadjti\] appear to be somewhat “exotic” (yet cf. Theorem 1 of [@zvon]) and possibly the $t$-structures constructed via the aforementioned “$R$-saturated” versions of the theorem are “more useful”. Still we also prove (using adjacent and [*$\Phi$-orthogonal*]{} weight structures in a crucial way) several properties of [compactly generated]{} $t$-structures.[^7]
To formulate the following theorem we need some definitions.
For a class of objects $S$ of a triangulated category ${\underline{C}}$ we will write $S{{}^{\perp}}$ (resp. ${{}^{\perp}}S$) for the class of those $M\in {\operatorname{Obj}}{\underline{C}}$ such that the morphism group ${\underline{C}}(N,M)$ (resp. ${\underline{C}}(M,N)$) is zero for all $N\in S$.
We will say that a $t$-structure $t=({\underline{C}}^{t\le 0},{\underline{C}}^{t\ge 0})$ on ${\underline{C}}$ is [*generated*]{} by a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ whenever ${\underline{C}}^{t\ge 0}=\cap_{i\ge 1}({\mathcal{P}}{{}^{\perp}}[i])$.
\[tgroth\] Let ${\underline{C}}$ be a triangulated category that is closed with respect to coproducts; let $t$ be a $t$-structure on it generated by a set of compact objects (we will say that $t$ is compactly generated if such a generating set ${\mathcal{P}}$ exists).[^8] 1. Then the heart ${{\underline{Ht}}}$ of $t$ has an injective cogenerator and satisfies the AB3\* axiom.
2\. Assume in addition either that ${\underline{C}}$ is the homotopy category of a proper simplicial stable model category or that $t$ is non-degenerate. Then ${{\underline{Ht}}}$ is a Grothendieck abelian category.
Past 2 of this theorem answers Question 3.8 of [@parrasao] “in all reasonable cases”;[^9] part 1 appears to be new also.
Now we describe a result that was important for the proof of part 1 of Theorem \[tgroth\].
One of the main topics of this paper is the study of the relation of (smashing) weight structures to perfect sets of objects (that we will now define; recall that these are also closely related to the Brown representability condition).[^10]
\[textw\] Assume that ${\underline{C}}$ is closed with respect to small coproducts. Let ${\mathcal{P}}$ be a [*perfect*]{} set of objects of ${\underline{C}}$ (i.e., assume that the class [*${\mathcal{P}}$-null*]{} of those morphisms that are annihilated by corepresentable functors of the type ${\underline{C}}(P,-)$ for $P\in {\mathcal{P}}$ is closed with respect to coproducts).[^11] Then the couple $w=(L,R)$ is a smashing weight structure, where $R=\cap_{i<0}({\mathcal{P}}{{}^{\perp}}[i])$ and $L=({{}^{\perp}}R)[1]$. Moreover, the class $L$ may be described “more explicitly” in terms of ${\mathcal{P}}$; cf. Corollary \[cwftw\] below.
Thus if we assume in addition that the Brown representability condition is fulfilled for ${\underline{C}}$[^12]then we also obtain the (right) adjacent $t$-structure $t=(R, (R{{}^{\perp}})[1])$. This $t$-structure is [*cogenerated*]{} by any class ${\mathcal{P}}'$ that is [*weakly symmetric*]{} to ${\mathcal{P}}$ (see Definition \[dsym\](\[iwsym\]) below; if the elements of ${\mathcal{P}}$ are compact then we can construct ${\mathcal{P}}'$ using [*Brown-Comenetz duality*]{}). Somewhat surprisingly to the author, a similar chain of arguments gives the existence of a weight structure $w$ that is [*right adjacent*]{} to a given compactly generated $t$-structure (i.e., ${\underline{C}}^{t\ge 0}={\underline{C}}_{w\le 0}$); this yields the proof of Theorem \[tgroth\](1).[^13] Moreover, the “opposite” weight structure $w{{^{op}}}$ in the category ${\underline{C}}{{^{op}}}$ is perfectly generated but not compactly generated (so, there exist plenty of perfectly generated weight structures that are not compactly generated; recall that the latter class of weight structures was introduced in [@paucomp]). We also prove the following “well generatedness” result for weight structures (obtaining in particular that all smashing weight structures on well generated categories can be obtained from Theorem \[textw\]).
\[pwgwstr\] Assume that ${\underline{C}}$ is a well generated triangulated category (i.e., there exists a [*regular*]{} cardinal ${\alpha}$ and a perfect set $S$ of [*${\alpha}$-small*]{} objects such that $S{{}^{\perp}}={\{0\}}$; see Definition \[dwg\](\[idpg\])). Then for any smashing weight structure $w$ on ${\underline{C}}$ there exists a cardinal ${\alpha}'$ such that for any regular ${\beta}\ge {\alpha}'$ the weight structure $w$ is strongly ${\beta}$-well generated in the following sense: the couple $({\underline{C}}_{w\le 0}\cap {\operatorname{Obj}}{\underline{C}}^{\beta}, {\underline{C}}_{w\ge 0}\cap {\operatorname{Obj}}{\underline{C}}^{\beta})$ is a weight structure on the triangulated subcategory ${\underline{C}}^{\beta}$ of ${\underline{C}}$ consisting of ${\beta}$-compact objects (see Definition \[dbecomp\](\[idcomp\])), the class ${\mathcal{P}}={\underline{C}}_{w\le 0}\cap {\operatorname{Obj}}{\underline{C}}^{\beta}$ is essentially small and perfect, and $w=(L,R)$ for $L$ and $R$ being the classes described in Theorem \[textw\] (if these conditions are fulfilled for $w$ and some ${\mathcal{P}}$ then we also say that $w$ is perfectly generated by ${\mathcal{P}}$).
A significant part of the “easier” results of the current paper is stated in terms of [*torsion pairs*]{} (as defined in [@aiya]; cf. Remark \[rgen\](4) below); these essentially generalize both weight structures and $t$-structures.[^14] This certainly makes the corresponding results more general; note also that (the main subject of) [@bkw] yields an interesting family of examples of torsion pairs that do not come either from weight structures or from $t$-structures. Probably the most interesting result about general torsion pairs proved in this paper is the classification of compactly generated ones (in Theorem \[tclass\]); we drop the assumption that ${\underline{C}}$ is a “stable derivator” category that was necessary for the proof of the closely related Corollary 3.8 of [@postov]. We also relate [*adjacent*]{} torsion pairs to “Brown-Comenetz-type symmetry”.
We also study [*dualities*]{} between triangulated categories and their relation to torsion pairs. We demonstrate that for a $t$-structure $t$ on ${\underline{C}'}$ that does not possess a left adjacent weight structure there still may exist $w$ on a category ${\underline{C}}$ that is [$\Phi$-orthogonal ]{} to $t$, where $\Phi:{\underline{C}}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$ is a (“bicontinuous”) duality of triangulated categories. Moreover, for an interesting sort of dualities that we study in §\[sdual2\] and for any compactly generated $t$ on ${\underline{C}'}$ the objects of the heart of the corresponding orthogonal $w$ give faithful exact “stalk” functors ${{\underline{Ht}}}\to {\underline{\operatorname{Ab}}}$ that respect coproducts; taking the functor $\Phi(P,-)$ for $P$ being a [*cogenerator*]{} of ${{\underline{Hw}}}$ we easily obtain that ${{\underline{Ht}}}$ is an AB5 abelian category (cf. Theorem \[tgroth\]). So we demonstrate once again that weight structures “shed some light on $t$-structures”.
Moreover, dualities are important for [@bgn] where they are applied to the study of coniveau spectral sequences and the homotopy $t$-structures on various motivic stable homotopy categories. This matter is also related to the study of the stable homotopy category in [@prospect]. Let us now describe the contents of the paper. Some more information of this sort may be found in the beginnings of sections.
In §\[sold\] we we define torsion pairs and prove several of their properties; these are mostly simple but new. We also relate torsion pairs to $t$-structures and study $t$-projective objects (essentially following [@zvon]); they are related to adjacent weight structures that we will study later.
We start §\[sws\] from recalling some basics on weight structures (among those are some properties of weight complexes; though these are not really new, we treat this subject more accurately than in [@bws] where weight complexes were originally defined). Next we introduce pure functors (in §\[sdetect\]; their “construction” in Proposition \[ppure\] appears to be a very useful statement). An “intrinsic” definition of pure functors is given by Proposition \[pwrange\]. We relate the weight range of functors to virtual $t$-truncations. We also study the properties of virtual $t$-truncations and weight complexes under the assumption that $w$ is smashing. As an application, we treat the representability for virtual $t$-truncations of representable functors; this result has important applications to the construction of (adjacent) $t$-structures below.
In §\[sadjbrown\] we investigate the question when weight structures and $t$-structures admit (right or left) adjacent $t$-structures or weight structures (respectively); we also study adjacent “structures” if they exist. To prove the existence of these adjacent structures on a triangulated category ${\underline{C}}$ we usually assume a certain Brown representability-type condition for ${\underline{C}}$. We also recall the definition of perfect and symmetric classes and relate them to Brown-Comenetz duality and torsion pairs. This gives a funny (general) criterion for the existence of an adjacent torsion pair in terms of “symmetry” along with a “new” description of a $t$-structure that is right adjacent to a given compactly generated weight structure. In §\[spgtp\] we study compactly generated torsion pairs and perfectly generated weight structures; we prove several new and interesting results about them (and some of these statements were formulated above). In particular, we prove that for any “symmetric” ${\mathcal{P}},{\mathcal{P}}'\subset {\operatorname{Obj}}{\underline{C}}$ there exist adjacent $t$ and $w$ (co)generated by them. This implies the existence of a weight structure right adjacent to a given compactly generated $t$; hence the category ${{\underline{Ht}}}$ has an injective cogenerator and satisfies the AB3\* axiom. Combining this fact with the results of [@humavit] we deduce that ${{\underline{Ht}}}$ is Grothendieck abelian whenever $t$ is non-degenerate. The results and arguments of the section are closely related to the properties of localizing subcategories of triangulated categories as studied by A. Neeman, H. Krause and other authors; see Remark \[rtst2\](\[it6\]) below for an “explanation” of this similarity. In §\[skan\] we study dualities between triangulated categories and torsion pairs orthogonal with respect to them. Considering a “very simple” duality we prove that for certain weight structures inside the bounded derived category of coherent sheaves on a scheme $X$ that is proper over ${{\operatorname{Spec}\,}}R$ (where $R$ is a noetherian ring) there necessarily exist (right or left) orthogonal $t$-structures.
Our main tools for constructing “more complicated” dualities are Kan extensions of (co)homological functors from a triangulated subcategory ${\underline{C}_0}$ to ${\underline{C}}$; their properties are rather interesting for themselves. We describe a duality between the homotopy category of filtered pro-objects for a stable proper Quillen model category ${\mathcal{M}}$ with ${\operatorname{Ho}}({\mathcal{M}})$. The properties of this duality imply that for any compactly generated $t$-structure on ${\operatorname{Ho}}({\mathcal{M}})$ its heart is a Grothendieck abelian category; they are also applied in [@bgn] to the study of [*motivic pro-spectra* ]{} and [*generalized coniveau spectral sequences*]{}.
For the convenience of the reader we also make a list of the main definitions and notation used in this paper. Regular cardinals, Karoubi-closures and related matters, $D\perp E$, ${{}^{\perp}}D$, and $D{{}^{\perp}}$, suspended, cosuspended, extension-closed, and strict classes of objects, extension-closures, envelopes, subcategories generated by classes of objects of triangulated categories, homological and cohomological functors, compact objects, localizing and colocalizing subcategories, cogenerators, Hom-generators, compactly generated categories, cc, cp, and pp functors, and the Brown representability condition (along with its dual) are defined in §\[snotata\]; (smashing, cosmashing, countably smashing, adjacent, and compactly generated) torsion pairs (often denoted by $s=({\mathcal{LO}},{\mathcal{RO}})$; sometimes we also use the notation $(L_sM,R_sM)$), generators for them, and ${\mathcal{P}}$-null morphisms are introduced in §\[shop\]; $t$-structures ($t=({\underline{C}}^{t\le 0},{\underline{C}}^{t\ge 0})$) and several of their “types” (including smashing and cosmashing $t$-structures), their hearts (${{\underline{Ht}}}$), associated torsion pairs, $t$-homology ($H_0^t$), and $t$-projective objects are defined in §\[sts\]; weight structures ($w=({\underline{C}}_{w\le 0},{\underline{C}}_{w\ge 0})$; we also define ${\underline{C}}_{w=i}$ and ${\underline{C}}_{[m,n]}$), their “types”, hearts (${{\underline{Hw}}}$), and associated (weighty) torsion pairs, adjacent weight and $t$-structures, $m$-weight decompositions, negative subcategories, and the weight structure ${w^{st}}$ are introduced in §\[ssws\]; (weight) Postnikov towers along with the corresponding filtrations and (weight) complexes, weakly homotopic morphisms of complexes, and the category ${K_{\mathfrak{w}}}({{\underline{Hw}}})$ are defined in §\[sswc\]; pure homological functors ($H^{{\mathcal{A}}}$) and purely $R$-representable homology (with values in ${\operatorname{PShv}^R}({\underline{B}})$) is introduced in §\[sdetect\], virtual $t$-truncations ($\tau^{\ge m}H$ and $\tau^{\le m}H$), weight range of functors, and pure cohomological functors ($H_{{\mathcal{A}}}$) are defined in §\[svtt\]; functors of $R$-finite type and $R$-saturated categories are introduced in §\[sadjt\]; ${\alpha}$-small objects, countably perfect and perfect classes of objects, perfectly generated and well generated categories, weakly symmetric and symmetric classes, and Brown-Comenetz duals of functors and objects are defined in §\[scomp\]; countable homotopy colimits $\operatorname{\varinjlim}Y_i$, strongly extension-closed classes and strong extension-closures, (naive) big hulls, and zero classes of (collections of) functors are introduced in §\[scoulim\]; perfectly generated weight structures, ${\mathcal{P}}$-approximations, and contravariantly finite classes are defined in §\[sperfws\]; weakly and strongly ${\beta}$-well generated weight structures (and torsion pairs) are studied in §\[swgws\]; coextended functors and coextensions are defined in §\[scoext\]; (nice) dualities and orthogonal structures are introduced in §\[sdual1\]; biextensions are defined in §\[sdual1\]; stalk functors are introduced in §\[sgengroth\]; certain model structures on pro-objects are recalled in §\[sprospectra\].
The author is deeply grateful to prof. F. Déglise, prof. G.C. Modoi, prof. Salorio M.J. Souto, and prof. J. Šťovíček for their very useful comments.
On torsion pairs and $t$-structures (“simple properties”) {#sold}
=========================================================
This section is dedicated to the basics on torsion pairs and $t$-structures. In §\[snotata\] we introduce some notation and recall several important properties of triangulated categories (mostly from [@neebook]).
In §\[shop\] we define and study torsion pairs (in the terminology of [@aiya]). Our results are rather easy; yet the author does not know any references for most of them. In §\[sts\] we recall some basics on $t$-structures and relate them to torsion pairs. We also study the notions of $t$-projective objects for the purpose of using them in §\[sadjw\] and later.
Some categorical preliminaries {#snotata}
------------------------------
When we will write $i\ge c$ or $i\le c$ (for some $c\in {{\mathbb{Z}}}$) we will mean that $i$ is an integer satisfying this inequality.
A cardinal ${\alpha}$ is said to be [*regular*]{} if it cannot be presented as a sum of less then ${\alpha}$ cardinals that are less than ${\alpha}$.
Most of the categories of this paper will be locally small. When considering a category that is not locally small we will usually say that it is (possibly) big; we will not need much of these categories. For categories $C,D$ we write $D\subset C$ if $D$ is a full subcategory of $C$.
Given a category $C$ and $X,Y\in{\operatorname{Obj}}C$ we will write $C(X,Y)$ for the set of morphisms from $X$ to $Y$ in $C$. We will say that $X$ is a [*retract*]{} of $Y$ if ${\operatorname{id}}_X$ can be factored through $Y$.[^15]
For a category $C$ the symbol $C^{op}$ will denote its opposite category.
For a subcategory $D\subset C$ we will say that $D$ is [*Karoubi-closed*]{} in $C$ if it contains all retracts of its objects in $C$. We call the smallest Karoubi-closed subcategory $\operatorname{\operatorname{Kar}}_C(D)$ of $C$ containing $D$ the [*Karoubi-closure*]{} of $D$ in $C$. The [*Karoubi envelope*]{} $\operatorname{\operatorname{Kar}}({\underline{B}})$ (no lower index) of an additive category ${\underline{B}}$ is the category of “formal images” of idempotents in ${\underline{B}}$ (so ${\underline{B}}$ is embedded into an idempotent complete category). ${\underline{\operatorname{Ab}}}$ is the category of abelian groups.
${\underline{C}}$, ${\underline{C}}'$, ${\underline{C}_0}$, ${\underline{D}}$, ${{\underline{D}}_0}$, and ${\underline{E}}$ will always denote certain triangulated categories. ${\underline{C}}$ will often be endowed with a weight structure $w$; we always assume that this is the case in those formulations where $w$ is mentioned without any explanations. For $f\in{\underline{C}}(X,Y)$, where $X,Y\in{\operatorname{Obj}}{\underline{C}}$, we call the third vertex of (any) distinguished triangle $X\stackrel{f}{\to}Y\to Z$ a cone of $f$.
We will often consider some representable and corepresentable functors and their restrictions. So for ${\underline{D}}'$ being a full triangulated subcategory of a triangulated category ${\underline{D}}$ and $M\in {\operatorname{Obj}}{\underline{D}}$ we will often write $H^M: {\underline{D}}'\to {\underline{\operatorname{Ab}}}$ for the restriction of the corepresentable (homological) functor ${\underline{D}}(M,-)$ to ${\underline{D}}'$ (yet $H^P$ in Proposition \[porthop\] will denote a certain [*coextension*]{}); $H_M: {\underline{D}}'{{^{op}}}\to {\underline{\operatorname{Ab}}}$ is the restriction of the functor ${\underline{D}}(-,M)$ to ${\underline{D}}'$. We will often be interested in the case ${\underline{D}}'={\underline{D}}$ in this notation; we assume that the domain of the functors $H^M$ and $H_M$ is the category ${\underline{C}}$ if not specified otherwise.
For $X,Y\in {\operatorname{Obj}}{\underline{C}}$ we will write $X\perp Y$ if ${\underline{C}}(X,Y)={\{0\}}$. For $D,E\subset {\operatorname{Obj}}{\underline{C}}$ we will write $D\perp E$ if $X\perp Y$ for all $X\in D,\ Y\in E$. For $D\subset{\operatorname{Obj}}{\underline{C}}$ we will write $D^\perp$ for the class $$\{Y\in {\operatorname{Obj}}{\underline{C}}:\ X\perp Y\ \forall X\in D\};$$ sometimes we will write $\perp_{{\underline{C}}}$ instead to indicate the category that we are considering. Dually, ${}^\perp{}D$ is the class $\{Y\in {\operatorname{Obj}}{\underline{C}}:\ Y\perp X\ \forall X\in D\}$.
In this paper all complexes will be cohomological, i.e., the degree of all differentials is $+1$; respectively, we use cohomological notation for their terms.
We will use the term [*exact functor*]{} for a functor of triangulated categories (i.e., for a functor that preserves the structures of triangulated categories).
We will say that a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ is [*suspended*]{} if ${\mathcal{P}}[1]\subset {\mathcal{P}}$; ${\mathcal{P}}$ is [*cosuspended*]{} if ${\mathcal{P}}[-1]\subset {\mathcal{P}}$.
A class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ is said to be [*extension-closed*]{} if $0\in {\mathcal{P}}$ and for any distinguished triangle $A\to B\to C$ in ${\underline{C}}$ we have the implication $A,C\in
{\mathcal{P}}\implies B\in {\mathcal{P}}$. In particular, an extension-closed ${\mathcal{P}}$ is [*strict*]{} (i.e., contains all objects of ${\underline{C}}$ isomorphic to its elements). The smallest extension-closed class ${\mathcal{P}}'$ containing a given ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ will be called the [*extension-closure*]{} of ${\mathcal{P}}$. The smallest extension-closed Karoubi-closed ${\mathcal{P}}'\subset {\operatorname{Obj}}{\underline{C}}$ containing ${\mathcal{P}}$ will be called the [*envelope*]{} of ${\mathcal{P}}$.
We call the smallest Karoubi-closed triangulated subcategory ${\underline{D}}$ of ${\underline{C}}$ such that ${\operatorname{Obj}}{\underline{D}}$ contains ${\mathcal{P}}$ the [*triangulated subcategory densely generated by*]{} ${\mathcal{P}}$; we will write ${\underline{D}}={\langle}{\mathcal{P}}{\rangle}_{{\underline{C}}}$.
We will say that ${\underline{C}}$ [*has coproducts*]{} (resp. products, resp. countable coproducts) whenever it contains arbitrary small coproducts (resp. products, resp. countable coproducts) of families of its objects.
${\underline{A}}$ will usually denote some abelian category; the case ${\underline{A}}={\underline{\operatorname{Ab}}}$ is the most important one for the purposes of this paper.
We will call a covariant additive functor ${\underline{C}}\to {\underline{A}}$ for an abelian ${\underline{A}}$ [*homological*]{} if it converts distinguished triangles into long exact sequences; homological functors ${\underline{C}}^{op}\to {\underline{A}}$ will be called [*cohomological*]{} when considered as contravariant functors from ${\underline{C}}$ into $ {\underline{A}}$.
For additive categories $C,D$ the symbol $\operatorname{\operatorname{AddFun}}(C,D)$ will denote the (possibly, big) category of additive functors from $C$ to $D$. Certainly, if $D$ is abelian then an $\operatorname{\operatorname{AddFun}}(C,D)$-complex $X\to Y\to Z$ is exact in $Y$ whenever $X(P)\to Y(P)\to Y(P)$ is exact (in $D$) for any $P\in {\operatorname{Obj}}C$.
Below we will sometimes need some properties of the Bousfield localization setting (cf. §9.1 of [@neebook]; most of these statements are contained in Propositions 1.5 and 1.6 of [@bondkaprserr]).
\[pbouloc\]
Let $F:{\underline{E}}\to {\underline{C}}$ be a full embedding of triangulated categories; assume that ${\underline{E}}$ is Karoubi-closed in ${\underline{C}}$. Denote by ${\underline{D}}$ the full subcategory of ${\underline{C}}$ whose object class is ${\operatorname{Obj}}{\underline{E}}{{}^{\perp}}$; denote the embedding ${\underline{D}}\to {\underline{C}}$ by $i$.
Then the following statements are valid.
I. A (left or right) adjoint to an exact functor is exact.
II\. ${\underline{D}}$ is a Karoubi-closed triangulated subcategory of ${\underline{C}}$.
III\. Assume that $F$ possesses a right adjoint $G$.
1. \[ibou1\] Then for any $N\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $$\label{ebou}
N'\stackrel{f}{\to} N\stackrel{g}{\to} N''\to N'[1]$$ with $N'\in {\operatorname{Obj}}{\underline{E}}$ and $ N'' \in{\operatorname{Obj}}{\underline{D}}$ (here we consider ${\underline{E}}$ as a subcategory of ${\underline{C}}$ via $F$), and the triangle (\[ebou\]) is unique up to a canonical isomorphism.
2. \[ibouort\] ${{}^{\perp}}{\operatorname{Obj}}{\underline{D}}={\operatorname{Obj}}{\underline{E}}$.
3. \[ibou2\] The functor $i$ possesses an (exact) left adjoint $A$ and the morphism $g$ in (\[ebou\]) is given by the unit of this adjunction. Moreover, this unit transformation yields an equivalence of the Verdier localization of ${\underline{C}}$ by ${\underline{E}}$ (that is locally small in this case) to ${\underline{D}}$.
4. \[ibou3\]The morphism $f$ in (\[ebou\]) is given by the counit of the adjunction $F\dashv G$, and this counit gives an equivalence ${\underline{E}}\cong {\underline{C}}/{\underline{D}}$.
IV\. A full embedding $F$ as above possesses a right adjoint if and only if for any $N\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle (\[ebou\]).
I. This is Lemma 5.3.6 of [@neebook]. II. Obvious.
III.\[ibou1\]. The existence of (\[ebou\]) is immediate from Proposition 9.1.18 (that says that the Verdier localization ${\underline{C}}\to {\underline{C}}/{\underline{E}}$ is a [*Bousfield localization*]{} functor in the sense of Definition 9.1.1 of ibid.) and Proposition 9.1.8 of ibid. (cf. also Proposition 1.5 of [@bondkaprserr]). Lastly, the essential uniqueness of (\[ebou\]) easily follows from [@bbd Proposition 1.1.9] (cf. also Proposition \[pbw\](\[icompl\]) below).
\[ibouort\]. According to the aforementioned Proposition 9.1.18 of [@neebook], we can deduce the assertion from Corollary 9.1.14 of ibid. \[ibou2\]. The “calculation” of $g$ is given by Proposition 9.1.8 of ibid. also. It remains to apply Theorem 9.1.16 of ibid.
\[ibou3\]. Corollary 9.1.14 of ibid. allows to deduce the assertion from the previous one. IV. The “only if” part of the assertion is given by assertion III.\[ibou1\]. The converse implication is given by Proposition 9.1.18 of ibid.
Now we recall some terminology, notation, and statements related to infinite coproducts and products in triangulated categories. Some of these definitions and results may will be generalized in §\[scomp\] below.
All the coproducts and products in this paper will be small. We will say that a subclass ${\mathcal{P}}$ of ${\operatorname{Obj}}{\underline{C}}$ is [*coproductive*]{} (resp. [*productive*]{}) if it is closed with respect to all (small) coproducts (resp. products) that exist in ${\underline{C}}$. For triangulated categories closed with respect to (all) coproducts or products we will just say that these categories have coproducts (resp. products).[^16]
We recall the following very useful statement.
\[pcoprtriang\] Assume that ${\underline{C}}$ has coproducts (resp. products, resp. countable coproducts). Then ${\underline{C}}$ is Karoubian and all (small) coproducts (resp. products, resp. countable coproducts) of distinguished triangles in ${\underline{C}}$ are distinguished.
The first of the assertions is given by Proposition 1.6.8 of [@neebook], and the second one is is given by Proposition 1.2.1 and Remark 1.2.2 of ibid.
\[dcomp\] Assume that ${\underline{C}}$ has coproducts; ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$.
1. \[idcompa\] An object $M$ of ${\underline{C}}$ is said to be [*compact*]{} if the functor $H^M={\underline{C}}(M,-):{\underline{C}}\to {\underline{\operatorname{Ab}}}$ respects coproducts.
2. \[idloc\] For ${\underline{D}}\subset {\underline{C}}$ (${\underline{D}}$ is a triangulated category that may be equal to ${\underline{C}}$) one says that ${\mathcal{P}}$ generates ${\underline{D}}$ [*as a localizing subcategory*]{} of ${\underline{C}}$ if ${\underline{D}}$ is the smallest full strict triangulated subcategory of ${\underline{C}}$ that contains ${\mathcal{P}}$ and is closed with respect to ${\underline{C}}$-coproducts.
If this is the case then we will also say that ${\mathcal{P}}$ [*cogenerates*]{} ${\underline{D}}{{^{op}}}$ as a [*colocalizing*]{} subcategory of ${\underline{C}}{{^{op}}}$.
3. \[idhg\] We will say that ${\mathcal{P}}$ [*Hom-generates*]{} a full triangulated category ${\underline{D}}$ of ${\underline{C}}$ containing ${\mathcal{P}}$ if ${\operatorname{Obj}}{\underline{D}}\cap(\cup_{i\in {{\mathbb{Z}}}}{\mathcal{P}}[i]){{}^{\perp}}={\{0\}}$.
4. \[idcg\] We will say that ${\underline{C}}$ is [*compactly generated*]{} if it is Hom-generated by a set of compact objects.
5. \[idcc\] It will be convenient for us to use the following somewhat clumsy terminology: a homological functor $H:{\underline{C}}\to {\underline{A}}$ (where ${\underline{A}}$ is an abelian category) will be called a [*cc*]{} functor if it respects coproducts (i.e., the image of any coproduct in ${\underline{C}}$ is the corresponding coproduct in ${\underline{A}}$); $H$ will be called a [*wcc*]{} functor if it respects countable coproducts.
A cohomological functor $H$ from ${\underline{C}}$ into ${\underline{A}}$ will be called a [*cp*]{} functor if it converts all (small) coproducts into ${\underline{A}}$-products.
Dually, for a triangulated category ${\underline{D}}$ that has products we will call a homological functor $H:{\underline{D}}\to {\underline{A}}$ a [*pp*]{} functor if its respects products.
6. \[idbrown\] We will say that ${\underline{C}}$ satisfies the [*Brown representability*]{} property whenever any cp functor from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$ is representable.
Dually, we will say that a triangulated category ${\underline{D}}$ satisfies the dual Brown representability property if it has products and any pp functor from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$ is corepresentable (i.e., if ${\underline{D}}{{^{op}}}$ satisfies the Brown representability property).
\[pcomp\] Assume that ${\underline{C}}$ has coproducts. I. Let ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$.
1\. If ${\mathcal{P}}$ generates ${\underline{C}}$ as its own localizing subcategory then it also Hom-generates ${\underline{C}}$. Conversely, if ${\mathcal{P}}$ Hom-generates ${\underline{C}}$ and the embedding into ${\underline{C}}$ of its localizing subcategory ${\underline{C}}'$ generated by ${\mathcal{P}}$ possesses a right adjoint then ${\underline{C}}'={\underline{C}}$.
2\. More generally, denote by $C$ the smallest coproductive extension-closed subclass of ${\operatorname{Obj}}{\underline{C}}$ containing ${\mathcal{P}}$. Then ${\mathcal{P}}{{}^{\perp}}\cap C={\{0\}}$. II.1. Assume that ${\underline{C}}$ is compactly generated. Then both ${\underline{C}}$ and ${\underline{C}}{{^{op}}}$ satisfy the Brown representability property.
2\. Assume that ${\underline{C}}$ satisfies the Brown representability property. Then it has products and any exact functor $F$ from ${\underline{C}}$ (into a triangulated category ${\underline{D}}$) that respects coproducts possesses an exact right adjoint $G$.
I.1. The first part of the assertion is essentially a part of [@neebook Proposition 8.4.1] (note that the simple argument used for the proof of this implication does not require ${\mathcal{P}}$ to be a set); it is also a particular case of assertion I.2.
Now assume that the embedding ${\underline{C}}'\to {\underline{C}}$ possesses a right adjoint. Then Proposition \[pbouloc\](III.\[ibou1\]) implies that ${\underline{C}}'={\underline{C}}$ whenever ${\operatorname{Obj}}{\underline{C}}'{{}^{\perp}}={\{0\}}$ (note here that ${\underline{C}}'$ is a strict subcategory of ${\underline{C}}$). Lastly, ${\operatorname{Obj}}{\underline{C}}'{{}^{\perp}}$ is certainly zero if ${\mathcal{P}}$ Hom-generates ${\underline{C}}$ (since ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}'$).
2\. Similarly to the proof of loc. cit., if $N$ belongs to ${\mathcal{P}}^{\perp}\cap C$ then the class $C\cap {}^{\perp}N$ contains ${\mathcal{P}}$, coproductive and extension-closed. Hence $N\perp N$ and we obtain $N=0$.
II.1. The Brown representability property for ${\underline{C}}$ is given by Proposition 8.4.2 of [@neebook]. The Brown representability for ${\underline{C}}{{^{op}}}$ is immediate from the combination of Theorem 8.6.1 with Remark 6.4.5 of ibid.
2\. The first part of the assertion is given by Proposition 8.4.6 of [@neebook]. The second part is immediate from Theorem 8.4.4 of ibid. (combined with Proposition \[pbouloc\](I)).
On torsion pairs {#shop}
----------------
As we have already said, this paper is mostly dedicated to the study of weight structures and $t$-structures. Now, these notions have much in common; so we start from recalling an (essentially) more general definition of a torsion pair (in the terminology of [@aiya Definition 1.4]; in [@postov Definition 3.2] torsion pairs were called complete Hom-orthogonal pairs).[^17]
\[dhop\]
A couple $s$ of classes ${\mathcal{LO}},{\mathcal{RO}}\subset{\operatorname{Obj}}{\underline{C}}$ (of $s$-left orthogonal and $s$-right orthogonal objects, respectively) will be said to be a [*torsion pair*]{} (for ${\underline{C}}$) if ${\mathcal{LO}}^{\perp}={\mathcal{RO}}$, ${\mathcal{LO}}={}^{\perp}{\mathcal{RO}}$, and for any $M\in{\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $$\label{swd}
L_sM\stackrel{a_M}{\to} M\stackrel{n_M}{\to} R_sM{\to} L_sM[1]$$ such that $L_sM\in {\mathcal{LO}}$ and $ R_sM\in {\mathcal{RO}}$. We will call any triangle of this form an [*$s$-decomposition*]{} of $M$; $a_M$ will be called an [*$s$-decomposition morphism*]{}.
We will also need the following auxiliary definitions.
\[dhopo\] Let $s=({\mathcal{LO}},{\mathcal{RO}})$ be a torsion pair.
1\. We will say that $s$ is [*coproductive*]{} (resp. [*productive*]{}) if ${\mathcal{RO}}$ is coproductive (resp. ${\mathcal{LO}}$ is productive). We will also say that $s$ is smashing (resp. cosmashing) if ${\underline{C}}$ in addition has coproducts (resp. products).
We will also use the following modification of the smashing condition: we will say that $s$ is [*countably smashing*]{} whenever ${\underline{C}}$ has coproducts and ${\mathcal{RO}}$ is closed with respect to countable ${\underline{C}}$-coproducts.[^18] 2. For another torsion pair $s'=({\mathcal{LO}}',{\mathcal{RO}}')$ for ${\underline{C}}$ we will say that $s$ is left adjacent to $s'$ or that $s'$ is right adjacent to $s$ if ${\mathcal{RO}}={\mathcal{LO}}'$.
3\. We will say (following [@postov Definition 3.1]) that $s$ is [*generated by ${\mathcal{P}}\subset {\mathcal{LO}}$*]{} if ${\mathcal{P}}^\perp={\mathcal{RO}}$.[^19]
We will say that $s$ is [*compactly generated*]{} if it is generated by some set of compact objects.
4\. For ${\underline{C}'}$ being a full triangulated subcategory of ${\underline{C}}$ we will say that $s$ restricts to it whenever $({\mathcal{LO}}\cap {\operatorname{Obj}}{\underline{C}'},{\mathcal{RO}}\cap {\operatorname{Obj}}{\underline{C}'})$ is a torsion pair for ${\underline{C}}'$.
5\. For a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ we will say that a ${\underline{C}}$-morphism $h$ is [*${\mathcal{P}}$-null*]{} whenever for all $M\in {\mathcal{P}}$ we have $H^M(h)=0$ (where $H^M={\underline{C}}(M,-):{\underline{C}}\to {\underline{\operatorname{Ab}}}$).[^20]
\[rgen\] 1. If ${\mathcal{P}}$ generates a torsion pair $s$ then ${\mathcal{RO}}= {\mathcal{P}}^\perp$ and ${\mathcal{LO}}={}^{\perp}{\mathcal{RO}}$; thus ${\mathcal{P}}$ determines $s$ uniquely. So we will say that $s$ is [**the**]{} torsion pair generated by ${\mathcal{P}}$.
2\. On the other hand, for a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ the corresponding couple $s=({\mathcal{LO}},{\mathcal{RO}})$ (where ${\mathcal{RO}}= {\mathcal{P}}^\perp$ and ${\mathcal{LO}}={}^{\perp}{\mathcal{RO}}$) certainly satisfies the orthogonality properties prescribed by Definition \[dhop\]. Yet simple examples demonstrate that the existence of $s$-decompositions can fail in general. In [@postov Definition 3.1] a couple that satisfies only the orthogonality properties in Definition \[dhop\] was called a [*Hom-orthogonal pair*]{} (in contrast to complete Hom-orthogonal pairs). The reader may easily note that several of our arguments below work for “general” Hom-orthogonal pairs; yet the author has chosen not to treat this more general definition in the current paper. 3. The object $M$ “rarely” determines its $s$-decomposition triangle (\[swd\]) canonically (cf. Remark \[rstws\](1) below). Yet we will often need some choices of its ingredients; so we will use the notation of (\[swd\]). 4. Our definition of torsion pair actually follows [@postov Definition 3.2] and differs from Definition 1.4(i) of [@aiya]. However, Proposition \[phop\](9) below yields immediately that these two definitions are equivalent.
As noted in [@postov], some other authors use the term “torsion pair” to denote the couple $s$ associated with a $t$-structure (see Remark \[rtst1\](\[it1\]) below). So, the term “complete Hom-orthogonal pair” would be less ambiguous; yet it does not fit well (linguistically) with the notion of $\Phi$-orthogonality that we will introduce below.
We make some simple observations.
\[phop\] Let $s=({\mathcal{LO}},{\mathcal{RO}})$ be a torsion pair for ${\underline{C}}$, $i, j\in {{\mathbb{Z}}}$. Then the following statements are valid.
1\. Both ${\mathcal{LO}}$ and ${\mathcal{RO}}$ are Karoubi-closed and extension-closed in ${\underline{C}}$.
2\. ${\mathcal{LO}}$ is coproductive and ${\mathcal{RO}}$ is productive.
3\. If $s$ is coproductive (resp. productive) then the class ${\mathcal{RO}}[i]\cap {\mathcal{LO}}[j]$ is coproductive (resp. productive) also.
4\. Assume that ${\underline{C}}$ has coproducts. Then $s$ is (countably) smashing if and only if the coproduct of any $s$-decompositions of $M_i\in {\operatorname{Obj}}{\underline{C}}$ gives an $s$-decomposition of $\coprod M_i$; here $i$ runs through any (countable) index set.
Dually, if ${\underline{C}}$ has products then $s$ is cosmashing if and only if the product of any $s$-decompositions of any small family of $M_i\in {\operatorname{Obj}}{\underline{C}}$ gives an $s$-decomposition of $\prod M_i$. 5. If $s$ is left adjacent to a torsion pair $s'$ (for ${\underline{C}}$) then $s$ is coproductive and $s'$ is productive.
6\. $s^{op}=({\mathcal{RO}},{\mathcal{LO}})$ is a torsion pair for ${\underline{C}}{{^{op}}}$.
7\. $s$-decompositions are “weakly functorial” in the following sense: any ${\underline{C}}$-morphism $g:M\to M'$ can be completed to a morphism between any choices of $s$-decompositions of $M$ and $M'$, respectively.
In particular, if $M\in {\mathcal{LO}}$ then it is a retract of any choice of $L_sM$ (see Remark \[rgen\](3)).
8\. A morphism $h\in {\underline{C}}(M,N)$ is ${\mathcal{LO}}$-null if and only if factors through an element of ${\mathcal{RO}}$. Moreover, the couple $({\mathcal{LO}},{\mathcal{LO}}-{\text{null}})$ (the latter is the class of ${\mathcal{LO}}$-null morphisms) is a projective class in the sense of [@christ] (see Remark \[rwsts\](3) below).
9\. For $L,R\subset {\operatorname{Obj}}{\underline{C}}$ assume that $L\perp R$ and that for any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $l\to M\to r\to l[1]$ for $l\in L$ and $r\in R$. Then $(\operatorname{\operatorname{Kar}}_{{\underline{C}}}(L),\operatorname{\operatorname{Kar}}_{{\underline{C}}}(R))$ is a torsion pair for ${\underline{C}}$. 10. Assume that $({\mathcal{LO}}',{\mathcal{RO}}')$ is a torsion pair in a triangulated category ${\underline{C}}'$; assume that $F:{\underline{C}}\to {\underline{C}}'$ is an exact functor that is essentially surjective on objects, and such that $F({\mathcal{LO}})\subset {\mathcal{LO}}'$ and $F({\mathcal{RO}})\subset {\mathcal{RO}}'$. Then $({\mathcal{LO}}',{\mathcal{RO}}')= (\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{LO}})),\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{RO}})))$.
1–3, 5, 6. Obvious. 4. The “only if” implication follows immediately from Proposition \[pcoprtriang\] (along with its dual).
Conversely, assume that (countable) coproducts of $s$-decompositions are $s$-decompositions also. Since for any (countable) set of $R_i\in {\mathcal{RO}}$ the distinguished triangles $0\to R_i\to R_i\to 0$ are $s$-decompositions of $R_i$, we obtain $\coprod R_i\in{\mathcal{RO}}$; hence $s$ is (countably) smashing. To prove the remaining (“cosmashing”) part of the assertion one can apply duality (along with assertion 6).
7\. According to [@bbd Proposition 1.1.9], to prove the first part of the assertion it suffices to verify the following: for any $s$-decomposition triangles (\[swd\]) and $L_sM'\to M'\to R_sM' {\to} L_sM'[1]$ the composition $L_sM\to M\to M'\to R_sM'$ vanishes. This certainly follows from ${\mathcal{LO}}\perp{\mathcal{RO}}$.
The “in particular” part of the assertion follows if we take $g={\operatorname{id}}_M$, $M'=M$, and take the triangle $M\stackrel{{\operatorname{id}}_M}{\to} M\to 0\to M[1]$ as (“the first”) $s$-decomposition of $M$.
8\. Since ${\mathcal{LO}}\perp{\mathcal{RO}}$, any morphism that factors through ${\mathcal{RO}}$ is ${\mathcal{LO}}$-null. Conversely, if $h:M\to N$ is ${\mathcal{LO}}$-null then for any choice of an $s$-decomposition of $M$ the composition $h\circ a_M$ is zero (see (\[swd\]) for the notation). It certainly follows that $h$ factors through $R_sM\in {\mathcal{RO}}$.
Next, for any $L\in {\mathcal{LO}}$ and $h\in {\mathcal{LO}}$-null we have $H^L(h)=0$ by definition (see §\[snotata\] or Definition \[dhopo\](5) for this notation). Arguing as in the proof [@christ Lemma 3.2], we easily obtain that (to prove the second part of the assertion) it remains to construct for $X\in {\operatorname{Obj}}{\underline{C}}$ a morphism $a:L\to X$ such that the cone morphism for it is a ${\mathcal{LO}}$-null one. According to the first part of the assertion, for this purpose we can take $a$ being any choice of $a_X$ (in the notation of (\[swd\])).
9\. Certainly, $\operatorname{\operatorname{Kar}}_{{\underline{C}}}(L)\perp \operatorname{\operatorname{Kar}}_{{\underline{C}}}(R)$.
Assume that $M\in {{}^{\perp}}R$. Then in the corresponding distinguished triangle $l\to M\stackrel{f}{\to} r\to l[1]$ we have $f=0$; hence $l\in \operatorname{\operatorname{Kar}}_{{\underline{C}}}(L)$ and ${{}^{\perp}}R \subset \operatorname{\operatorname{Kar}}_{{\underline{C}}}(L)$. Dually (cf. assertion 6) if $M \in \operatorname{\operatorname{Kar}}_{{\underline{C}}}(R)$ then it is a retract of the corresponding $r$; thus $L^\perp\subset \operatorname{\operatorname{Kar}}_{{\underline{C}}}(R)$. This concludes the proof.
10\. For any object $M'$ of ${\underline{C}}'$ choose $M\in {\operatorname{Obj}}{\underline{C}}$ such that $F(M)\cong M'$; choose an $s$-decomposition $L_sM\to M \to R_sM\to L_sM[1]$ of $M$. Then we obtain a distinguished triangle $F(L_sM)\to M' \to F(R_sM)\to F(L_sM)[1]$. Next, the relation of $s$ to $s'$ implies that $F({\mathcal{LO}})\perp F({\mathcal{RO}})$. Hence $(\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{LO}})),\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{RO}}))$ is a torsion pair in ${\underline{C}}'$.
Next, ${\mathcal{LO}}'$ and ${\mathcal{RO}}'$ are Karoubi-closed in ${\underline{C}}'$; hence $\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{LO}}))\subset {\mathcal{LO}}'$ and $\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{RO}}))\subset {\mathcal{RO}}'$. On the other hand, $$\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{LO}}))={{}^{\perp}}(\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{RO}})))\supset {{}^{\perp}}{\mathcal{RO}}'={\mathcal{LO}}';$$ dualizing we obtain the remaining inclusion $\operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\mathcal{RO}}))\supset {\mathcal{RO}}$.
\[rwsts\] 1. We have to pay a certain price for uniting weight structures and $t$-structures in a single definition. The problem is that we have ${\mathcal{LO}}[1]\subset {\mathcal{LO}}$ for $t$-structures, whereas for weight structures the opposite inclusion is fulfilled (and we will see below that these inclusions actually characterize $t$-structures and weight structures). So, the left orthogonal class for a weight structure is actually “a right one with respect to shifts”. Also, the definition of left and right adjacent (weight and $t$-) structures in [@bws] was “symmetric”, i.e., $w$ being left adjacent to $t$ and $t$ being left adjacent to $w$ were synonyms; in contrast, our current convention follows Definition 3.10 of [@postov]. So, $w$ and $t$ being left adjacent in the sense introduced in the previous papers is equivalent to the torsion pair associated with $w$ (see Remark \[rwhop\](1) below) being left adjacent for the torsion pair associated with $t$ “up to a shift” (see Remark \[rstws\](4)).[^21] Lastly, we will study the [*hearts*]{} both for weight structures and $t$-structures. The corresponding definitions are very much similar; yet we are not able to give a single definition in terms of torsion pairs.
2\. Certainly, part 6 of the proposition essentially (see part 1 of this remark) generalizes Proposition \[pbw\](\[idual\]) below, whereas Proposition \[phop\](7) is closely related to Proposition \[pbw\](\[icompl\]).
In particular, Proposition \[phop\](7) easily implies that for any functor $H:{\underline{C}}\to {\underline{A}}$ (for any abelian category ${\underline{A}}$) the correspondence $M\mapsto \operatorname{\operatorname{Im}}(H(L_sM)\to H(M))$ for $M\in {\operatorname{Obj}}{\underline{C}}$ gives a well-defined subfunctor of $H$; cf. Proposition 2.1.2(1) of [@bws]. Thus $s$ yields a certain (two-step) filtration on any (co)homology theory defined on ${\underline{C}}$. This is a certain generalization of the weight filtration defined in ibid. (cf. Remark \[rwhop\] below). The main distinction is that we don’t know (in general) the relation between $s$ and the “shifted” torsion pairs $s[i]=({\mathcal{LO}}[i],{\mathcal{RO}}[i])$; thus there appears to be no reasonable way to obtain “longer” filtrations using this observation.
3\. Recall (see Proposition 2.6 of [@christ]; cf. also [@modoi]) that a projective class in a triangulated category ${\underline{C}}$ is a couple $({\mathcal{P}},I)$ for ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ and $I$ being the class of ${\mathcal{P}}$-null morphisms that satisfy the following additional conditions: ${\mathcal{P}}$ is the largest class such that all elements of $I$ are ${\mathcal{P}}$-null and for any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $L\to M\stackrel{n}{\to} R\to L[1]$ such that $L\in {\mathcal{P}}$ and $n\in I$.
The author has proved the relation of torsion pairs to projective classes for the purposes of applying it in §\[swgws\] below. He was not able to get anything useful from this relation yet; so the reader may ignore projective classes in this text. Note however that knowing the notion of a projective class is necessary to trace the (close) relation of our proof of Theorem \[tpgws\] to Lemma 2.2 of [@modoi].
We also prove some simple statements on torsion pairs in categories that have coproducts.
\[phopft\] Let $s=({\mathcal{LO}},{\mathcal{RO}})$ be a torsion pair generated by some ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ in a triangulated category ${\underline{C}}$. Then the following statements are valid.
I. Assume that ${\underline{C}}$ has coproducts and ${\mathcal{P}}$ is a set of compact objects. Denote by ${\underline{D}}$ the localizing subcategory of ${\underline{C}}$ generated by ${\mathcal{P}}$; $E=(\cup_{i\in {{\mathbb{Z}}}}{\mathcal{P}}[i])^\perp$.
1\. ${\mathcal{LO}}\subset {\operatorname{Obj}}{\underline{D}}$, whereas ${\mathcal{RO}}$ is precisely the class of extensions of elements of $E$ by that of $D={\operatorname{Obj}}{\underline{D}}\cap {\mathcal{RO}}$.
2\. $({\mathcal{LO}},D)$ is a torsion pair for ${\underline{D}}$.
II\. Let $F:{\underline{C}}\to {\underline{C}}'$ be an exact functor that possesses a right adjoint $G$. 1. Assume in addition that $F$ is a full embedding. Then $s'=({\mathcal{LO}}', {\mathcal{RO}}')$ is the torsion pair generated by $F({\mathcal{P}})$ in ${\underline{C}}'$, where ${\mathcal{RO}}'$ is the class of extensions of elements of $F({\operatorname{Obj}}{\underline{C}})^{\perp_{{\underline{C}}'}}$ by that of $F({\mathcal{RO}})$, and ${\mathcal{LO}}'$ is the closure of $F({\mathcal{LO}})$ with respect to ${\underline{C}}'$-isomorphisms.
2\. Let $s'=({\mathcal{LO}}', {\mathcal{RO}}')$ be an arbitrary torsion pair in ${\underline{C}}'$. Then the following conditions are equivalent:
a). $F({\mathcal{LO}})\subset {\mathcal{LO}}'$;
b.) $F({\mathcal{P}})\subset {\mathcal{LO}}'$; c). $G({\mathcal{RO}}')\subset {\mathcal{RO}}$.
3\. For $s'$ as in the previous assertion assume in addition that $F({\mathcal{LO}})\subset {\mathcal{LO}}'$ and $F({\mathcal{RO}})\subset {\mathcal{RO}}'$, and $F$ is essentially surjective on objects. Then $s'$ is generated by $F({\mathcal{P}})$.
III\. Assume that ${\underline{C}}$ has coproducts and all elements of ${\mathcal{P}}$ are compact in it. Then $s$ is smashing.
I.1. Certainly, any extension of an element of $E$ by an object of ${\underline{D}}$ belongs to ${\mathcal{P}}^\perp={\mathcal{RO}}$.
Next, Proposition \[pcomp\](II.2) gives the existence of an exact right adjoint $G$ to the embedding of ${\underline{C}'}$ into ${\underline{C}}$. Moreover, any object of ${\underline{C}}$ is an extension of an element of ${\underline{C}}$ by an object of ${\underline{D}}$, and $G$ is equivalent to the localization of ${\underline{C}}$ by the full triangulated subcategory ${\underline{E}}$ whose object class is $E$ according to Proposition \[pbouloc\](III.\[ibou1\],\[ibou2\]). Hence ${\mathcal{LO}}\subset {\operatorname{Obj}}{\underline{D}}$ and we obtain that any element of ${\mathcal{RO}}$ can be presented as an extension of the aforementioned form.
2\. Obviously, $D={\mathcal{LO}}^{\perp_{{\underline{C}'}}}$. Next, the presentation of elements of ${\mathcal{RO}}$ by extensions as above yields that ${\mathcal{LO}}={}^{\perp_{{\underline{C}}'}} D$.
It remains to verify the existence of the corresponding decompositions. For $M\in {\operatorname{Obj}}{\underline{C}'}$ we apply $G$ to (any) its $s$-decomposition; this is easily seen to yield a decomposition of $G(M)\cong M$ with respect to the couple $({\mathcal{LO}},D)$.
II.1. We can certainly assume that ${\underline{C}}$ is a (full) strict subcategory of ${\underline{C}}'$. We apply Proposition \[phop\](9).
Obviously, ${\mathcal{LO}}'\perp{\mathcal{RO}}'$. Since ${\underline{C}}$ is Karoubi-closed in ${\underline{C}}'$, ${\mathcal{LO}}'$ is Karoubi-closed in ${\underline{C}}'$. Next, for any $N\in {\operatorname{Obj}}{\underline{C}}'$ there exists a essentially unique distinguished triangle $$\label{eglud}
N'\to N \to N''\to N'[1]$$ with $N'=G(N)\in{\operatorname{Obj}}{\underline{C}}$ and $N''\in {\underline{C}}^{\perp_{{\underline{C}}'}}$; see Proposition \[pbouloc\](III,\[ibou1\]). Since ${\mathcal{LO}}\perp N''$, we obtain ${\mathcal{LO}}^\perp={\mathcal{RO}}'$.
So it remains to verify that any $N$ (as above) possesses an $s'$-decomposition. We choose an $s$-decomposition $L_sN'\to N'\to R_sN'\to L_sN'[1]$ of $N'$. Applying the octahedral axiom to this distinguished triangle along with (\[eglud\]) we obtain a presentation of $N$ as an extension of $R$ by $L_sN'$, where $R$ is some extension of $N''$ by $R_sN'$. Thus we obtain an $s'$-decomposition of $N$. 2. Certainly, c) is equivalent both to ${\mathcal{P}}\perp G({\mathcal{RO}}')$ and to ${\mathcal{LO}}\perp G({\mathcal{RO}}')$. Applying the adjunction $F \dashv G$ we obtain the equivalences in question.
3\. For $M'\in {\operatorname{Obj}}{\underline{C}}$ we obviously have the following chain of equivalences: $$F({\mathcal{P}})\perp M'\iff {\mathcal{P}}\perp G(M') \iff G(M')\in {\mathcal{RO}}\iff {\mathcal{LO}}\perp G(M')\iff F({\mathcal{LO}}')\perp M'.$$ Now, Proposition \[phop\](10) implies immediately that $F({\mathcal{LO}}){{}^{\perp}}={\mathcal{RO}}'$; hence $F({\mathcal{P}}){{}^{\perp}}={\mathcal{RO}}'$ indeed.
III\. Obvious; cf. also Proposition \[psym\](\[isymcomp\]) below.
$t$-structures: recollection, relation to torsion pairs, and $t$-projectives {#sts}
----------------------------------------------------------------------------
Now we pass to $t$-structures. Certainly, one can easily define them in terms of torsion pairs; still we give the “classical” definition of a $t$-structure here for fixing the notation and for recalling its relation to Definition \[dhop\] (explicitly).
\[dtstr\]
A couple of subclasses ${\underline{C}}^{t\ge 0},{\underline{C}}^{t\le 0}\subset{\operatorname{Obj}}{\underline{C}}$ will be said to be a $t$-structure $t$ on ${\underline{C}}$ if they satisfy the following conditions:
\(i) ${\underline{C}}^{t\ge 0},{\underline{C}}^{t\le 0}$ are strict, i.e., contain all objects of ${\underline{C}}$ isomorphic to their elements.
\(ii) ${\underline{C}}^{t\ge 0}\subset {\underline{C}}^{t\ge 0}[1]$, ${\underline{C}}^{t\le
0}[1]\subset {\underline{C}}^{t\le 0}$.
\(iii) ${\underline{C}}^{t\le 0}[1]\perp {\underline{C}}^{t\ge 0}$.
\(iv) For any $M\in{\operatorname{Obj}}{\underline{C}}$ there exists a [*$t$-decomposition*]{} distinguished triangle $$\label{tdec}
L_tM\to M\to R_tM{\to} L_tM[1]$$ such that $L_tM\in {\underline{C}}^{t\le 0}, R_tM\in {\underline{C}}^{t\ge 0}[-1]$.
We also need the following auxiliary definitions.
\[dtsto\] Let $n\in {{\mathbb{Z}}}$; let $t$ be a $t$-structure on ${\underline{C}}$.
1. \[ito1\] ${\underline{C}}^{t\ge n}$ (resp. ${\underline{C}}^{t\le n}$) will denote ${\underline{C}}^{t\ge
0}[-n]$ (resp. ${\underline{C}}^{t\le 0}[-n]$); ${\underline{C}}^{t=0}={\underline{C}}^{t\ge 0}\cap {\underline{C}}^{t\le 0}$.
2. \[ito2\] ${{\underline{Ht}}}$ will be the full subcategory of ${\underline{C}}$ whose object class is ${\underline{C}}^{t=0}$.
3. \[ito3\] We will say that $t$ is [*right non-degenerate*]{} if $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}^{t\ge i} ={\{0\}}$.
We will say that $t$ is (just) [*non-degenerate*]{} if we also have $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}^{t\le i} ={\{0\}}$.
4. \[ito4\] We will say that $t$ is [*bounded above*]{} if ${\underline{C}}=\cup_{i\in {{\mathbb{Z}}}}{\underline{C}}^{t\le i}$.
5. \[itoe\] Let ${\underline{C}}'$ be a triangulated category endowed with a $t$-structure $t'$. We will say that an exact functor $F:{\underline{C}}\to {\underline{C}}'$ is [*$t$-exact*]{} whenever $F({\underline{C}}^{t\le 0})\subset {\underline{C}}'^{t'\le 0}$ and $F({\underline{C}}^{t\ge 0})\subset {\underline{C}}'^{t'\ge 0}$.
\[rtst1\]
1. \[it1\] Recall that ${\underline{C}}^{t\le n}={}^{\perp}({\underline{C}}^{t\ge n+1})$ and ${\underline{C}}^{t\ge n}={\underline{C}}^{t\le n-1}{}^{\perp}$ (for $t$ and $n$ as above). Thus for ${\mathcal{LO}}={\underline{C}}^{t\le 0}$ and ${\mathcal{RO}}={\underline{C}}^{t\ge 1}$ the couple $s=({\mathcal{LO}},{\mathcal{RO}})$ is a torsion pair for ${\underline{C}}$ that we will call the torsion pair associated with $t$. Conversely, if for a torsion pair $s$ we have ${\mathcal{LO}}[1]\subset {\mathcal{LO}}$ then $({\mathcal{LO}},{\mathcal{RO}}[1])$ is a $t$-structure (that we will say to be associated with $s$). For a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ we will say that ${\mathcal{P}}$ generates $t$ whenever it generates the associated $s$ (certainly, then we have ${\mathcal{P}}\subset {\mathcal{LO}}={\underline{C}}^{t\le 0}$); we will say that $t$ is compactly generated whenever $s$ is.
We will say that $t$ is coproductive (resp., productive; resp., smashing or cosmashing) whenever $s$ is; we will say that $t$ restricts to ${\underline{C}'}$ (see Definition \[dhopo\](4)) whenever $s$ does.
2. \[itd\] It is well known (and also follows from the previous part of this remark along with Proposition \[phop\](6)) that $({\mathcal{LO}},{\mathcal{RO}}[1])$ is a $t$-structure on ${\underline{C}}$ if and only if $({\mathcal{RO}}, {\mathcal{LO}}[-1])$ gives a $t$-structure on ${\underline{C}}{{^{op}}}$.
3. \[it3\] Recall that the triangle (\[tdec\]) is canonically (and functorially) determined by $M$. So for $M'=M[n]$ and $L_tM'\to M'\to R_tM'{\to} L_tM'[1]$ being its $t$-decomposition we will write $t^{\le n}M$ for $L_tM'[-n]$ (and this notation is ${\underline{C}}$-functorial). Moreover, the functor $t^{\le n}: {\underline{C}}\to {\underline{C}}^{t\le n}$ (considered as a full subcategory of ${\underline{C}}$) is right adjoint to the embedding $ {\underline{C}}^{t\le n}\to {\underline{C}}$. Dually, we obtain a functor $t^{\ge n+1}:M\mapsto R_tM'[-n-1]$, and $t^{\ge n+1}$ is left adjoint to the embedding $ {\underline{C}}^{t\ge n+1}\to {\underline{C}}$.
It certainly follows that any $t$-exact functor “commutes with these adjoints”. In particular, this fact may be applied in the case where ${\underline{C}}^{t\le 0}$ is the class of objects of a full triangulated subcategory ${\underline{E}}$ of ${\underline{C}}$ (cf. Proposition \[prtst\](\[itp4\]) below); in this case the triangle (\[tdec\]) is just the triangle (\[ebou\]).
4. \[it4\] Recall that ${{\underline{Ht}}}$ is necessarily an abelian category with short exact sequences corresponding to distinguished triangles in ${\underline{C}}$.
Moreover, have a canonical isomorphism of functors $t^{\le 0}\circ t^{\ge 0}=t^{\le 0}\circ t^{\le 0}$ (if we consider these functors as endofunctors of ${\underline{C}}$). This composite functor $H_0^t$ actually takes values in ${{\underline{Ht}}}\subset {\underline{C}}$, and it is homological if considered this way.
We prove a few properties of $t$-structures related to these observations.
\[prtst\] Let $t$ be a $t$-structure on ${\underline{C}}$. Then the following statements are valid.
1. \[it2\] Assume that $t$ is smashing. Then ${{\underline{Ht}}}$ is an AB4 category that is closed with respect to ${\underline{C}}$-coproducts, and $H_0^t$ is a cc functor (see Definition \[dcomp\](\[idcc\])).
2. \[it2d\] Dually, if $t$ is cosmashing then ${{\underline{Ht}}}$ is an AB4\* category, the embedding ${{\underline{Ht}}}\to {\underline{C}}$ respects products, and $H_0^t$ is a pp functor.
3. \[itp4\] Triangulated subcategories $L\subset {\underline{C}}$ possessing a right adjoint $G$ to the embedding functor $L\to {\underline{C}}$ are in one-to-one correspondence with torsion pairs $({\mathcal{LO}},{\mathcal{RO}})$ such that ${\mathcal{LO}}[1]={\mathcal{LO}}$; the correspondence sends $L$ into $({\operatorname{Obj}}L,{\operatorname{Obj}}L{{}^{\perp}})$.
Moreover, any $({\mathcal{LO}},{\mathcal{RO}})$ of this sort is a $t$-structure.[^22]
4. \[it4sm\] Assume that ${\underline{C}}$ has coproducts. Then smashing torsion pairs $({\mathcal{LO}},{\mathcal{RO}})$ such that ${\mathcal{LO}}[1]={\mathcal{LO}}$ are in one-to-one correspondence with those exact embeddings $L\to {\underline{C}}$ such that the corresponding $G$ (exists and) respects coproducts.
5. \[ittadj\] Assume that ${\underline{C}}$ is compactly generated; let $F:{\underline{C}}\to {\underline{C}}'$ be a full exact embedding respecting coproducts. Then $({\underline{C}}'^{t'\le 0}, {\underline{C}}'^{t'\ge 0})$ is the $t$-structure generated by $F({\underline{C}}^{t\le 0})$ in ${\underline{C}}'$, where ${\underline{C}}'^{t'\ge 0}$ is the class of extensions of elements of $F({\operatorname{Obj}}{\underline{C}})^{\perp_{{\underline{C}}'}}$ by that of $F({\underline{C}}^{t\ge 0})$, and ${\underline{C}}'^{t'\le 0}$ is the closure of $F({\underline{C}}^{t\le 0})$ with respect to ${\underline{C}}'$-isomorphisms.
\[it2\]. ${{\underline{Ht}}}$ is closed with respect to ${\underline{C}}$-coproducts according to Proposition \[phop\](3). Thus ${{\underline{Ht}}}$ is an AB4 category according to Proposition \[pcoprtriang\]. Next, the endofunctors $t^{\le 0} $ and $ t^{\ge 0}$ of ${\underline{C}}$ respect coproducts according to Proposition \[phop\](4); hence their composition also does.
\[it2d\]. This is just the categorical dual to the previous assertion (see Remark \[rtst1\](\[itd\]).
\[itp4\]. If $G$ exists then $({\operatorname{Obj}}L,{\operatorname{Obj}}L{{}^{\perp}})$ is easily seen to be a torsion pair; combine Proposition \[pbouloc\](III.\[ibou1\], II) with Proposition \[phop\](9). Moreover, we certainly have ${\operatorname{Obj}}L[1]={\operatorname{Obj}}L$.
Conversely, if ${\mathcal{LO}}={\mathcal{LO}}[1]$ then the corresponding $L$ is triangulated since it is extension-closed (see Proposition \[phop\](1)). Next, $({\mathcal{LO}},{\mathcal{RO}})$ is a $t$-structure according to Remark \[rtst1\](\[it1\]). Hence $G$ exists according to part \[it3\] of the remark.
\[it4sm\]. We should check which torsion pairs $s=({\mathcal{LO}},{\mathcal{RO}})$ with ${\mathcal{LO}}={\mathcal{LO}}[1]$ are smashing. According to Proposition \[phop\](4) we should check whether coproducts of $s$-decompositions are $s$-decompositions.
Since $s$ is a $t$-structure, $t$-decompositions are canonical (see Remark \[rtst1\](\[it1\]) hence we should check when the endofunctor $t^{\le 0}$ respects coproducts. Now, the embedding $i:L\to {\underline{C}}$ respects coproducts since it possesses a right adjoint; thus $ G$ respects coproducts if and only if $i\circ G=-^{t\le 0}$ does.
\[ittadj\]. Since ${\underline{C}}$ is compactly generated and $F$ respects coproducts, it possesses a right adjoint (see Proposition \[pcomp\](II)). Hence it remains to apply Proposition \[phopft\](II.1) (we take ${\mathcal{P}}={\underline{C}}^{t\le 0}$ in it and invoke Remark \[rtst1\](\[itd\])).
\[rtst2\]
1. \[ismashs\] If ${\underline{C}}$ has coproducts and the embedding $i:L\to {\underline{C}}$ possesses a right adjoint respecting coproducts then $L$ is called a [*smashing subcategory*]{} of ${\underline{C}}$; see [@kellerema]. Moreover, these conditions are equivalent to the perfectness of the class ${\operatorname{Obj}}L$ in ${\underline{C}}$ (see Definition \[dwg\](\[idpc\]) below) according to Proposition \[psym\](\[iperftp\]).
2. \[it5s1\] Both $t$-structures and weight structures are essentially particular cases of torsion pairs corresponding to the cases ${\mathcal{LO}}[1]\subset {\mathcal{LO}}$ and ${\mathcal{LO}}\subset {\mathcal{LO}}[1]$, respectively (see Remark \[rtst1\](\[it1\]) and Remark \[rwhop\](1) below). So, the “shift-stable” torsion pairs described in Proposition \[prtst\](\[itp4\]) yield the “intersection” of these cases.
3. \[it6\] So it is no wonder that the results and arguments of §\[sperfws\] below are closely related to the properties of localizing subcategories of triangulated categories as studied by A. Neeman, H. Krause and others.
4. \[ialsmash\] Proposition \[prtst\](\[it4sm\]) can easily be generalized as follows: if ${\alpha}$ is a regular cardinal (see §\[snotata\]) and ${\underline{C}}$ is closed with respect to coproducts of less than ${\alpha}$ objects then for an exact embedding $L\to {\underline{C}}$ possessing a right adjoint $G$ the functor $G$ respects coproducts of less than ${\alpha}$ objects if and only if the class ${\operatorname{Obj}}L^{\perp_{{\underline{C}}}}$ is closed with respect to ${\underline{C}}$-coproducts of less than ${\alpha}$ elements.
Now we introduce $t$-projective objects (essentially following [@zvon]).
\[dpt\] Let $t$ be a $t$-structure on ${\underline{C}}$.
Then we will write $P_t$ for the class ${}^\perp ({\underline{D}}^{t\ge 1}\cup {\underline{D}}^{t\le -1})$; we will say that its elements are [*$t$-projective*]{}.[^23]
We prove some simple statements relating $t$-projectives to exact functors from ${{\underline{Ht}}}$ into abelian groups.
\[pgen\] Let $t$ be a $t$-structure for ${\underline{C}}$.
Then the following statements are valid.
1. \[ipgen1\] Let $N\in {\underline{C}}^{t\le 0}$. Then we have $t^{\ge 0}N\cong H_0^t(N)$, and the object $H_0^t(N)$ corepresents the restriction of the functor $H^N={\underline{C}}(N,-)$ to ${{\underline{Ht}}}$.
2. \[ipgen2\] $P_t\subset {\underline{C}}^{t\le 0}$, and for any $P\in P_t$ we have natural isomorphisms of functors $$\label{ept}
{\underline{C}}(P,-)\cong {\underline{C}}(P,H_0^t(-))\cong {{\underline{Ht}}}(H_0^t(P), H_0^t(-));$$ the first of them is induced by the transformations $ {\operatorname{id}}_{{\underline{C}}}\to t^{\ge 0}$ and $H_0^t\to t^{\ge 0}$.
3. \[ipgen25\] For any $P\in P_t$ the object $H_0^t(P)$ is projective in ${{\underline{Ht}}}$.
4. \[ipgen4\] Assume that ${\underline{C}}$ (has products and) satisfies the dual Brown representability condition; assume that $t$ is cosmashing. Then $H_0^t$ gives an equivalence of (the full subcategory of ${\underline{C}}$ given by) $P_t$ with the subcategory of projective objects of ${{\underline{Ht}}}$.
\[ipgen1\]. The first part of the assertion is just a particular case of the definition of $H_0^t(-)$ (see Remark \[rtst1\](\[it4\])).
The second part is very easy also (cf. the proof of [@zvon Lemma 2(1)]); just apply the fact that the functor $t^{\ge 0}$ is left adjoint to the embedding $ {\underline{C}}^{t\ge 0}\to {\underline{C}}$ (see Remark \[rtst1\](\[it3\])).
\[ipgen1\]. The first part of the assertion is immediate from Remark \[rtst1\](\[it1\]) (since it gives ${\underline{C}}^{t\le 0}= {{}^{\perp}}{\underline{C}}^{t\ge 1}$). The first isomorphism in (\[ept\]) follows easily from the definitions of $P_t$ and $H^t_0$, whereas the second one is given by assertion \[ipgen1\].
\[ipgen25\]. The statement is given by Lemma 2(1) of [@zvon] (and also easily follows from assertion \[ipgen1\]). \[ipgen4\]. We should prove that any projective object $P_0$ of ${{\underline{Ht}}}$ “lifts” to $P_t$. Consider the functor $H^{P_0}={{\underline{Ht}}}(P_0,-)\circ H_0^t:{\underline{C}}\to {\underline{\operatorname{Ab}}}$. Since $H^t_0$ is a (homological) pp functor according to (the dual to) Proposition \[prtst\](\[it2\]), $H^{P_0}$ is a pp functor also. Hence it is corepresentable by some $P\in {\operatorname{Obj}}{\underline{C}}$ that obviously belongs to $P_t$. It remains to apply assertions \[ipgen1\] and \[ipgen2\] to prove that $H_0^t(P)\cong P_0$.
\[rcompgen\] Now assume that $t$ is generated by the class $\cup_{i\ge 0}{\mathcal{P}}[i]$ for some ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ (i.e., ${\underline{C}}^{t\ge 0}=\cap_{i\ge 1} {\mathcal{P}}{{}^{\perp}}[i]$; it certainly follows that ${\mathcal{P}}\subset {\underline{C}}^{t\le 0}$). Then part \[ipgen1\] of our proposition implies that ${\underline{C}}^{t=0}\cap (\{H_0^t(P):\ P\in {\mathcal{P}}\}{{}^{\perp}})={\{0\}}$, i.e., this class Hom-generates ${{\underline{Ht}}}$. Indeed, if $M$ is a non-zero element of ${\underline{C}}^{t=0}$ then $M\notin {\underline{C}}^{t\ge 1}$. Hence there exists $P\in {\mathcal{P}}$ such that $P\not\perp M$; thus $H_0^t(P) \not\perp M$ also.
Moreover, if ${\mathcal{P}}$ is a set and $t$ is smashing then the ${\underline{C}}$-coproduct $\coprod_{P\in {\mathcal{P}}} H_0^t(P)$ Hom-generates ${{\underline{Ht}}}$ also (see Proposition \[phop\](3)).
Weight structures: reminder and pure functors {#sws}
=============================================
In §\[ssws\] we recall some basics on weight structures (so, the only new result of the subsection is the remark on the relation of weight structures to torsion pairs). In §\[sswc\] we recall some properties of the weight complex functors. Our treatment of this subject is “more accurate” than the original one in [@bws].
In §\[sdetect\] we construct [*pure*]{} (cf. Remark \[rwrange\](5)) homological functors from ${\underline{C}}$ starting from additive functors from ${{\underline{Hw}}}$ into abelian categories. We also study conditions ensuring that a functor of this sort “detects weights of objects”. The results of this section are important for the study of Picard groups of triangulated categories (endowed with weight structures) carried over in [@bontabu].
In §\[svtt\] we recall the notion of virtual $t$-truncations of (co)homological functors from ${\underline{C}}$ and relate them to functors of limited weight range (that are important for [@bkw]).
In §\[sprcoprod\] we prove that weight decompositions and weight complexes “respect coproducts” whenever $w$ is smashing; it follows that virtual $t$-truncations of cc and cp functors are cc and cp functors, respectively (in this case). We also study generators (see Remark \[rgenw\]) for ${{\underline{Hw}}}$.
Weight structures: basics {#ssws}
--------------------------
Let us recall the definition of one of the main notions of this paper.
\[dwstr\]
A couple of subclasses ${\underline{C}}_{w\le 0},{\underline{C}}_{w\ge 0}\subset{\operatorname{Obj}}{\underline{C}}$ will be said to define a [*weight structure*]{} $w$ for a triangulated category ${\underline{C}}$ if they satisfy the following conditions.
\(i) ${\underline{C}}_{w\le 0}$ and ${\underline{C}}_{w\ge 0}$ are Karoubi-closed in ${\underline{C}}$ (i.e., contain all ${\underline{C}}$-retracts of their elements).[^24] (ii) [**Semi-invariance with respect to translations.**]{}
${\underline{C}}_{w\le 0}\subset {\underline{C}}_{w\le 0}[1]$ and ${\underline{C}}_{w\ge 0}[1]\subset
{\underline{C}}_{w\ge 0}$.
\(iii) [**Orthogonality.**]{}
${\underline{C}}_{w\le 0}\perp {\underline{C}}_{w\ge 0}[1]$.
\(iv) [**Weight decompositions**]{}.
For any $M\in{\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $LM\to M\to RM {\to} LM[1]$ such that $LM\in {\underline{C}}_{w\le 0} $ and $ RM\in {\underline{C}}_{w\ge 0}[1]$.
We will also need the following definitions.
\[dwso\]
Let $i,j\in {{\mathbb{Z}}}$.
1. \[id1\] The full subcategory ${{\underline{Hw}}}\subset {\underline{C}}$ whose object class is ${\underline{C}}_{w=0}={\underline{C}}_{w\ge 0}\cap {\underline{C}}_{w\le 0}$ is called the [*heart*]{} of $w$.
2. \[id2\] ${\underline{C}}_{w\ge i}$ (resp. ${\underline{C}}_{w\le i}$, ${\underline{C}}_{w= i}$) will denote ${\underline{C}}_{w\ge
0}[i]$ (resp. ${\underline{C}}_{w\le 0}[i]$, ${\underline{C}}_{w= 0}[i]$).
3. \[id3\] ${\underline{C}}_{[i,j]}$ denotes ${\underline{C}}_{w\ge i}\cap {\underline{C}}_{w\le j}$; so, this class equals ${\{0\}}$ if $i>j$.
${\underline{C}}^b\subset {\underline{C}}$ will be the category whose object class is $\cup_{i,j\in {{\mathbb{Z}}}}{\underline{C}}_{[i,j]}$.
4. \[id4\] We will say that $({\underline{C}},w)$ is [*bounded*]{} if ${\underline{C}}^b={\underline{C}}$ (i.e., if $\cup_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w\le i}={\operatorname{Obj}}{\underline{C}}=\cup_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w\ge i}$).
Respectively, we will call $\cup_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w\le i}$ (resp. $\cup_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\ge
i}$) the class of $w$-[*bounded above*]{} (resp. $w$-[*bounded below*]{}) objects; we will say that $w$ is bounded above (resp. bounded below) if all the objects of ${\underline{C}}$ satisfy this property.
5. \[ideg\] We will call the elements of $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\le i}$ (resp. of $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\ge i}$) [*right degenerate*]{} (resp. [*left degenerate*]{}).
Respectively, we will say that $w$ is [*non-degenerate*]{} if $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\le i}=\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\ge i}={\{0\}}$ (i.e., if all degenerate objects of ${\underline{C}}$ are trivial). We will say that $w$ is [*right non-degenerate*]{} (resp. [*left non-degenerate*]{}) if $\bigcap\limits_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w \le i} = \{0\}$ (resp. $\bigcap\limits_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w \ge i} = \{0\}$).
6. \[idadj\] If $t$ is a $t$-structure on ${\underline{C}}$ then we will say that $w$ is left adjacent to $t$ or that $t$ is right adjacent to $w$ if ${\underline{C}}_{w\ge 0}={\underline{C}}^{t\le 0}$. Dually, $w$ is right adjacent to $t$ whenever ${\underline{C}}_{w\le 0}={\underline{C}}^{t\ge 0}$.
7. \[ineg\] An additive subcategory $D\subset {\operatorname{Obj}}{\underline{C}}$ will be called [*negative*]{} if for any $i>0$ we have ${\operatorname{Obj}}D\perp {\operatorname{Obj}}D[i]$.
8. \[idwe\] Let ${\underline{C}}'$ be a triangulated category endowed with a weight structure $w'$; let $F:{\underline{C}}\to {\underline{C}}'$ be an exact functor.
We will say that $F$ is [*left weight-exact*]{} (with respect to $w,w'$) if it maps ${\underline{C}}_{w\le 0}$ to ${\underline{C}}'_{w'\le 0}$; it will be called [*right weight-exact*]{} if it maps ${\underline{C}}_{w\ge 0}$ to ${\underline{C}}'_{w'\ge 0}$. $F$ is called [*weight-exact*]{} if it is both left and right weight-exact.
\[rstws\]
1\. Similarly to Remark \[rgen\](3), we will sometimes need a choice of a weight decomposition of $M[-m]$ shifted by $[m]$. So we take a distinguished triangle $$\label{ewd} w_{\le m}M\to M\to w_{\ge m+1}M$$ with some $ w_{\ge m+1}M\in {\underline{C}}_{w\ge m+1}$, $ w_{\le m}M\in {\underline{C}}_{w\le m}$; we will call it an [*$m$-weight decomposition*]{} of $M$, and call arbitrary choices of $w_{\ge m+1}M$ and $ w_{\le m}M$ [*weight truncations of $M$*]{} (for all $m\in {{\mathbb{Z}}}$). We will use this notation below (though $w_{\ge m+1}M$ and $ w_{\le m}M$ are not canonically determined by $M$). Moreover, when we will write arrows of the type $w_{\le m}M\to M$ or $M\to w_{\ge m+1}M$ we will always assume that they come from some $m$-weight decomposition of $M$.
2\. A simple (and still useful) example of a weight structure comes from the stupid filtration on the homotopy categories of cohomological complexes $K(B)$ for an arbitrary additive $B$. In this case $K(B)_{{w^{st}}\le 0}$ (resp. $K(B)_{{w^{st}}\ge 0}$) will be the class of complexes that are homotopy equivalent to complexes concentrated in degrees $\ge 0$ (resp. $\le 0$); see Remark 1.2.3(1) of [@bonspkar] for more detail. The heart of this weight structure is the Karoubi-closure of $B$ in $K(B)$ (that is actually equivalent to $\operatorname{\operatorname{Kar}}(B)$).
3\. In the current paper we use the “homological convention” for weight structures; it was previously used in [@wildcons], [@wildab], [@brelmot], [@bkw], [@bonspkar], and [@bgn], whereas in [@bws], [@bger], [@bach], and [@bontabu] the “cohomological convention” was used.[^25] In the latter convention the roles of ${\underline{C}}_{w\le 0}$ and ${\underline{C}}_{w\ge 0}$ are interchanged, i.e., one considers ${\underline{C}}^{w\le 0}={\underline{C}}_{w\ge 0}$ and ${\underline{C}}^{w\ge 0}={\underline{C}}_{w\le 0}$. Note also that in this paper we will (following [@bbd] and coherently with all the papers of the author cited here) use the “cohomological” convention for $t$-structures. This “discrepancy between conventions” will force us to put somewhat weird “$-$” signs in some of the formulas (cf. also Definition \[dwso\](\[idadj\])); however, it is coherent with Definition \[dhopo\](2) (of adjacent torsion pairs).
Lastly, in [@bws] both “halves” of $w$ were required to be additive. Yet this additional restriction is easily seen to follow from the remaining axioms; see Remark 1.2.3(4) of [@bonspkar].
4\. As we had already noted in Remark \[rwsts\], the current definition of right and left adjacent “structures” is somewhat different from the one used in previous papers of the author. Also, if $w$ is left or right adjacent to $t$ then the associated torsion pairs are only “adjacent up to a shift”; yet this is easily seen to make no difference in the proofs.
Now we recall some basic properties of weight structures.
\[pbw\] Let ${\underline{C}}$ be a triangulated category endowed with a weight structure $w$, $M, M',M''\in {\operatorname{Obj}}{\underline{C}}$, $i,m,n\in {{\mathbb{Z}}}$. Then the following statements are valid.
1. \[idual\] The axiomatics of weight structures is self-dual, i.e., for ${\underline{D}}={\underline{C}}^{op}$ (so ${\operatorname{Obj}}{\underline{C}}={\operatorname{Obj}}{\underline{D}}$) there exists the (opposite) weight structure $w'$ for which ${\underline{D}}_{w'\le 0}={\underline{C}}_{w\ge 0}$ and ${\underline{D}}_{w'\ge 0}={\underline{C}}_{w\le 0}$.
2. \[iextw\] ${\underline{C}}_{w\le i}$, ${\underline{C}}_{w\ge i}$, and ${\underline{C}}_{w=i}$ are Karoubi-closed and extension-closed in ${\underline{C}}$ (and so, additive).
3. \[iort\] ${\underline{C}}_{w\ge i}=({\underline{C}}_{w\le i-1})^{\perp}$ and ${\underline{C}}_{w\le i}={}^{\perp} ({\underline{C}}_{w\ge i+1})$.
4. \[igenlm\] The class ${\underline{C}}_{[m,n]}$ is the extension-closure of $\cup_{m\le j\le n}{\underline{C}}_{w=j}$.
5. \[iwd0\] If $M\in {\underline{C}}_{w\ge m}$ then $w_{\le n}M\in {\underline{C}}_{[m,n]}$ (for any $n$-weight decomposition of $M$). Dually, if $M\in {\underline{C}}_{w\le n}$ then $w_{\ge m}M\in {\underline{C}}_{[m,n]}$.
6. \[icompl\] Assume that $ m\le n$. The for any (fixed) $m$-weight decomposition of $M$ and an $n$-weight decomposition of $M'$ (see Remark \[rstws\](1)) any morphism $g\in {\underline{C}}(M,M')$ can be extended to a morphism of the corresponding distinguished triangles: $$\label{ecompl} \begin{CD} w_{\le m} M@>{c}>>
M@>{}>> w_{\ge m+1}M\\
@VV{h}V@VV{g}V@ VV{j}V \\
w_{\le n} M'@>{}>>
M'@>{}>> w_{\ge n+1}M' \end{CD}$$
Moreover, if $m<n$ then this extension is unique (provided that the rows are fixed).
7. \[iwdext\] For any distinguished triangle $M\to M'\to M''\to M[1]$ and any weight decompositions $LM\stackrel{a_{M}}{\to} M\stackrel{n_{M}}{\to} R_M\to LM[1]$ and $LM''\stackrel{a_{M''}}{\to} M''\stackrel{n_{M''}}{\to} R_M''\to LM''[1]$ there exists a commutative diagram $$\begin{CD}
LM @>{}>>LM'@>f>> LM''@>{}>>LM[1]\\
@VV{a_M}V@VV{a_{M'}}V @VV{a_{M''}}V@VV{a_{M}[1]}V\\
M@>{}>>M'@>{}>>M''@>{}>>M[1]\\
@VV{n_M}V@VV{n_{M'}}V @VV{n_{M''}}V@VV{n_{M}[1]}V\\
RM@>{}>>RM'@>{}>>RM''@>{}>>M[1]\end{CD}$$ in ${\underline{C}}$ whose rows are distinguished triangles and the second column is a weight decomposition (along with the first and the third one).
8. \[isplit\] If an ${{\underline{Hw}}}$-morphism $f:A\to B$ is split surjective then there exists an object $C\in {\underline{C}}_{w=0}$ such that $f$ is isomorphic to the canonical epimorphism $B \bigoplus C\to C$.
9. \[iwdmod\] If $M$ belongs to $ {\underline{C}}_{w\le 0}$ (resp. to ${\underline{C}}_{w\ge 0}$) then it is a retract of any choice of $w_{\le 0}M$ (resp. of $w_{\ge 0}M$).
All of the assertions except the two last ones were essentially established in [@bws] (pay attention to Remark \[rstws\](3)!). Assertion \[isplit\] is immediate from assertion \[iextw\] (that says that ${\underline{C}}_{w=0}$ is Karoubi-closed in ${\underline{C}}$).
\[iwdmod\]. The case $M\in {\underline{C}}_{w\le 0}$ of the assertion follows immediately from Proposition \[phop\](7) (see Remark \[rwhop\](1) below). The case $M\in {\underline{C}}_{w\ge 0}$ is just the categorical dual of the first case; see assertion \[idual\].
\[rwhop\]
1\. Similarly to Remark \[rtst1\](\[it1\]), part \[iort\] of our proposition yields that for ${\mathcal{LO}}={\underline{C}}_{w\le 0}$ and ${\mathcal{RO}}={\underline{C}}_{w\ge 1}$ the couple $s=({\mathcal{LO}},{\mathcal{RO}})$ is a torsion pair; we will call it the torsion pair associated with $w$. Conversely, if for a torsion pair $s$ we have ${\mathcal{LO}}\subset {\mathcal{LO}}[1]$ then $({\mathcal{LO}},{\mathcal{RO}}[-1])$ is a weight structure that we will say to be associated with $s$. In this case we will say that $s$ is [*weighty*]{}; for a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ we will say that ${\mathcal{P}}$ generates $w$ whenever it generates $s$ (certainly, then we have ${\mathcal{P}}\subset {\mathcal{LO}}={\underline{C}}_{w\le 0}$, ${\mathcal{P}}{{}^{\perp}}={\underline{C}}_{w\ge 1}$, and $w$ is determined by ${\mathcal{P}}$); $w$ is compactly generated whenever $s$ is.
Respectively, we will say that $w$ is smashing (resp. cosmashing, countably smashing; resp. generated by ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$) whenever the corresponding $s$ is. Lastly, we will say that $w$ restricts to a triangulated subcategory ${\underline{C}'}$ of ${\underline{C}}$ whenever the associated $s$ does; cf. Definition \[dhopo\](4). 2. Certainly, the “most interesting” weight and $t$-structures are the non-degenerate ones. However, this non-degeneracy condition can be quite difficult to check (and it actually fails for certain “important” weight structures; see Remark 2.4.4(1) of [@bkw]). So we prefer not to avoid degenerate weight and $t$-structures in this paper; this makes Proposition \[phopft\](II.1) an important tool.
Now we describe (some) consequences of this proposition for a torsion pair associated with a weight structure $w$ (resp. with a $t$-structure $t$; see Remark \[rtst1\](\[it1\])).
So, assume that ${\underline{C}}$ is a full strict triangulated subcategory of a triangulated category ${\underline{C}}'$, and assume that the embedding ${\underline{C}}\to {\underline{C}}'$ possesses a right adjoint $G$. Then there exists a weight structure $w'$ (resp. a $t$-structure $t'$) on ${\underline{C}}'$ such that ${\underline{C}}'_{w'\le 0}={\underline{C}}_{w\le 0}$ (resp. ${\underline{C}}'^{t'\le 0}={\underline{C}}^{t\le 0}$). It certainly follows that ${\underline{C}}_{w'\ge 0}\cap {\operatorname{Obj}}{\underline{C}}={\underline{C}}_{w\ge 0}$ (resp. ${\underline{C}}^{t'\ge 0}\cap {\operatorname{Obj}}{\underline{C}}={\underline{C}}^{t\ge 0}$); hence ${{\underline{Hw}}}'={{\underline{Hw}}}$ (resp. ${{\underline{Ht}}}'={{\underline{Ht}}}$). Moreover, if $w$ (resp. $t$) is generated by a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ in ${\underline{C}}$ then $w'$ (resp. $t'$) is generated by ${\mathcal{P}}$ in ${\underline{C}}'$.
3\. Now we apply to weight structures Proposition \[phopft\](II.2). Assume that $F:{\underline{C}}\to {\underline{C}}'$ is an exact functor, $G$ is its (exact) right adjoint, and $w$ is generated by some ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$. Then Proposition \[phopft\](II.2) (combined with the first part of this remark) yields the following: $F$ is left weight-exact (with respect to $w$ and some weight structure $w'$ on ${\underline{C}}'$) if and only if $F({\mathcal{P}})\subset {\underline{C}}'_{w'\le 0}$; these conditions are equivalent to the right weight-exactness of $G$.
Note that this statement can certainly be applied for ${\mathcal{P}}={\underline{C}}_{w\le 0}$; hence $F$ is left weight-exact if and only if $G$ is right weight-exact.
4\. Recall also that for a compactly generated triangulated category ${\underline{C}}$ and $F:{\underline{C}}\to {\underline{C}}'$ being an exact functor respecting coproducts Proposition \[pcomp\](II) gives the existence of $G$. Hence these assumptions imply that $F$ is left weight-exact if and only if $F({\mathcal{P}})\subset {\underline{C}}'_{w'\le 0}$.
5\. We will also need the categorical dual of the latter observation. Certainly, it is formulated is follows: if ${\underline{C}}$ is cocompactly cogenerated (i.e., there exists a set of objects of ${\underline{C}}$ that compactly generates ${\underline{C}}{{^{op}}}$), $F$ is an exact functor ${\underline{C}}\to {\underline{C}}'$ that respects products, $w$ and $w'$ are are weight structures on ${\underline{C}}$ and ${\underline{C}}'$, respectively, and $w$ is cogenerated by some ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ (i.e., ${\underline{C}}_{w\le -1}={{}^{\perp}}{\mathcal{P}}$) then $F$ is right weight-exact if and only if $F({\mathcal{P}})\subset {\underline{C}}'_{w'\ge 0}$.
6\. Applying Proposition \[phop\](10) to weighty torsion pairs we certainly obtain the following: if $F:{\underline{C}}\to {\underline{C}}'$ is a weight-exact functor that is surjective on objects then ${\underline{C}}'_{w'\ge 0} = \operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\underline{C}}_{w\ge 0}))$ and ${\underline{C}}'_{w'\le 0} = \operatorname{\operatorname{Kar}}_{{\underline{C}}'}(F({\underline{C}}_{w\le 0}))$.
7\. Lastly, assume that $F$ as above possesses a right adjoint. Then Proposition \[phopft\](II.3) implies that the weight structure $w'$ is generated by $F({\mathcal{P}})$.
Once again, the existence of right adjoint is automatic whenever ${\underline{C}}$ is compactly generated and $F$ respects coproducts.
8\. Assume that $w$ is generated by a class ${\mathcal{P}}$. Then $\cap_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w\ge i}=\cap_{i\in {{\mathbb{Z}}}}{\mathcal{P}}{{}^{\perp}}$. Thus ${\mathcal{P}}$ is left non-degenerate if and only if ${\mathcal{P}}$ Hom-generates ${\underline{C}}$.
On weight Postnikov towers and weight complexes {#sswc}
-----------------------------------------------
To define the weight complex functor we will need the following definitions.
\[dfilt\]
Let $M\in {\operatorname{Obj}}{\underline{C}}$.
1\. A datum consisting of $M_{\le i}\in {\operatorname{Obj}}{\underline{C}}$, $h_i\in {\underline{C}}(M_{\le i},M)$, $j_i\in {\underline{C}}(M_{\le i},M_{\le i+1})$ for $i$ running through integers will be called a [*filtration on $M$*]{} if we have $h_{i+1}\circ j_i=h_i$ for all $i\in {{\mathbb{Z}}}$; we will write ${{\operatorname{Fil}}}_*M$ for this filtration.
A filtration will be called [*bounded*]{} if there exist $l\le m\in {{\mathbb{Z}}}$ such that $M_{\le i}=0$ for all $i<l$ and $h_i$ are isomorphisms for all $i\ge m$.
2\. A filtration as above equipped with distinguished triangles $M_{\le i-1}\stackrel{j_{i-1}}{\to}M_{\le i}\to M_i$ for all $i\in {{\mathbb{Z}}}$ will be called a [*Postnikov tower*]{} for $M$ or for ${{\operatorname{Fil}}}_*M$; this tower will be denoted by $Po_{{\operatorname{Fil}}}$.
We will use the symbol $M^p$ to denote $M_{-p}[p]$; we will call $M^p$ the [*factors*]{} of $Po_{{\operatorname{Fil}}}$. 3. If ${{\operatorname{Fil}}}_*M'=(M'_{\le i}, h'_i, j_i)$ is a filtration of $M'\in {\operatorname{Obj}}{\underline{C}}$ and $g\in {\underline{C}}(M,M')$ then we will call $g$ along with a collection of $g_{\le i}\in {\underline{C}}(M_{\le i}, M'_{\le i})$ a [*morphism of filtrations compatible with $g$*]{} if $g\circ h_i=h'_i\circ g_{\le i}$ and $j'_i\circ g_{\le i} =g_{\le i+1}\circ j_i$ for all $i\in {{\mathbb{Z}}}$.
\[rwcomp\] 1. Composing (and shifting) arrows from triangles in $Po_{{\operatorname{Fil}}}$ for all pairs of two subsequent $i$s one can construct a complex whose $i$th term equals $M^i$ (it is easily seen that this is a complex indeed; cf. Proposition 2.2.2 of [@bws]). We will call it a complex [*associated with*]{} $Po_{{\operatorname{Fil}}}$.
2\. Certainly, any filtration yields a Postnikov tower (uniquely up to a non-unique isomorphism). Furthermore, it is easily seen that any morphism of filtrations extends to a morphism of the corresponding Postnikov towers (defined in the obvious way). Besides, any morphism of Postnikov towers yields a morphism of the associated complexes.
Lastly, note that morphisms of filtrations and Postnikov towers can certainly be added and composed.
3\. The triangles in $Po_{{\operatorname{Fil}}}$ also give the following statement immediately: if a filtration of $M$ is bounded then $M$ belongs to the extension-closure of $\{M_i\}$.
\[dwpt\] Assume that ${\underline{C}}$ is endowed with a weight structure $w$.
1\. We will call a filtration (see Definition \[dfilt\]) ${{\operatorname{Fil}}}_*M$ of $M\in {\operatorname{Obj}}{\underline{C}}$ a [*weight filtration*]{} (of $M$) if the morphisms $h_i:M_{\le i}\to M$ yield $i$-weight decompositions for all $i\in {{\mathbb{Z}}}$ (in particular, $M_{\le i}=w_{\le i}M$).
We will call the corresponding $Po_{{\operatorname{Fil}}}$ a [*weight Postnikov tower*]{} for $M$.
2\. ${\operatorname{Post}_w({\underline{C}})}$ will denote the category whose objects are objects of ${\underline{C}}$ endowed with arbitrary weight Postnikov towers and whose morphisms are morphisms of Postnikov towers.
${{\underline{C}}_w}$ will be the category whose objects are the same as for ${\operatorname{Post}_w({\underline{C}})}$ and such that ${{\underline{C}}_w}(Po_{{{\operatorname{Fil}}}_M},Po_{{{\operatorname{Fil}}}_{M'}})=\operatorname{\operatorname{Im}}({\operatorname{Post}_w({\underline{C}})}(Po_{{{\operatorname{Fil}}}_M},Po_{{{\operatorname{Fil}}}_{M'}})\to {\underline{C}}(M,M'))$ (i.e., we kill those morphisms of towers that are zero on the underlying objects).
3\. For an additive category ${\underline{B}}$, complexes $A,B\in {\operatorname{Obj}}K({\underline{B}})$, and morphisms $m_1,m_2\in C({{\underline{Hw}}})(A,B)$ we will write $m_1\backsim m_2$ if $m_1-m_2=d_Bh+jd_A$ for some collections of arrows $j^*,h^*:A^*\to B^{*-1}$.
We will call this relation the [*weak homotopy one*]{}.[^26]
The following statements were essentially proved in [@bws]. Moreover, the first two of them easily follow from Proposition \[pbw\](\[icompl\]) (along with the corresponding definitions).
\[pwt\] In addition to the notation introduced above assume that ${\underline{B}}$ is an additive category.
1. \[iwpt1\] Any choice of $i$-weight decompositions of $M$ for $i$ running through integers naturally yields a canonical weight filtration for $M$ (with $M_{\le i}=w_{\le i}M$).
Moreover, we have $\operatorname{\operatorname{Cone}}(Y_i\to M)\in {\underline{C}}_{w\ge i+1}$ and $M^i\in {\underline{C}}_{w=0}$.
2. \[iwpt2\] Any $g\in {\underline{C}}(M,M')$ can be extended to a morphism of (any choice of) weight filtrations for $M$ and $M'$, respectively; hence it also extends to a morphism of weight Postnikov towers.
3. \[iwpt3\] The natural functor ${{\underline{C}}_w}\to {\underline{C}}$ is an equivalence of categories.
4. \[iwhecat\] Factoring morphisms in $K({\underline{B}})$ by the weak homotopy relation yields an additive category ${K_{\mathfrak{w}}}({\underline{B}})$. Moreover, the corresponding full functor $K({\underline{B}})\to {K_{\mathfrak{w}}}({\underline{B}})$ is (additive and) conservative.
5. \[iwhefu\] Let ${{\mathcal{A}}}:{\underline{B}}\to {\underline{A}}$ be an additive functor, where ${\underline{A}}$ is any abelian category. Then for any $B,B'\in {\operatorname{Obj}}K({\underline{B}})$ any pair of weakly homotopic morphisms $m_1,m_2\in C({{\underline{Hw}}})(B,B')$ induce equal morphisms of the homology $H_*({{\mathcal{A}}}(B^i))\to H_*({{\mathcal{A}}}(B'^i))$.
6. \[iwhefun\] Sending an object of ${{\underline{C}}_w}$ into the complex described in Remark \[rwcomp\](1) yields a well-defined additive functor $t=t_w:{{\underline{C}}_w}\to {K_{\mathfrak{w}}}({{\underline{Hw}}})$.
We will call this functor the [*weight complex*]{} one.[^27] We will often write $t(M)$ for $M\in {\operatorname{Obj}}{\underline{C}}$ assuming that some weight Postnikov tower for $M$ is chosen; we will say that $t(M)$ is [*a choice of a weight complex*]{} for $M$.
7. \[iwcex\] If $M_0\stackrel{f}{\to} M_1\to M_2$ is a distinguished triangle in ${\underline{C}}$ then any possible “lift” of $f$ (along with $M_0$ and $M_1$) to ${{\underline{C}}_w}$ can be completed to a lift of the couple of morphisms $M_0\stackrel{f}{\to} M_1\to M_2$ to ${{\underline{C}}_w}$ such that corresponding morphisms $t(M_0)\to t(M_1)\to t(M_2)$ yield a distinguished triangle in $K({{\underline{Hw}}})$.
\[rwc\] So, for an object $M$ of ${\underline{C}}$ its weight complex $t(M)$ is well-defined up to a $K({{\underline{Hw}}})$-endomorphism that is weakly homotopic to zero; thus it is defined in $K({{\underline{Hw}}})$ up to a (not necessarily unique) isomorphism.
In particular, if $M\in {\underline{C}}_{w\ge -n}$ for some $n\in {{\mathbb{Z}}}$ then we can take $M_{\le i}=w_{\le i}M=0$ for all $i<-n$; hence any choice of $t(M)$ is homotopy equivalent to a complex concentrated in degrees at most $n$. Similarly, if $M\in {\underline{C}}_{w\le -n}$ then we can take $M_{\le i}=w_{\le i}M=M$ for all $i\le -n$; thus $t(M)$ is homotopy equivalent to a complex concentrated in degrees at least $n$. Hence $t(M)\cong 0$ whenever $M$ is left or right degenerate.
On pure functors and detecting weights {#sdetect}
--------------------------------------
\[ppure\] Assume that ${\underline{C}}$ is endowed with a weight structure $w$.
Let ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}$ be an additive functor, where ${\underline{A}}$ is any abelian category. Choose a weight complex $t(M)=(M^j)$ for each $M\in {\operatorname{Obj}}{\underline{C}}$, and denote by $H(M)=H^{{{\mathcal{A}}}}(M)$ the zeroth homology of the complex ${{\mathcal{A}}}(M^{j})$. Then $H(-)$ yields a homological functor that does not depend on the choices of weight complexes. Moreover, the assignment ${{\mathcal{A}}}\mapsto H^{{\mathcal{A}}}$ is natural in ${{\mathcal{A}}}$.
Immediate from Proposition \[pwt\] (\[iwpt3\], \[iwhefu\],\[iwcex\]).
\[rpure\] 1. (Co)homological functors of this type have already found interesting applications in [@kellyweighomol], [@bach], [@bscwh], [@bontabu], and [@bgn]. We will prove some statements relevant for the latter paper just now. We will not apply all the remaining results of this subsection elsewhere in the paper. 2. We will call a functor $H:{\underline{C}}\to {\underline{A}}$ [*pure*]{} (or $w$-pure) if it equals $H^{{\mathcal{A}}}$ for a certain ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}$. In the next subsection we will prove that this definition of purity is equivalent to another (“intrinsic”) one.
So we prove that pure functors can be used to “detect weights”; these results are crucial for [@bontabu]. This notion of detecting weights is closely related to the one of [*weight-conservativity*]{} that was introduced in [@bach].
To prove the most general case of we have to recall a result from [@bkw].[^28]
\[lbkw\] Let $m\in {{\mathbb{Z}}}$, $M\in {\operatorname{Obj}}{\underline{C}}$, where ${\underline{C}}$ is endowed with a weight structure $w$.
Then $t(M)$ belongs to $K({{\underline{Hw}}})_{{w^{st}}\ge -m }$ (resp. to $K({{\underline{Hw}}})_{{w^{st}}\le -m }$; see Remark \[rstws\](2)) if and only if $M\bigoplus M[1]$ is an extension of an element of ${\underline{C}}_{w\ge -m}$ by a right degenerate object (resp. $M\bigoplus M[-1]$ is an extension of a left degenerate object by an element of ${\underline{C}}_{w\le -m}$).
This statement is contained Corollary 3.1.5 of ibid.
For a homological functor $H$ the symbol $H_i$ will be used to denote the composite functor $H\circ [i]$.
\[pdetect\] Adopt the notation of Proposition \[ppure\] and assume that the following conditions are fulfilled:
\(i) the image of ${{\mathcal{A}}}$ consists of ${\underline{A}}$-projective objects only;
\(ii) if an ${{\underline{Hw}}}$-morphism $h$ does not split (i.e., it is not a retraction) then ${{\mathcal{A}}}(h)$ does not split also.
Then for any $M\in {\operatorname{Obj}}{\underline{C}}$, $m\in {{\mathbb{Z}}}$, and $H=H^{{\mathcal{A}}}$ we have the following: $M$ is $w$-bounded below and $H_i(M)=0$ for all $i> m$ if and only if $M\in {\underline{C}}_{w\ge -m}$.
The “if” part of the statement is very easy (for any ${{\mathcal{A}}}$); just combine the definition of $H^{{\mathcal{A}}}$ with Remark \[rwc\]. Now we prove the converse application. If $M$ is $w$-bounded below then the object $M'=M\bigoplus M[1]$ is $w$-bounded below also. We also have $H_i(M')=0$ for all $i> m$, and it suffices to prove that $M'\in {\underline{C}}_{w\ge -m}$ (by the axiom (i) of weight structures).
Now chose the minimal integer $n\ge m$ such that $t(M')\in K(B)_{{w^{st}}\ge -n}$ (see Remark \[rstws\](2)). Then Lemma \[lbkw\] yields the existence of a distinguished triangle $M_1'\stackrel{g}{\to} M'{\to} M'_2\to M'_1[1]$ such that $M'_1$ is right $w$-degenerate and $M'_2\in {\underline{C}}_{w\ge -n}$. Since $M'$ is bounded, $g=0$; hence $M'$ is a retract of $M'_2$ and so belongs to ${\underline{C}}_{w\ge -n}$ itself.
It remains to prove that $n=m$. Assume the converse. Then $t(M')$ is homotopy equivalent to an ${{\underline{Hw}}}$-complex $(N^i)$ concentrated in degrees $\le n$, and Proposition \[pbw\](\[isplit\]) yields that the boundary morphism $d^{n-1}_N:N^{n-1}\to N^n$ does not split. Thus ${{\mathcal{A}}}(d^{n-1}_N)$ does not split also. Since ${{\mathcal{A}}}(N^n)$ is projective, this non-splitting implies that $H_n(M')\neq 0$. Thus $n\le m$.
Now we formulate a simple corollary from the proposition that will be applied in [@bontabu].
\[cbontabu\] Let ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}$ be a full additive conservative functor whose target is semi-simple. Then for a $w$-bounded object $M$ of ${\underline{C}}$ we have $M\in {\underline{C}}_{w=0}$ if and only if $H^{{\mathcal{A}}}_i(M)=0$ for all $i\neq 0$.
Once again, the “if” part of the statement is simple (and can easily be deduced from the previous proposition). So we prove the converse implication.
Now, Lemma \[lbkw\] allows us to assume that ${\underline{C}}=K({{\underline{Hw}}})$ and $w={w^{st}}$ (since it enables us to treat $t(M)$ instead of $M$). Hence it suffices to prove that $M\in {\underline{C}}_{w\ge 0}$ since then we will also have $M\in {\underline{C}}_{w\le 0}$ by duality (see Proposition \[pbw\](\[idual\]; note that our assumptions on ${{\mathcal{A}}}$ and ${\underline{A}}$ are self-dual and the construction of $H^{{\mathcal{A}}}$ is also so in this case).[^29] Hence it suffices to verify that the functor ${{\mathcal{A}}}$ satisfies the assumptions of Proposition \[pdetect\]. The latter is very easy: all elements of ${{\mathcal{A}}}({{\underline{Hw}}})$ are projective in ${\underline{A}}$ since all objects of ${\underline{A}}$ are (recall that ${\underline{A}}$ is semi-simple), and a full conservative functor obviously does not send non-split morphisms into split ones.
\[rdetect\] 1. As we have (essentially) just noted, condition (ii) of Proposition \[pdetect\] is obviously fulfilled both for ${{\mathcal{A}}}$ and for the opposite functor ${{\mathcal{A}}}^{op}:{{\underline{Hw}}}^{op}\to {\underline{A}}^{op}$ whenever ${{\mathcal{A}}}$ is a full conservative functor. In particular, it suffices to assume that ${{\mathcal{A}}}$ is a full embedding.
Hence it may be useful to assume (in addition to assumption (i) of the proposition) that the image of ${{\mathcal{A}}}$ consists of injective objects only. 2. Now we describe a general method for constructing a full embedding ${{\mathcal{A}}}$ whose image consists of ${\underline{A}}$-projective objects.
Assume that ${{\underline{Hw}}}$ is an essentially small (additive) $R$-linear category, where $R$ is a commutative unital ring (certainly, one may take $R={{\mathbb{Z}}}$ here). Denote some small skeleton of ${{\underline{Hw}}}$ by ${\underline{B}}$ (to avoid set-theoretical difficulties).
Consider the abelian category ${\operatorname{PShv}^R}({\underline{B}})$ of $R$-linear contravariant functors from ${\underline{B}}$ into the category of $R$-modules (cf. §\[snotata\]). Then ${\underline{B}}$ (and so, also ${{\underline{Hw}}}$) embeds into the full subcategory of projective objects of ${\operatorname{PShv}^R}({\underline{B}})$ (by the Yoneda lemma; see Lemma 5.1.2 of [@neebook]). Hence this functor “detects weights” (in the sense of Proposition \[pdetect\]).
3\. The objects in the essential image of this functor may be called [*purely $R$-representable homology*]{}. Since they are usually not injective in ${\operatorname{PShv}^R}({\underline{B}})$, a dual construction may be useful for checking whether $M\in {\underline{C}}_{w\le -m}$.
4\. It is easily seen that in the proof of our corollary (and so, also of the bounded case of Proposition \[pdetect\]) one can replace the usage of Lemma \[lbkw\] by that of [@bws Theorem 3.3.1(IV)].
5\. Condition (ii) of Proposition \[pdetect\] is certainly necessary. Indeed, if $h\in {\operatorname{Mor}}({{\underline{Hw}}})$ does not split whereas ${{\mathcal{A}}}(h)$ does, then one can easily check that $\operatorname{\operatorname{Cone}}(h)\in {\underline{C}}_{w\ge 0}\setminus {\underline{C}}_{w\ge 1}$ and $H_i(\operatorname{\operatorname{Cone}}(h))=0$ for $i\neq -1$.
Now we prove a certain unbounded version of Proposition \[pdetect\]. Note that its proof may be (slightly) simplified if we assume that ${{\underline{Hw}}}$ is Karoubian (cf. [@bkw §3.1] that demonstrates that the general case can be “reduced” to this one).
\[pdetectsse\] Assume that ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}$ (as in the previous proposition) is a full functor, for any $N\in {\underline{C}}_{w=0}$ the ideal $\operatorname{\operatorname{Ker}}({{\underline{Hw}}}(N,N )\to {\underline{A}}({{\mathcal{A}}}(N),{{\mathcal{A}}}(N))$ is a nilpotent ideal of the endomorphism ring ${{\underline{Hw}}}(N,N)$, and ${\underline{A}}$ is abelian semi-simple.
1\. Then $H_i(M)=0$ for all $i> m$ (resp. for all $i<m$) if and only if $M\bigoplus M[1]$ is an extension of an element of ${\underline{C}}_{w\ge -m}$ by a right degenerate object (resp. $M\bigoplus M[-1]$ is an extension of an element of ${\underline{C}}_{w\le -m}$ by a left degenerate object).
2\. Assume in addition that $w$ is non-degenerate. Then these two conditions are equivalent to $M\in {\underline{C}}_{w\ge -m}$ (resp. $M\in {\underline{C}}_{w\le -m}$). In particular, if $H_i(M)=0$ for all $i
\neq m$ then $M\in {\underline{C}}_{w= -m}$.
1\. It suffices to verify that $t(M)$ belongs to $K({{\underline{Hw}}})_{{w^{st}}\le -m }$ (resp. to $K({{\underline{Hw}}})_{{w^{st}}\ge -m }$). Indeed, then applying Lemma \[lbkw\] once again we will certainly obtain the result. Note also that it suffices to verify the “main” version of this assertion since the “resp.” one is its dual.
Now, by Proposition \[pbw\](\[iort\]) it suffices to check that $t(M)$ is $K(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))$-isomorphic to a complex concentrated in degrees $\le m$. To construct the latter we extend ${{\mathcal{A}}}$ to an additive functor ${{\mathcal{A}}}': \operatorname{\operatorname{Kar}}({{\underline{Hw}}})\to {\underline{A}}$ and find a complex $T=(t^i)\in {\operatorname{Obj}}C(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))$ that is $K(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))$-isomorphic to $t(M)$ such that the boundary $d_t^{m}:t^{m}\to t^{m+1}$ is killed by ${{\mathcal{A}}}'$. This is easily seen to be possible according to Theorem 1.3 of [@wildab] (cf. also Theorem 2.2 of ibid.). Indeed, our assumptions on ${{\mathcal{A}}}$ imply immediately that ${{\underline{Hw}}}$ is [*semi-primary*]{} in the sense of . Then $\operatorname{\operatorname{Kar}}({{\underline{Hw}}})$ is semi-primary also according to Proposition 2.3.4(c) of ibid.; hence ${{\mathcal{A}}}': \operatorname{\operatorname{Kar}}({{\underline{Hw}}})\to {\underline{A}}$ satisfies the “kernel nilpotence assumption” similar to that for ${{\mathcal{A}}}$. Thus Theorem 1.3 of [@wildab] says that the cone of any $\operatorname{\operatorname{Kar}}({{\underline{Hw}}})$-morphism is $K(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))$-isomorphic to a cone of a morphism killed by ${{\mathcal{A}}}'$, and so we can “replace” the $m$th boundary of $t(M)$ by a morphism satisfying this condition.
For this complex $T$ the result of the applying ${{\mathcal{A}}}'$ termwisely to its stupid truncation complex ${w^{st}}_{\le -m-1}T$ is certainly zero. Hence the identity of ${w^{st}}_{\le -m-1}T$ is homotopy equivalent to an endomorphism killed by the termwise application of ${{\mathcal{A}}}'$. Applying our nilpotence assumption to a sufficiently high power of this ($K(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))$-invertible) endomorphism we obtain that for any $j\in {{\mathbb{Z}}}$ the complex ${w^{st}}_{\le -m-1}T$ is homotopy equivalent to a complex concentrated in degrees $\ge j$. Hence ${w^{st}}_{\le -m-1}T$ is contractible and we obtain $t(M)$ is ${K(\operatorname{\operatorname{Kar}}({{\underline{Hw}}}))}$-isomorphic to its stupid truncation ${w^{st}}_{\ge -m}T$. 2. Assertion 1 implies that $M\bigoplus M[1]$ belongs to ${\underline{C}}_{w\ge -m}$ (resp. $M\bigoplus M[-1]$ belongs to ${\underline{C}}_{w\le -m}$) itself. Hence $M$ belongs to ${\underline{C}}_{w\ge -m}$ (resp. to ${\underline{C}}_{w\le -m}$) also.
The “in particular” part of the assertion follows immediately.
\[rpsh\] 1. Both Propositions \[pdetect\] and \[pdetectsse\] are easily seen to imply certain generalizations of [@wildcons Theorem 1.5].
2\. Proposition \[pdetectsse\] can probably generalized. In particular, it appears to be sufficient to assume for any $N\in {\underline{C}}_{w=0}$ that all the endomorphisms in $\operatorname{\operatorname{Ker}}({{\underline{Hw}}}(N,N )\to {\underline{A}}({{\mathcal{A}}}(N),{{\mathcal{A}}}(N)))$ are nilpotent.
However, it is demonstrated in [@bontabu] that the case where ${{\mathcal{A}}}$ is a full embedding (into a semi-simple category) is quite interesting (already). In this case the proof can be simplified since then $t(M)$ is obviously a retract of ${w^{st}}_{\ge -m}t(M)$ (resp. of ${w^{st}}_{\le -m}t(M)$).
On virtual $t$-truncations and cohomology of bounded weight range {#svtt}
-----------------------------------------------------------------
We recall the notion of virtual $t$-truncations for a cohomological functor $H:{\underline{C}}\to {\underline{A}}$ (as defined in §2.5 of [@bws] and studied in more detail in §2 of [@bger]). These truncations allow us to “slice” $H$ into $w$-pure pieces. These truncations behave as if they were given by truncations of $H$ in some triangulated “category of functors” ${\underline{D}}$ with respect to some $t$-structure (whence the name). Moreover, this is often actually the case (and we will discuss this matter below); yet the definition does not require the existence of ${\underline{D}}$ (and so, does not depend on its choice). Our choice of the numbering for them is motivated by the cohomological convention for $t$-structures; this convention combined with the homological numbering for weight structures causes the (somewhat weird) “$-$” signs in the definitions and formulations of this section.
\[dvtt\] Let $H$ be a cohomological functor from ${\underline{C}}$ into an abelian category ${\underline{A}}$; assume that ${\underline{C}}$ is endowed with a weight structure $w$ and $ n\in {{\mathbb{Z}}}$.
We define the [*virtual $t$-truncation*]{} functors $\tau^{\ge -n }(H)$ (resp. $\tau^{\le -n }(H)$) by the correspondence $$M\mapsto\operatorname{\operatorname{Im}}(H(w_{\le n+1}M)\to H(w_{\le n}M)) ;$$ (resp. $M\mapsto\operatorname{\operatorname{Im}}(H(w_{\ge n}M)\to H(w_{\ge n-1}M)) $); here we take arbitrary choices of the corresponding weight truncations of $M$ and connect them using Proposition \[pbw\](\[icompl\]).
We recall the main properties of these constructions that were established in [@bws §2.5 and Theorem 4.4.2(7,8)].
\[pwfil\] In the notation of the previous definition the following statements are valid. 1. The objects $\tau^{\ge -n}(H)(M)$ and $\tau^{\le -n}(H)(M)$ are ${\underline{C}}$-functorial in $M$ (and so, the virtual $t$-truncations of $H$ are well-defined functors).
2\. The functors $\tau^{\ge -n }(H)$ and $\tau^{\le -n }(H)$ are cohomological.
3\. There exist natural transformations that yield a long exact sequence $$\label{evtt}
\begin{gathered}
\dots \to \tau^{\ge -n +1}(H)\circ [-1] \to \tau^{\le -n }(H)\to H \\ \to \tau^{\ge -n +1}(H)\to \tau^{\le -n }(H)\circ [1]\to \dots\end{gathered}$$ (i.e., the result of applying this sequence to any object of ${\underline{C}}$ is a long exact sequence); the shift of this exact sequence by $3$ positions is given by composing the functors with $-[1]$. 4. Assume that there exists a $t$-structure $t$ that is right adjacent to $w$. Then for any $M\in {\operatorname{Obj}}{\underline{C}}$ and $H_M={\underline{C}}(-,M):{\underline{C}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ the functors $ \tau^{\le -n }(H_M)$ and $\tau^{\ge -n}(H_M)$ are represented by $t^{\le -n}M$ and $t^{\ge -n}M$, respectively.
Moreover, a stronger and more general statement than part 4 of this proposition is given by [@bger Proposition 2.5.4(1)]; it will be applied in the proof of Proposition \[psaturdu\](2) below.
Now we define weight range and introduce notation for pure cohomological functors. Some of these statements will be applied below, whereas other ones are proved here for the purpose of applying them in [@bkw].
\[drange\] 1. Let $m,n\in {{\mathbb{Z}}}$; let $H$ be as above. Then we will say that $H$ is [*of weight range*]{} $\ge m$ (resp. $\le n$, resp. $[m,n]$) if it annihilates ${\underline{C}}_{w\le m-1}$ (resp. ${\underline{C}}_{w\ge n+1}$, resp. both of these classes).
2\. Let ${{\mathcal{A}}}:{{\underline{Hw}}}^{op}\to {\underline{A}}$ be an additive functor. Then for ${{\mathcal{A}}}^{op}$ being the opposite functor ${{\underline{Hw}}}\to {\underline{A}}^{op}$ we will write $H_{{\mathcal{A}}}$ for the cohomological functor from ${\underline{C}}$ into ${\underline{A}}$ obtained from $H^{{{\mathcal{A}}}^{op}}$ (see Proposition \[ppure\]) by means of reversion of arrows.
We will functors obtained using this construction [*pure cohomological*]{} ones.
\[pwrange\] In the notation of the previous definition the following statements are valid.
1. \[iwrvt\] The functor $\tau^{\ge -n}(H)$ is of weight range $\le n$, and $\tau^{\le -m}(H)$ is of weight range $\ge m$.
2. \[iwrcrit\] Assume that $w$ is bounded. Then $H$ is of weight range $\le n$ (resp. of weight range $\ge m$) if and only if it kills ${\underline{C}}_{w=i}$ for all $i>n$ (resp. for $i>m$).
3. \[iwridemp\] We have $\tau^{\ge -n}(H)\cong H$ (resp. $\tau^{\le -m}(H)\cong H$) if and only if $H$ is of weight range $\le n$ (resp. of weight range $\ge m$).
4. \[iwrcomm\] We have $\tau^{\ge -n}(\tau^{\le -m})(H)\cong \tau^{\le -m}(\tau^{\ge -n})(H)$.
5. \[iwfil4\] If $H$ is of weight range $\ge m$ then $\tau^{\ge -n}(H)$ is of weight range $[m,n]$.
6. \[iwrd\] Dually, if $H$ is of weight range $\le n$ then $\tau^{\le -m}(H)$ is of weight range $[m,n]$.
7. \[iwrvan\] If $m>n$ then the only functors of weight range $[m,n]$ are zero ones.
8. \[iwrpure\] The functors of weight range $[m,m]$ are exactly those of the form $H_{{\mathcal{A}}}\circ [-m]$ (see Definition \[drange\](2)), where ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}^{op}$ is an additive functor.
9. \[iwfil3\] The (representable) functor $H_M:{\underline{C}}\to {\underline{\operatorname{Ab}}}$ if of weight range $\ge m$ if and only if $M\in {\underline{C}}_{w\ge m}$.
10. \[iwfil5\] If $H$ is of weight range $[m,n]$ then the morphism $H(w_{\ge m}M)\to H(M)$ is surjective and the morphism $H(M)\to H(w_{\le n}M)$ is injective (here we take arbitrary choice of the corresponding weight decompositions of $M$ and apply $H$ to their connecting morphisms).
Let $M\in {\underline{C}}_{w\ge n+1}$. Then we can take $w_{\le n}(M)=0$. Thus $\tau^{\ge -n}(H)(M)=0$, and we obtain the first part of assertion \[iwrvt\]. It second part is easily seen to be dual to the first part. The “only if” part of assertion \[iwrcrit\] is immediate from the definition of weight range. The converse implication easily follows from Proposition \[pbw\](\[igenlm\]).
Assertion \[iwridemp\] is precisely Theorem 2.3.1(III2,3) of [@bger]; assertion \[iwrcomm\] is given by part II.3 of that theorem.
Now let $H$ be of weight range $\ge m$. Then $\tau^{\ge -n}(H)\cong \tau^{\ge -n}(\tau^{\le -m})(H)\cong \tau^{\le -m}(\tau^{\ge -n})(H)$ (according to the two previous assertions). It remains to apply assertion \[iwrvt\] to obtain assertion \[iwfil4\].
Assertion \[iwrd\] can be proved similarly; it is also easily seen to be dual to assertion \[iwfil4\].
Next, for any $l\in {{\mathbb{Z}}}$ and any cohomological $H$ any choice of an $l$-weight decomposition triangle (cf. (\[ewd\])) for $M$ gives the long exact sequence $$\label{eles}
\begin{gathered}
\dots \to H((w_{\le l}M)[1])\to H(w_{\ge l+1}M)\to H(M)\\
\to H(w_{\le l}M)\to H((w_{\ge l+1}M)[-1])\to\dots
\end{gathered}$$ The exactness of this sequence in $H(M)$ for $l=n$ immediately gives assertion \[iwrvan\].
\[iwrpure\]. It certainly suffices to verify the statement for $m=0$. Now, the functor $H^{{{\mathcal{A}}},op}$ is easily seen to be of weight range $[0,0]$ for any additive functor ${{\mathcal{A}}}:{\underline{C}}\to {\underline{A}}^{op}$; see Remark \[rwc\].
Conversely, let $H$ be of weight range $[0,0]$. We take the functor ${{\mathcal{A}}}$ being the restriction of $H^{op}:{\underline{C}}\to {\underline{A}}^{op}$ to ${{\underline{Hw}}}$. Then we should check that $H$ sends $M\in {\operatorname{Obj}}{\underline{C}}$ into the homology in $H(M^0)$ of the complex $H(M^{-*})$ (where $t(M)=(M^*)$). This is an immediate consequence of the properties of the [*weight spectral sequence*]{} converging to $H^*(M)$; see Theorem 2.4.2 of [@bws]. Indeed, this spectral sequence converges according to part II(ii) of this theorem, and it remains to apply the vanishing for $H(M^i[j])$ for $j\neq 0$.
Assertion \[iwfil3\] is immediate from Proposition \[pbw\](\[iort\]).
Assertion \[iwfil5\] is an immediate consequence of assertion \[iwfil3\]; just apply (\[eles\]) for $l=m$ and for $l=m-1$, respectively.
\[rwrange\] 1. So, we call cohomological functors of weight range $[0,0]$ and their opposite homological functors $w$-pure ones; this terminology is compatible with Remark \[rpure\](2) according to part \[iwrpure\] of our proposition.
2\. Actually, the arguments used in the proof of this statement are easily seen to be functorial enough to yield an equivalence of the (possibly) big category of pure (cohomological) functors ${\underline{C}}\to {\underline{A}}$ with the one of additive contravariant functors ${{\underline{Hw}}}\to {\underline{A}}$.
3\. Sending $H$ into the pure functor $\tau^{\ge -m}(\tau^{\le -m})(H\circ [m])\cong \tau^{\le -m}(\tau^{\ge -m})(H\circ [m])$ (for $m\in {{\mathbb{Z}}}$) yields a sort of “pure homology” for $H$. It will correspond to the homology of an object representing (or, more generally, [*$\Phi$-representing*]{}) $H$ in the settings that we will consider below. 4. Certainly, a homological functor $H$ from ${\underline{C}}$ into ${\underline{A}}$ may be considered as a cohomological functor from ${\underline{C}}$ into ${\underline{A}}^{op}$. Thus one can easily dualize the aforementioned results; we give some more detail for this here to refer to them later.
Obviously, the virtual $t$-truncation functor $\tau^{\ge -n }(H)$ (resp. $\tau^{\le -n }(H)$) will be defined by the correspondence $M\mapsto\operatorname{\operatorname{Im}}(H(w_{\le n}M)\to H(w_{\le n+1}M)) $ (resp. $M\mapsto\operatorname{\operatorname{Im}}(H(w_{\ge n}M) \to H(w_{\ge n-1}M)$), whereas the arrows in (\[evtt\]) should be reversed.
5\. The author is using the term “pure” due to the relation of pure functors to Deligne’s purity of cohomology.
To explain it we recall that various categories of Voevodsky motives are endowed with so-called Chow weight structures; the first of these weight structures was constructed in [@bws] where it was proved that the category ${\operatorname{DM_{gm}}}(k)$ of geometric motives over a characteristic zero field $k$ is endowed with a weight structure $w_{{\operatorname{Chow}}}(k)$ whose heart is the category of Chow motives over $k$.[^30] Now, for any $r\in {{\mathbb{Z}}}$ the $r$th level of the Deligne’s weight filtration of either of singular of étale cohomology of motives certainly kills ${\operatorname{Chow}}[i]$ for all values of $i$ except one (and the remaining value of $i$ is either $r$ or $-r$ depending on the choice of the convention for Deligne’s weights).[^31] Thus (the corresponding shifts of) Deligne’s pure factors of (singular and étale) cohomology are pure with respect to $w_{{\operatorname{Chow}}(k)}$.
We also formulate a simple statement that will be applied in a succeeding paper.
\[pcrivtt\] For $M\in {\operatorname{Obj}}{\underline{C}}$ the following conditions are equivalent.
\(i) $M\in {\underline{C}}_{w\ge 0}$.
\(ii) $H(M)=0$ for any cohomological $H$ from ${\underline{C}}$ into (an abelian category) ${\underline{A}}$ that is of weight range $\le -1$.
\(iii) $(\tau^{\ge 1}H_N)(M) ={\{0\}}$ for any $N\in {\operatorname{Obj}}{\underline{C}}$.
\(iv) $(\tau^{\ge 1}H_M)(M) ={\{0\}}$.
Condition (i) implies condition (ii) by definition; certainly, (iii) $\implies$ (iv). Next, condition (ii) implies condition (iii) according to Proposition \[pwrange\](\[iwrvt\]).
Now we prove that condition (iv) implies that $M\in {\underline{C}}_{w\ge 0}$. We consider the commutative diagram $$\label{epr} \begin{CD} w_{\le -1} M@>{c_{-1}}>> M@>{}>> w_{\ge 0}M\\
@VV{h}V@VV{{\operatorname{id}}_M}V@ VV{}V \\
w_{\le 0} M@>{c_0}>> M@>{}>> w_{\ge 1}M \end{CD}$$ given by Proposition \[pbw\](\[icompl\])) (we take $g={\operatorname{id}}_M$ in the proposition). Since $(\tau^{\ge 1}H_M)(M) ={\{0\}}$, $c_0\circ h=0$; hence $c_{-1}=0$ also. Hence the upper distinguished triangle in (\[epr\]) yields that $M$ is a retract of $w_{\ge 0}M\in {\underline{C}}_{w\ge 0}$.
On the relation of smashing weight structures to cc and cp functors {#sprcoprod}
-------------------------------------------------------------------
We prove a collection of properties of smashing weight structures (see Definition \[dhop\]).
\[ppcoprws\] Let $w$ be a smashing weight structure (on ${\underline{C}}$), $i,j\in {{\mathbb{Z}}}$; let ${\underline{A}}$ be an AB4 abelian category, and ${\underline{A}}'$ be an AB4\* abelian category. Then the following statements are valid.
1. \[icopr1\] The classes ${\underline{C}}_{w\le j}$, ${\underline{C}}_{w\ge i}$, and ${\underline{C}}_{[i,j]}$ are closed with respect to ${\underline{C}}$-coproducts.
2. \[icoprhw\] In particular, the category ${{\underline{Hw}}}$ is closed with respect to ${\underline{C}}$-coproducts, and the embedding ${{\underline{Hw}}}\to {\underline{C}}$ respects coproducts.
3. \[icopr2\] Coproducts of weight decompositions are weight decompositions.
4. \[icopr3\] Coproducts of weight Postnikov towers are weight Postnikov towers.
5. \[icopr4\] The categories ${{\underline{C}}_w}$ and ${K_{\mathfrak{w}}}({{\underline{Hw}}})$ are closed with respect to coproducts, and the functor $t$ respects coproducts.
6. \[icopr5\] Pure functors ${\underline{C}}\to {\underline{A}}$ respecting coproducts are exactly the functors of the form $H^{{\mathcal{A}}}$ (see Proposition \[ppure\]), where ${{\mathcal{A}}}:{{\underline{Hw}}}\to {\underline{A}}$ is an additive functor respecting coproducts. Moreover, this correspondence is an equivalence of (possibly, big) categories.
7. \[icopr6\] If $H:{\underline{C}}\to {\underline{A}}$ is a cc functor (see Definition \[dcomp\](\[idcc\])) then $\tau^{\ge i }(H)$ and $\tau^{\le i }(H)$ are cc functors also.
8. \[icopr6p\] $H':{\underline{C}}^{op}\to {\underline{A}}'$ is a cp functor then $\tau^{\ge i }(H')$ and $\tau^{\le i }(H')$ are cp functors also.
9. \[icopr5p\] Pure cohomological (see Definition \[drange\](2)) cp functors from ${\underline{C}}$ into ${\underline{A}}'$ are exactly those of the form $H_{{\mathcal{A}}}$ for ${{\mathcal{A}}}:{{\underline{Hw}}}^{op}\to {\underline{A}}'$ being an additive functor that sends ${{\underline{Hw}}}$-coproducts into products.
10. \[icopr7\] Let ${\underline{D}}$ be the localizing subcategory of ${\underline{C}}$ generated by a class of objects $\{D_l\}$, and assume that for any of the $D_l$ a choice of (the terms of) its weight complex $t(D_l)=(D_l^k)$ is fixed. Then any element of ${\underline{C}}_{w=0}\cap {\operatorname{Obj}}{\underline{D}}$ is a retract of a coproduct of a family of $D_l^k$.
11. \[icopr7p\] For $\{D_l\}$ and ${\underline{D}}$ as in the previous assertions assume that for any of the $D_l$ a choice of $w_{\le k}D_l$ and of $w_{\ge k}D_l$ for $k\in {{\mathbb{Z}}}$ is fixed (we do not assume any relation between these choices). Then for any $D\in{\operatorname{Obj}}{\underline{D}}$ and any $m\in {{\mathbb{Z}}}$ there exists a choice of $(w_{\le m}D)[-m]$ (resp. of $(w_{\ge m}D)[-m]$) belonging to the smallest coproductive extension-closed subclass $D_1$ (resp. $D_2$) of ${\operatorname{Obj}}{\underline{C}}$ containing $(w_{\le k}D_l)[-k]$ (resp. $(w_{\ge k}D_l)[-k]$) for all $l$ and all $k\in{{\mathbb{Z}}}$. Moreover, ${\underline{C}}_{w\le 0}\cap {\operatorname{Obj}}{\underline{D}}$ and ${\underline{C}}_{w\ge 0}\cap {\operatorname{Obj}}{\underline{D}}$ lie in $D_1$ and $D_2$, respectively. Furthermore, if ${\alpha}$ is a regular cardinal and $D$ belongs to the smallest triangulated category of ${\underline{C}}$ that contains $D_l$ and closed under ${\underline{C}}$-coproducts of less than ${\alpha}$ objects then all $(w_{\le m}D)[-m]$ can be chosen to belong to the smallest extension-closed subclass of ${\operatorname{Obj}}{\underline{C}}$ containing $(w_{\le k}D_l)[-k]$ and closed under coproducts of less than ${\alpha}$ objects.
<!-- -->
1. This is essentially a particular case of Proposition \[phop\](2,3); see Remark \[rwhop\](1).
2. Immediate from the previous assertion.
3. Recalling Remark \[rwhop\](1) once again, we reduce the statement to Proposition \[phop\](4).
4. Immediate from the previous assertion.
5. Immediate from assertions \[icopr1\] and \[icopr3\].
6. It certainly follows from Proposition \[pwrange\](\[iwrpure\]) that pure functors are exactly those of the type $H^{{\mathcal{A}}}$. Since ${{\mathcal{A}}}$ is the restriction of $H^{{\mathcal{A}}}$ to ${{\underline{Hw}}}$, this (restriction) correspondence is functorial. Next, if $H$ respects coproducts then its restriction to ${{\underline{Hw}}}$ also does according to assertion \[icoprhw\]. Conversely, if ${{\mathcal{A}}}$ respects coproducts then $H^{{\mathcal{A}}}$ also does according to assertion \[icopr4\].
\[icopr6\], \[icopr6p\]. Immediate from assertion \[icopr2\].
\[icopr5p\]. Similarly to assertion \[icopr5\], ${{\mathcal{A}}}$ is the restriction of $H_{{\mathcal{A}}}$ to ${{\underline{Hw}}}$; hence it sends coproducts into products according to assertion \[icoprhw\]. Conversely, if ${{\mathcal{A}}}$ sends coproducts into products then $H^{{\mathcal{A}}}$ also does according to assertion \[icopr4\].
\[icopr7\]. Combining assertion \[icopr4\] with Proposition \[pwt\](\[iwcex\]) we obtain that for any $M\in {\operatorname{Obj}}{\underline{D}}$ there exists a choice of $t(M)$ all of whose terms are coproducts of $D_l^k$. If $M$ also belongs to ${\underline{C}}_{w=0}$ then we obtain that $M$ is a retract of an element of this form according to Remark \[rwc\].
\[icopr7p\]. Combining assertion \[icopr2\] with Proposition \[pbw\](\[iwdext\]) we obtain that the class $C$ of those $D\in {\operatorname{Obj}}{\underline{C}}$ such that for any $m\in {{\mathbb{Z}}}$ there exist a choice of $(w_{\le m}D)[-m]$ and of $(w_{\ge m}D)[-m]$ belonging to $D_1$ (resp. to $D_2$) is a coproductive class of objects of a full triangulated subcategory of ${\underline{C}}$ (i.e., there exists a triangulated ${\underline{C}'}\subset {\underline{C}}$ such that ${\operatorname{Obj}}{\underline{C}'}=C$ and $C$ is coproductive). Thus $C$ contains ${\operatorname{Obj}}{\underline{D}}$.
Next, if $d$ belongs to $ {\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}_{w\le 0}$ (resp. to $ {\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}_{w\ge 0}$) then the existence of $w_{\le 0}M$ belonging to $D_1$ (resp. of $w_{\ge 0}M$ belonging to $D_2$) implies that $d$ belongs to the Karoubi-closure of $D_1$ (resp. of $D_2$) according to Proposition \[pbw\](\[iwdmod\]). Thus to prove the “moreover” part of the assertion it remains to note that $D_1$ and $D_2$ are Karoubian according to Remark \[rcoulim\](4) below.
The proof of the “furthermore” part of the assertion is similar.
\[ral\] In all the part of our proposition one can replace arbitrary small coproducts by coproducts of less than ${\alpha}$ objects in all occurences (where ${\alpha}$ is any regular infinite cardinal). In particular, in assertions \[icopr4\] and \[icopr6p\] one can assume that ${\underline{C}}$ and ${\underline{C}}_{w\le 0}$ are closed with respect to ${\underline{C}}$-coproducts of less than ${\alpha}$ of their objects; then $t$ respects these coproducts also, and virtual $t$-truncations of cohomological functors that convert coproducts of less than ${\alpha}$ objects into the corresponding products fulfil this condition as well.[^32]
Part \[icopr6p\] of our proposition immediately implies the following corollary that will be important for us below.
\[cvttbrown\] Assume that $w$ is smashing.
I.1. If ${\underline{C}}$ satisfies the Brown representability condition (see Definition \[dcomp\](\[idbrown\])) then virtual $t$-truncations of representable functors are representable.
2\. If ${\underline{C}}$ is generated by a set of objects as its own localizing subcategory[^33] then ${{\underline{Hw}}}$ has a generator, i.e., there exists $P\in {\underline{C}}_{w=0}$ such that any object of ${{\underline{Hw}}}$ is a retract of a coproduct of (copies of) $P$.
II\. Let $F:{\underline{C}}\to {\underline{C}}'$ be an exact functor respecting coproducts. Adopt the notation of Proposition \[ppcoprws\](\[icopr7p\]) and let $w'$ be a weight structure for ${\underline{C}}'$;
1\. Then $F$ is left (resp. right) weight-exact if and only if $F(w_{\le k}D_l)[-k]\in {\underline{C}}'_{w'\le 0}$ (resp. $F(w_{\ge k}D_l)[-k]\in {\underline{C}}'_{w'\ge 0}$) for all $l$ and all $k\in {{\mathbb{Z}}}$.
2\. Assume that all $D_l$ are $w$-bounded; let ${\mathcal{P}}\subset {\underline{C}}_{w=0}$ be a generating class for ${{\underline{Hw}}}$ (i.e., any object of ${{\underline{Hw}}}$ is a retract of a coproduct of elements of ${\mathcal{P}}$). Then $F$ is left (resp. right) weight-exact if and only if $F({\mathcal{P}})\subset {\underline{C}}'_{w'\le 0}$ (resp. $F({\mathcal{P}})\subset {\underline{C}}'_{w'\ge 0}$).
I.1. The Brown representability condition says that ${\underline{C}}$-representable functors are precisely all cp functors from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$. Hence the statement follows from Proposition \[ppcoprws\](\[icopr6p\]) indeed.
2\. Just take ${\underline{D}}={\underline{C}}$ in part \[icopr7\] of the proposition; then we can take $P$ being the coproduct of the corresponding $D_l^k$ (recall that ${{\underline{Hw}}}$ has coproducts!).
II\. If $F$ is left (resp. right) weight-exact then the images of all elements of ${\underline{C}}_{w\le 0}$ (resp. of ${\underline{C}}_{w\ge 0}$) belong to ${\underline{C}}'_{w'\le 0}$ (resp. to ${\underline{C}}'_{w'\ge 0}$), and we obtain “one half” of the implications in question.
The converse implication for assertion II.1 follows from Proposition \[ppcoprws\](\[icopr7p\]) immediately.
Now we check the converse implication for assertion II.2. Since $F$ respects coproducts and $F({\mathcal{P}})\subset {\underline{C}}'_{w'\le 0}$ (resp. $F({\mathcal{P}})\subset {\underline{C}}'_{w'\ge 0}$), we obtain $F({\underline{C}}_{w=0})\subset {\underline{C}}'_{w'\le 0}$ (resp. $F({\underline{C}}_{w=0})\subset {\underline{C}}'_{w'\ge 0}$). Thus Proposition \[pbw\](\[igenlm\]) yields the result easily.
\[rgenw\] 1. Note that our definition of a generator for ${{\underline{Hw}}}$ is much more “restrictive” than the assumption that $\{P\}$ Hom-generates ${{\underline{Hw}}}$ (cf. Remark \[rcompgen\]) and even than the (usual) “abelian version” of the definition of generators (cf. Theorem \[tab5\](1)). Moreover, it is much easier to specify generators in ${{\underline{Hw}}}$ (if $w$ is smashing) than in ${{\underline{Ht}}}$. 2. A certain “finite dimensional” analogue of part I.1 of this corollary is given by Proposition \[psatur\](1) below.
On adjacent weight and $t$-structures and Brown representability-type conditions {#sadjbrown}
================================================================================
In this section we study some general conditions ensuring that a torsion pair admits a (right or left) torsion pair.
In §\[sadjt\] we prove that a weight structure admits a right adjacent $t$-structure if and only if the virtual $t$-truncations of representable functors are representable. It easily follows that in a triangulated categories satisfying the Brown representability condition a weight structure admits a right adjacent $t$-structure if and only if it is smashing. Certainly, the dual to this statement is also true. Moreover, a similar argument demonstrates that if the representable functors from an $R$-linear category ${\underline{C}}$ are precisely the [*$R$-finite type*]{} ones (i.e., if ${\underline{C}}$ is [*$R$-saturated*]{}) then all bounded weight structures on ${\underline{C}}$ admit right adjacent $t$-ones. Note here that if $R$ is Noetherian then for $X$ being a regular separated finite-dimensional scheme that is proper over ${{\operatorname{Spec}\,}}R$ its bounded derived category of coherent sheaves (as well as its dual) is $R$-saturated according to a recent result of Neeman. In §\[sadjw\] we study when a $t$-structure $t$ admits a (left or right) adjacent weight structure $w$; however, the results of this section are “not as nice” as their “mirror” ones in §\[sadjt\] (at least, in the case where ${\underline{C}}$ has coproducts; cf. Remark \[rnondeg\]).
In §\[scomp\] we recall the notions of perfectly generated and well generated triangulated categories along with their (Brown representability) properties; we also relate perfectness to smashing torsion pairs. Next we define symmetric classes and study their relation to perfect classes, Brown-Comenetz duality, and adjacent torsion pairs (obtaining a new criterion for the existence of the latter). This gives one more “description” of a $t$-structure that is right adjacent to a given compactly generated weight structure.
On the existence of adjacent $t$-structures {#sadjt}
-------------------------------------------
\[phadj\] Let $w$ be left adjacent to a $t$-structure $t$ on ${\underline{C}}$. Then ${{\underline{Ht}}}$ is a full exact subcategory of the (possibly, big) abelian category $\operatorname{\operatorname{AddFun}}({{\underline{Hw}}}^{op},{\underline{\operatorname{Ab}}})$ (see §\[snotata\]).
Moreover, ${\underline{C}}_{w=0}=P_t$ and the functor $H^P={\underline{C}}(P,-)$ is isomorphic to ${{\underline{Ht}}}(H_0^t(P), H_0^t(-))$ for any $P\in {\underline{C}}_{w=0}$.
The first part of the assertion is given by part 4 of [@bws Theorem 4.4.2].
The equality ${\underline{C}}_{w=0}=P_t$ is immediate from Remark \[rtst1\](\[it1\]) (recall Definition \[dwso\](\[idadj\])).
The last of the assertions is given by Proposition \[pgen\](\[ipgen2\]).
As we have essentially already noted (see Proposition \[phop\](5)), if a weight structure possesses a left adjacent $t$-structure then it is coproductive (i.e., the associated torsion pair is coproductive). Now we prove that the converse implication is valid also if ${\underline{C}}$ satisfies the Brown representability property (in particular, if it is compactly generated or [*perfectly generated*]{} in the sense of Definition \[dwg\](\[idpc\]) below).
\[tadjt\] Assume that ${\underline{C}}$ satisfies the Brown representability condition. 1. Then for a weight structure $w$ on ${\underline{C}}$ there exists a $t$-structure right adjacent to it if and only if $w$ is smashing. 2. If a right adjacent $t$ exists then ${{\underline{Ht}}}$ is equivalent to the full subcategory of $\operatorname{\operatorname{AddFun}}({{\underline{Hw}}}^{op},{\underline{\operatorname{Ab}}})$ consisting of those functors that sends ${{\underline{Hw}}}$-coproducts into products.
1\. The “only if” assertion is immediate from Proposition \[phop\](5) (and very easy for itself).
Conversely, assume that $w$ is smashing. According to Proposition \[padjt\](\[ile4\]) below it suffices to verify that for any $M\in {\operatorname{Obj}}{\underline{C}}$ the functor $\tau^{\le 0}H_M$ is representable (in ${\underline{C}}$; recall that $H_M$ denotes the functor ${\underline{C}}(-,M)$). The latter statement is given by Corollary \[cvttbrown\](I.1).
2\. Applying Proposition \[phadj\] we obtain that is suffices to find out which functors ${{\underline{Hw}}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ are represented by objects of ${{\underline{Ht}}}$. Since the embedding ${{\underline{Hw}}}\to {\underline{C}}$ respects coproducts (see Proposition \[ppcoprws\](\[icopr1\])), all these functors send ${{\underline{Hw}}}$-coproducts into ${\underline{\operatorname{Ab}}}$-products.
Conversely, let ${{\mathcal{A}}}:{{\underline{Hw}}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ be an additive functor converting coproducts into products. Then the corresponding $H_{{\mathcal{A}}}$ is a cp functor (see Definition \[dcomp\](\[idcc\])) according to Proposition \[ppcoprws\](\[icopr5p\]). Hence it is representable by some $M\in {\operatorname{Obj}}{\underline{C}}$. Since $H_{{\mathcal{A}}}$ is also of weight range $[0,0]$ (see Proposition \[pwrange\](\[iwrpure\])), we have $M\in {\underline{C}}^{t=0}$.
\[rstable\]
1\. Proposition \[pwsym\](\[iwsymcgwt\]) below gives some more information on the adjacent weight structure $t$ whenever $w$ is compactly generated (see Remark \[rwhop\](1)). 2. Recall from Proposition \[prtst\](\[itp4\], \[it4sm\]) that “shift-stable” weight structures are in one-to-one correspondence with exact embeddings $i:L\to {\underline{C}}$ possessing right adjoints. Hence applying our theorem in this case we obtain the following: if $i$ possesses a right adjoint respecting coproducts and ${\underline{C}}$ satisfies the Brown representability condition then for the full triangulated subcategory $R$ of ${\underline{C}}$ with ${\operatorname{Obj}}R=L{{}^{\perp}}$ the embedding $R\to {\underline{C}}$ possesses a right adjoint also. Thus $R$ is [*admissible*]{} in ${\underline{C}}$ in the sense of [@bondkaprserr] and the embedding $R\to{\underline{C}}$ may be completed to a [*gluing datum*]{} (cf. [@bbd §1.4] or [@neebook §9.2]). So we re-prove Corollary 2.4 of [@nisao].
\[cdualt\] Let ${\underline{C}}$ be a category satisfying the dual Brown representability property (recall that this is the case if ${\underline{C}}$ is compactly generated; see Proposition \[pcomp\](II.1)). 1. Then for a weight structure $w$ on ${\underline{C}}$ there exists a $t$-structure left adjacent to it if and only if $w$ is cosmashing. 2. If these equivalent conditions are fulfilled then ${{\underline{Ht}}}$ is anti-equivalent to the subcategory of $\operatorname{\operatorname{AddFun}}({{\underline{Hw}}},{\underline{\operatorname{Ab}}})$ consisting of those functors that respect products.
This is just the categorical dual to Theorem \[tadjt\].
Now we re-formulate the existence of a $t$-structure right adjacent to $w$ in terms of virtual $t$-truncations; this finishes the proof of Theorem \[tadjt\].
\[padjt\] Let $w$ be a weight structure for ${\underline{C}}$, $M\in {\operatorname{Obj}}{\underline{C}}$, and assume that for the functor $H_M={\underline{C}}(-,M)$ its virtual $t$-truncation $\tau^{\le 0}H_M$ is represented by some object $M^{\le 0}$ of ${\underline{C}}$.
Then the following statements are valid.
1. \[ile1\] $M^{\le 0}$ belongs to ${\underline{C}}_{w\ge 0}$.
2. \[ile2\] The natural transformation $\tau^{\le -n }(H_M)\to H_M$ mentioned in (\[evtt\]) is induced by some $f\in {\underline{C}}(M^{\le 0},M)$.
3. \[ile3\] The object $\operatorname{\operatorname{Cone}}(f)$ belongs to ${\underline{C}}_{w\ge 0}^{\perp}$.
4. \[ile4\] There exists a $t$-structure $t$ (on ${\underline{C}}$) right adjacent to $w$ if and only if the functor $\tau^{\le 0}H_{M'}$ is ${\underline{C}}$-representable for any object $M'$ of ${\underline{C}}$.
5. \[ile5\] For $M'\in {\operatorname{Obj}}{\underline{C}}$ the representability of the functor $\tau^{\le 0}H_{M'}$ is equivalent to that of $\tau^{\ge 1}H_{M'}$.
1\. $\tau^{\le 0}H_M$ is of weight range $\ge 0$ (see Proposition \[pwrange\](\[iwrvt\])). Hence the assertion follows from part \[iwfil3\] of the same proposition.
2\. Immediate from the Yoneda lemma.
3\. For any $N\in {\operatorname{Obj}}{\underline{C}}$ applying the functor $H^N={\underline{C}}(N,-)$ to the distinguished triangle $M^{\le 0}\to M \to M^{\ge 1} \to M^{\le 0}[1]$ one obtains a long exact sequence that yields the following short one: $$\label{eshort}
\begin{gathered} 0\to \operatorname{\operatorname{Coker}}(H^N(M^{\le 0})\stackrel{h^1_N}{\to} H^N(M))\to H^N(\operatorname{\operatorname{Cone}}(f)) \\
\to \operatorname{\operatorname{Ker}}(H^N(M^{\le 0}[1])\stackrel{h^2_N}{\to} H^N(M[1]))\to 0.
\end{gathered}$$ So, for any $N\in {\underline{C}}_{w\ge 0}$ we should check that $h^1_N$ is surjective and $h^2_N$ is injective.
Applying (\[evtt\]) to the functor $H_M$ (in the case $n=0$) we obtain a long exact sequence of functors $$\label{evttp}
\dots\to \tau^{\le 0 }(H_M)\to H_M\to \tau^{\ge 1}(H_M)\to \tau^{\le 0}(H_M)\circ [1]\to H_M\to \dots$$ Applying this sequence of functors to $N$ we obtain that the surjectivity of $h^1_N$ along with the injectivity of $h^2_N$ is equivalent to $\tau^{\ge 1}(H_M)(N)={\{0\}}$. So, recalling Proposition \[pwrange\](\[iwrvt\]) once again (to obtain that $\tau^{\ge 1}(H_M)$ is of weight range $\le -1$) we conclude the proof.
4\. The “only if” part of the assertion is immediate from Proposition \[pwfil\](4).
To prove the converse implication we should check that the couple $({\underline{C}}_{w\ge 0}, {\underline{C}}_{w\ge 0}^{\perp}[1])$ is a $t$-structure if our representability assumption is fulfilled. It is easily seen that the only non-trivial axiom check here is the existence of $t$-decompositions (see Definition \[dtstr\]), which is given by the previous assertion.
5\. If $\tau^{\le 0}H_{M'}$ is representable then the previous assertions imply the existence of a distinguished triangle $M'^{\le 0}\to M' \to M'^{\ge 1} \to M'^{\le 0}[1]$ with $M'\in {\underline{C}}_{w\ge 0}^{\perp}$. Then the object $M'^{\ge 1} $ represents the functor $\tau^{\ge 1}H_{M'}$ according to Theorem 2.3.1(III.4) of [@bger] (and so, $\tau^{\ge 1}H_{M'}$ is representable). The proof of the converse implication is similar.
Now we describe one more application of Proposition \[padjt\]. It relies on a modified version of the Brown representability property that we will now define.
\[dsatur\] Let $R$ be an associative commutative unital ring, and assume that ${\underline{C}}$ is $R$-linear.
1\. We will say that an $R$-linear cohomological functor $H$ from ${\underline{C}}$ into $R-{\operatorname{Mod}}$ is [*of $R$-finite type*]{} whenever for any $M\in {\operatorname{Obj}}{\underline{C}}$ the $R$-module $H(M)$ is finitely generated and $H(M[i])={\{0\}}$ for almost all $i\in {{\mathbb{Z}}}$.
2\. We will say that ${\underline{C}}$ is [*$R$-saturated*]{} if the representable functors from ${\underline{C}}$ are exactly all the $R$-finite type ones.
3\. The symbol $\operatorname{\operatorname{AddFun}_R}(C,D)$ will denote the (possibly, big) category of $R$-linear (additive) functors from $C$ into $D$ whenever $C$ and $D$ are $R$-linear categories.
4\. We will write $R-{\operatorname{mod}}$ for the category of finitely generated $R$-modules.
\[psatur\] Assume that ${\underline{C}}$ is $R$-linear and endowed with a bounded weight structure $w$.
I. Then all virtual $t$-truncations of functors of $R$-finite type are of $R$-finite type also.
II\. Assume in addition that ${\underline{C}}$ is $R$-saturated. Then the following statements are valid.
1\. For any $i\in {{\mathbb{Z}}}$ and $M\in {\operatorname{Obj}}{\underline{C}}$ the functors $\tau^{\le -i }(H_M)$ and $\tau^{\ge - i }(H_M)$ are representable.
2\. There exists a $t$-structure right adjacent to $w$.
3\. Its heart ${{\underline{Ht}}}$ naturally embeds into the category $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}{{^{op}}},R-{\operatorname{mod}})$. This embedding is essentially surjective whenever $R$ is noetherian.
I. Recall that virtual $t$-truncations (of cohomological functors) are cohomological. Moreover, virtual $t$-truncations of $R$-linear functors are obviously $R$-linear.
Now, let $H$ be a functor of $R$-finite type. It obviously follows that the values of $\tau^{\ge 0}(H)$ are finitely generated $R$-modules. Next, for any $N\in {\operatorname{Obj}}{\underline{C}}$ we can take $w_{\ge 0}(N[j])$ being $0$ for $j$ small enough and being equal to $N[j]$ for $j$ large enough (recall that $w$ is bounded); hence the functor $\tau^{\ge 0}(H)$ is of $R$-finite type. Applying this argument to $H\circ [-i]$ (for $i\in {{\mathbb{Z}}}$) we obtain that the functor $\tau^{\ge - i}(H)$ is of $R$-finite type also. The proof for $\tau^{\le - i }(H)$ is similar.
II.1. Immediate from assertion I combined with the definition of saturatedness. 2. According to Proposition \[padjt\](\[ile4\]), the assertion follows from the previous one.
3\. Certainly, restricting functors of $R$-finite type from ${\underline{C}}$ to ${{\underline{Hw}}}$ gives functors of the type described. This restriction gives an embedding of ${{\underline{Ht}}}$ into $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}{{^{op}}},R-{\operatorname{mod}})$ according to Proposition \[phadj\]. Lastly, if a functor $A$ belongs to $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}{{^{op}}},R-{\operatorname{mod}})$ and $R$ is noetherian then the pure functor $H_{{\mathcal{A}}}$ (see Definition \[drange\](2)) is easily seen to be of $R$-finite type (as a cohomological functor from ${\underline{C}}$ into $R-{\operatorname{mod}}$), whereas the object representing it belongs to ${\underline{C}}^{t=0}$ according to Proposition \[pwrange\](\[iwrpure\]).
\[rsatur\] 1. Note now that Corollary 0.5 of [@neesat] easily implies that ${\underline{C}}$ is $R$-saturated whenever $R$ is a noetherian ring, $X$ is a regular separated finite-dimensional scheme that is proper over ${{\operatorname{Spec}\,}}R$, and ${\underline{C}}=D^b(X)$ (the bounded derived category of coherent sheaves). Indeed, the assumptions on $X$ imply that in this case the derived category of perfect complexes of sheaves equals ${\underline{C}}$, and then loc. cit. gives the result immediately (cf. Remark \[roq\](1) below). Moreover, in this case ${\underline{C}}\cong {\underline{C}}{{^{op}}}$ (since there exists a dualizing complex on $X$). Thus $t$-structures left adjacent to bounded weight structures on $D^b(X)$ exist also.
We recall also that in the case where $R$ is a field the saturatedness is question is given by Corollary 3.1.5 of [@bvdb]. Moreover, Theorem 4.3.4 of ibid. is a certain a “non-commutative geometric” analogue of this statement.
2\. Now we discuss possible weight structures on the category ${\underline{C}}=D^b(X)$ (as above).
We recall that bounded weight structures for ${\underline{C}}$ are determined by their hearts (see Proposition \[pbw\](\[igenlm\])), whereas the latter are precisely all the negative additive Karoubi-closed subcategories of ${\underline{C}}$ that densely generate it (see Corollary 2.1.2 of [@bonspkar]).
Moreover, the author suspects that in the “geometric” examples mentioned above there exists a single “dense” generator $P$ of ${{\underline{Hw}}}$, i.e., all objects of ${{\underline{Hw}}}$ are retracts of (finite) powers of $P$.
One can construct a rich family of bounded weight structures on $D^b(X)$ at least in the case where $X={\mathbb{P}}^n$ (for some $n>0$; the author does not know whether it is necessary to assume that $R$ is a field here) since one can use gluing for constructing weight structures (cf. [@bws §8.2]). More generally, it suffices to assume that $D^b(X)$ possesses a [*full exceptional collection*]{} of objects.
Note however that degenerate weight structures are certainly possible in triangulated categories of this type (cf. Proposition \[prtst\](\[itp4\])); so, these weight structures are not bounded. Still the author conjectures that the boundedness restriction is not actually necessary (the proof may rely on the existence of a [*strong generator*]{} in the sense of [@bvdb] for ${\underline{C}}$).
3\. The only examples of $R$-saturated triangulated categories for a non-noetherian $R$ known to the author are direct sums of $R/J_i$-saturated categories, where $\{J_i\}$ is a finite collection of ideals of $R$ and all $R/J_i$ are noetherian. So, a general $R$ was mentioned in the proposition just for the sake of generality. Our definition of $R$-saturatedness may be “not optimal”; it could make sense to put finitely presented modules instead of finitely generated ones into the definition. Note however that any length two ${{\underline{Hw}}}$-complex is a weight complex of some object of ${\underline{C}}$. This allows to describe ${{\underline{Ht}}}$ completely (cf. Proposition \[psatur\](II.3)) for any possible definition of $R$-saturatedness (and can possibly help choosing among the definitions).
4\. Actually, all of the statements of this paper may easily be proved in the $R$-linear context (cf. [@zvon]). This is formally a generalization (since one can take $R={{\mathbb{Z}}}$); yet Propositions \[psatur\] and \[psaturdu\] (along with the examples to them) appear to be the only statements for which this setting is really actual. Note also that below we define Brown-Comenetz duals (of objects and functors) “using” the group ${{\mathbb{Q}}}/{{\mathbb{Z}}}$. However, the only property of this group that we will actually apply is that it is an injective cogenerator of the category ${\underline{\operatorname{Ab}}}$. Hence for an $R$-linear category ${\underline{C}}$ one can replace our Definition \[dsym\](\[ibcomf\],\[ibcomo\]) below by any its “$R$-linear analogue”; this may make sense if $R$ is a field.
On adjacent weight structures {#sadjw}
-----------------------------
In this subsection we will assume that ${\underline{C}}$ is endowed with a $t$-structure $t$. For the construction of certain weight structures we will need the following statement.
\[lconstws\] Assume that certain extension-closed classes ${\underline{C}}_-$ and ${\underline{C}}_+$ of objects of ${\underline{C}}$ satisfy the axioms (i)—(iii) of Definition \[dwstr\] (for ${\underline{C}}_{w\le 0}$ and $ {\underline{C}}_{w\ge 0}$, respectively). Let us call a ${\underline{C}}$-distinguished triangle $X\to M\to Y[1]$ a [*pre-weight decomposition*]{} of $M$ if $X$ belongs to ${\underline{C}}_-$ and $Y$ belongs to ${\underline{C}}'_+$.
Then the following statements are valid.
1\. The class $C$ of objects possessing pre-weight decompositions is extension-closed (in ${\underline{C}}$).
2\. Assume that $C$ contains a subclass ${\mathcal{P}}$ such that ${\mathcal{P}}={\mathcal{P}}[1]$. Then $C$ also contains the object class of the smallest strict subcategory of ${\underline{C}}$ containing ${\mathcal{P}}$. 3. Assume that ${\underline{C}}$ has coproducts, and ${\underline{C}}_-$ and ${\underline{C}}_+$ are coproductive. Then $C$ is coproductive also.
1\. Immediate from Theorem 2.1.1(I.1) of [@bonspkar] (cf. also Remark 1.5.5(1) of [@bws]).
2\. Immediate from assertion 1.
3\. Once again, it suffices to recall Proposition \[pcoprtriang\].
Now we prove a simple statement on the existence of $w$ that is left adjacent to $t$.
\[pconstrwfromt\] I. Assume that there exists a weight structure $w$ left adjacent to $t$. Then for any $M\in {\underline{C}}^{t=0}$ there exists an ${{\underline{Ht}}}$-epimorphism from the ${{\underline{Ht}}}$-projective object $H_0^t(P)$ into $ M$ for some $P\in P_t$, and the functor $H_0^t$ induces an equivalence of $\operatorname{\operatorname{Kar}}({{\underline{Hw}}})$ with the category of projective objects of ${{\underline{Ht}}}$. II. The converse implication is valid under any of the following additional assumptions.
1\. $t$ is bounded above (see Definition \[dtstr\]).
2\. There exists an integer $n$ such that ${\underline{C}}^{t\ge n}\perp {\underline{C}}^{t\le 0}$.
I. Fix $M\in {\underline{C}}^{t=0}$ and consider its weight decomposition $P\stackrel{p}{\to} M\to M'\to P[1]$. Since $M\in {\underline{C}}^{t\le 0}={\underline{C}}_{w\ge 0}$, we have $P\in {\underline{C}}_{w=0}$ according to Proposition \[pbw\](\[iwd0\])). Next, since $P\in {\underline{C}}_{w\ge 0}={\underline{C}}^{t\le 0}$, the object $P_0=t^{\ge 0}P$ equals $H_0^t(P)$; hence $P_0$ is projective in ${{\underline{Ht}}}$ according to Proposition \[pgen\](\[ipgen25\]). The adjunction property for the functor $t^{\ge 0}$ (see Remark \[rtst1\](\[it3\])) implies that $p$ factors through the $t$-decomposition morphism $P\to P_0$. Now we check that the corresponding morphism $P_0\to M$ is an ${{\underline{Ht}}}$-epimorphism. This is certainly equivalent to its cone $C$ belonging to ${\underline{C}}^{t\le -1}$. The octahedral axiom of triangulated categories gives a distinguished triangle $(t^{\le -1}P)[1]\to M'\to C\to (t^{\le -1}P)[2]$; it yields the assertion in question since $M'\in {\underline{C}}_{w\ge 1}={\underline{C}}^{t\le -1}$ and the class ${\underline{C}}^{t\le -1}$ is extension-closed.
Next, the category of projective objects of ${{\underline{Ht}}}$ is certainly Karoubian. According to Proposition \[phadj\] it remains to verify that for any projective object $Q$ of ${{\underline{Ht}}}$ there exists $R\in P_t$ such that $Q $ is a retract of $ H_0^t(R)$. Now, the first part of the assertion implies the existence of an ${{\underline{Ht}}}$-epimorphism $H_0^t(R')\to Q$ for some $R'\in P_t$. Since $H_0^t(R')$ is projective in ${{\underline{Ht}}}$, this epimorphism splits, i.e., $Q$ equals the image of some idempotent isomorphism of $H_0^t(R')$. Lifting this endomorphism to ${{\underline{Hw}}}$ we obtain the result.
II.1. We set ${\underline{C}}_{w\ge 0}={\underline{C}}^{t\le 0}$ and take ${\underline{C}}_{w\le 0}$ to be the envelope of $\cup_{i<0} P_t[i]$.[^34]We should prove that this couple yields a weight structure for ${\underline{C}}$, since this weight structure would certainly be adjacent to $t$. Now, this “candidate weight structure” obviously satisfies axioms (i) and (ii) in Definition \[dwstr\]. Next, since $P_t[i]\perp {\underline{C}}^{t\le 0}$ for any $i<0$; hence the orthogonality axiom (iii) is fulfilled also.
It remains to verify the existence of a weight decomposition for any $M\in {\underline{C}}^{t\le i}$ by induction on $i$. The statement is obvious for $i< 0$ since $M\in {\underline{C}}_{w\ge 1}={\underline{C}}^{t\le -1}$ and we can take a “trivial” weight decomposition $0\to M\to M\to 0$. Now assume that existence of $w$-decompositions is known for any $M\in {\underline{C}}^{t\le j}$ for some $j\in {{\mathbb{Z}}}$. We should verify the existence of weight decomposition of an element $N$ of ${\underline{C}}^{t\le j+1}$. Certainly, $N$ is an extension of $N'[-j-1]=H_0^t(N[j+1])[-j-1]$ by $t^{\le j}N$ (see Remark \[rtst1\](\[it3\]) for the notation). Since the latter object possesses a weight decomposition, Lemma \[lconstws\](1) allows us to verify the existence of a weight decomposition of $N'[-j-1]$ (instead of $N$). Now we choose a surjection $t^{=0}P\to N'$ whose existence is given by our assumptions. Then a cone $C $ of the corresponding composed morphism $P\to N'$ is easily seen to belong to $t^{\le -1}{\underline{C}}$. Since both $P$ and $C$ possess weight decompositions, applying loc. cit. once again we obtain the assertion in questions.
2\. We take ${\underline{C}}_{w\ge 0}={\underline{C}}^{t\le 0}$ and ${\underline{C}}_{w\le 0}={}^{\perp}{\underline{C}}^{t\le -1}$. Once again it suffices to verify the existence of a $w$-decomposition for an object $M$ of ${\underline{C}}$.
We consider the (full) triangulated subcategory ${\underline{C}}'$ consisting of $t$-bounded below objects, i.e., ${\operatorname{Obj}}{\underline{C}}'=\cup_{i\in {{\mathbb{Z}}}}{\underline{C}}^{t\le i}$. According to the previous assertion, any object of ${\underline{C}}'$ possesses a weight decomposition with respect to the corresponding weight structure; thus it also possesses a $w$-decomposition.
Now, the $t$-decomposition of the object $M[n-2]$ yields a presentation of $M$ as extension of an element $M'$ of ${\underline{C}}^{t\ge n-1}$ by an element $M''$ of ${\underline{C}}^{t\le n-2}$. Since $M''\in {\operatorname{Obj}}{\underline{C}}'$, it possesses a $w$-decomposition. Next, our “extra” orthogonality assumption on $t$ yields that $M''\in {\underline{C}}_{w\le 0}$; hence one take the triangle $M''\to M''\to 0\to M''[1]$ as a $w$-decomposition of $M''$. Lastly, applying Lemma \[lconstws\](1) once again we obtain that $M$ possesses a $w$-decomposition also.
\[rexenproj\] 1. Assume that the category ${\underline{C}}{{^{op}}}$ is $R$-saturated (see Definition \[dsatur\]; in particular, ${\underline{C}}$ may equal the category $D^b(X)$ or $D^b(X){{^{op}}}$ for $X$ being regular separated finite-dimensional scheme that is proper over ${{\operatorname{Spec}\,}}R$ for a Noetherian $R$) and $t$ is a bounded above $t$-structure on ${\underline{C}}$. Then our proposition (combined with Proposition \[phadj\]) easily implies that there exists a weight structure left adjacent to $t$ if and only if ${{\underline{Ht}}}$ has enough projectives (since the corresponding pure functors are corepresented by elements of $P_t$). Moreover, in this case ${{\underline{Hw}}}$ is equivalent to ${\operatorname{Proj}}{\underline{A}}$.
2\. The assumption of the existence of an ${{\underline{Ht}}}$-epimorphism $t^{=0}P\to M$ with $P\in P_t$ for any $M\in {\underline{C}}^{t=0}$ naturally generalizes the condition of the existence of enough projectives that allows to relate the derived category of ${\underline{A}}$ to $K({\operatorname{Proj}}{\underline{A}})$. Note however that in this setting we have ${{\underline{Hw}}}\subset {{\underline{Ht}}}$; this is not the case in general. Moreover, the condition ${\underline{C}}^{t\ge n}\perp {\underline{C}}^{t\le 0}$ for $n\gg 0$ is a natural generalization of the finiteness of the cohomological dimension condition (for an abelian category).
3\. One can easily see that $P_t$ is [*negative*]{} for any $t$. So our existence of $w$ results are closely related to the statements on “constructing $w$ from a negative subcategory”; see §2.2 of [@bsnew], [@bws §4.3,4.5], [@bonspkar Corollary 2.1.2], and Remark \[rsatur\](2) above.
\[tadjw\] Assume that ${\underline{C}}$ has coproducts and is endowed with a $t$-structure $t$ such that its localizing subcategory ${\underline{C}}'$ generated by ${\underline{C}}^{t\le 0}$ satisfies the dual Brown representability condition. 1. Then there exists a weight structure left adjacent to $t$ if and only if $t$ is productive and the category ${{\underline{Ht}}}$ has enough projectives. 2. If such a left adjacent $w$ exists then ${{\underline{Hw}}}$ is equivalent to the subcategory of projective objects of ${{\underline{Ht}}}$.
1\. If $w$ exists then $t$ is productive according to Proposition \[phop\](5). Next, the existence of enough projectives in ${{\underline{Ht}}}$ follows from Proposition \[pconstrwfromt\](I). Conversely, assume that $t$ is productive and the category ${{\underline{Ht}}}$ has enough projectives. Once again, for the “candidates” ${\underline{C}}_{w\ge 0}={\underline{C}}^{t\le 0}$ and ${\underline{C}}_{w\le 0}={{}^{\perp}}({\underline{C}}_{w\ge 0}[1])$ it suffices to verify the existence of a weight decomposition for any $Y\in {\operatorname{Obj}}{\underline{C}}$.
For each projective object $P_0$ of ${{\underline{Ht}}}$ Proposition \[pgen\](\[ipgen4\]) gives the existence of $P\in P_t$ such that $H_0^t(P)\cong P_0$ (here $H^t_0$ is the $t$-homology on ${\underline{C}}$; see Remark \[rtst1\](\[it4\])). Thus the existence of enough projectives in ${{\underline{Ht}}}$ is equivalent to the fact that for any $M\in {\underline{C}}^{t=0}$ there exists an ${{\underline{Ht}}}$-epimorphism $H_0^t(P)\to M$ for $P\in P_t$.
We take ${\underline{C}_0}\subset {\underline{C}}$ being the triangulated category of $t$-bounded below objects (i.e., ${\operatorname{Obj}}{\underline{C}_0}=\cup_{i\in{{\mathbb{Z}}}}{\underline{C}}^{t\le i}$). According to Proposition \[pconstrwfromt\](II.1), there exists a weight structure $w_0$ for ${\underline{C}_0}$ with ${\underline{C}}_{{w_0}\ge 0}={\underline{C}}^{t\le 0}$. Now we study the class $C$ of objects possessing pre-weight decompositions with respect to $w$ in the terms of Lemma \[lconstws\]. The existence of ${w_0}$ certainly implies that $C$ contains ${\operatorname{Obj}}{\underline{C}_0}$. Applying parts 2 and 3 of the lemma we obtain that $C$ actually contains ${\operatorname{Obj}}{\underline{C}'}$. On the other hand, since the class $C'={{}^{\perp}}{\operatorname{Obj}}{\underline{C}'}$ is contained in ${\underline{C}}_{w\le 0}$, $C$ also contains $C'$. According to Lemma \[lconstws\](1), it remains to verify that any object of ${\underline{C}}$ can be presented as an extension of an object of ${\underline{C}}'$ by an element of $C'$. The latter is immediate from Proposition \[pcomp\](II) combined with Proposition \[pbouloc\](III.\[ibou1\]).
2\. Similarly to the proof of Theorem \[tadjt\](II), Proposition \[phadj\] gives an embedding of ${{\underline{Hw}}}$ into the category of projective objects of ${{\underline{Ht}}}$. Hence the arguments used in the proof of assertion 1 (when $P$ was constructed from $P_0$) allow us to conclude the proof.
\[restrt\]
1\. It appears that one is “usually” interested in the case where ${\underline{C}'}={\underline{C}}$. 2. It can be easily seen that in Theorem \[tadjt\] we could have replaced the Brown representability assumption for ${\underline{C}}$ by that for the category ${\underline{C}}'$ being its localizing category generated by ${\underline{C}}_{w\ge 0}$ (and so also by ${\underline{C}_0}=\cup_{i\in {{\mathbb{Z}}}} {\underline{C}}_{w\le i}$).
3\. It is actually not necessary to assume that the whole ${\underline{C}}$ has coproducts when defining ${\underline{C}}'$ (in both of these settings). Indeed, it suffices to assume that ${\underline{C}}^{t\le 0}$ is contained in some triangulated category ${\underline{C}}''\subset {\underline{C}}$ that has coproducts such that the embedding ${\underline{C}}''\to {\underline{C}}$ respects them.
Certainly the dual to Theorem \[tadjw\] is also valid; it is formulated as follows.
\[cadjw\] Assume that ${\underline{C}}$ has products and is endowed with a $t$-structure $t$ such that its colocalizing subcategory ${\underline{C}}'$ cogenerated by ${\underline{C}}^{t\ge 0}$ (see Definition \[dcomp\](\[idloc\])) satisfies the Brown representability condition. 1. Then there exists a weight structure right adjacent to $t$ if and only if $t$ is coproductive and the category ${{\underline{Ht}}}$ has enough injectives. 2. If such a left adjacent $w$ exists then ${{\underline{Hw}}}$ is equivalent to the subcategory of injective objects of ${{\underline{Ht}}}$.
\[rnondeg\] 1. As it often happens when dealing with “large” triangulated categories, the “roles” of Theorem \[tadjw\] and Corollary \[cadjw\] seem to be somewhat different. This “asymmetry” occurs when one tries to apply these statements to a compactly generated (or more generally, [*well generated*]{}; see Definition \[dwg\](\[idpg\]) below) ${\underline{C}}$; recall that these condition are far from being self-dual (see Corollary E.1.3 and Remark 6.4.5 of [@neebook]).
Now, if ${\underline{C}}$ is well generated then it appears to be “quite reasonable” to consider smashing $t$-structures (only); moreover, the existence of enough injectives seems to be a rather “reasonable” restriction on $t$ (see Corollary \[csymt\] below; note however that is not clear how to prove it in general without constructing a right adjacent weight structure [**first**]{}). Yet it may be difficult to check the Brown representability condition for ${\underline{C}}'$ (unless ${\underline{C}'}={\underline{C}}$;[^35] still checking the latter could be difficult also).
On the other hand, it appears that for all “known” cosmashing $t$-structures the left adjacent weight structures can be easily described without using the results of this subsection.
So, both Theorem \[tadjw\] and Corollary \[cadjw\] do not appear to be really “practical”.
2\. One results on the existence of adjacent weight and $t$-structures are certainly “not symmetric”: constructing a (right or left) adjacent $t$-structure is “much easier”. On the other hand, Theorem \[tpgws\] below is a tool of constructing weight structures that appears not to possess a $t$-structure analogue (see Remark \[rigid\](1)). Rather funnily, these two “asymmetries” appear to “compensate” each other. So, the “main general” sorts of weight structures for compactly generated categories that are “easy to construct” are the compactly generated ones (that are smashing) and the (cosmashing) weight structures right adjacent to compactly generated $t$-structures, whereas the main sorts of $t$-structures are the compactly generated ones and the ones right adjacent to compactly generated weight structures (these are smashing and cosmashing, respectively).
3\. Note also that there do exist smashing examples that are not compactly generated (inside a compactly generated ${\underline{C}}$). In [@kellerema] for a ring $R$ along with its two-sided ideal $I$ satisfying certain conditions possesses a right adjoint respecting coproducts but $L$ is not generated by compact objects of ${\underline{C}}$; a family of couples $(R,I)$ satisfying the conditions the following was proved: the embedding $i$ of the localizing subcategory $L$ generated by $I$ into the derived category ${\underline{C}}$ of right $R$-modules proved in Proposition \[prtst\](\[it4sm\]), the corresponding couple $s_L=({\operatorname{Obj}}L,{\operatorname{Obj}}L^{\perp_{{\underline{C}}}})$ is a “shift-stable” torsion pair for ${\underline{C}}$ (cf. Remark \[rtst2\](\[it5s1\]); so, it is a weight structure and a $t$-structure simultaneously); still it is certainly not compactly generated.
One can probably obtain other examples of non-compactly generated weight structures for ${\underline{C}}$ of this type “starting from” $s_L$ since one can (considering it as a weight structure and) “join” it with any compactly generated weight structure for ${\underline{C}}$ (see Remark \[revenmorews\](1) below).
Moreover, in Remark \[rsymt\](\[irsymt1\],\[irsymt4\]) below a rich family of smashing weight structures on categories dual to compactly generated ones is described; these weight structures are not compactly generated.
On (countably) perfect and symmetric classes; their relation to Brown-Comenetz duality and adjacent torsion pairs {#scomp}
-----------------------------------------------------------------------------------------------------------------
Now we recall the notion of perfectly generated triangulated categories; categories of this type satisfy the Brown representability property according to results of A. Neeman and H. Krause.
\[dwg\] Let ${\alpha}$ be a regular infinite cardinal.[^36]
1. \[idsmall\] An object $M$ of ${\underline{C}}$ is said to be [*${\alpha}$-small*]{} if for any set of $N_i\in {\operatorname{Obj}}{\underline{C}}$ any morphism $M\to \coprod N_i$ factors through the coproduct of a subset of $\{N_i\}$ of cardinality less then ${\alpha}$.
2. \[idpc\] We will say that a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ is [*countably perfect*]{} if the class of ${\mathcal{P}}$-null morphisms is closed with respect to countable coproducts (recall that $h$ is ${\mathcal{P}}$-null if for all $M\in {\mathcal{P}}$ we have $H^M(h)=0$; see Definition \[dhopo\](5)). We will say that ${\mathcal{P}}$ is [*perfect*]{} if the class of ${\mathcal{P}}$-null morphisms is closed with respect to arbitrary coproducts.
3. \[idpg\] We will say that ${\underline{C}}$ is [*perfectly generated*]{} if there exists a countably perfect [**set**]{} ${\mathcal{P}}_0\subset {\operatorname{Obj}}{\underline{C}}$ that Hom-generates it.
We will also say that ${\underline{C}}$ is ${\alpha}$-[*well generated*]{} (or just well generated) if (in addition) all elements of ${\mathcal{P}}_0$ are ${\alpha}$-small and ${\mathcal{P}}_0$ is perfect.
Now we prove a few properties of these definitions.
\[psym\] Let ${\mathcal{P}}$ be a class of objects of a triangulated category ${\underline{C}}$ that has coproducts. Denote by ${\underline{C}}'$ the localizing subcategory of ${\underline{C}}$ generated by ${\mathcal{P}}$ and denote by ${\underline{D}}$ the full subcategory of ${\underline{C}}$ whose object class equals ${\operatorname{Obj}}{\underline{C}'}{{}^{\perp}}$; denote the embedding ${\underline{C}}'\to {\underline{C}}$ by $i$. Then the following statements are valid.
1. \[isymcomp\] If a set ${\mathcal{Q}}$ Hom-generates ${\underline{C}}$ then it also ${{\aleph_0}}$-well generates it if and only if all elements of ${\mathcal{Q}}$ are compact.
2. \[isymeu\] ${\underline{D}}$ is triangulated and ${\operatorname{Obj}}{\underline{D}}=\cap_{i\in {{\mathbb{Z}}}}({\mathcal{P}}{{}^{\perp}}[i])$. Moreover, if ${\mathcal{P}}$ is (countably) perfect then ${\underline{D}}$ is closed with respect to (countable) ${\underline{C}}$-coproducts.
3. \[isymbr\] Assume that ${\mathcal{P}}$ is a (countably) perfect set. Then ${\underline{C}}'$ is perfectly generated by ${\mathcal{P}}$ and has the Brown representability property. Moreover, the embedding $i$ possesses an exact right adjoint $G$ that gives an equivalence ${\underline{C}}/{\underline{D}}\to {\underline{C}}'$ and respects (countable) coproducts.[^37]
4. \[isymuni\] If ${\mathcal{P}}_i$ is a collection of (countably) perfect subclasses of ${\operatorname{Obj}}{\underline{C}}$ then $\cup {\mathcal{P}}_i$ is (countably) perfect also.
5. \[iperftp\] Assume that $s=({\mathcal{LO}},{\mathcal{RO}})$ is a torsion pair for ${\underline{C}}$. Then $s$ is (countably) smashing if and only if ${\mathcal{LO}}$ is (countably) perfect.
\[isymcomp\]. Certainly, ${{\aleph_0}}$-small objects are precisely the compact ones. Hence any ${{\aleph_0}}$-well generating set consists of compact objects. To get the converse implication it suffices to note that any class of compact objects is perfect (see assertion \[isymuni\]).
\[isymeu\]. The first part of the assertion is obvious; the second one follows immediately from Proposition \[pwsym\](\[iwsmor\]). \[isymbr\]. The set ${\mathcal{P}}$ Hom-generates the category ${\underline{C}'}$ according to Proposition \[pcomp\](I.1); certainly it is perfect in it. Hence Lemma \[lperf\](\[ipereq\]) below (see Remark \[requivdef\]) implies that ${\mathcal{P}}$ perfectly generates ${\underline{C}'}$ (also) in the sense of [@kraucoh Definition 1]. Hence the Brown representability condition for ${\underline{C}'}$ is given by Theorem A of [@kraucoh]. Given this condition the existence of $G$ follows from Proposition \[pcomp\](II.2). Lastly, Proposition \[prtst\](\[it4sm\]) says that $G$ respects (countable) coproducts.
\[isymuni\]. It suffices to note that uniting ${\mathcal{P}}_i$ corresponds to intersecting the corresponding classes of null morphisms.
\[iperftp\]. Immediate from Proposition \[phop\](8).
\[rwg\] 1. Recall also that any well generated triangulated category possessing a combinatorial model satisfies the dual Brown representability property; see §0 of [@neefrosi] (the statement is given by the combination of Theorems 0.17 and 0.14 of ibid.).
2\. It is well known that the class of well generated triangulated categories is “much bigger” than that of compactly generated ones; the class of perfectly generated categories is even bigger.[^38] Moreover, if ${\underline{C}}$ is well generated then any its subcategory ${\underline{C}}'$ generated by a set of objects as a localizing subcategory is well generated also; the Verdier localization ${\underline{C}}/{\underline{C}'}$ exists and is well generated (see Theorem 4.4.9 of [@neebook]). Note that the obvious analogue of this result for compactly generated categories is wrong (so, if ${\underline{C}}$ is compactly generated then its set-generated localizing subcategory ${\underline{C}}'$ along with the quotient ${\underline{C}}/{\underline{C}}'$ is only well generated in general). In particular, in [@neeshman] it was proved that the derived category of sheaves on a non-compact manifold is well generated but not compactly generated, whereas it is a localization of the compactly generated derived category of presheaves. However, the reader may assume that all the perfectly generated triangulated categories we study are actually compactly generated since most of our results are quite interesting in this case also. 3. Certainly, a set ${\mathcal{P}}$ is (countably) perfect if and only if the coproduct of its elements forms a (countably) perfect set.
Now, constructing perfect sets (and classes) “without using compact objects” is rather difficult. The main source of “non-compact” perfect classes in our paper is a certain Brown-Comenetz-type “symmetry”; the idea originates from [@kraucoh]. We start with some definitions.
\[dsym\]
Let ${\mathcal{P}}$ and ${\mathcal{P}}'$ be subclasses of ${\operatorname{Obj}}{\underline{C}}$, $P\in {\operatorname{Obj}}{\underline{C}}$.
1. \[iwsym\] We will say that ${\mathcal{P}}$ is [*weakly symmetric*]{} to ${\mathcal{P}}'$ if ${\mathcal{P}}{{}^{\perp}}={{}^{\perp}}{\mathcal{P}}'$.
2. \[iconull\] We will say that a ${\underline{C}}$-morphism $h$ is [*${\mathcal{P}}$-conull*]{} whenever for all $M\in {\mathcal{P}}$ we have $H_M(h)=0$ (where $H_M={\underline{C}}(-,M):{\underline{C}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$).
3. \[isym\] We will say that ${\mathcal{P}}$ is [*symmetric*]{} to ${\mathcal{P}}'$ if the class ${\mathcal{P}}$-null (see Definition \[dhopo\](5)) coincides with the class of ${\mathcal{P}}'$-conull morphisms.
4. \[ibcomf\] Let $H:{\underline{C}}\to {\underline{\operatorname{Ab}}}$ be a homological functor. Then we will call the functor ${\widehat{H}}:M\mapsto {\underline{\operatorname{Ab}}}(H(M),{{\mathbb{Q}}}/{{\mathbb{Z}}}):{\underline{C}}\to {\underline{\operatorname{Ab}}}$ the [*Brown-Comenetz dual*]{} of $H$.
5. \[ibcomo\] We will call an object of ${\underline{C}}$ the [*Brown-Comenetz dual*]{} of $P$ and denote it by $\hat{P}$ if it represents the Brown-Comenetz dual of the functor $H^M={\underline{C}}(M,-):{\underline{C}}\to {\underline{\operatorname{Ab}}}$.
\[psymb\] Assume that ${\mathcal{P}}$, ${\mathcal{P}}'$, along with certain ${\mathcal{P}}_i$ and ${\mathcal{P}}'_i$ for $i$ running through some $I$, are subclasses of ${\operatorname{Obj}}{\underline{C}}$; let $P\in {\operatorname{Obj}}{\underline{C}}$ and $h$ be a ${\underline{C}}$-morphism. Assume that $H:{\underline{C}}\to {\underline{\operatorname{Ab}}}$ is a homological functor respecting coproducts.
I. Then the following statements are valid.
1. \[iwsmor\] $P$ belongs to ${\mathcal{P}}{{}^{\perp}}$ if and only if the morphism ${\operatorname{id}}_P$ is ${\mathcal{P}}$-null; dually, $P\in {{}^{\perp}}{\mathcal{P}}$ if and only if ${\operatorname{id}}_M$ is ${\mathcal{P}}$-conull.
2. \[iwsym1\] If ${\mathcal{P}}$ is symmetric to ${\mathcal{P}}'$ then it is also weakly symmetric to ${\mathcal{P}}'$.
If we assume in addition that ${\underline{C}}$ has coproducts then ${\mathcal{P}}$ is perfect.
3. \[iws2\] ${\mathcal{P}}$ is (weakly) symmetric to ${\mathcal{P}}'$ if and only if ${\mathcal{P}}'$ is (weakly) symmetric to ${\mathcal{P}}$ in the category ${\underline{C}}{{^{op}}}$.[^39]
4. \[iws1\] If ${\mathcal{P}}_i$ is (weakly) symmetric to ${\mathcal{P}}'_i$ for any $i\in I$ then $\cup{\mathcal{P}}_i$ is (weakly) symmetric to $\cup{\mathcal{P}}'_i$.
5. \[isbcd\] The Brown-Comenetz dual functor ${\widehat{H}}$ is a cohomological functor that converts ${\underline{C}}$-coproducts into ${\underline{\operatorname{Ab}}}$-products. Moreover, if ${\widehat{H}}$ is represented by some $N\in {\operatorname{Obj}}{\underline{C}}$ then $h$ is $\{N\}$-conull if and only if $H(h)=0$.
6. \[isym4\] Assume that ${\underline{C}}$ has coproducts, ${\mathcal{P}}$ is symmetric to ${\mathcal{P}}'$, and both of them are sets. Then the localizing subcategory ${\underline{C}'}$ generated by ${\mathcal{P}}$ satisfies both the Brown representability condition and its dual, the embedding $i:{\underline{C}}'\to {\underline{C}}$ has an (exact) right adjoint $G$, and ${\mathcal{P}}$ is symmetric to the class $G({\mathcal{P}}')$ in ${\underline{C}}'$. Moreover, for ${\underline{D}}$ being the full subcategory of ${\underline{C}}$ whose object class equals ${\operatorname{Obj}}{\underline{C}'}{{}^{\perp}}$, the functor $G$ gives an equivalence to ${\underline{C}}'$ of the full subcategory ${\underline{D}}'$ of ${\underline{C}}$ whose object class equals ${\operatorname{Obj}}{\underline{D}}{{}^{\perp}}$. Furthermore, the embedding ${\underline{D}}'\to {\underline{C}}$ respects products and possesses a left adjoint, and ${\underline{D}}'^{op}$ (has coproducts and) is perfectly generated by ${\mathcal{P}}'$.
II\. Assume in addition that ${\underline{C}}$ (has coproducts) and satisfies the Brown representability condition.
1. \[iws21\] Then ${\widehat{H}}$ is representable by some $N\in {\operatorname{Obj}}{\underline{C}}$.
2. \[iws22\] Assume that $P$ is compact in ${\underline{C}}$. Then its Brown-Comenetz dual object $\hat{P}$ exists.
3. \[iws23\] Assume that all objects of ${\mathcal{P}}$ are compact. Then ${\mathcal{P}}$ is symmetric to the set of the Brown-Comenetz duals of elements of ${\underline{C}}$.
I.\[iwsmor\]. Obvious.
\[iwsym1\]. The first part of the assertion is immediate from the previous one; the second one is obvious.
\[iws2\], \[iws1\]. Obvious.
\[isbcd\]. Certainly, ${\widehat{H}}$ converts ${\underline{C}}$-coproducts into products of abelian groups. It is cohomological since ${{\mathbb{Q}}}/{{\mathbb{Z}}}$ is an injective object of ${\underline{\operatorname{Ab}}}$. Since it also cogenerates ${\underline{\operatorname{Ab}}}$, we obtain that $H(h)=0$ if ${\widehat{H}}(h)=0$, whereas the converse implication is automatic.
\[isym4\]. The set ${\mathcal{P}}$ is perfect according to assertion I.\[iwsym1\]; hence ${\underline{C}}'$ is perfectly generated by ${\mathcal{P}}$. Thus ${\underline{C}'}$ satisfies Brown representability and $i$ possesses an exact right adjoint $G$ that respects coproducts and gives an equivalence ${\underline{C}}/{\underline{D}}\to {\underline{C}}'$ (see Proposition \[psym\](\[isymbr\])). Thus $G$ restricts to a fully faithful functor $j$ from ${\underline{D}}'$ into ${\underline{C}'}$. Next, the adjunction of $i$ to $G$ immediately yields that ${\mathcal{P}}$ is symmetric to the class $G({\mathcal{P}}')$ in ${\underline{C}}'$ indeed. Hence ${\underline{C}'}$ has the Brown representability property; thus it has products according to Proposition \[pcomp\](II.2). Thus $G({\mathcal{P}}')$ perfectly generates the category ${\underline{C}}'^{op}$; therefore ${\underline{C}'}$ satisfies the dual Brown representability condition.[^40] Now, ${\underline{D}}'$ is certainly a triangulated subcategory of ${\underline{C}}$ that is closed with respect to ${\underline{C}}$-products. We have ${\operatorname{Obj}}{\underline{D}}\perp {\mathcal{P}}'$ since ${\mathcal{P}}\perp {\operatorname{Obj}}{\underline{D}}$; hence ${\mathcal{P}}'$ perfectly generates ${\underline{D}}'^{op}$ (since this statement becomes true after we apply $j$) and $j$ is an equivalence. It remains to apply (the dual to) Proposition \[pcomp\](II.2).
II.\[iws21\],\[iws22\]. By the definition of Brown representability, it suffices to note that both ${\widehat{H}}$ and $\widehat{H^P}$ are cp functors.
\[iws23\]. Easy; combine assertions II.\[iws22\], I.\[isbcd\], and I.\[iws1\].
Now we establish a (somewhat funny) general criterion for a torsion pair $s'$ to be adjacent to $s$.
\[pwsym\] Let ${\mathcal{P}}$, ${\mathcal{P}}'$ be subclasses of ${\operatorname{Obj}}{\underline{C}}$; for torsion pairs $s=({\mathcal{LO}},{\mathcal{RO}})$ and $s'=({\mathcal{LO}},{\mathcal{RO}})$ assume that a class ${\mathcal{P}}_s\subset {\operatorname{Obj}}{\underline{C}}$ generates $s$ and some ${\mathcal{P}}_{s'}$ [*cogenerates*]{} $s'$ (i.e., ${\mathcal{LO}}={{}^{\perp}}{\mathcal{P}}_{s'}$ and so $s'\subset {\mathcal{RO}}'$). Then the following statements are valid.
1. \[iws3\] ${\mathcal{P}}_s$ is weakly symmetric to ${\mathcal{P}}'$ if and only if ${\mathcal{LO}}$ is weakly symmetric to ${\mathcal{P}}'$.
2. \[iws3p\] ${\mathcal{P}}$ is weakly symmetric to ${\mathcal{P}}_{s'}$ if and only if ${\mathcal{P}}$ is weakly symmetric to ${\mathcal{RO}}'$.
3. \[iws4\] The following conditions are equivalent:
\(i) ${\mathcal{P}}_s$ is weakly symmetric to ${\mathcal{P}}_{s'}$.
\(ii) ${\mathcal{LO}}$ is symmetric to ${\mathcal{RO}}'$.
\(iii) There exist a class ${\mathcal{Q}}$ that generates $s$ and ${\mathcal{Q}}'$ that cogenerates $s'$ such that ${\mathcal{Q}}$ is symmetric to ${\mathcal{Q}}'$.
\(iv) $s$ is left adjacent to $s'$.
4. \[iwsymcgatp\] Assume that ${\underline{C}}$ (has coproducts) and satisfies the Brown representability property, ${\mathcal{P}}_s$ is a class of compact objects,[^41] and $s'$ is right adjacent to $s$. Then $s$ is smashing, $s'$ is cosmashing, and $s'$ is cogenerated by the Brown-Comenetz duals of elements of ${\mathcal{P}}_s$ (i.e., ${\mathcal{LO}}'={{}^{\perp}}\widehat{{\mathcal{P}}_s}$ where $\widehat{{\mathcal{P}}_s}=\{\hat{P}:\ P\in {\mathcal{P}}_s\}$).
5. \[iwsymcgwt\] For ${\underline{C}}$ as in the previous assertion assume that ${\mathcal{P}}$ is a class of compact objects and it generates a weight structure $w$. Then $w$ is smashing, there exists a cosmashing $t$-structure $t$ right adjacent to $w$, and $t$ is cogenerated by the Brown-Comenetz duals of elements of ${\mathcal{P}}$ (i.e., ${\underline{C}}^{t\le -1}={{}^{\perp}}\widehat{{\mathcal{P}}}$, where $\widehat{{\mathcal{P}}}=\{\hat{P}:\ P\in {\mathcal{P}}\}$).
\[iws3\], \[iws3p\]: obvious (recall the corresponding definitions).
\[iws4\]. By definition, $s$ is left adjacent to $s'$ if and only if ${\mathcal{LO}}{{}^{\perp}}={{}^{\perp}}{\mathcal{RO}}'$; hence (i) is equivalent to (iv). Applying assertion \[iwsmor\] we also obtain that (iii) implies (i). Next, the equivalence of (ii) to (iv) is immediate from Proposition \[phop\](8). Since (ii) implies (iii), we obtain the result.
\[iwsymcgatp\]. $s$ is smashing and $s'$ is cosmashing according to Proposition \[phop\](5) (recall that ${\underline{C}}$ has products according to Proposition \[pcomp\](II.2)). Next, ${\mathcal{P}}_s$ is symmetric to $\widehat{{\mathcal{P}}_s}$ according to Proposition \[psymb\](II.\[iws23\]). Hence it remains to apply the previous assertion.
\[iwsymcgwt\]. Certainly, $w$ is smashing (see Proposition \[phopft\](III)). Hence $t$ exists according to Theorem \[tadjt\], and it remains to apply the previous assertion.
\[rwsym\]
1\. Part \[iwsymcgwt\] of corollary is the first application of part \[iwsymcgatp\]; another application is (essentially) Corollary \[csymt\] below. Note also that Theorem 3.11 of [@postov] states that in a compactly generated [*algebraic*]{} triangulated category ${\underline{C}}$ for any compactly generated torsion pair $s$ (i.e., we assume ${\mathcal{P}}_s$ to be a set in assertion \[iwsymcgatp\]) there exists a torsion pair right adjacent to it. Hence this torsion pair $s'$ is cogenerated by the corresponding $\widehat{{\mathcal{P}}_s}$ (also). 2. In theory, (part \[iws4\] of) our proposition gives a complete description of all couples of adjacent torsion pairs: one can start with $s$, take a generating class ${\mathcal{P}}_s$ for it (that may be equal to ${\mathcal{LO}}$), find a class ${\mathcal{P}}_{s'}$ that is weakly symmetric to ${\mathcal{P}}_s$ (if any), and “cogenerate” $s'$ (if ${\mathcal{P}}_{s'}$ does cogenerate some torsion pair).
Still constructing (weakly) symmetric classes and cogenerating torsion pairs by them appears to be rather difficult in general; the author can only “do” this by applying the Brown-Comenetz duality. So, one may say that we construct (weakly) symmetric classes “elementwisely”; the author wonders whether a “more involved” method exists.
However, our proposition demonstrates the relation of adjacent torsion pairs to “Brown-Comenetz-type symmetry”; this point of view appears to be new.
3\. The problem with the symmetry condition is that the class ${\mathcal{P}}_s$-null is not determined by $s$; so even if $s$ and $s'$ are adjacent, it may be difficult to find “small” symmetric ${\mathcal{Q}}$ and ${\mathcal{Q}}'$ as in condition \[iws4\](iii). So, we only have two (rather) “extreme” (“basic”) types of symmetric classes: the ones coming from classes of compact objects (that “usually” have bounded cardinality) via Brown-Comenetz duality and the “big” ones of the “type” $({\mathcal{LO}},{\mathcal{RO}}')$ (see condition \[iws4\](ii))).[^42]
Note also that in the proof of Theorem \[tsymt\] we do not actually need ${\mathcal{P}}$ to be symmetric; it suffices to assume that ${\mathcal{P}}$ is weakly symmetric to ${\mathcal{P}}'$ and ${\mathcal{P}}'$ is a set that is perfect in the category ${\underline{C}}{{^{op}}}$.
On perfectly generated weight structures and torsion pairs {#spgtp}
==========================================================
In this section we will always assume that ${\underline{C}}$ has coproducts and ${\mathcal{P}}$ is a subclass of ${\operatorname{Obj}}{\underline{C}}$. Our goal is to study the case when ${\mathcal{P}}$ is a (countably) perfect class and it generates a torsion pair $s$ for ${\underline{C}}$. Our results are more satisfactory when $s$ is weighty; in particular, we prove that all weight structures on well generated triangulated categories are ([*perfectly generated*]{} and) [*strongly well generated*]{}.
In §\[scoulim\] we recall the notion of countable homotopy colimit in ${\underline{C}}$ (that is one of the main tools of this section) and introduce several related notions and facts.
In §\[scghop\] we prove that compactly generated torsion pairs are in one-to-one correspondence with extension-closed Karoubi-closed essentially small classes of compact objects (in ${\underline{C}}$). This result (slightly) generalizes Theorem 3.7 of [@postov] (along with Corollary 3.8 of ibid.). In §\[sperfws\] we study the (naturally defined) perfectly generated weight structures. The existence Theorem \[tpgws\] is absolutely new; still its proof has some “predecessors”. It appears to be rather difficult to construct examples of perfectly generated weight structures that are not compactly generated; still we construct a curious family of those using suspended symmetric sets $({\mathcal{P}},{\mathcal{P}}')$ (see Definition \[dsym\](\[isym\])). For any couple of this sort the set ${\mathcal{P}}'$ is perfect in the category ${\underline{C}}{{^{op}}}$; so we obtain a certain weight structure $w$ on ${\underline{C}}$ whereas its left adjacent $t$-structure (whose existence is essentially given by Corollary \[cdualt\]) is generated by ${\mathcal{P}}$. The author does not know how to construct “new” $t$-structures using this result; it implies however that for any compactly generated $t$-structure there exists a right adjacent weight structure.[^43] Now, the opposite to this weight structure (in ${\underline{C}}{{^{op}}}$) is perfectly generated but (“almost never”) compactly generated, whereas the existence of $w$ implies that the category ${{\underline{Ht}}}$ has an injective cogenerator and satisfies the AB3\* axiom. Moreover, we deduce (using the results of [@humavit]) that ${{\underline{Ht}}}$ is Grothendieck abelian whenever $t$ is non-degenerate.
In §\[swgws\] we develop a certain theory of well generated torsion pairs. We prove several relations between torsion pairs and (countably) perfect classes. Probably, the most interesting result in this section is the fact that a smashing weight structure on a well generated triangulated category is [*strongly well generated*]{} (i.e., it restricts to the subcategory of [*${\beta}$-compact*]{} objects for any large enough regular ${\beta}$ and it can be “recovered” from this restriction); in particular, it is perfectly generated (i.e., it may be constructed using Theorem \[tpgws\]).
On countable homotopy colimits {#scoulim}
------------------------------
We recall the basics of the theory of countable (filtered) homotopy colimits in triangulated categories (as introduced in [@bokne]; some more detail can be found in [@neebook]; cf. also §4.2 of [@bws]). We will only apply the results of this subsection to triangulated categories that have coproducts; so we will not mention this restriction below.
\[dcoulim\] For a sequence of objects $Y_i$ of ${\underline{C}}$ for $i\ge 0$ and maps $\phi_i:Y_{i}\to Y_{i+1}$ we consider the morphism $a:\oplus {\operatorname{id}}_{Y_i}\bigoplus \oplus (-\phi_i): D\to D$ (we can define it since its $i$-th component can be easily factorized as the composition $Y_i\to Y_i\bigoplus Y_{i+1}\to D$). Denote a cone of $a$ by $Y$. We will write $Y=\operatorname{\varinjlim}Y_i$ and call $Y$ a [*homotopy colimit*]{} of $Y_i$ (we will not consider any other homotopy colimits in this paper).
Moreover, $\operatorname{\operatorname{Cone}}(\phi_i)$ will be denoted by $Z_{i+1}$ and we set $Z_0=Y_0$.
\[rcoulim\] 1. Note that these homotopy colimits are not really canonical and functorial in $Y_i$ since the choice of a cone is not canonical. They are only defined up to non-canonical isomorphisms; still this is satisfactory for our purposes.
Note also that the definition of $Y$ gives a canonical morphism $D\to Y$.
2\. By Lemma 1.7.1 of [@neebook], a homotopy colimit of $Y_{i_j}$ is the same (up to an isomorphism) for any subsequence of $Y_i$. In particular, we can discard any (finite) number of first terms in $(Y_i)$.
3\. By Lemma 1.6.6 of [@neebook], $M$ is a homotopy colimit of $M\stackrel{{\operatorname{id}}_M}{\to}M\stackrel{{\operatorname{id}}_M}{\to}
M\stackrel{{\operatorname{id}}_M}{\to} M\stackrel{{\operatorname{id}}_M}{\to}\dots$. 4. More generally, if $p$ is an idempotent endomorphism of $M$ then $p$ is isomorphic to a retraction of $M$ onto a homotopy colimit $N$ of $M\stackrel{p}{\to}M\stackrel{p}{\to} M\stackrel{p}{\to} M\stackrel{p}{\to}\dots$ (see the proof of Proposition 1.6.8 of ibid). It easily follows that any extension-closed coproductive subclass of ${\operatorname{Obj}}{\underline{C}}$ that is closed either with respect $[1]$ or $[-1]$ is Karoubian; see also Corollary 2.1.3(2) of [@bsnew] for an alternative proof of this fact.
5\. Below we will often want to say something on some (co)homology of $Y$ along with morphisms from it.
We start from treating representable cohomology.
Let $T$ be an object of ${\underline{C}}$ and consider the cp functor $H_T={\underline{C}}(-,T)$ from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$. Then the distinguished triangle defining $Y$ certainly yields a long exact sequence $$\begin{aligned} \dots\to H_T(D[1])\stackrel{H_T(a[1])} \to H_T(D[1])(\cong \prod H_T(Y_i[1]))\to
H_T(Y) \\
\to H_T(D)(\cong \prod H_T(Y_i)) \stackrel{H_T(a)}{\to} H_T(D)\to\dots \end{aligned}$$ According to Remark A.3.6 of [@neebook] this gives a short exact sequence $$0\to \operatorname{\varprojlim}^1 H_T(Y_i[1])\to H_T(Y)\to \operatorname{\varprojlim}H_T(Y_i) \to 0.$$ Thus any morphism $f\in {\underline{C}}(Y,T)$ gives a “coherent” system of morphisms $Y_i\to T$. Conversely, for any coherent system $(f_i)$ of this sort there exists its “lift” to some $f\in {\underline{C}}(Y,T)$ that we will say to be [*compatible*]{} with $(f_i)$. Certainly, the compatibility of $f$ with $(f_i)$ is fulfilled if and only if the composition $Y_i\to \coprod Y_i\to Y\stackrel{f}{\to} T$ equals $f_i$ for any $i\ge 0$. Hence for any functor $F$ from ${\underline{C}}$ the composition $F(Y_i)\to F(\coprod Y_i)\to F(Y)\stackrel{F(f)}{\to} F(T)$ equals $F(f_i)$.
We study the behaviour of homotopy colimits under cp and wcc functors.
\[lcoulim\] Assume that $Y=\operatorname{\varinjlim}Y_i$ (in ${\underline{C}}$), ${\underline{A}}$ is an abelian category; let $H$ (resp. $H'$) be a cp (resp. a wcc; see Definition \[dcomp\](\[idcc\])) functor from ${\underline{C}}$ into ${\underline{A}}$. Then the following statements are valid.
1\. The obvious connecting morphisms $Y_i\to Y$ give an epimorphism $H(Y)\to \operatorname{\varprojlim}H(Y_i)$.
2\. This epimorphism $H(Y)\to \operatorname{\varprojlim}H(Y_i)$ is an isomorphism whenever ${\underline{A}}$ is an AB4\* category and all the morphisms $H(\phi_i[1])$ are epimorphic for $i\gg 0$.
3\. $H'(Y) $ naturally surjects onto $ \operatorname{\varinjlim}H'(Y_i)$.
This surjection is an isomorphism if either (i) there exist objects $A$ and $A_i\in {\operatorname{Obj}}{\underline{A}}$, along with compatible isomorphisms $H'(Y_i[1])\cong A_i\bigoplus A$ and $H(\phi_i[1])\cong (0:A_i\to A_{i+1}) \bigoplus {\operatorname{id}}_A$, for $i\gg 0$
or
\(ii) ${\underline{A}}$ is an AB5 category.
In particular, if $C$ is compact then ${\underline{C}}(C,Y)\cong \operatorname{\varinjlim}{\underline{C}}(C,Y_i)$.
1\. We argue as in Remark \[rcoulim\](5).
We have a long exact sequence $$\dots\to H(D[1])\stackrel{H(a[1])} \to H(D[1])\to
H(Y)\to H(D)\stackrel{H(a)}{\to} H(D)\to\dots.$$
Since $H(D)\cong \prod H(Y_i)$, the kernel of $H(a)$ equals $\operatorname{\varprojlim}H(Y_i)$ (and this inverse limit exists in ${\underline{A}}$). This yields the result.
2\. Remark A.3.6 of [@neebook] yields that (in this case) the cokernel of $H(a[1])$ equals the $1$-limit of the objects $H(Y_i[1])$. Next, by Remark \[rcoulim\](1) we can assume that the homomorphisms $\phi[1]^*$ are surjective for all $i$. Hence the statement is given by Lemma A.3.9 of ibid.
3\. Similarly to the proof of assertion 1, we consider the long exact sequence $$\dots\to H'(D)\stackrel{H'(a)}{\to} H'(D)\to H'(Y)\to
H'(D[1]) \stackrel{H'(a[1])} \to H'(D[1]) \to\dots.$$
Since $H'(D)\cong \coprod H'(Y_i)$, it easily follows that the cokernel of $H'(a)$ is $\operatorname{\varinjlim}H'(Y_i)$, this gives the first part of the assertion.
To prove its second part we should verify that $H'(a[1])$ is monomorphic (if either of the two additional assumption is fulfilled). We will write $B_i$ and $f_i$ for $H'(Y_i[1])$ and $H(\phi_i[1])$, respectively, whereas $H'(a[1])$ (that can certainly be expressed in terms of ${\operatorname{id}}_{B_i}$ and $f_i$) will be denoted by $h$.
If (i) is valid then we can certainly assume that $B_i\cong A_i\bigoplus A$ and $f_i\cong {\operatorname{id}}_A\bigoplus 0$ for all $i\ge 0$. Moreover, the additivity of the object $\operatorname{\operatorname{Ker}}(h)$ with respect to direct sums of $(B_i,f_i)$ reduces its calculation to the following to cases: (1) $f_i=0$ and (2) $f_i\cong {\operatorname{id}}_A$. In case (1) $h$ is isomorphic to ${\operatorname{id}}_{\coprod B_i}$; so it is monomorphic. In case (2) $h$ is monomorphic also since its composition with the projection of $\coprod B_i$ onto $\coprod_{i>0}B_i$ is an isomorphism. Indeed, the inverse to the latter is given by the morphism matrix $$\begin{pmatrix}{\operatorname{id}}_A &{\operatorname{id}}_A &{\operatorname{id}}_A &\dots \\
0 & {\operatorname{id}}_A &{\operatorname{id}}_A &\dots\\
0 & 0 &{\operatorname{id}}_A &\dots\\
0 & 0 &0 &\dots\\
\dots & \dots &\dots &\dots \end{pmatrix}$$ (cf. the proof of [@neebook Lemma 1.6.6]). To prove version (ii) of the assertion note that the composition of $H'(a[1])$ with the obvious monomorphism $\coprod_{i\le j} H'(Y_i[1])\to \coprod_{i\ge 0} H'(Y_i[1])$ is easily seen to be monomorphic for each $j\ge 0$. If ${\underline{A}}$ is an AB5 category then it follows that the morphism $H'(a[1])$ is monomorphic itself.
We will also need the following definitions.[^44]
\[dses\]
1\. A class ${{\tilde{\mathcal{P}}}}\subset {\operatorname{Obj}}{\underline{C}}$ will be called [*strongly extension-closed*]{} if it contains $0$ and for any $\phi_i:Y_{i}\to Y_{i+1}$ such that $Y_0\in {{\tilde{\mathcal{P}}}}$ and $\operatorname{\operatorname{Cone}}(\phi_i)\in {{\tilde{\mathcal{P}}}}$ for all $i\ge 0$ we have $\operatorname{\varinjlim}_{i\ge 0} Y_i\in {{\tilde{\mathcal{P}}}}$ (i.e. ${{\tilde{\mathcal{P}}}}$ contains all possible cones of the corresponding distinguished triangle; note that these are isomorphic).
2\. The smallest strongly extension-closed Karoubi-closed class of objects of ${\underline{C}}$ that contains a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ and is closed with respect to arbitrary (small) coproducts will be called the [*strong extension-closure*]{} of ${\mathcal{P}}$.
3\. We will write ${\underline{\coprod}{\mathcal{P}}}$ for the closure of ${\mathcal{P}}$ with respect to ${\underline{C}}$-coproducts (and in §\[sperfws\] below we will also use this notation for the full subcategory of ${\underline{C}}$ formed by these objects). Also, we will call the class of the objects of ${\underline{C}}$ that may be presented as homotopy limits of $Y_i$ with $Y_0$ and $\operatorname{\operatorname{Cone}}(\phi_i)\in {\underline{\coprod}{\mathcal{P}}}$ the [*naive big hull*]{} of ${\mathcal{P}}$. We will call the Karoubi-closure of the naive big hull of the class ${\mathcal{P}}$ its [*big hull*]{}.
Now we prove a few simple properties of these notions.
\[lbes\] Let ${\mathcal{P}}$ be a class of objects of ${\underline{C}}$; denote its strong extension-closure by ${{\tilde{\mathcal{P}}}}$.
1. \[iseses\] ${{\tilde{\mathcal{P}}}}$ is extension-closed in ${\underline{C}}$; it contains the big hull of ${\mathcal{P}}$.
2. \[isesperp\] Let $H$ be a cp functor from ${\underline{C}}$ into a AB4\*-category ${\underline{A}}$, and assume that the restriction of $H$ to ${\mathcal{P}}$ is zero. Then $H$ kills ${{\tilde{\mathcal{P}}}}$ also.
In particular, if for some $D\subset {\operatorname{Obj}}{\underline{C}}$ we have $ {\mathcal{P}}\perp D$ then ${{\tilde{\mathcal{P}}}}\perp D$ also.
3. \[isescperp\] Let $H'$ be a cc functor from ${\underline{C}}$ into a AB5-category. Then $H'$ kills ${{\tilde{\mathcal{P}}}}$ whenever it kills ${\mathcal{P}}$.
In particular, if $D$ is a class of compact objects in ${\underline{C}}$ and $D\perp {\mathcal{P}}$ then $D\perp {{\tilde{\mathcal{P}}}}$ also.
4. \[izs\] [*Zero classes*]{} of arbitrary families of cp and cc functors (into AB4\* and AB5 categories, respectively) are strongly extension-closed (i.e. for any cp functors $H_i$ and cc functors $H'_i$ of this sort the classes $\{M\in {\operatorname{Obj}}{\underline{C}}:\ H_i(M)=0\ \forall i\}$ and $\{M\in {\operatorname{Obj}}{\underline{C}}:\ H'_i(M)=0\ \forall i\}$ are strongly extension-closed).
In particular, if $({\mathcal{LO}},{\mathcal{RO}})$ is a torsion pair in ${\underline{C}}$ then ${\mathcal{LO}}$ is strongly extension-closed.
5. \[ilwd\] Adopt the notation of Definition \[dcoulim\]; let $w$ be a countably smashing weight structure on ${\underline{C}}$. Choose some $w$-decompositions $LZ_i\to Z_i\to RZ_i\to LZ_i[1]$ of $Z_i$ (see Definition \[dcoulim\]) for $i\ge 0$.
Then there exists some $w$-decompositions $LY_i\to Y_i\to RY_i\to LY_i[1]$ for $i\ge 0$ and connecting morphisms $l_i:LY_i\to LY_{i+1}$ for $i\ge 0$ such that the corresponding squares commute, $LY_0=LZ_0$, $\operatorname{\operatorname{Cone}}(l_i)\cong LZ_{i+1}$, and there exists a weight decomposition $\operatorname{\varinjlim}LY_i\to Y\to RY\to (\operatorname{\varinjlim}LY_i)[1]$ (for some $RY\in {\underline{C}}_{w\ge 1}$).
\[iseses\]. For any distinguished triangle $X\to Y\to Z$ for $X,Z\in {{\tilde{\mathcal{P}}}}$ the object $Y$ is the colimit of $X\stackrel{f}{\to} Y\stackrel{{\operatorname{id}}_Y}{\to} Y \stackrel{{\operatorname{id}}_Y}{\to} Y \stackrel{{\operatorname{id}}_Y}{\to} Y\to \dots$ (see Remark \[rcoulim\](3)). Since a cone of $f$ is $Z$, whereas a cone of ${\operatorname{id}}_Y$ is $0$, ${{\tilde{\mathcal{P}}}}$ is extension-closed indeed. It contains the big hull of ${\mathcal{P}}$ by definition.
\[isesperp\]. Since for any $d\in D$ the functor $H_d:{\underline{C}}\to {\underline{\operatorname{Ab}}}$ converts arbitrary coproducts into products, it suffices to verify the first part of the statement.
Thus it suffices to verify that $H(Y)=0$ if $Y=\operatorname{\varinjlim}Y_i$ and $H$ kills cones of the connecting morphisms $\phi_i$.
Now, $H(Y_j)={\{0\}}$ for any $j\ge 0$ (by obvious induction). Next, the long exact sequence $$\dots \to H(Y_{i+1}[1]) \stackrel{H(\phi_i[1])}{\to} H(Y_i[1]) \to H(\operatorname{\operatorname{Cone}}(\phi_i)) (=0) \to H(Y_{i+1}) \to H(Y_i)\to \dots$$ gives the surjectivity of $H(\phi_i[1])$. Hence $H(Y)\cong \operatorname{\varprojlim}H(Y_i)=0$ according to Lemma \[lcoulim\](1,2).
\[isescperp\]. Once again, it suffices to verify the first part of the assertion. Similarly to the previous argument the result easily follows from Lemma \[lcoulim\](3). \[izs\]. The first part of the assertion is immediate from the previous assertions. To deduce the “in particular” part we note that ${\mathcal{LO}}$ is precisely the zero class of the cp functors $\{H_N\}$ for $N$ running through ${\mathcal{RO}}$ and $H_N={\underline{C}}(-,N)$.
\[ilwd\]. We can construct $LY_i$ and $l_i$ satisfying the conditions in question expect the last (“colimit”) one inductively using Proposition \[pbw\](\[iwdext\]).
Now we consider the commutative square $$\begin{CD}
\coprod LY_i@>{La}>>\coprod LY_i\\
@VV{\coprod a_{Y_i}}V@VV{\coprod a_{Y_i}}V \\
D@>{a}>>D
\end{CD}$$ where $La=\oplus {\operatorname{id}}_{LY_i}\bigoplus \oplus (-l_i): \coprod LY_i\to \coprod LY_i$ is the morphism corresponding to $\operatorname{\varinjlim}LY_i$, and the remaining notation is from Definition \[dcoulim\]. According to Proposition 1.1.11 of [@bbd], we can complete it to a commutative diagram $$\label{ely}\begin{CD}
\coprod LY_i @>{La}>>\coprod LY_i@>{}>> LY@ >{}>>\coprod LY_i[1] \\
@VV{\coprod a_{Y_i}}V@VV{\coprod a_{Y_i}}V @VV{}V@VV{\coprod a_{Y_i}[1]}V\\
D @>{a}>>D@>{}>> Y@>{}>> D[1]\\
@VV{}V @VV{}V @VV{}V@VV{}V \\
\coprod RY_i @>{}>>\coprod RY_i@>{}>> RY @>{}>> \coprod RY_i[1] \\
\end{CD}$$ whose rows and columns are distinguished triangles. Then $LY$ is a homotopy colimit of $LY_i$ (with respect to $l_i$) by definition. Since $LY_0$ and cones of $l_i$ belong to ${\underline{C}}_{w\le 0}$, we also have $LY\in {\underline{C}}_{w\le 0}$ according to assertion \[izs\]. On the other hand, the bottom row of (\[ely\]) gives $RY\in {\underline{C}}_{w\ge 1}$ (since $\coprod RY_i\in {\underline{C}}_{w\ge 1}$). Thus the third column of our diagram is a weight decomposition of $Y$ of the type desired.
A classification of compactly generated torsion pairs {#scghop}
-----------------------------------------------------
Now we generalize (and extend) Theorem 3.7 of [@postov] to arbitrary triangulated categories that have coproducts.[^45]
\[tclass\] Assume that ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ is a set of compact objects (recall that ${\underline{C}}$ has coproducts).
Then the following statements are valid.
1. \[iclass1\] The strong extension-closure ${\mathcal{LO}}$ of ${\mathcal{P}}$ and ${\mathcal{RO}}={\mathcal{P}}^\perp$ give a smashing torsion pair $s$ for ${\underline{C}}$ (so, $s$ is the torsion pair generated by ${\mathcal{P}}$). Moreover, ${\mathcal{LO}}$ equals the big hull of ${\mathcal{P}}$, and for any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a choice of $L_sM$ (see Remark \[rwsts\](3)) belonging to the naive big hull of ${\mathcal{P}}$.
2. \[iclass2\] The class of compact objects in ${\mathcal{LO}}$ equals the ${\underline{C}}$-envelope of ${\mathcal{P}}$ (see §\[snotata\]).
3. \[iclassts\] The correspondence sending a compactly generated torsion pair $s=({\mathcal{LO}},{\mathcal{RO}})$ for ${\underline{C}}$ into ${\mathcal{LO}}\cap{\underline{C}}^{{{\aleph_0}}}$, where ${\underline{C}}^{{{\aleph_0}}}$ is the class of compact objects of ${\underline{C}}$, gives a one-to-one correspondence between the following classes: the class of compactly generated torsion pairs for ${\underline{C}}$ and the class of essentially small Karoubi-closed extension-closed subclasses of ${\underline{C}}^{{{\aleph_0}}}$.[^46]
4. \[iclasst\] If the torsion pair $s$ generated by ${\mathcal{P}}$ is associated to a $t$-structure then the class ${\mathcal{LO}}$ ($={\underline{C}}^{t\le 0}$) equals the naive big hull of ${\mathcal{P}}$.
5. \[iclass5\] Let $H$ be a cp (resp. a cc) functor from ${\underline{C}}$ into an AB4\* (resp. AB5) category ${\underline{A}}$ whose restriction to ${\mathcal{P}}$ is zero. Then $H$ kills all elements of ${\mathcal{LO}}$ also.
<!-- -->
1. If $s$ is a torsion pair indeed then it is smashing according to Proposition \[phopft\](III).
Since ${\mathcal{P}}\perp {\mathcal{RO}}$, for any $N\in {\mathcal{RO}}$ the cp functor $H_N$ (from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$) also kills ${\mathcal{LO}}$ according to Lemma \[lbes\](\[isescperp\]). Hence ${\mathcal{LO}}\perp{\mathcal{RO}}$.[^47] Since ${\mathcal{LO}}$ is Karoubi-closed by definition, Proposition \[phop\](9) (along with Lemma \[lbes\](\[iseses\])) reduces the assertion to the existence for any $M\in {\operatorname{Obj}}{\underline{C}}$ of an $s$-decomposition such that the corresponding $L_sM$ belongs to the naive big hull of ${\mathcal{P}}$. We apply (a certain modification of) the method used in the proof of [@bws Theorem 4.5.2(I)]. We construct a certain sequence of $M_k\in {\operatorname{Obj}}{\underline{C}}$ for $k\ge 0$ by induction in $k$ starting from $M_0=M$. Assume that $M_k$ (for some $k\ge 0$) is constructed; then we take $P_k=\coprod_{(P,f):\,P\in {\mathcal{P}},f\in {\underline{C}}(P,M_k)}P$; $M_{k+1}$ is a cone of the morphism $\coprod_{(P,f):\,P\in {\mathcal{P}},f\in {\underline{C}}(P,M_k)}f:P_k\to M_k$.
Now we ’assemble’ $P_k$. The compositions of the morphisms $h_k:M_{k}\to M_{k+1}$ given by this construction yields morphisms $g_i:M\to M_i$ for all $i\ge 0$. Besides, the octahedral axiom of triangulated categories immediately yields $\operatorname{\operatorname{Cone}}(h_k)\cong P_k[1]$. Now we complete $g_k$ to distinguished triangles $$\label{etrproof}
L_k\stackrel{b_k}{\to}M \stackrel{g_k}{\to}M_k\stackrel{f_k}{\to} L_k[1].$$ Certainly, $L_0=0$ and the octahedral axiom yields the existence of morphisms $s_i:L_i\to L_{i+1}$ that are compatible with $b_k$ such that $\operatorname{\operatorname{Cone}}(s_i)\cong P_i$ for all $i\ge 0$.
We consider $L=\operatorname{\varinjlim}L_k$ and choose a morphism $b: L\to M$ compatible with the morphism system $(b_k)$ (see Remark \[rcoulim\](5)). We complete $b$ to a distinguished triangle $L\stackrel{b}{\to} M\stackrel{a}{\to} R\stackrel{f}{\to} L[1]$. This triangle will be our candidate for an $s$-decomposition of $M$.
First we note that $L_0=0$; since $\operatorname{\operatorname{Cone}}(s_i)\cong P_i$ we have $L\in {\mathcal{LO}}$ by the definition of the latter.
It remains to prove that $R\in {\mathcal{RO}}$, i.e., that ${\mathcal{P}}\perp R$. For an element $P $ of ${\mathcal{P}}$ we should check that ${\underline{C}}(P,R)={\{0\}}$. The long exact sequence $$\dots \to {\underline{C}}(P,L)\to {\underline{C}}(P,M)\to {\underline{C}}(P,R)\to {\underline{C}}(P, L[1])\to {\underline{C}}(P,M[1])\to\dots$$ translates this into the following assertion: $H^P(b)$ is surjective and $H^P(b[1])$ is injective. Now, by Lemma \[lcoulim\](3(ii)) we have ${\underline{C}}(P,L)\cong\operatorname{\varinjlim}{\underline{C}}(P,L_i)$ and ${\underline{C}}(P,L[1])\cong\operatorname{\varinjlim}{\underline{C}}(P,L_i[1])$. Hence the long exact sequences $$\dots \to H^P(L_k)\to H^P(M)\to H^P(M_k)\to H^P(L_k[1])\to H^P(M[1])\to\dots$$ yield: it suffices to verify that $\operatorname{\varinjlim}{\underline{C}}(P,M_k) ={\{0\}}$ (note here that $h_k$ are compatible with $s_k$). Lastly, ${\underline{C}}(P,P_k)$ surjects onto ${\underline{C}}(P,M_k)$; hence the group ${\underline{C}}(P,M_k)$ dies in ${\underline{C}}(P,M_{k+1})$ for any $k\ge 0$ and we obtain the result.
2. Given the previous assertion, the argument used in the proof of [@postov Theorem 3.7(ii)] goes through without any changes. We will describe another proof of our statement (that does not depend on ibid.) in Remark \[rnz\](1) below.
3. Recall that for any set $D\subset {\operatorname{Obj}}{\underline{C}}$ the smallest strict (full) triangulated subcategory of ${\underline{C}}$ containing $D$ is essentially small by Lemma 3.2.4 of [@neebook]; hence ${\langle}D{\rangle}_{{\underline{C}}}$ (see §\[snotata\] for the notation) is essentially small also (cf. Proposition 3.2.5 of ibid.). Thus for a set ${\mathcal{P}}$ of compact objects of ${\underline{C}}$ its ${\underline{C}}$-envelope ${\mathcal{P}}'$ is essentially small; its elements are compact according to Lemma 4.1.4 of [@neebook]. Since ${\mathcal{P}}'{{}^{\perp}}={\mathcal{P}}{{}^{\perp}}$ (see Proposition \[phop\](1)), the torsion pair $s$ given by assertion \[iclass1\] is also generated by ${\mathcal{P}}'$; hence it suffices to note that ${\mathcal{LO}}\cap {\underline{C}}^{{{\aleph_0}}}={\mathcal{P}}'$ according to assertion \[iclass2\].
4. Recall from assertion \[iclass1\] that for any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a choice of $L_sM$ that belongs to the naive big hull of ${\mathcal{P}}$. Now, if $s$ is associated to a $t$-structure and $M\in {\mathcal{LO}}={\underline{C}}_{t\le 0}$ then we certainly have $L_sM=M$ (see Remark \[rtst1\](\[it1\],\[it3\])); this concludes the proof.
\[iclass5\]. Immediate from Lemma \[lbes\](\[isesperp\]) (resp. \[isescperp\]).
\[rnewt\] 1. As a particular case of part \[iclass1\] of our theorem we obtain that for any set ${\mathcal{P}}$ (of compact objects of ${\underline{C}}$) such that ${\mathcal{P}}[-1]\subset {\mathcal{P}}$ there exists a weight structure on ${\underline{C}}$ such that ${\underline{C}}_{w\ge 1}={\mathcal{P}}^{\perp}$ (cf. Remark \[rwhop\](1); note that this statement was originally proved in [@paucomp]). Thus if ${\underline{C}}$ is compactly or perfectly generated (see Definition \[dwg\](\[idpg\]); in particular, this is certainly the case if ${\underline{C}}={\underline{C}'}$, where the latter is the localizing subcategory of ${\underline{C}}$ generated by ${\mathcal{P}}$) then Theorem \[tadjt\] implies that $({\mathcal{P}}^{\perp}[-1], ({\mathcal{P}}^{\perp})^{\perp})$ is a $t$-structure on ${\underline{C}}$. Moreover, the couple $({\mathcal{P}}^{\perp}[-1]\cap {\operatorname{Obj}}{\underline{C}'}, {\mathcal{P}}^{\perp}\cap {\operatorname{Obj}}{\underline{C}'})^{\perp_{{\underline{C}}'}})$ is a $t$-structure on ${\underline{C}}'$ regardless of any extra restrictions on ${\underline{C}}$ (here one should invoke Propositions \[phopft\](I) and \[pcomp\](II)).
So, we obtain a statement on the existence of (cosmashing) $t$-structures that does not mention weight structures! This result appears to be new (unless ${\underline{C}}$ is a compactly generated algebraic triangulated category; see Remark \[rwsym\](1) and Theorem 3.11 of [@postov]).
Moreover, the results of the next subsection (see Remark \[revenmorews\]) give an even vaster source of smashing weight structures (and so, of their adjacent $t$-structures as well).
2\. Recall that Theorem 3.7 and Corollary 3.8 of [@postov] give parts \[iclass1\]–\[iclassts\] of our theorem in the case where ${\underline{C}}$ is a “stable derivator” triangulated category ${\underline{C}}$.
Moreover, as a consequence of part \[iclassts\] we certainly obtain a bijection between compactly generated $t$-structures (resp. weight structures) and those essentially small Karoubi-closed extension-closed subclasses of ${\underline{C}}^{{{\aleph_0}}}$ that are closed with respect to $[1]$ (resp. $[-1]$); this generalizes Theorem 4.5 of ibid. to arbitrary triangulated categories having coproducts.
3\. Part \[iclasst\] of our theorem generalizes Theorem A.9 of [@kellerw] where “stable derivator” categories were considered (similarly to the aforementioned results of [@postov]).
4\. The question whether all smashing weight structures on a given compactly generated category ${\underline{C}}$ are compactly generated is a certain weight structure version of the (generalized) telescope conjecture (that is also sometimes called the smashing conjecture) for ${\underline{C}}$; this question generalizes its “usual” stable version (see Proposition \[prtst\](\[it4sm\])). As we have noted in Remark \[rnondeg\](3), the main result of [@kellerema] demonstrates that the answer to the shift-stable version of the question is negative for a general ${\underline{C}}$; hence this is only more so for our weight structure version. On the other hand, the answer to our question for ${\underline{C}}=SH$ (the topological stable homotopy category) is not clear.
5\. The description of compact objects in ${\mathcal{LO}}$ provided by part \[iclass2\] of our theorem is important for the continuity arguments in [@bcons].
On perfectly generated weight structures and symmetrically generated $t$-structures {#sperfws}
-----------------------------------------------------------------------------------
Now we prove that an arbitrary cosuspended countably perfect set ${\mathcal{P}}$ gives a weight structure; this is an interesting modification of Theorem \[tclass\](\[iclass1\]) (note that any class of compact objects is perfect).
\[tpgws\] Let ${\mathcal{P}}$ be a countably perfect (see Definition \[dwg\]) cosuspended (i.e., ${\mathcal{P}}[-1]\subset {\mathcal{P}}$) set of objects of ${\underline{C}}$. Then the strong extension-closure ${\mathcal{LO}}$ of ${\mathcal{P}}$ and ${\mathcal{RO}}={\mathcal{P}}^\perp$ give a weighty torsion pair for ${\underline{C}}$ (i.e., $({\mathcal{LO}},{\mathcal{RO}}[-1])$ is a weight structure). Moreover, ${\mathcal{LO}}$ equals the big hull of ${\mathcal{P}}$.
As we have noted in Remark \[rwhop\](1), $s$ is weighty whenever $s$ is a torsion pair and ${\mathcal{LO}}\subset {\mathcal{LO}}[1]$; the latter certainly follows from ${\mathcal{P}}[-1]\subset {\mathcal{P}}$.
So, we should prove the remaining assertions. Repeating the beginning of the proof of Theorem \[tclass\](\[iclass1\]), we reduce them to the existence for any $M\in {\operatorname{Obj}}{\underline{C}}$ an $s$-decomposition $L_sM\to M\to R_sM\to L_sM[1]$ with $L_sM$ belonging to the naive big hull of ${\mathcal{P}}$.
For this purpose we construct a distinguished triangle $L\stackrel{b}{\to} M\stackrel{a}{\to} R\stackrel{f}{\to} L[1]$ using the method described in the proof of Theorem \[tclass\]; so, $L$ belongs to the naive big hull of ${\mathcal{P}}$ by construction. To finish the proof we should check that $R\in {\mathcal{RO}}$; for this purpose we “mix” the proof of Theorem \[tclass\](\[iclass1\]) with that of [@kraucoh Theorem A]; cf. also Remark \[rigid\](1) below. The idea is to replace the collection of functors $H^P$ for $P\in {\mathcal{P}}$ with a single “more complicated” functor (that would be a wcc one in contrast with the functors $H^P$ in the case of “general” $P$).
We will write ${\underline{\coprod}{\mathcal{P}}}$ for the full subcategory of ${\underline{C}}$ formed by the $\coprod$-closure of ${\mathcal{P}}$; following [@kraucoh] (see also [@neebook Definition 5.1.3] and [@auscoh]) we consider the full subcategory ${\operatorname{Coh}_{{\mathcal{P}}}}\subset {\operatorname{PShv}}^{{\mathbb{Z}}}(\underline{\coprod}{\mathcal{P}})=\operatorname{\operatorname{AddFun}}(({\underline{\coprod}{\mathcal{P}}})^{op},{\underline{\operatorname{Ab}}})$ (cf. Remark \[rdetect\]; we will omit the index ${{\mathbb{Z}}}$ in this notation below) of [*coherent functors*]{}. We recall (see [@kraucoh]) that a functor $H\in {\operatorname{Obj}}{\operatorname{PShv}}({\underline{\coprod}{\mathcal{P}}})$ is said to be coherent whenever there exists a ${\operatorname{PShv}}({\underline{\coprod}{\mathcal{P}}})$-short exact sequence ${\underline{\coprod}{\mathcal{P}}}(-,X)\to {\underline{\coprod}{\mathcal{P}}}(-,Y)\to H\to 0$, where $X$ and $Y$ are some objects of ${\underline{\coprod}{\mathcal{P}}}$ (note that this is a projective resolution of $H$ in ${\operatorname{PShv}}({\underline{\coprod}{\mathcal{P}}})$; see [@neebook Lemma 5.1.2]).
According to [@kraucoh Lemma 2], the category ${\operatorname{Coh}_{{\mathcal{P}}}}$ is abelian; it has coproducts according to Lemma 1 of ibid. Since any morphisms of (coherent) functors is compatible with some morphism of their (arbitrary) projective resolutions, a ${\operatorname{Coh}_{{\mathcal{P}}}}$-morphism is zero (resp. surjective) if and only if it is surjective in ${\operatorname{PShv}}({\underline{\coprod}{\mathcal{P}}})$.
Next, the Yoneda correspondence ${\underline{C}}\to {\operatorname{PShv}}({\underline{\coprod}{\mathcal{P}}})$ (sending $M\in {\operatorname{Obj}}{\underline{C}}$ to the restriction of ${\underline{C}}(-,M)$ to ${\underline{\coprod}{\mathcal{P}}}$) gives a homological functor $H^{{\mathcal{P}}}:{\underline{C}}\to {\operatorname{Coh}_{{\mathcal{P}}}}$ (see Lemma 3 of ibid.). $H^{{\mathcal{P}}}$ is a wcc functor since ${\mathcal{P}}$ is countably perfect (according to that lemma); it also respects arbitrary ${{\underline{\coprod}{\mathcal{P}}}}$-coproducts (very easy; see Lemma 1 of ibid.). Lastly, our discussion of zero and surjective ${\operatorname{Coh}_{{\mathcal{P}}}}$-morphisms certainly yields that $H^{{\mathcal{P}}}(h)$ is zero (resp. surjective) for $h$ being a ${\underline{C}}$-morphism if and only if ${\underline{C}}(N,-)(h)=0$ (resp. surjective) for any $N\in {\operatorname{Obj}}{{\underline{\coprod}{\mathcal{P}}}}$; it is certainly suffices to take $N\in {\mathcal{P}}$ in these “criteria” only.
Now we prove that $R\in {\mathcal{RO}}$ using the notation introduced in the proof of Theorem \[tclass\](\[iclass1\]) (see (\[etrproof\])). As we have just proved, $R\in {\mathcal{RO}}$ whenever $H^{{\mathcal{P}}}(R)=0$. Hence the long exact sequence $$\to H^{{\mathcal{P}}}(L)\stackrel{H^{{\mathcal{P}}}(b)} \to H^{{\mathcal{P}}}(M) \to H^{{\mathcal{P}}}(R[1])\to H^{{\mathcal{P}}}(L[1])\stackrel{H^{{\mathcal{P}}}(b[1])} \to H^{{\mathcal{P}}}(M[1])\to$$ reduces the assertion to the surjectivity of $H^{{\mathcal{P}}}(b)$ along with the injectivity of $H^{{\mathcal{P}}}(b[1])$.
Next, the vanishing of ${\underline{C}}(P,-)(h_k)$ for all $k\ge 0$ and $P\in {\mathcal{P}}$ implies that the morphisms $H^{{\mathcal{P}}}(h_k)$ are zero also. Thus the corresponding argument used in the proof of Theorem \[tclass\](\[iclass1\]) would carry over to our setting (to yield the assertion) if we knew that $H^{{\mathcal{P}}}(L[1])\cong \operatorname{\varinjlim}H^{{\mathcal{P}}}(L_k[1])$ and $H^{{\mathcal{P}}}(L)$ surjects onto $ \operatorname{\varinjlim}H^{{\mathcal{P}}}(L_k)$.[^48] Now, the surjectivity in question follows immediately from Lemma \[lcoulim\](3). To prove the injectivity statement it suffices to verify that condition (i) of the lemma is fulfilled (for $Y_i=L_i[1]$, $H'=H^{{\mathcal{P}}}$, and some $A,A_i\in {\operatorname{Obj}}{\operatorname{Coh}_{{\mathcal{P}}}}$). Now for all $k\ge 0$ consider the following morphism of ${\operatorname{Coh}_{{\mathcal{P}}}}$-exact sequences: \[1.0\] [$\begin{CD} H^{{\mathcal{P}}}(M[1])@>{H^{{\mathcal{P}}}(g_k[1])}>> H^{{\mathcal{P}}}(M_k[1])@>{H^{{\mathcal{P}}}(f_k[1])}>>H^{{\mathcal{P}}}(L_k[2])@>{H^{{\mathcal{P}}}(b_k[2])}>>H^{{\mathcal{P}}}(M[2])@>{H^{{\mathcal{P}}}(g_k[2])}>>H^{{\mathcal{P}}}(M_k[2]) \\
@VV{=}V @VV{H^{{\mathcal{P}}}(h_k[1])}V@VV{H^{{\mathcal{P}}}(s_k[2])}V@VV{=}V @VV{H^{{\mathcal{P}}}(h_k[2])}V \\
H^{{\mathcal{P}}}(M[1])@>{H^{{\mathcal{P}}}(g_{k+1}[1])}>> H^{{\mathcal{P}}}(M_{k+1}[1])@>{H^{{\mathcal{P}}}(f_{k+1}[1])}>>H^{{\mathcal{P}}}(L_{k+1}[2])@>{H^{{\mathcal{P}}}(b_{k+1}[2])}>>H^{{\mathcal{P}}}(M[2]) @>{H^{{\mathcal{P}}}(g_{k+1}[2])}>>H^{{\mathcal{P}}}(M_{k+1}[2])
\end{CD}$]{} Note that functors $H^{{\mathcal{P}}}\circ [1]$ and $H^{{\mathcal{P}}}\circ [2]$ can be expressed in terms of restricting functors represented by objects of ${\underline{C}}$ to ${\underline{\coprod}{\mathcal{P}}}[-1]$ and ${\underline{\coprod}{\mathcal{P}}}[-2]$, respectively. Since these categories lie in ${\underline{\coprod}{\mathcal{P}}}$, we have $H^{{\mathcal{P}}}(h_k[1])=0=H^{{\mathcal{P}}}(h_k[2])$.[^49] It follows that $H^{{\mathcal{P}}}(g_{k+1}[1])=0=H^{{\mathcal{P}}}(g_{k+1}[2])$ for all $k\ge 0$. Hence for $k\ge 1$ the morphism $H^{{\mathcal{P}}}(f_{k+1}[1])\bigoplus H^{{\mathcal{P}}}(s_{k}[1])$ gives an isomorphism of $\operatorname{\operatorname{Im}}(b_k[2])\bigoplus H^{{\mathcal{P}}}(M_{k+1}[1])$ with $H^{{\mathcal{P}}}(M_{k+1}[1])$, whereas $\operatorname{\operatorname{Im}}(b_k[2])$ is isomorphic to $H^{{\mathcal{P}}}(M[2])$. Thus for $A=H^{{\mathcal{P}}}(M[2])$ and $A_i=H^{{\mathcal{P}}}(M_{i}[1])$ we have $H^{{\mathcal{P}}}(L_i[2])\cong A_i\bigoplus A$ for $i\ge 2$ and these isomorphisms are compatible with $H^{{\mathcal{P}}}(s_i[2])\cong(0:A_i\to A_{i+1}) \bigoplus {\operatorname{id}}_A$. Hence Lemma \[lcoulim\](3)(i) implies that $H^{{\mathcal{P}}}(L[1])\cong \operatorname{\varinjlim}H^{{\mathcal{P}}}(L_k[1])$.
\[rigid\] 1. The author was inspired to apply coherent functors in this context by [@salorio]; yet the proof of Theorem 2.2 of ibid. (where coherent functors are applied to the construction of $t$-structures) appears to contain a gap.[^50] The author believes that applying arguments of the sort used in the proof of our theorem in the case of a “general” (countably) perfect set ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ such that ${\mathcal{P}}\subset {\mathcal{P}}[1]$ one can (only) obtain a “semi-$t$-structure” for ${\underline{C}}$, i.e., for any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $L\to M\to R\to L[1]$ such that $L$ belongs to the big hull of ${\mathcal{P}}$ and $R\in {\mathcal{P}}{{}^{\perp}}[1]$.[^51] The author wonders whether this result can be improved, and also whether semi-$t$-structures can be “useful”.[^52] Note also that in Theorem \[tsymt\] below we will prove the existence of a $t$-structure generated by a suspended ${\mathcal{P}}$ whenever ${\mathcal{P}}$ is symmetric to some set ${\mathcal{P}}'\subset {\operatorname{Obj}}{\underline{C}}$; however, the author does not know whether ${\underline{C}}^{t\le 0}$ equals the big hull of ${\mathcal{P}}$ whenever ${\mathcal{P}}$ satisfies these conditions.
2\. The arguments from the proof of our theorem can also be used (and significantly simplified) if instead of requiring ${\mathcal{P}}$ being cosuspended (and countably perfect) we assume that ${{\underline{\coprod}{\mathcal{P}}}}$ is [*rigid*]{}, i.e., ${{\underline{\coprod}{\mathcal{P}}}}\perp {{\underline{\coprod}{\mathcal{P}}}}[1]$. Indeed, then the distinguished triangle $P_0\to M\to M_1\to P_0[1]$ (see the proof of Theorem \[tclass\](\[iclass1\])) is easily seen to be an $s$-decomposition of $M$ (and if we proceed as above then this triangle will actually be equal to $L\to M\to R\to L[1]$).
Note also that ${{\underline{\coprod}{\mathcal{P}}}}$ is rigid if and only if ${\mathcal{P}}\perp {{\underline{\coprod}{\mathcal{P}}}}[1]$. Moreover, if ${\mathcal{P}}$ is perfect then these conditions are equivalent to ${\mathcal{P}}\perp {\mathcal{P}}[1]$.
The author does not know whether any formulation of this sort is known.
3\. The case ${\mathcal{P}}={\mathcal{P}}[1]$ of our theorem (cf. Remark \[rtst2\]) was essentially treated in the proof of [@kraucoh Theorem A].
4\. We will say that a weighty torsion pair and the corresponding weight structure are perfectly generated whenever they can be obtained by means of our theorem. Remark \[revenmorews\](1) below will give a “more natural” equivalent of this definition. Note also that instead of assuming that ${\mathcal{P}}$ is a set in the theorem it certainly suffices to assume that ${\mathcal{P}}$ is essentially small.
Moreover, Theorem \[twgws\](III.2) states that any smashing weight structure on a well generated triangulated category is perfectly generated. 5. In Theorem \[tsymt\] and Corollary \[csymt\] below we will study examples for Theorem \[tpgws\] that are constructed using “symmetry”; this will yield some new results on $t$-structures. The idea to relate $t$-structures to symmetric sets and Brown-Comenetz duals comes from [@salorio] also; however the author doubts that one can get a “simple description” of a $t$-structure obtained using arguments of this sort (cf. Corollary 2.5 of ibid.).
Now we prove a few simple definitions and statements related to countably perfect classes (without claiming much originality in these results); recall that ${\mathcal{P}}$-null morphisms (for ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$) are the ones annihilated by the functors $H^P$ for all $P\in {\mathcal{P}}$ (see Definition \[dhopo\](5)).
\[dapprox\] Let ${\mathcal{P}}$ be a subclass of ${\operatorname{Obj}}{\underline{C}}$, $h\in {\underline{C}}(M,N)$ (for some $M,N\in {\operatorname{Obj}}{\underline{C}}$).
1\. We will say that $h$ is a [*${\mathcal{P}}$-epic*]{} whenever for any $P\in {\mathcal{P}}$ the homomorphism $H^P(h)$ is surjective, i.e., if any $g\in {\underline{C}}(P,N)$ for $P\in {\mathcal{P}}$ factors through $h$. We will say that $h$ is [*${\mathcal{P}}$-monic*]{} if all $H^P(h)$ are injective.[^53]
2\. We will say that $h$ is a [*${\mathcal{P}}$-approximation*]{} (of $N$) if it is a ${\mathcal{P}}$-epic and $M$ belongs to ${\operatorname{Obj}}{\underline{\coprod}{\mathcal{P}}}$.
3\. We will say that ${\mathcal{P}}$ is [*contravariantly finite* ]{} (in ${\underline{C}}$) if for any $N\in {\operatorname{Obj}}{\underline{C}}$ there exists its ${\mathcal{P}}$-approximation.[^54]
\[lperf\] Let ${\mathcal{P}}$ be a subclass of ${\operatorname{Obj}}{\underline{C}}$; denote by ${{\tilde{\mathcal{P}}}}$ the Karoubi-closure of ${\underline{\coprod}{\mathcal{P}}}$ in ${\underline{C}}$.
1. \[ipercl\] If ${\mathcal{P}}'$ is a subclass of ${{\tilde{\mathcal{P}}}}$ containing ${\mathcal{P}}$ then a ${\underline{C}}$-morphism $h$ is ${\mathcal{P}}'$-null (resp. a ${\mathcal{P}}'$-epic, resp. a ${\mathcal{P}}$-approximation) if and only if it is ${\mathcal{P}}$-null (resp. a ${\mathcal{P}}$-epic, resp. a ${\mathcal{P}}$-approximation).
2. \[iperot\] In a ${\underline{C}}$-distinguished triangle $M\stackrel{h}{\to}N \stackrel{f}{\to}Q\stackrel{g}{\to} M[1]$ the morphism $h$ is a ${\mathcal{P}}$-epic if and only if $f$ is ${\mathcal{P}}$-null; this is also equivalent to $g$ being ${\mathcal{P}}$-monic.
3. \[ipereq\] The class ${\mathcal{P}}$ is (countably) perfect if and only if any (countable) coproduct of ${\mathcal{P}}$-epic morphisms is ${\mathcal{P}}$-epic.
Moreover, if this is the case then any (countable) coproduct of ${\mathcal{P}}$-approximations is a ${\mathcal{P}}$-approximation.
4. \[ipertest\] If $h:M\to N$ is a ${\mathcal{P}}$-approximation of $N$ then a ${\underline{C}}$-morphism $g:N\to N'$ is ${\mathcal{P}}$-null if and only if $g\circ h=0$.
5. \[ipercrit\] Assume that for any (countable) collection of $N_i\in {\operatorname{Obj}}{\underline{C}}$ the object $\coprod N_i$ possesses a ${\mathcal{P}}$-approximation being the coproduct of some ${\mathcal{P}}$-approximations of $M_i$. Then ${\mathcal{P}}$ is (countably) perfect.
6. \[iperfcovarf\] If ${\mathcal{P}}$ is a set then it is contravariantly finite.
7. \[iperloc\] Let $F:{\underline{D}}\to {\underline{C}}$ be an exact functor that possesses a right adjoint $G$ respecting (countable) coproducts. Then for any (countably) perfect class ${\mathcal{P}}'$ of objects of ${\underline{D}}$ the class ${\mathcal{P}}=F({\mathcal{P}}')$ is (countably) perfect also.
All of the assertions are rather easy.
\[ipercl\], \[iperot\]: obvious.
Assertion \[ipereq\] follows from assertion \[iperot\] immediately according to Proposition \[pcoprtriang\].
\[ipertest\]. Since $M$ is a coproduct of elements of ${\mathcal{P}}$, the composition of $h$ with any ${\mathcal{P}}$-null morphism is zero. Conversely, since any morphism from ${\mathcal{P}}$ into $N$ factors through $h$, if $g\circ h=0$ then $g$ is ${\mathcal{P}}$-null.
\[ipercrit\]. Let $f_i:N_i\to P_i$ be some ${\mathcal{P}}$-null morphisms; choose ${\mathcal{P}}$-approximations $h_i:M_i\to N_i$ such that $\coprod h_i$ is a ${\mathcal{P}}$-approximation of $\coprod N_i$. Then $\coprod f_i \circ \coprod h_i=\coprod (f_i\circ h_i)=0$. Hence $\coprod f_i$ is ${\mathcal{P}}$-null according to assertion \[ipertest\].
\[iperfcovarf\]. Easy and standard: a ${\mathcal{P}}$-approximation of $M\in {\operatorname{Obj}}{\underline{C}}$ is given by $\coprod_{P,h_P}\stackrel{\bigoplus h_P}{\to}M$, where $P$ runs through all elements of ${\mathcal{P}}$ and $h_P$ runs through ${\underline{C}}(P,M)$.
\[iperloc\]. The adjunction immediately yields that a ${\underline{C}}$-morphism $h$ is ${\mathcal{P}}$-null if and only if $G(h)$ is ${\mathcal{P}}'$-null. It remains to recall that $G$ respects (countable) coproducts.
\[requivdef\] Our definition of perfect classes essentially coincides with the one used in [@modoi]. Moreover, part \[ipereq\] of the lemma gives the equivalence of our definition (\[dcomp\](\[idpc\])) of countably perfect classes to conditions (G1) and (G2) in the definition of perfect generators in [@kraucoh]; hence a category is perfectly generated in the sense of ibid. if and only if it is so in the sense of Definition \[dwg\]. Similarly, our condition \[isym\] in Definition \[dsym\] is equivalent to condition (G3) in [@kraucoh Definition 2]; hence ${\underline{C}}$ is symmetrically generated in the terms of loc.cit. whenever it has products and contains a set ${\mathcal{P}}$ that Hom-generates ${\underline{C}}$ and is symmetric to some set ${\mathcal{P}}'\subset {\operatorname{Obj}}{\underline{C}}$.
Furthermore, any class that is $\aleph_1$-perfect in the sense of [@neebook Definition 3.3.1] is countably perfect. We also obtain that our definition of ${\alpha}$-well generated categories is equivalent to the one given in [@krauwg]. Moreover, recall that the latter definition is equivalent to the definition given in [@neebook] according to Theorem A of ibid.
We deduce some immediate consequences from the lemma.
\[cwftw\] 1. Assume that ${\mathcal{P}}$ is a (countably) perfect cosuspended set of objects of ${\underline{C}}$. Then the weight structure $w$ constructed in Theorem \[tpgws\] is (countably) smashing. 2. Assume that $\{{\mathcal{P}}_i\}$ is a set of (countably) perfect sets of objects of ${\underline{C}}$. Then the couple $w=({\underline{C}}_{w\le 0}, {\underline{C}}_{w\ge 0})$ is a (countably) smashing weight structure on ${\underline{C}}$, where ${\underline{C}}_{w\le 0}$ is the big hull of $\cup_{j\ge 0,i} {\mathcal{P}}_i[-j]$ and ${\underline{C}}_{w\ge 0}=\cap_{j\ge 1,i} ({\mathcal{P}}_i{{}^{\perp}}[-j])$.
3\. Assume that $\{w_i\}$ is a set of perfectly generated weight structures on ${\underline{C}}$, i.e., assume that there exist countably perfect sets ${\mathcal{P}}_i\subset {\operatorname{Obj}}{\underline{C}}$ that generate $w_i$ (see Remark \[rwhop\](1)). Then the couple $w=({\underline{C}}_{w\le 0}, {\underline{C}}_{w\ge 0})$ is a weight structure, where ${\underline{C}}_{w\le 0}$ is the big hull of $\cup_i {\underline{C}}_{w_i\le 0}$ and ${\underline{C}}_{w\ge 0}=\cap_i {\underline{C}}_{w_i\ge 0}$. Moreover, $w$ is perfectly generated in the sense of Remark \[rigid\](4); it is smashing whenever all $w_i$ are.
1\. Recall that we should check whether ${\underline{C}}_{w\ge 0}={\mathcal{P}}{{}^{\perp}}[-1]$ is closed with respect to small (resp. countable) ${\underline{C}}$-coproducts. Hence the statement follows immediately from Proposition \[psym\](\[isymeu\]). 2. $\cup_{j\ge 0;i} {\mathcal{P}}_i[-j]$ is a (countably) perfect set according to Proposition \[psym\](\[isymuni\]); it is certainly cosuspended. Hence $w$ is a weight structure according to Theorem \[tpgws\]. Lastly, the smashing property statements follow immediately from the previous assertion.
3\. According to the previous assertion, the couple $({\underline{C}}_{w'\le 0}, {\underline{C}}_{w'\ge 0})$ is a weight structure on ${\underline{C}}$, where ${\underline{C}}_{w'\le 0}$ is the big hull of $\cup_{j\ge 0,i} {\mathcal{P}}_i[-j]$ and ${\underline{C}}_{w'\ge 0}=\cap_{j\ge 1;i} ({\mathcal{P}}_i{{}^{\perp}}[-j])$. Now we compare $w$ with $w'$. Since ${\mathcal{P}}_i$ generate $w_i$, we certainly have ${\underline{C}}_{w\ge 0}={\underline{C}}_{w'\ge 0}$. Next, ${\underline{C}}_{w\le 0}\perp {\underline{C}}_{w\ge 0}[1]$ according to Lemma \[lbes\](\[isesperp\]). Since ${\underline{C}}_{w\le 0}$ contains ${\underline{C}}_{w'\le 0}$, these classes are equal. Thus $w$ is a perfectly generated weight structure. It is smashing if all $w_i$ are; indeed, ${\underline{C}}_{w\ge 0}$ is coproductive as being the intersection of coproductive classes.
\[revenmorews\] 1. In particular, we obtain that $w$ is perfectly generated in the sense of Remark \[rigid\](4) if and only if it is generated by a countably perfect set (i.e., if ${\mathcal{RO}}={\underline{C}}_{w\le -1}$ equals ${\mathcal{P}}^\perp$ for some countably perfect set ${\mathcal{P}}$). Note here that we do not have to assume that ${\mathcal{P}}$ is cosuspended, since $\cup_{i\le 0}{\mathcal{P}}[i]$ generates $w$ whenever ${\mathcal{P}}$ does.
Moreover, part 3 of the corollary gives a certain “join” operation on perfectly generated $t$-structures (and so, we obtain a monoid). Note also that the join of any set of smashing weight structures is smashing also.
2\. Thus our corollary gives a vast source of smashing weight structures. Now, the results of §\[sadjt\] allow to construct “new” $t$-structures that are right adjacent to these weight structures (cf. Remark \[rnewt\](1)) and also describe their hearts. Note also that for $t_i$ being right adjacent to (perfectly generated) $w_i$ (for $i\in I$) and the corresponding $w$ and $t$ we obviously have ${\underline{C}}^{t\ge 0}=\cap_{i\in I} {\underline{C}}^{t_i\ge 0}$, whereas ${\underline{C}}^{t\le 0}$ is the big hull of $\cup_{i\in I} {\underline{C}}^{t_i\le 0}$.
Now we prove that [**suspended**]{} symmetric sets generate $t$-structures.
\[tsymt\] Assume that ${\underline{C}}$ (also) has products; for a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ assume that ${\mathcal{P}}$ is [*suspended*]{} (i.e., ${\mathcal{P}}[1]\subset {\mathcal{P}}$) and ${\mathcal{P}}$ is symmetric (see Definition \[dsym\](\[isym\])) to a [**set**]{} ${\mathcal{P}}'$ (of objects of ${\underline{C}}$). Then the following statements are valid.
1\. There exists a $t$-structure $t$ on ${\underline{C}}$ such that ${\underline{C}}^{t\ge 1}={\mathcal{P}}{{}^{\perp}}$ (i.e., $t$ is generated by ${\mathcal{P}}$) and also a cosmashing weight structure $w$ that is right adjacent to $t$.
2\. ${{\underline{Ht}}}$ has an injective cogenerator and satisfies the AB3\* axiom. Moreover, ${{\underline{Ht}}}$ is naturally anti-equivalent to the subcategory of $\operatorname{\operatorname{AddFun}}({{\underline{Hw}}},{\underline{\operatorname{Ab}}})$ consisting of those functors that respect products; ${{\underline{Hw}}}$ is naturally equivalent to the subcategory of injective objects of ${{\underline{Ht}}}$.
3\. If ${\mathcal{P}}'$ is suspended also then ${\underline{C}}_{w\ge 0}$ equals the big hull of ${\mathcal{P}}'$ in ${\underline{C}}{{^{op}}}$.
Firstly we note that ${\mathcal{P}}$ is also symmetric to $\cup_{i\ge 0}{\mathcal{P}}'[i]$ (see Proposition \[psymb\](I.\[iws1\])); hence it suffices to consider the case where ${\mathcal{P}}'$ is suspended.
Now we adopt the notation of Proposition \[psymb\](I.\[isym4\]). Assume first that ${\underline{D}}={\{0\}}$ (in the notation of the proposition), and so, ${\underline{C}}'={\underline{C}}$. According to (parts I.\[iwsym1\] and I.\[iws2\] of) the aforementioned proposition, the set ${\mathcal{P}}'$ is perfect in the category ${\underline{C}}{{^{op}}}$. Since ${\mathcal{P}}'$ is certainly cosuspended in ${\underline{C}}{{^{op}}}$, Theorem \[tpgws\] yields a smashing weight structure $w_{{\mathcal{P}}'}^{op}$ that is ${\underline{C}}{{^{op}}}$-generated by ${\mathcal{P}}'$ with ${\underline{C}}{{^{op}}}_{w^{op}\le 0}$ being the ${\underline{C}}{{^{op}}}$-big hull of ${\mathcal{P}}'$. The corresponding weight structure $w_{{\mathcal{P}}'}$ on ${\underline{C}}$ (see Proposition \[pbw\](\[idual\])) will be our candidate for $w$. $w_{{\mathcal{P}}'}$ is certainly cosmashing; hence ${{\underline{Hw}}}_{{\mathcal{P}}'}$ is Karoubian (easy from Remark \[rcoulim\](4)) and closed with respect to ${\underline{C}}$-products. Moreover, ${{\underline{Hw}}}_{{\mathcal{P}}'}$ has a [*cogenerator*]{}, i.e., any its object is a retract of a product of (copies of) some $M\in {\underline{C}}_{{{\underline{Hw}}}_{{\mathcal{P}}'}=0}$; here we apply (the dual to) Corollary \[cvttbrown\](I.2).
Next we apply Corollary \[cdualt\] to obtain that there exists a $t$-structure $t_{{\mathcal{P}}'}$ on ${\underline{C}}$ that is left adjacent to $w_{{\mathcal{P}}'}$. Moreover, ${{\underline{Ht}}}_{{\mathcal{P}}'}$ is anti-equivalent to the subcategory of $\operatorname{\operatorname{AddFun}}({{\underline{Hw}}}_{{\mathcal{P}}'},{\underline{\operatorname{Ab}}})$ consisting of those functors that respect products (according to the corollary); ${{\underline{Hw}}}_{{\mathcal{P}}'}$ is equivalent to the subcategory of injective objects of ${{\underline{Ht}}}_{{\mathcal{P}}'}$. Hence ${{\underline{Ht}}}_{{\mathcal{P}}'}$ has an injective cogenerator; it satisfies the AB3\* axiom since ${{\underline{Hw}}}_{{\mathcal{P}}'}$ has products.
So to finish the proof in this case it remains to note that $t_{{\mathcal{P}}'}$ is precisely the $t$-structure generated by ${\mathcal{P}}$ according to Proposition \[pwsym\](\[iws4\]).
Now we proceed to the general case of our setting (using Proposition \[psymb\](I.\[isym4\]) also). We recall that the corresponding category ${\underline{D}}'$ is closed with respect to ${\underline{C}}$-products and ${\underline{D}}'^{op}$ is perfectly generated by ${\mathcal{P}}'$. Thus (by Theorem \[tpgws\]) there exists a weight structure $w_{{\underline{D}}'}$ on ${\underline{D}}'$ with ${\underline{D}}'_{w_{{\underline{D}}'}\le 0}$ equal to the ${\underline{D}}'^{op}$-big hull of ${\mathcal{P}}'$ and ${\underline{D}}'_{w_{{\underline{D}}'}\ge 0}=({}^{\perp_{{\underline{D}}'}} {\mathcal{P}}')[1]$. Once again, we apply Corollary \[cdualt\] to obtain a $t$-structure $t_{{\underline{D}}'}$ on ${\underline{D}}$ that is left adjacent to $w_{{\underline{D}}'}$. We also obtain that ${{\underline{Ht}}}_{{\underline{D}}'}$ and ${{\underline{Hw}}}_{{\underline{D}}'}$ are related similarly to assertion 2.
Now we “extend” $t_{{\underline{D}}'}$ and ${{\underline{Hw}}}_{{\underline{D}}'}$ to ${\underline{C}}$. Recall that the embedding ${\underline{D}}'\to {\underline{C}}$ respects products and possesses a left adjoint; hence (the dual to) Remark \[rwhop\](2) gives the existence of a weight structure $w_{{\mathcal{P}}'}$ as above. Next, we consider the equivalence $j:{\underline{D}}'\to {\underline{C}}'$ induced by the functor $G$ (that is right adjoint to $i:{\underline{C}'}\to{\underline{C}}$) and denote by $t_{{\underline{C}'}}$ the $t$-structure on ${\underline{C}'}$ obtained from $t_{{\underline{D}}'}$ via $j$. Since the embedding $i:{\underline{C}'}\to {\underline{C}}$ possesses an (exact) right adjoint, we obtain (using Proposition \[phopft\](II.1)) that $({\underline{C}'}^{t_{{\underline{C}'}}\le 0},({\underline{C}'}^{t_{{\underline{C}'}}\le 0}){{}^{\perp}}[1])$ is a $t$-structure $t$ on ${\underline{C}}$. Note also that ${{\underline{Ht}}}$ equals $j({{\underline{Ht}}}_{{\underline{D}}'})$, ${{\underline{Hw}}}_{{\mathcal{P}}'}={{\underline{Hw}}}_{{\underline{D}}'}$, and ${\underline{C}}_{w_{{\mathcal{P}}'}\ge 0}={\underline{D}}'_{w_{{\underline{D}}'}\le 0}$; hence assertions 2 and 3 follow from assertion 1, and to prove the latter it suffices to verify that $({\underline{C}'}^{t_{{\underline{C}'}}\le 0}){{}^{\perp}}={\mathcal{P}}{{}^{\perp}}$. Since $i$ possesses a right adjoint, for the latter purpose one should compare ${\mathcal{P}}{{}^{\perp}}\cap {\operatorname{Obj}}{\underline{C}'}$ with $({\underline{C}'}^{t_{{\underline{C}'}}\le 0}){{}^{\perp}}\cap {\operatorname{Obj}}{\underline{C}'}={\underline{C}'}^{t_{{\underline{C}'}}\ge 1}=j({\mathcal{P}}')^{\perp_{{\underline{C}'}}}$ (here we apply Remark \[rwhop\](2)). It remains to recall that ${\mathcal{P}}$ is symmetric to $j({\mathcal{P}}')$ in ${\underline{C}'}$ and apply Proposition \[pwsym\](\[iws4\]) once again.
To demonstrate the relevance of our theorem, we apply it to the study of compactly generated $t$-structures.
\[csymt\] Let ${\mathcal{Q}}$ be a set of compact objects of ${\underline{C}}$.
1\. Then $t=({\underline{C}}^{t\le 0},{\underline{C}}^{t\ge 0})$ is a smashing $t$-structure on ${\underline{C}}$, where ${\underline{C}}^{t\le 0}$ is the smallest coproductive extension-closed subclass of ${\operatorname{Obj}}{\underline{C}}$ containing $\cup_{i\ge 0}{\mathcal{Q}}[i]$ and ${\underline{C}}^{t\ge 0}=\cap_{i\ge 1} {\mathcal{Q}}{{}^{\perp}}[i]$.[^55] Moreover, ${{\underline{Ht}}}$ has an injective cogenerator and satisfies the AB3\* axiom; it is anti-equivalent to the subcategory of $\operatorname{\operatorname{AddFun}}(({\operatorname{Inj}{{\underline{Ht}}}}){{^{op}}},{\underline{\operatorname{Ab}}})$ consisting of those functors that send ${\operatorname{Inj}{{\underline{Ht}}}}$-products into products of abelian groups.[^56] 2. Assume in addition that ${\underline{C}}$ satisfies the Brown representability condition. Then there exists a weight structure $w$ that is right adjacent to $t$ with ${{\underline{Hw}}}\cong {\operatorname{Inj}{{\underline{Ht}}}}$, and ${\underline{C}}_{w\ge 0}$ equals the ${\underline{C}}{{^{op}}}$-big hull of $\{\hat{Q}[i]_{{\underline{C}}}:\ Q\in {\mathcal{Q}}, i\ge 0\}$, where $\hat{Q}$ is the Brown-Comenetz dual of $Q$ (that represents the functor $M\mapsto {\underline{\operatorname{Ab}}}({\underline{C}}(Q,M),{{\mathbb{Q}}}/{{\mathbb{Z}}})$).[^57]
3\. Assume that the $t$-structure $t$ mentioned in assertion 1 is non-degenerate. Then ${{\underline{Ht}}}$ is Grothendieck abelian.
1\. $t$ is a $t$-structure on ${\underline{C}}$ according to Theorem A.1. of [@talosa]; it is certainly smashing. Next we take ${\underline{C}}'$ being the localizing subcategory of ${\underline{C}}$ generated by ${\mathcal{Q}}$. Then ${\mathcal{Q}}$ is symmetric to the set ${\mathcal{Q}}'_{{\underline{C}'}}$ of ${\underline{C}}'$-Brown-Comenetz duals of elements of ${\mathcal{Q}}$ according to Proposition \[psymb\](II.\[iws23\]). Hence we can apply Theorem \[tsymt\] to the category ${\underline{C}}'$ with the corresponding ${\mathcal{P}}$ and ${\mathcal{P}}'$ being equal to $\cup_{i\ge 0}{\mathcal{Q}}[i]$ and to $\cup_{i\ge 0}{\mathcal{Q}}'_{{\underline{C}'}}[i]$, respectively. This yields the result since for the corresponding $t_{{\underline{C}}'}$ we have ${\underline{C}}^{t\le 0}={\underline{C}}'^{t_{{\underline{C}}'}\le 0}$ and ${\underline{C}}^{t\ge 0}\cap {\operatorname{Obj}}{\underline{C}}'={\underline{C}}'^{t_{{\underline{C}}'}\ge 0}$ (here we note that the embedding ${\underline{C}}'\to {\underline{C}}$ has a right adjoint according to Proposition \[pcomp\](II.2), and apply Remark \[rwhop\](2)).
2\. Once again, it suffices to combine Proposition \[psymb\](II.\[iws23\]) with Theorem \[tsymt\].
3\. Since $t$ is non-degenerate, the set ${\mathcal{Q}}$ Hom-generates ${\underline{C}}$. Hence the previous assertion gives the existence of $w$ that is right adjacent to $t$.
Now to prove the result it suffices to repeat the argument used in the proof of [@humavit Corollary 4.9]. We do so briefly here (without recalling the corresponding definitions).
Firstly, the corresponding shifts of the classes ${\underline{C}}^{t\le 0}$, ${\underline{C}}^{t\ge 0}={\underline{C}}_{w\le 0}$, and ${\underline{C}}_{w\ge 0}$ give a [*cosuspended TTF triple*]{}; see Definition 2.3 of ibid.[^58] Next, ${\underline{C}}_{w\le 0}$ is definable in the sense of Definition 4.1 (since it is the zero class of the set $\{{\underline{C}}(Q[i],-),\ Q\in {\mathcal{Q}},\ i>0\}$ of coherent functors; see Lemma \[lbes\](\[izs\]) for the definition of zero classes). Applying Theorem 4.8 of ibid. we obtain that ${\underline{C}}^{t\le 0}={{}^{\perp}}\{I[j],\ j< 0\}$ for some [*pure-injective cosilting object*]{} $I$ of ${\underline{C}}$ (this is where we use the non-degeneracy assumption on $t$). Thus it remains to apply Theorem 3.6 of ibid.
\[rsymt\]
1. \[irsymt2\] The author does not know how to deduce the existence of $w$ (in part 2 of our corollary) from the results of §\[sadjw\]. Another interesting fact related to this statement is Theorem 3.11 of [@postov] (where algebraic triangulated categories were considered); cf. Remark \[rwsym\](1) above.[^59] So, we generalize “the $t$-structure case” of loc. cit.; this immediately yields the corresponding generalization of [@humavit Corollary 4.9] (in part 3 of our corollary). Recall also that ${\underline{C}}^{t\le 0}={{}^{\perp}}\{I[j],\ j<0\}$ for some pure-injective cosilting object $I$ under the assumptions of our corollary (see its proof). It is easily seen that any cogenerator of ${{\underline{Hw}}}$ (see the proof of Theorem \[tsymt\]) can be taken for $I$ in this statement.
2. \[irsymtab5\] In Theorem \[tab5\] below we prove under certain restrictions on ${\underline{C}}$ (and without assuming that $t$ is non-degenerate) that ${{\underline{Ht}}}$ is an AB5 category; this argument is completely independent from ibid. (and gives some interesting additional information on ${{\underline{Ht}}}$). On the other hand, the generators of ${{\underline{Ht}}}$ given by part 1 of that theorem are the same as the ones given by the proof of [@humavit Theorem 3.6] (yet they were not specified explicitly in loc. cit.).
3. \[irsymt1\] It appears to be quite difficult to produce perfect and symmetric sets “out of nothing”; note that the existing literature on this subject mostly concentrates on the search of shift-stable sets of perfect generators of triangulated categories. So, the weight structures opposite to those given by Corollary \[csymt\](2) appear to be (essentially) the only known “type” of perfectly generated weight structures that are not compactly generated (yet cf. Remark \[rnondeg\](3)).
Let us prove that the corresponding weight structure $w^{op}$ is not compactly generated if the set ${\mathcal{Q}}$ Hom-generates ${\underline{C}}$ (certainly, this condition is fulfilled if and only if ${\mathcal{Q}}$ generates ${\underline{C}}$ as its own localizing subcategory). Indeed, in this case the symmetric set $\hat{Q}$ Hom-generates ${\underline{C}}^{op}$. Then Remark \[rwhop\](8) says that $w^{op}$ is left non-degenerate; thus any class generating $w^{op}$ also Hom-generates ${\underline{C}}^{op}$. On the other hand, ${\underline{C}}^{op}$ is not compactly generated according to Corollary E.1.3 (combined with Remark 6.4.5) of [@neebook].
Certainly, one can also consider the direct sum of an example of this sort with a “compactly generated” one.
4. \[irsymt4\] Probably the most interesting case of Theorem \[tsymt\] and Corollary \[csymt\] is the one where ${\mathcal{P}}$ (resp. ${\mathcal{Q}}$) Hom-generates ${\underline{C}}$; note that in this case ${\underline{C}}$ has products and satisfies the Brown representability condition automatically. Note also that ${\underline{C}}$ is Hom-generated by the corresponding set if and only if $t$ is right non-degenerate (see Definition \[dtstr\](\[ito3\]); this is certainly equivalent to the right non-degeneracy of $w$).
5. \[irsymt5\] It is easily seen that the Brown-Comenetz duals of any family $\{F_i\}$ of cc functors ${\underline{C}}\to {\underline{\operatorname{Ab}}}$ that are also pp ones form a perfect class in ${\underline{C}}{{^{op}}}$. Yet this observation can scarcely give any “new” weight structures since all “known” functors satisfying these conditions appear to be corepresented by compact objects of ${\underline{C}}$ (cf. [@krause Proposition 2.9]). Moreover, when we pass from the weight structure $w$ to its left adjacent $t$ we apply the dual Brown representability condition, whereas the latter says that all pp functors are corepresentable.
On well generated weight structures and torsion pairs {#swgws}
-----------------------------------------------------
Now we study the relation of (countably) perfect classes to torsion pairs and (especially) to weight structures. In particular, we obtain a complete “description” of smashing weight structures on well generated triangulated categories (see Theorem \[twgws\](III)). We will need some new definitions to deal with well generated categories. Most of them are simple variations of the notions described above; we also recall the notion of ${\beta}$-compact objects.
\[dbecomp\] Let ${\beta}$ be a regular infinite cardinal.
1. \[idclass\] We will say that a class ${{\tilde{\mathcal{P}}}}$ of objects of ${\underline{C}}$ is [*${\beta}$-coproductive*]{} if it is closed with respect to ${\underline{C}}$-coproducts of less than ${\beta}$ objects.
2. \[idchop\] We will say that a torsion pair $s=({\mathcal{LO}}',{\mathcal{RO}}')$ for a full triangulated subcategory ${\underline{C}}'$ of ${\underline{C}}$ is [*${\beta}$-coproductive*]{} if both ${\operatorname{Obj}}{\underline{C}}'$ and ${\mathcal{RO}}'$ are ${\beta}$-coproductive.
3. \[idcomp\] We will say that an object $M$ of ${\underline{C}}$ is [*${\beta}$-compact*]{} if it belongs to the maximal perfect class of ${\beta}$-small objects of ${\underline{C}}$ (whose existence is given by Proposition \[psym\](\[isymuni\])). We will write ${\underline{C}}^{{\beta}}$ for the full subcategory of ${\underline{C}}$ formed by ${\beta}$-compact objects.
\[rbecomp\] 1. Our definition of ${\beta}$-compact objects is equivalent to the one used in [@krauwg]. Indeed, coproducts of less than ${\beta}$ of ${\beta}$-small objects are obviously ${\beta}$-small; thus ${\operatorname{Obj}}{\underline{C}}^{\beta}$ is ${\beta}$-coproductive. Hence the equivalence of definitions follows from Lemma 4 of ibid. Furthermore, Lemma 6 of ibid. states that (both of) these definitions are equivalent to Definition 4.2.7 of [@neebook] if we assume in addition that ${\underline{C}}^{{\beta}}$ is essentially small.
2\. Now we recall some more basic properties of ${\beta}$-compact objects in an ${\alpha}$-well generated category ${\underline{C}}$ assuming that ${\beta}\ge {\alpha}$.
Theorem A of [@krauwg] yields immediately that ${\underline{C}}^{\beta}$ is an essentially small triangulated subcategory of ${\underline{C}}$.
Moreover, the union of ${\underline{C}}^{\gamma}$ for $\gamma$ running through all regular cardinals ($\ge {\alpha}$) equals ${\underline{C}}$ (see the Corollary in loc. cit. or Proposition 8.4.2 of [@neebook]). 3. Lastly, we recall a part of [@krauwg Lemma 4]. For any ${\beta}$-coproductive essentially small perfect class ${\mathcal{P}}$ of ${\beta}$-small objects of a triangulated category ${\underline{C}}$ (that has coproducts) it says the following: for any $P\in {\mathcal{P}}$ and any set of $N_i\in {\operatorname{Obj}}{\underline{C}}$ any morphism $P\to \coprod N_i$ factors through the coproduct of some ${\underline{C}}$-morphisms $M_i\to N_i$ with $M_i\in {\mathcal{P}}$.
\[twgws\] Let $s=({\mathcal{LO}},{\mathcal{RO}})$ be a torsion pair for ${\underline{C}}$ that is countably smashing, ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$. I. Consider the class $J$ of ${\underline{C}}$-morphisms characterized by the following condition: $h\in {\underline{C}}(M,N)$ (for $M,N\in {\operatorname{Obj}}{\underline{C}}$) belongs to $J$ whenever for any chain of morphisms $ L_sP\stackrel{a_P}{\to} P\stackrel{g}{\to}M \stackrel{h}{\to}N$ its composition is zero if $P\in {\mathcal{P}}$ and $a_P$ is an $s$-decomposition morphism.
Then the following statements are valid.
1. \[indep\] The class $J$ will not change if we will fix $a_P$ for any $P\in {\mathcal{P}}$ in this definition.
2. \[icontraf\] Assume that ${\mathcal{P}}$ is contravariantly finite and $s$ is smashing. Then $h$ belongs to $J$ if and only if there exists a ${\mathcal{P}}$-approximation morphism $AM\stackrel{g}{\to} M$ and an $s$-decomposition morphism $a_{AM}: L_sAM\to AM$ such that $h\circ g \circ a_{AM}=0$. Moreover, the latter is equivalent to the vanishing of all compositions of this sort.
3. \[icoprcl\] Assume that ${\mathcal{P}}$ is contravariantly finite and (countably) perfect and $s$ is smashing. Then the class $J$ is closed with respect to (countable) coproducts.
4. \[ilscp\] Assume that for any $P\in {\mathcal{P}}$ there exists a choice of $ L_sP\in {\mathcal{P}}$; denote the class of these choices by ${L_s\mathcal{P}}$. Then $J$ coincides with the class of ${L_s\mathcal{P}}$-null morphisms.
5. \[ilscper\] Assume in addition (to the previous assumption) that ${\mathcal{P}}$ is a (countably) perfect contravariantly finite class and $s$ is smashing. Then ${L_s\mathcal{P}}$ is a (countably) perfect contravariantly finite class also.
6. \[ilscperw\] Assume in addition that $s$ is weighty; assume that the class ${\mathcal{P}}$ is essentially small, contains ${\mathcal{P}}[-1]$, and Hom-generates ${\underline{C}}$. Then the class $L_s{\mathcal{P}}$ generates $s$ and ${\mathcal{LO}}$ is the big hull of $L_s{\mathcal{P}}$; thus $s$ is perfectly generated in the sense of Remark \[rigid\](4).
II\. For a regular cardinal ${\beta}$ let $s'=({\mathcal{LO}}',{\mathcal{RO}}')$ be a ${\beta}$-coproductive torsion pair for a full triangulated category ${\underline{C}'}$ of ${\underline{C}}$ such that ${\operatorname{Obj}}{\underline{C}'}$ is a perfect essentially small class of ${\beta}$-small objects. Then ${\mathcal{LO}}'$ is perfect also.
Moreover, if $s'$ is weighty in ${\underline{C}'}$ then ${\mathcal{LO}}'$ generates a weighty smashing torsion pair for ${\underline{C}}$.
III\. Assume in addition that ${\underline{C}}$ is ${\alpha}$-well generated for some regular cardinal ${\alpha}$, and that $s$ is smashing. 1. Assume that $s$ restricts (see Definition \[dhopo\](4)) to ${\underline{C}}^{{\beta}}$ for a regular cardinal ${\beta}\ge {\alpha}$. Then ${\mathcal{LO}}\cap {\operatorname{Obj}}{\underline{C}}^{\beta}$ is an essentially small perfect class. 2. If $s$ is weighty then it restricts to ${\underline{C}}^{{\beta}}$ for all large enough regular ${\beta}\ge {\alpha}$. Moreover, the class ${\mathcal{LO}}\cap {\operatorname{Obj}}{\underline{C}}^{\beta}$ perfectly generates $s$ for these ${\beta}$.
I.\[indep\]. It suffices to note that any $s$-decomposition morphism for $M$ factors through any other one according to Proposition \[phop\](7).
\[icontraf\]. We fix $h$ (along with $M$ and $N$).
The definition of approximations along with Proposition \[phop\](7) implies that any composition $ L_sP\stackrel{a_P}{\to} P\stackrel{g}{\to}M$ as in the definition of $J$ factors through the composition morphism $L_sAM\to M$. Hence if the composition $L_sAM\to N$ is zero then $h\in J$.
Conversely, assume that $h\in J$. Since ${\mathcal{P}}$ is contravariantly finite, we can choose a ${\mathcal{P}}$-approximation morphism $g\in {\underline{C}}(AM, M)$. Present $AM$ as a coproduct of some $P_i\in {\mathcal{P}}$; choose some $s$-decomposition morphisms $L_sP_i\stackrel{a_{P_i}}\to P_i$. Since $s$ is smashing, the morphism $a_{AM}^0=\coprod a_{P_i}$ is an $s$-decomposition one also according to Proposition \[phop\](4). Since $h \circ g\circ a_{P_i}=0$ for all $i$, we also have $h \circ g\circ a_{AM}^0=0$. Lastly, any other choice of $a_{AM}$ factors through $a_{AM}^0$ (by Proposition \[phop\](7); cf. the proof of assertion I.\[indep\]); this gives the “moreover” part of our assertion.
\[icoprcl\]. This is an easy consequence of the previous assertion. Indeed, to prove that $\coprod h_i\in J$ for a small (resp. countable) collection of $h_i\in J\cap {\underline{C}}(M,N)$ note that for any choices of ${\mathcal{P}}$-approximations $AM_i\to M_i$ their coproduct is a ${\mathcal{P}}$-approximation of $\coprod M_i$ (by Lemma \[lperf\](\[ipereq\])). The assertion follows easily since the coproduct of any choices of $L_sAM_i\to AM_i$ of $s$-decomposition morphisms is an $s$-decomposition morphism also (according to Proposition \[phop\](4)); thus it remains to apply assertion I.\[icontraf\].
\[ilscp\]. Assertion I.\[indep\] certainly implies that any ${L_s\mathcal{P}}$-null morphism belongs to $J$. The converse implication is immediate from ${L_s\mathcal{P}}\subset {\mathcal{P}}$.
\[ilscper\]. This is an obvious combination of the previous two assertions.
\[ilscperw\]. Since ${\mathcal{LO}}$ contains ${L_s\mathcal{P}}$, it also contains its big hull (see Lemma \[lbes\](\[iseses\], \[isesperp\])). Thus it suffices to verify the converse inclusion. Now, since ${\mathcal{P}}$ is essentially small, countably perfect, and ${\mathcal{P}}\subset {\mathcal{P}}[1]$, the big hull of ${\mathcal{P}}$ along with ${\mathcal{P}}^\perp$ is a (weighty) torsion pair according to Theorem \[tpgws\]. Since ${\mathcal{P}}^\perp={\{0\}}$, we obtain any object of ${\underline{C}}$ belongs to this big hull.
Now let $P$ belong ${\mathcal{LO}}$. As we have just proved, it is a retract of some $P'$ belonging to the naive big hull of ${\mathcal{P}}$. So we present $P'$ as $\operatorname{\varinjlim}Y_i$ so that $Y_0$ and cones of the connecting morphisms $\phi_i$ belong to ${\underline{\coprod}{\mathcal{P}}}$. Thus for $Z_i$ being as in Lemma \[lbes\](\[ilwd\]) we can choose $LZ_i\in {\underline{\coprod} {L_s\mathcal{P}}}$. Applying the lemma we obtain the existence of an $s$-decomposition triangle $L'\to P'\stackrel{n_{P'}}{\to} R'\to L[1]$ with $L'$ belonging to the naive big hull of ${L_s\mathcal{P}}$. Now, the distinguished triangle $P\to P\to 0\to P[1]$ is an $s$-decomposition of $P$. Since ${\operatorname{id}}_P$ can be factored through $P'$, applying Proposition \[phop\](7) to the corresponding morphisms $P\to P'\to P$ we obtain that ${\operatorname{id}}_P$ can be factored through $L'$.
II\. Let $f_i\in {\underline{C}}(N_i,Q_i)$ for $i\in J$ be a set of ${\mathcal{LO}}'$-null morphisms; for $N=\coprod N_i$, $f=\coprod f_i$, and $P\in {\mathcal{LO}}'$ we should check that the composition of any $e\in {\underline{C}}(P,N)$ with $f$ vanishes. The ${\beta}$-smallness of $P$ allows us to assume that $J$ contains less than ${\beta}$ elements. Next, Remark \[rbecomp\](3) gives a factorization of $e$ through the coproduct of some $h_i\in {\underline{C}}(M_i,N_i)$ with $M_i\in {\operatorname{Obj}}{\underline{C}'}$. We choose some $s'$-decompositions $L_i\to M_i\to R_i\to L_i[1]$ of $M_i$. Our assumptions easily imply that $\coprod L_i\to \coprod M_i\to \coprod R_i$ is an $s'$-decomposition of $\coprod M_i$ (cf. Proposition \[phop\](4)). Hence part 7 of the proposition implies that $e$ factors through the coproduct $g$ of the corresponding morphisms $L_i\to N_i$. Now, since $f_i$ are ${\mathcal{LO}}'$-null and $L_i\in {\mathcal{LO}}'$ then $f\circ g=0$; hence $f\circ e=0$ also.
Lastly, if $s'$ is weighty then ${\mathcal{LO}}'$ is cosuspended. Since it is also essentially small it remains to apply Theorem \[tpgws\] (along with Corollary \[cwftw\]).
III\. For ${\beta}\ge {\alpha}$ being a regular cardinal we take ${\mathcal{P}}= {\operatorname{Obj}}{\underline{C}}^{\beta}$. This is certainly a perfect essentially small class that Hom-generates ${\underline{C}}$; we also have ${\mathcal{P}}\subset {\mathcal{P}}[1]$. To prove assertion III.1 it suffices to note that ${\mathcal{LO}}\cap {\operatorname{Obj}}{\underline{C}}^{\beta}$ is a possible choice of $L_s{\mathcal{P}}$ (in the notation of assertion I) and apply assertion I.\[ilscper\].
Next, assertion I.\[ilscperw\] (combined with Remark \[revenmorews\](1)) implies that to prove assertion III.2 it suffices to verify that $s$ restricts to ${\underline{C}}^{{\beta}}$ for all large enough regular ${\beta}\ge {\alpha}$.
Now we choose some $L_sM$ for all $M\in {\operatorname{Obj}}{\underline{C}}^{\alpha}$, and choose for ${\alpha}'$ a regular cardinal such that all elements of $L_s{\mathcal{P}}$ belong to ${\underline{C}}^{{\alpha}'}$ (see Remark \[rbecomp\](2)). Then for any regular ${\beta}\ge {{\alpha}'}$ the pair $s$ restricts to ${\underline{C}}^{{\beta}}$, since the corresponding weight decompositions exist according to the “furthermore” part of Proposition \[ppcoprws\](\[icopr7p\]).
\[rtkrau\] 1. Our theorem suggests that it makes sense to define (at least) two distinct notions of ${\beta}$-well generatedness for smashing torsion pairs and weight structures in an ${\alpha}$-well generated category ${\underline{C}}$. One may say that $s$ is [*weakly ${\beta}$-well generated*]{} for some regular ${\beta}\ge {\alpha}$ if it is generated by a perfect set of ${\beta}$-compact objects. $s$ is [*strongly ${\beta}$-well generated*]{} if in addition to this condition, $s$ restricts to ${\underline{C}}^{\beta}$.
Certainly, compactly generated torsion pairs (see Definition \[dhopo\](3)) are precisely the weakly ${{\aleph_0}}$-well generated ones (since any set of compact objects is perfect; cf. Proposition \[psym\](\[isymcomp\])). Hence our two notions of ${\beta}$-well generatedness are not equivalent (already) in the case ${\alpha}={\beta}={{\aleph_0}}$; this claim follows from [@postov Theorems 4.15, 5.5] (cf. also Corollary 5.6 of ibid.) where (both weakly and strongly ${{\aleph_0}}$-well generated) weight structures on ${\underline{C}}=D({\operatorname{Mod}}-R)$ were considered in detail.
Moreover, for $k$ being a field of cardinality $\gamma$ the main subject of [@bgn] gives the following example: the opposite (see Proposition \[pbw\](\[idual\])) to (any version of) the Gersten weight structure over $k$ (on the category ${\underline{C}}$ that is opposite to the corresponding category of [*motivic pro-spectra*]{}; note that ${\underline{C}}$ is compactly generated) is weakly ${{\aleph_0}}$-well generated (by definition) and it does not restrict to the subcategory of ${\beta}$-compact objects for any ${\beta}\le \gamma$. On the other hand, this example is “as bad is possible” for weakly ${{\aleph_0}}$-well generated weight structures in the following sense: combining the arguments used the proof of part IV.2 of our theorem with that for Theorem \[tclass\] one can easily verify that any ${{\aleph_0}}$-well generated weight structure is ${\alpha}$-well generated whenever the set of (all) isomorphism classes of morphisms in the subcategory ${\underline{C}}^{{\aleph_0}}$ compact objects of ${\underline{C}}$ is of cardinality less than ${\alpha}$.
2\. Obviously the join (see Remark \[revenmorews\](1) and Corollary \[cwftw\](3)) of any set of weakly ${\beta}$-well generated weight structures is weakly ${\beta}$-well generated; so, we obtain a filtration (respected by joins) on the “join monoid” of weight structures. The natural analogue of this fact for strongly ${\beta}$-well generated weight structures is probably wrong. Indeed, it is rather difficult to believe that for a general compactly generated category ${\underline{C}}$ the class of weight structures on the subcategory ${\underline{C}}^{{{\aleph_0}}}$ of compact objects (cf. Proposition \[psym\](\[isymcomp\])) would be closed with respect to joins; note that joining compactly generated weight structures $w_i$ on ${\underline{C}}$ corresponds to intersecting the classes ${\underline{C}}_{w_i\ge 0}\cap {\operatorname{Obj}}{\underline{C}}^{{{\aleph_0}}}$.
On the other hand, Corollary 4.7 of [@krause] suggests that the filtration of the class of smashing weight structures by the sets of weakly ${\beta}$-well generated ones (for ${\beta}$ running through regular cardinals) may be “quite short”.
3\. According to part III.2 of our theorem, any weight structure on a well generated ${\underline{C}}$ is strongly ${\beta}$-well generated for ${\beta}$ being large enough. Combining this part of the theorem with its part II we also obtain a bijection between strongly ${\beta}$-well generated weight structures on ${\underline{C}}$ and ${\beta}$-coproductive weight structures on ${\underline{C}}^{\beta}$. Note that (even) the restrictions of these results to compactly generated categories appear to be quite interesting. 4. Now assume that a weight structure $w$ on ${\underline{C}}$ is strongly ${{\aleph_0}}$-well generated; this certainly means that ${\underline{C}}$ is compactly generated (see Proposition \[psym\](\[isymcomp\])) and $w$ restricts to its subcategory ${\underline{C}}^{{\aleph_0}}$ of compact objects. Then Proposition \[ppcoprws\](\[icopr7p\]) implies that ${\underline{C}}_{w\ge 0}$ is the smallest coproductive extension-closed subclass of ${\operatorname{Obj}}{\underline{C}}$ that contains ${\underline{C}}_{w\ge 0}\cap {\operatorname{Obj}}{\underline{C}}^{{\aleph_0}}$ (cf. the proof of Theorem \[twgws\](III.2)). Thus the $t$-structure right adjacent to $w$ is generated by the essentially small class ${\underline{C}}_{w\ge 0}\cap {\operatorname{Obj}}{\underline{C}}^{{\aleph_0}}$; so it is compactly generated (and hence smashing).
5\. For ${\underline{C}}$ as above and a weakly ${\beta}$-well generated weight structure $w$ on it one can easily establish a natural weight structure analogue of [@krauwg Theorem B] that will “estimate the size” of an element $M$ of ${\underline{C}}_{w\le 0}$ in terms of the cardinalities of ${\underline{C}}(P,M)$ for $P$ running through ${\beta}$-compact elements of ${\underline{C}}_{w\le 0}$ (modifying the proof of loc. cit. that is closely related to our proof of Theorem \[tpgws\]). Moreover, this result should generalize loc. cit. Note also that there is a “uniform” estimate of this sort that only depends on ${\underline{C}}$ (and does not depend on $w$). This argument should also yield that a weakly ${\beta}$-well generated weight structure is always strongly ${\beta}'$-well generated for a regular cardinal ${\beta}'$ that can be described explicitly.
Moreover, similar arguments can possibly yield that any smashing weight structure on a perfectly generated triangulated category ${\underline{C}}$ is perfectly generated (cf. Theorem \[twgws\](III.2)).
6\. Our understanding of “general” well generated torsion pairs is much worse than the one of (well generated) weight structures. In particular, the author does not know which properties of weight structures proved in this section can be carried over to $t$-structures.
On torsion pairs orthogonal with respect to dualities {#skan}
=====================================================
In this section we study dualities between triangulated categories (generalizing the bifunctor ${\underline{C}}(-,-)$ along with its restrictions to pairs of triangulated subcategories of ${\underline{C}}$). Our main construction tool are Kan extensions of homological functors from triangulated subcategories of ${\underline{C}}$ (to ${\underline{C}}$; we call the resulting functors [*coextensions*]{}); these are interesting for themselves.
In §\[scoext\] we study coextensions of homological functors (into an AB5 category ${\underline{A}}$) from a triangulated subcategory ${\underline{C}_0}$ to ${\underline{C}}$ following [@krause]. If ${\underline{C}}$ is compactly generated and ${\underline{C}_0}$ is its subcategory of compact objects then the [*coextended*]{} functors are precisely the cc ones. As an application we demonstrate that for any compactly generated torsion pair $s=({\mathcal{LO}},{\mathcal{RO}})$ there exists an object $N$ of ${\underline{C}}$ such that the functor $H^N$ kills precisely those compact objects of ${\underline{C}}$ that (also) belong to ${\mathcal{LO}}$.
In §\[sdual1\] we recall (from [@bger]) the definition of a duality $\Phi:{\underline{C}}{{^{op}}}\times {\underline{C}'}\to {\underline{A}}$ (we are mostly interested in the case ${\underline{C}}={\underline{\operatorname{Ab}}}$). The corresponding notion of [$\Phi$-orthogonal]{} torsion pairs generalizes the notion of adjacent ones. In the case where ${\underline{C}}\subset {\underline{C}'}$ and $\Phi$ is just the restriction of ${\underline{C}'}(-,-)$ to ${\underline{C}}{{^{op}}}\times {\underline{C}'}$ we are able to prove two natural generalizations of Proposition \[psatur\]; so we prove (assuming some additional conditions) that for any weight structure $w$ on ${\underline{C}}$ (resp. on ${\underline{C}}'$) there exists a $t$-structure on ${\underline{C}'}$ (resp. on ${\underline{C}}$) that is right (resp. left) $\Phi$-orthogonal to $w$. The results of [@neesat] and [@roq] demonstrate that these results can be applied for ${\underline{C}}$ being the derived category of perfect complexes and ${\underline{C}}'$ being the bounded derived category of coherent sheaves on a scheme proper over the spectrum of a noetherian ring. In §\[sdual2\] we study in detail the case where ${\underline{C}}$ and ${\underline{C}'}$ contain a common triangulated subcategory ${\underline{C}_0}$ whose objects are cocompact in ${\underline{C}}$ and compact in ${\underline{C}'}$, and $\Phi$ is a certain “bicontinuous biextension” of the bi-functor ${\underline{C}_0}(-,-)$. We obtain that any set ${\mathcal{P}}$ of objects of ${\underline{C}_0}$ gives a couple of $\Phi$-orthogonal torsion pairs (in ${\underline{C}}$ and ${\underline{C}}'$, respectively).
For a suspended ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}_0}$ this gives (in §\[sgengroth\]) a $t$-structure $t$ on ${\underline{C}'}$ that is generated by ${\mathcal{P}}$ and a weight structure $w$ on ${\underline{C}}$ that is cogenerated by ${\mathcal{P}}$ and is (left) $\Phi$-orthogonal to $t$. A particular case of this setting is considered in [@bgn] (where our results are applied to the study of various motivic homotopy categories, homotopy $t$-structures, and coniveau spectral sequences). We also apply this result to the study of compactly generated $t$-structures The elements of the heart of $w$ give a faithful family of exact functors ${{\underline{Ht}}}\to {\underline{\operatorname{Ab}}}$ that respect coproducts; hence ${{\underline{Ht}}}$ is an AB5 category. Next an easy argument yields the existence of a generator for ${{\underline{Ht}}}$ (so, it is a Grothendieck abelian category).
In §\[sprospectra\] we prove (using the results of [@tmodel]) that taking ${\underline{C}'}$ to be the homotopy category of a proper simplicial stable model category ${\mathcal{M}}$ and ${\underline{C}}$ to be the homotopy category of the category ${\operatorname{Pro}-\mathcal{M}}$ of (filtered) pro-objects of ${\mathcal{M}}$ one obtains an example of the aforementioned setting. Thus we obtain that ${{\underline{Ht}}}$ is an Grothendieck abelian category for any compactly generated $t$-structure on a “topological” triangulated category. Moreover, this pro-object construction is used in [@bgn] for the description of various triangulated categories of [*motivic pro-spectra*]{} (and [*comotives*]{}) and for the study of their relation to the corresponding motivic stable homotopy categories. Another example of the couple $({\underline{C}},{\underline{C}'})$ that may be obtained this way is $(\operatorname{SH}^{op},\operatorname{SH})$; this observation is closely related to the main subject of [@prospect].
In §\[slocoeff\] we recall a few results on “localizing coefficients” in (compactly generated) triangulated categories and the relate this matter to torsion pairs and their orthogonality. We also study (from a similar perspective) decompositions of triangulated categories as direct sums of their subcategories. The section is included here for the purpose of using it in [@bgn]; still some of its results may be interesting for themselves.
On Kan extensions of homological functors {#scoext}
------------------------------------------
Now we will study a method of extending of homological functors from a triangulated subcategory of ${\underline{C}}$; this is a version of left Kan extensions that was studied in detail in [@krause §2]. In particular, we obtain a description of all cc functors from ${\underline{C}}$ (into an AB5 abelian category) if it is compactly generated. Since the construction is dual to the one applied in [@bger] and [@bgn] (and it “usually respects coproducts”) we will call the resulting functors [*coextended*]{} ones; we will explain that they are actually Kan extensions below.
To formulate some of the properties of the construction we will use (a few times) the following definition.
\[drelim\] Let ${\underline{C}_0}$ be a subcategory of ${\underline{C}}$ (or just a class of objects), $M\in {\operatorname{Obj}}{\underline{C}}$; let $L$ be a small (index) category and fix a functor $L\to {\underline{C}}: l\mapsto N_l$.
1\. Let $M$ be a co-cone of this functor (i.e., $M$ is equipped with compatible morphisms from $N_l$ for $l\in {\operatorname{Obj}}L$). Then we will say that $M$ is a [*${\underline{C}_0}$-colimit*]{} of $(N_l)$ if the restriction $H_M$ of the functor ${\underline{C}}(-,M)$ to ${\underline{C}_0}$ equals the colimit of $ H_{N_l}$ (in ${\operatorname{PShv}}({\underline{C}_0})$, i.e., if for any $Y\in {\operatorname{Obj}}{\underline{C}_0}$ the connecting morphisms induce an isomorphism ${\underline{C}}(Y,N)\cong \operatorname{\varinjlim}{\underline{C}}(Y,N_l)$).
2\. If $M$ is a cone of $N$ then we will say that $M$ is a [*${\underline{C}_0}$-limit*]{} of $(N_l)$ if $M$ is a ${\underline{C}_0}^{op}$-colimit of $(N_l)$ in ${\underline{C}}^{op}$.
\[rlim\] We will not need much of this definition in the current paper. Moreover, it appears that the most “useful” case of part 1 (resp. 2) of the definition is the filtered one, i.e., the one where $(N_l)$ is an inductive (resp. projective) system.
Now let ${\underline{C}_0}$ be an essentially small triangulated subcategory of ${\underline{C}}$; we will also assume that ${\underline{C}}$ has coproducts. We consider the category ${\operatorname{PShv}}({\underline{C}_0})=\operatorname{\operatorname{AddFun}}({\underline{C}}_0^{op},{\underline{\operatorname{Ab}}})$ (cf. the proof of Theorem \[tpgws\]); recall that this is a locally small abelian category. It it easily seen that any $H\in {\operatorname{Obj}}{\operatorname{PShv}({\underline{C}}_0)}$ possesses a (projective) resolution $ \coprod{\underline{C}}_0(-, C_i)\to \coprod {\underline{C}}_0(-,C_j)\to H\to 0 $ where $\{C^i\}$ and $\{C^j\}$ are some families of objects of ${\underline{C}}_0$; cf. Lemma 2.2 of [@krause].
\[pkrause\] Let $H_0:{\underline{C}}_0\to {\underline{A}}$ be a homological functor, where ${\underline{A}}$ is an AB5 abelian category; fix some $N\in {\operatorname{Obj}}{\underline{C}}$. For any $M\in {\operatorname{Obj}}{\underline{C}}$ we fix a resolution (as above) $$\label{ekrause}
\coprod_{i\in I}H_{C_M^i}\to \coprod_{j\in J} H_{C_M^j}\to H_M\to 0,$$ where we use the notation $H_M$ for the restriction of the functor ${\underline{C}}(-,M)$ to ${\underline{C}_0}$.
Then for the association $H:M\mapsto \operatorname{\operatorname{Coker}}(\coprod H_0(C_M^i)\to \coprod H_0(C_M^j))$ the following statements are valid.
1. \[ikr1\] $H$ is a homological functor ${\underline{C}}\to {\underline{A}}$ that is essentially independent on the choice of resolutions. In particular, the restriction of $H$ to ${\underline{C}_0}$ is canonically isomorphic to $H_0$.
We will call $H$ the [*coextension*]{} of $H_0$ to ${\underline{C}}$, and say that it is a [*coextended*]{} (from ${\underline{C}_0}$) functor.
2. \[ikrchar\] For any ${\operatorname{PShv}}({\underline{C}_0})$-resolution $ \coprod H_{C'{}_M^{i}}\to \coprod H_{C'{}_M^{j}} \to H_M\to 0$ of $H_M$ with $C'{}_M^{i}$ and $C'{}_M^{j}$ being some objects of ${\underline{C}_0}$, the object $\operatorname{\operatorname{Coker}}(\coprod H_0(C'{}_M^{i})\to \coprod H_0(C'{}_M^{j}))$ is canonically isomorphic to $H(M)$. Moreover, $H$ is canonically characterized by this condition along with its restriction to ${\underline{C}_0}$.
3. \[ikres\] More generally, we have $H(M)\cong \operatorname{\operatorname{Coker}}(\coprod H(C'{}_{M^i})\to \coprod H(C'{}_{M^j}))$ (also) if the ${\operatorname{PShv}({\underline{C}}_0)}$-sequence $ \coprod H_{C'{}_M^{i}}\to \coprod H_{C'{}_M^{j}} \to H_M\to 0$ is exact for some objects $C'{}_M^{i}$ and $C'{}_M^{j}$ of ${\underline{C}}$.
4. \[ikrtriv\] If $H_0:{\underline{C}_0}\to {\underline{\operatorname{Ab}}}$ is corepresented by some $M_0\in {\operatorname{Obj}}{\underline{C}_0}$ then $H$ is ${\underline{C}}$-corepresented by $M_0$ also.
5. \[ikr7\] For any exact sequence $\coprod H^i_0\to \coprod H^j_0\to H_0\to 0$ the corresponding ${\operatorname{PShv}}({\underline{C}})$-sequence of coextensions $\coprod H^i\to \coprod H^j\to H\to 0$ is exact also. In particular, the coextension of $\coprod H^j_0$ equals $\coprod H^j$.
6. \[ikr2\] $H(N)$ for $N\in {\operatorname{Obj}}{\underline{C}}$ only depends on the restriction of ${\underline{C}}(-,N)$ to ${\underline{C}_0}$.
7. \[ikradj\] Let ${\underline{E}}$ be a full triangulated subcategory of ${\underline{C}}$ that contains ${\underline{C}_0}$ and assume that there exists a right adjoint $G$ to the embedding ${\underline{E}}\to {\underline{C}}$ (so, ${\underline{E}}$ has coproducts). Then we have $H\cong H^{{\underline{E}}}\circ G$,where the functor $H^{{\underline{E}}}:{\underline{E}}\to {\underline{A}}$ is defined on ${\underline{E}}$ using the same construction as the one used for the definition of ${\underline{C}}$.
8. \[ikr3\] If $N$ is a ${\underline{C}_0}$-colimit of some $(N_l)$ (see Definition \[drelim\](1)), then $H(N)\cong \operatorname{\varinjlim}H(N_l)$.
Moreover, such a set of $(X_l,f_l)$ exists for any $X\in {\operatorname{Obj}}{\underline{C}}$.
9. \[ikrtransf\] For $H'_0$ being another homological functor ${\underline{C}_0}\to {\underline{A}}$ and the corresponding coextension $H'$ we have the following: the restriction of natural transformations $H\implies H'$ to the subcategory ${\underline{C}_0}$ gives a one-to-one correspondence between them and the transformations $H_0\implies H'_0$.
10. \[ikr6\] Let $H_0\stackrel{f_0}{\to} H'_0 \stackrel{g_0}{\to} H''_0$ be a (three-term) complex of homological functors ${\underline{C}_0}\to {\underline{\operatorname{Ab}}}$ that is exact in the term $H'_0$. Then the complex $H\stackrel{f}{\to} H' \stackrel{g}{\to} H''$ (here $H,H',H'',f,g$ are the corresponding coextensions) is exact in the middle also.
11. \[ikr8\] Assume that all objects of ${\underline{C}_0}$ are compact. Then $H$ is determined (up to a canonical isomorphism) by the following conditions: it respects coproducts, it restriction to ${\underline{C}_0}$ equals $H_0$, and it kills ${\underline{C}_0}^\perp$.
\[ikr1\]. Immediate from [@krause Lemma 2.2] (see also Proposition 2.3 of ibid.).
\[ikrchar\] –\[ikr3\]. The proofs are straightforward (and very easy); cf. also Remark \[rkrause\](I.1) below.
Assertion \[ikrtransf\] is easy also. The injectivity of this restriction correspondence follows easily from the previous assertion and the surjectivity is immediate from the naturality of the coextension construction in $H_0$. \[ikr6\]. We should check that the sequence $H(M)\stackrel{f(M)}{\to} H'(M) \stackrel{g(M)}{\to} H''(M)$ of abelian groups is exact (in the middle term) for any $M\in {\operatorname{Obj}}{\underline{C}}$. Certainly, the functoriality of coextensions gives $g(M)\circ f(M)=0$.
Our exactness assertion is obviously valid if $M\in {\operatorname{Obj}}{\underline{C}_0}$. We reduce the general case of this statement to this one.
We start from analysing the sequence (\[ekrause\]). The Yoneda lemma immediately implies that the natural transformations in it are given by certain ${\underline{C}}$-morphisms $f_j: C_M^j \to M$ for all $j\in L$ and $g_{ij}:C_M^i\to C_M^j $ for all $i\in I$, $j\in J$; moreover, for any $i\in I$ almost all $g_{ij}$ are $0$ and we have $\sum_{j\in J} f_j\circ g_{ij}=0$.
We should check that if for $a\in H'(M)$ we have $g(M)(a)=0$, then $a=f(M)(x)$ for some $x\in H(M)$.
Using the additivity of ${\underline{C}_0}$ we can gather finite sets of $C_M^i$ and $C_M^i$ in (\[ekrause\]) into single objects. Hence we can assume that the following assumptions are fulfilled: $a=H'_0(f_{j_0})(a_0)$ for some $j_0\in J$ and $a_0\in H'_0(C_M^{j_0} )$, $g_0(C_M^{j_0})(a_0)=H''_0(g_{i_0j_0}) (b_0)$ for some $i_0\in I$ and $b_0\in H''_0(C_M^{i_0} )$ (recall that $H''(M)$ is defined as the corresponding cokernel!). Moreover, we can assume that $g_{i_0j}=0$ for any $j\neq j_0$; thus $f_{j_0}\circ g_{i_0j_0}=0$. Complete $g_{i_0j_0}$ to a distinguished triangle $C_M^{i_0}$( g\_[i\_0j\_0]{})$ by $Y$
\to C_M^{j_0}\stackrel{{\alpha}}{\to} Y$; then we can assume $Y\in {\operatorname{Obj}}{\underline{C}_0}$, and the equality $f_{j_0}\circ g_{i_0j_0}=0$ implies that $f_{j0}$ can be decomposed as ${\beta}\circ {\alpha}$ for some ${\beta}\in {\underline{C}}(Y,M)$.
Since $H''_0$ is homological, $H''({\alpha})(g_0(C_M^{j_0})(a_0))=0$. Applying the exact sequence $H_0\to H_0'\to H_0''$ to $Y$ we obtain that $H'_0({\alpha})(a_0)\in H'(Y)$ can be presented as $f_0(Y)(x_Y)$ for some $x_Y\in H_0(Y)=H(Y)$. Hence $a=f(M)(x)$ for $x=H({\beta})\in H(M)$; see the commutative diagram $$\begin{CD}
H_0(C_M^{j_0}) @>{H_0({\alpha})}>> H_0(Y)=H(Y)@>{H({\beta})}>>H(M)\\
@VV{f_0(C_M^{j_0})}V@VV{f_0(Y)=f(Y)}V@VV{f(M)}V \\
H'_0(C_M^{j_0})@>{H'_0({\alpha})}>>H'_0(Y)=H'(Y) @>{H'({\beta})}>>H'(M)\\
@VV{g_0(C_M^{j_0})}V@VV{g_0(Y)=g(Y)}V@VV{g(M)}V \\
H''_0(C_M^{j_0})@>{H''_0({\alpha})}>>H''_0(Y)=H''(Y)@>{H''({\beta})}>>H''(M)
\end{CD}$$
\[ikr8\]. In the case where ${\underline{C}_0}$ generates ${\underline{C}}$ as its own localizing category the assertion is given by Proposition 2.3 of [@krause]. Now recall that (in the general case) the embedding of the localizing category generated by ${\underline{C}_0}$ into ${\underline{C}}$ possesses an (exact) right adjoint $G$ that respects coproducts (according to Proposition \[pcomp\](II)). Hence the general case of the assertion reduces to loc. cit. if we apply assertion \[ikradj\].
\[rkrause\] I.1. Now we explain that $H$ is actually the left Kan extension of $H_0$ along the inclusion ${\underline{C}_0}\to {\underline{C}}$.
Indeed, we can certainly assume that ${\underline{C}_0}$ is small. Then the standard pointwise construction of the left Kan extension is easily seen to correspond to “the most obvious” resolutions of the functors $H_M$ in ${\operatorname{PShv}({\underline{C}}_0)}$.
This observation certainly gives an alternative proof of part \[ikrtransf\] of the proposition.
Below we will also mention extensions of cohomological functors ${\underline{C}}\to {\underline{A}}$ obtained via applying the coextension construction to the corresponding (homological) functors ${\underline{C}}{{^{op}}}\to {\underline{A}}$. So they can also be described via right Kan extensions of the opposite homological functors ${\underline{C}}\to {\underline{A}}{{^{op}}}$.
2\. It appears that it is not necessary to assume that ${\underline{C}_0}$ is triangulated to construct $H$ from $H_0$.
So, assume that ${\underline{C}_0}$ is an essentially small additive subcategory of ${\underline{C}}$ that satisfies the following condition: for any ${\underline{C}}$-distinguished triangle $$\label{edi}
Z[-1]\to X\to Y\to Z$$ the object $X$ belongs to ${\operatorname{Obj}}{\underline{C}_0}$ whenever $Y$ and $Z$ do. Note that any ${\underline{C}_0}$ satisfying this condition is cosuspended, and the extension-closure of any cosuspended subcategory of ${\underline{C}}$ satisfies this condition.
Now, our restriction on ${\underline{C}_0}$ certainly implies that it has [*weak kernels*]{} (for all morphisms); hence the subcategory of coherent functors inside ${\operatorname{PShv}}({\underline{C}_0})$ is abelian (see §1.2 of [@krause]). Take an additive functor $H_0:{\underline{C}_0}\to {\underline{A}}$ for an AB5 category ${\underline{A}}$ such that for any triangle (\[edi\]) with $Y,Z\in {\operatorname{Obj}}{\underline{C}_0}$ the sequence $H(X)\to H(Y)\to H(Z)$ is exact (in the middle term). Using [@krause Lemmas 2.1, 2.2] (cf. also the proof of Proposition 2.3 of ibid.) one can easily prove that the (“pointwise Kan”) recipe described in Proposition \[pkrause\] gives a homological functor $H:{\underline{C}}\to {\underline{A}}$ whose restriction to ${\underline{C}_0}$ is isomorphic to $H_0$.
Assume in addition that ${\underline{C}_0}$ consists of compact objects of ${\underline{C}}$. Then the functor $H$ (coming from any $H_0$ as above) is certainly a cc functor. Moreover, ${\operatorname{Obj}}{\underline{C}_0}$ generates a weight structure $w$ (see Remark \[rnewt\](1)) and $H$ obviously annihilates ${\mathcal{RO}}={\underline{C}}_{w\ge 1}$.
Now we prove that the functors $H$ that can be obtained using coextensions of this sort are precisely the cc functors satisfying this vanishing condition (similarly to Definition \[drange\] one may say that these functors are of weight range $\le 0$). It follows that for $H_0$ being the restriction to ${\underline{C}_0}$ of a cc functor $H':{\underline{C}}\to {\underline{A}}$ we have $H\cong \tau^{\ge 0}H'$ (see Remark \[rwrange\](4)).
We start from noting that for any $H'$ as above the definition of $H$ gives a canonical transformation $\Psi: H\to H'$. Now assume that $H'$ annihilates ${\mathcal{RO}}$. First we prove that $\Psi(M_0)$ is injective for any $M_0\in {\operatorname{Obj}}{\underline{C}}$. Considering a $w$-decomposition $(w_{\ge 1}M_0)[1]\to w_{\le 0}M_0\to M\to w_{\ge 1}M_0$ we obtain that $H'(M_0)\cong H'(w_{\le 0}M_0)$ and $H(M_0)\cong H(w_{\le 0}M_0)$. Hence it suffices to verify the injectivity of $\Psi(M)$ for any $M\in {\underline{C}}_{w\le 0}$. We will apply a certain inductive argument with the base being the obvious fact that $\Psi(M)$ is an isomorphism if $M$ is a coproduct of objects of ${\underline{C}_0}$.
We use the construction (and adopt the notation) used in the proof of Theorem \[tclass\](\[iclass1\]). Since $M$ belongs to $M\in {\underline{C}}_{w\le 0}$, it is a retract of the corresponding $L$ (according to Proposition \[phop\](7)), whereas the latter is a homotopy colimit of $L_i$ with cones of the connecting morphisms belonging to ${\underline{\coprod}{\mathcal{P}}}$.
So, for all $k\ge 0$ we have commutative diagrams $$\begin{CD}
H(P_k[-1])@>{}>>H(L_k)@>{}>>H(L_{k+1})@>{}>>H(P_k) \\
@VV{\cong }V@VV{}V@VV{}V @VV{\cong}V\\
H'(P_k[-1])@>{}>>H'(L_k)@>{}>>H'(L_{k+1})@>{}>>H'(P_k)
\end{CD}$$ whose rows are exact. It easily follows by induction (starting from $\Psi(L_0)=\Psi(0)=0$) that $\Psi(L_k)$ is injective for any $k\ge 0$. Passing to the colimit (see Lemma \[lcoulim\](3(ii))) we obtain that $\Psi(L)$ is monomorphic; hence $\Psi(M)$ is monomorphic also. Next we prove that $\Psi(M)$ is also epimorphic (for any $M\in {\operatorname{Obj}}{\underline{C}}$). We set $H''(M)= \operatorname{\operatorname{Coker}}(\Psi(M))$. Since $\Psi(M)$ is monomorphic for all $M\in {\operatorname{Obj}}{\underline{C}}$, we obtain that $H''(M)$ is a homological functor (${\underline{C}}\to{\underline{A}}$); certainly it is a cc functor. So we should prove that $H''$ is zero. For any $M$, $L$, $L_k$, and $P_k$ as above we certainly have $H''(P_k)=H''(P_k[-1])=0$ for any $k\ge 0$; hence obvious induction (similar to that used in the proof of the injectivity of $\Psi(M)$) gives the vanishing of $H''(L_k)$ for $k\ge 0$. Passing to the colimit once again we obtain $H''(L)=H''(M)=0$ (for any $M\in {\underline{C}}_{w\le 0}$). Since $H''$ also annihilates ${\mathcal{RO}}$, we obtain that it is zero.
Possibly, the author will write these arguments in more detail in a new version of this paper to obtain a set of generators of a (compactly generated) $t$-structure $t$ under the assumptions of Theorem \[tab5\](2) below (this set of generators would be much smaller than the one given by part 1 of this theorem; see Remark \[rab5\](\[ismgen\])).
II\. Now assume (once again) that all objects of ${\underline{C}_0}$ are compact in ${\underline{C}}$. 1. Then part \[ikr8\] of our proposition easily implies that $\operatorname{\operatorname{AddFun}}({\underline{C}_0},{\underline{A}})$ is equivalent to the category of those cc functors ${\underline{C}}\to {\underline{A}}$ that kill ${\underline{C}_0}^\perp$. 2. Let $w$ be a smashing weight structure for ${\underline{C}}$ that restricts to ${\underline{C}_0}^\perp$. Since the corresponding virtual $t$-truncations of cc functors are cc ones according to Proposition \[ppcoprws\](\[icopr6\]), we also obtain that virtual $t$-truncations of coextended functors are coextended.
It certainly follows that (in this case) virtual t-truncations of coextended functors satisfy the “continuity” property described in part \[ikr3\] of our proposition. Recall also that we have a similar continuity for natural transformations between coextended functors according to part \[ikrtransf\] of the proposition.
3\. Now we describe an interesting application of these observations. We recall that for any (co)homological functor $J:{\underline{C}}\to {\underline{A}}$ and any $N\in {\operatorname{Obj}}{\underline{C}}$ a certain [*weight spectral sequence*]{} $T_w(J,N)$ for $J_*(N)$ is defined. We will not need its definition here. We will only recall (see Theorem 2.3.2 of [@bws]) that this spectral sequence is defined starting from the $E_1$-page; $T_w(J,N)$ becomes independent from any choices (and functorial in $N$) starting from $E_2$ (and we will write $T^{\ge 2}_w(J,N)$ for this “part” of $T_w(J,N)$). Moreover, Theorem 2.4.2(II) of [@bger] immediately implies the following: $T^{\ge 2}_w(J,N)$ can be $N$-functorially described in terms of certain virtual $t$-truncations of $J$ along with canonical transformations between them (and the transformations of the second level essentially come from (\[evtt\]); note that this statement follows from the description of the derived exact couple $(E_2^{**},D_2^{**})$ for the exact couple $(E_1^{**},D_1^{**})$ that gives $T_w(J,N)$). Hence we obtain the following: if $J$ is a coextended functor (and so, ${\underline{A}}$ is an AB5 category) and $N$ is a ${\underline{C}_0}$-colimit of an [**inductive system**]{} $(N_l)$ then $T^{\ge 2}_w(J,N)$ is the direct limit of $T^{\ge 2}_w(J,N_l)$.
The author plans to apply (the dual to) this statement in [@bgn] (where weight spectral sequences are described and studied in much more detail). It will be applied to extended (see part I of this remark) cohomological functors from a cocompactly cogenerated category ${\underline{E}}$, where ${\underline{E}}$ is a certain category of [*motivic pro-spectra*]{} (or of [*comotives*]{}). This allows to compute (“generalized”) coniveau spectral sequences for the corresponding cohomology of a projective limits of smooth varieties $X_l$ (over the base field $k$ that is perfect).
4\. Now we give one more proof of the fact that $\operatorname{\varinjlim}T^{\ge 2}_w(J,N_l)\cong T^{\ge 2}_w(J,N) $ (if $N$ is a ${\underline{C}_0}$-colimit of an inductive system $(N_l)$ and $J$ is extended); this argument avoids the consideration of $D_*^{**}$.
The functoriality of $T^{\ge 2}_w(J,-)$ yields canonical compatible morphisms between $T^{\ge 2}_w(J,N_l)$ and from them into $T^{\ge 2}_w(J,N)$. Now, since $(N_l)$ form an inductive system, the inductive limit $T'^{\ge 2}_w(J,N)$ of $T^{\ge 2}_w(J,N_l)$ is a spectral sequence (that starts from $E_2$); we also have a canonical morphism $T'^{\ge 2}_w(J,N)\to T^{\ge 2}_w(J,N)$. Thus it remains to verify that this morphism is an isomorphism at $E_2$, i.e., that $\operatorname{\varinjlim}E_2^{**}T_w(J,N_l)\cong E_2^{**}T(J,N_l)$.
Now, Theorem 2.3.2 of [@bws] implies immediately that the $E_2$-terms of $T_w(J,-)$ are given by the pure functors $H^{A_{J\circ [i]}}\circ [j]$ for $i,j\in {{\mathbb{Z}}}$ (see Proposition \[ppure\]). It is easily seen that these functors kill ${\underline{C}_0}^\perp$, and they are cc according to Proposition \[ppcoprws\](\[icopr5\]). Hence they are extended according to Proposition \[pkrause\](\[ikr8\])[^60], and it remains to apply Proposition \[pkrause\](\[ikr3\]).
Now we combine the results of this section with the ones of [@bsnull]; this gives an alternative proof of the “classification” of compactly generated torsion pairs given by Theorem \[tclass\].
\[rnz\] 1. So, we want to give one more proof of the existence of a one-to-one correspondence between compactly generated torsion pairs for ${\underline{C}}$ and essentially small Karoubi-closed extension-closed subclasses of the class ${\underline{C}}^{{{\aleph_0}}}$ of compact objects of ${\underline{C}}$; the argument should not depend on Theorem \[tclass\](\[iclass2\]).
Recall that any essentially small subclass ${\mathcal{P}}$ of ${\underline{C}}^{{{\aleph_0}}}$ generates a compactly generated torsion pair $s=({\mathcal{LO}},{\mathcal{RO}})$ for ${\underline{C}}$ according to Theorem \[tclass\](\[iclass1\]) (this statement is also given by the easier Theorem 4.3 of [@aiya]). Since ${\underline{C}}^{{{\aleph_0}}}$ gives a triangulated subcategory of ${\underline{C}}$ (see Lemma 4.1.4 of [@neebook]), the class ${\underline{C}}^{{{\aleph_0}}}\cap {\mathcal{LO}}$ contains the envelope of ${\mathcal{P}}$. Next (cf. Theorem \[tclass\](\[iclasst\]) or its proof) this envelope is essentially small also. Thus we should prove that ${\underline{C}}^{{{\aleph_0}}}\cap {\mathcal{LO}}$ equals ${\mathcal{P}}$ whenever the generating class ${\mathcal{P}}$ is essentially small, Karoubi-closed, and extension-closed in ${\underline{C}}$. For the latter purpose it suffices to prove the existence of $I\in {\mathcal{RO}}={\mathcal{P}}{{}^{\perp}}$ such that ${{}^{\perp}}I\cap {\underline{C}}^{{{\aleph_0}}}={\mathcal{P}}$.
Now, in [@bsnull] zero classes (see Lemma \[lbes\](\[izs\])) of (co)homological functors were studied. Corollary 3.11 of ibid. (applied to the category ${\underline{C}_0}{{^{op}}}$) gives the following remarkable statement: if ${\underline{C}_0}$ is a small triangulated category then a set ${\mathcal{P}}_0$ of its objects is the zero class of some “detecting” homological functor $H_0:{\underline{C}_0}\to {\underline{\operatorname{Ab}}}$ if and only if ${\mathcal{P}}_0$ is extension-closed and Karoubi-closed in ${\underline{C}_0}$. The author believes that this statement will become an important tool of studying (compactly generated) torsion pairs.
We take ${\mathcal{P}}_0$ to be a small skeleton of ${\mathcal{P}}$ and take ${\underline{C}_0}$ to be a small skeleton of ${\langle}{\mathcal{P}}{\rangle}_{{\underline{C}}}$ such that ${\operatorname{Obj}}{\underline{C}_0}\cap {\mathcal{P}}={\mathcal{P}}_0$. Since all objects of ${\underline{C}_0}$ are compact in ${\underline{C}}$, the coextension $H$ of the aforementioned functor $H_0$ to ${\underline{C}}$ is a cc functor according to Proposition \[pkrause\](\[ikr8\]). Next we consider the Brown-Comenetz dual functor ${\widehat{H}}$ (see Definition \[dsym\](\[ibcomf\]); recall that ${\widehat{H}}: M\mapsto {\underline{\operatorname{Ab}}}(H(M),{{\mathbb{Q}}}/{{\mathbb{Z}}})$ is a cp functor from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$ whose zero class coincides with that of $H$).
If we assume in addition that ${\underline{C}}$ equals its own localizing subcategory ${\underline{C}}'$ generated by ${\mathcal{P}}$ then it satisfies the Brown representability condition according to Proposition \[pcomp\](II.1). Thus the functor ${\widehat{H}}$ is ${\underline{C}}$-representable by some $I_0\in {\operatorname{Obj}}{\underline{C}}$ (see Proposition \[psymb\](II.\[iws21\])). Since the zero class of ${\widehat{H}}$ contains ${\mathcal{P}}$, we have $I_0\in {\mathcal{RO}}$. It remains to compute ${{}^{\perp}}I_0\cap {\underline{C}}^{{{\aleph_0}}}$. Now, ${\underline{C}}^{{{\aleph_0}}}$ equals ${\operatorname{Obj}}{\langle}{\mathcal{P}}{\rangle}$ (in this case) according to Lemma 4.4.5 of [@neebook]. Hence ${{}^{\perp}}I_0\cap {\underline{C}}^{{{\aleph_0}}}={\mathcal{P}}$ by construction (since ${\mathcal{P}}$ is the isomorphism-closure of ${\mathcal{P}}_0$ in ${\underline{C}}$). Lastly, to reduce the general case of our assertion (i.e., of ${\underline{C}}^{{{\aleph_0}}}\cap {\mathcal{LO}}={\mathcal{P}}$) to the case ${\underline{C}}={\underline{C}}'$ it suffices to recall that $({\mathcal{LO}},{\mathcal{RO}}\cap {\operatorname{Obj}}{\underline{C}}')$ is a torsion pair for ${\underline{C}}'$ according to Proposition \[phopft\](I.2) . Note also that the existence of a “detector object” $I_0$ is a new result. 2. The aforementioned result of [@bsnull] (along with some of its variations also proved in ibid.) is a sort of Nullenstellensatz for (co)homological functors from (small) triangulated categories (whence the name of the paper). It would certainly be interesting to obtain some analogue of this statement for cc and cp functors; Theorem \[tclass\](\[iclass5\]) is certainly related to this matter. Note however that a priori the intersection of zero classes of all those smashing virtual $t$-truncations of representable functors that contain a given ${\mathcal{P}}\subset {\underline{C}}$ (for ${\underline{C}}$ having coproducts) may be bigger than the strong extension-closure of ${\mathcal{P}}$.
Dualities between triangulated categories and orthogonal torsion pairs; applications to categories of coherent sheaves {#sdual1}
----------------------------------------------------------------------------------------------------------------------
Now we study certain pairings between triangulated categories and define the notion of orthogonality of torsion pairs (as well as of weight and $t$-structures) generalizing the one of adjacent structures. We also define the notion of a nice duality; yet we will not use it in the current paper (anywhere except in Proposition \[pnice\]).
\[dort\] Let ${\underline{C}}'$ be a triangulated category.
1. \[idu\] We will call a (covariant) bi-functor $\Phi:{\underline{C}}^{op}\times{\underline{C}}'\to {\underline{A}}$ a [*duality*]{} if it is bi-additive, homological with respect to both arguments, and is equipped with a (bi)natural bi-additive transformation $\Phi(A,Y)\cong \Phi (A[1],Y[1])$.
2. \[iorthop\] Suppose that ${\underline{C}}$ is endowed with a torsion pair $s=({\mathcal{LO}},{\mathcal{RO}})$ and ${\underline{C}}'$ is endowed with a torsion pair $s'=({\mathcal{LO}}',{\mathcal{RO}}')$. Then we will say that $s$ is (left) [*orthogonal*]{} to $s'$ with respect to $\Phi$ (or just [*left $\Phi$-orthogonal*]{} to it) if the following [*orthogonality condition*]{} is fulfilled: $\Phi (X,Y)=0$ whenever $X\in {\mathcal{LO}}$ and $Y\in {\mathcal{LO}}'$ or if $X\in {\mathcal{RO}}$ and $Y\in {\mathcal{RO}}'$.
3. \[iorthtw\] Suppose that ${\underline{C}}$ is endowed with a weight structure $w$, ${\underline{C}}'$ is endowed with a $t$-structure $t$. Then we will say that $w$ is (left) [*orthogonal*]{} to $t$ with respect to $\Phi$ if the following condition is fulfilled: $\Phi (X,Y)=0$ if $X\in {\underline{C}}_{w\ge 0}$ and $Y\in {\underline{C}}'^{t \ge 1}$ or if $X\in {\underline{C}}_{w\le 0}$ and $Y\in {\underline{C}}'^{t \le -1}$.
4. \[ini\] We will say that $\Phi$ is [*nice*]{} if for any distinguished triangles $T=A\stackrel{l}{\to} B \stackrel{m}{\to} C\stackrel{n}{\to} A[1]$ in ${\underline{C}}$ and $X\stackrel{f}{\to} Y\stackrel{g}{\to} Z\stackrel{h}{\to}X[1]$ in ${\underline{C}}'$ we have the following: the natural morphism $p$: $$\begin{gathered} \operatorname{\operatorname{Ker}}(\Phi(A,X)\bigoplus \Phi(B,Y) \bigoplus \Phi(C,Z))\\
\xrightarrow{\begin{pmatrix}
\Phi(A,-)(f) & -\Phi(-,Y)(l) &0 \\
0& g(B) &-\Phi(-,Z)(m) \\
- \Phi(-,X)([-1](n)) & 0 &\Phi(C,-)(h)
\end{pmatrix}}
\\ (\Phi(A,Y) \bigoplus \Phi(B,Z) \bigoplus \Phi(C[-1],X))
\stackrel{p}{\to} \operatorname{\operatorname{Ker}}((\Phi(A,X)\bigoplus \Phi(B,Y))\\ \xrightarrow{\Phi(A,-)(f)\oplus - \Phi(-,Y)(l)}
\Phi(A,Y)) \end{gathered}$$ is epimorphic.
\[rort\] 1. For ${\underline{C}}'={\underline{C}}$ and $\Phi={\underline{C}}(-,-)$ the definition of orthogonal torsion pairs restricts to the one of adjacent ones. Indeed, in this case the $\Phi$-orthogonality of $s$ and $s'$ means that ${\mathcal{LO}}\perp_{{\underline{C}}} {\mathcal{LO}}'$ and ${\mathcal{RO}}\perp_{{\underline{C}}} {\mathcal{RO}}'$; these inclusions are certainly equivalent to ${\mathcal{LO}}'\subset {\mathcal{RO}}$ and ${\mathcal{RO}}\subset {\mathcal{LO}}'$, respectively.
2\. Similarly to the notions of adjacent (weight and $t$-) structures and torsion pairs, part \[iorthtw\] of Definition \[dort\] is essentially a particular case of part \[iorthop\] if one takes $w$ and $t$ associated with $s$ and $s'$, respectively, and shifts one of them; see Remarks \[rstws\](4), \[rwhop\](1), and \[rtst1\](\[it1\]).
Now we give some definitions
\[dortt\] 1. We will say that a weight structure $w$ on ${\underline{C}'}$ is [*right $\Phi$-orthogonal*]{} to a $t$-structure $t$ on ${\underline{C}}$ whenever $w$ is left orthogonal to $t$ with respect to the (obviously defined) duality “opposite to $\Phi$”. 2. For an $R$-linear triangulated category ${\underline{C}}$ (where $R$ is an associative commutative unital ring) we will say that an $R$-linear cohomological functor $H$ from ${\underline{C}}$ into $R-{\operatorname{Mod}}$ is [*almost of $R$-finite type*]{} whenever for any $M\in {\operatorname{Obj}}{\underline{C}}$ the $R$-module $H(M)$ is finitely generated and $H(M[i])={\{0\}}$ for $i\ll 0$ (cf. Definition \[dsatur\] and Definition 0.1 of [@neesat]).
Now we prove a generalization of Proposition \[psatur\](II) that enables constructing $t$-structures orthogonal to certain weight ones; its formulation is motivated by related results of [@neesat] and [@roq].
\[psaturdu\] Assume that ${\underline{C}}\subset {\underline{C}'}$ are $R$-linear categories (for $R$ being a unital commutative ring; see Definition \[dsatur\] for some of our notation). Denote by $\Phi$ the restriction of the bifunctor ${\underline{C}'}(-,-)$ to ${\underline{C}}{{^{op}}}\times {\underline{C}'}$.
1\. Assume that functors ${\underline{C}'}\to R-{\operatorname{Mod}}$ that are corepresented by objects of ${\underline{C}}$ are precisely those ones that are of $R$-finite type as functors from ${\underline{C}}'{{^{op}}}$. Then for any bounded weight structure $w'$ on ${\underline{C}'}$ the couple $t=(t_1,t_2) =({{}^{\perp}}({\underline{C}'}_{w'\ge 1})\cap {\operatorname{Obj}}{\underline{C}}, {{}^{\perp}}({\underline{C}'}_{w'\le -1})\cap {\operatorname{Obj}}{\underline{C}})$ is a $t$-structure on ${\underline{C}}$. Moreover, $w'$ is right $\Phi$-orthogonal (see Definition \[dortt\]) to this $t$-structure.
2\. Assume that ${\underline{C}}$ is essentially small and there exists a triangulated category ${\underline{D}}$ that satisfies the following conditions: it has coproducts, ${\underline{C}'}\subset {\underline{D}}$, all objects of ${\underline{C}}$ are compact in ${\underline{D}}$, and for $N\in {\operatorname{Obj}}{\underline{D}}$ the restriction of the functor ${\underline{D}}(-,N)$ to ${\underline{C}}$ is of $R$-finite type (resp. almost of $R$-finite type) if and only if $N\in {\operatorname{Obj}}{\underline{C}'}$. Then for any bounded weight structure $w$ on ${\underline{C}}$ the couple $t'=(t_3,t_4) =(({\underline{C}}_{w\ge 1})^{\perp_{{\underline{C}'}}}, ({\underline{C}}_{w\le -1})^{\perp_{{\underline{C}'}}})$ is a $t$-structure on ${\underline{C}'}$.
Moreover, $w$ is left $\Phi$-orthogonal to this $t'$. Furthermore, if $R$ is noetherian and the correspondence $N\mapsto H_N$ is a full functor from ${\underline{C}'}$ into $\operatorname{\operatorname{AddFun}_R}({\underline{C}}{{^{op}}},R-{\operatorname{mod}})$ (resp. gives an equivalence of ${\underline{C}'}$ with the subcategory of $\operatorname{\operatorname{AddFun}_R}({\underline{C}}{{^{op}}},R-{\operatorname{mod}})$ consisting of functors of $R$-finite type) then the obvious functor from ${{\underline{Ht}}}'$ into $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}{{^{op}}},R-{\operatorname{mod}})$ is full (resp. gives an equivalence of ${{\underline{Ht}}}'$ with the category $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}, R-{\operatorname{mod}})$).
In both assertions the corresponding $\Phi$-orthogonalities are automatic; so we should only check that the corresponding couples are $t$-structures indeed, and study ${{\underline{Ht}}}'$ in assertion 2.
1\. This statement is rather similar to (the dual to) Proposition \[psatur\].
Axioms (i) and (ii) of Definition \[dtstr\] are obvious for $t$.
Next, ${}^{\perp_{{\underline{C}}'}}({\underline{C}'}_{w'\ge 1})={\underline{C}}'_{w'\le 0}$; thus $t_1[1]\perp t_2$.
It remains to verify the existence of $t$-decompositions. For $M\in {\operatorname{Obj}}{\underline{C}}$ we note that all the virtual $t$-truncations of the functors ${\underline{C}'}(M,-)$ (see Remark \[rwrange\](4)) are of $R$-finite type as considered as cohomological functors from ${\underline{C}}'{{^{op}}}$ according to Proposition \[psatur\](I). Hence they are representable by objects of ${\underline{C}}$ according to our assumptions. Thus arguments similar to that used for the proof of Proposition \[padjt\](\[ile4\]) easily allow us to conclude the proof (note here that similarly to Proposition \[padjt\](\[ile2\]) we can apply the Yoneda lemma here since ${\underline{C}}\subset {\underline{C}}'$).
2\. Since $w$ is bounded, the category ${\underline{C}}$ is densely generated by ${\underline{C}}_{w=0}$ (by Proposition \[pbw\](\[igenlm\]); cf. Remark \[rsatur\](2)) and ${{\underline{Hw}}}$ is negative in it. Moreover, we can assume that ${\underline{D}}$ is generated by ${\operatorname{Obj}}{\underline{C}}$ as its own localizing subcategory (see Remark \[rwhop\](2)); thus ${\operatorname{Obj}}{\underline{C}}^{\perp_{{\underline{D}}}}=(\cup_{i\in {{\mathbb{Z}}}}{\underline{C}}_{w=i})^{\perp_{{\underline{D}}}}={\{0\}}$. Hence we can apply [@bws Theorem 4.5.2(I)] to obtain that $t_{{\underline{D}}}=(({\underline{C}}_{w\ge 1})^{\perp_{{\underline{D}}}}, ({\underline{C}}_{w\le -1})^{\perp_{{\underline{D}}}})$ is a $t$-structure on ${\underline{D}}$.[^61]
Next, $w$ is certainly left orthogonal to $t_{{\underline{D}}}$ with respect to the restriction of ${\underline{D}}(-,-)$ to ${\underline{C}}{{^{op}}}\times {\underline{D}}$. Thus Proposition 2.5.4(1) of [@bger] (cf. Proposition \[pwfil\](4)) implies that for $N\in {\operatorname{Obj}}{\underline{D}}$ the restriction of ${\underline{D}}(-,N^{t_{{\underline{D}}}\le 0})$ to ${\underline{C}}$ is isomorphic to $ \tau^{\le 0 }(H_N)$ (where $H_N$ is the restriction of ${\underline{D}}(-,N)$ to ${\underline{C}}$). Now, if $H_N$ is (almost) of $R$-finite type then this virtual $t$-truncation is so as well according to (the corresponding obvious modification of) Proposition \[psatur\](I). Hence $N^{t_{{\underline{D}}}\le 0}$ is an object of $ {\underline{C}'}$ whenever $N$ is. It obviously follows that $t_{{\underline{D}}}$ restricts to ${\underline{C}}'$, i.e., $t'=(t_3,t_4)$ is a $t$-structure on ${\underline{C}}'$.
The proof of the “furthermore” part of the assertion is an easy application of Proposition \[phadj\]; see the proof of Proposition \[psatur\](3) where a particular case of this fact is established.
\[roq\] 1. Let $X$ be a scheme, and take ${\underline{C}}$ being the triangulated category of perfect complexes on $X$, ${\underline{C}}_1'=D^b(X)$ (the bounded derived category of coherent sheaves on $X$), ${\underline{C}}_2'=D^-(X)$ (the bounded above category), ${\underline{D}}=D(QCoh)$ (the unbounded derived category of quasi-coherent sheaves on $X$). Certainly, ${\underline{C}}\subset {\underline{C}}_1'\subset {\underline{C}}_2'\subset {\underline{D}}$, and objects of ${\underline{C}}$ compactly generate ${\underline{D}}$.
Next, assume that $X$ is proper over ${{\operatorname{Spec}\,}}R$, where $R$ is a commutative unital noetherian ring. Then Corollary 0.5 of [@neesat] says that for $N\in {\operatorname{Obj}}{\underline{D}}$ the restriction $H_N$ of the functor ${\underline{D}}(-,N)$ to ${\underline{C}}$ is of $R$-finite type (resp. almost of $R$-finite type) if and only if $N\in {\operatorname{Obj}}{\underline{C}}'_1$ (resp. $N\in {\operatorname{Obj}}{\underline{C}}'_2$). Thus we obtain the existence of the corresponding $t$-structures $t'_i$ both on ${\underline{C}}_1'$ and ${\underline{C}}_2'$; certainly, these $t$-structures are the restrictions of the $t$-structure $t_{{\underline{D}}}$ on ${\underline{D}}$ (see the proof of part 2 of the previous proposition). Moreover, the properties of the correspondence $N\mapsto H_N$ allow us to apply this part 2 to obtain a full functor from ${{\underline{Ht}}}'_2$ into $\operatorname{\operatorname{AddFun}_R}({{\underline{Hw}}}, R-{\operatorname{mod}})$ that restricts to an equivalence of ${{\underline{Ht}}}'_1$ with the latter category. It easily follows that we have ${{\underline{Ht}}}'_1={{\underline{Ht}}}'_2$ in this case.
2\. Assume in addition that $R$ is a field and $X$ is projective over ${{\operatorname{Spec}\,}}R$. Then Lemma 7.49 and Corollary 7.51(ii) of [@roq] (cf. also [@bvdb]\[Theorem A.1\]) also enable us to apply Proposition \[psaturdu\](1) for ${\underline{C}}'={\underline{C}}'_1$.
3\. So, to prove the existence of orthogonal $t$-structures when there is a duality $\Phi:{\underline{C}}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$ using our methods one needs some version of the Brown representability condition or its dual; i.e., one should have a description of the functors $\Phi(M,-)$ for $M\in {\operatorname{Obj}}{\underline{C}}$ (resp. of $\Phi(-,N)$ for $N\in {\operatorname{Obj}}{\underline{C}}'$) that should be respected by the corresponding virtual $t$-truncations (cf. Remark \[rkrause\](II)).
Now we give two simple recipes for constructing nice dualities.
\[pnice\] 1. If $F:{\underline{C}}\to {\underline{D}}$ and $F':{\underline{C}}'\to {\underline{D}}$ are exact functors, then $\Phi(X,Y)={\underline{D}}(F(X),F'(Y)):{\underline{C}}{{^{op}}}\times{\underline{C}}\to {\underline{\operatorname{Ab}}}$ is a nice duality.
2\. For triangulated categories ${\underline{C}}$, ${\underline{C}_0}\subset {\underline{C}}'$, ${\underline{C}_0}$ is skeletally small, let $\Phi_0:{\underline{C}}^{op}\times {\underline{C}_0}\to {\underline{A}}$ be a duality. For any $P\in {\operatorname{Obj}}{\underline{C}}'$ denote by $H^P$ the coextension (see Proposition \[pkrause\]) to ${\underline{C}'}$ of the functor $\Phi_0(-,Y)$; denote by $\Phi$ the pairing ${\underline{C}}^{op}\times {\underline{C}'}\to {\underline{\operatorname{Ab}}}:\ \Phi(P,Y)=H^P(Y)$. Then $\Phi$ is a duality (${\underline{C}}^{op}\times {\underline{C}}'\to {\underline{A}}$); it is nice whenever $\Phi_0$ is.
1\. Easy; it suffices to note that the niceness restriction is a generalization of the axiom (TR3) of triangulated categories (any commutative square can be completed to a morphism of distinguished triangles) to the setting of dualities of triangulated categories. 2. This is the categorical dual to Proposition 2.5.6(3) of [@bger].
On “bicontinuous” dualities {#sdual2}
---------------------------
Now we describe a interesting type of (nice) dualities of triangulated categories and orthogonal torsion pairs in them. We will apply it (below and in [@bgn]) in the case where ${\underline{C}}$ is a certain category of pro-objects; so it is no wonder that we will consider co-compact objects in it.
So we will say that a cohomological functor ${\underline{C}}\to {\underline{A}}$ (${\underline{A}}$ is an abelian category) is pc one if it converts ${\underline{C}}$-products into ${\underline{A}}$-coproducts; recall that an object $M$ of ${\underline{C}}$ is said to be cocompact if the functor ${\underline{C}}(-,M)$ is a pc one (cf. Remark \[rwhop\](5)).
\[porthop\] Let ${\underline{C}_0}$ be an essentially small common subcategory of (triangulated categories) ${\underline{C}}$ and ${\underline{C}'}$ whose objects are compact in ${\underline{C}'}$ and cocompact in ${\underline{C}}$ (and so, ${\underline{C}}$ has products and ${\underline{C}'}$ has coproducts); let ${\mathcal{P}}$ be a set of objects of ${\underline{C}_0}$, $M\in {\operatorname{Obj}}{\underline{C}}$, $N\in {\operatorname{Obj}}{\underline{C}'}$. For each $P\in {\operatorname{Obj}}{\underline{C}}$ denote by $H^P$ the coextension (see Proposition \[pkrause\]) to ${\underline{C}'}$ of the restriction to ${\underline{C}_0}$ of the functor ${\underline{C}}(P,-)$; denote by $\Phi$ the pairing ${\underline{C}}^{op}\times {\underline{C}'}\to {\underline{\operatorname{Ab}}}:\ \Phi(P,Y)=H^P(Y)$.
Then the following statements are valid.
1. \[icudupa\] $\Phi$ is a nice duality of triangulated categories.
2. \[icuduadj\] Denote by ${\underline{E}}$ the colocalizing triangulated subcategory of ${\underline{C}}$ cogenerated by ${\underline{C}_0}$. Then there exists a left adjoint $L$ to the embedding ${\underline{E}}\to {\underline{C}}$ and $\Phi(-,-)\cong \Phi^{{\underline{E}}}(L(-),-)$, where $\Phi^{{\underline{E}}}$ is the restriction of $\Phi$ to ${\underline{E}}{{^{op}}}\times {\underline{C}'}$.
3. \[icontcu\] Assume that $M\in {\operatorname{Obj}}{\underline{C}}$ is a ${\underline{C}_0}$-limit of some $(M_l)$ (see Definition \[drelim\](2)). Then $\Phi(M,N)\cong \operatorname{\varinjlim}\Phi(M_l,N)$. In particular, the functor $\Phi(-,N)$ is a pc one.
4. \[icontdu\] Assume that $N\in {\operatorname{Obj}}{\underline{C}}$ is a ${\underline{C}_0}$-colimit of some $(N_l)$ (see Definition \[drelim\](1)). Then $\Phi(M,N)\cong \operatorname{\varinjlim}\Phi(M,N_l)$. In particular, the functor $\Phi(M,-):{\underline{C}'}\to {\underline{\operatorname{Ab}}}$ is a cc one.
5. \[ibiext\] The functor $\Phi$ is determined (up to a canonical isomorphism) by the following conditions: it converts ${\underline{C}}$-products and ${\underline{C}'}$-coproducts into direct sums of abelian groups, it restriction to ${\underline{C}_0}{{^{op}}}\times {\underline{C}_0}$ equals ${\underline{C}_0}(-,-)$, and it annihilates both ${}^{\perp_{{\underline{C}}}}{\underline{C}_0}\times {\operatorname{Obj}}{\underline{C}'}$ and ${\operatorname{Obj}}{\underline{C}}\times {\underline{C}_0}^{\perp_{{\underline{C}'}}}$. So, in this case we will say that $\Phi$ is the [*biextension*]{} of ${\underline{C}_0}(-,-)$ to ${\underline{C}}^{op}\times {\underline{C}'}$.
6. \[icupr\] There exists a smashing torsion pair $s'=({\mathcal{LO}}',{\mathcal{RO}}')$ for ${\underline{C}}'$ such that ${\mathcal{LO}}'$ is the ${\underline{C}'}$-strong extension-closure of ${\mathcal{P}}$ and ${\mathcal{RO}}'={\mathcal{P}}^{\perp} $.
7. \[icup\] There exists a cosmashing torsion pair $s=({\mathcal{LO}},{\mathcal{RO}})$ for ${\underline{C}}$ such that ${\mathcal{LO}}={{}^{\perp}}{\mathcal{P}}$ and ${\mathcal{RO}}$ is the strong extension-closure of ${\mathcal{P}}$ in ${\underline{C}}{{^{op}}}$. Moreover, $s$ restricts to a torsion pair $s_{{\underline{E}}}$ for ${\underline{E}}$ (see assertion \[icuduadj\] for the latter notation).
8. \[ihoport\] The torsion pairs $s$ and $s_{{\underline{E}}}$ mentioned in the previous assertion are (respectively) $\Phi$-orthogonal and $\Phi^{{\underline{E}}}$-orthogonal to the torsion pair $s'$ from assertion \[icupr\]. Moreover, ${\mathcal{LO}}={}^{\perp_{\Phi}}({\mathcal{LO}}')$ and ${\mathcal{LO}}_{{\underline{E}}}={}^{\perp_{\Phi^{{\underline{E}}}}}({\mathcal{LO}}')$.
9. \[inters\] Both ${\mathcal{LO}}'\cap {\operatorname{Obj}}{\underline{C}_0}$ and ${\mathcal{RO}}\cap {\operatorname{Obj}}{\underline{C}_0}$ are equal to the ${\underline{C}_0}$-envelope of ${\mathcal{P}}$.
<!-- -->
1. Immediate from Proposition \[pnice\].
2. The existence of $L$ is immediate from Proposition \[pcomp\](II). The rest of the assertion follows from the adjunction property of $L$ immediately.
3. It suffices to note that the coextension construction respects colimits; this is obvious (cf. also Proposition \[pkrause\](\[ikr7\])).
4. Immediate from Proposition \[pkrause\](\[ikres\]).
5. The previous assertions easily imply that $\Phi$ satisfies all the properties desired.
Next, Proposition \[pkrause\](\[ikr8\]) implies that $\Phi$ is determined by its restriction to ${\underline{C}_0}{{^{op}}}\times {\underline{C}'}$ along with the conditions that it respects ${\underline{C}'}$-coproducts and annihilates ${\operatorname{Obj}}{\underline{C}}\times {\underline{C}_0}^{\perp_{{\underline{C}'}}}$. Thus one can easily conclude the proof by applying the categorical dual of the aforementioned statement.
6. This is just Theorem \[tclass\](\[iclass1\]).
7. The first part of the assertion is the dual of assertion \[icupr\].
To prove the “moreover” statement we note that the embedding ${\underline{D}}\to {\underline{C}}$ admits an (exact) left adjoint whose kernel is ${\operatorname{Obj}}{\underline{D}}^{\perp_{{\underline{C}}}}={\mathcal{P}}^{\perp_{{\underline{C}}}}$ (by the dual to Proposition \[pcomp\](II.2)). Thus it remains to apply (the dual to) Proposition \[phopft\](II.1).
8. To verify the first part of the assertion it certainly suffices to prove that $s\perp_{\Phi}s'$.
Now recall that all the functors of the type $\Phi(M,-):{\underline{C}'}\to {\underline{\operatorname{Ab}}}$ are cc ones and functors of the type $\Phi(-,N):{\underline{C}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ are pc-ones (see assertions \[icontcu\] and \[icontdu\]). Thus Theorem \[tclass\](\[iclass5\]) (along with its dual) reduces the $\Phi$-orthogonality checks to the following ones: $\Phi(X,Y)=0$ if either $X\in {\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ and $Y\in {\mathcal{P}}^{\perp_{{\underline{C}}}}$ or if $X\in {}^{\perp_{{\underline{D}}}}{\mathcal{P}}$ and $Y\in {\mathcal{P}}\subset {\operatorname{Obj}}{\underline{D}}$. Thus it suffices to note that $\Phi(A,B)$ is isomorphic to ${\underline{C}'}(A,B)$ if $A\in {\operatorname{Obj}}{\underline{C}_0}(\subset {\operatorname{Obj}}{\underline{C}})$ and is isomorphic to ${\underline{C}}(A,B)$ if $B\in {\operatorname{Obj}}{\underline{C}_0}(\subset {\operatorname{Obj}}{\underline{C}'})$. The latter statements are immediate from Proposition \[pkrause\] (parts I.\[ikr1\] and I.\[ikrtriv\], respectively).
Now, to prove the “moreover” statement it remains to note that ${}^{\perp_{\Phi^{{\underline{E}}}}}({\mathcal{LO}}')\subset {\mathcal{LO}}$ since ${\mathcal{LO}}'$ contains ${\mathcal{P}}$.
9. Immediate from Theorem \[tclass\](\[iclass2\]).
\[rcudu\]
1. \[ircocomp\] One may say that all objects of ${\underline{C}}$ are “compact with respect to $\Phi$” and objects of ${\underline{C}'}$ are “cocompact with respect to $\Phi$”; see Proposition \[porthop\](\[icontcu\], \[icontdu\]). Note that both of these properties fail for the duality ${\underline{C}}(-,):{\underline{C}}{{^{op}}}\times {\underline{C}}\to {\underline{\operatorname{Ab}}}$ (and ${\underline{C}'}={\underline{C}}$); so one may say that $\Phi$ is a “regularized Hom bifunctor” that is “bicontinuous”.
2. \[ibcf\] Now assume that $\Phi:{\underline{C}}{{^{op}}}\times {\underline{C}'}\to {\underline{\operatorname{Ab}}}$ is any duality satisfying these bicontinuity conditions and ${\underline{C}'}$ satisfies the Brown representability condition (this is certainly the case whenever ${\underline{C}'}$ is compactly generated). Then one can define a curious functor ${\underline{C}}\to {\underline{C}'}$ as follows.
We note that for any $P\in {\operatorname{Obj}}{\underline{C}}$ there exists its “$\Phi$-Brown-Comenetz dual” ${\mathcal{BCD}}(P)\in {\operatorname{Obj}}{\underline{C}'}$ that ${\underline{C}'}$-represents the functor ${\underline{\operatorname{Ab}}}(\Phi(P,-), {{\mathbb{Q}}}/{{\mathbb{Z}}})$ (since this functor is obviously a cp one). Moreover, the correspondence ${\mathcal{BCD}}$ is easily seen to be exact; it also respects products.
Next, for the torsion pairs $s$ and $s'$ as above we obviously have ${\mathcal{BCD}}({\mathcal{LO}})\subset {\mathcal{RO}}'{{}^{\perp}}$ and ${\mathcal{BCD}}({\mathcal{RO}})\subset {\mathcal{LO}}'{{}^{\perp}}={\mathcal{RO}}'$ (see Proposition \[psymb\](I.\[isbcd\]). Lastly, if ${\mathcal{P}}$ is suspended then for the corresponding $w$ and $t$ (see Corollary \[cwt\] below) and for the weight structure $w'$ right adjacent to $t$ (whose existence is given by Corollary \[csymt\](2)) we obtain that ${\mathcal{BCD}}$ is weight-exact (with respect to $w$ and $w'$; see Definition \[dwso\](\[idwe\])). Note however that the existence of $\Phi$ is a stronger assumption (in this case) then the existence of a weight-exact functor ${\mathcal{BCD}}$ that respects products since the former condition implies that the image of ${\mathcal{BCD}}$ consists of “Brown-Comenetz duals” only.
We will demonstrate the utility of dualities in Theorem \[tab5\] below.
3. \[ir1\] For any ${\underline{C}}$ and ${\underline{C}'}$ as in our proposition and their small subcategories ${\underline{C}}_0$ and ${\underline{C}}_0'$ respectively, one can extend any (nice) duality ${\underline{C}}_0{{^{op}}}\times {\underline{C}}_0'\to {\underline{\operatorname{Ab}}}$ first to a duality ${\underline{C}}{{^{op}}}\times {\underline{C}}_0'\to {\underline{\operatorname{Ab}}}$ using the dual to Proposition \[pkrause\] and next proceed as above to obtain $\Phi: {\underline{C}}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$. It is easily seen that some of the statements proved above have natural (and easy to prove) analogues for this setting.[^62]
Note also that a nice duality ${\underline{C}}_0{{^{op}}}\times {\underline{C}}_0'\to {\underline{\operatorname{Ab}}}$ can be obtained from any exact functors ${\underline{C}}_0\to {\underline{D}}$ and ${\underline{C}}'_0\to {\underline{D}}$ (see Proposition \[pnice\](1)).
4. \[ir2\] Dually, one may start from “coextending” a duality ${\underline{C}}_0{{^{op}}}\times {\underline{C}}_0'\to {\underline{\operatorname{Ab}}}$ to a duality ${\underline{C}_0}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$ and next extend the result to a duality ${\underline{C}}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$ (using the dual to Proposition \[pkrause\]). Combining part \[ikrchar\] and \[ikr7\] of this proposition we obtain the duality obtained this way is isomorphic to the one described above (and obtained in the “reverse order”). Note also that this statement easily follows from the corresponding generalization of Proposition \[porthop\](\[ibiext\]) whenever all objects of ${\underline{C}_0}'$ are compact in ${\underline{C}'}$ and objects of ${\underline{C}_0}$ are cocompact in ${\underline{C}}$.
5. \[ir3\] If one applies the “reverse biextension method” of part \[ir2\] of this remark to the duality ${\underline{C}_0}(-,-)$ (that was the “starting one” in our proposition) then the “intermediate” duality ${\underline{C}_0}{{^{op}}}\times {\underline{C}}'\to {\underline{\operatorname{Ab}}}$ would be the restriction of the bi-functor ${\underline{C}'}(-,-)$ to ${\underline{C}_0}{{^{op}}}\times {\underline{C}}'$; see Proposition \[pkrause\](\[ikrtriv\]).
6. \[ikrausl\] An interesting family of examples for our proposition can easily be constructed using Theorem 4.9 of [@krauslender].
An application: hearts of compactly generated $t$-structures are “usually” Grothendieck abelian {#sgengroth}
-----------------------------------------------------------------------------------------------
First we list the consequences of Proposition \[porthop\] (essentially) in the case where $s$ is a weight structure and $s'$ is a $t$-structure.
\[cwt\] In the setting of the previous proposition (and adopting its notation including the one of its part \[icuduadj\]) assume that ${\mathcal{P}}$ is a suspended subset of ${\operatorname{Obj}}{\underline{C}_0}$ (i.e., ${\mathcal{P}}[1]\subset {\mathcal{P}}$).
Then the following statements are valid.
1. \[ic1\] For any $N'\in {\operatorname{Obj}}{\underline{C}'}$ denote by $H^{N'}$ the extension to ${\underline{C}}$ of the functor $H_0^{N'}:{\underline{C}_0}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ (that is the restriction of ${\underline{C}'}(-,N')$ to ${\underline{C}_0}$) obtained via the dual to Proposition \[pkrause\]. Then the bi-functor $\Phi:{\underline{C}}{{^{op}}}\times {\underline{C}'}\to {\underline{\operatorname{Ab}}}: (M,N')\mapsto H^{N'}(M)$ is a nice duality that is naturally isomorphic to the one given by Proposition \[porthop\](\[icudupa\]).
2. \[iccc\] For any $M\in {\operatorname{Obj}}{\underline{C}}$ the functor $\Phi(M,-):{\underline{C}'}\to {\underline{\operatorname{Ab}}}$ is a cc one.
3. \[ic2\] The functor $L:{\underline{C}}\to{\underline{E}}$ (left adjoint to the embedding ${\underline{E}}\to {\underline{C}}$) respects products, ${\underline{C}_0}$-limits, and is identical on ${\operatorname{Obj}}{\underline{E}}$.
4. \[ic15\] The restriction $\Phi^{{\underline{E}}}$ of $\Phi$ to ${\underline{E}}{{^{op}}}\times {\underline{C}'}$ is a nice duality also, and we have $\Phi(-,-)\cong \Phi^{{\underline{E}}}(L(-),-)$.
5. \[ic25\] For any set of $M_i\in {\operatorname{Obj}}{\underline{C}}$ and $N\in {\operatorname{Obj}}{\underline{C}'}$ we have $\Phi(\prod M_i,N)\cong \bigoplus \Phi (M_i,N)$.
6. \[ict\] There exists a smashing $t$-structure $t$ on ${\underline{C}'}$ such that ${\underline{C}'}^{t\le 0}$ is the smallest coproductive extension-closed subclass of ${\operatorname{Obj}}{\underline{C}'}$ containing ${\mathcal{P}}$ and ${\underline{C}'}^{t\ge 0}={\mathcal{P}}{{}^{\perp}}[1]$. Moreover, ${\underline{C}'}^{t\le 0}$ also equals the big hull of ${\mathcal{P}}$ in ${\underline{C}'}$.
7. \[icw\] There exists a cosmashing weight structure $w$ on ${\underline{C}}$ such that ${\underline{C}}_{w\ge 0}$ is the big hull of ${\mathcal{P}}$ in ${\underline{C}}{{^{op}}}$ and ${\underline{C}}_{w\le 0}=({}^{\perp_{{\underline{C}}}}{\mathcal{P}})[1]$. Moreover, ${\underline{C}}_{w\le 0}$ is the extension-closure of $({}^{\perp_{{\underline{E}}}}{\mathcal{P}})[1]\cup {{}^{\perp}}(\cup_{i\in {{\mathbb{Z}}}}{\mathcal{P}})$.
8. \[icwe\] The couple $w_{{\underline{E}}}= (({}^{\perp_{{\underline{E}}}}{\mathcal{P}})[1], {\underline{C}}_{w\ge 0})$ is a cosmashing weight structure on ${\underline{E}}$, $L$ is weight-exact (with respect to $w$ and $w_{{\underline{E}}}$, respectively), and ${\underline{C}}_{w=0}={\underline{E}}_{w_{{\underline{E}}}=0}$. Moreover, if ${\mathcal{P}}$ densely generates ${\underline{C}_0}$ then $w_{{\underline{E}}}$ is right non-degenerate.
9. \[icort\] $w$ and $w_{{\underline{E}}}$ are orthogonal to $t$ with respect to $\Phi$ and $\Phi^{{\underline{E}}}$, respectively. Moreover, ${\underline{C}}_{w\le 0}={}^{\perp_{\Phi}}{\underline{C}'}^{t\le -1}$ and ${\underline{E}}_{w_{{\underline{E}}}\le 0}={}^{\perp_{\Phi^{{\underline{E}}}}}{\underline{C}'}^{t\le -1}$.
10. \[iwheart\] Choose some $w_{{\underline{E}}}$-weight complexes for elements of ${\mathcal{P}}$; denote their terms by $P^k_l$. Then the object $I=\prod P^k_l$ cogenerates ${{\underline{Hw}}}$ (cf. Corollary \[cvttbrown\](I.2)), i.e., any object of ${{\underline{Hw}}}$ is a retract of a product of copies of $I$.
11. \[icint\] The ${\underline{C}_0}$-envelope of ${\mathcal{P}}$ equals both ${\underline{C}'}^{t\le 0}\cap {\operatorname{Obj}}{\underline{C}_0}$ and ${\underline{C}}_{w\ge 0}\cap {\operatorname{Obj}}{\underline{C}_0}$.
12. \[icvirt\]If ${\underline{A}}$ is an AB5 category then for the extension (obtained via the dual to Proposition \[pkrause\]) $H:{\underline{C}}{{^{op}}}\to {\underline{A}}$ of a cohomological functor $H_0$ from ${\underline{C}_0}$ into ${\underline{A}}$ all its virtual $t$-truncations are extended functors (in the same sense as $H$ is).
\[ic1\]. Easy from Proposition \[porthop\](\[ibiext\]); see Remark \[rcudu\] (\[ir2\],\[ir3\]).
\[iccc\]. Immediate from the previous assertion combined with Proposition \[porthop\](\[icontdu\]).
\[ic2\]. Obviously, $L$ respects ${\underline{C}_0}$-limits and is identical on ${\operatorname{Obj}}{\underline{E}}$. Dualizing Proposition \[prtst\](\[it4sm\]) we obtain that $L$ respects products.
\[ic15\]. Certainly, (arbitrary) restrictions of nice dualities are nice dualities also. It remains to apply Proposition \[porthop\](\[icuduadj\]).
\[ic25\]. Immediate from Proposition \[porthop\](\[icontcu\]).
\[ict\]. Certainly, ${\mathcal{P}}$ generates a smashing torsion pair $s'=({\mathcal{LO}}',{\mathcal{RO}}')$ for ${\underline{C}'}$ with ${\mathcal{RO}}'={\underline{C}'}^{t\ge 0}$ and ${\mathcal{LO}}'$ being the big hull of ${\mathcal{P}}[1]$; see Proposition \[porthop\](\[icupr\]). Hence the assertion follows from Corollary \[csymt\](1) easily.
\[icw\]. The first part of the assertion is immediate from Proposition \[porthop\](\[icup\]); see Remark \[rwhop\](1). The second part is immediate from Proposition \[phopft\](II.1).
\[icwe\]. Proposition \[porthop\](\[icup\]) gives the existence of $w_{{\underline{E}}}$, which certainly implies the weight-exactness of $L$ and ${\underline{C}}_{w=0}={\underline{E}}_{w_{{\underline{E}}}=0}$. To prove the “moreover” part one should note that ${}^{\perp_{{\underline{E}}}}{\operatorname{Obj}}{\langle}{\mathcal{P}}{\rangle}={}^{\perp_{{\underline{E}}}}{\operatorname{Obj}}{\underline{C}_0}={\{0\}}$ according to (the dual to) Proposition \[pcomp\](I.1); it remains to apply (the dual to) Remark \[rwhop\](8).
\[icort\]. Immediate from Proposition \[porthop\](\[ihoport\]) (see Remark \[rort\](2)).
\[iwheart\]. Immediate from (the dual to) Proposition \[ppcoprws\](\[icopr7\]) (cf. Corollary \[cvttbrown\](I.2)).
\[icint\]. This is just a particular case of Proposition \[porthop\](\[inters\]).
\[icvirt\]. Immediate from Remark \[rkrause\](II.2).
Now we apply this corollary to the study of compactly generated $t$-structures. First we study the AB5 condition.
\[tab5\] Let ${\underline{D}}$ be a triangulated category having coproducts and $t$ be a smashing $t$-structure on it.
Then the following statements are valid.
1\. Assume moreover that ${{\underline{Ht}}}$ is an AB5 category and ${\underline{D}}$ is compactly generated. Then the (essentially small) class $H_0^t({\underline{D}}^{{{\aleph_0}}})$ generates ${{\underline{Ht}}}$ (and so, the coproduct of any small skeleton of $H_0^t({\underline{D}}^{{{\aleph_0}}})$ generates it also).[^63]
2\. Assume that $t$ is generated by a suspended set ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{D}}$ of compact objects. Denote by ${{\underline{D}}_0}$ the subcategory of ${\underline{D}}$ that is densely generated (see §\[snotata\]) by ${\mathcal{P}}$ and assume that there exists a triangulated category ${\underline{D}}'$ that contains ${\underline{D}}_0^{op}$ as a full subcategory of compact objects. Then ${{\underline{Ht}}}$ is a Grothendieck abelian category, and there exists a faithful exact functor ${\mathcal{S}}:{{\underline{Ht}}}\to {\underline{\operatorname{Ab}}}$ that respects coproducts.
1\. Proposition \[pkrause\](\[ikr8\]) implies that the functor $H_0^t:{\underline{D}}\to {{\underline{Ht}}}$ is coextended (from the subcategory ${\underline{D}}'$ of ${\underline{D}}$ whose object class equals ${\underline{D}}^{{{\aleph_0}}}$). Hence for any $M\in {\operatorname{Obj}}{\underline{D}}$ and a family $C_M^j\in {\underline{D}}^{{{\aleph_0}}}$ as in (\[ekrause\]) we obtain that $H_0^t(M)$ is a ${{\underline{Ht}}}$-quotient of $\coprod H_0^t(C_M^j)$. Hence the class $H_0^t({\underline{D}}^{{{\aleph_0}}})$ generates ${{\underline{Ht}}}$ indeed. 2. The embedding into ${\underline{D}}$ of its localizing subcategory generated by ${{\underline{D}}_0}$ possesses an exact right adjoint (by Proposition \[pcomp\](II.2)); hence Remark \[rwhop\](2) allows us to assume that ${\underline{D}}$ equals this subcategory. Thus ${\underline{D}}$ is compactly generated. Hence assertion 1 says that it suffices to verify whether ${{\underline{Ht}}}$ is an AB5 category.
Next, ${{\underline{Ht}}}$ is an AB4 category according to Proposition \[prtst\](\[it2\]). Hence it is cocomplete; thus to check that it is AB5 we should prove that filtered colimits of ${{\underline{Ht}}}$-monomorphisms are ${{\underline{Ht}}}$-monomorphic. Thus it suffices to construct ${\mathcal{S}}$ (since small colimits can be expressed in terms of coproducts). Note also that an exact functor (of abelian categories) is faithful if and only if it is conservative (and this is equivalent to the assumption that ${\mathcal{S}}$ does not kill non-zero objects).
Next we note that exact functors ${{\underline{Ht}}}\to {\underline{\operatorname{Ab}}}$ are precisely the restrictions to ${{\underline{Ht}}}$ of those homological functors ${\underline{C}}\to {\underline{\operatorname{Ab}}}$ that annihilate ${\underline{C}}^{t\le -1}\cup {\underline{C}}^{t\ge 1}$. Hence it suffices to find a cc functor ${\mathcal{S}}^{{\underline{D}}}:{\underline{D}}\to {\underline{\operatorname{Ab}}}$ that kills ${\underline{C}}^{t\le -1}\cup {\underline{C}}^{t\ge 1}$ and whose restriction to ${{\underline{Ht}}}$ is conservative.
We start with constructing a “big” conservative family of cc functors ${\underline{D}}\to {\underline{\operatorname{Ab}}}$ satisfying the vanishing condition above; we call them [*stalk*]{} ones for the reason that will be explained in Remark \[rsheaves\](1) below.[^64] For this purpose we apply the previous corollary for ${\underline{C}}'$ equal to ${\underline{D}}$ and ${\underline{C}}={\underline{D}}'^{op}$. We obtain the existence of a duality $\Phi:{\underline{C}}{{^{op}}}\times {\underline{D}}\to {\underline{\operatorname{Ab}}}$ and a weight structure $w$ on ${\underline{C}}$ that is (left) $\Phi$-orthogonal to $t$.
Our stalk functors are the functors $\Phi(P,-):{\underline{D}}\to {\underline{\operatorname{Ab}}}$ for $P$ running through ${\underline{C}}_{w=0}$. The stalk functors are certainly exact (by the definition of a duality); they annihilate ${\underline{C}}^{t\le -1}\cup {\underline{C}}^{t\ge 1}$ since $w\perp_{\Phi}t$. The stalk functors are cc according to Corollary \[cwt\](\[iccc\]).
Let us verify the conservativity of our family. Let $N$ be a non-zero element of ${\underline{D}}^{t=0}$; we should verify that the restriction $A$ of the functor $\Phi_N=\Phi(-, N):{\underline{D}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ to ${{\underline{Hw}}}{{^{op}}}$ is not zero. Now, the $\Phi$-orthogonality of $w$ to $t$ along with Proposition \[pwrange\](\[iwrpure\]) implies that $\Phi_N$ is a pure functor; hence it equals $H_{{\mathcal{A}}}$ in the notation of Definition \[drange\](2). Thus to check that $A\neq 0$ it suffices to prove that $\Phi_N\neq 0$. Now, the class ${\operatorname{Obj}}{{\underline{D}}_0}$ Hom-generates ${\underline{D}}$; hence there exists $M_0\in {\operatorname{Obj}}{{\underline{D}}_0}$ such that ${\underline{D}}(M_0,N)\cong \Phi(M_0, N)\neq{\{0\}}$.
Lastly, we recall that ${{\underline{Hw}}}$ has a cogenerator (see Corollary \[cwt\](\[iwheart\])). Thus according to Corollary \[cwt\](\[ic25\]) one can take ${\mathcal{S}}^{{\underline{D}}}$ to be the stalk functor corresponding to this cogenerator.
\[rab5\]
1. \[irrel\] Note that in the case where ${{\underline{D}}_0}{{^{op}}}$ embeds into the subcategory of compact objects of ${\underline{D}}$ (in particular, this is the case if the latter category is anti-isomorphic to itself) one can take ${\underline{D}}'={\underline{D}}$. A toy example of this situation is ${\underline{D}}$ being the derived category $D(R-{\operatorname{Mod}})$, where $R$ is a commutative ring; ${{\underline{D}}_0}$ is the category of perfect complexes. More generally, for ${\underline{D}}=R-{\operatorname{Mod}}$, where $R$ is a not (necessarily) commutative ring, one may take ${\underline{D}}'$ being the derived category of right $R$-modules; this example has natural “differential graded” (see §5.2 of [@bger]) and probably “spectral” (see [@schwmod]) generalizations.
On the other hand, Corollary \[cgdb\](\[iab5\]) below demonstrates that our theorem can be applied for ${\underline{D}}$ being the homotopy category of an arbitrary proper simplicial stable (Quillen) model category.[^65] Hence our theorem gives a positive answer to Question 3.8 of [@parrasao] for a really wide range of triangulated categories. Recall also that Theorem 3.7 of loc. cit. says that [**countable**]{} colimits in ${{\underline{Ht}}}$ are exact for any compactly generated $t$.[^66]
Thus our theorem demonstrates once again that “weight structures can shed some light” on $t$-structures; cf. Corollary \[csymt\]. The author wonders whether one can mimic the argument above (and so, obtain certain stalk functors) using the “naive” category ${\operatorname{Pro}}-{{\underline{D}}_0}$ instead of ${\underline{D}}$ (note that ${\operatorname{Pro}}-{{\underline{D}}_0}$ is a [*pro-triangulated*]{} category that is “very rarely” triangulated).
2. \[ismgen\] The author also wonders whether one can describe a set of generators for ${{\underline{Ht}}}$ that is smaller than that described in part 1 of our theorem. If $t$ is generated by a set ${\mathcal{P}}$ of compact objects then a natural candidate here is the (essentially small) class of $H_0^t(P)$ for $P$ running through elements of ${\underline{D}}^{t\le 0}\cap {\underline{D}}^{{{\aleph_0}}}$. Note that to prove that this class generates ${{\underline{Ht}}}$ is suffices to describe a generating family of stalk functors that could be presented as colimits of functors corepresented by elements of ${\underline{D}}^{t\le 0}\cap {\underline{D}}^{{{\aleph_0}}}$ (see Proposition \[pgen\](\[ipgen1\])). In particular, the proof of [@bondegl Proposition 4.2.10] is essentially an argument of this sort.
Moreover, the argument described in Remark \[rkrause\](I.2) above yields that ${{\underline{Ht}}}$ is generated by the (even smaller) class $H_0^t({\mathcal{Q}})$, where $Q$ is the smallest subclass of ${\operatorname{Obj}}{\underline{C}}$ containing ${\mathcal{P}}$ such that for any triangle $Z[-1]\to X\to Y\to Z$ with $Y,Z\in {\mathcal{Q}}$ the object $X$ belongs to ${\mathcal{Q}}$ also.
3. \[irpostov\] Proposition \[porthop\](\[inters\]) (cf. also Remark \[rnewt\](2)) certainly gives a one-to-one correspondence between weight structures (resp. $t$-structures) generated by subsets of ${\operatorname{Obj}}{{\underline{D}}_0}$ in ${\underline{D}}$ and $t$-structures (resp. weight structures) generated by these sets in ${\underline{D}}'$. This observation was very successively applied in [@postov §4]. However, (the proof of) our theorem (cf. also the next part of this remark) demonstrates that introducing a duality between ${\underline{D}}'^{op}$ and ${\underline{D}}$ can give information that can hardly be obtained if ${\underline{D}}$ and ${\underline{D}}'$ are considered “separately” only.
4. \[iremb\] Proposition 6.2.1 of [@bger] suggests the following conjecture: ${{\underline{Ht}}}$ is equivalent to the category of those functors ${{\underline{Hw}}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ that respect coproducts. The conjecture certainly contains more information on ${{\underline{Ht}}}$ than Theorem \[tab5\] (along with its proof); however, this description of ${{\underline{Ht}}}$ may be not that “useful” (since ${{\underline{Hw}}}$ can be rather “complicated”).
5. \[ir4\] An interesting question (that is closely related to the aforementioned conjecture) is which homological functors from ${\underline{D}}$ (resp. from ${\underline{D}}$) are [*$\Phi$-corepresentable*]{}, i.e., have the form $\Phi(-,N)$ (resp. $\Phi(M,-)$) for some $N\in {\operatorname{Obj}}{\underline{D}}$ (resp. $M\in {\operatorname{Obj}}{\underline{D}}'$). Now, the functors of the form $\Phi(-,N):{\underline{D}}'\to {\underline{\operatorname{Ab}}}$ are precisely the “extensions” (obtained using the dual to Proposition \[pkrause\]) of the functors that are ${\underline{D}}$-represented on ${{\underline{D}}_0}$. In the case where ${{\underline{D}}_0}$ is [*countable*]{} (i.e., its object class is essentially countable and all morphism sets are at most countable) all cohomological functors from ${{\underline{D}}_0}$ into ${\underline{\operatorname{Ab}}}$ are represented by objects of ${\underline{D}}$ according to Theorem 5.1 of [@neebrh]; so, we obtain a complete description of $\Phi$-representable functors in this case. Unfortunately, this argument cannot be extended to the case of a general ${{\underline{D}}_0}$.
6. \[istalks\] The stalk functors $\Phi(P,-)$ for $P\in {\underline{D}}'^{op}_{w=0}$ in the proof of Theorem \[tab5\] essentially play the role of functors corepresented by $t$-projective objects of ${\underline{C}}$ (see Definition \[dpt\]). Note however we cannot have “enough” of the latter unless ${{\underline{Ht}}}$ has enough projectives (see Proposition \[pgen\](\[ipgen25\])). If we assume in addition the existence of (a set of) “compact generators” for $P_t$ (that are necessary to obtain enough cc functors) then it would imply that ${{\underline{Ht}}}$ is isomorphic to $\operatorname{\operatorname{AddFun}}({\underline{B}},{\underline{\operatorname{Ab}}})$ for ${\underline{B}}$ being the corresponding small additive category. Certainly, ${{\underline{Ht}}}$ “rarely” can be presented in this form; (cf. Remark \[rsheaves\](1) below). Hence constructing a duality of ${\underline{D}}$ with an “auxiliary” category ${\underline{D}}'$ is necessary for the proof of Theorem \[tab5\].
7. [irotimes]{} In Proposition \[porthop\] the “starting duality” was ${\underline{C}_0}(-,-)$. It is an interesting question whether Theorem \[tab5\] can be generalized by treating (biextensions of) other dualities (cf. Remark \[rcudu\](\[ir1\],\[ir2\])).
Anyway, it appears that one can construct “interesting” dualities using various tensor products. Assume that we are given triangulated categories ${\underline{C}}$ and ${\underline{C}'}$ as above, a triangulated category ${\underline{E}}$ having coproducts, a bi-exact functor $\otimes: {\underline{C}}{{^{op}}}\times {\underline{C}'}\to {\underline{E}}$ that respects coproducts when any of the arguments is fixed, and a cc functor $H:{\underline{E}}\to {\underline{A}}$ (for some abelian ${\underline{A}}$). Then $\Phi(-,-)=H(-\otimes -)$ is a duality ${\underline{C}}{{^{op}}}\times {\underline{C}'}\to {\underline{A}}$ that converts ${\underline{C}}$-products and ${\underline{C}'}$-coproducts into ${\underline{A}}$-coproducts. Certainly, one may take $H$ being the functor corepresented by a compact object of ${\underline{E}}$ (and ${\underline{A}}={\underline{\operatorname{Ab}}}$). Note also that $\Phi$ is canonically characterized by its “coproductivity properties” along with its restriction $\Phi_0$ to ${\underline{C}}'_0{}{{^{op}}}\times {\underline{C}_0}$ whenever ${\underline{C}_0}$ and ${\underline{C}}'_0{}^{op}$ are categories of compact objects in ${\underline{C}}{{^{op}}}$ and ${\underline{C}}'$ (respectively) that Hom-generate these categories (and so, generate them as their own localizing subcategories). The author believes that dualities of this sort may be useful for “computing” dualities of the type treated in Proposition \[porthop\] in the case where ${\underline{C}}{{^{op}}}={\underline{C}'}$, ${\underline{C}_0}$ is self-dual with respect to the tensor product on ${\underline{C}'}$ (cf. part \[irrel\] of this remark) and Hom-generates ${\underline{C}'}$, and the unit object ${{\pmb{1}}}_{{\underline{C}'}}$ is compact in ${\underline{C}'}$. Note that all these condition are fulfilled for ${\underline{C}'}$ being the stable homotopy category of (“topological”) spectra; we will say more on this setting (that was essentially treated in [@prospect]) in Remark \[rsheaves\](4) below.
8. \[krauseideals\] Recall that Corollary 4.7 of [@krause] gives a description of all smashing (see Remark \[rtst2\](\[ismashs\])) subcategories of a compactly generated triangulated category ${\underline{D}}$ in terms of certain ideals of morphisms in its subcategory ${{\underline{D}}_0}$ of compact objects. Along with Theorem 4.9 of ibid. and the example from [@kellerema] this appears to yield an example of a not compactly generated smashing triangulated subcategory $L$ of ${\underline{D}}$ such that the corresponding “shift-stable $t$-structure” (see Remark \[rtst2\](\[ismashs\]) again) possesses a (left) $\Phi$-orthogonal shift-stable weight structure in the corresponding ${\underline{D}}'$. The authors wonders whether one can also obtain similar non-shift-stable examples.
Relation to triangulated categories of pro-objects {#sprospectra}
--------------------------------------------------
Now we describe a method for constructing a vast family of examples for Proposition \[porthop\] (and so, also of Corollary \[cwt\]) along with Theorem \[tab5\]. Its main ingredient is a construction of a triangulated category of “homotopy pro-objects” for a stable model category that is a straightforward application of the results of [@tmodel].
So let ${\mathcal{M}}$ be a proper simplicial stable (Quillen) model category; denote its homotopy category by ${\underline{C}'}$. We construct another model category ${\operatorname{Pro}-\mathcal{M}}$ whose underlying category is the category of (filtered) pro-objects of ${\mathcal{M}}$ (cf. §5 of [@tmodel]).
\[rproobj\] In [@tmodel] pro-objects were defined via filtered diagrams, i.e., via contravariant functors from filtered small categories. However the author prefers (for the reasons of minor notational convenience) to consider inverse limits indexed by filtered sets instead. The latter notion is certainly somewhat more restrictive formally; however, Theorem 1.5 of [@adr] says that any “categorical” filtered colimit can be presented as a certain “set-theoretic” one. It easily follows that it is no difference between the usage of these two notions (for the purposes of the current papers as well as for that of [@bgn]); cf. the discussion in §2 of [@isalim].
We endow ${\operatorname{Pro}-\mathcal{M}}$ with the [*strict*]{} model structure; see §5.1 of ibid. (so, weak equivalences and cofibrations are [*essential levelwise*]{} weak equivalences and cofibrations of pro-objects). An important observation here is that this model structure is a particular case of a [*$t$-model structure*]{} in the sense of §6 of ibid. if one takes the following “totally degenerate” $t$-structure $t_{deg}$ on ${\underline{C}'}$: ${\underline{C}'}^{t_{deg}\ge 0}={\operatorname{Obj}}{\underline{C}'}$, ${\underline{C}'}^{t_{deg}\le 0}={\{0\}}$; see Remark 6.4 of ibid. Indeed, one can take the following functorial factorization of morphisms: for $f\in {\mathcal{M}}(X,Y)$ (for $X,Y\in {\operatorname{Obj}}{\mathcal{M}}$) we can present it as $f\circ {\operatorname{id}}_X$; note that ${\operatorname{id}}_X$ is an [*$n$-equivalence*]{} and $f$ is a [*co-$n$-equivalence*]{} in the sense of Definition 3.2 of ibid. for any $n\in {{\mathbb{Z}}}$ (pay attention to Remark \[rstws\](3)!).
Now let us describe some basic properties of ${\operatorname{Pro}-\mathcal{M}}$ and its homotopy category ${\underline{C}}$ (we will apply some of them below, whereas other ones are important for [@bgn]). The pro-object corresponding to a projective system $M_i$ for $i\in I$ where $I$ is an inductive set and $M_i\in {\operatorname{Obj}}{\mathcal{M}}$, will be denoted by $(M_i)$. Note that $(M_i)$ is precisely the (inverse) limit of the system $M_i$ in ${\operatorname{Pro}-\mathcal{M}}$ (by the definition of morphisms in this category).
\[pgdb\]
Let $X_i,Y_i,Z_i\ i\in I$, be projective systems in ${\mathcal{M}}$. Then the following statements are valid.
1\. ${\operatorname{Pro}-\mathcal{M}}$ is a proper stable simplicial model category; hence ${\underline{C}}$ is triangulated. 2. If some morphisms $X_i\to Y_i$ for all $i\in I$ yield a compatible system of cofibrations (resp. of weak equivalences; resp. some couples of morphisms $X_i\to Y_i\to Z_i$ yield a compatible system of homotopy cofibre sequences) then the corresponding morphism $(X_i)\to (Y_i)$ is a cofibration also (resp. a weak equivalence; resp. the couple of morphisms $(X_i)\to (Y_i)\to (Z_i)$ is a homotopy cofibre sequence). 3. The natural embedding $c:{\mathcal{M}}\to {\operatorname{Pro}-\mathcal{M}}$ is a left Quillen functor; it also respects weak equivalences and fibrations.
4\. For any $N\in {\operatorname{Obj}}{\mathcal{M}}$ we have ${\underline{C}}((X_i),c(N))\cong \operatorname{\varinjlim}{\underline{C}'}(X_i,N)$.
In particular, the homotopy functor ${\operatorname{Ho}}(c):{\underline{C}'}\to {\underline{C}}$ is a full embedding, and $(X_i)$ is a ${\underline{C}'}$-limit of $X_i$ in ${\underline{C}}$ (with respect to this embedding; see Definition \[drelim\](2)).
5\. More generally, for any projective system $\{M_i\}$ in $ {\operatorname{Pro}-\mathcal{M}}$ and any $N\in {\operatorname{Obj}}{\underline{C}'}$ the inverse limit of $M_i$ exists in ${\operatorname{Pro}-\mathcal{M}}$ and we have ${\underline{C}}(\operatorname{\varprojlim}M_i,c(N))\cong \operatorname{\varinjlim}_{i\in I} {\underline{C}}(M_i,c(N))$.
6\. ${\underline{C}}$ has products and all objects of ${\operatorname{Ho}}(c)({\underline{C}'})$ are cocompact in ${\underline{C}}$. 7. The class ${\operatorname{Ho}}(c)({\operatorname{Obj}}{\underline{C}'})$ cogenerates ${\underline{C}}$.
1\. Theorems 6.3 and 6.13 of [@tmodel] yield everything except the existence of functorial factorizations for morphisms in ${\operatorname{Pro}-\mathcal{M}}$. The existence of functorial factorizations is given by Theorem 1.3 of [@fufa] (see the text following Remark 1.5 of ibid.). One can also deduce this statement from [@prospect Remark 4.5].
2\. The first two parts of the assertion are contained in the definition of the strict model structure. The last part follows from the previous ones immediately (recall that pushouts can be computed levelwisely); this fact is also mentioned in the proof of [@tmodel Proposition 9.4].
3\. The first part of the assertion is given by Lemma 8.1 (an also by §5.1) of [@tmodel]. The second part is immediate from the description of weak equivalences in ${\operatorname{Pro}-\mathcal{M}}$ given in loc. cit.
4\. The first of the statements is immediate from Corollary 8.7 of ibid.; the other ones are its obvious consequences.
5\. The first part of the assertion is provided by Theorem 4.1 of [@isalim]. Since loc. cit. also (roughly) says that filtered limits in ${\operatorname{Pro}-\mathcal{M}}$ can be naturally expressed in terms of limits in ${\mathcal{M}}$, combining this statement with assertion 4 we obtain the second part of assertion 5 immediately.
6\. ${\underline{C}}$ has products since it is the homotopy category of a model category (see Example 1.3.11 of [@hovey]). Hence we should verify that ${\underline{C}}(\prod_{i\in I} Y^i,X)=\bigoplus_{i\in I} {\underline{C}}(Y^i,X)$ for $Y^i$ being fibrant objects of ${\operatorname{Pro}-\mathcal{M}}$, $X\in {\operatorname{Ho}}(c)({\operatorname{Obj}}{\underline{C}'})$. Now (see loc. cit.) the product of $Y^i$ in ${\underline{C}}$ comes from their product in ${\operatorname{Pro}-\mathcal{M}}$. Certainly, if $Y^i=(Y_{j}^i)$ then the product of $Y^i$ in ${\operatorname{Pro}-\mathcal{M}}$ can be presented by the projective system of all $\prod_{i\in J} Y^i_{j_i}$ for finite $J\subset I$. Hence the statement follows from the previous assertion (recall that products are particular cases of inverse limits).[^67]
7\. Theorem 6.1 of [@chor] implies that ${\operatorname{Pro}-\mathcal{M}}$ admits a non-functorial version of the generalized cosmall object argument with respect to $c(f)$. Hence we can apply the dual of the argument used in the proof of Theorem 7.3.1 of [@hovey].
We deduce some consequences from this statement (mostly) using Proposition \[porthop\]. We will consider ${\underline{C}'}$ as a full subcategory of ${\underline{C}}$ (via the embedding $c$ that we will not mention) in this corollary.
\[cgdb\]
In the setting of the previous proposition assume that ${\underline{C}_0}$ is an essentially small subcategory of ${\underline{C}'}$ consisting of compact objects; let ${\mathcal{P}}$ be a suspended subset of ${\operatorname{Obj}}{\underline{C}_0}$. Denote by ${\underline{E}}$ the colocalizing subcategory of ${\underline{C}}$ cogenerated by ${\operatorname{Obj}}{\underline{C}_0}$; let $X_i$ be a projective system in ${\operatorname{Pro}-\mathcal{M}}$ and $N\in {\operatorname{Obj}}{\underline{C}'}$.
Then the following statements are valid.
1. \[iprev\] One can apply Corollary \[cwt\] to this setting.
2. \[iab5\] ${{\underline{Ht}}}$ is a Grothendieck abelian category and there exists a faithful exact functor ${\mathcal{S}}:{{\underline{Ht}}}\to {\underline{\operatorname{Ab}}}$ that respects coproducts.
3. \[ic3\] $\Phi^{{\underline{E}}}(L(\operatorname{\varprojlim}X_i),N)\cong \operatorname{\varinjlim}\Phi^{{\underline{E}}} (L(X_i),N)$.
4. \[icf\] More generally, if ${\underline{A}}$ is an AB5 category and $H$ from ${\underline{C}}$ into ${\underline{A}}$ is an extended functor (i.e., it is obtained via the dual to Proposition \[pkrause\] from a cohomological functor $H_0$ from ${\underline{C}_0}$ into ${\underline{A}}$) then we have $H(\operatorname{\varprojlim}X_i)\cong H(L(\operatorname{\varprojlim}X_i))\cong \operatorname{\varinjlim}H(L(X_i))\cong \operatorname{\varinjlim}H(X_i)$.
5. \[icfig\] $\Phi^{{\underline{E}}}$ is isomorphic to the restriction to ${\underline{E}}{{^{op}}}\times {\underline{C}}'$ of ${\underline{C}}(-,-)$.
6. \[icharw\] We have ${\underline{E}}_{w_{{\underline{E}}}\le 0}=({}^{\perp_{{\underline{C}}}}{\underline{C}'}^{t\le -1})\cap {\operatorname{Obj}}{\underline{E}}$.
\[iprev\]. Immediate from the previous proposition.
\[iab5\]. According to the previous assertion, we can apply Theorem \[tab5\] to our setting.
Assertion \[ic3\] is a particular case of assertion \[icf\] indeed (by the definition of $\Phi$). Next, since $L$ respects ${\underline{C}_0}$-limits, assertion \[icf\] follows from Proposition \[pgdb\](6) according to (the dual to) Proposition \[pkrause\](\[ikres\]).
\[icfig\]. According to Proposition \[pkrause\](\[ikr8\]) (cf. also Proposition \[porthop\](\[ibiext\])) it suffices to verify that for any $M\in {\operatorname{Obj}}{\underline{E}}$ the functor ${\underline{C}}(M,-):{\underline{C}}\to {\underline{\operatorname{Ab}}}$ is a cc one. Now, this condition is certainly fulfilled if $M\in {\operatorname{Obj}}{\underline{C}_0}$. Next, the class of objects of ${\underline{E}}$ satisfying this condition is shift-stable; hence it also closed with respect to extensions. Thus it remains to verify for a set of $M_i\in {\operatorname{Obj}}{\underline{E}}$ that the functor ${\underline{C}}(\prod M_i,-):{\underline{C}'}\to {\underline{\operatorname{Ab}}}$ is a cc one if all ${\underline{C}}(M_i,-)$ are. The latter implication follows immediately from Proposition \[pgdb\](6).
\[icharw\]. The previous assertion certainly implies that $({}^{\perp_{{\underline{C}}}}{\underline{C}'}^{t\le -1})\cap {\operatorname{Obj}}{\underline{E}}={}^{\perp_{\Phi^{{\underline{E}}}}}{\underline{C}'}^{t\le -1}$. Hence it remains to apply Corollary \[cwt\] (\[icort\]).
This proposition is applied in [@bgn] to various motivic homotopy categories. The corresponding $t$ is a (version of) the Voevodsky-Morel homotopy $t$-structure, whereas $w$ is called (a version of) the [*Gersten*]{} weight structure; the corresponding weight filtrations and weight spectral sequences generalize coniveau ones.
\[rsheaves\]
1\. Certainly, these methods can be applied for ${\underline{C}'}$ being some (other) triangulated category “constructed from sheaves”; one can use Proposition 8.16 (and other results) of [@jardloc] to present ${\underline{C}'}$ as the homotopy category of a proper stable model category. Note also that a category ${\underline{C}'}$ of this sort is “usually” compactly generated (still cf. [@neeshman] and Remark \[rwg\](2) above); in this case there exist plenty of possible ${\mathcal{P}}$.
On the other hand, the heart of $t$ is “quite rarely” of the form $\operatorname{\operatorname{AddFun}}({\underline{B}},{\underline{\operatorname{Ab}}})$ for these examples; this justifies the claim made in Remark \[rab5\](\[istalks\]). Note also that any stalk functor in this case should send a complex of sheaves into a certain “stalk” of the zeroth (co)homology sheaf of this complex; whence the name. Moreover, for the motivic examples considered in [@bgn] the stalk functors are (retracts of coproducts of) certain “twists” of “actual stalks” (cf. also [@bondegl Theorem 3.3.1] for a certain “relative” version of this observation). On the other hand, for other triples $({\underline{D}},t,{\underline{D}}')$ as in Theorem \[tab5\] the stalk functors may give a new (and non-trivial) object of study. In particular, we obtain stalk functors for arbitrary compactly generated $t$-structures on arbitrary motivic homotopy categories (and on their localizing subcategories); these may have quite non-trivial “geometrical meaning” (and do not require any resolution of singularities assumptions in contrast to loc. cit.).
2\. Part \[iab5\] of our corollary generalizes Corollary 4.9 of [@humavit] (where ${\underline{C}}$ was assumed to be [*algebraic*]{} and $t$ was assumed to be non-degenerate; cf. Corollary \[csymt\](3) and Remark \[rsymt\](\[irsymt2\]) above). Note that both of these conditions are rather restrictive if one studies motivic homotopy categories.
3\. Since all objects of ${\underline{C}'}$ are cocompact in ${\underline{C}}$, the class of cocompact objects of ${\underline{C}}$ is not essentially small (in contrast to that for ${\underline{E}}$).
4\. Recall that ${\underline{E}}$ can be presented as the (Verdier) localization of ${\underline{C}}$ by the subcategory ${{}^{\perp}}{\underline{E}}$. In [@prospect] the case ${\underline{C}'}=SH$ (the topological stable homotopy category) was considered, and the corresponding ${\underline{E}}$ was constructed as the homotopy category of a certain right Bousfield localization of ${\underline{C}}$. Moreover, an exact equivalence $F:{\underline{E}}\to {\underline{C}'}{{^{op}}}$ was constructed (in this case).
Furthermore, Remark 6.9 of ibid. appears to imply that the duality $\Phi^{{\underline{E}}}$ in this case is isomorphic to the bifunctor $SH(S^0,F(-)\otimes -)$ ($S^0={{\pmb{1}}}_{SH}$ is the sphere spectrum); note that it suffices to construct the restriction of this isomorphism to finite spectra (see Remark \[rab5\](\[irotimes\])). Next, if we take ${\mathcal{P}}=\{S^0[i]: i\ge 0\}$ then the corresponding $t$ is certainly the Postnikov $t$-structure for $SH$, whereas $w_{{\underline{E}}}$ is easily seen to be the opposite (see Proposition \[pbw\](\[idual\])) to the [*spherical*]{} weight structure on ${\underline{E}}{{^{op}}}\cong SH$ (see [@bws §4.6] and [@bkw §2.4]).
Moreover, the author hopes that applying Remark \[rab5\](\[irotimes\], \[krauseideals\]) in this context may shed some light on the seminal telescope conjecture.
5\. We conjecture that an isomorphism $F:{\underline{E}}\to {\underline{C}'}{{^{op}}}$ (in the setting of this section) exists for a wide range of stable monoidal model categories such that ${\underline{C}_0}$ is self-dual with respect to $\otimes$ (and it generates ${\underline{C}}'$ as its own localizing subcategory). Note that in the aforementioned particular case ${\underline{C}'}=SH$ the existence of $F$ may be deduced from Theorem 5.3 of [@schwemarg];[^68] cf. also [@bger §6.4] for a certain motivic observation related to our conjecture. Note also that in this case the conjecture stated in Remark \[rab5\](\[iremb\]) is easily seen to be fulfilled also.
6\. Part \[icharw\] of our corollary can easily be “axiomatized” somewhat similarly to Corollary \[cwt\]. However, the existence of a category ${\underline{C}}\supset {\underline{C}}'$ such that all objects of ${\underline{C}}'$ are cocompact in it seems to be rather “exotic”.
On localizations of coefficients and “splittings” for triangulated categories {#slocoeff}
-----------------------------------------------------------------------------
In this subsection we gather a few results related to localizations of coefficients and decomposing triangulated categories into direct summands for the purpose of applying these statements in [@bgn].
First we recall the “naive” method of localizing coefficients (for a triangulated category).
\[rcgws\] 1. Let $S\subset {\mathbb{P}}$ be a set of prime numbers; denote the ring ${{\mathbb{Z}}}[S{^{-1}}]$ by ${\Lambda}$.
Then for any triangulated ${\underline{C}}$ one can consider the category ${\underline{C}}\otimes {\Lambda}$ with the same object class and ${\underline{C}}\otimes {\Lambda}(M,N)={\underline{C}}(M,N)\otimes_{{\mathbb{Z}}}{\Lambda}$ for all $M,N\in {\operatorname{Obj}}{\underline{C}}$. The category ${\underline{C}}\otimes {\Lambda}$ has a natural structure of a triangulated category; see Proposition A.2.3 of [@kellyth]. Next, if $s$ is a torsion pair for ${\underline{C}}$ then the Karoubi-closures of the classes $({\mathcal{LO}},{\mathcal{RO}})$ in ${\underline{C}}\otimes {\Lambda}$ give a torsion pair for ${\underline{C}'}$ according to Proposition \[phop\](9).
2\. Certainly, for ${\underline{C}}$ being an $R$-linear category (see Remark \[rsatur\](4)) we can also localize (using any of the methods described in this section) by any multiplicative subset of $R$.
However, this method of “localizing coefficients” of a triangulated category does not seem to be “appropriate” if ${\underline{C}}$ is has coproducts. So we describe an alternative construction that “works fine” for compactly generated categories.
\[plocoeff\] Assume that ${\underline{C}}$ is compactly generated; denote its subcategory of compact objects by ${\underline{C}}^c$.
For a set $S$ as above denote by ${\underline{E}}$ the localizing subcategory of ${\underline{C}}$ generated by cones of $p{\operatorname{id}}_M$ for $M\in {\operatorname{Obj}}{\underline{C}},\ p\in S$; denote by ${\underline{D}}$ the full subcategory of ${\underline{C}}$ whose object class is ${\operatorname{Obj}}{\underline{E}}{{}^{\perp}}$.
I. Then the following statements are valid.
1. \[icge\] ${\underline{E}}$ is compactly generated by cones of $p{\operatorname{id}}_N$ for $N$ running through ${\operatorname{Obj}}{\underline{C}}^c$ (and $p\in S$).
2. \[icgd\] ${\operatorname{Obj}}{\underline{D}}$ equals the class of ${\Lambda}$-linear objects of ${\underline{C}}$, i.e., of those $M\in {\operatorname{Obj}}{\underline{C}}$ such that $p{\operatorname{id}}_M$ is an automorphism for any $p\in S$; so, it is closed with respect to ${\underline{C}}$-coproducts.
3. \[icgadj\] The embedding $i:{\underline{D}}\to {\underline{C}}$ possesses an exact left adjoint $l_S$ that gives an equivalence ${\underline{C}}/{\underline{E}}\to {\underline{D}}$.
4. \[icg3\] For any $M\in {\operatorname{Obj}}{\underline{C}}$ and $N\in {\operatorname{Obj}}{\underline{C}}^c$ we have ${\underline{D}}(l(N),l(M))\cong {\underline{C}}(N,M)\otimes_{{\mathbb{Z}}}{\Lambda}$; thus ${\underline{D}}$ contains ${\underline{C}}^c\otimes{\Lambda}$ as a full subcategory.
5. \[icgls\] $l_S$ respects coproducts and converts compact objects into ${\underline{D}}$-compact ones. Moreover, ${\underline{D}}$ is generated by $l_S({\operatorname{Obj}}{\underline{C}}^c)$ as its own localizing subcategory, and the class of compact objects of ${\underline{D}}$ equals ${\operatorname{Obj}}{\langle}l_S ({\operatorname{Obj}}{\underline{C}}^c{\rangle}){\rangle}={\operatorname{Obj}}\operatorname{\operatorname{Kar}}_{{\underline{D}}} ({\underline{C}}^c\otimes {\Lambda})$.
6. \[icgtr\] For any $M\in {\operatorname{Obj}}{\underline{C}}$ there exists a distinguished triangle $$\label{ecdec}
N\to M\to i\circ l_S(M)\to N[1]$$ for some $N\in {\operatorname{Obj}}{\underline{E}}$; this triangle is unique up to a canonical isomorphism.
7. \[icgfun\] Let $H:{\underline{C}}\to {\underline{\operatorname{Ab}}}$ be a cc-functor. Then for any $M\in {\operatorname{Obj}}{\underline{C}}$ we have $H( i\circ l_S(M))\cong H(M)\otimes_{{\mathbb{Z}}}{\Lambda}$.
8. \[icgdi\] Assume that ${\underline{C}}'$ is also a compactly generated (triangulated) category; define ${\underline{D}}'$, $i'$ and $l'_S$ as the ${\underline{C}}'$-versions of ${\underline{D}}$, $i$, and $l_S$, respectively. Then any functor $F:{\underline{C}}\to {\underline{C}}'$ that respects coproducts can be canonically completed to a diagram $$\label{ediaf}
\begin{CD}
{\underline{C}}@>{l_S}>>{\underline{D}}@>{i}>>{\underline{C}}\\
@VV{F}V@VV{G}V@VV{F}V \\
{\underline{C}}'@>{l'_S}>>{\underline{D}}' @>{i'}>>{\underline{C}}'
\end{CD}$$ where $G$ is a certain exact functor respecting coproducts.
II\. Assume in addition that ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ is a class of compact objects. Denote by $s=({\mathcal{LO}},{\mathcal{RO}})$ the torsion pair generated by ${\mathcal{P}}$ (whose existence is given by Theorem \[tclass\](\[iclass1\]); note that ${\mathcal{P}}$ is essentially small).
1\. Then the couple $s[S{^{-1}}]=(\operatorname{\operatorname{Kar}}_{{\underline{D}}}(l_S({\mathcal{LO}})),\operatorname{\operatorname{Kar}}_{{\underline{D}}}(l_S({\mathcal{RO}})))$ gives the torsion pair generated by $l_S({\mathcal{P}})$ in ${\underline{D}}$.
2\. If $s$ is weighty then $s[S{^{-1}}]$ also is and the functor $l_S$ is weight-exact with respect to the corresponding weight structures $w$ and $w[S{^{-1}}]$, respectively (and so $l_S({\underline{C}}_{w=0})\subset {\underline{D}}_{w[S{^{-1}}]=0}$).
3\. If $s$ is associated to a $t$-structure $t$ then $s[S{^{-1}}]$ also is. Moreover, both $l_S$ and $i$ are $t$-exact (with respect to the corresponding $t$-structures), and so ${{\underline{Ht}}}[S{^{-1}}]\subset {{\underline{Ht}}}$. Furthermore, ${\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}^{t\le 0}={\underline{D}}^{t[S{^{-1}}]\le 0}$ and ${\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}^{t\ge 0}={\underline{D}}^{t[S{^{-1}}]\ge 0}$.
4\. If $N$ is a ${\mathcal{P}}$-colimit of some $(N_j)$ (see Definition \[drelim\]) then $l_S(N)$ is a $l_S({\mathcal{P}})$-colimit of the corresponding $(l_S(N_j))$.
III\. Adopt the notation and conventions of Proposition \[porthop\]. Then for ${\underline{D}}'\subset {\underline{C}}'$ (resp. for ${\underline{D}}\subset {\underline{C}}$) being the corresponding subcategories of ${\Lambda}$-linear object the restriction of $\Phi(-,-)$ to ${\underline{D}}{{^{op}}}\times {\underline{D}}'$ is a nice duality $\Phi[S{^{-1}}]:{\underline{D}}{{^{op}}}\times {\underline{D}}'\to {\underline{\operatorname{Ab}}}$. Moreover, one can obtain it by “bi-extending” (see Remark \[rcudu\](\[ir1\])) the duality ${\underline{D}}_0(-,-)$, where ${\underline{D}}_0=l_S({\underline{C}_0})\subset {\underline{D}}'$ is isomorphic to the “naive” $S$-localization of ${\underline{C}_0}$ as described in Remark \[rcgws\](1). Furthermore, the corresponding $w[S{^{-1}}]$ and $t[S{^{-1}}]$ are $\Phi[S{^{-1}}]$-orthogonal.
I. This construction was described in detail in Appendix A.2 of [@kellyth]. So ibid. would yield our assertions \[icgd\]–\[icgls\] if we replace ${\underline{E}}$ in it by the subcategory ${\underline{E}}'$ defined by means of assertion \[icge\]. Now, we certainly have ${\underline{E}}'\subset {\underline{E}}$, and the converse inclusion follows from ibid. since $l_S$ is easily seen to kill all $\operatorname{\operatorname{Cone}}(p{\operatorname{id}}_M)$ (for $M\in {\operatorname{Obj}}{\underline{C}},\ p\in S$). Hence ${\underline{E}}={\underline{E}}'$ and we obtain assertions \[icge\]–\[icgadj\].
Assertion \[icgtr\] easily follows from Proposition \[pbouloc\](III.\[ibou1\]).
\[icgfun\]. First we note that $H({\operatorname{Obj}}{\underline{D}})$ consists ${\Lambda}$-linear objects of ${\underline{\operatorname{Ab}}}$ (i.e., of ${\Lambda}$-modules) immediately from assertion \[icgd\].
Next, the group $H(\operatorname{\operatorname{Cone}}(p{\operatorname{id}}_M))$ is certainly killed by the multiplication by $p^2$ (for any $M\in {\operatorname{Obj}}{\underline{C}},\ p\in S$). Since the subcategory ${\underline{A}}$ of ${\underline{\operatorname{Ab}}}$ consisting of groups all of whose elements are $S$-torsion (i.e., annihilated by multiplying by some products of elements of $S$) is a Serre subcategory closed with respect to coproducts, and $H$ is a cc functor, we obtain $H({\operatorname{Obj}}{\underline{D}})\subset {\operatorname{Obj}}{\underline{A}}$.
Now, for any $M\in {\operatorname{Obj}}{\underline{C}})$ we apply $H$ to the triangle (\[ecdec\]) to obtain an exact sequence $H(N)\to H(M)\to H(i\circ l_S(M))\to H(N[1])$ (for some $N\in {\operatorname{Obj}}{\underline{E}}$). Since ${\Lambda}$ is a flat ${{\mathbb{Z}}}$-module, we can tensor this sequence by ${\Lambda}$ to obtain the exact sequence $H(N)\otimes_{{{\mathbb{Z}}}}{\Lambda}\to H(M)\otimes_{{{\mathbb{Z}}}}{\Lambda}\to H(i\circ l_S(M)) \otimes_{{{\mathbb{Z}}}}{\Lambda}\to H(N[1]) \otimes_{{{\mathbb{Z}}}}{\Lambda}$. Since $H(N)$ and $H(N[1])$ belong to ${\operatorname{Obj}}{\underline{A}}$, they are annihilated by $-\otimes_{{{\mathbb{Z}}}}{\Lambda}$. Lastly, since $H(i\circ l_S(M))$ is ${\Lambda}$-linear, we obtain $H(M)\otimes_{{{\mathbb{Z}}}}{\Lambda}\cong H(i\circ l_S(M)) \otimes_{{{\mathbb{Z}}}}{\Lambda}\cong H(i\circ l_S(M))$.
\[icgdi\]. Certainly, $F$ maps ${\Lambda}$-linear objects of ${\underline{C}}$ into ${\Lambda}$-linear ones; hence we can define $G$ as the corresponding restriction of $F$. It remains to verify that the left hand square in (\[ediaf\]) is essentially commutative. Applying assertion I.\[icgtr\] we obtain the following: it suffices to check that for any $M\in {\operatorname{Obj}}{\underline{C}}$ the functor $F$ maps the distinguished triangle (\[ecdec\]) into the corresponding triangle for $F(M)$. Hence it remains to note that $F(N)$ belongs to the corresponding localizing subcategory ${\underline{E}}'$ of ${\underline{C}}$ since $F$ respects coproducts. II.1. For any $M\in {\operatorname{Obj}}{\underline{C}}$ the functor $H_{M,S}: N\mapsto {\underline{D}}(l_S(N),l_S(M))$ is a cp functor from ${\underline{C}}$ into ${\underline{\operatorname{Ab}}}$. Assertion \[icg3\] implies that for any $M\in {\mathcal{RO}}$ the functor $H_{M,S}$ kills ${\mathcal{P}}$; thus it kills ${\mathcal{LO}}$ also (by Theorem \[tclass\](\[iclass5\]). Hence $\operatorname{\operatorname{Kar}}_{{\underline{D}}}(l_S({\mathcal{LO}}))\perp_{{\underline{D}}} \operatorname{\operatorname{Kar}}_{{\underline{D}}}(l_S({\mathcal{RO}}))$. Next, objects of ${\underline{D}}$ certainly possess $s[S{^{-1}}]$-decompositions (since all objects come from ${\underline{C}}$ and one can apply $l_S$ to $s$-decompositions).
According to Proposition \[phop\](9), it remains to prove that $l_S({\mathcal{RO}})$ is Karoubi-closed in ${\underline{D}}$.[^69] This certainly reduces to $l_S({\mathcal{RO}})=l_S({\mathcal{P}})^{\perp_{{\underline{D}}}}$ since $l_S$ respects the compactness of objects (see assertion I.\[icgls\]). Now, $l_S({\mathcal{P}})^{\perp_{{\underline{D}}}}\subset {\mathcal{RO}}$ by assertion I.\[icg3\], and it remains to note that $l_S$ maps objects of ${\underline{D}}$ into isomorphic ones. 2. We certainly have ${\mathcal{LO}}[S{^{-1}}]\subset {\mathcal{LO}}[S{^{-1}}][1]$; thus $s[S{^{-1}}]$ is weighty indeed (see Remark \[rwhop\](1)). The remaining parts of the assertion follow immediately.
3\. Similarly to the previous assertion, Remark \[rtst1\](\[it1\]) implies that $s[S{^{-1}}]$ is associated to a $t$-structure. Thus $l$ is $t$-exact.
According to assertion I.\[icgd\], to check the $t$-exactness of $i$ we should prove that for any ${\Lambda}$-linear object $M$ of ${\underline{C}}$ its $t$-truncations are ${\Lambda}$-linear also. Now, for any $p\in S$ the functoriality of the $t$-decomposition triangle $L\to M\to R\to L[1]$ implies that ${\underline{C}}(L,L)$ contains a morphism $1/p{\operatorname{id}}_L$ inverse to $p{\operatorname{id}}_M$ and ${\underline{C}}(R,R)$ contains a morphism $1/p{\operatorname{id}}_R$ inverse to $p{\operatorname{id}}_R$. Thus $L$ and $R$ are ${\Lambda}$-linear indeed.
It certainly follows that ${{\underline{Ht}}}[S{^{-1}}]\subset {{\underline{Ht}}}$. Lastly, ${\underline{D}}^{t[S{^{-1}}]\le 0}\subset {\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}^{t\le 0}$ and ${\underline{D}}^{t[S{^{-1}}]\ge 0}\subset {\operatorname{Obj}}{\underline{D}}\cap {\underline{C}}^{t\ge 0}$. The converse implication is immediate from the $t$-exactness of $l_S$ (along with the fact that $l_S$ maps objects of ${\underline{E}}$ into isomorphic ones). 4. Immediate from assertion I.\[icg3\].
III\. $\Phi[S{^{-1}}]$ is a nice duality since it is a restriction of a nice duality. Moreover, assertion I.\[icg3\] gives the “description” of ${\underline{D}}_0$ in question.
Now, ${\underline{D}}$ has products, ${\underline{D}}'$ has coproducts, and the objects of ${\underline{D}}_0$ are compact in ${\underline{D}}'$ and cocompact in ${\underline{D}}$ according to assertions I.\[icgd\]–\[icgadj\] along with their duals. So we will prove that $\Phi[S{^{-1}}]$ is isomorphic to the corresponding biextension of ${\underline{D}}_0(-,-)$ using Proposition \[porthop\](\[ibiext\]). Since the embeddings ${\underline{D}}'\to {\underline{C}}'$ and ${\underline{D}}{{^{op}}}\to {\underline{C}}{{^{op}}}$ respect coproducts (see assertion I.\[icgd\]), we obtain that $\Phi[S{^{-1}}]$ respects ${\underline{D}}'$-coproducts as well as ${\underline{D}}{{^{op}}}$-ones. Next, assertion I.\[icg3\] (along with assertion I.\[icge\] and the duals of these two) implies that $\Phi[S{^{-1}}]$ annihilates both ${}^{\perp_{{\underline{D}}}}{{\underline{D}}_0}\times {\operatorname{Obj}}{\underline{D}}'$ and ${\operatorname{Obj}}{\underline{D}}\times {{\underline{D}}_0}^{\perp_{{\underline{D}}'}}$.
Hence Proposition \[porthop\](\[ibiext\]) implies the “moreover” part of the assertion. It remains to apply part \[ihoport\] of that proposition to obtain the orthogonality result in question.
\[rloc\] 1. Combining part II.4 of our proposition (and adopting its notation) with Proposition \[pkrause\](\[ikr3\]) we obtain the following: if ${\underline{C}_0}$ is a full triangulated subcategory of ${\underline{C}}$, ${\operatorname{Obj}}{\underline{C}_0}\subset {\mathcal{P}}$, and $H:{\underline{D}}\to {\underline{A}}$ is an exact functor coextended from $l_S({\underline{C}_0})$ then $H(l_S(N))\cong \operatorname{\varinjlim}H(l_S(N_j))$.
Since $i$ is a full embedding, we also obtain that $H'(i(l_S(N)))\cong \operatorname{\varinjlim}H'(i(l_S(N_j)))$ for any functor exact functor $H':{\underline{C}}\to {\underline{A}}$ coextended from ${\underline{C}_0}$.
2\. One can easily generalize part I.\[icgfun\] of our proposition to cc functors into arbitrary AB5 categories.
3\. Certainly, those results of of this subsection that concern compactly generated categories can easily be dualized.
Now we study certain “splittings” of triangulated categories.
\[psplit\] I. Assume that ${\underline{C}}$ is compactly generated, ${\underline{C}_0}$ is its subcategory of compact objects, and that decomposes into the direct sum of two triangulated subcategories ${\underline{C}_0}^1$ and ${\underline{C}_0}^2$. Denote by ${\underline{C}}^1$ the localizing subcategory of ${\underline{C}}$ generated by ${\underline{C}_0}^1$.
Then the following statements are valid.
1. \[idec\] ${\underline{C}}^1$ is a direct summand of the category ${\underline{C}}$; so, there exists an exact functor $l$ projecting ${\underline{C}}$ onto ${\underline{C}}^1$. Moreover, $l$ is essentially the only exact projector functor that respects coproducts and restricts to the projection ${\underline{C}_0}\to {\underline{C}_0}^1$. Furthermore, if $N\in {\operatorname{Obj}}{\underline{C}}$ is a $C$-colimit of some $(N_j)$ (for some $C\in {\operatorname{Obj}}{\underline{C}}$) then $l(N)$ is a $l(C)$-colimit of the corresponding $(l(N_j))$.
2. \[isplcg\] $l$ respects the compactness of objects, ${\underline{C}}^1$ is compactly generated, and ${\underline{C}_0}^1$ is its subcategory of compact objects.
3. \[isplitp\] Assume that $s=({\mathcal{LO}},{\mathcal{RO}})$ is a torsion pair generated by a class ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$. Then $l({\mathcal{LO}})={\mathcal{LO}}\cap {\operatorname{Obj}}{\underline{C}}^1$, $l({\mathcal{RO}})={\mathcal{RO}}\cap {\operatorname{Obj}}{\underline{C}}^1$, and the couple $l(s)=(l({\mathcal{LO}}),l({\mathcal{RO}}))$ is a torsion pair for ${\underline{C}}^1$. In particular, if $s$ is weighty then the functor $l$ is weight-exact with respect to the corresponding weight structures. Moreover, the pair $l(s)$ is smashing whenever $s$ is.
II\. Adopt the notation and conventions of Proposition \[porthop\]; assume in addition that ${\underline{C}_0}$ generates ${\underline{C}}'$ and cogenerates ${\underline{C}}$, and that ${\underline{C}_0}\cong {\underline{C}_0}^1\bigoplus {\underline{C}_0}^2$. Denote by ${\underline{C}}'^1$ (resp. ${\underline{C}}^1$) the subcategory of ${\underline{C}}'$ (resp. of ${\underline{C}}$) that is (co)generated by ${\underline{C}_0}^1$. Then the restriction $\Phi^1$ of $\Phi(-,-)$ to ${\underline{C}}^1{}{{^{op}}}\times{\underline{C}}'^1$ is a nice duality, and one can obtain it by “bi-extending” (see Remark \[rcudu\](\[ir1\])) the duality ${\underline{C}}^1_0(-,-)$. Moreover, the corresponding $w^1$ is $\Phi^1$-orthogonal to $t^1$.
I.\[idec\]. Since ${\operatorname{Obj}}{\mathcal{P}}^1\perp {\operatorname{Obj}}{\mathcal{P}}^2$ and vice versa, our compactness assumptions easily imply that the natural functor ${\underline{C}}^1\bigoplus {\underline{C}}^2\to {\underline{C}}$ is an equivalence, where ${\underline{C}}^2$ is the localizing subcategory of ${\underline{C}}$ generated by ${\mathcal{P}}^2$. So, we obtain the existence of $l$. Next, $l$ is essentially unique since it should be identical on ${\underline{C}}^1$ and should kill ${\operatorname{Obj}}{\underline{C}}^1{}^\perp={\operatorname{Obj}}{\underline{C}}^2$. The “furthermore” part of the assertion follows immediately. \[isplcg\]. Immediate from the previous assertion. \[isplitp\]. Obvious from the fact that $l$ is a projection functor. II. The proof is rather similar to that of Proposition \[plocoeff\](III).
So, $\Phi^1$ is a nice duality since it is a restriction of a nice duality.
Assertion I.\[idec\] easily implies that $\Phi^1$ is isomorphic to the corresponding biextension of ${\underline{C}}_0^1(-,-)$. ${\underline{D}}_0(-,-)$ using Proposition \[porthop\](\[ibiext\]). It remains to apply part \[ihoport\] of that proposition to obtain that $w_1\perp_{\Phi^1}t_1$.
[1]{} Adámek J., Rosicky J., Locally presentable and accessible categories, London Math. Soc. Lect. Notes, vol. 189. Cambridge University Press, 1994.
Aihara T., Iyama O., Silting mutation in triangulated categories// J. Lond. Math. Soc. (2) 85(3), 2012, 633–668.
Alonso L., Jeremías A., Souto M.J., Construction of $t$-structures and equivalences of derived categories// Trans. of the AMS, 355(6), 2003, 2523–2543.
André Y., Kahn B., O’Sullivan P., Nilpotence, radicaux et structures monoïdales //Rendiconti del Seminario Matematico della Universitá di Padova, vol. 108 (2002), 107–291.
Auslander M., Coherent functors, in: Proc. conf. Categorical Algebra (La Jolla, 1965), Springer, Berlin, 1966, 189–231.
Bachmann T., On the invertibility of motives of affine quadrics// Doc. Math. 22 (2017), 363–395.
Barnea I., Schlank T., A new model for pro-categories// J. Pure Appl. Algebra 219(4), 2015, 1175–1210.
Barr M., Absolute homology// Theory and Applications of Categories vol. 14 (2005), 53–59.
Beilinson A., Bernstein J., Deligne P., Faisceaux pervers, Asterisque 100 (1982), 5–171.
Beligiannis A., Reiten I., Homological and homotopical aspects of torsion theories, Mem. Amer. Math. Soc. 188(883), 2007, viii+207.
Bökstedt M., Neeman A., Homotopy limits in triangulated categories// Comp. Math. 86 (1993), 209–234.
Bondal A.I., Kapranov M.M., Representable functors, Serre functors, and mutations// Izvestiya Rossiiskoi Akademii Nauk. Seriya Matematicheskaya, 53(6), 1989, 1183–1205, transl. in Izvestiya: Mathematics 35.3 (1990): 519–541.
Bondal A.I., Van den Bergh M., Generators and representability of functors in commutative and noncommutative geometry// Mosc. Math. J. 3(1), 2003, 1–36.
Bondarko M.V., Weight structures vs. $t$-structures; weight filtrations, spectral sequences, and complexes (for motives and in general)// J. of K-theory, v. 6(3), 2010, 387–504, see also <http://arxiv.org/abs/0704.4003>
Bondarko M.V., Motivically functorial coniveau spectral sequences; direct summands of cohomology of function fields// Doc. Math., extra volume: Andrei Suslin’s Sixtieth Birthday, 2010, 33–117; see also <http://arxiv.org/abs/0812.2672>
Bondarko M.V., $\mathbb{Z}[\frac{1}{p}]$-motivic resolution of singularities// Comp. Math., vol. 147(5), 2011, 1434–1446.
Bondarko M.V., Weights for relative motives: relation with mixed complexes of sheaves// Int. Math. Res. Notes, vol. 2014(17), 2014, 4715–4767.
Bondarko M.V., On morphisms killing weights, weight complexes, and Eilenberg-Maclane (co)homology of spectra, preprint, 2015, <http://arxiv.org/abs/1509.08453>
Bondarko M.V., On infinite effectivity of motivic spectra and the vanishing of their motives, preprint, 2016, <https://arxiv.org/abs/1602.04477>
Bondarko M.V., Gersten weight structures for motivic homotopy categories; retracts of cohomology of function fields, motivic dimensions, and coniveau spectral sequences, preprint, 2018, <https://arxiv.org/abs/1803.01432>
Bondarko M.V., Deglise F., Dimensional homotopy $t$-structures in motivic homotopy theory// Adv. in Math., vol. 311 (2017), 91–189. Bondarko M.V., Luzgarev A.Ju., On relative $K$-motives, weights for them, and negative $K$-groups, preprint, 2016, <http://arxiv.org/abs/1605.08435>
Bondarko M.V., Sosnilo V.A., Detecting the $c$-effectivity of motives, their weights, and dimension via Chow-weight (co)homology: a “mixed motivic decomposition of the diagonal”, preprint, 2014, <http://arxiv.org/abs/1411.6354>
Bondarko M.V., Sosnilo V.A., A Nullstellensatz for triangulated categories// Algebra i Analiz, v. 27(6), 2015, 41–56. Bondarko M.V., Sosnilo V.A., Non-commutative localizations of additive categories and weight structures; applications to birational motives// J. Math. Jussie, published online: 24 May 2016, DOI: <http://dx.doi.org/10.1017/S1474748016000207>
Bondarko M.V., Sosnilo V.A., On purely generated ${\alpha}$-smashing weight structures and weight-exact localizations, preprint, 2017, <http://arxiv.org/abs/1712.00850>
Bondarko M.V., Sosnilo V.A., On constructing weight structures and extending them to idempotent extensions, Homology, Homotopy and Appl., vol 20(1), 2018, 37–57.
Bondarko M.V., Tabuada G., Picard groups, weight structures, and (noncommutative) mixed motives// Doc. Math. 22 (2017), 45–66. Chorny B., A generalization of Quillen’s small object argument// J. Pure and Appl. Alg., vol. 204 (2006), 568–583.
Christensen J., Ideals in triangulated categories: phantoms, ghosts and skeleta// Adv. in Math. 136.2 (1998), 284–339.
Christensen J.D., Isaksen D. C., Duality and pro-spectra// Algebraic and Geometric Topology 4.2 (2004), 781–812.
Fausk H., Isaksen D., t-model structures// Homology, Homotopy and Appl. 9(1), 2007, 399–438.
Gillet H., Soulé C., Descent, motives and $K$-theory// J. f. die reine und ang. Math. v. 478 (1996), 127–176.
Hoshino M., Kato Y., Miyachi J.-I., On t-structures and torsion theories induced by compact objects// J. of Pure and Appl. Algebra, vol. 167(1), 2002, 15–35.
Hovey M., Model categories, Mathematical Surveys and Monographs, vol. 63, AMS, 2003.
Hügel L. A., Marks F., Vitória J., Torsion pairs in silting theory // Pacific Journal of Math. 291.2 (2017), 257–278. Isaksen D., Calculating limits and colimits in pro-categories// Fundamenta Math. 2.175 (2002), 175–194.
Jardine R., Local homotopy theory, Springer, 2015. Keller B., A remark on the generalized smashing conjecture// Manuscripta Math. 84.1 (1994), 193–198.
Keller B., Nicolas P., Weight structures and simple dg modules for positive dg algebras //Int. Math. Res. Not. vol. 2013(5), 2013, 1028–1078.
Kelly S., Triangulated categories of motives in positive characteristic, Ph.D. thesis of Université Paris 13 and of the Australian National University, 2012, <http://arxiv.org/abs/1305.5349>
Kelly S., Saito S., Weight homology of motives// Int. Math. Res. Notices, v. 2017(13), 2017, 3938–3984. Krause H., Smashing subcategories and the telescope conjecture — an algebraic approach // Invent. math. 139 (2000), 99–133.
Krause H., On Neeman’s well generated triangulated categories// Doc. Math. 6 (2001), 121–125.
Krause H., A Brown representability theorem via coherent functors// Topology 41(4), 2002, 853–861.
Krause H., Deriving Auslander’s formula // Doc. Math. 20 (2015), 669–688.
Modoi G.C., On perfectly generating projective classes in triangulated categories// Comm. Alg. 38(3), 2010, 995–1011.
Neeman A., On a theorem of Brown and Adams// Topology 36 (1997), 619–645.
Neeman A., Triangulated Categories. Annals of Mathematics Studies 148 (2001), Princeton University Press, viii+449 pp.
Neeman A., On the derived category of sheaves on a manifold// Doc. Math. 6 (2001), 483–488.
Neeman A., Brown representability follows from Rosický’s theorem// J. of Topology, 2.2 (2009), 262–276.
Neeman A., Triangulated categories with a single compact generator and a Brown representability theorem, preprint, 2018, <https://arxiv.org/abs/1804.02240>
Nicolas P., Saorin M., Parametrizing recollement data for triangulated categories// J. of Algebra 322 (2009), 1220–1250.
Nicolas P., Saorin M., Zvonareva A., Silting theory in triangulated categories with coproducts, preprint, 2015, <http://arxiv.org/abs/1512.04700>
Parra C.E., Saorin M., Direct limits in the heart of a $t$-structure: the case of a torsion pair// J. of Pure and Applied Algebra 219.9 (2015), 4117–4143.
Pauksztello D., A note on compactly generated co-t-structures// Comm. in Algebra, vol. vol. 40(2), 2012, 386–394.
Pospisil D., Šťovíček J., On compactly generated torsion pairs and the classification of co-t-structures for commutative Noetherian rings// Trans. Amer. Math. Soc. 368 (2016), 6325–6361.
Rosicky J., Generalized Brown representability in homotopy categories// Theory and applications of categories, vol. 14(19), 2005, 451–479.
Rouquier R., Dimensions of triangulated categories //J. of K-theory 1.02 (2008), 193–256.
Schwede S., A uniqueness theorem for stable homotopy theory //Math. Zeitschrift 239.4 (2002), 803–828.
Schwede S., Shipley B., Stable model categories are categories of modules// Topology 42(1), 2003, 103–153.
Souto M. J., On the cogeneration of t-structures// Archiv der Mathematik 83.2, 2004, 113–122.
Wildeshaus J., Weights and conservativity, preprint, 2015, <http://arxiv.org/abs/1509.03532>
Wildeshaus J., Intermediate extension of Chow motives of Abelian type //Adv. in Math., vol. 305 (2017), 515–600.
[^1]: The author’s work on sections 1–2 was supported by the RFBR grant No. 15-01-03034-a and by Dmitry Zimin’s Foundation “Dynasty”, whereas the work on sections 3–5 was supported by the Russian Science Foundation grant no. 16-11-10073.
[^2]: The relation of pure functors to Deligne’s purity of (singular and étale) cohomology is recalled in Remark \[rwrange\](5).
[^3]: In [@bws] $t$ was said to be left adjacent to $w$ in this case; we discuss this distinction in “conventions” in §\[rwsts\] below.
[^4]: Thanks to the foundational results of A. Neeman and others, this property is known to hold for several important classes of triangulated categories; in particular, it suffices to assume that either ${\underline{C}}$ or ${\underline{C}}{{^{op}}}$ is compactly generated.
[^5]: Here ${{\underline{Hw}}}$ is the heart of $w$; note also that $G: {{\underline{Hw}}}{{^{op}}}\to {\underline{\operatorname{Ab}}}$ respects products whenever it converts ${{\underline{Hw}}}$-coproducts into products of groups.
[^6]: The author suspects that the content of the paper will not be evenly interesting to the readers. So he suggests the readers not (much) interested in “large” categories to ignore all matters related to infinite coproducts on the first reading (this includes compact objects and smashing torsion pairs). On the other hand, §\[sswc\] (and so, weight complexes and Postnikov towers) are mentioned explicitly in §\[sws\] only. So, the reader is encouraged to look for his personal trajectory through this paper.
[^7]: $t$-structures of these type appear to be originally introduced in [@talosa]. They have become a popular object of study recently, with plenty of examples important to various areas of mathematics.
[^8]: Recall that $P\in {\operatorname{Obj}}{\underline{C}}$ is said to be compact if the corepresentable functor ${\underline{C}}(P,-)$ respects coproducts. Now, any set of compact objects generates a $t$-structure according to Theorem A.1 of [@talosa]; the corresponding class ${\underline{C}}^{t\le 0}$ is the smallest subclass of ${\operatorname{Obj}}{\underline{C}}$ that is closed with respect to $[1]$, extensions, and coproducts, and contains ${\mathcal{P}}$.
[^9]: Recall that Theorem 3.7 of loc. cit. says that countable colimits in the heart ${{\underline{Ht}}}$ are exact for any compactly generated $t$-structure $t$.
[^10]: More generally, perfect classes are closely related to smashing torsion pairs; see Proposition \[psym\](\[iperftp\]).
[^11]: See Remark \[requivdef\] below for a comparison of this definition with other ones in the literature.
[^12]: Recall that this is always the case if the class ${\mathcal{P}}$ [*Hom-generates*]{} ${\underline{C}}$, i.e., $\cap_{i\in {{\mathbb{Z}}}}({\mathcal{P}}[i]{{}^{\perp}})={\{0\}}$. Since all elements of $\cap_i({\mathcal{P}}[i]{{}^{\perp}})$ are [*left degenerate*]{} with respect to $w$, if the Brown representability condition is not fulfilled for ${\underline{C}}$ then $w$ is necessarily “somewhat pathological”.
[^13]: Note here that our proof of Theorem \[textw\] (that is somewhat similar to the corresponding proof from [@kraucoh]) does not allow constructing $t$-structures “directly from perfect sets” since it relies crucially on ${\underline{C}}_{w\le 0}[-1]\subset {\underline{C}}_{w\le 0}$ (and does not work for general torsion pairs that will be discussed soon; yet cf. Remark \[rigid\](1)). Still we are able to prove the existence of $t$-structure generated by ${\mathcal{P}}$ whenever ${\mathcal{P}}$ is [*symmetric*]{} to some ${\mathcal{P}}'\subset {\operatorname{Obj}}{\underline{C}}$ (see Theorem \[tsymt\]; note that any such ${\mathcal{P}}$ is certainly perfect).
[^14]: Note that in [@postov] torsion pairs were called complete Hom-orthogonal pairs.
[^15]: Certainly, if $C$ is triangulated then $X$ is a retract of $Y$ if and only if $X$ is its direct summand.
[^16]: These are axioms \[TR5\] and \[TR5\*\] of [@neebook], respectively.
[^17]: Another reason of passing to this more general notion is that some of our results are valid for arbitrary torsion pairs.
[^18]: In the current paper we are not interested in triangulated categories that only have countable coproducts; yet in [@bsnew] we demonstrate this (weaker) assumption is rather useful also.
[^19]: It suffices to assume that ${\mathcal{P}}\subset {\operatorname{Obj}}{\underline{C}}$ and ${\mathcal{P}}^\perp={\mathcal{RO}}$ since then ${\mathcal{P}}$ certainly lies in ${\mathcal{LO}}$.
[^20]: Note that the class of ${\mathcal{P}}$-null morphisms is not necessarily shift-stable in contrast to the main examples of the paper [@christ] where this notion was introduced.
[^21]: Another way to deal with this discrepancy is to modify the definitions of $t$-structures and weight structures by the corresponding shifts; see Definition 2.1 of [@humavit].
[^22]: So, we don’t have to “shift” ${\mathcal{RO}}$ as we did in Remark \[rtst1\](\[it1\]).
[^23]: The class $P_t$ was called the [*coheart*]{} of $t$ in §3 of [@zvon].
[^24]: In [@bws] the axioms of a weight structure also required ${\underline{C}}_{w\le 0}$ and ${\underline{C}}_{w\ge 0}$ to be additive. Yet this is not necessary; see Remark 1.2.3(4) of [@bonspkar].
[^25]: Recall also that D. Pauksztello has introduced weight structures independently (see [@paucomp]); he called them co-t-structures.
[^26]: This relation was earlier introduced in [@barrabs]; $m_1$ is [*absolutely homologous*]{} to $m_2$ in the terminology of that paper. Respectively, some of the results below concerning this equivalence relation were proved in ibid.
[^27]: The term comes from [@gs]; yet the domain of the weight complex functor in that paper was not triangulated, whereas the target was (“the ordinary”) $K^b({\operatorname{Chow^{eff}}})$.
[^28]: Note however that the bounded cases of Proposition \[pdetect\] and Proposition \[pdetectsse\] can also be easily deduced from Theorem 3.3.1(IV) of [@bws].
[^29]: Alternatively, one can apply Proposition \[pwrange\](\[iwrpure\]) below to obtain this duality assertion.
[^30]: This result was extended to the case where $k$ is a perfect field of characteristic $p>0$ in [@bzp]; note however that one is forced to invert $p$ in the coefficient ring in this setting.
[^31]: Certainly, singular (co)homology (of motives) is only defined if $k$ is a subfield of complex numbers; then it is endowed with Deligne’s weight filtration that can also be computed using $w_{{\operatorname{Chow}}}(k)$ (see Remark 2.4.3 of [@bws]). On the other hand, Deligne’s weight filtration for étale (co)homology can be defined (at least) for $k$ being any finitely generated field; the comparison of the corresponding weight factors with the ones computed in terms of $w_{{\operatorname{Chow}}(k)}$ is carried over in Proposition 4.3.1 of [@bkl]. Note also that in ibid. and in [@brelmot §3.4,3.6] certain “relative perverse” versions of these weight calculations were discussed.
[^32]: Certainly, the AB4\* condition for the target category ${\underline{A}}'$ can also be weakened respectively.
[^33]: Certainly, in this case ${\underline{C}}$ is also generated by the coproduct of these objects (as its own localizing subcategory).
[^34]: Actually, this envelope also equals the extension-closure of $\cup_{i<0} P_t[i]$; see Proposition \[pbw\](\[igenlm\]).
[^35]: This is a rather “natural” additional assumption for the corollary (as well as for Theorem \[tadjw\]) since otherwise the corresponding weight structure $w$ would be “rather degenerate” in the following sense: the class $\cap_i{\underline{C}}_{w\ge i}$ would contain the non-zero class $ {}^{\perp_{{\underline{C}}}} {\underline{C}'}$
[^36]: Recall that ${\alpha}$ is said to be regular if it cannot be presented as a sum of less then ${\alpha}$ cardinals that are less than ${\alpha}$.
[^37]: In particular, the localization ${\underline{C}}/{\underline{D}}$ is a category, i.e., its Hom-classes are sets.
[^38]: Recall that for every combinatorial stable Quillen model category $K$ its homotopy category is well generated; see [@rosibr Proposition 6.10].
[^39]: This is why we use the word “symmetric”.
[^40]: Alternatively, Remark \[requivdef\] below allows to deduce this fact from [@kraucoh Theorem B].
[^41]: Recall that any set ${\mathcal{P}}$ of compact objects generates a torsion pair according to Theorem 4.1 of [@aiya]; cf. Theorem \[tclass\](\[iclass1\]) below.
[^42]: One can also “unite” symmetric classes (see Proposition \[psymb\](I.\[iws1\])); yet this does not give “really new” essentially small symmetric classes (if ${\underline{C}}$ has coproducts).
[^43]: One has to assume in addition that ${\underline{C}}$ satisfies the Brown representability condition; however this is “almost automatic”.
[^44]: The terminology we introduce is new; yet big hulls were essentially considered in (Theorem 3.7 of) [@postov].
[^45]: Note however that our reasoning is somewhat more clumsy than that of Pospisil and Šťovíček since we cannot apply their Proposition 2.7 to a general ${\underline{C}}$. Moreover, the proof of Theorem \[tpgws\] below can also be simplified if we assume (following loc. cit.) that ${\underline{C}}$ is a “stable derivator” triangulated category.
[^46]: Actually, ${\underline{C}}^{{{\aleph_0}}}$ is essentially small itself in “reasonable” cases; in this case the essential smallness of its classes of objects is automatic.
[^47]: This statement was previously proved in [@postov] and our argument is just slightly different from the one of Pospisil and Šťovíček; see Lemma 3.9 of ibid.
[^48]: Actually our argument also yields that $H^{{\mathcal{P}}}(L)\cong \operatorname{\varinjlim}H^{{\mathcal{P}}}(L_k)$ easily.
[^49]: This is where we need ${\mathcal{P}}$ to be cosuspended!
[^50]: An argument even more closely related to our one was used in the proof of [@modoi Lemma 2.2]; yet the assumptions of that lemma appear to require a correction.
[^51]: Certainly, the big hull of ${\mathcal{P}}$ is contained in its big extension-closure, whereas the latter (for ${\mathcal{P}}\subset {\mathcal{P}}[1]$) equals the smallest coproductive extension-closed subclass of ${\operatorname{Obj}}{\underline{C}}$ containing ${\mathcal{P}}$; cf. Corollary \[cgdb\](\[ict\]) below.
[^52]: Note however that [*weak weight structures*]{} (one replaces the orthogonality axiom in Definition \[dwstr\] by ${\underline{C}}_{w\le 0}\perp{\underline{C}}_{w\ge 2}$) were essentially considered in [@bsosn] (cf. Remark 2.1.2 of ibid.) and in Theorem 3.1.3(2,3) of [@bsnew], whereas in Proposition 3.17 of [@brelmot] it was shown that they are relevant for the study of mixed étale ${{\mathbb{Q}_l}}$-adic sheaves over varieties over finite fields (actually, it was demonstrated that the weight filtration for the category ${D^b_m}(X_0,{{\mathbb{Q}_l}})$ satisfies the somewhat stronger Definition 3.11 of ibid., where $X_0$ is a variety over a finite field of characteristic $\neq l$).
[^53]: These definitions (along with the definition of ${\mathcal{P}}$-null morphisms) is taken from [@christ].
[^54]: Actually, the standard convention is to say that ${\underline{\coprod}{\mathcal{P}}}$ is contravariantly finite if this condition is fulfilled; yet our version of this term is somewhat more convenient for the purposes of the current paper.
[^55]: Certainly, $t$ is generated by the set $\cup_{i\ge 0}{\mathcal{Q}}[i]$. Hence Theorem \[tclass\](\[iclass1\],\[iclasst\]) gives a “more precise” description of the class ${\underline{C}}^{t\le 0}$.
[^56]: This relation of ${{\underline{Ht}}}$ with its subcategory ${\operatorname{Inj}{{\underline{Ht}}}}$ of injective objects does not mention weight structures; yet it appears to follow from the existence of an injective cogenerator along with the AB4 property.
[^57]: Recall here that ${\underline{C}}{{^{op}}}$ has coproducts according Proposition \[pcomp\](II.2).
[^58]: Note that in Definition 2.1 of ibid. $t$-structures and co-$t$-structures (i.e., weight structures) were defined as the corresponding types of torsion pairs; so our definition differs from loc. cit. by shifts of the corresponding ${\mathcal{RO}}$ (cf. Remarks \[rtst1\](\[it1\]) and \[rwhop\](1)).
[^59]: Note also that the reasoning of Pospisil and Šťovíček in the proof of loc. cit. works for arbitrary torsion pairs; this is certainly not the case for our arguments.
[^60]: This fact can also be proved by noting that $H^{A_J}$ can be obtained from $J$ “by means” of the corresponding virtual $t$-truncations; see Theorem 2.4.2(II) of [@bger].
[^61]: One can also prove this statement using Remark \[rtkrau\](4); note that is actually not necessary to assume that ${\operatorname{Obj}}{\underline{C}}^{\perp_{{\underline{D}}}}={\{0\}}$ to get $t_{{\underline{D}}}$. Note also that Theorem 1.3 of [@hoshimi] is essentially an important particular case of [@bws Theorem 4.5.2(I.1)].
[^62]: Moreover, it would be interesting to prove some version of part \[ihoport\] of the proposition without assuming that objects of ${\underline{C}_0}$ are compact in ${\underline{C}'}$.
[^63]: Recall that ${\underline{D}}^{{{\aleph_0}}}$ is the (essentially small) class of compact objects of ${\underline{D}}$. Following Remark 4.2.11 of [@bondegl], one may call elements of $H_0^t({\underline{D}}^{{{\aleph_0}}})$ [*strongly constructible*]{} objects of ${{\underline{Ht}}}$.
[^64]: Note also that our argument has inspired the proof of [@bondegl Corollary 4.2.4].
[^65]: Possibly, a somewhat more general statement of this sort may be obtained by using stable $\infty$-categories; see Denis Nardin’s answer at <http://mathoverflow.net/q/255440>; yet the author does not know much about these matters. Respectively, the author is not sure that the construction of ${\underline{D}}'$ in §\[sprospectra\] is “optimal” for the purposes of Theorem \[tab5\]; yet this argument enables certain “computations” that are important for [@bgn].
[^66]: The author wonders whether some version of our argument can work for arbitrary ${\underline{D}}$. For this purpose it seems necessary to “get rid of ${\underline{D}}'$” in the proof. One may try to construct certain version of the stalk functors $\Phi(P,-)$ “directly”. It appears that a minor modification of the reasoning used in the proof of [@bsnull Proposition 2.1] gives the result whenever ${{\underline{D}}_0}$ is [*countable*]{} (i.e., ${\operatorname{Mor}}{{\underline{D}}_0}$ is a countable set). It is not clear whether the result can be extended to the general case (possibly, using the arguments of §3.1 of ibid.).
[^67]: Alternatively, one can combine the argument dual to the one in the proof of Theorem 7.4.3 of [@hovey] with the fact that $c(f)$ for $f$ running through all fibrations in ${\mathcal{M}}$ yield a set of generating fibrations for ${\operatorname{Pro}-\mathcal{M}}$ (see Theorem 6.1 of [@chor]).
[^68]: For this purpose one should use the fact that ${\underline{E}}$ has a model; however the Margolis’ uniqueness conjecture (see [@schwemarg §3]) predicts (in particular) that this conditions is fulfilled automatically.
[^69]: Note that $l_S({\mathcal{LO}})$ does not necessarily have this property.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A notion of the Berry phase is a powerful means to unravel the non-trivial role of topology in various novel phenomena observed in chiral magnetic materials and structures. A celebrated example is the intrinsic anomalous Hall effect (AHE) driven by the non-vanishing Berry phase in the momentum space. As the AHE is highly dependent on details of the band structure near the Fermi edge, the Berry phase and AHE can be altered in thin films whose chemical potential is tunable by dimensionality and disorder. Here, we demonstrate that in ultrathin SrRuO$_3$ films the Berry phase can be effectively manipulated by the effects of disorder on the intrinsic Berry phase contribution to the AHE, which is corroborated by our numerically exact calculations. In addition, our findings provide ample experimental evidence for the superficial nature of the topological Hall effect attribution to the protected spin texture and instead lend strong support to the multi-channel AHE scenario in ultrathin SrRuO$_3$.'
author:
- Liang Wu
- Fangdi Wen
- Yixing Fu
- 'Justin H. Wilson'
- Xiaoran Liu
- Yujun Zhang
- 'Denis M. Vasiukov'
- 'Mikhail S. Kareev'
- 'J. H. Pixley'
- Jak Chakhalian
bibliography:
- 'reference.bib'
title: 'Berry phase manipulation in ultrathin SrRuO$_3$ films'
---
During the last decade, a search for topologically non-trivial modalities in both real and momentum space has become a dominant driver in condensed matter physics [@RevModPhys.82.3045; @RevModPhys.83.1057; @RevModPhys.90.015001; @RMP1; @ref03; @RN485]. A metallic magnet entwined with a non-collinear spin texture like skyrmions, domain walls, or helical order can demonstrate interesting phenomena due to the interaction of conduction carriers with the localized spins [@ref03; @RN485; @RN490; @PhysRevB.88.220405]. Microscopically, such non-trivial magneto-transport response stems from the emergent electromagnetic fields (EEMFs) linked to the finite Berry phase accumulation. From the experimental standpoint, the challenge is to devise clear signatures of the EEMFs linked to the topologically protected spin texture with a non-zero winding number or skyrmions. Recently, it has been realized that such skyrmionic contribution can be revealed as the extra features to the transverse Hall resistivity $\rho_{xy}$ arising from a fictitious Lorentz force [@RN485; @RN486]. This contribution to the anomalous Hall effect (AHE) is collectively known as the topological Hall effect or THE. Because of the relative simplicity of the detection method, THE has quickly become a popular means to interrogate bulk, thin films and hetero-junctions of various materials for the presence of skyrmionic matter [@ref06; @PhysRevLett.102.186601; @RN488; @RN489; @PhysRevB.91.245115; @RN493; @PhysRevLett.110.117202; @ref09; @PhysRevLett.119.176809; @ref10; @ref08; @ref11; @ref12; @RN342].
The criterion for the observation of THE is the extra bumps/dips along with the AHE, however, which is only valid for *homogenous materials with a single conduction channel* [@ref13; @ref14; @arXiv1]. Moreover, the effects of disorder which are always present in real materials, can drastically alter the Berry phase and AHE, and yet are typically removed from the consideration. In this work, we interrogate the validity of the THE for a prototypical magnet SrRuO$_3$ (SRO) grown in the ultrathin form to amplify the effects of confinement, and magnetic and structural inhomogeneity on the Berry phase and AHE. As the Berry phase and AHE behavior in SRO are highly dependent on details of the band structure near the Fermi surface, we demonstrate how both entities can be effectively manipulated in the ultrathin limit of SRO. Using a phenomenological model that displays the THE-like Hall signal, we theoretically capture the experimentally observed tunability of the AHE with film-thickness, disorder, and temperature. More specifically, by including the effect of inhomogeneity to analyze the experimental data in the form of a two-channel AHE with opposite signs, we are able reproduce both the overall transverse Hall effect and the universal scaling behavior between AHE conductivity $ \sigma_{\mathrm{AHE}} $ and longitudinal conductivity $ \sigma_{xx} $ without resorting to THE.
In this work we focus on SrRuO$_3$ (SRO), an exemplary ferromagnetic metal with perovskite structure with a Curie temperature $T_c \approx 150 $K [@RN491; @PRB1; @ref25]. The investigation of the Hall effect in SRO not only played an important role in separating the intrinsic nature of AHE from extrinsic contributions [@ref15; @ref24; @PhysRevB.88.125110; @PhysRevB.70.180407], but also accelerated the research on THE and skyrmions [@ref08; @ref11; @ref12; @RN342]. However, the AHE in SRO also exhibits a complex non-monotonous dependence on thickness and temperature [@ref15; @ref08; @ref11; @ref23], which in turn demands scrutiny of the experimental results.
{width="16cm"}
{width="17cm"}
To investigate the temperature dependence of the AHE as a function of confinement, we have acquired a series of temperature (T)-dependent resistivity curves shown in Fig. \[Fig1\]a. In addition to the pristine samples we have developed a protocol to introduce controlled disorder by time-exposing the as-grown samples (labeled as series A samples of 4 u.c. and 5 u.c., which is short for 4A and 5A) to ambient conditions (labeled as series B samples, 4B and 5B). As seen, all films display a common feature, namely, a metallic state at high temperature and a small kink at around 100 K indicating the paramagnetic to ferromagnetic phase transition, which is lower than that of the bulk [@ref21; @ref08; @ref11]. In addition, a characteristic upturn appears below 40 K, which is due to Anderson weak localization in the ultrathin limit. Further, as immediately seen in Fig.s \[Fig1\]b and 1c for both 4B and 5B samples the R-T curves still retain the characteristic shape of the as-grown samples 4A and 5A, albeit with larger resistivity and steeper upturn at the low temperature characteristic of enhanced disorder. Interestingly, the magnetoresistance (MR) measurements, however, show only a negligible difference between the A and B series samples (see Fig. \[Fig1\]b, c) implying that disorder barely impacts ferromagnetism in SRO. With this understanding, we conduct T-dependent Hall measurements to investigate the thickness- and disorder-dependent AHE at different temperatures. Figures \[Fig1\]b and 1c show the representative low-temperature Hall resistance data. Following the convention, a magnitude of anomalous Hall resistance is extrapolated from the high-field linear part of the data, whose sign defines the sign of AHE, namely, the AHE in 4A (or 5A) is refereed to positive (or negative). This attribution is consistent with the previous reports [@ref08; @ref11; @ref23]. A direct inspection of Fig. \[Fig1\]b and \[Fig1\]c, however, reveals an unexpected result that the Hall data undergo a strong change for the B-series samples.
Next we apply the data analysis which is conventionally used to extract THE and separate various contributions to the Hall effect in SRO. Under the assumption of an idealized sample and in the presence of skyrmions, the Hall resistance can be decomposed as $ R_{xy}=R_{\mathrm{OHE}}+R_{\mathrm{AHE}}+R_{\mathrm{THE}}$ [@ref08; @ref11; @ref23], where the $R_{\mathrm{OHE}}$ stands for ordinary Hall effect (OHE). Alternatively, the THE-like Hall resistance can also be decomposed as $ R_{xy}=R_{\mathrm{OHE}}+R^+_{\mathrm{AHE}}+R^-_{\mathrm{AHE}} $, where the last two terms denote a positive and negative sign AHEs of the two-channel AHE [@ref13; @arXiv1]. In the following discussions, all OHE has been subtracted by fitting the high field linear slope. Given that complex oxide perovskites in the film form often exhibit intrinsic propensity to defects and layer non-uniformity during the step-flow growth, we demonstrate that the minimal two-channel AHE without THE can successfully capture all the AHE features across the whole temperature range. First, we remind the reader that a thickness variation of at least 1 u.c. practically always exists in a real film since step edges on the surface of an SRO film cannot ideally replicate those of the substrate after the growth. This intrinsic thickness non-uniformity, though often ignored in thicker films, upon approaching the ultrathin limit can strongly alter the Hall transport. Experimentally, the thickness variation in SRO was recently observed by magnetic force microscopy [@ref11]. In addition, the two-step transitions found in magnetic hysteresis loops [@ref21; @ref22] and MR shown in Fig. \[Fig1\]b land strong support for the thickness variation of the nominal SRO thickness. As seen in Fig. \[Fig2\], remarkably, for the 4A film the sign of AHE is *opposite* to that of the 5A sample below the crossover temperature around 90K [@ref08; @ref11; @ref23]. Given the above discussed intrinsic variation of the thickness, we are able to successfully reproduce the total AHE of the nominal 5 u.c. sample as the sum of the positive (4 u.c.) and negative AHEs (5 u.c.) originating from the thickness variation throughout the SRO film, without the ’superficial’ THE feature. This result strongly suggests that the thickness-dependent AHE sign reversal and disorder are the most probable underlying physical mechanisms for the extra features of the AHE. Next, we quantify the individual AHE contributions within the two-channel model in 5 u.c. SRO (4 u.c. SRO data are given in Supplementary Fig. S2) to elucidate the role of disorder and dimensionality on the AHE as: $R^{\mathrm{tot}}_{\mathrm{AHE}}= R^\mathrm{I}_{\mathrm{AHE}}\tanh(\omega_\mathrm{I}(H-H^\mathrm{I}_\mathrm{c})) + R^{\mathrm{II}}_{\mathrm{AHE}}\tanh(\omega_{\mathrm{II}}(H-H^{\mathrm{II}}_\mathrm{c}))$, where $H_\mathrm{c}$ and $\omega$ denote coercive field and slope related parameter at $H_c$ for each channel. Figure \[Fig2\]a (top panel) shows the experimental AHE data and results of the fit. A direct inspection of Fig. \[Fig2\] confirm that the fitting curves are indeed in excellent agreement with the experimental data and require no additional THE-like contribution. Another backing for the proposed two-channel model is found in the T-dependence of the total and the two-channel AHE shown in Fig. \[Fig2\]. As seen, of the data reveals that at 90 K for the sample 5A the THE-like features completely disappear. For the disorder induced sample 5B, however, the total AHE remain positive through the whole temperature range. This comparison implies that the disorder indeed makes a strong positive contribution to the Berry phase part of the AHE. The physical mechanism of how disorder interacts with the Berry phase will be discussed in detail below.
![**Comparison of the scaling relation.** Blue filled and red open symbols are scaling relation derived in THE scenario and two-channel AHE scenario, respectively. The black and green dash circles indicate the large deviation from the universal scaling relation of BaTiO$_3$-capped 4 u.c. SRO [@ref11] and 5B, where large THE-like Hall signal appears simultaneously.[]{data-label="Fig3"}](Fig3.pdf){width="\columnwidth"}
Further, to verify the validity of the two-channel AHE model, we apply the scaling relation between the magnitude of anomalous Hall conductivity $| \sigma_{\mathrm{AHE}} |$ vs. longitudinal conductivity $\sigma_{xx}$. This universal scaling relation broadly subdivides all materials systems into three regimes [@PRL1; @ref01; @ref14; @RN191]. For ultrathin SRO, $\sigma_{xx}\leq$ $10^4 \ \Omega^{-1}\textnormal{cm}^{-1}$ implies that the material is a subject to disorder that smears the contribution from intrinsic Berry phase driven AHE, resulting in the scaling relation for $\sigma_{\mathrm{AHE}}\propto(\sigma_{xx})^{1.6}$.
Next, we map our data on to the universal scaling curve to verify the THE and two-channel AHE scenarios. Here, we note that in the THE scenario, the extra THE contribution to the Hall measurements vanish at a high magnetic field due to the spin texture alignment by a sufficiently large magnetic field [@ref08; @ref11; @ref23]. As a result, if SRO contains a real THE contribution, the data extracted from AHE in the high magnetic field regime still should follow the universal scaling relation. Figure \[Fig3\] shows the experimental results based on both THE and our two-channel AHE scenarios (see the Supplementary, section 3) along with the summary of previously published SRO data [@ref08; @ref11]. As immediately seen, the THE scenario sharply diverges from the universal scaling, whereas our scaling data derived from the two-channel scenario show excellent consistency with the universal scaling relation.
Finally, we discuss the physical mechanism of the observed thickness, disorder, and temperature dependence on the AHE sign reversal in ultrathin SRO films (Fig. \[Fig1\]) with numerically exact calculations on a phenomenological lattice model [@ref15]. In bulk three-dimensional models, it is known that temperature can induce a sign reversal in the AHE [@ref15]. Here, we demonstrate that in disordered thin film samples varying the thickness and or the strength of disorder can similarly induce a reversal of sign in the AHE.
The model we use is a spin-polarized tight-binding model with $d_{xz}$ and $d_{yz}$ orbitals (originating from the $t_{2g}$ orbitals) from Ref.[@ref15] that is known to host an AHE and apply it to disordered thin films. The model is on a square lattice with spin-orbit coupling, with a finite magnetization, and with orbitals that induce nearest and next-nearest neighbor hoppings (with strength $t_1$ and $t_2$ respectively) on each individual u.c. while each u.c. is tunnel-coupled (with strength $t_1$). We consider slab sizes of $L_x=L_y=L=233$ and vary the number of u.c. $L_z = 1 - 7$ with open boundary conditions in the $z$-direction and periodic in the $x$- and $y$-direction. While the features we see appear generic over different parameters and different types of disorder \[Supplementary Section 4\], we focus here on on-site potential disorder with strength $W$. Using the kernel polynomial method [@weisse2006KPM] we compute the full conductivity tensor [@garcia2015conductivity] and hence the resistivity. The exact model, details of the calculation, and a broader consideration of the parameters and finite size can be found in Supplementary Section 4.
{width="16cm"}
The numerical results are presented in Fig. \[fig:theory\] for the Hall resistivity $\rho_{xy}$. In the theory section, the only contribution for $\rho_{xy}$ is $\rho_{\mathrm{AHE}}$. At a particular disorder strength that we label $W^*$, the sign of the AHE changes. We precisely define this as the point at $E_F=0$ when the slope of $\rho_{xy}$ changes signs. Our results clearly demonstrate that the sign of $\rho_{xy}$ changes due to varying the strength of disorder (Fig. \[fig:theory\]a), the number of layers (Fig. \[fig:theory\]b), and the temperature (Fig. \[fig:theory\]c). In Fig. \[fig:theory\]d we fix the temperature and find $W^*$ as a function of number of u.c.. If we contrast the change in the AHE between disorder and temperature, we notice that increasing disorder is providing an enhancement of $\rho_{xy}$ near $E_F\approx 0.5t_1$ in Fig. \[fig:theory\]a whereas increasing temperature is tending to suppress this feature Fig. \[fig:theory\]c. In the experiment, thinner films have larger disorder, and thus the experiment is taking a cut across the $W^*$ boundary in Fig. \[fig:theory\]d. As we cannot disentangle these two effects, the theoretical calculations demonstrate that the experimental results exhibit a sign reversal in the AHE due to varying both the number of layers and the strength of disorder.
In this few u.c. model, the density of states (Supplementary Section 6) shows that each u.c. creates a pair of van Hove peaks, with an offset due to hopping in the $z$-direction. Using the van Hove peaks as a guide, the distance between neighboring peaks near zero energy, denoted as $\Delta E$, reduces with increased number of u.c. and Fig. \[fig:theory\]d shows that this trend is correlated with $W^*$. As the number of u.c. increases the sign reversal in $\rho_{xy}$ occurs when the disorder strength smears the two van Hove peaks closest to $E_F=0$ into one. This is a signal of the general phenomena: due to the nonuniform distribution of Berry curvature in the bands, the inclusion of terms which sample states with different Berry curvature (e.g. temperature, disorder, and thickness) can significantly alter the AHE, even changing its sign.
In conclusion, the presented results unveil the unexpected thickness-dependent sign reversal in the AHE signal at the ultrathin limit. Such extreme sensitivity stems from the effects of disorder of the intrinsic Berry phase contribution of the AHE as confirmed by numerically exact calculations with the kernel polynomial method on a model that hosts the AHE. In addition, our findings provide strong experimental evidence for the superficial nature of the THE attribution to topologically protected spin texture and instead lend strong support to the two-channel AHE in SRO. The proposed multi-channel magneto-transport framework can be readily extended to many other ultrathin chiral magnets with spin-order where disorder, and the effects of surface and interface are critically important.
**Acknowledgements**
We deeply acknowledge fruitful and insightful discussions with Weida Wu, Daniel I. Khomskii, and X. Renshaw Wang. This work was supported by the Gordon and Betty Moore Foundations EPiQS initiative through Grant No. GBMF4534. J.H.W. and J.H.P. acknowledge the Aspen Center for Physics where some of this work was completed, which is supported by National Science Foundation grant PHY-1607611.
**Author contributions**
J.C. and L.W. conceived the original idea. J.C., L.W., F.W., Y.Z., X.L., D.M.V and M.S.K. planned and designed the experiments. F.W., M.K. and L.W. fabricated the samples. L.W., F.W. and J.C. performed transport measurements and data analysis. Y.F., J.H.W. and J.H.P performed numerically exact calculations and the analysis of relevant data. All the authors discussed the results and commented on the manuscript.
**Competing interests**
The authors declare no competing interests.
**Methods**
**Thin film growth and sample treatment.** Ultrathin SRO films were grown on SrTiO$_3$ (001) substrates using pulsed laser deposition monitored by *in-situ* reflection high-energy electron diffraction (RHEED) (Supplementary Section 1). Prior to the film growth, SrTiO$_3$(001) substrates were treated to obtain TiO$_2$-terminated step-and-terrace surfaces \[1\]. A controlled sequence of time exposure to ambient air treatment was employed to invoke the disorder in the as-grown films.
**Transport measurements.** Transport measurements for the films were conducted using the d.c. transport option of a physical property measurement system (PPMS, Quantum Design). All the Hall transport data were anti-symmetrized as a function of the magnetic field.
**Numerical Calculations.** We use the kernel polynomial method to compute the conductivity tensor and the density of states \[2\]. We average over 100 disorder samples for all of the results presented here.
1\. M. Kareev, S. Prosandeev, J. Liu, C. Gan, A. Kareev, J. W. Freeland, M. Xiao, and J. Chakhalian. Atomic control and characterization of surface defect states of $\mathrm{TiO}_2$ terminated $\mathrm{SrTiO}_3$ single crystals. *Appl. Phys. Lett.*, 93(6):061909, 2008.
2\. A. Wei[ß]{}e, G. Wellein, A. Alvermann, and H. Fehske. The kernel polynomial method. *Rev. Mod. Phys.*, 78(1):275, 2006.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Oliver Korten [^1]'
title: 'On the Complexity of 2-Player Packing Games'
---
Abstract {#abstract .unnumbered}
========
[10pt]{}[10pt]{} We analyze the computational complexity of two 2-player games involving packing objects into a box. In the first game, players alternate drawing polycubes from a shared pile and placing them into an initially empty box in any available location; the first player who can’t place another piece loses. In the second game, there is a fixed sequence of polycubes, and on a player’s turn they drop the next piece in through the top of the box, after which it falls until it hits a previously placed piece (as in Tetris); the first player who can’t place the next piece loses. We prove that in both games, deciding the outcome under perfect play is PSPACE-complete.
Introduction {#Intro}
============
Many NP-complete problems take the form of 1-player packing puzzles: there are a set of “pieces” and a board, and the player’s goal is to pack all of the pieces into the board, subject to the restriction that all of them “fit”. In polyomino packing, proved hard in [@dem], the pieces are polyominoes and the board is a rectangle, and “fitting” in this context means being placed with no overlap. Tetris, first proved hard in [@tetris] and analyzed further in [@totaltetris], uses the same kinds of pieces, but in this puzzle the pieces must be dropped in from the top of a box and fall until they hit other pieces, and the order of the available pieces is a fixed part of the input. Since Robertson and Munro published their analysis of Instant Insanity [@insanity], it has been known that PSPACE-complete problems often take the form of 2-player generalizations of 1-player puzzles. Thus, a natural question to ask is the complexity of 2-player packing games, where players take turns placing pieces into an arrangement, and the first player who cannot extend the packing loses.
In this paper we analyze two 2-player games involving packing physical pieces into a box. In Section \[polypacking\], we introduce 2-Player Polycube Packing, a 2-player generalization of the known NP-hard polyomino packing problem. In 2-Player Polycube Packing, players take turns placing polycubes into an initially empty box, and the first player who can’t place another polycube loses. We prove that deciding the winner of this game under perfect play is PSPACE-complete.
In Section \[tetris\], we introduce 2-Player 3D $n$-tris, a 2-player generalization of Tetris. In this game there is a set sequence of polycubes, and players take turns placing the next polycube in the sequence into an initially empty box by continuous downward motion until it rests on top of a previously placed piece. The first player who can’t place a piece loses. Once again, we show that deciding a winner under optimal play in this game is PSPACE-complete. Finally, in Section \[open\], we conclude with some open problems.
For both games defined below, membership in PSPACE is justified by the fact that these games are perfect information and end after a polynomial number of moves [@sch]. In order to establish PSPACE-hardness of our games, we will reduce from a game called Node Kayles. In this game, the input is an undirected graph $G$. Players take turns marking vertices in $G$, with the restriction that marked vertices cannot be adjacent. The first player who cannot make a move loses. Deciding a winner in this game was shown to be PSPACE-complete in [@sch].
2-Player Polycube Packing {#polypacking}
=========================
2-Player Polycube Packing is defined as follows: the input is an $N \times M \times K$ box and a set of polycubes. Two players take turns placing any remaining polycube into any available grid-aligned location in the box using translations, rotations, and (optionally) reflections. The first player who can’t place any more polycubes loses.
It is PSPACE-complete to determine the winner of a Polycube Packing game from an initially empty $N \times M \times 3$ box.
We will prove PSPACE-hardness of Polycube Packing by reducing from the problem of deciding a winner in Node Kayles. The proof holds regardless of if we allow rotations only, or rotations and reflections. The proof also uses only simply-connected polycubes.
Overview of Reduction
---------------------
Given an instance of Node Kayles in the form of a graph $G$, we will construct a set of polycubes and a complex initial board state such that each remaining polycube can only be placed in exactly one spot on the board. Further, we will construct this initial state such that if one were to superimpose all of these remaining polycubes onto the board in their unique spots, the intersection graph of these polycubes would be isomorphic to $G$. In this way, each playable polycube plays the role of a vertex in Node Kayles, and playing a polycube (by putting it in its unique spot) removes from future play exactly those polycubes which represent the neighbors of its respective vertex in $G$. Next, it is shown that this complex initial state can actually be arrived at after the placement of a single polycube, which we call the $G$-mold. Finally, we prove that by adding the $G$-mold and a few other pieces into the set of playable polycubes, we can get this reduction to work even from an initially empty board state.
***Note:*** In every figure, upwards is the positive $y$-direction, right is the positive $x$-direction, and out of the page is the positive $z$-direction, unless otherwise specified by a 3-dimensional axis icon. In this icon, the arrows pointing straight up and straight right indicate dimensions oriented upwards and rightwards respectively, and the arrow bisecting these two orthogonal arrows indicates the dimension oriented into the page. An example appears in Figure \[fig:crossover\].
Hardness from a Complex Initial Board State
-------------------------------------------
### Wiring Diagram
Given a graph $G = (V, E)$, where $V = \{v_1, ..., v_n\}$ and $E = \{e_1, ..., e_m\}$, first we will assign to each $v_i$ a horizontal line segment with $x$-coordinate $4i$, spanning the $y$-interval $[0, 2m]$. Now, for each edge $e_i$ which connects $v_j$ to $v_k$, draw a horizontal line segment with endpoints $(4j, 2i)$ and $(4k, 2i)$. We will refer to this 2-dimensional layout of $G$ as its wiring diagram. Figure \[fig:wiring diagram\] shows the wiring diagram of $H = (V', E')$ where\
$V' = \{v_1, v_2, v_3, v_4, v_5, v_6, v_7\}$
$E' = \{v_1v_4, v_6v_7, v_2v_4, v_4v_6, v_3v_5, v_1v_3, v_1v_4, v_5v_7\}$
![Wiring diagram of H[]{data-label="fig:wiring diagram"}]("Wiring_Diagram".pdf)
### Cavity
We will now construct a polycubal structure which mimics the wiring diagram of $G$. The first step is to build a vertical $1 \times 1 \times k$ column for each vertex, where $k$ is the height of the vertical line corresponding to that vertex in the wiring diagram (dotted lines in Figure \[fig:wiring diagram\]). Now place these vertex columns with their left boundaries along those corresponding vertical lines in the plane.
Now, for each edge $e_j$, create a horizontal $1 \times 1 \times k$ column where $k$ is the length of the horizontal segment corresponding to $e_j$ in the wiring diagram, and place this at the same $y$ coordinate as the corresponding segment in the wiring diagram. Say this edge connects $v_i$ to $v_h$, $i < h$. So we will glue the ends of it to the vertical columns for $v_i$ and $v_h$ by laying the two vertical columns and the horizontal column out in the plane as described, and then taking the union of these three polycubes. We then make the following modification to this construction: for every $v_k$ where $i < k < h$, bend $e_j$ to cross over $v_k$ via the construction shown in Figure \[fig:crossover\].
![Edge crossing over vertex[]{data-label="fig:crossover"}]("crossover".pdf)
This part of the contruction will be refered to as the “wiring region” of the cavity. The wiring region for the example graph $H$ is shown in Figure \[fig:wiring region\].
![Wiring Region for $H$[]{data-label="fig:wiring region"}]("Wiring_Region_Corrected".pdf)
Next, we will add “support beams" to each of the vertex columns as follows. First, extend the tops of the vertex columns by $2n$ units. Now, for vertex $v_i$, add two horizontal beams $2i$ units above the bottom of the extension, one reaching from the vertex column of $v_i$ to one unit past the left limit of the wiring region, and the other reaching to one unit past the right limit. These horizontal pieces cross over other vertex columns in the same way edges cross over vertex columns (as shown in Figure \[fig:crossover\]). The support beams will be used to ensure that playable pieces must be placed with the correct $x$-shift.
The final step will be to create “keys” at the bottom of each vertex column. To do this, extend each vertex column at the bottom by $2n$ units, and attach a single cube hanging off of the left-hand side at the top of this extension. Now, we have completed construction of what we will refer to as the $G$-cavity. Figure \[fig:full cavity\] shows the complete cavity construction for the example graph $H$ including the wiring region, keys, and support beams.
![Full Cavity Construction for $H$[]{data-label="fig:full cavity"}]("Full_Cavity_Corrected".pdf)
### Mold
The purpose of the $G$-cavity is to serve as the effective board, constraining which of the remaining pieces can be played and where. In order to achieve this, we will create a piece such that when it is placed in an $N \times M \times 3$ bounding box, it leaves the cavity as the only remaining open space. First, create an $N \times M \times 3$ bounding box which is exactly large enough the fit the $G$-cavity; take $M$ to be the height of the $G$-cavity in the $y$ direction and $N$ to be its width in the $x$ direction. By construction, the height of the $G$-cavity in the $z$ direction is 2; it’s flat everywhere except at the crossovers which bulge upwards one unit in the z direction. We will now place the $G$-cavity in the bounding box such that the crossover “bulges” face upwards in the $z$ direction, and the bottom of the cavity is flush with the bottom of the box. Now, take the $G$-mold to consist of every voxel in the bounding box which is not occupied by the cavity. Figure 6 shows this construction one $y$-valued layer at a time. It is straightforward to see that since each $y$-valued layer of this piece is simply connected and each layer connects to the next by proper face-to-face contact, the entire construction is simply connected.
[.5]{} ![Constructing the mold at a fixed $y$-coordinate. Dotted line shows box boundary, and shaded region shows the mold.[]{data-label="fig:mold"}]("Mold_Construction_1".pdf "fig:"){width=".8\linewidth"}
[.5]{} ![Constructing the mold at a fixed $y$-coordinate. Dotted line shows box boundary, and shaded region shows the mold.[]{data-label="fig:mold"}]("Mold_Construction_2".pdf "fig:"){width=".8\linewidth"}
[.5]{} ![Constructing the mold at a fixed $y$-coordinate. Dotted line shows box boundary, and shaded region shows the mold.[]{data-label="fig:mold"}]("Mold_Construction_3".pdf "fig:"){width=".8\linewidth"}
### Vertex Pieces
The final step in the reduction is to create pieces corresponding to each vertex. To do this, for each $v_i$, take the part of the $G$-cavity construction corresponding to $v_i$ and all of its incident edges, along with its support beams and its key. An example construction for the vertex piece for $v_4$ in $H$ is shown in Figure \[fig:piece\].
![Vertex Piece for $v_4$ in $H$. Note that this piece still contains the crossovers on its edges/support beams despite appearing flat in this figure.[]{data-label="fig:piece"}]("Vertex_Piece".pdf)
\[placement\] Once the $G$-mold has been placed in the $N \times M \times 3$ bounding box, each vertex piece fits only into its corresponding location in the cavity.
Say that the $G$-mold has been placed in the bounding box such that the cavity has the orientation shown in Figure \[fig:full cavity\]. This is without loss of generality since the $G$-mold spans every dimension of the bounding box and thus has only one valid placement up to rotation and reflection. First, note that by construction, the $G$-mold occupies every cube in the $z = 3$ plane. Further, the $z = 2$ plane is almost entirely full except for a set of disjoint $1 \times 1 \times 3$ tubes which are the result of edge-vertex crossovers. Since every vertex piece has a vertex column which is the length of the entire box in the $y$-dimension, and the $y$-dimension is the longest dimension of the box by construction, this column must be placed spanning the $y$-dimension of the box. Moreover, since the $z = 3$ and $z = 2$ planes are almost entirely filled by the $G$-mold, the vertex piece must be placed with its column spanning the $y$-dimension of the $z=1$ plane.
Next, we show that every vertex piece must be placed with its key section at the $-y$ region of the board and its support beam section at the $+y$ region. Since the support beam region is the same height as the key region, if a vertex piece were placed upside-down, its support beams would be in the key section of the $G$-mold. But since in the key section of the $G$-mold, the entire $z=2$ plane is filled (no crossovers occur), and since at least one of the support beams on every vertex piece crosses over some other vertex column, the cross-over section of the piece’s support beam would intersect the filled $z=2$ plane in the key region of the $G$-mold. So every vertex piece must be placed with its vertex column in the $z=1$ plane such that its key faces the bottom ($-y$) end, and its support beams faces the top end. Since every vertex piece has at least one support beam longer than 3 units, every piece must be placed with its support beams spanning left and right across the $x$-dimension. Further, the edge crossovers must be facing upwards since if they faced down they would intersect the bottom of the bounding box. Thus, if we only allow rotations, the proof is complete; the vertex column must sit in the $z=1$ plane spanning the $y$-dimension, the key side must face $-y$ side of the box, and since the edge crossovers must face upwards in the $z$ direction, the fact that the support beams span the $x$-dimension fixes the position of the piece along the $x$-axis.
If reflections are allowed, the only other option is for the piece to be reflected across the $x=0$ plane from its “correct” position (the position we just proved it is forced into if reflections aren’t used). However, if a vertex $v_i$ is placed in such a way, then it occupies the vertex column for vertex $v_{n-i+1}$ in the $G$-cavity. Since it has been reflected, its key peg now protrudes from the right of its vertex column. But in the key region, the column immediately to right of every vertex column is completely filled in the $z=1$ plane by the $G$-mold, so the key peg would intersect the $G$-mold. So such a placement is not possible.
### Play
\[hard with mold\] Once the $G$-mold has been placed, the next player to move has a winning strategy in this instance of Polycube Packing if and only if the first player has a winning strategy in Node Kayles on $G$.
Imagine that the $G$-mold has already been placed in the $N \times M \times 3$ box, and the remaining pieces are exactly the vertex pieces for each $v_i$. Due to Lemma \[placement\], each $v_i$ can be played only in its corresponding location in the $G$-mold. Once some $v_i$ piece is played in its unique spot, the pieces which it prevents from being played in the future are exactly the pieces corresponding to the vertices $v_i$ is adjacent to in $G$. This is because the $v_i$ piece is constructed to contain the edge pieces for all of its edges, so if $v_i$ has some neighbor $v_j$, the vertex pieces for $v_i$ and for $v_j$ will both occupy the cavity for the edge $v_iv_j$, so playing either one prevents the other from being played. By construction, the forced positions of the vertex pieces are otherwise completely disjoint, so any two pieces corresponding to non-adjacent vertices do not prevent one another from being played. Thus, this game plays exactly like Node Kayles, so winning positions in one game correspond precisely to winning positions in the other.
![Playing a Vertex Piece[]{data-label="fig:play"}]("playing_a_vertex_piece".pdf)
Hardness From an Initially Empty Board
--------------------------------------
The above proof shows that it is PSPACE-hard to determine if a player can win once a single piece, the $G$-mold, has been placed. We will now show that it is also PSPACE-hard to determine a winner from an initially empty $N \times M \times 3$ box. Given an instance of Kayles just as before, we will use the same reduction to create an instance of Polycube Packing, except now we reverse the roles of who plays first, and add some pieces.
### New Pieces
First, we will add the $G$-mold into the set of playable pieces. Next, for each vertex piece $p_i$, we will add two new pieces, $q_i^-$ and $q_i^+$. The purpose is to create these pieces such that if $p_i$ is played first, either $q_i^-$ or $q_i^+$ can be played second and the game immediately ends, and conversely if $q_i^-$ or $q_i^+$ is played first, $p_i$ can be played second and the game immediately ends. If we can construct $q_i^-$ and $q_i^+$ such that this is accomplished, than the first player will be forced to play the $G$-mold piece in order to avoid immediate defeat, and if we can ensure that playing the $G$-mold makes all $q_i^-$ and $q_i^+$ unplayable, then the remainder of the game will play out exactly as in the original reduction.
Because of the use of support beams which cause each vertex piece to span the $x$ and $y$ dimensions of the bounding box, we know that each vertex piece $p_i$ can only be played in the empty bounding box in one of two grid positions, modulo a reflection/rotation: either it is placed with its vertex column in the $z=1$ plane and its edge crossovers rising into the $z=2$ plane, or it is placed with its vertex column in the $z=2$ plane and its edge crossovers in the $z=3$ plane We will call the first such position “normal” position, and the second “raised” position. We can ignore reflections and $180^{\circ}$ rotations since if we can construct the $q_i^-$ and $q_i^+$ to fill the rest of the box once $p_i$ is played in a given orientation, then if $p_i$ is reflected or rotated by $180^{\circ}$, $q_i^-$ and $q_i^+$ can be reflected/rotated in the same way.
Imagine that $p_i$ is placed in normal position. We will construct $q_i^-$ to (almost) fill the rest of the box. The construction is very similar to that of the $G$-mold. First, fill the $z=3$ plane completely, then fill every voxel in the $z=2$ plane not occupied by an edge crossover, and finally, fill every voxel in the $z=1$ layer which is unoccupied and is not below an edge crossover. By not filling the gaps under the edge crossovers we ensure that this construction is simply connected.
Next, imagine that $p_i$ is placed in raised position. We construct $q_i^+$ as follows. First, fill the entire $z=1$ plane. Now, fill every unoccupied voxel in the $z=2$ plane. Finally, fill every voxel in the $z=3$ plane which is not occupied and which is not directly above an edge, support beam, or key stub. Again, by not filling areas directly above edges/beams/keys we ensure simply connected pieces. These new piece constructions are shown in Figure \[fig:q pieces\].
[.5]{} ![Construction of $q_i^-$ and $q_i^+$. Top figures show the cross-sectional view of a vertex piece in normal (raised) position. Bottom figures show construction of $q_i^-$ ($q_i^+$) based on the position of the vertex piece. Dotted line shows box boundary, and shaded region shows $q_i^-$ ($q_i^+$) piece.[]{data-label="fig:q pieces"}]("normal_position".pdf "fig:"){width=".8\linewidth"}
[.5]{} ![Construction of $q_i^-$ and $q_i^+$. Top figures show the cross-sectional view of a vertex piece in normal (raised) position. Bottom figures show construction of $q_i^-$ ($q_i^+$) based on the position of the vertex piece. Dotted line shows box boundary, and shaded region shows $q_i^-$ ($q_i^+$) piece.[]{data-label="fig:q pieces"}]("raised_position".pdf "fig:"){width=".8\linewidth"}
\[lemma normal\] If $p_i$ is placed in the empty bounding box in normal position, $q_i^-$ can be placed subsequently, after which no other pieces can fit into the board.
$q_i^-$ was constructed around $p_i$’s placement in the box in normal position, so we know once $p_i$ is placed in normal position $q_i^-$ will fit. Once they are placed together, $q_i^-$ fills the entire $z=3$ plane, and every voxel in the $z=2$ plane is filled either by an edge crossover or by $q_i^-$ (this is how we constructed $q_i^-$). So if any piece were to fit, it would have to fit entirely in the $z=1$ plane. But by construction, every piece has thickness at least 2 in every dimension so no piece will fit.
\[lemma raised\] If $p_i$ is placed in the empty bounding box in raised position, $q_i^+$ can be placed subsequently, after which no other pieces can fit into the board.
The proof is almost identical to that of Lemma \[lemma normal\]. Again, $q_i^+$ was constructed around $p_i$’s placement in the box in raised position, so we know once $p_i$ is placed in raised position $q_i^+$ will fit. Once they are placed together, $q_i^+$ fills the entire $z=1$ plane, and every voxel in the $z=2$ plane is filled either by $q_i^+$ or by $p_i$. Again, this is because we constructed $q_i^+$ such that it filled every voxel in the $z=2$ plane not occupied by $p_i$ in raised position. So for a piece to fit it would have to squeeze into the $z=3$ plane, and for the reason stated above no piece meets this requirement.
If $q_i^-$ or $q_i^+$ is placed in the empty bounding box, $p_i$ can be placed subsequently, after which no other pieces can fit into the board.
Since $q_i^-$ and $q_i^+$ span every dimension of the bounding box they only have one valid placement up to rotation and reflection, so this is an immediate consequence of Lemmas \[lemma normal\] and \[lemma raised\].
\[mold blocks blockers\] Once the $G$-mold is placed, no $q_i^-$ or $q_i^+$ can fit in the board.
Note that every $q$ piece fills one of the $z$-planes entirely. Once the $G$-mold is placed, there is at least one voxel filled in every $z$-plane by construction, so it is impossible to place a $q$ piece since there is no empty $z$-plane remaining.
### Play
The first player to move from an initially empty board state has a winning strategy in this instance of Polycube Packing if and only if the second player has a winning strategy in Node Kayles on $G$.
First, say Player I has a winning strategy in Node Kayles on $G$ (where they play first). Then they will have a winning strategy in this instance of Polycube Packing where Player II plays first, since if Player II plays a non-$G$-mold piece on the first move, Player I wins immediately by Lemmas \[lemma normal\] - \[mold blocks blockers\], and if Player II plays the $G$-mold, then Lemma \[hard with mold\] tells us that on the next turn this game is equivalent to Node Kayles on $G$ where Player I goes first, so Player I can force a win. On the other hand, if Player I can’t force a win in Kayles when they play first, then in the Polycube Packing game where Player II plays first, on their first turn Player II can play the $G$-mold, and again due to Lemma \[hard with mold\] the game becomes equivalent to Kayles on $G$ with Player I moving first and thus Player I cannot force a win.
Partisan Polycube Packing
-------------------------
In [@sch], Bigraph Node Kayles is also proved PSPACE-complete. This game is identical to Node Kayles except the vertices are partitioned into two independent sets, one for each player, and a player is only allowed to play vertices from their set. It is easy to see that the same exact reduction shows that the partisan version of Polycube Packing is hard in the case where the $G$-mold has already been placed; we construct the same reduction as before, and give each player only the vertex pieces corresponding to their vertices in $G$. To extend to the case of an initially empty box, we will first generate the $q_i^-$ and $q_i^+$ and switch the roles of who plays first, just as before. Then, we put the $G$-mold in the piece set of the first player, and the $q_i^-$ and $q_i^+$ in the piece set of the second player. By Lemmas 3.4-3.7, we see that the first player is forced to play the $G$-mold, at which point the game becomes identical to Bigraph Node Kayles on $G$. So Partisan Polycube Packing is also PSPACE-hard, even from an initially empty box.
2-Player 3D $n$-tris {#tetris}
====================
One simple way to convert Tetris into a 2-player packing game is as follows: players alternate taking control of the keyboard in a standard game of Tetris, and whichever player causes the game to end on their turn loses. In this section we will show that if we generalize this game by introducing a third dimension and allowing the pieces to be arbitrarily-sized polycubes, deciding a winner is PSPACE-complete. The main distinctions between this game and 2-Player Polycube Packing are that in this game the pieces have a fixed order of play, and the rules for piece placement are much more restricted. Following [@tetris], we will study the “offline” version of 2-Player Tetris, where the entire sequence of pieces is known to all parties ahead of time. The complexity of a rather different 2-player version of Tetris was analyzed in [@2ptetris], in which players have separate boards, and when one player clears a line, their “garbage” appears on the other player’s screen.
2-Player 3D $n$-tris is defined as follows: the input is an $N \times M \times K$ box, oriented so that one face is considered the “top”, and a fixed sequence of polycubes $S = (s_1, ..., s_n)$. On turn $i$, the current player places $s_i$ into the box by a continuous downward motion (using any number of rotations) from the top face without intersecting any other pieces. Every piece must be placed with support, meaning its bottom must touch the top of some previously placed piece; another way to phrase this is that there is gravity in the box and pieces fall until they are caught by other pieces. Player 1 moves on odd-numbered turns, Player 2 moves on even-numbered turns, and the first player who cannot move loses. To make this game more similar to traditional Tetris, one could add the feature of plane-clearing: when some $xy$-plane is filled completely, its contents disappear and everything above it falls down by one unit.
Deciding a winner in an instance of 2-Player 3D $n$-tris starting from an initially empty board is PSPACE-complete, with or without plane-clearing.
Overview of Reduction
---------------------
Given an instance of Node Kayles in the form of a graph $G$ with $n$ vertices, we will construct a 2-Player 3D $n$-tris game which will play out in $n$ phases, each phase corresponding to a single turn in Node Kayles. First, we construct a very large piece called the mold which fills most of the box and leaves several cavities open. This will be the first piece in the piece sequence and the first player will be forced to place it with the cavities facing up under threat of immediate loss. After the mold is placed, the phases begin. In the $i^{th}$ phase, the player who moves on turn $i$ in Kayles will be presented with a sequence of $n$ pieces, one for each vertex in $G$. The player will be forced to place one of these pieces into the “region of contest”, and the rest into the “phase $i$ dump”. The piece placed into the region of contest represents the vertex they select on that turn in Kayles. In a similar fashion to the 2-Player Polycube Packing reduction, the region of contest will represent the current position of the Kayles game. In order to allow one player to make $n$ moves in a row while the other player does nothing, we will construct a “garbage chute”, which is a very deep $1 \times 1 \times k$ cavity in the mold, and present the other player with “garbage pieces” which are $1 \times 1 \times \frac{k}{n^2}$ columns. If we make $k$ sufficiently large relative to the rest of the construction, a player presented with garbage will be forced to throw it down the garbage chute, and this allows us to effectively skip that player’s turn, thus completing the reduction. Finally, to extend hardness to the case with plane-clearing, we will modify the mold so that it is impossible for any plane in the box to ever be full.
Hardness Without Plane-Clearing
-------------------------------
Given a graph $G$, we will construct an instance of 2-Player 3D $n$-tris such that the first player has a winning strategy in this game if and only if the first player in Kayles does. Let $n$ denote the number of vertices in $G$ and let $m$ denote the number of edges.
### Phases
As described above, this game will play out in $n$ phases, each corresponding to a player’s turn in Kayles. The odd phases correspond to Player 1’s turns in Kayles, and the even phases correspond to Player 2’s turns, so we will say that Player 1 (2) is “in phase” during the odd (even) phases and “out of phase” during the even (odd) phases. The phase that the game is currently in will be fully defined by the next piece in the piece sequence during a certain position of 2-Player 3D $n$-tris; Section \[sequence\] specifies how the piece sequence is partitioned into phases.
### Mold: General Structure
First, we will construct a polycube which fills nearly the entire box, but leaves open several deep cavities on one side, which we will call the “play side”. We present the general structure of the mold here, and we will specify more precisely the dimensions of the cavities after we construct the other relevant pieces. Essentially, the mold contains a series of $n + 1$ large cavities, plus a small hole in the corner. At the bottom is a cavity we will refer to as the region of play, then above that is a cavity we will call the phase 1 dump, then above that the phase 2 dump, continuing all the way up to the phase $n$ dump at the top. Additionally, at the top left corner there is a narrow and very deep cavity called the garbage chute. Figure \[fig:tetris frame\] shows the mold with its play side facing out of the page.
![Mold with play side up. Grey region is filled in all the way to the top of the box, while white regions are cavities reaching towards (but not touching) the bottom of the box.[]{data-label="fig:tetris frame"}]("tetris_frame".pdf)
### Vertex Pieces
We will construct a vertex piece for each vertex and each phase. For the phase $i$ vertex piece for a vertex $v$, we will construct a polycube which is the union of the following smaller polycubes:
1. For each edge $e_h$ incident on $v$, construct an “edge column”: $[h-1,h] \times [0,1] \times [0,n]$
2. Construct a “binding” which holds the edge columns together: $[0,m] \times [-1,0] \times [i-1, i]$
3. Construct a “handle” which extends off the binding: $[\lceil{\frac{m}{2}}\rceil, \lceil{\frac{m}{2}}\rceil+1] \times [i - (n + 3), -1] \times [i-1,i]$
4. Construct a “stub” which hangs off the side of the handle, one unit away from where the handle meets the binding: $[\lceil{\frac{m}{2}}\rceil-1, \lceil{\frac{m}{2}}\rceil] \times [-3, -2] \times [i-1,i]$. In general we will consider the stub to be a part of the handle.
This construction is shown for an example vertex in Figure \[fig:tetris vertex\].
[.475]{} ![Phase 8 vertex piece for a vertex incident on edges $e_1, e_2, e_7, e_{12}, e_{14}$.[]{data-label="fig:tetris vertex"}]("tetris_vertex_piece_side".pdf "fig:"){width=".8\linewidth"}
[.525]{} ![Phase 8 vertex piece for a vertex incident on edges $e_1, e_2, e_7, e_{12}, e_{14}$.[]{data-label="fig:tetris vertex"}]("tetris_vertex_piece_top".pdf "fig:"){width=".8\linewidth"}
### Mold: Specific Dimensions
First we will construct the cavity for the region of contest. Figure \[fig:region of contest\] shows the dimensions of an $xy$ cross section of this cavity; extending this along the $z$-dimension by $n$ units gives us the complete construction.
Now we will construct the dumps. For each phase $i$, Figure \[fig:dump\] shows the $xy$ cross section of the phase $i$ dump; extending this along the $z$-dimension by $m(n-1)$ units gives us the complete construction.
[.5]{} ![Construction of Region of Contest and Dumps[]{data-label="fig:contest and dumps"}]("region_of_contest".pdf "fig:"){width=".8\linewidth"}
[.5]{} ![Construction of Region of Contest and Dumps[]{data-label="fig:contest and dumps"}]("dump".pdf "fig:"){width=".8\linewidth"}
Now, arrange these cavities as indentations along the top of a sufficiently large box as shown in Figure \[fig:tetris frame\]; the exact arrangement of the various cavities along the surface does not matter, so long as the openings of the cavities are on the same face and they are all pairwise disjoint. This box with various cavities on one side will will be called the “mold”, and the side of the mold where the cavities open up is the “play side”. This gives us an almost complete construction of our mold; to finish it we must add a few more very narrow cavities called “garbage chutes”.
### Garbage and Garbage Chutes
Take $L$ to be 1 unit greater than maximum length of any dimension over all cavities constructed thus far. Now, extend a solid region from the bottom of the mold along the $z$-dimension until the mold’s profile along that dimension is at least $Ln(n-1) + 1$ units long. Next, build an additional cavity that opens up onto the play side whose $xy$ cross section is $1 \times 1$ and which has a depth of $Ln(n-1)$ along the $z$-dimension. This cavity will be referred to as the “garbage chute”, and an example placement is shown in Figure \[fig:tetris frame\]. Now, construct $n(n-1) + 1$ “garbage pieces” which are simply $1 \times 1 \times L$ columns. Finally, add one additional garbage chute on each face of the mold which isn’t the play side; these cavities will be shallower than the main garbage chute, with dimensions $1 \times 1 \times L$. The dimensions of the mold can be extended slightly if necessary to ensure these extra garbage chutes don’t intersect each other or any other previously constructed cavities.
### Piece Sequence {#sequence}
We will construct the piece sequence $S$ as follows: for $1 \leq i \leq n$, let $P_i$, the $i^{th}$ phase sequence, consist of all phase $i$ vertex pieces (in any order), with one garbage piece between every vertex piece. Now construct $S$ to contain the following elements in order:
1. The mold
2. One garbage piece
3. All of the elements of each $P_i$, in ascending order of $i$
### Play
The first player is forced to play the mold with its play side facing upwards on the first turn.
Since every non-play side of the mold has garbage chutes exactly large enough to accommodate one garbage piece, and the second piece in $S$ is garbage, if the first player plays the mold with a non-play side facing up the second player can fill the chute with garbage after which there is no available space exposed to the top of the box, so the first player then loses.
From now on, every claim we make will assume a game state in which the mold has been placed with its play side facing upwards.
If a player is presented with garbage, they must place it in the garbage chute.
Since $L$ was taken to be longer than the longest dimension of all non-garbage-chute cavities, and each garbage piece has length $L$, garbage pieces can only fit into the garbage chute.
This effectively allows us to skip the turns of the player who is out of phase during a given phase, since if we interweave vertex pieces on the in-phase player’s turns with garbage pieces on the out-of-phase player’s turns, the out-of-phase player has no choice but to place their pieces into the garbage, and the contents of the garbage chute have no effect on the rest of the board.
Vertex pieces have a unique placement into the region of contest.
It is clear that the handle of a vertex pieces is forced to lie in the $1 \times n + 2$ strip of the region’s cavity, so it suffices to show that only one of the four placements of the edge column section of the vertex piece into the $m \times 2$ section of the cavity is possible. We can assume without loss of generality that $n < m$; if this is not the case, we can create “dummy edges” which are not incident on any vertex, thus padding the dimensions of any piece whose size depends on $m$ but without effecting game-play. This tells us that the edge column region cannot be placed sideways, since the binding has height $m$ so orienting it vertically would have it rise above the depth of the cavity (which is only $n$). The stub on the handle of each vertex piece and the corresponding indentation on one side of the region of contest prevents any vertex piece from being placed upside down.
During phase $i$, the player who is in phase must place exactly one vertex piece into the region of contest.
Say that one phase $i$ vertex piece has been placed in the region of contest, and that a second phase $i$ vertex piece is then placed on top of it. As seen in Figure \[fig:double placement\], the handle and binding of the second piece must rest on top of the handle and binding of the first piece. By the previous lemma we see both pieces must be placed facing up, and since they are of the same phase, both have the same length of edge columns rising above their bindings. So the tops of the edge columns of the second piece stick out one unit above the tops of the edge columns of the first, so together the two pieces have a vertical height of at least $n + 1$ and thus cannot fit into the region of contest.
![Attempting to place two phase 8 vertex pieces into the region of contest. Handles have been removed for clarity. First white was placed, then grey, so grey now sticks out of the top of the cavity.[]{data-label="fig:double placement"}]("double_placement_3d".pdf)
Next, we show that for any phase $i$, exactly $n-1$ vertex pieces fit into the phase $i$ dump. Since we showed above that we can assume $n < m$, a vertex piece cannot be placed into a dump in the same orientation as it is placed into the region of contest. This is because the thickest end of the dump has length $n$, and if a vertex piece is placed in the same orientation used for placement in the region of contest, its thickest end will have length $m$ and thus will not fit into the dump. When the first vertex piece is placed sideways into a dump its binding will rise from the floor up to height $m$, and when further pieces are placed, the bindings stack consecutively on top of one another, each with height $m$. Since the dump has depth $m(n-1)$, this means that exactly $n-1$ vertex pieces will fit into the phase $i$ dump.
Now we can prove the main claim. We do so by proving the following stronger statement by induction on $i$: during phase $i$, the player who is in phase must place 1 vertex piece into the region of contest and $n-1$ vertex pieces into the phase $i$ dump. First, note that a phase $i$ vertex piece does not fit into any phase $k$ dump for $k > i$, since the handle of a phase $i$ vertex piece has length $n - i + 2$ while the handle region of the phase $k$ dump is $n - k + 2$ and is thus too short to accommodate the vertex piece. Thus, the base case for our inductive claim is easy: by the above arguments, only one phase 1 vertex piece can fit into the region of contest, and the only available cavity that can fit the rest of the pieces is the phase 1 dump, which fits only $n-1$ pieces, forcing us to put at least one into the region of contest, and the rest into the dump. Now for the inductive case, assume the hypothesis holds up to and including some number $i$; we will now prove it for $i + 1$. As proven above, at most one phase $i + 1$ vertex piece can be placed into the region of contest. Due to our inductive hypothesis, we know that for all $h < i + 1$, the phase $h$ dump contains $n - 1$ phase $h$ vertex pieces and is thus full, and for any $k > i + 1$, the phase $k$ dump is too small to fit a phase $i + 1$ vertex piece. So the only cavity into which the phase $i + 1$ vertex pieces can fit is the phase $i + 1$ dump, and as proven in the previous paragraph, this cavity only fits $n - 1$ pieces. So exactly one phase $i + 1$ vertex piece must be placed into the region of contest, and the other $n - 1$ must be placed into the phase $i + 1$ dump, thus completing the proof.
The first player to win in this instance of 2-Player 3D $n$-tris has a winning strategy if and only if the first player in Node Kayles on $G$ does.
By Lemma 3.5 we know that the strategy of the player who is in phase during phase $i$ is fully defined by which vertex piece they choose to place into the region of contest. By Lemma 3.4 we see that pieces must be placed into the region of contest in their correct upright orientation. In this orientation it is clear that a set of pieces can be placed together in the region of contest in increasing order of phase if and only if they occupy disjoint sets of edge columns, since if two vertices occupy the same edge column they will intersect along that column, and if they do not occupy any common edge columns they will not otherwise intersect by construction. In other words, a vertex piece can be placed into the region of contest if and only if its corresponding vertex does not share an edge with the corresponding vertex of any vertex piece that is already in the region of contest. So choosing which vertex piece to place into the region of contest on phase $i$ corresponds to choosing which vertex in $G$ to mark on turn $i$ in Node Kayles, and a vertex piece corresponding to vertex $v$ can be chosen in a given phase if and only if none of the vertex pieces corresponding to $v$’s neighbors have already been chosen. So we see that this game mimics Node Kayles in its moves and its win conditions, so the first player has a winning strategy in this game if and only if they did in Node Kayles on $G$.
Hardness With Plane-Clearing
----------------------------
To extend hardness to the case with plane clearing, we simply modify the non-play sides of the mold to have small indentations to prevent any plane from every being full. Note that in our construction of the mold, each $z$-valued, $xy$-parallel plane has its lower left and right corners filled (see the lower boundary in Figure \[fig:tetris frame\]). Now, for every even value of $z$, modify that layer of the mold by removing the cube in the lower left corner, and for every odd value of $z$, modify that layer by removing the cube in the lower right corner. This modification does not effect play since the only new cavities are 1-unit cubes into which no piece fits, and after this modification is made, no $z$-valued plane can ever get filled and clear since one of its lower corners will always be empty. So our reduction holds regardless of the presence of plane-clearing.
Open Problems {#open}
=============
#### 2-Player Polycube Packing with Constant Sized Pieces:
An interesting question raised by our Polycube Packing reduction is whether we can establish hardness of polycube packing using pieces of constant complexity, since our reduction relied on the use of highly complex pieces. It seems plausible that this problem is also PSPACE-hard, provided the bounding box is allowed to be an arbitrary polycube. The analogous 1-player 2D puzzle of optimally packing constant sized polyominoes into a larger polyomino has been proven NP-complete [@squarepacking].
#### 2-Player Polyomino Packing:
Another natural extension of our polycube packing result would be to prove PSPACE-hardness of 2-Player polyomino packing into a rectangle. Key to our reduction was the fact (implicitly proven by our reduction) that every graph can be represented as the intersection graph of a set of polycubes in $\mathbb{R}^3$. The same is not true for polyominoes lying in the plane. Indeed for any arrangement of polyominoes in $\mathbb{R}^2$, we can replace each with a Jordan arc tracing out its interior without effecting pairwise intersections. This implies that polyomino intersection graphs are a subset of the family of “string graphs”, the graphs representable as the intersection graph of a set of curves in the plane. String graphs properly contain the planar graphs, but beyond that they are not well understood, and their recognition is NP-complete [@ss]. In order to extend this proof technique to the planar case, hardness of Node Kayles would thus need to be known for some subset of string graphs, for example planar graphs. Currently, Kayles is not known to be hard on any restricted class of graphs, and in fact admits a polynomial time algorithm for a subset of the string graphs known as cocomparability graphs [@nim].
#### 2-Player Tetris:
Another problem that remains open is settling the complexity of restricted versions of 2-Player 3D $n$-tris, such as 2-Player 2D $n$-tris, 3D $O(1)$-tris (where the pieces are polycubes of constant size), or ideally, 2-Player 2D Tetris.
Acknowledgements
================
This paper began as a final project for 6.892, an MIT course on Algorithmic Lower Bounds taught by Erik Demaine. The author would like to thank Erik Demaine along with Jeffrey Bosboom, Adam Hesterberg, and Jayson Lynch for their useful criticisms and commentary on earlier drafts of this paper, and for introducing the author to the study of games and puzzles.
[^1]: Tufts University, Medford, MA, USA
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We extend results of Bhagwat and Rajan on a strong multiplicity one property for length spectrum to hyperbolic manifolds with cusps, showing that for two even dimensional hyperbolic manifolds of finite volume, if all but finitely many closed geodesics have the same length, then all closed geodesics have the same length. When the set of exceptional lengths is infinite, but sufficiently sparse, we can show that the two manifolds must have the same volume, and in low dimensions also the same number of cusps. A main ingredient in our proof is a generalization of a result of Selberg on the distribution of poles of Eisenstein series to hyperbolic manifolds.'
address: 'Department of Mathematics, 301 Carney Hall, Boston College Chestnut Hill, MA 02467'
author:
- Dubi Kelmer
title: On distribution of poles of Eisenstein series and the length spectrum of hyperbolic manifolds
---
[^1]
Introduction {#introduction .unnumbered}
============
The length spectrum of a hyperbolic manifold is the set of lengths of primitive closed geodesics listed with their multiplicities. It is an interesting question how much of the geometry of the manifold can be extracted from (partial) information on the length spectrum. For compact hyperbolic surfaces a classical result of Huber [@Huber59], using the trace formula, states that the length spectrum and the Laplace spectrum determine each other, as well as the area of the surface. This result was extended to noncompact finite area hyperbolic surfaces by M[ü]{}ller [@Muller92], where one needs to consider the Laplace spectrum together with the residual spectrum coming from poles of the Eisenstein series.
In higher dimensions the situation is more complicated since the geometric side of the trace formula depends on the complex length spectrum (i.e. lengths and holonomy) and not just lengths. Nevertheless, using the analytic continuation of the Ruelle Zeta function, Bhagwat and Rajan [@BhagwatRajan11] showed that if two compact even dimensional hyperbolic manifolds have the same multiplicities for all but possibly finitely many exceptional lengths, then they must have the same length spectrum. In [@Kelmer12Spectra], we refined their result and showed that one can allow an infinite, but sparse, set of possible exceptional lengths. Moreover, we showed that this data determines the Laplace spectrum and volume of the manifold (the question of whether the Laplace spectrum determines the length spectrum remains open). The main ingredient used in [@Kelmer12Spectra] was a more general version of the trace formula having the length spectrum appear directly on its geometric side obtained by combining different trace formulas corresponding to several representations of the Holonomy group.
The purpose of this note is to extend these results to hyperbolic manifolds with cusps. Instead of working with the trace formula directly as in [@Kelmer12Spectra], we adopt the approach of [@BhagwatRajan11] and use the Ruelle Zeta function. Applying the results of Gon and Park on Selberg Zeta functions [@GonPark08; @GonPark10] we can extend the result of [@BhagwatRajan11] to this setting. However, the refinement in [@Kelmer12Spectra] allowing an infinite exceptional set is more problematic. In this setting we only obtain a partial result showing that if the exceptional set is sufficiently sparse the two manifolds must have the same discrete Laplace spectrum and the same volume. In low dimensions we can also deduce that they have the same number of cusps. Even these partial results already require new results regarding the distribution of poles of Eisenstein series, which are of independent interest. To describe our results in more detail we need to introduce some notation.
Let $G\cong { \mathrm{SO}}_0(d,1)$ denote the group of isometries of hyperbolic $d$-space, ${\mathbb{H}^{d}}$. Any finite volume hyperbolic manifold is of the form $X_{\Gamma}={\Gamma}{\backslash}{\mathbb{H}^{d}}$ where ${\Gamma}<G$ is a torsion free lattice. Given a hyperbolic manifold $X_{\Gamma}$, for every $\ell\in (0,\infty)$ we denote by ${\mathrm{m}}_{\Gamma}(\ell)$ the number of primitive (i.e., wrapping once around) closed geodesics of length $\ell$ in $X_\Gamma$. For any two lattices $\Gamma_1,\Gamma_2< G$ let $$\label{e:Dls}{\mathrm{D}_{\mathrm{L}}}(\Gamma_1,\Gamma_2;T)=\sum_{\ell\leq T}|{\mathrm{m}}_{\Gamma_1}(\ell)-{\mathrm{m}}_{\Gamma_2}(\ell)|,$$ and $$\label{e:dls}{\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)=\limsup_{T\to\infty}\frac{\log({\mathrm{D}_{\mathrm{L}}}(\Gamma_1,\Gamma_2;T))}{T}.$$ One can think of ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)$ as measuring the scaled density of the exceptional set of lengths having different multiplicities in the two manifolds, in particular, if this exceptional set is finite then ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)=0$. The result of [@Kelmer12Spectra] states that for two compact even dimensional hyperbolic manifolds, the condition that $d_L(\Gamma_1,\Gamma_2)<\frac{1}{2}$ already implies that the two manifolds have the same length spectrum. Moreover, it was shown there that (in any dimension) the weaker condition that $\mathrm{d}_L(\Gamma_1,\Gamma_2)<\frac{d-1}{2}$ implies that the two manifolds have the same Laplace spectrum (and hence, by Weyl’s law, also the same volume).
Our first result extends [@BhagwatRajan11] to finite volume non-compact hyperbolic manifolds. Since we rely on results of [@GonPark10] we need to impose a certain technical condition on the cusps of $X_{\Gamma}$. We say that $X_{\Gamma}$ has neat cusps if for any parabolic subgroup $P<G$ with unipotent radical $N<P$ we have ${\Gamma}\cap P={\Gamma}\cap N$.
\[t:1\] Let $X_{\Gamma_1},X_{\Gamma_2}$ denote two even dimensional hyperbolic manifolds of finite volume with neat cusps. If $m_{\Gamma_1}(\ell)=m_{\Gamma_2}(\ell)$ for all $\ell\in {\mathbb R}$ except perhaps some finite exceptional set, then $m_{\Gamma_1}(\ell)=m_{\Gamma_2}(\ell)$ for all $\ell$.
Next, we want to generalize the results of [@Kelmer12Spectra] allowing an infinite set of exceptions. The relation between the length spectrum and discrete Laplace spectrum follows by more or less the same arguments as in the compact case. We note that when the manifold is not compact, the corresponding Weyl law includes both discrete and continuous spectrum, hence, the fact that the two manifolds have the same discrete spectrum no longer implies that they have the same volumes. Nevertheless, we show:
\[t:2\] Let $X_{\Gamma_1},X_{\Gamma_2}$ denote two $d$-dimensional hyperbolic manifolds of finite volume with neat cusps.
1. If ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)<\frac{d-1}{2}$ then the two manifolds have the same discrete Laplace spectrum and the same volume.
2. For $d=2,3$, if ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)<1/4$ then the two manifolds have the same number of cusps.
The proof of Theorem \[t:2\] relies on results on the distribution of poles of Eisenstein series associated to hyperbolic manifolds, generalizing previous results of Selberg [@Selberg90] for hyperbolic surfaces. To describe these results we first recall some definitions and facts regarding these Eisenstein series.
Fix an Iwasawa decomposition, $G=NAK$, with $K$ maximal compact, $A$ Cartan, and $N$ unipotent, and let $P=NAM$ be a minimal parabolic where $M=Z_K(A)$ is the centralizer of $A$ in $K$. The cusps of $\Gamma$ are the $\Gamma$-conjugacy classes of minimal parabolic subgroups of $G$ intersecting $\Gamma$ nontrivially. Let $P_1,\ldots, P_\kappa$ denote a full set of representatives for these classes and let $k_i\in K$ such that $P_i=k_iPk_i^{-1}$. For each cusp $P_i=N_iA_iM_i$, let $\Gamma_{P_i}=\Gamma\cap P_i$ and $\Gamma_{N_i}=\Gamma\cap N_i$. The spherical Eisenstein series corresponding to the $i$’th cusp is the function on the upper half space defined for $\Re(s)>d-1$ by the convergent series $$\label{e:EisensteinSph}E_i(s,z)=\sum_{\gamma\in \Gamma_{\!P_i}{\backslash}\Gamma} y_i(\gamma.z)^s,$$ where we use the coordinates $z=(x,y)\in {\mathbb R}^{d-1}\times {\mathbb R}^+$ for the upper half space and set $y_i(z)=y(k_i^{-1}.z)$. The constant term of $E_i(s,z)$ with respect to the $j$’th cusp is defined by $$\label{e:ConstSph} E_{ij}(s,z)=\frac{1}{v_j}\int_{\Gamma_{N_j}{\backslash}N_j}E_i(s,n.z)dn,$$ where $v_j={\mathrm{vol}}(\Gamma_{N_j}{\backslash}N_j)$, and $dn$ is Haar measure on $N_j$. These constant terms satisfy $$\label{e:ConstSph1}
E_{ij}(s,z)=\delta_{ij}y_j(z)^s+\phi_{ij}(s)y_j(z)^{d-1-s},$$ with $\phi_{ij}(s)$ the coefficients of the scattering matrix.
The Eisenstein series $E_j(s,z)$, the scattering matrix $\phi(s)=(\phi_{ij}(s))$, and its determinant ${{\varphi}}(s)=\det(\phi(s))$ (a priori defined for $\Re(s)>d-1$) have a meromorphic extension to the complex plane and satisfy the functional equation $\phi(s)\phi(d-1-s)=I$. The poles of ${{\varphi}}(s)$ which are also the poles of the Eisenstein series, are all in the half plane $\Re(s)<\tfrac{d-1}{2}$ except for at most finitely many poles in the interval $(\tfrac{d-1}{2},d-1]$. From the functional equation ${{\varphi}}(s){{\varphi}}(d-1-s)=1$, we can understand the distribution of these poles by looking at the zeroes of ${{\varphi}}(s)$ in the half plane $\Re(s)>\tfrac{d-1}{2}$. For these we show the following.
\[t:ZeroDist1\] The zeroes of the scattering determinant, ${{\varphi}}(s)$, in the half plane $\Re(s)>\tfrac{d-1}{2}$ are all located in some vertical strip, and writing these zeroes as $\rho=\beta+i\gamma$ (with multiplicities) we have:
1. There is a constant $A_{\Gamma}$ such that $$\label{e:ZeroDist1}
\mathop{\sum_{|\gamma|<T}}_{\beta>\tfrac{d-1}{2}}(\beta-\tfrac{d-1}{2})=\frac{\kappa(d-1)}{2\pi} T\log(T)+A_\Gamma T+ O(\log(T))$$
2. For any $\alpha\geq \alpha_0=d-\tfrac{5}{4}$ $$\label{e:ZeroDist2}
\mathop{\sum_{|\gamma|<T}}_{\beta>\alpha}(\beta-\alpha)\ll T\min\{\log(\tfrac{1}{(\alpha-\alpha_0)}),\log\log T\}$$
Using the relation between the zeroes and poles one can interpret this result as saying that a hundred percent of the poles $\tilde{\rho}=\tilde{\beta}+i\gamma$ of ${{\varphi}}(s)$ in the half plane $\Re(s)<\tfrac{d-1}{2}$ are concentrated in the strip $\tfrac{1}{4}\leq \Re(s)<\tfrac{d-1}{2}$, in the sense that, for any $\alpha<\tfrac{1}{4}$, as $T\to\infty$ $$\mathop{\sum_{|\gamma|<T}}_{\alpha<\tilde{\beta} <\tfrac{d-1}{2}}\!\!\!(\tfrac{d-1}{2}-\tilde{\beta})\sim \mathop{\sum_{|\gamma|<T}}_{\tilde{\beta} <\tfrac{d-1}{2}}(\tfrac{d-1}{2}-\tilde{\beta}).$$
For hyperbolic surfaces this result is due to Selberg [@Selberg90] and the value of $\alpha_0=\tfrac{3}{4}$ is best possible. Indeed, for $\Gamma={ \mathrm{SL}}_2({\mathbb{Z}})$ the scattering determinant can be computed explicitly in terms of the Riemann Zeta function and its poles are located at the zeroes of $\zeta(1-2s)$, hence, a positive proportion[^2] are on the line $\Re(s)=\tfrac{1}{4}$. For $3$-manifolds, when $\Gamma={ \mathrm{SL}}_2({\mathcal{O}}_K)$ with ${\mathcal{O}}_K$ the ring of integers of a quadratic complex number field, the poles of the scattering determinant are at the zeros of the Dedekind Zeta function $\zeta_K(1-s)$ and holds with $\alpha_0=\tfrac{3}{2}$. However, for general $\Gamma< { \mathrm{PSL}}_2({\mathbb C})$ our method only gives a weaker result with $\alpha_0=\tfrac{7}{4}$.
Zeta functions of Selberg and Ruelle
====================================
Selberg Zeta functions
----------------------
We recall the correspondence between closed geodesics on $X_{\Gamma}$ and hyperbolic conjugacy classes in ${\Gamma}$. Let ${\mathfrak{g}}={\mathfrak{n}}\oplus {\mathfrak{a}}\oplus {\mathfrak{k}}$ denote the decomposition of the Lie algebra ${\mathfrak{g}}$ corresponding to the Iwasawa decomposition $G=NAK$. Fix once and for all an element $H_0\in {\mathfrak{a}}$ such that $\rho(H_0)=\frac{d-1}{2}$, where $\rho$ denotes half the sum of positive roots, and let $a_t=\exp(tH_0)\in A$. Any hyperbolic $\gamma\in \Gamma$ is conjugated in $G$ to an element $m_\gamma a_{\ell_\gamma}\in MA^+$ where $A^+=\{a_t|t>0\}$ and $M$ is the centralizer of $A$ in $K$. The pair $(\ell_\gamma,m_{\gamma})$ is the length and holonomy class of the closed geodesic corresponding to $\gamma$, where $\ell_\gamma$ is uniquely determined by the conjugacy class of $\gamma$ and $m_\gamma$ is determined up to conjugacy in $M$.
The Selberg Zeta function, $Z_{\Gamma}(\sigma,s)$, corresponding to an irreducible representation $\sigma\in \hat{M}$ is defined on the half plane $\Re(s)>d-1$ by $$\begin{aligned}
\label{e:SZF}\nonumber Z_{\Gamma}(\sigma,s)
=\exp\left(-\sum_{\gamma\in \Gamma_h'}\sum_{j=1}^\infty\frac{\overline{\chi_{\sigma}(m_{\gamma^j})}}{j D(\gamma^j)}e^{-(s-\frac{d-1}{2})j\ell_\gamma}\right)\end{aligned}$$ where ${\Gamma}_h'$ denotes the set of primitive hyperbolic conjugacy classes and $$D(\gamma)=e^{\frac{d-1}{2}\ell_\gamma}|\det\big(\mathrm{Ad}(m_\gamma a_\gamma)^{-1}-\mathrm{Id})|_{\mathfrak{n}}\big)|.$$
Before we present the result of [@GonPark10] on the analytic continuation of these Zeta functions we need to recall some more background on Eisenstein series. In addition to the spherical Eisenstein series given in , for any representation class $\sigma\in \hat{M}$ one can define corresponding Eisenstein series, scattering matrix, and scattering determinant. We refer the reader to [@Warner79; @GonPark10] for the precise definition, and just note here the following general result from [@Muller89 Section 6] regarding the poles of these scattering determinants.
\[p:Muller\] Let ${{\varphi}}_{\Gamma,\sigma}(s)$ denote the scattering determinant corresponding to $\sigma\in \hat{M}$. Then ${{\varphi}}_{{\Gamma},\sigma}(s)$ is a meromorphic function with all of its poles in the half plane $\Re(s)<\tfrac{d-1}{2}$ except for finitely many poles in the interval $(\tfrac{d-1}{2},d-1]$. Moreover if we denote by $S_{{\Gamma},\sigma}$ the set of poles in $\Re(s)<\tfrac{d-1}{2}$ then $\sum_{\eta\in S_{{\Gamma},\sigma}}\frac{\Re(\eta-\tfrac{d-1}{2})}{|\eta-\tfrac{d-1}{2}|^2}$ converges.
We can now state the result of Gon and Park [@GonPark10 Theorem 4.6]. For $k=0,\ldots, [\frac{d-1}{2}]$, let $\sigma_k\in \hat{M}$ correspond to the irreducible representation of $M$ on the space $\bigwedge^k({\mathbb C}^{d-1})$ (when $d=2n+1$ and $k=n$ we denote by $\sigma_n^{\pm}$ the two unramified irreducible representations acting on $\bigwedge^n({\mathbb C}^{2n})$. We also denote by ${\triangle}_k$ the Laplacian acting on $k$-forms (where ${\triangle}_0={\triangle}$ is just the hyperbolic Laplacian).
\[p:GonPark\] The Selberg Zeta function $Z_{\Gamma}(\sigma_k,s)$ has a meromorphic continuation to the complex plane. It has poles at the points $s=\frac{d-1}{2}-\ell$, $\ell\in {\mathbb N}\cup \{0\}$ (and additional zeroes/poles at negative integers when $d$ is even). It has spectral zeroes at the points $s=\tfrac{d-1}{2}\pm ir$ with $\lambda=r^2+(\tfrac{d-1}{2}-k)^2$ an eigenfunction of ${\triangle}_k$, and residual zeroes at $\eta\in S_{\Gamma,\sigma_k}$. It also has finitely many residual poles in the interval $(\tfrac{d-1}{2},d-1]$ at the poles of ${{\varphi}}_{\sigma_k}$. The order of the spectral zeroes equals the multiplicity of the corresponding eigenvalue and the order of the residual poles and zeroes are equals to $\dim(\sigma_k)$ times the order of the corresponding pole of ${{\varphi}}_{\sigma_k}$.
Ruelle Zeta function
--------------------
Information about the length spectrum is captured directly by the Ruelle Zeta function, defined in the half plane $\Re(s)>\frac{d-1}{2}$ by the Euler product $$R_{\Gamma}(s)=\prod_{\gamma\in \Gamma_h'}(1-e^{-s\ell_\gamma}).$$ As shown in [@GonPark10], when $d$ is even it is related to the Selberg Zeta functions via $$\label{e:RF1} R_\Gamma(s)=\prod_{k=0}^{d/2-1}\left[\frac{Z_\Gamma(\sigma_k,s+k)}{Z_\Gamma(\sigma_k,s+d-1-k)}\right]^{(-1)^{k+1}},$$ and when $d$ is odd we have $$\label{e:RF2}R_\Gamma(s)=\prod_{k=0}^{\tfrac{d-1}{2}}\left[ Z_\Gamma(\sigma_k,s+k) Z_\Gamma(\sigma_k,s+d-1-k)\right]^{(-1)^{k+1}},$$ where in the odd case $d=2n+1$ we denoted by $Z_\Gamma(\sigma_n,s):=Z_\Gamma(\sigma_n^+,s)Z_\Gamma(\sigma_n^-,s)$.
We also consider a variant of the Ruelle Zeta function that is similar to the Selberg Zeta function for surfaces, that is, $$\label{e:Zeta}Z_\Gamma(s)=\prod_{\gamma\in \Gamma_h'}\prod_{a=0}^\infty(1-e^{-(s+a)\ell_\gamma}).$$ When $d$ is even, a simple manipulation of shows that this Zeta function can be expressed as a finite product of Selberg Zeta functions and their inverses as $$\label{e:Zeta2}
Z_\Gamma(s)=\prod_{k=0}^{d/2-1}\left[\prod_{j=k}^{d-2-k}Z_\Gamma(\sigma_k,s+j)\right]^{(-1)^{k+1}}.$$
Proof of Theorem 1
==================
Let $\Gamma_1,\Gamma_2\subset { \mathrm{SO}}(d,1)$ be two torsion free lattices with $d$ even. Assume that $m_{\Gamma_1}(\ell)=m_{\Gamma_2}(\ell)$ for all $\ell\in {\mathbb R}$ except perhaps for a finite exceptional set $\{\ell_1,\ldots, \ell_N\}$. For these exceptional lengths let $\Delta m(\ell_j)=m_{\Gamma_1}(\ell_j)-m_{\Gamma_2}(\ell_j)$. We will show that $\Delta m(\ell_j)=0$ as well.
To do this consider the quotient $F(s)=\frac{Z_{\Gamma_1}(s)}{Z_{\Gamma_2}(s)}$ of the corresponding Zeta functions given in and note that for $\Re(s)>d-1$ we have $$F(s)=\prod_{j=1}^N \prod_{a=0}^\infty(1-e^{-(s+a)\ell_j})^{\Delta m(\ell_j)}.$$ This product absolutely and uniformly converges on any compact set away from $\{s\in {\mathbb C}| e^{-(s+a)\ell_j}= 1,\; j=1,\ldots, N,\; a\in {\mathbb N}\cup \{0\}\}$. Consequently, $F(s)$ is a meromorphic functions with all its zeros and poles located at the points $\rho_{a,b,j}=-a+\frac{2\pi i b}{\ell_j}$ with $a,b\in {\mathbb{Z}}$, $a\geq 0$. The order of the pole/zero $\rho_{a,b,j}$ does not depend on $a$ and is given by $$\sum_{\{i|\ell_i b\in \ell_j{\mathbb{Z}}\}}\Delta m(\ell_i).$$ In order to avoid any possible cancelation in this sum, let $\ell_1$ be the largest for which $\Delta m(\ell_1)\neq 0$. If there are other $\ell_i$ with $\frac{\ell_i}{\ell_1}\in{\mathbb{Q}}$ and $\Delta m(\ell_i)\neq 0$ let $q\in {\mathbb N}$ denote their least common multiple. Now for all $b\in {\mathbb N}$ with $(b,q)=1$ we have that $\ell_i b/\ell_1\in {\mathbb{Z}}$ if and only if $\ell_i=\ell_1$, and hence $F(s)$ has poles/zeroes at all points $\rho_{a,b}=-a+\frac{2\pi i b}{\ell_1}$ with $a\geq 0$ and $(b,q)=1$, with the same order $\Delta m(\ell_1)\neq 0$.
Next use to express $F(s)$ as a finite product of quotients of Selberg Zeta functions. Since $Z_{\Gamma_i}(\sigma_k,s+j)$ have no complex poles in the half plane $\Re(s)<-j$, all of the poles/zeroes of $F(s)$ come from the residual zeroes of $Z_{\Gamma_i}(\sigma_k,s+j)$ with $0\leq k\leq d/2-1$, and $k\leq j\leq d-2-k,\; i=1,2$. This means that for any pair $a,b\in {\mathbb N}$ with $a\geq d-1$ and $(b,q)=1$, there is some $j_{a,b}\leq d-1$ such that $\rho_{a-j_{a,b},b}\in S_{\Gamma_i,\sigma_k}$ is a residual zero of $Z_{\Gamma_i}(\sigma_k,s)$, for some $0\leq k\leq d/2-1$ and $i\in\{1,2\}$. We thus get a bound $$\sum_{a=d}^\infty \mathop{\sum_{b=1}^\infty}_{(b,q)=1}\frac{a-j_{a,b}+\frac{d-1}{2}}{(a-j_{a,b}+\frac{d-1}{2})^2+4\pi^2 b^2/\ell_1^2}\leq \sum_{i=1}^2\sum_{k=0}^{d/2-1}\sum_{\eta\in S_{\Gamma_i,\sigma_k}}\frac{|\Re(\eta-\frac{d-1}{2})|}{|\eta-\frac{d-1}{2}|^2},$$ noting that all terms are positive and every summand on the left also appears on the right. However, the sum on the right converges by Proposition \[p:Muller\] while the sum on the left clearly diverges, implying that $\Delta m(\ell_1)= 0$ as claimed.
Proof of Theorem 2
==================
To prove Theorem 2 consider the quotient of the Ruelle Zeta functions $$F(s)=\frac{R_{\Gamma_1}(s)}{R_{\Gamma_2}(s)}=\prod_{\ell } (1-e^{-s\ell})^{\Delta m(\ell)}.$$ A priori, this equality holds for $\Re(s)>d-1$, however, if we know that $$\label{e:expbound}
\sum_{\ell\leq T}|\Delta m(\ell)|=O(e^{cT}),$$ then the right hand side absolutely converges for $\Re(s)>c$ and hence defines an analytic function that has no zeroes or poles on that half plane. The condition ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)<\tfrac{d-1}{2}$ implies that holds with some $c<\tfrac{d-1}{2}$ and hence the zeroes and poles of $R_{\Gamma_1}(s)$ and $R_{\Gamma_2}(s)$ in the half pane $\Re(s)>c$ must be the same. From the factorization of $R_{\Gamma_i}(s)$ as a product of Selberg Zeta functions in together with the location of the poles and zeroes of the Zeta functions $Z_\Gamma(\sigma_k,s)$ given in Proposition \[p:GonPark\], we see that $R_{\Gamma_i}(s)$ has no zeroes in $\Re(s)\geq \frac{d-1}{2}$ and its poles there all come from the spectral zeroes of $Z_{\Gamma}(\sigma_0,s)$ with $\sigma_0$ the trivial representation. That is, $R_{\Gamma_i}(s)$ has poles at $s=\frac{d-1}{2}+ir$ whenever $\lambda=\frac{(d-1)^2}{4}+r^2$ is an eigenvalue of ${\triangle}$ with eigenfunction in $L^2(X_{\Gamma_i})$ with order given by the multiplicity of the eigenvalue. In particular, this shows that the discrete spectrum of $\Gamma_1$ and $\Gamma_2$ must be the same.
To see that the volumes are also equal we use the Weyl law, which for non-compact hyperbolic manifolds takes the form, $$N_{\Gamma}(T)-\frac{1}{2\pi}\int_{-T}^T \frac{{{\varphi}}_{{\Gamma}}'}{{{\varphi}}_{\Gamma}}(\tfrac{d-1}{2}+it)dt=C_d{\mathrm{vol}}(X_{\Gamma})T^d+O(T^{d-1})+O(T\log(T)),$$ where $N_{\Gamma}(T)$ denotes the number of Laplace eigenvalues $\lambda< \tfrac{(d-1)^2}{4}+T^2$ and $C_d$ is a constant depending only on $d$. The term involving the scattering determinant can be further evaluated by counting poles. Specifically if we let $$S_{\Gamma}(T)=\{\eta\in S_{\Gamma,\sigma_0}:|{\mathrm{Im}}(\eta|\leq T\},$$ following the same argument as in [@Selberg90 equation (0.15)], we see that $$|S_{\Gamma}(T)|=-\frac{1}{2\pi}\int_{-T}^T \frac{{{\varphi}}_{{\Gamma}}'}{{{\varphi}}_{\Gamma}}(\tfrac{d-1}{2}+it)dt+O(T),$$ and the Weyl law takes the form $$\label{e:Weyl}
N_{\Gamma}(T)+|S_{\Gamma}(T)|=C_d{\mathrm{vol}}(X_{\Gamma})T^d+O(T^{d-1})+O(T\log(T)).$$ Comparing this for the two lattices, noting that $N_{\Gamma_1}(T)=N_{\Gamma_2}(T)$ and bounding all error terms by, say $O(T^{d-1/2})$, we see that $$|S_{{\Gamma}_1}(T)|-|S_{{\Gamma}_2}(T)|=C_d({\mathrm{vol}}(X_{\Gamma_1})-{\mathrm{vol}}(X_{\Gamma_2}))T^d+O(T^{d-1/2}).$$
Now let $c'=\max\{c,\tfrac{d-3}{2}\}$ so that $\frac{R_{\Gamma_1}(s)}{R_{\Gamma_2}(s)}$ has no poles or zeros in $c'<\Re(s)<\tfrac{d-1}{2}$. In this range all zeroes of $R_{\Gamma_i}(s)$ are the residual zeros of $Z_{\Gamma_i}(\sigma_0,s)$, and hence the residual zeroes of $Z_{\Gamma_1}(\sigma_0,s)$, and $Z_{\Gamma_2}(\sigma_0,s)$ in $\Re(s)>c'$ must be the same. Fix some $c'<c_1<\tfrac{d-1}{2}$ and let $$S_{\Gamma}(c_1,T)=\{\eta\in S_{\Gamma}(T): \Re(\eta)\leq c_1\}.$$ Recalling the relation between the residual zeroes and the poles of the scattering matrix we get that $$|S_{{\Gamma}_1}(T)|-|S_{{\Gamma}_2}(T)|=|S_{{\Gamma}_1}(c_1,T)|-|S_{{\Gamma}_2}(c_1,T)|,$$ where all sets are counted with multiplicities. Finally, recalling that $\eta\in S_{\Gamma}$ if and only if $d-1-\eta$ is a zero of ${{\varphi}}_{\Gamma}$ and using we get that for each lattice $$\begin{aligned}
|S_{{\Gamma}_i}(c_1,T)|&\leq&\frac{1}{\tfrac{d-1}{2}-c_1}\sum_{\eta \in S_{{\Gamma}_i}(c_1,T)}(\tfrac{d-1}{2}-\Re \eta)\\
&\ll& \sum_{\eta \in S_{{\Gamma}_i}(T)}(\tfrac{d-1}{2}-\Re\eta)
\ll T\log(T)\end{aligned}$$ We thus get that $||S_{{\Gamma}_1}(T)|-|S_{{\Gamma}_2}(T)||\ll T\log T$ and hence $$\begin{aligned}
C_d(|{\mathrm{vol}}(X_{\Gamma_1})-{\mathrm{vol}}(X_{\Gamma_2})|)T^d&\ll& T^{d-1/2},\end{aligned}$$ implying that ${\mathrm{vol}}(X_{\Gamma_1})={\mathrm{vol}}(X_{\Gamma_2})$.
For the results on the number of cusps, let $d\leq 3$ and assume that ${\mathrm{d}_{\mathrm{L}}}({\Gamma}_1,{\Gamma}_2)<\tfrac{1}{4}$, so that $D_L(\Gamma_1,\Gamma_2,T)=O(e^{cT})$ with some $c<\tfrac{1}{4}$. Comparing the Zeta functions as before we see that the poles of ${{\varphi}}_{\Gamma_1}$ and ${{\varphi}}_{\Gamma_2}$ in the half plane $\Re(s)>c$ are the same. Hence $$\begin{aligned}
\lefteqn{\sum_{\eta\in S_{\Gamma_1}(T)}(\tfrac{d-1}{2}-\Re\eta)-\sum_{\eta\in S_{\Gamma_2}(T)}(\tfrac{d-1}{2}-\Re\eta)}\\
&&=\sum_{\eta\in S_{\Gamma_1}(c,T)}(\tfrac{d-1}{2}-\Re\eta)-\sum_{\eta\in S_{\Gamma_2}(c,T)}(\tfrac{d-1}{2}-\Re\eta).\end{aligned}$$ By , the left hand side equals $\frac{(\kappa_1-\kappa_2)(d-1)}{2\pi}T\log(T)+O(T)$, hence, if we can bound each one of the sums on the right by $O(T\log\log(T))$ we would get that $\kappa_1=\kappa_2$ as claimed.
To bound the sum on the right, for $\Gamma={\Gamma}_i$ any one of the two lattices we can write each pole as $\eta=d-1-\beta-i\gamma$ with $\beta+i\gamma$ a zero of ${{\varphi}}_{\Gamma}(s)$. Setting $\alpha=d-1-c$ we get $$\begin{aligned}
\sum_{\eta\in S_{\Gamma}(c,T)}(\tfrac{d-1}{2}-\Re\eta)=\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}(\beta-\tfrac{d-1}{2}).\end{aligned}$$ The condition $c<\tfrac{1}{4}$ implies $\alpha>\alpha_0$ and we can bound $$\begin{aligned}
\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha_0}(\beta-\alpha_0)&\geq& \mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}(\beta-\alpha_0)
\geq(\alpha-\alpha_0)\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}1\\
&= & \left(\tfrac{\alpha_0-\alpha}{\alpha-\frac{d-1}{2}}\right)\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}(\alpha-\beta+\beta-\tfrac{d-1}{2}),\end{aligned}$$ hence $$\begin{aligned}
\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}(\beta-\tfrac{d-1}{2})\leq \left(\tfrac{\alpha-\frac{d-1}{2}}{\alpha_0-\alpha}\right)\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha_0}(\beta-\alpha_0)+\mathop{\sum_{|\gamma|<T}}_{\beta \geq \alpha}(\beta-\alpha),\end{aligned}$$ and from the sums on the right are bounded by $O(T\log\log(T))$ concluding the proof.
If we assume that $d_L(\Gamma_1,\Gamma_2)<\tfrac{1}{2}$ we get that all the poles and zeroes of $R_{\Gamma_1}(s)$ and $R_{\Gamma_2}(s)$ cancel out in the half plane $\Re(s)\geq \tfrac{1}{2}$. However, in contrast to the compact case, we cannot deduce from this that Zeta functions are the same because we cannot exclude the possibility that the spectral zeroes of $Z_{\Gamma_i}(\sigma_k,s)$ will cancel out with residual poles of $Z_{\Gamma_i}(\sigma_{k'},s_j)$ for some $k'<k$.
The result on the number of cusps is restricted to dimensions $d=2,3$ for a similar reason. In general, for $d>3$ our result shows that almost all poles of $Z_{\Gamma}(\sigma_0,s)$ are in the half plane $\Re(s)\geq \frac{1}{4}$, but even if we assume that ${\mathrm{d}_{\mathrm{L}}}(\Gamma_1,\Gamma_2)<\frac{1}{4}$ we can’t conclude that the number of cusps is the same because of possible cancelation with zeroes of the other Zeta functions.
Distribution of Poles of Eisenstein series {#s:DistPoles}
==========================================
We conclude with the proof of Theorem \[t:ZeroDist1\], generalizing the results of [@Selberg90] on the distribution of poles of Eisenstein series to hyperbolic manifolds of higher dimensions. Our proof is very similar to Selberg’s original proof, that is, we express the scattering coefficients as certain Dirichlet series with positive coefficients and then use general results about such Dirichlet series to study the distribution of the zeroes of the scattering determinant. The only new difficulty is coming from the fact that in higher dimensions these Dirichlet series have a faster growth rate on the critical line, requiring some modifications of the arguments in [@Selberg90].
Dirichlet series with positive coefficients
-------------------------------------------
We prove some results about general Dirichlet series with positive coefficients of the form, $$\label{e:fseries}
f(s)=\sum_{n=1}^\infty \frac{a_n}{\lambda_n^s},$$ with all $\lambda_n,a_n>0$. In particular, we need the following generalization of Selberg’s [@Selberg90 Lemma 3]
\[l:boundintf\] Let $f(s)$ be as in . Assume that the series absolutely converges for $\Re(s)>1$, that it has an analytic continuation to $\Re(s)>0$ with a simple pole at $s=1$ and at most finitely many poles in the strip $0<\Re(s)<1$, all in the interval (0,1). Further assume that $f(s)$ has a continuous extension to $\Re(s)=0$ and it satisfies the growth condition $$\label{e:tbound}
|f(\sigma+it)|=O(|t|^{r}),\quad 0\leq \sigma <3/2,\; t\gg 1.$$ for some $r\geq 1$. Let $\sigma_1=\frac{4r-1}{4r}$, then for all $\sigma\geq \sigma_1$ we have $$\frac{1}{T}\int_1^T|f(\sigma+it)|^2dt\ll \min(\frac{1}{(\sigma-\sigma_1)^2},\log^2(T)).$$
For $r< 1$ our proof gives a similar result with $\sigma_1=\tfrac{3}{4}$ and an error of $\log(T)$ instead of $\log^2(T)$. This is worse than the original result of Selberg’s [@Selberg90 Lemma 3] who showed this for $r=1/2$ with a better value of $\sigma_1=\tfrac{1}{2}$. Our proof goes along the same line as his, except in one point where his argument seems to work only when $r\leq 1/2$ and we had to make some modifications, resulting in a slightly worse value for $\sigma_1$.
Before we proceed with the proof we recall a couple of standard results on these Dirichlet series.
\[l:Ax\] Let $f(s)$ be as in Proposition \[l:boundintf\] and let $$\label{e:Ax}
A_f(x)=\sum_{\lambda_n\leq x} a_n.$$ Then $$\label{e:Af}
A_f(x)=a x+\sum_{j=1}^k x^{\rho_j}p_j(\log(x))+O(x^{1-\frac{1}{2r}}\log(x)),$$ where $a=\mathrm{Res}_{s=1}f(s)$, $\rho_1,\ldots,\rho_k$ are the poles of $f(s)$ in $(0,1)$, and $p_j$ are certain polynomials.
This follows from standard techniques of analytic number theory. Specifically, let $$B_f(x)=\sum_{\lambda_n\leq x}a_n(1-\tfrac{\lambda_n}{x}),$$ that is, $B_f(x)=\frac{1}{x}\int_1^x A_f(t)dt$. For any $c>1$ we can write $B_f(x)=\frac{1}{2\pi i}\int_{\Re(s)=c}\frac{f(s)x^s}{s(s+1)}ds$. Taking $c=1+\frac{1}{\log(x)}$ and noting that $|f(s)|\ll \frac{1}{c-1}\ll \log(x)$ is uniformly bounded there we get that $$B_f(x)=\frac{1}{2\pi i}\int_{c-iT}^{c+iT}\frac{f(s)x^s}{s(s+1)}ds+O(\frac{x\log(x)}{T}).$$ Shifting the contour of integration to $\Re(s)=\epsilon$ (and taking $\epsilon\to 0$) we get that $$\begin{aligned}
\frac{1}{2\pi i}\int_{c-iT}^{c+iT}\frac{f(s)x^s}{s(s+1)}ds &=& \frac{ax}{2}+\sum_j x^{\rho_j}\tilde{p}(\log(x))\\
&&+O(T^{r-1}\log(T))+O(\frac{x\log(T)}{T^2\log(x/T^r)})$$ where the explicit terms come from the residues at the poles, we used $f(\epsilon+it)\ll t^r$ to bound the integral on the vertical line $\Re(s)=\epsilon$ by $O(T^{r-1}\log(T))$, and the convexity bound $f(\sigma+iT)\ll T^{r(1-\sigma)}\log(T)$ to bound the horizontal integral from $\epsilon+iT$ to $c+iT$ by $O(\frac{x\log(T)}{T^2\log(x/T^r)})$. (For $r<1$ the first error term would be $O(1)$). Putting it all together with a choice of $T=\left(\frac{x}{2}\right)^{1/r}$ gives $$\label{e:Bx}
xB_f(x)=F(x)+O(x^{2-\frac{1}{r}}\log(x)),$$ where $F(x)=\frac{ax^2}{2}+\sum_j x^{1+\rho_j}\tilde{p}_j(\log(x)$. Finally, since $\int_{1}^xA_f(t)dt=xB_f(x)$ and $A_f(x)$ is increasing we get that implies $$A_f(x)=F'(x)+O(x^{1-\frac{1}{2r}}\log(x)),$$ concluding the proof.
\[l:ff\*\] Let $f(s)$ be as above. For a large parameter $x\geq1$ and $k\in {\mathbb N}$ let $f^*(s)=f^*_{x,k}(s)$ be defined by the finite series $$\begin{aligned}
\label{e:f*}
f^*(s)&=&\sum_{\lambda_n\leq x}a_n(1-\frac{\lambda_n}{x})^k\lambda_n^{-s}$$ Let $0<\sigma_0<1$ and $k>r$. Then for any $\sigma_0\leq \sigma\leq 3/2$ and $x^{\frac{1-\sigma_0}{k+1}}\leq t\leq x^{\frac{\sigma_0}{r}}$ we have $$\label{e:f*f}f_{x,k}^*(s)=f(s)+O(1).$$
For any $c>1$ we can write $$f^*(s)=\frac{k!}{2\pi i}\int_{c-i\infty}^{c+i\infty} \frac{x^{z-s}f(z)}{(z-s)(z-s+1)\cdots(z-s+k)}dz.$$ Shifting the contour of integration to $\Re(z)=\epsilon$ (and taking $\epsilon\to 0$) the pole at $z=s$ will contribute $f(s)$ (if $\sigma>1$ the pole at $z=s-1$ will contribute $-\frac{kf(s-1)}{x}=O(t^{r}x^{-1})$ and if $\sigma=1$ we avoid the pole on the imaginary axis by integrating over half a circle centered at $s-1$ with some small fixed radius). The contribution of a pole $\rho_j\in(0,1)$ of $f$ with multiplicity $1+n_j$ is bounded by $\frac{x^{\rho_j-\sigma}(\log(x))^{n_j}}{|\rho_j-s||\rho_j-s+1|\cdots|\rho_j-s+k|}=O(\frac{x^{1-\sigma}}{t^{k+1}})$ which also bounds the contribution of the simple pole at one. Finally, using and the fact that $k>r$ we can bound the remaining integral by $O(t^{r}x^{-\sigma})$ giving that $$f_{x,k}^*(s)=f(s)+O(\frac{x^{1-\sigma}}{t^{k+1}})+O(t^{r}x^{-\sigma})+O(t^rx^{-1}),$$ where the last error term occurs only when $\sigma>1$. Consequently, for any $\sigma\geq \sigma_0$ and $x^{\frac{1-\sigma_0}{k+1}}\leq t\leq x^{\frac{\sigma_0}{r}}$ we have $f_{x,k}^*(s)=f(s)+O(1)$ as claimed.
Let $\sigma_1=\frac{4r-1}{4r}$ and fix some $\tfrac{1}{2}\leq \sigma_0<\sigma_1$. For $x=T^{r/\sigma_0}$ and $k>r$ large enough so that $\frac{r(1-\sigma_0)}{(k+1)\sigma_0}<\frac{1}{2r+1}$ let $f^*(s)=f^*_{x,k}(s)$ as in . By Lemma \[l:ff\*\] we have that $f^*(s)=f(s)+O(1)$ for all $\sigma>\sigma_0$ and $T^{\frac{1}{2r+1}}\leq t\leq T$. We can thus bound for all $\sigma> \sigma_0$ $$\begin{aligned}
\int_1^T|f(\sigma+it)|^2&=& \int_1^{T^{\frac{1}{2r+1}}}|f(\sigma+it)|^2+\int_{T^{\frac{1}{2r+1}}}^T |f(\sigma+it)|^2\\
&\ll& \int_{-T}^T |f^*(\sigma+it)|^2+O(T),\end{aligned}$$ where we used to bound $$\int_1^{T^{\frac{1}{2r+1}}}|f(\sigma+it)|^2\ll \int_1^{T^{\frac{1}{2r+1}}}t^{2r}dt\ll T.$$
Next, using that $1\ll\frac{\sin(x)}{x}$ for $|x|\leq\tfrac{1}{2}$ we bound $$\begin{aligned}
\int_{-T}^T |f^*(\sigma+it)|^2dt &\ll & \int_{-T}^T|f^*(\sigma+it)|^2\left(\frac{2T}{t}\sin(\frac{t}{2T})\right)^2dt\\
&\leq & \int_{-\infty}^\infty|f^*(\sigma+it)|^2\left(\frac{2T}{t}\sin(\frac{t}{2T})\right)^2dt\\
&=& 2T\sum_{\lambda_n, \lambda_m<x}\frac{a_n^*a_m^*}{(\lambda_n\lambda_m)^\sigma}\int_{-\infty}^\infty \left(\frac{\sin(t)}{t}\right)^2e^{2iTt\log(\frac{\lambda_m}{\lambda_n})}dt\\
&\ll& T\mathop{\sum_{\lambda_n, \lambda_m<x}}_{|\log(\frac{\lambda_m}{\lambda_n})|\leq \frac{1}{T}}\frac{a_n^*a_m^*}{(\lambda_n\lambda_m)^\sigma}\left(1-T|\log(\frac{\lambda_m}{\lambda_n})|\right),\end{aligned}$$ where $a_n^*=a_n(1-\frac{\lambda_n}{x})^k\leq a_n$. Using the condition that $|\log(\frac{\lambda_m}{\lambda_n})|\leq \frac{1}{T}$ we may restrict the double sum to an outer sum over $\lambda_n\leq x$ and an inner sum over $\lambda_ne^{-1/T}\leq \lambda_m\leq \lambda_ne^{1/T}$ so $$\begin{aligned}
\int_{-T}^T |f^*(\sigma+it)|^2dt&\ll & T\sum_{\lambda_n\leq x}\frac{a_n}{\lambda_n^{\sigma}} \left(\sum_{e^{-1/T}\leq \frac{\lambda_m}{\lambda_n}\leq e^{1/T}}\frac{a_m}{\lambda_m^\sigma}\right).\\\end{aligned}$$ Now use and summation by parts to bound the inner sum by $$\begin{aligned}
\sum_{\lambda_ne^{-1/T}\leq \lambda_m\leq \lambda_ne^{1/T}}\frac{a_m}{\lambda_m^\sigma}\ll \frac{\lambda_n^{1-\sigma}}{T}+\log(\lambda_n)\lambda_n^{\frac{2r-1}{2r}-\sigma}\end{aligned}$$ hence $$\begin{aligned}
\int_{-T}^T |f^*(\sigma+it)|^2dt &\ll & \sum_{\lambda_n\leq x}\frac{a_n}{\lambda_n^{2\sigma-1}}+T\sum_{\lambda_n\leq x}\frac{a_n\log(\lambda_n)}{\lambda_n^{1+2(\sigma-\sigma_1)}}\\
&\ll&x^{2(1-\sigma)}+T\min\{ \frac{1}{(\sigma-\sigma_1)^2},\log^2(x)\}\end{aligned}$$ where we used and summation by parts to bound the first term by $O(x^{2(1-\sigma)})$, and bounded the sum in the second term by $f'(1+2(\sigma-\sigma_1))\ll \frac{1}{(\sigma-\sigma_1)^2}$ when $\sigma>\sigma_1$ and by $\log(x)\sum_{\lambda_n\leq x}\frac{a_n}{\lambda_n}\ll \log^2(x)$ in general. Recalling that $x=T^{r/\sigma_0}$, for any $\sigma\geq \sigma_1$ we can bound the first term by $x^{2(1-\sigma)}\leq T^{\frac{2r(1-\sigma_1)}{\sigma_0}}\leq T$ while the second term is bounded by $O(T\min\{\frac{1}{(\sigma-\sigma_1)^2},\log^2(T)\}$.
Zeroes of scattering determinant
--------------------------------
In order to apply the above results on Dirichlet series we express the scattering coefficients as certain Dirichlet series with positive coefficients. Explicitly, $$\label{e:phiLij}
\phi_{ij}(s)=c_{j} \frac{\Gamma(s-\frac{d+1}{2})}{\Gamma(s)}L_{ij}(s),$$ where $L_{ij}(s)$ is a Dirichlet series of the form $$\label{e:Lij}
L_{ij}(s)=\sum_{n=0}^\infty \frac{a_{ij}(n)}{\lambda_{ij}(n)^s},$$ with $a_{ij}(n)\in {\mathbb N}$ and $\lambda_{ij}(n)>0$, converging in the half plane $\Re(s)>d-1$ with a simple pole at $s=d-1$. Since sums and products of Dirichlet series is also a Dirichlet series, the scattering determinant has an expression of the form $$\label{e:phitoL}
{{\varphi}}(s)=\left(\frac{\Gamma(s-\tfrac{d-1}{2})}{\Gamma(s)}\right)^{\kappa}L(s),$$ where $L(s)$ is another a Dirichlet series with real but not necessarily positive coefficients, and the zeroes of ${{\varphi}}(s)$ in the half plane $\Re(s)>\tfrac{d-1}{2}$ are the same as the zeroes of $L(s)$. We can also rewrite it as $$\label{e:Ltof}
L(s)=ab^{d-1-2s}L^*(s)$$ for some $a,b>0$ with $$\label{e:Lseries}
L^*(s)=1+\sum_{n=1}^\infty\frac{a_n}{\lambda_n^s}$$ with all $a_n\in {\mathbb R}$ and $1<\lambda_1<\lambda_2<\ldots$
The fact that the scattering coefficients can be expressed as such Dirichlet series is crucial for understanding the distribution of the poles. It seems that when there is no such expression (e.g., for asymptotically hyperbolic manifolds) the pole distribution is different (see [@Muller92]). Even though this result is well known to experts, since we did not find a proof for a general hyperbolic manifold in the literature we include a proof in an appendix.
We now give a few estimates on $L^*(s)$ and then apply the general results on Dirichlet series to prove Theorem \[t:ZeroDist1\].
\[p:L\*\] The function $L^*(s)$ is holomorphic in $\Re(s)>\frac{d-1}{2}$ except for finitely many poles in $(\tfrac{d-1}{2},d-1]$ and satisfies there $$\label{e:fboundsigma}
|L^*(\sigma+it)|=1+O(e^{-c\sigma}),$$ for some $c>0$ and all $\sigma>>1$ sufficiently large, $$\label{e:fboundt}
L^*(\sigma+it)=O(|t|^{(d-1)\frac{\kappa}{2}}),$$ for $\sigma\geq\frac{d-1}{2}$ and $t>>1$ sufficiently large, and $$\label{e:fcritical}
|L^*(\tfrac{d-1}{2}+it)|=a_\Gamma \left|\frac{\Gamma(\tfrac{d-1}{2}+it)}{\Gamma(it)}\right|^\kappa.$$
The first part follows from the holomorphic extension of ${{\varphi}}(s)$. The bound follows from the expansion which absolutely converges for $\sigma>d-1$ (one can take the constant $c=\log(\lambda_1)$).
Next, and give $$\label{e:ftophi}
L^*(s)=(ac)^{-1}b^{2s+1-d}\left(\frac{\Gamma(s)}{\Gamma(s-\tfrac{d-1}{2})}\right)^\kappa {{\varphi}}(s),$$ and since $|{{\varphi}}(\tfrac{d-1}{2}+it)|=1$ we get .
Finally to show , for $Y>0$ sufficiently large (but fixed) let $$E_i^Y(z,s)=\left\lbrace\begin{array}{cc} E_i(z,s) & y_j(z)<Y,\; j=1,\ldots,\kappa\\
E_i(z,s)-\delta_{ij}y_j^s-\phi_{ij}(s)y_j^{d-1-s} & y_j(z)\geq Y
\end{array}\right.,$$ and let $E^Y(z,s)$ denote the column vector with components $E_i^Y(z,s)$. From the Maass-Selberg relations (see [@SelbergHarmonic Equation (7.44)] and [@CohenSarnak80 1.62]) we get the matrix equation $$\begin{aligned}
\int_{{\mathcal{F}}_{\Gamma}}E^Y(z,s)E^Y(z,s)^*dz&=&\frac{1}{2\sigma-d-1}(Y^{2\sigma+1-d}I-Y^{d-1-2\sigma}\phi(s)\phi(s)^*)\\
&&+\frac{\phi(s)^*Y^{2it}-\phi(s)Y^{-2it}}{2it},\end{aligned}$$ where $s=\sigma+it$ and $I$ is the identity matrix. Since the matrix on the left hand side is positive so is the matrix on the right, implying that $$\phi(s)\phi(s)^*\leq Y^{4\sigma-2(d-1)}\left(\sqrt{1+(\tfrac{2\sigma+1-d}{2t})^2}+\frac{2\sigma+1-d}{2t}\right)^2I.$$ Taking the trace gives for $\tfrac{d-1}{2}\leq\sigma\leq d$ the uniform bound $$\label{e:vphibound}
|\phi_{ij}(\sigma+it)|\ll \left(\sqrt{1+(\tfrac{2\sigma+1-d}{2t})^2}+\frac{2\sigma+1-d}{2t}\right).$$ In particular, ${{\varphi}}(\sigma+it)=O(1)$ for $\sigma\in (\tfrac{d-1}{2},d)$ and $|t|>1$. Combining this with for $\tfrac{d-1}{2}<\sigma\leq d$ gives the bound $$|L^*(\sigma+it)|\ll \left|\frac{\Gamma(\sigma+it)}{\Gamma(\sigma-\tfrac{d-1}{2}+it)}\right|^\kappa\ll |t|^{\frac{(d-1)\kappa}{2}}.$$ Since $L^*(\sigma+it)=O(1)$ for $\sigma\geq d$ (from the series expression) this concludes the proof of .
\[l:intL\*half\] There are constants $B_{\Gamma},C_{\Gamma}$ depending on ${\Gamma}$ such that $$\begin{aligned}
\lefteqn{\frac{1}{2\pi}\int_{-T}^T(T-|t|)\log|L^*(\tfrac{d-1}{2}+it)|dt=}\\&&
\tfrac{\kappa(d-1)}{4\pi}T^2\log T+B_{\Gamma}T^2+C_{\Gamma}T +O(\log(T)).\end{aligned}$$
Let $d-1=2m+\nu$ with $\nu\in\{0,1\}$, and expand $$\Gamma(\tfrac{d-1}{2}+it)=\prod_{j=1}^m(\tfrac{d-1}{2}-j+it)\Gamma(\tfrac{\nu}{2}+it).$$ Using we evaluate the integral $$\begin{aligned}
\label{e:intlogf}
\frac{1}{2\pi}\int_{-T}^T(T-|t|)\log|L^*(\tfrac{d-1}{2}+it)|dt=\frac{\log(a_\Gamma)}{2\pi}T^2 \\
\nonumber +\frac{\kappa}{\pi}\sum_{j=1}^{m}\int_{0}^T(T-t)\log|it+\tfrac{d-1}{2}-j|dt\\ \nonumber
+\frac{\kappa}{\pi}\int_{0}^T(T-t)\log|\frac{\Gamma(\tfrac{\nu}{2}+it)}{\Gamma(it)}|dt.\end{aligned}$$ The second term on the right hand side of can be evaluated as $$\begin{aligned}
\lefteqn{
\frac{\kappa}{\pi}\sum_{j=1}^{m}\int_{0}^T(T-t)\log|it+\tfrac{d-1}{2}-j|dt=} \\
&&=\frac{\kappa m}{\pi}(\frac{T^2}{2}\log T-\frac{3T^2}{4})
+\frac{\kappa}{2\pi}\sum_{j=1}^{m}\int_0^T(T-t)\log(1+\frac{(\tfrac{d-1}{2}-j)^2}{t^2})dt\\
&&=\frac{\kappa m}{2\pi}T^2\log T-\frac{3\kappa m}{4\pi}T^2+\frac{\kappa m(d-m-2)}{8}T+O(\log(T)),\end{aligned}$$ and the last term is given by $$\begin{aligned}
\lefteqn{
\frac{\nu\kappa}{\pi}\int_{0}^T(T-t)\log|\frac{\Gamma(\tfrac{1}{2}+it)}{\Gamma(it)}|dt
=\frac{\nu\kappa}{2\pi}\int_{0}^T(T-t)\log|\frac{t\tanh(\pi t)}{\pi}|dt}\\
&&=\frac{\nu\kappa}{4\pi}T^2\log(T)-(\frac{3\nu\kappa}{8\pi}+\frac{\nu\kappa\log(\pi)}{4\pi})T^2-
\frac{\nu\kappa}{16}T+O(1).\end{aligned}$$ Plugging these back in proves the claim with $B_{\Gamma}=\tfrac{4\log(a_\Gamma)-\kappa(d-1+2\nu\log\pi)}{8\pi}$ and $C_{\Gamma}=\tfrac{\kappa(2m(m+\nu-1)-\nu)}{16}$.
\[l:boundintL\*\] For $\alpha\geq\alpha_0=d-\frac{5}{4}$ we have $$\begin{aligned}
\frac{1}{T}\int_{-T}^T\log|L^*(\alpha+it)|dt\ll \min\{\log\left(\tfrac{1}{\alpha-\alpha_0}\right),\log\log(T)\}\end{aligned}$$
First, from and we get that for $|t|<1$ and $\tfrac{d-1}{2}<\alpha<d$ $$\log|L^*(\alpha+it)|\ll 1+\log(1+\frac{1}{t}),$$ so it is enough to bound $\int_1^T \log|L^*(\alpha+it)|dt$. Next, since $L(s)=\det(L_{ij}(s))$ we get the bound $$|L^*(\alpha+it)|\leq \frac{b^{2\alpha+1-d}}{a}\left(\frac{1}{\kappa}\sum_{i,j}|L_{ij}(\alpha+it)|^2\right)^{\kappa/2},$$ and from the inequality between geometric and arithmetic mean we get $$\begin{aligned}
\label{e:boundintL*1}
\frac{1}{T}\int_1^T \log(|L^*(\alpha+it)|)dt\leq (2\alpha+1-d)\log b-\log a \\ \nonumber+\frac{\kappa}{2}\log\left(\frac{1}{T}\int_1^T \frac{1}{\kappa}\sum_{i,j}|L_{ij}(\alpha+it)|^2dt\right).\end{aligned}$$
For each pair $(i,j)$ consider $f(s)=L_{ij}(\tfrac{d-1}{2}(s+1))$. This function is still given by a Dirichlet series with positive coefficients, it converges absolutely for $\Re(s)>1$ with a simple pole at $s=1$, it has an analytic continuation to $\Re(s)\geq 0$ except perhaps finitely many poles in $(0,1)$, and it satisfies that $|f(\sigma+it)|=O(t^{\frac{d-1}{2}})$ for $\sigma\geq 0$ and $t>1$ (by and ). We can thus apply Proposition \[l:boundintf\] with $r=\tfrac{d-1}{2}$ to $L_{ij}(\tfrac{d-1}{2}(s+1))$ and get that for $\alpha\geq \alpha_0$, $$\frac{1}{T}\int_1^T|L_{ij}(\alpha+it)|^2dt\ll \min\{\tfrac{1}{(\alpha-\alpha_0)^2},\log^2(T)\}.$$ Using this bound with gives $$\begin{aligned}
\frac{1}{T}\int_1^T \log(|L^*(\alpha+it)|)dt&\ll&\min\{\log(\tfrac{1}{\alpha-\alpha_0}),\log\log(T)\}\\\end{aligned}$$ as claimed.
\[l:Arg\] For any $\alpha>\frac{d-1}{2}$ we have $$\int_\alpha^\infty\mathrm{arg}(L^*(\sigma+iT))d\sigma =O(\log(T)).$$
Since $L^*(\sigma+iT)=1+O(e^{-c\sigma})$ as $\sigma\to\infty$ we have $\mathrm{arg}(L^*(\sigma+iT))\ll e^{-c\sigma}$, so for $\sigma_1>d$ large enough (but fixed), we have $$\int_\alpha^\infty\mathrm{arg}(L^*(\sigma+iT))d\sigma=\int_\alpha^{\sigma_1}\mathrm{arg}(L^*(\sigma+iT))d\sigma+O(1).$$ For the remaining integral, using the bound $|L^*(\sigma+iT)|=O(T^{\tfrac{d-1}{2}})$ and Titchmarsh’s [@Titchmarsh86 Lemma 9.2] we get that $\mathrm{arg}(L^*(\sigma+iT))=O(\log(T))$ and hence the whole integral is bounded by $O(\log(T))$.
We can now prove Theorem \[t:ZeroDist1\], the argument is almost identical to Selberg’s [@Selberg90] and we include it for the sake of completeness.
The zeroes and poles of ${{\varphi}}(s)$ in $\Re(s)>\tfrac{d-1}{2}$ are the same as the zeroes and poles of $L^*(s)$ in $\Re(s)>\tfrac{d-1}{2}$. By Proposition \[p:L\*\], $L^*(s)$ satisfies all the assumptions needed for [@Selberg90 Lemma 1,2], stating that for any $\alpha\geq\tfrac{d-1}{2}$, $$\begin{aligned}
\label{e:variation}
\mathop{\sum_{|\gamma|\leq T}}_{\beta>\alpha}(T-|\gamma|)(\beta-\alpha)=\frac{1}{2\pi}\int_{-T}^T(T-|t|)\log|L^*(\alpha+it)|dt\\
\nonumber +T\sum_{\sigma_j>\alpha}(\sigma_j-\alpha)+O(\log(T)),\end{aligned}$$ where the last sum is over the finitely many poles in $(\tfrac{d-1}{2},d]$. Let $F(\alpha,T)$ denote the left hand side of , and let $$\label{e:F1}
F_1(\alpha,T)=\mathop{\sum_{|\gamma|\leq T}}_{\beta>\alpha}(\beta-\alpha).$$ One easily sees that $$\label{e:diff}
F(\alpha, T)-F(\alpha, T-1)\leq F_1(\alpha,T)\leq F(\alpha,T+1)-F(\alpha,T).$$ From with $\alpha=\tfrac{d-1}{2}$, together with Lemma \[l:intL\*half\] we get $$\begin{aligned}
F(\tfrac{d-1}{2};T)=\tfrac{\kappa(d-1)}{4\pi}T^2\log T+B_{\Gamma}T^2+(C_{\Gamma}+\sum_{\sigma_j>\alpha}(\sigma_j-\tfrac{d-1}{2})) T +O(\log(T)),\end{aligned}$$ which together with implies that $$F_1(\tfrac{d-1}{2};T)=\tfrac{\kappa(d-1)}{2\pi}T\log T+A_{\Gamma}T+O(\log(T)),$$ with $A_{\Gamma}=2(C_{\Gamma}+\sum_{\sigma_j>\alpha}(\sigma_j-\tfrac{d-1}{2}))+B_{\Gamma}$. This confirms
Next, to show , we use Littlewood’s formula $$\begin{aligned}
\label{e:Littlewood}
\mathop{\sum_{|\gamma|\leq T}}_{\beta>\alpha}(\beta-\alpha)&=&\frac{1}{2\pi}\int_{-T}^T\log|L^*(\alpha+it)|dt+\frac{1}{\pi}\int_\alpha^\infty\mathrm{arg}(L^*(\sigma+iT))d\sigma\\
\nonumber &&+\sum_{\sigma_j>\alpha}(\sigma_j-\alpha).\\
$$ For $\alpha\geq\alpha_0$, we use Lemma \[l:boundintL\*\] to bound the first integral, Lemma \[l:Arg\] to bound the second integral, and bound the sum over the poles by $O(1)$ to get that$$\begin{aligned}
\mathop{\sum_{|\gamma|\leq T}}_{\beta>\alpha}(\beta-\alpha)\ll T\min\{\log(\tfrac{1}{\alpha-\alpha_0}),\log\log(T)\})\end{aligned}$$ confirming .
Scattering determinant as a Dirichlet series
============================================
In this section we verify the formula , expressing the scattering coefficients as Dirichlet series with positive coefficients. Recall the definition of the Eisenstein series, it’s constant terms, and the scattering coefficients given in ,, and . We will show
\[p:scatterDirichlet\] The coefficients of the scattering matrix can be written as $$\label{e:phiLijA}
\phi_{ij}(s)=c_{j} \frac{\Gamma(s-\frac{d+1}{2})}{\Gamma(s)}L_{ij}(s),$$ where $L_{ij}(s)$ is a Dirichlet series of the form $$\label{e:LijA}
L_{ij}(s)=\sum_{n=0}^\infty \frac{a_{ij}(n)}{\lambda_{ij}(n)^s},$$ with $a_{ij}(n)\in {\mathbb N}$ and $\lambda_{ij}(n)>0$, converging in the half plane $\Re(s)>d-1$ with a simple pole at $s=d-1$.
We follow the same proof given in [@SelbergHarmonic] for hyperbolic surfaces, that is, we explicitly compute the terms $y_i(\gamma.z)^s$ appearing in , and then integrate along the cusps. We will need the following explicit formula for the hyperbolic action.
\[l:HypAction\] Let $G\cong{ \mathrm{SO}}_0(d,1)$ denote the group of isometries of hyperbolic space ${\mathbb{H}^{d}}$. For any $g\in G$ there is $\lambda=\lambda(g)\geq 0$ such that for any $z=(x,y)\in {\mathbb{H}^{d}}$ $$\label{e:HypAction}
y(z)=y(g.z)\left\lbrace\begin{array}{cc} \lambda\big(y^2+{\left\Vertx+\eta\right\Vert}^2\big) & \lambda>0\\
\alpha & \lambda=0\end{array}\right.,$$ for some $\eta\in {\mathbb R}^{d-1}$ and $\alpha>0$ (depending only on $g$).
We start with the hyperboloid model for hyperbolic space $$\label{e:HmodelL}
\mathbb{L}^d=\{(\xi_0,\xi_1,\ldots,\xi_d)| \xi_0^2+\ldots \xi_{d-1}^2-\xi_{d}^2=-1,\; \xi_{d}>0\}.$$ In this model the group of isometries is just the group of linear maps sending $\mathbb{L}^d$ onto itself (see [@CannonFloydKenyonParry97 Section 7] for more details on the various models of hyperbolic space and the action of the group of isometries on each). Explicitly, this is the identity component of the group $$\label{e:SOd1}
{ \mathrm{SO}}(d,1)=\{A\in { \mathrm{SL}}_{d+1}({\mathbb R})|A^tJA=J=AJA^t\},$$ with $J=\mathrm{diag}(1,\dots,1,-1)$, acting linearly on $\mathbb{L}^d\subseteq {\mathbb R}^{d+1}$.
In order to see how this action looks like in the upper half space model we use the isometry $\iota:\mathbb{L}^d\to {\mathbb{H}^{d}}$ given by $$\label{e:IsoLH}
\iota(\xi_0,\ldots,\xi_d)=(\frac{2\xi_1}{\xi_0+\xi_d},\ldots, \frac{2\xi_{d-1}}{\xi_0+\xi_d},\frac{2}{\xi_0+\xi_d})=(x,y),$$ with inverse map given by $$\label{e:IsoHL}
\iota^{-1}(x,y)=(\tfrac{1-\frac{1}{4}(y^2+\sum_j x_j^2)}{y},\tfrac{x_1}{y},\ldots,\tfrac{x_{d-1}}{y},\tfrac{1+\tfrac{1}{4}(y^2+\sum_j x_j^2)}{y}).$$
Now fix some $g\in \mathrm{Isom}({\mathbb{H}^{d}})$ and let $A\in { \mathrm{SO}}_0(d,1)$ denote the corresponding linear map acting on ${\mathbb{L}^{d}}$. Given $z=(x,y)\in {\mathbb{H}^{d}}$ let $\xi=\iota^{-1}(z)\in {\mathbb{L}^{d}}$ and $\tilde{\xi}=A\xi$ so that $\iota(\tilde{\xi})=(\tilde{x},\tilde{y})=g.z$. Using we have that $$\tilde{y}=\frac{2}{\tilde{\xi}_0+\tilde{\xi}_d},\quad \tilde{x}_j=\tilde{\xi_j}\tilde{y},$$ and since $A=(a_{i,j})$ acts linearly we can write $$\tilde{\xi}_0+\tilde{\xi}_d=\sum_{j=0}^d \alpha_j\xi_j, \mbox{ with } \alpha_j=a_{0,j}+a_{d,j}.$$ Now use to rewrite $\xi_j$ back in terms of $(x,y)$ to get $$\tilde{\xi}_0+\tilde{\xi}_d=\frac{\alpha_0+\alpha_d}{y}+\frac{\alpha_0-\alpha_d}{4y}(y^2+\sum_{j=1}^{d-1}x_j^2)+\sum_{j=1}^{d-1} \frac{\alpha_j x_j}{y}.$$
We first, consider the case where $\alpha_0\neq \alpha_d$ and we define $\lambda=\frac{\alpha_d-\alpha_0}{8}$ and $\eta_j=\frac{2\alpha_j}{\alpha_d-\alpha_0}$. Completing the squares we get $$\tilde{\xi}_0+\tilde{\xi}_d=\frac{1}{y}\left[(\alpha_d+\alpha_0)-\frac{1}{\alpha_d-\alpha_0}\sum_{j=1}^{d-1}\alpha_j^2+2\lambda\left(y^2+\sum_{j=1}^{d-1}(x_j+\eta_j)^2\right) \right].$$ A direct computation using the fact that $A\in { \mathrm{SO}}(d,1)$ shows that $$(\alpha_d+\alpha_0)=\frac{1}{\alpha_d-\alpha_0}\sum_{j=1}^{d-1}\alpha_j^2,$$ hence, $$\tilde{y}=\frac{2}{\tilde{\xi}_0+\tilde{\xi}_d}=\frac{y}{\lambda\left(y^2+{\left\Vertx+\eta\right\Vert}^2\right)}.$$ Note that $\lambda$ depends only on $A$ (and hence only on $g$) and not on $(x,y)$, in particular, the fact that $\tilde{y}>0$ implies that $\lambda>0$ as well.
Next, consider the case where $\alpha_0= \alpha_d=\alpha$ so that $$\tilde{\xi}_0+\tilde{\xi}_d=\frac{2\alpha}{y}+\sum_{j=1}^{d-1} \frac{\alpha_j x_j}{y}.$$ Using again that $\tilde{y}>0$ we must have that $\alpha>0$ and that $\alpha_j=0$ for $j=1,\ldots, d-1$, (otherwise one can always choose $x_j$ to make $\tilde{y}=\frac{2}{\tilde{\xi}_0+\tilde{\xi}_d}<0$). We thus see that $\tilde{\xi}_0+\tilde{\xi}_d=\frac{2\alpha}{y}$, hence, $\tilde{y}=y/\alpha$ concluding the proof.
With this formula we can compute the scattering matrix and express it in terms of Dirichlet series following the same argument as [@SelbergHarmonic].
We first consider the top left coefficient $\phi_{11}$, given by $$\label{e:constant}
\frac{1}{{\mathrm{vol}}(\Gamma_N{\backslash}N)}\int_{\Gamma_N{\backslash}N}E_1(s,n.z)dn=y^s+\phi_{11}(s)y^{d-1-s}.$$ We note that $\tau\in \Gamma_N$ acts on $z=(x,y)\in {\mathbb{H}^{d}}$ via $(x,y)\mapsto(x+u_\tau,y)$ and that the set, $L_1\subset {\mathbb R}^{d-1}$, of all $u_\tau$’s occurring in this way is a lattice in ${\mathbb R}^{d-1}$. Equation can be written explicitly as $$\label{e:constant1}\
\frac{1}{v_1}\int_{{\mathcal{F}}_1}E(s,(x,y))dx=y^s+\phi_{11}(s)y^{d-1-s},$$ where ${\mathcal{F}}_1\subseteq {\mathbb R}^{d-1}$ is a fundamental domain for $L_1{\backslash}{\mathbb R}^{d-1}$ and $v_1={\mathrm{vol}}({\mathcal{F}}_1)$. Using the explicit formula for $y(\gamma.z)$ we can compute this integral directly (for $\Re(s)>d-1$) as $$\begin{aligned}
\label{e:explicit1}\nonumber
\int_{{\mathcal{F}}_{1}}E(s,(x,y))dx &=& y^s v_1\big(\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma}}_{\lambda(\gamma)=0}\frac{1}{\alpha(\gamma)}\big)\\
&&+\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma}}_{\lambda(\gamma)>0}\frac{y^s}{\lambda(\gamma)^s}\int_{{\mathcal{F}}_1}\frac{dx}{(y^2+{\left\Vertx+\eta\right\Vert}^2)^s}\end{aligned}$$ For any $\gamma\in \Gamma_P{\backslash}\Gamma$ with $\lambda(\gamma)>0$ and $\tau\in \Gamma_N$ we have that $\gamma$ and $\gamma\tau$ are distinct classes in $\Gamma_P{\backslash}\Gamma$ with $\lambda(\gamma)=\lambda(\gamma\tau)$. Indeed, a direct computation using shows that $y(\gamma.z)=y(\gamma\tau.z)$ for all $z$ if and only if $\tau=1$, hence, $\gamma\tau\gamma^{-1}\not\in \Gamma_P$ when $\tau\neq 1$; To see that $\lambda(\gamma)=\lambda(\gamma\tau)$ use to get the identity $$\begin{aligned}
\frac{1}{\lambda(\gamma \tau)}&=&y(z)\cdot y(\gamma\tau.z)=\frac{y^2}{\lambda(\gamma)(y^2+{\left\Vertx+u_\tau+\eta(\gamma)\right\Vert}^2)}\\
&=&\frac{1}{\lambda(\gamma)(1+y^{-2}{\left\Vertx+u_\tau+\eta(\gamma)\right\Vert}^2)}\end{aligned}$$ and take $y\to\infty$.
We can thus write the second sum in as $$\begin{aligned}
\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N}}_{\lambda(\gamma)>0}\sum_{u\in L_1}\frac{y^s}{\lambda(\gamma)^s}\int_{{\mathcal{F}}_1}\frac{dx}{(y^2+{\left\Vertx+u+\eta\right\Vert}^2)^s}\\
=\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N}}_{\lambda(\gamma)>0}\frac{y^s}{\lambda(\gamma)^s}\int_{{\mathbb R}^{d-1}}\frac{dx}{(y^2+{\left\Vertx\right\Vert}^2)^s}\\
=\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N}}_{\lambda(\gamma)>0}\frac{y^{d-1-s}}{\lambda(\gamma)^s}\int_{{\mathbb R}^{d-1}}\frac{dx}{(1+{\left\Vertx\right\Vert}^2)^s}\\
=c(d)\frac{\Gamma(s-\frac{d-1}{2})}{\Gamma(s)}\mathop{\sum_{\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N}}_{\lambda(\gamma)>0}\frac{y^{d-1-s}}{\lambda(\gamma)^s}\\\end{aligned}$$ Consider the set $$\Lambda_{11}=\{\lambda(\gamma)|\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N\}\cap (0,\infty),$$ from the discreteness of ${\Gamma}$ we get that $\Lambda_{11}\subset (0,\infty)$ is discrete and we can order it $$\begin{aligned}
\Lambda_{11}=\{\lambda_{11}(0)<\lambda_{11}(1)<\lambda_{11}(2)<\cdots\},\end{aligned}$$ and let $$a_{11}(n)=\#\{\gamma\in \Gamma_P{\backslash}\Gamma/\Gamma_N|\lambda(\gamma)=\lambda_{11}(n)\}.$$ With these notation we get that $$\label{e:explicit2}
\frac{1}{v_1}\int_{{\mathcal{F}}_1}E_1(s,z)dx= y^s \big(\mathop{\sum_{\gamma\in \Gamma_\infty{\backslash}\Gamma}}_{\lambda(\gamma)=0}\frac{1}{\alpha(\gamma)}\big)+
\frac{c(d)}{v_1}\frac{\Gamma(s-\frac{d-1}{2})}{\Gamma(s)}\sum_{n=0}^\infty \frac{a_{11}(n)}{\lambda_{11}(n)^s}$$ Comparing with we see that $$\mathop{\sum_{\gamma\in \Gamma_\infty{\backslash}\Gamma}}_{\lambda(\gamma)=0}\frac{1}{\alpha(\gamma)}=1,$$ and that $$\phi_{11}(s)=c_1\frac{\Gamma(s-\frac{d-1}{2})}{\Gamma(s)}L_{11}(s),$$ with $c_1=\frac{c(d)}{v_1}$. Finally, since the Eisenstein series $E_1(s,z)$ absolutely converges for $\Re(s)>d-1$ and has a simple pole at $s=d-1$, the series $L_{11}(s)$ also absolutely converges in this region and has a simple pole at $s=d-1$.
The formula for the other coefficients $\phi_{ij}(s)$ follows from the same arguments where we denote by $\lambda_{ij}(\gamma)=\lambda(k_i^{-1}\gamma k_j)$ and let $$\begin{aligned}
\Lambda_{ij}&=&\{\lambda_{ij}(\gamma)>0|\gamma\in \Gamma_{P_i}{\backslash}\Gamma/\Gamma_{N_j}\}\\
&=&\{\lambda_{ij}(0)<\lambda_{ij}(1)<\lambda_{ij}(2)<\cdots\},\end{aligned}$$ and $a_{ij}(n)=\#\{\gamma\in \Gamma_{P_i}{\backslash}\Gamma/\Gamma_{N_j}|\lambda_{ij}(\gamma)=\lambda_{ij}(n)\}$. Note that the fact that the cusps are distinct implies that when $i\neq j$ we have that $\lambda_{ij}(\gamma)>0$ for all $\gamma\in \Gamma$ and the first sum in vanishes.
[CFKP97]{}
Chandrasheel Bhagwat and C. S. Rajan. On a multiplicity one property for the length spectra of even dimensional compact hyperbolic spaces. , 131 2339–2344, 2011.
James W. Cannon, William J. Floyd, Richard Kenyon, and Walter R. Parry. Hyperbolic geometry. In [*Flavors of geometry*]{}, volume 31 of [*Math. Sci. Res. Inst. Publ.*]{}, pages 59–115. Cambridge Univ. Press, Cambridge, 1997.
P. Cohen and P. Sarnak. Notes on the Selberg trace formula, Stanford University, 1980,
Yasuro Gon and Jinsung Park. Ruelle zeta function for odd dimensional hyperbolic manifolds with cusps. , 84(1):1–4, 2008.
Yasuro Gon and Jinsung Park. The zeta functions of [R]{}uelle and [S]{}elberg for hyperbolic manifolds with cusps. , 346(3):719–767, 2010.
H. Huber. Zur analytischen [T]{}heorie hyperbolischen [R]{}aumformen und [B]{}ewegungsgruppen. **138** 1–26, 1959.
D. [Kelmer]{}. . , 366: 5925–5961 ,2014.
Werner M[ü]{}ller. The trace class conjecture in the theory of automorphic forms. , 130(3):473–529, 1989.
Werner M[ü]{}ller. Spectral geometry and scattering theory for certain complete surfaces of finite volume. 109, 265-–305, 1992.
Atle Selberg. . Springer-Verlag, Berlin, 1989.
Atle Selberg. Remarks on the distribution of poles of [E]{}isenstein series. In [*Festschrift in honor of [I]{}. [I]{}. [P]{}iatetski-[S]{}hapiro on the occasion of his sixtieth birthday, [P]{}art [II]{} ([R]{}amat [A]{}viv, 1989)*]{}, volume 3 of [*Israel Math. Conf. Proc.*]{}, pages 251–278. Weizmann, Jerusalem, 1990.
E. C. Titchmarsh. . The Clarendon Press Oxford University Press, New York, second edition, 1986.
Garth Warner. Selberg’s trace formula for nonuniform lattices: the [$R$]{}-rank one case. In [*Studies in algebra and number theory*]{}, volume 6 of [*Adv. in Math. Suppl. Stud.*]{}, pages 1–142. Academic Press, New York, 1979.
[^1]:
[^2]: The Riemann hypothesis implies that all the zeroes are on that line.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, based on a fuzzy entropy feature selection framework, different methods have been implemented and compared to improve the key components of the framework. Those methods include the combinations of three ideal vector calculations, three maximal similarity classifiers and three fuzzy entropy functions. Different feature removal orders based on the fuzzy entropy values were also compared. The proposed method was evaluated on three publicly available biomedical datasets. From the experiments, we concluded the optimized combination of the ideal vector, similarity classifier and fuzzy entropy function for feature selection. The optimized framework was also compared with other six classical filter-based feature selection methods. The proposed method was ranked as one of the top performers together with the Correlation and ReliefF methods. More importantly, the proposed method achieved the most stable performance for all three datasets when the features being gradually removed. This indicates a better feature ranking performance than the other compared methods.'
author:
-
bibliography:
- 'sample.bib'
title: 'Performance Optimization of a Fuzzy Entropy based Feature Selection and Classification Framework\'
---
Introduction
============
Due to the rapid development and wide application of information technology, increasing amount of data with rich information are generated. Discovering the information that concealed in these datasets becomes essential and challenging. In real world applications, datasets often contain irrelevant and redundant features that do not provide useful or additional information for subsequent decision makings [@kumar2014feature]. According to Occam’s razor, it is important and necessary to eliminate those irrelevant and redundant features [@domingos1999role]. Therefore, feature selection has always been an active research area, particularly when more and more ’big data’ become available in many application areas.
Recently, machine learning methods have achieved superior performance in many application areas such as diagnostic decision making and disease classification [@kononenko2001machine]. However, the high dimensionality and complexity of the data often make the methods suffering from the problem of curse of dimensionality [@bellman2013dynamic]. With the high dimensionality, the limited number of training samples are sparsely distributed in the feature space. This makes the machine learning methods difficult to learn the underlying relationship accurately. This also leads to an over-fitting problem that the learned model is not generic enough for unseen data samples. In addition, high dimensional datasets significantly increase the memory usage and computational cost for data analysis, which result in low efficiency of algorithms [@li2016feature].
Dimensionality reduction methods aim to deal with the aforementioned issues, which are mainly classified into feature extraction and feature selection methods. The feature extraction methods tend to project the original high dimensional feature into a new low dimensional feature space, where the contributions of features are combined. In contrast, feature selection methods maintain a subset of the original features that are highly relevant to the subsequent decision makings [@li2016feature]. This provides the model with better readability and interpretability, which are essential for our main application area (biomedical dataset). Our proposed method belongs to the feature selection category.
Practical biomedical datasets are usually imperfect which contain uncertain texts and incomplete features. The uncertainty will increase after applying some data analysis processes [@bandemer2012fuzzy]. Fuzzy logic algorithms are designed to model the vagueness, imprecision and uncertainty. In order to overcome the practical problems embedded in the biomedical datasets, it becomes a natural choice to integrate fuzzy logic methods into the feature selection process.
Various fuzzy methods have been proposed for feature selection. In 1999, Rezaee [@rezaee1999fuzzy] presented a method to automatically identify the reduced linguistic fuzzy set of a labeled multi-dimensional dataset. The optimal subset of fuzzy features is determined by projecting the original data set onto a fuzzy space. In 2002, Rui-Ping Li [@li2002fuzzy] proposed a fuzzy neural network method for pattern classification and feature selection. The proposed neural network attempts to select the important features from the original features while maintaining the maximum recognition rate. In 2008, Tsang [@tsang2008attributes] introduced a concept of attributes reduction with fuzzy rough sets and developed an algorithm using discernibility matrix to compute all the attributes reductions. More recently, Luukka [@luukka2011feature] introduced a fuzzy entropy feature selection framework based on a maximal similarity classifier. The framework is computationally efficient, readily comprehensible and easily adapted to different applications compared with other fuzzy feature selection methods. The original framework consists of three fundamental components, namely ideal vector calculation, similarity measurement and fuzzy entropy calculation. Different measurements can be used in these components, which affect the feature selection performance. To the authors’ best knowledge, a comprehensive comparison using different measurements within these components has not been reported, so as the comparison to other feature selection methods.
In this paper, we implemented three different measurements for each of the key components in the framework and comprehensively compared the performances based on different combinations of the measures. Additionally, it is also compared with other six feature selection methods in the literature. All the evaluations were tested based on three widely used publicly available biomedical datasets [@Lichman2013].
Methodology
===========
Based on the method in [@luukka2011feature], we proposed a data driven framework to deal with feature selection and classification. The overall structure is illustrated in Fig. \[flowchartFS\].
![Overview of the proposed framework. Blue dotted line and red solid line are the data flows for training and testing processes respectively.[]{data-label="flowchartFS"}](Figures/FeatureSelDiag.png){width="45.00000%"}
The proposed method aims to classify a total number of $M$ subjects into $N$ different classes $C_k, k\in[1, N]$ by their feature vector $\vec{x_i}$. $i$ is the index of the subjects, and the number of features in $\vec{x_i}$ is denoted by D. The procedure of the proposed method is described as follows.
1. For the training set, normalize each feature value to the range of \[0 1\] using min-max normalization method [@jain2011min]. The maximum value of each feature needs to be carefully determined to avoid using outliers and consistently applied to both training and testing datasets.
2. Based on the normalized values from Step 1, calculate the ideal vector $\vec{v}_k$ for the $k^{th}$ class.
3. Apply the same normalization process in Step 1 to the testing set.
4. Calculate the fuzzy similarity between the feature vector $\vec{x_i}$ of the testing subjects and the ideal vector $\vec{v}_k$ obtained in Step 2.
5. Based on the similarity values in Step 4, construct a $MN \times D$ similarity matrix. Calculate the fuzzy entropy value of each feature (column of the matrix) and subsequently rank the features according to the values.
6. Select the features and classify the testing set based on the ranked feature sequence from Step 5.
The detailed descriptions of ideal vector calculation, similarity measurement, fuzzy entropy calculation and classification are given in the following subsections.
**Ideal vector calculation based on training dataset** {#Idealvector}
------------------------------------------------------
Ideal vector is used to represent the “mean” property of the subjects in each class. Different methods of calculating the ideal vector, namely arithmetic mean, geometric mean and harmonic mean have been implemented and compared as follows. In equations (1-3), $N_k$ is the number of subjects for the $k^{th}$ class. $j$ is the index of features. The remaining notations are the same as previously introduced. The performance comparing different ideal vector calculations is reported in section \[expCombinations\].
#### Arithmetic mean {#eqAm}
$$\vec{v}^A_k(j) = \frac{\sum_{i=1}^{N_k}\vec{x_i}(j)}{N_k}, \ j \in [1,D]$$
#### Geometric mean {#eqGm}
$$\vec{v}^G_k(j) = \sqrt[N_k]{\prod_{i=1}^{N_k}\vec{x_i}(j)}, \ j \in [1,D]$$
#### Harmonic mean {#eqHm}
$$\vec{v}^H_k(j) = \frac{N_k}{\sum_{i=1}^{N_k} \frac{1}{\vec{x_i}(j)}}, \ j \in [1,D]$$
**Fuzzy similarity measurement** {#SimilarityMeasures}
--------------------------------
In this section, the similarity measurement is presented in the form of generalized $L$ukasiewicz algebra [@saastamoinen2003testing]. It measures the similarity between the $j^{th}$ element of feature vector $\vec{x_i}$ and the corresponding $j^{th}$ element of the ideal vector in each class. The calculation is described mathematically in equation (\[eqSim\]).
$$\label{eqSim}
Sim \langle \vec{x_i}, \vec{v}_k, j \rangle = \sqrt[p]{1 - |\vec{x_i}(j)^p - \vec{v}_k(j)^p|}$$
$p$ is a hyper parameter which is optimized in section \[expCombinations\]. A similarity value is calculated for each feature in each class for each subject. Subsequently, a $MN \times D$ similarity matrix $\mathbf{P}$ is constructed as shown in Table \[similarityMatrix\]. The fuzzy entropy calculation for each feature is described in the next subsection.
**Data** **Feature 1** **Feature 2** **...** **Feature D**
------------- ----------------------------------------------- ----------------------------------------------- --------- -----------------------------------------------
$\vec{x_1}$ $Sim \langle \vec{x_1}, \vec{v_1}, 1 \rangle$ $Sim \langle \vec{x_1}, \vec{v_1}, 2 \rangle$ ... $Sim \langle \vec{x_1}, \vec{v_1}, D \rangle$
$\vec{x_1}$ $Sim \langle \vec{x_1}, \vec{v_2}, 1 \rangle$ $Sim \langle \vec{x_1}, \vec{v_2}, 2 \rangle$ ... $Sim \langle \vec{x_1}, \vec{v_2}, D \rangle$
... ... ... ... ...
$\vec{x_1}$ $Sim \langle \vec{x_1}, \vec{v_N}, 1 \rangle$ $Sim \langle \vec{x_1}, \vec{v_N}, 2 \rangle$ ... $Sim \langle \vec{x_1}, \vec{v_N}, D \rangle$
$\vec{x_2}$ $Sim \langle \vec{x_2}, \vec{v_1}, 1 \rangle$ $Sim \langle \vec{x_2}, \vec{v_1}, 2 \rangle$ ... $Sim \langle \vec{x_2}, \vec{v_1}, D \rangle$
... ... ... ... ...
$\vec{x_M}$ $Sim \langle \vec{x_M}, \vec{v_N}, 1 \rangle$ $Sim \langle \vec{x_M}, \vec{v_N}, 2 \rangle$ ... $Sim \langle \vec{x_M}, \vec{v_N}, D \rangle$
: Similarities matrix
\[similarityMatrix\]
**Fuzzy entropy based feature selection** {#fuzzyEntropy}
-----------------------------------------
In order to reduce the dimensionality and discard the non-important features, the fuzzy entropy based feature selection process [@luukka2011feature] is used to rank the features. Fuzzy entropy is the basic definition of the fuzzy information process and widely used to measure the degree of vagueness among various areas [@kosko1986fuzzy].
Based on the previously constructed similarity matrix, we calculate the fuzzy entropy value for each feature (each column of matrix $\mathbf{P}$) using the fuzzy entropy functions described below. $\mathbf{P}(r,j)$ is used to represent the value of the $r^{th}$ row and $j^{th}$ column in the similarity matrix. These similarity values are utilized as the membership function of fuzzy set in the fuzzy entropy calculation. Three different fuzzy entropy methods are implemented as expressed in equation (\[eq1\]), (\[eq2\]) and (\[eq3\]).
#### Non Probabilistic Entropy (Luca’s method)
De Luca and Termini [@de1972definition] axiomatized non-probabilistic fuzzy entropy functions and defined a fuzzy entropy measurement based on Shannon’s entropy as below. $$\label{eq1}
\begin{split}
H_1(j) =& - \sum_{r = 1}^{MN} [(\mathbf{P}(r,j) \log \mathbf{P}(r,j)) \\
&+ (1 - \mathbf{P}(r,j))\log (1 - \mathbf{P}(r,j))]
\end{split}$$
#### Weighted Measurement of Fuzzy Entropy (Parkash’s method)
Parkash [@parkash2008new] proposed a new measurement of fuzzy entropy as in equation (\[eq2\]).
$$\label{eq2}
H_2(j) = \sum_{r = 1}^{MN} sin\frac{\pi \mathbf{P}(r,j)}{2} + sin\frac{\pi(1- \mathbf{P}(r,j))}{2} - 1$$
#### Geometry of Fuzzy Set and Entropy (Kosko’s method)
Kosko [@kosko1986fuzzy] utilized the concepts of overlap and underlap to define the fuzzy entropy based on the geometry of hypercube:
$$\label{eq3}
H_3(j)= \frac {\sum_{r = 1}^{MN}(\mathbf{P}(r,j) \land (1-\mathbf{P}(r,j)) )}
{\sum_{r = 1}^{MN}(\mathbf{P}(r,j) \lor (1-\mathbf{P}(r,j)))}$$
Subsequently the fuzzy entropy values are used for feature ranking and selection. We then perform classification based on the selected features in the next subsection.
**Classification based on the selected features** {#classification}
-------------------------------------------------
The classification method is based on the maximal fuzzy similarity measures proposed in [@luukka2001classifier]. Corresponding to the three methods for idea vector calculation, three similarity measurements are implemented here.
#### Similarity measure based on arithmetic mean
$$\label{eqAri}
S^A \langle \vec{x_i}, \vec{v}_k \rangle = \frac{1}{D'} \sum_{j=1}^{D'} \sqrt[p]{1 - |\vec{x_i}(j)^p - \vec{v}_k(j)^p|}$$
#### Similarity measure based on geometric mean
$$\label{eqGeo}
S^G \langle \vec{x_i}, \vec{v}_k \rangle = \sqrt[D']{\prod_{j=1}^{D'} \sqrt[p]{1 - |\vec{x_i}(j)^p - \vec{v}_k(j)^p|} }$$
#### Similarity measure based on harmonic mean
$$\label{eqHar}
S^H \langle \vec{x_i}, \vec{v}_k \rangle = \frac{D'}{ \sum_{j =1}^{D'} \frac{1}{ \sqrt[p]{1 - |\vec{x_i}(j)^p - \vec{v}_k(j)^p|}}}$$
In equation (\[eqAri\]), (\[eqGeo\]) and (\[eqHar\]), $\vec{x_i}$ represents the feature vector of the $i^{th}$ subject in testing set after feature selection. $\vec{v}_k(j)$ stands for the recalculated ideal vector with the reduced dimension in the training set. $D'$ is the number of the selected features. The parameter $p$ is the same as in equation (\[eqSim\]). Each testing subject is then classified into the class that produces the highest similarity value. It is noteworthy to mention that, based on the reduced feature subset, other classifiers can also be applied and compared, e.g. random forest, support vector machine etc. The comparison of different classifiers is not the main focus of this paper.
Experiments
===========
**Materials**
-------------
The proposed method was tested on three publicly available biomedical datasets with binary classifications. Those widely tested datasets were all extracted from real clinical problems with different sample to feature ratios. The key properties of those datasets are shown in Table \[descriptionData\].
**Dataset** **Nb. Features** **Nb. Samples** **Samples/Features**
------------- ------------------ ----------------- ----------------------
WBC 9 699 77.719
WDBC 31 569 18.4
Parkinsons 22 197 9.0
: Description of the biomedical datasets
\[descriptionData\]
### Wisconsin Breast Cancer (WBC)
Wisconsin breast cancer dataset was generated by Dr. Wolberg from his clinical cases. For data preprocessing, the sample code ID and the rows with nan values were removed. Then the number of samples became 683 after the preprocessing. Nine visually assessed features were considered to predict benign or malignant [@wolberg1990multisurface].
### Wisconsin Diagnostic Breast Cancer (WDBC)
The features in Wisconsin Diagnostic Breast Cancer dataset were computed from a digitized image of a fine needle aspirate of a breast mass. The features described characteristics of the cell nuclei presented in the image [@Lichman2013].
### Parkinsons
The dataset, created by Max Little at the University of Oxford, is composed of a range of biomedical voice measurements from healthy people and the people with Parkinson’s disease (PD). The main aim of this data is to discriminate healthy people from those with PD [@little2009suitability].
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
**Evaluation of different combinations of ideal vector calculation and classification methods** {#expCombinations}
-----------------------------------------------------------------------------------------------
The combination of using three ideal vector calculations and three similarity functions for classification were colour coded and listed in Table \[combinations\]. In this experiment, the full sets of features were used for both training and testing without performing feature selection. Without affecting by the feature selection results, this allows a fair comparison of different ideal vector calculations combined with different classification methods, as well as $p$ value (equation (\[eqSim\])) optimization. The methods were tested and compared on three biomedical datasets by evaluating the classification accuracy. The classification accuracy was defined as the number of correctly classified subjects divided by the total number of subjects.
**Ideal vector** **Classification methods** **Name** **Line**
------------------ ---------------------------- ---------- -------------------------------------------------
Arithmetic mean A-A {width="2.00000%"}
Geometric mean A-G {width="2.00000%"}
Harmonic mean A-H {width="2.00000%"}
Arithmetic mean G-A {width="2.00000%"}
Geometric mean G-G {width="2.00000%"}
Harmonic mean G-H {width="2.00000%"}
Arithmetic mean H-A {width="2.00000%"}
Geometric mean H-G {width="2.00000%"}
Harmonic mean H-H {width="2.00000%"}
: Different classifier combinations used[]{data-label="combinations"}
Same as the evaluation in Luukka’s work [@luukka2011feature], the datasets were divided into two halves. One half was used for training and the other half for testing. Additional to the experiment in [@luukka2011feature], we also repeated the experiments for 1000 times for each $p$ value (equation (\[eqSim\])) with random two-half group splitting. Note that all the remaining experiments in this paper for classification accuracy calculation were tested based on the same evaluation mechanism, if not explicitly described. The mean classification accuracy of those combinations on the three biomedical datasets were evaluated and presented in Fig. \[meanAccComparison\].
Fig. \[meanAccComparison\]-a shows the results of the mean classification accuracies for the WBC dataset. In this case, the idea vector calculation using the arithmetic and geometric mean have achieved similar results, which are much higher and more stable than using the harmonic mean. The curves of using the harmonic mean methods vary dramatically when $p$ value is greater than 5.
Fig. \[meanAccComparison\]-b shows the mean classification accuracies for WDBC dataset at different $p$ values. It is seen that different classification methods with the same ideal vector method achieved similar performances. The ideal vector calculation using arithmetic mean and geometric mean methods decreased slowly when $p$ value increased. However, in the case of harmonic mean method, the accuracies increased sharply and peaked at $p=3$. Then the mean classification accuracies decreased slowly along with the other two ideal vector calculation methods.
Fig. \[meanAccComparison\]-c presents the results for the Parkinsons dataset. Arithmetic mean method for calculating the idea vector produced quite stable mean classification accuracy around 0.73. The accuracies of the methods using geometric mean for ideal vector calculation were maximized at the value around 0.78 and then dropped down quickly when $p$ was greater than 2. The harmonic mean methods for ideal vector calculation produced the worst and unstable results.
Overall, the geometric mean method for calculating the idea vector have achieved the maximal classification accuracies when $p$ is around 2 for all of the three datasets. There are not much differences by using the three different similarity functions for classification. Therefore in the following experiments, geometric mean methods were utilized for both ideal vector calculation and maximal similarity classification. The $p$ value in equation (\[eqSim\]) and (\[eqGeo\]) was set to be 2.
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
**Evaluation of different fuzzy entropy methods**
-------------------------------------------------
The aim of this experiment was to compare the feature ranking sequence produced by three different fuzzy entropy methods in section \[fuzzyEntropy\]. The fuzzy entropy values were used to rank the features from high to low. In order to compare different methods, the entropy values were then normalized to the range of \[0 1\] by min-max normalization. For ease of comparison, we randomly chose one method (Luca’s method in this section) as the reference ranking sequence in the horizontal axis of Fig. \[entropyFeatureComparison\]. All the indices of features were sorted according to the reference ranking.
It is observed from Fig. \[entropyFeatureComparison\] that different fuzzy entropy functions produced similar ranking sequences for the three datasets. Luca’s method and Parkash’s method resulted in almost identical ranking sequence for all the datasets. The result from Kosko’s method showed disagreement at multiple points with the other two, especially in Parkinsons dataset.
According to our experiment results, the ranking differences of the three methods did not make a significant impact on the final classification performance. We chose Luca’s method for the fuzzy entropy function in our final framework, as it produced the highest consistency with the other two methods.
**Evaluation on different removing order of fuzzy entropy methods**
-------------------------------------------------------------------
In order to explore the optimal feature selection process, we also compared different feature removal order according to the entropy values. Two different feature selection approaches were compared. One method removed the feature with the highest entropy value each time. The other method removed the feature with the lowest entropy value each time. The mean classification accuracies using the two different feature removal orders for the three datasets are shown in Fig. \[RemovingOrder\].
It is observed from Fig. \[RemovingOrder\] that the method that removed the feature with the lowest entropy value each time maintained a high performance even when half of the features were removed for all three datasets. In contrast, as soon as one feature with the highest entropy value was removed, the performance dropped significantly for all three datasets. Therefore, we concluded that the feature selection approach that eliminated the feature with the lowest entropy value each time should be used. However, this conclusion is contradictory with Luukka’s suggestion in [@luukka2011feature], which removes the feature with the highest fuzzy entropy value each time.
{width="120.00000%"}
{width="120.00000%"}
{width="120.00000%"}
**Comparison with other feature selection methods**
---------------------------------------------------
Based on the previous experiments, we have found the optimized combination of different methods within the proposed feature selection and classification framework. The optimal choice and settings are: geometric mean method for ideal vector calculation and classification function with $p=2$ and Luca’s method for fuzzy entropy calculation. In this section, we compared the proposed method with other six classical filter based feature selection methods in the literature including Chi square based [@jin2006machine], Correlation based [@hall1999correlation] , Gain ratio based [@karegowda2010comparative], Information gain based [@lee2006information], ReliefF based [@liu2007computational] and Symmetrical uncertainty based [@yu2004efficient]. All the filter based feature selection methods rank the features from higher to lower values. The same maximal similarity classifier with geometric method was used for the classification task after the feature selection process for all the compared methods. The mean classification accuracies of the different feature selection methods on three datasets are presented in Fig. \[FSComparison\].
Fig.\[FSComparison\]-(a) has shown that in WBC dataset, the proposed method maintains the highest classification accuracies among all the methods with the number of removed features increasing from 0 to 6. In WDBC dataset (Fig.\[FSComparison\]-(b)), the top two performers are the proposed method and ReliefF method. The classification accuracies keep increasing even when about 20 features have been removed by using the proposed method. For Parkinsons dataset (Fig.\[FSComparison\]-(c)), the proposed method maintained a stable performance with arguably the highest classification accuracies until 14 features being removed.
Another important observation is that the classification accuracies of the proposed method generally follow the trend of gradually increasing, achieving peak performance and decreasing when features were gradually removed for all three datasets. This is a good indication of the features were ranked, from the least to the most importance, reasonably well. However, the performances of other methods changed dramatically when features were gradually removed (Fig. \[FSComparison\]-(b) and Fig. \[FSComparison\]-(c)).
For performance comparison of feature selection methods, it has not been standardized in the literature. One option is to report the highest classification accuracy despite the number of features selected. Alternatively, the classification accuracies are compared based on the same number of selected features. Arguably, if the classification result is more important, the first option should be applied. In this paper, we aims to compare different feature selection methods, where the compactness, representative and relevance of the selected features are more important in this case. Therefore, we adopted the second option and proposed the following comparison criteria.
------------- ----------- ------- ----------- ------- ----------- -------
Acc.(%) Nb. Acc. (%) Nb. Acc. (%) Nb.
Proposed **96.97** **7** 94.86 8 78.23 9
Chi square 95.86 7 93.67 5 77.70 2
Correlation 96.95 7 93.86 4 **78.72** **3**
Gain Ratio 95.83 6 93.73 5 78.09 6
Info. Gain 96.53 7 93.71 5 77.43 2
ReliefF 96.96 7 **95.21** **4** 78.26 8
Sym. Unc. 95.84 7 93.68 5 77.19 3
------------- ----------- ------- ----------- ------- ----------- -------
: Mean classification accuracy of different feature selection methods[]{data-label="comFS"}
We chose the proposed method as the reference method to compare with each of the other competitors. The selected number of features (denoted as $S$) that produced the highest mean classification accuracy of our method was used as the reference. For other methods, the highest mean classification accuracies were reported with the selected number of features less or equal to $S$. For comparison, higher classification accuracy indicates better feature selection performance. Additionally, McNemar’s test [@bennett1970283] was applied to test the statistical significance of the binary classification results for each of the two compared methods.
Methods **WBC** **WDBC** **Parkinsons**
------------- ---------- ---------- ----------------
Chi square $<$0.001 $<$0.001 $<$0.001
Correlation 1.000 $<$0.001 $<$0.001
Gain Ratio $<$0.001 $<$0.001 $<$0.001
Info. Gain $<$0.001 $<$0.001 $<$0.001
ReliefF 1.000 $<$0.001 $<$0.001
Sym. Unc. $<$0.001 $<$0.001 $<$0.001
: $P$ values of McNemar’s test for the pairwise tests between the proposed method and each of the competitors[]{data-label="Pvalue"}
The mean classification accuracies (Acc.) and the selected number of features (Nb.) for the three datasets are listed in Table \[comFS\]. The $P$ values of McNemar’s test for the pairwise tests between the proposed method and each of the competitors are presented in Table \[Pvalue\]. For the results of the WBC dataset in Table IV, it is observed that the proposed method produced the best mean classification accuracy compared with other methods with $S=7$. According to Table V, it is statistically better than the Chi square, Gain Ratio, Info. Gain and Sym. Unc. methods but no statistical differences to the Correlation and ReliefF methods.
For WDBC dataset, the proposed method produced the second best classification accuracy with $S=8$. Each of other methods achieved individually higher performance of using about 4 or 5 features rather than 8 features. However, from Fig. 5-(b), it is seen that the proposed method was still the second best, if 5 features were used (the value corresponding to 26 in the horizontal axes of Fig. 5-(b)). According to Table V, the proposed method is statistically worse than the ReliefF method, but statistically better than all other methods.
For Parkinsons dataset, the proposed method ($S=9$) ranked the third, which was statistically worse than the Correlation (3 features) and RefliefF methods (8 features). The other methods were statistically worse than the proposed method.
Discussion & Conclusion
=======================
In this paper, based on Luukka’s [@luukka2011feature] fuzzy entropy feature selection framework, we have implemented and compared different methods within each of the key components of the framework. They include the combinations of using three ideal vector calculations, three maximal similarity classifiers and three fuzzy entropy functions. All the evaluations were performed on three widely used publicly available biomedical datasets. All these three datasets were generated from challenging clinical applications with different feature to subject ratios. All experiments were thoroughly tested by evenly and randomly splitting the dataset into a training and a testing group, and repeated for 1000 times. From the experiments, we found that the use of geometric method for ideal vector calculation ($p=2$), geometric method for similarity classifier ($p=2$) and Luca’s method for fuzzy entropy calculation achieved the most stable performance and highest classification accuracy. Additionally, we concluded that features with the lowest entropy value should be removed each time to achieve the best performance.
We further compared the proposed method with other six classical filter-based feature selection methods. The mean classification accuracies were compared by fixing the number of selected features. McNemar’s test was also applied to evaluate the statistical differences between the pairwise comparisons. The proposed method produced the highest classification accuracy for WBC dataset, ranked the $2^{nd}$ best and $3^{rd}$ best for WDBC and Parkinsons datasets respectively. Correlation method, ReliefF method and the proposed method are the top performers among the compared methods. More importantly, from the results, it is shown that the proposed method achieved the most stable performance for all three datasets when the features being gradually removed. This indicates a better feature ranking performance.
For future work, we will test our method on different datasets for various applications with more features and more subjects. The robustness of the proposed method that handles outliers and incomplete data will be investigated.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The spin wave spectra of multiferroic BiFeO$_3$ films is calculated using a phenomenological Landau theory that includes magnetostatic effects. The lowest frequency magnon dispersion is shown to be quite sensitive to the angle between spin wave propagation vector and the Néel moment. Since electrical switching of the Néel moment has recently been demonstrated in this material, the sensitivity of the magnon dispersion permits direct electrical switching of spin wave propagation. This effect can be used to construct spin wave logical gates without current pulses, potentially allowing reduced power dissipation per logical operation.'
author:
- Rogerio de Sousa
- 'Joel E. Moore'
title: 'Electrical control of magnon propagation in multiferroic BiFeO$_3$ films'
---
One of the challenges of current research in microelectronic devices is the development of a fast logic switch with minimal power dissipation per cycle. Devices based on spin wave interference [@kostylev05; @khitun05] may provide an interesting alternative to conventional semiconductor gates by minimizing the need for current pulses. Recently, a spin wave NOT gate was demonstrated experimentally [@kostylev05]. The device consisted of a current-controlled phase shifter made by a ferromagnetic (FM) film on top of a copper wire. The application of a current along the wire creates a local magnetic field on the film, leading to a phase shift of its spin waves.
In this letter we predict an effect that allows the design of similar spin wave devices without the need for external current pulses or applied time-dependent magnetic fields. We show that the dispersion of the lowest frequency spin-wave branch of a canted antiferromagnet depends strongly on the direction of spin wave propagation. This occurs because of the long-ranged (dipolar) interactions of the magnetic excitations, which creates a gap for spin waves propagating with non-zero projection along the Néel axis. This effect allows electrical control of spin waves in multiferroic materials that possess simultaneous ferroelectric (FE) and canted antiferromagnetic (AFM) order.
Our model is applicable to the prominent room temperature multiferroic BiFeO$_3$ (BFO) [@wang03]. BFO films have homogeneous AFM order [@bai05; @bea07], in contrast to the inhomogeneous (cycloidal) AFM order present in bulk BFO [@sosnowska82]. The canted AFM order in BFO films is constrained to be in the plane perpendicular to the FE polarization $\bm{P}$. Recently, Zhao [*et al.*]{} [@zhao06] demonstrated room temperature switching of the Neél moment $\bm{L}=\bm{M}_1-\bm{M}_2$ in BFO films after the orientation of the ferroelectric moment was changed electrically. As we show here, spin wave propagation along $\bm{P}$ has high group velocity ($\sim
10^5$ cm/s), in contrast to spin wave propagation along $\bm{L}$ which has zero group velocity at $\bm{k}=0$. Hence switching $\bm{P}$ for a fixed spin wave propagation direction allows electrical control of the spin wave dispersion, which assuming some loss rate will effectively stop long-wavelength spin waves such as those created in [@khitun05].
Although a theory of AFM resonance for canted magnets was developed some time ago[@herrmann63; @tilley82], we are not aware of calculations of spin wave dispersion including magnetostatic effects. The electromagnon spectra for a ferromagnet with quadratic magnetoelectric coupling was discussed without magnetostatic effects in Ref. , and with magnetostatic effects in Ref. . Recently we developed a theory of spin wave dispersion in bulk BFO, a cycloidal (inhomogeneous) multiferroic [@desousa07]. The lowest frequency spin wave mode was shown to depend sensitively on the $\bm{P}$ orientation because of the inhomogeneous nature of the antiferromagnetic order. Interestingly, we show here that BFO films with a homogeneous order display a similar effect, albeit due to a completely different physical reason: the magnetostatic effect.
Our calculation is based on a dynamical Ginzburg-Landau theory for the coupled magnetic and ferroelectric orders. We assume a model free energy given by $$\begin{aligned}
F &=& \frac{a P_{z}^{2}}{2} + \frac{u P_{z}^{4}}{4}+
\frac{a_{\perp}(P_{x}^{2}+P_{y}^{2})}{2}-\bm{P}\cdot \bm{E}\nonumber\\
&&+ \sum_{j=1,2}\left[
\frac{r\bm{M}_{j}^{2}}{2}+\frac{G\bm{M}_{j}^{4}}{4}+
\frac{\alpha \sum_i\left(\nabla M_{ji}\right)^{2}}{2}\right]
\nonumber\\&&+\left(J_0+\eta P^2\right)\bm{M}_{1}\cdot \bm{M}_{2}+ d\bm{P}
\cdot \bm{M}_{1}\times \bm{M}_{2}.
\label{f}\end{aligned}$$ Here $\bm{M}_j$ is the magnetization of one of the two sublattices $j=1,2$, and $\bm{P}$ is a ferroelectric polarization. The coordinate system is such that $\hat{\bm{z}}$ points along one of the cubic (111) directions in BFO. The exchange interaction $J=\left(J_0+\eta
P^2\right)$ is assumed to have a quadratic dependence on $P$ due to magnetostriction. The last contribution to Eq. (\[f\]) is a Dzyaloshinskii-Moriya (DM) interaction, with a DM vector given by $d\bm{P}$. Note that this changes sign under inversion symmetry, hence Eq. (\[f\]) is invariant under spatial inversion at a point in between the two sublattices.
![Spin and polarization waves in a canted multiferroic, such as a BiFeO$_3$ film. The sublattice magnetizations $\bm{M}_1$, $\bm{M}_2$ lie in the plane perpendicular to the FE polarization $\bm{P}$. Fluctuations $\delta\bm{P}$ denote polar phonons associated to vibrations of the FE moment. (a) Depicts the low frequency (soft) spin wave mode. (b) High-frequency (gapped) mode. The dots in the circle denote the position of the spins one quarter cycle later. The soft mode leaves the canting angle $\beta$ invariant, while the gapped mode modulates $\beta$. (c) Coordinate system.[]{data-label="fig1"}](canted_waves){width="3in"}
The design of multiferroic materials with enhanced couplings of this type was recently discussed [@fennie07]. Although BiFeO$_3$ has no inversion center, its crystal structure is quite close to an inversion-symmetric one, and the above free energy is derived by assuming that both the DM vector and polarization $\bm{P}$ are associated with the same distortion of the lattice. An alternative model for BiFeO$_3$ assumes the DM vector to be independent of $\bm{P}$ [@ederer05], i.e., requires Eq. (\[f\]) be invariant under spatial inversion at a point on top of one of the magnetic ions. Later we will discuss the implications of this alternative assumption for the electromagnon spectra, and show how optical experiments may determine which model is appropriate.
The free energy is minimized by a homogeneous ferroelectric and antiferromagnetic state, with FE moment (at $\bm{E}=0$) given by $\bm{P}=P_0\hat{\bm{z}}$, with $P_{0}^{2}=\frac{-a}{u}+{\cal O}(d^3)$. The magnetic moments are perpendicular to $\bm{P}$,
$$\begin{aligned}
\bm{M}_{01}&=&M_0 \left(\sin{\beta} \hat{\bm{x}} +\cos{\beta}\hat{\bm{y}}\right),\\
\bm{M}_{02}&=&-M_0 \left(-\sin{\beta} \hat{\bm{x}} +\cos{\beta}\hat{\bm{y}}\right),\end{aligned}$$
with canting angle $\beta$ and magnetization $M_0$ determined by $\tan{\beta}=(dP_0)/(\tilde{J}+J)$, and $M_{0}^{2}=(\tilde{J}-r)/G$, with $\tilde{J}^{2}=(dP_0)^{2}+J^{2}$. Below the Curie and Néel temperatures we have $a<0$, and $J>-r>0$ respectively.
Small oscillations away from the ground state are described by the Landau-Lifshitz equations, $$\frac{\partial \bm{M}_i}{\partial t}=\gamma \bm{M}_{i}\times
\frac{\delta F}{\delta \bm{M}_{i}},\label{ll}$$ where $\gamma$ is a gyromagnetic ratio. A corresponding set of equations is written for $\bm{P}$ in order describe the high frequency optical phonon spectra. Keeping only the lowest order in the deviations $\delta \bm{M}_i$ and $\delta\bm{P}$, and focusing on the low requency magnetic oscillations we seek plane wave solutions of the type $$\bm{M}_i=\bm{M}_{0i}+\delta \bm{M}_i \textrm{e}^{i(\bm{k}\cdot \bm{r}-\omega t)}, \;
\bm{P}=P_0 \hat{\bm{z}}+\delta\bm{P}\textrm{e}^{i(\bm{k}\cdot \bm{r}-\omega t)}.
\label{pw}$$ >From Eq. (\[ll\]) we see that $\delta\bm{M}_i$ must be perpendicular to $\bm{M}_i$. Hence we may reduce the number of variables by using a parametrization for $\delta\bm{M}_{i}$ shown in Fig. 1(c), with further definitions $Y=y_1+y_2$, $Z=z_1+z_2$, $y=y_1-y_2$, $z=z_1-z_2$.
From Maxwell’s equations we see that any macroscopic wave producing nonzero fluctuations of $\delta\bm{M}=\delta\bm{M}_1+\delta\bm{M}_{2}$ must induce an AC magnetic $\bm{h}$ field. In the magnetostatic approximation this is obtained from $\nabla\cdot
\bm{h}=-4\pi\nabla\cdot\delta \bm{M}$ and $\nabla\times \bm{h}\approx
0$. The latter assumes the time variations are negligible in Maxwell’s equations, which is a good approximation for spin waves provided $k\gg \omega_{\rm{AFM}}/c$, with $c$ the speed of light. For a canted AFM this is a good approximation provided the domain sizes are smaller than a few centimeters. The self induced field is therefore $$\bm{h}=-4\pi \left(\delta \bm{M}\cdot \hat{\bm{n}}\right)\hat{\bm{n}},
\label{sif}$$ where $\hat{\bm{n}}$ is a propagation direction for the spin waves, $\bm{k}=k\hat{\bm{n}}$. The self-induced field contribute a term $2\pi
(\delta \bm{M}\cdot \hat{\bm{n}})^2$ to the free energy, tending to increase the spin wave frequencies whenever the quantity $\delta\bm{M}
= (-\cos{(\beta)} y, \sin{(\beta)}Y, Z)$ has a finite projection along $\hat{\bm{n}}$.
In the magnetostatic approximation the linearized equations of motion are obtained by substituting Eqs. (\[pw\])-(\[sif\]) into Eq. (\[ll\]), and using the explicit expressions for $\tan{\beta}$ and $M_0$. After some algebra the Landau-Lifshitz equations become
$$\begin{aligned}
\!\!\!\!\!\!\!\!\!\!-i\tilde{\omega}Y +(\tilde{J}+J+\alpha k^2)Z - 2h_z &=& -2 d' \cos{\beta} \delta P_x, \label{fc}\\
\!\!\!\!\!\!\!\!\!\!\alpha k^2 Y +i\tilde{\omega}Z- 2\sin{\beta} h_y &=&-4\eta' \sin{2\beta} \delta P_z,\label{sw1}\\
\!\!\!\!\!\!\!\!\!\!i\tilde{\omega}z +(2\tilde{J}+\alpha k^2)y +2\cos{\beta}h_x &=& -2d' \cos{2\beta}\delta P_z,\label{sw2}\\
\!\!\!\!\!\!\!\!\!\!(\tilde{J}-J+\alpha k^2) z -i\tilde{\omega}y &=& -2d' \sin{\beta}\delta P_y,\label{sw3}\end{aligned}$$
where we defined $\tilde{\omega}=\omega/(\gamma M_0)$, $d'=dM_0$, and $\eta'=\eta P_0M_0$.
Consider the pure spin waves in the limit $\delta\bm{P}\rightarrow 0$. This case may be solved analytically, because the system of four equations decouples into two independent sets of equations on the variables $(Y,Z)$ and $(y,z)$. The former is a low frequency mode, because it corresponds to spin vibrations that leaves the canting angle $\beta$ unchanged \[the spins vibrate in phase, see Fig 1(a)\]. The latter corresponds to spin vibrations half-cycle out of phase, leading to modulations of $\beta$, and a high frequency gap equal to the DM interaction $dP_0$ \[Fig. 1(b)\]. Neglecting terms to second order in $(dP_0)/J$, we may get an analytical expression for the low frequency mode, $$\begin{aligned}
\tilde{\omega}^{2}(\bm{k})&\approx&
2J \left( 1+\frac{4\pi}{J}n_{z}^{2}\right) \alpha k^{2} +
\frac{4\pi (dP_0)^{2}}{J}n_{y}^{2}.
\label{soft}\end{aligned}$$ This dispersion is anisotropic with respect to the polarization ($\hat{\bm{z}}$) axis: For $\bm{k}$ along the $x-z$ plane, we have a truly gapless mode to all orders in $dP_0/J$, with $\tilde{\omega}\approx \sqrt{2J\alpha}k$. For $\bm{k}$ along $\hat{\bm{y}}$ we find a gap equal to a fraction of the DM interaction, $\approx \sqrt{4\pi/J}(dP_0)$. This gap is a result of the *magnetostatic correction in the presence of DM weak ferromagnetism*.
![(a) Low frequency magnetostatic spin wave dispersion for a BiFeO$_3$ film, for propagation angles $\theta=0^{\circ}$ (propagation along the electric polarization direction $\hat{\bm{z}}$), $10^{\circ}$, $30^{\circ}$, $60^{\circ}$, $90^{\circ}$ (propagation along the Néel direction $\hat{\bm{y}}$). The high frequency mode (not shown) has a gap equal to the Dyzyaloshinskii-Moriya coupling ($5\times
10^{10}$ rad/s), and is nearly isotropic with respect to the direction of spin wave propagation. (b) Dispersion including electrodynamical effects in the $k<\omega/c$ region. Note the relationship between the magnetostatic gap in (a) and the photon-magnon anticrossing in (b).[]{data-label="fig2"}](spectra_film){width="3in"}
The physical origin of the magnetostatic gap is found by noting that $\delta \bm{M}$ for a pure soft mode $Y,Z\neq 0, y,z=0$ as $k\rightarrow 0$ is approximately given by a rigid rotation around the $\hat{\bm{z}}$ axis. In this limit, $\delta \bm{M}$ points exclusively along $\hat{\bm{y}}$, hence only propagation with some projection in this direction leads to a gap.
A small anisotropy is also found for the high frequency mode ($y,z\neq
0,Y=Z=0$). For example, when $\hat{\bm{k}}\parallel \hat{\bm{x}}$ the high frequency mode gap increases to $dP_0\sqrt{1+4\pi/J}$.
We calculated the coupled spin and polarization wave spectra solving the full set of Eqs. (\[fc\])-(\[sw3\]) numerically, with parameters extracted from experiment [@wang03; @bai05; @bea07]. The low frequency spin wave branch within the magnetostatic approximation is shown in Fig. 2(a). The inset \[Fig. 2(b)\] shows the low frequency spectra beyond the magnetostatic approximation, including electrodynamical corrections (For numerical convenience the speed of light was rescaled to $10^6$ cm/s). Note the anticrossing of the spin wave modes with the photon dispersion $\omega=ck$, and the orientation dependence of the photon gap. As expected, we see that the strict $k\rightarrow 0$ limit has no orientation dependence. We emphasize that the latter low $k$ limit is only observable for domain sizes of one cm or larger. The magnetostatic propagation anisotropy discussed in this work arises precisely because the spin waves travel with finite $k>\omega/c$.
Finally, we discuss the selection rules for the excitation or detection of magnon modes using an AC electric field. From inspecting the right hand side of Eqs. (\[fc\]) and (\[sw1\]) we see that the low frequency magnon may be excited electrically by the application of an AC field in the $x$ or $z$ direction. The former has a strong response in the presence of linear magnetoelectric effect ($d\neq
0$), while the latter has a weak response ($\propto \sin{\beta}$) due to magnetostriction.
The high frequency magnon $(x,y)$ has a dielectric response only in the presence of the linear magnetoelectric effect, as seen in Eqs. (\[sw2\]) and (\[sw3\]). This mode responds to electric fields in the $y-z$ plane, with the $z$ direction response larger by a factor of $\cos{2\beta}/\sin{\beta}\sim 2J/dP_0\gg 1$. The presence or absense thereof of this electromagnon using an optical probe may be used to discern whether the DM vector is linear in $P$ as proposed e.g. in [@zhdanov06] or if it is independent of $P$ as suggested in [@ederer05].
In conclusion, we predicted a magnetostatic gap anisotropy for the propagation of spin waves in a canted antiferromagnet. This effect may allow the electrical switching of magnons in multiferroic materials such as BiFeO$_3$ films.
The authors acknowledge useful conversations with J. Orenstein and R. Ramesh. This work was supported by WIN (RdS) and by NSF DMR-0238760 (JEM).
[99]{}
M.P. Kostylev, A.A. Serga, T. Schneider, B. Leven, and B. Hillebrands, , 153501 (2005).
A. Khitun and K. L. Wang, Superlattices and Microstructures [**38**]{}, 184 (2005).
J. Wang, J. B. Neaton, H. Zheng, V. Nagarajan, S. B. Ogale, B. Liu, D. Viehland, V. Vaithyanathan, D. G. Schlom, U. V. Waghmare, N. A. Spaldin, K. M. Rabe, M. Wuttig, and R. Ramesh, Science [**299**]{}, 1719 (2003).
F. Bai, J. Wang, M. Wuttig, J. Li, and N. Wang, Appl. Phys. Lett. [**86**]{}, 032511 (2005).
H. Béa, M. Bibes, S. Petit, J. Kreisel, and A. Barthélémy, Phil. Mag. Lett. [**87**]{}, 165 (2007).
I. Sosnowska, T. Peterlin-Neumaier, and E. Steichele, J. Phys. C: Solid State Phys. [**15**]{}, 4835 (1982).
T. Zhao, A. Scholl, F. Zavaliche, K. Lee, M. Barry, A. Doran, M. P. Cruz, Y. H. Chu, C. Ederer, N. A. Spaldin, R. R. Das, D. M. Kim, S. H. Baek, C. B. Eom, R. Ramesh, Nature Materials [**5**]{}, 823 (2006).
G.F. Herrmann, J. Phys. Chem. Solids [**24**]{}, 597 (1963).
D.R. Tilley and J.F. Scott, , 3251 (1982).
V.G. Bar’yakhtar and I.E. Chupis, Sov. Phys. Solid State [**10**]{}, 2818 (1969); [*ibid*]{} [**11**]{}, 2628 (1970).
G.A. Maugin, , 4608 (1981).
R. de Sousa and J.E. Moore (preprint), arXiv:0706.1260 (2007).
C. Fennie (preprint), arXiv:0711.1331 (2007).
C. Ederer and N.A. Spaldin, , 060401(R) (2005).
A.G. Zhdanov, A.K. Zvezdin, A.P. Pyatakov, T.B. Kosykh, and D. Viehland, Phys. Solid State [**48**]{}, 88 (2006).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
No. It is simply not plausible that cosmic acceleration could arise within the context of general relativity from a back-reaction effect of inhomogeneities in our universe, without the presence of a cosmological constant or “dark energy.” We point out that our universe appears to be described very accurately on all scales by a Newtonianly perturbed FLRW metric. (This assertion is entirely consistent with the fact that we commonly encounter $\delta \rho/\rho
> 10^{30}$.) If the universe is accurately described by a Newtonianly perturbed FLRW metric, then the back-reaction of inhomogeneities on the dynamics of the universe is negligible. If not, then it is the burden of an alternative model to account for the observed properties of our universe. We emphasize with concrete examples that it is [*not*]{} adequate to attempt to justify a model by merely showing that some spatially averaged quantities behave the same way as in FLRW models with acceleration. A quantity representing the “scale factor” may “accelerate” without there being any physically observable consequences of this acceleration. It also is [*not*]{} adequate to calculate the second-order stress energy tensor and show that it has a form similar to that of a cosmological constant of the appropriate magnitude. The second-order stress energy tensor is gauge dependent, and if it were large, contributions of higher perturbative order could not be neglected. We attempt to clear up the apparent confusion between the second-order stress energy tensor arising in perturbation theory and the “effective stress energy tensor” arising in the “shortwave approximation.”
author:
- |
Akihiro Ishibashi$^{\dag}$ and Robert M. Wald$^{\dag \ddag}$\
\
[*Enrico Fermi Institute$^{\dag}$ and Department of Physics$^{\ddag}$*]{}\
[*The University of Chicago, Chicago, IL 60637, USA*]{}
title: ' Can the Acceleration of Our Universe Be Explained by the Effects of Inhomogeneities? '
---
(\#1/\#2) (\#1/\#2)[()]{}
\#1[\^[(\#1)]{}]{}
\#1[[[\#1]{}\^]{}]{} \#1[[[\#1]{}\^]{}]{} \#1[[[\#1]{}\^]{}]{} \#1
Introduction
============
The apparent acceleration of our universe is one of the most striking cosmological observations of recent times. In the context of Friedmann-Lemaitre-Robertson-Walker (FLRW) models in general relativity, the acceleration of our universe would require either the presence of a cosmological constant or a new form of matter (“dark energy”) with large negative pressure. However, since there does not appear to be any natural explanation for the presence of a cosmological constant of the necessary size, nor does there appear to be any natural candidate for dark energy, it is tempting to look for alternative explanations. In recent years, there have been at least two approaches that have attempted to account for the observed acceleration of our universe within the framework of general relativity as being a consequence of deviations of our universe from exact FLRW symmetry, without invoking the presence of a cosmological constant or dark energy.
One approach notes that the mass density of our universe is, in fact, extremely inhomogeneous on scales much smaller than the Hubble radius. In order to get an effective homogeneous, isotropic universe, one needs to average and/or smooth out the inhomogeneities on some appropriate choice of spatial slicing. In such an “averaged” (or effective) FLRW universe, one can then define “effective cosmological parameters” [@Buchert00; @Buchert01; @BC03]. One then finds that the equations of motion for these effective cosmological parameters differ, in general, from the equations satisfied by these parameters in FLRW models. If they differ in a way that corresponds to adding a cosmological constant (or dark energy) of the right magnitude, then one may hope to have explained the acceleration of our universe in the context of general relativity, without invoking the presence of a cosmological constant or dark energy [@Rasanen03; @KMNR05; @KMR05; @BMR05; @Nambu05a; @NT05; @Moffat05].
A second approach has attempted to account for the acceleration of our universe as a back-reaction effect of long-wavelength cosmological perturbations. Here one constructs an “effective energy-momentum tensor” for these perturbations from second-order perturbation theory, and adds this as a source term in Einstein’s equation [@MAB1997Lett; @ABM1997PRD; @Nambu02; @BL2004]. If this effective energy-momentum tensor has a form similar to that of a cosmological constant (or, at least, provides a negative pressure term) and is of the appropriate magnitude, then one may hope that this could explain the acceleration of our universe.
Given that our universe appears to be very accurately described by a FLRW model, with very small deviations from homogeneity and isotropy, it would seem extremely implausible that cosmologically important effects could result from the second (or higher) order corrections produced by these small departures from a FLRW model. The main purposes of this paper are to explain this point with somewhat more precision and to point out some significant flaws in the arguments that have been made in the context of the above two approaches. In particular, we emphasize that one cannot justify a model by merely showing that spatially averaged quantities behave the same way in FLRW models with acceleration; rather one must show that [*all*]{} of the predictions of the model are compatible with observations. We illustrate this point by showing that spatial averaging can yield “acceleration” for the case of a universe that consists of two disconnected components, each of which is decelerating! We also show that spatial averaging can yield acceleration for suitably chosen slices of Minkowski spacetime. With regard to the back-reaction effects of long-wavelength perturbations, we note that the effective energy-momentum tensor is highly gauge dependent, as has previously been pointed out by Unruh [@Unruh1998]. Even in the long-wavelength limit, we show that one can get essentially any answer one wishes for the effective energy-momentum tensor, even though one cannot get any new physical phenomena beyond those already present in FLRW models. We also emphasize the difference between the use of an effective energy-momentum tensor for gravitational perturbations in the context of second-order perturbation theory and the use of a similar effective energy-momentum tensor in the context of the “shortwave approximation”[@BH1964; @Isaacson1968]. The former is highly gauge dependent (and, thus, not easily interpreted) and must be “small” if higher order perturbative corrections are to be neglected. The latter is essentially gauge independent and need not be “small.” However, the “shortwave approximation” clearly is not valid for analyzing long-wavelength cosmological perturbations.
In the next section, we point out that our universe appears to be very accurately described on all scales by a Newtonianly perturbed FLRW metric, despite the presence of large density contrasts. If this is correct, then higher order corrections to this metric resulting from inhomogeneities would be negligible. Thus, any model that attempts to explain the acceleration of our universe as a consequence of higher order effects of inhomogeneities will have to overcome the seemingly impossible burden of explaining why the universe appears to be so well described by a model that has only very small departures from a FLRW metric.[^1] In section 3, we illustrate that spatial averaging can produce an entirely spurious “acceleration” that is not associated with any physical observations. In section 4, we emphasize the distinction between second-order perturbation theory and the shortwave approximation, and we analyze the gauge dependence of the effective energy-momentum tensor arising in second-order perturbation theory.
The Newtonianly Perturbed FLRW Metric
=====================================
By a Newtonianly perturbed FLRW metric, we mean a metric of the form d\^2 = -(1 + 2 )dt\^2 + a\^2(t)(1- 2 ) \_[ij]{}dx\^idx\^j . \[metric:Newton-FLRW\] where $\gamma_{ij}$ denotes the metric of a space of constant curvature (3-sphere, flat, or hyperboloid), and $\Psi$ satisfies || 1 , | |\^2 D\^iD\_i , (D\^i D\_i )\^2 (D\^iD\^j) D\_iD\_j, \[condi:Newtonian\] where $D_i$ denotes the derivative operator associated with $\gamma_{ij}$. Suppose that the stress-energy content of this spacetime consists of fluid components that are very nearly homogeneously and isotropically distributed but may have arbitrary equation of state (such as radiation, dark energy, and/or a cosmological constant) together with components that may be very inhomogeneously distributed on scales small compared with the Hubble radius but are nearly pressureless and move with velocity much smaller than light relative to the Hubble flow (such as ordinary matter and dark matter). The stress-energy of the smoothly distributed components take the form T\^[(s)]{}\_[ab]{} \^[(s)]{}(t) dt\^2 + P\^[(s)]{}(t) a\^2(t) \_[ij]{}dx\^idx\^j , \[Ts\] where $P^{(s)} = P^{(s)}(\rho^{(s)})$ is arbitrary, whereas the inhomogeneously distributed components have a stress-energy of the “dust” form T\^[(m)]{}\_[ab]{} \^[(m)]{}(t,x\^i) dt\^2 . \[Tm\]
If one plugs the metric form eq. (\[metric:Newton-FLRW\]) into Einstein’s equation and uses eqs. (\[condi:Newtonian\]), the spatial average yields the usual FLRW equations for the scale factor $a$ with stress-energy source consisting of the sum of eq. (\[Ts\]) and the spatial average of eq. (\[Tm\]), 3 ()\^2 &=& \^2 ( \^[(s)]{} + |\^[(m)]{} ) - 3 , \[Einstein:tt\]\
3 &=& -( \^[(s)]{} + |\^[(m)]{} + 3 P\^[(s)]{} ) , \[Einstein:ij\] where $\bar{\rho}$ denotes the spatial average of $\rho$ (taken on a $t = {\rm const.}$ time slice with respect to the underlying FLRW metric with $K=\pm 1,\, 0$). The dominant remaining terms in Einstein’s equation then yield \_[(3)]{} = , \[eq:Poisson\] where $\Delta_{(3)} \equiv \gamma^{ij}D_i D_j$, and where $\delta \rho = \rho^{(m)} - \bar{\rho}^{(m)} $ denotes the deviation of the density from the spatial average. In the following we assume the metric $\gamma_{ij}$ to be a flat spatial metric, i.e., $K=0$.
It should be emphasized that the above discussion does [*not*]{} constitute a [*derivation*]{} that a metric form, eq. (\[metric:Newton-FLRW\]), is a good approximation to a solution to Einstein’s equation for stress-energy eq. (\[Ts\]) and (\[Tm\]) when eqs (\[Einstein:tt\])–(\[eq:Poisson\]) hold. Rather, all that has been shown is that if one [*assumes*]{} a metric of the form eq. (\[metric:Newton-FLRW\]) with eqs. (\[condi:Newtonian\]) holding and with the stress-energy given by eqs. (\[Ts\]) and (\[Tm\]), then eqs. (\[Einstein:tt\]), (\[Einstein:ij\]) and (\[eq:Poisson\]) must hold. The above discussion is thus analogous to the usual textbook “derivations” of the ordinary Newtonian limit of general relativity, where one also postulates a spacetime metric of a suitable “Newtonian form” and assumes that the matter distribution is approximately of the form (\[Tm\]) (see, e.g., section 4.4a of [@w84]). One can derive the ordinary Newtonian limit more systematically by considering one-parameter families of solutions to Einstein’s equation with suitable limiting properties (see, e.g., [@FS83] and [@ehlers] and references cited therein). It would be more difficult to provide an analogous analysis here, but we see no reason to doubt that eq. (\[metric:Newton-FLRW\]) is a good approximation to a solution to Einstein’s equation when eqs (\[Einstein:tt\])-(\[eq:Poisson\]) hold, provided, of course, that conditions (\[condi:Newtonian\]) are satisfied.
We now assert that the metric, eq. (\[metric:Newton-FLRW\]), appears to very accurately describe our universe on [*all scales*]{}, except in the immediate vicinity of black holes and neutron stars. The basis for this assertion is simply that the FLRW metric appears to provide a very accurate description of all phenomena observed on large scales, whereas Newtonian gravity appears to provide an accurate description of all phenomena observed on small scales. The metric (\[metric:Newton-FLRW\]) together with eqs (\[condi:Newtonian\]), predicts that large scale phenomena will be accurately described by a FLRW model, whereas the metric (\[metric:Newton-FLRW\]) together with eq. (\[eq:Poisson\]) predicts that Newtonian gravity will hold in regions small compared with the Hubble radius in which the stress-energy (\[Tm\]) dominates over (\[Ts\]) (see [@HolzWald97]).
Note that the validity of eq. (\[metric:Newton-FLRW\]) for accurately describing phenomena on [*small*]{} scales holds despite the fact that the density contrast of matter is commonly quite large [@Barrow88; @Futamase89] 1 . Indeed, for the solar system, galaxies, clusters of galaxies, we can estimate respectively, $\delta \rho / \rho \approx 10^{30},\, \approx 10^{5},\, \approx 10^{2}
\gg 1$. Nevertheless, in all of these cases, we have $\Psi \approx 10^{-6} \sim 10^{-5} \ll 1$, and the other conditions appearing in eq. (\[condi:Newtonian\]) also hold. Even for neutron stars, $\Psi \approx 10^{-1}$, so the metric eq. (\[metric:Newton-FLRW\]) is probably not too bad an approximation even in the vicinity of neutron stars.
The key point of this section is that if our assertion is correct that a metric of the form of eq. (\[metric:Newton-FLRW\]) accurately describes our universe, and if it also is true that conditions (\[condi:Newtonian\]) hold, then the nonlinear correction terms occurring in eqs. (\[Einstein:tt\]) and (\[Einstein:ij\]) are negligibly small. It therefore is manifest that nonlinear corrections[^2] to the dynamics of the universe will be negligible, i.e., there will be no important “back-reaction” effects of the inhomogeneities on the observed expansion of the universe on large scales. In particular, accelerated expansion cannot occur if the smoothly distributed matter satisfies the strong energy condition. However, our assertion that the metric, eq. (\[metric:Newton-FLRW\]), very accurately describes our universe is merely an assertion, and we cannot preclude the possibility that other models (e.g., with large amplitude, long-wavelength gravitational waves or with matter density inhomogeneities of a different type) might also fit observations. Our main point of this paper, however, is that if one wishes to propose an alternative model, then it is necessary to show that all of the predictions of this model are compatible with observations such as the observed redshift-luminosity relation for type Ia supernovae and the various observed properties of the cosmological microwave background (CMB) radiation. As we shall illustrate in the next two sections, it does not suffice to show merely that the spatially averaged scale factor behaves in a desired way or that an effective stress-energy tensor is of a desired form.
Cosmic Acceleration via Averaging
=================================
The type of spatial averaging in the context of Newtonianly perturbed FLRW models that was done to derive eqs. (\[Einstein:tt\]) and (\[Einstein:ij\]) above is not problematical. The metric very nearly has FLRW symmetry, so there is a natural choice of spatial slices on which one can take spatial averages. Since $\Psi \ll 1$, it makes negligible difference if one uses the spacetime metric (\[metric:Newton-FLRW\]) or the corresponding “background” FLRW metric (i.e., eq. (\[metric:Newton-FLRW\]) with $\Psi$ set equal to zero) to define the averaging.
If one has a metric that does not nearly have FLRW symmetry, one can, of course, still define spatial averaging procedures. However, these will, in general, be highly dependent on the choice of spatial slicing, and the results obtained from spatial averaging need not be interpretable in a straightforward manner. We now illustrate these comments with concrete examples.
For simplicity and definiteness, we consider an inhomogeneous universe with irrotational dust. In the comoving synchronous gauge,[^3] the metric takes the form ds\^2 = -d t\^2 + q\_[ij]{}(t,x\^m) dx\^idx\^j . Let $\Sigma$ denote a hypersurface of constant $t$, let ${{\mathcal D}}$ denote a compact region of $\Sigma$ and let $\phi$ be a scalar field on $\Sigma$. The average, $\langle \phi \rangle_{{\mathcal D}}$, of $\phi$ over the domain ${{\mathcal D}}\subset \Sigma$ may be defined by \_[[D]{}]{} \_[[D]{}]{}d, \[def:averaging\] where $V_{{\mathcal D}}$ denotes the volume of ${{\mathcal D}}$ and $d\Sigma$ is the proper volume element of $\Sigma$. [^4] Define the averaged scale factor, $a_{{\mathcal D}}$, by a\_[[D]{}]{}(V\_[[D]{}]{})\^[1/3]{} . We “time evolve” ${{\mathcal D}}$ by making it be comoving with the dust, i.e., the (comoving) coordinates of the boundary of ${{\mathcal D}}$ remain constant with time. Following Buchert [@Buchert00], one then obtains from Einstein’s equation the following equations of motion for $a_{{\mathcal D}}$, 3 &=& - \_[[D]{}]{}+ Q\_[[D]{}]{}, \[eq:evolve\]\
3 ( )\^2 &=& \^2\_[[D]{}]{}-[R]{} \_[[D]{}]{}- Q\_[[D]{}]{}, \[eq:friedman\] together with + a\_[[D]{}]{}\^4 =0 . \[condi:integrability\] Here ${\cal R}$ is the scalar curvature of $\Sigma$ and Q\_[[D]{}]{}&& ( \^2 \_[[D]{}]{} - \_[[D]{}]{}\^2 ) - \_[ij]{}\^[ij]{} \_[[D]{}]{}, In deriving these equations, it is important to bear in mind that the averaging of the time derivative $\langle {\dot \phi} \rangle_{{\mathcal D}}\equiv
\langle { \partial \phi}/\partial t \rangle_{{\mathcal D}}$ of a locally defined quantity $\phi$ differs in general from the time derivative of the averaged quantity $\Tdot{\langle \phi \rangle}_{{\mathcal D}}\equiv
\partial {\langle \phi \rangle}_{{\mathcal D}}/\partial t$, since the volume element and $V_{{\mathcal D}}$ may depend on the time $t$. Indeed, we have \_[[D]{}]{} = \_[[D]{}]{}+ \_[[D]{}]{} - \_[[D]{}]{}\_[[D]{}]{}, where $\theta$ denotes the expansion of the world lines of the dust.
It is immediately seen from (\[eq:evolve\]) that “averaged acceleration” $\ddot a_{{\mathcal D}}>0$ is achieved if Q\_[[D]{}]{}> \_[[D]{}]{}. \[condi:acceleration\] Buchert [@Buchert05] has discussed cosmological implications of the condition (\[condi:acceleration\]).
A number of authors (see, e.g., Refs. [@KMR05; @NT05; @Moffat05]) have sought to account for the observed acceleration of our universe by means of inhomogeneous models that satisfy eq. (\[condi:acceleration\]). However, our main point of this section is that even if our universe (or a suitable spatial domain of our universe) satisfies eq. (\[condi:acceleration\]) and thus has $\ddot
a_{{\mathcal D}}>0$, this does [*not*]{} imply that the model will possess any physically observable attributes of an accelerating FLRW model. Indeed Nambu and Tanimoto [@NT05] have shown that in a cosmological model in which ${{\mathcal D}}$ can be written as a union of regions each of which is locally homogeneous and isotropic, we have a\_[[D]{}]{}\^2 = a\_1\^2[a]{}\_1 + a\_2\^2[a]{}\_2 + + \_[ij]{} a\_i\^3a\_j\^3 ( - )\^2 , \[nt\] where $a_i$ denotes the locally defined scale factor in the $i$-th patch, and, in this case, $a_{{\mathcal D}}\equiv (a_1^3 + a_2^3 +
\cdots)^{1/3}$. Consider, now, a model where at time $t$ the universe consists of two disconnected(!) dust filled FLRW models, one of which is expanding and the other of which is contracting. Both components of the universe are, of course, decelerating, i.e., ${\ddot a}_1 < 0$, ${\ddot a}_2 < 0$. Nevertheless, it is not difficult to see from eq. (\[nt\]) that ${\ddot a}_{{\mathcal D}}> 0$ can easily be satisfied. For example, if we take $a \equiv a_1 = a_2 $ and $\dot{a}_1 = - \dot{a}_2$ we obtain a\_[[D]{}]{}\^2 [a]{}\_[[D]{}]{} = 2a\^3 { + 4()\^2 } = \^2 a\^3 > 0 . We thereby obtain a very simple model where the universe is accelerating according to the definition eq. (\[condi:acceleration\]), but all observers see only deceleration. This graphically illustrates that satisfaction of eq. (\[condi:acceleration\]) in a model is far from sufficient to account for the physically observed effects of acceleration in our universe. We see no reason to believe that the spatially averaged acceleration found, e.g., in the models of [@NT05; @Moffat05] directly corresponds to any physical effects of acceleration such as would be observed in type Ia supernovae data. The only way to tell if a model displays physically observable effects of acceleration is to calculate these effects.
As already mentioned above, the averaging procedure defined by eq. (\[def:averaging\]) also has ambiguities both with regard to the choice of time slicing and the choice of domain ${{\mathcal D}}$. One may also artificially produce an averaged cosmic acceleration as a result of a suitably chosen time-slicing. To show this explicitly, we give an example of accelerated expansion in Minkowski spacetime. We note first that for a general inhomogeneous universe (i.e., with no assumption concerning the form of the stress-energy), the equation of motion for $a_{{\mathcal D}}$ can be expressed as 3 = - \_[[D]{}]{} - 6( )\^2 + (G\_[ab]{} + g\_[ab]{}G\^c\_c)t\^at\^b \_[[D]{}]{}, \[average:acceleration\] where $G_{ab}$ denotes the Einstein tensor. Therefore, for any vacuum spacetime, if there is a domain ${{\mathcal D}}$ such that $- \langle {\cal R} \rangle_{{\mathcal D}}> 6\left({\dot a_{{\mathcal D}}}/{a_{{\mathcal D}}} \right)^2 $, then ${{\mathcal D}}$ describes an accelerated expansion insofar as $a_{{\mathcal D}}$ is concerned.
To construct an accelerating ${{\mathcal D}}$ in Minkowski spacetime, we start with two hyperboloidal slices, one of which corresponds to an expanding time-slice in the Milne chart covering the future of the origin ds\^2 = -da\^2 + a\^2 (d\^2 + \^2 d\^2) , and the other of which is a similar hyperboloidal slice that is contracting. The idea of the construction is to join these two hyperboloids at some radius, smooth out the join region, and choose ${{\mathcal D}}$ so that ${\dot a}_{{\mathcal D}}= 0$ (see Figure \[fig:Accele-slicing\]). Since the hyperboloids have negative scalar curvature, we thereby have ${\cal R} < 0$ except near the radius where the hyperboloids are joined. However, we can show that this construction can be done so that the contribution from the join region can be made arbitrarily small. Consequently, we obtain $3{\ddot a_{{\mathcal D}}}/{a_{{\mathcal D}}}
= - \langle {\cal R} \rangle_{{\mathcal D}}> 0$. Details of this construction are given in the appendix. Since Minkowski spacetime does not display any physical effects associated with accelerated expansion, this example shows quite graphically that “acceleration” as defined by the above averaging procedure can easily arise as a gauge artifact produced by a suitable choice of time slicing.
= 7cm
\[fig:Accele-slicing\]
Cosmological Back-reaction in the Long-wavelength Limit
=======================================================
A number of authors have considered the back-reaction effects of perturbations of FLRW models, particularly with regard to modes whose wavelength is comparable to or larger than the Hubble radius [@MAB1997Lett; @ABM1997PRD; @Nambu02; @BL2004]. The basic strategy has been to compute the second order terms in Einstein’s equation arising from these perturbations, thereby obtaining an “effective stress-energy tensor” for the perturbations. If the form of this effective stress-energy tensor corresponds to that of a positive cosmological constant of the correct magnitude, then one might hope to have provided a mechanism for obtaining the observed acceleration of our universe as a back-reaction effect of long-wavelength perturbations, without the need to introduce a cosmological constant or dark energy.[^5] We comment, first, that, even without extensive analysis, there is an intrinsic implausibility to this type of explanation. If one considers perturbations of wavelength less than the Hubble radius, it is hard to imagine that the perturbations could be so small that we do not notice any significant deviations from homogeneity and isotropy, yet so large that their [*second order*]{} effects produce very significant changes to the dynamics of our universe. On the other hand, if one goes to the long-wavelength limit, then the perturbation should correspond closely to a perturbation to a spatially homogeneous cosmological model. But, given the severe constraints on anisotropy arising from CMB observations, the perturbation should, in fact, correspond to a perturbation towards another FLRW model. Thus, one should not be able to obtain any new phenomena (such as acceleration without a cosmological constant or dark energy) that are not already present in FLRW models.
At least part of the confusion with regard to the calculation of the back-reaction effects of cosmological perturbations appears to stem from the fact that the notion of an “effective stress-energy tensor” for perturbations arises in two quite different contexts, namely (i) ordinary perturbation theory and (ii) the “shortwave approximation.” We now explain this distinction. For simplicity, we restrict consideration in the following discussion to the vacuum case; cosmological perturbations for the Einstein-scalar-field system will be considered later in this section.
Ordinary perturbation theory (see, e.g., section 7.5 of [@w84]) arises by considering a one-parameter family of metrics $g_{ab}(\alpha)$ that is jointly analytic in its dependence on $\alpha$ and the spacetime point. We refer to $g^{(0)}_{ab} \equiv g_{ab}(0)$ as the “background metric.” Roughly speaking, as $\alpha \rightarrow 0$, $g_{ab}(\alpha)$ differs from $g^{(0)}_{ab}$ by a perturbation that becomes of arbitrarily small amplitude but maintains a fixed profile. One expands $g_{ab}(\alpha)$ in a power series in $\alpha$ about $\alpha = 0$ g\_[ab]{}() = \_n \^n g\^[(n)]{}\_[ab]{} . The perturbation equations for $g^{(n)}_{ab}$ for the vacuum Einstein equation, G\_[ab]{} = 0 , \[ein\] are then obtained by differentiating the Einstein tensor, $G_{ab} (\alpha)$, of $g_{ab}(\alpha)$ $n$ times with respect to $\alpha$ at $\alpha = 0$. The zeroth order equation is just Einstein’s equation for $g^{(0)}_{ab}$ G\_[ab]{} \[g\^[(0)]{}\] = 0 . \[0th\] The first order equation is G\^[(1)]{}\_[ab]{} \[g\^[(1)]{}\] = 0 , \[1st\] where $G^{(1)}_{ab}$ denotes the linearized Einstein tensor off of the background metric $g^{(0)}_{ab}$. The second order equation is G\^[(1)]{}\_[ab]{} \[g\^[(2)]{}\] = - G\^[(2)]{}\_[ab]{} \[g\^[(1)]{}\] , \[2nd\] where $G^{(2)}_{ab}[g^{(1)}]$ denotes the second-order Einstein tensor constructed from $g^{(1)}_{ab}$.
As can be seen from eq. (\[2nd\]), minus the second-order Einstein tensor (divided with the gravitational constant), $- \kappa^{-2}G^{(2)}_{ab} [g^{(1)}]$, plays the role of an “effective stress-energy tensor” associated with the perturbation $g^{(1)}_{ab}$ in the sense that it acts as a source term for the second-order metric perturbation $g^{(2)}_{ab}$. However, this does [*not*]{} mean that one can treat $- \kappa^{-2}G^{(2)}_{ab} [g^{(1)}]$ as though it were a new form of matter stress-energy that can be inserted into the right side of the exact Einstein equation (\[ein\]) as opposed to the right side of eq. (\[2nd\]). For one thing, the second-order Einstein tensor is highly gauge dependent (as we shall illustrate explicitly below), so it is not straightforward to interpret its meaning. The key point, however, is that eq. (\[2nd\]) arises only in the context of perturbation theory. If $G^{(2)}_{ab} [g^{(1)}]$ is very small, then its effects on the spacetime metric can be reliably calculated from eq. (\[2nd\]). But if $G^{(2)}_{ab} [g^{(1)}]$ is large enough to produce cosmologically interesting effects (such as acceleration), then the third and higher order contributions to $g_{ab}(\alpha)$ will also be large, and one cannot reliably compute back-reaction effects from second-order perturbation theory.
This situation occurring in perturbation theory contrasts sharply with the situation that arises when one uses the “shortwave approximation” [@Isaacson1968; @MTW; @Burnett1989]. Here, one wishes to develop a formalism in which the self-gravitating effects of gravitational radiation—and the consequent effects on the spacetime metric on scales much larger than the wavelength of the radiation—can be reliably obtained, even when these effects are “large.” Again, one considers a one-parameter family of metrics $g_{ab}(\beta)$ that has a continuous limit to the metric $g^{(0)}_{ab} \equiv g_{ab}(0)$. Thus, as in ordinary perturbation theory, as $\beta \rightarrow 0$, $g_{ab}(\beta)$ differs from $g^{(0)}_{ab}$ by a perturbation of arbitrarily small amplitude. However, one now requires $g_{ab}(\beta)$ to be such that, roughly speaking, as $\beta \rightarrow 0$, the ratio of the amplitude to the wavelength of the perturbation goes to a finite, non-zero limit; see [@Burnett1989] for a precise statement of what is required in this limit. Thus, in this scheme, the dominant terms in Einstein’s equation as $\beta \rightarrow 0$ are actually the linear terms in the second derivatives of the first order perturbation, which diverge as $1/\beta$. One thereby obtains G\^[(1)]{}\_[ab]{} \[g\^[(1)]{}\_[ab]{}\] = 0 . The quadratic terms in the first order perturbation are of zeroth order in $\beta$, so they make a contribution to the Einstein tensor that is comparable to that of $g^{(0)}_{ab}$. The linear terms in the second-order perturbation also contribute to this order, but these contributions can be eliminated by averaging over a spacetime region that is large compared with the wavelength of the perturbation. One thereby obtains G\_[ab]{} \[g\^[(0)]{}\] = - G\^[(2)]{}\_[ab]{} \[g\^[(1)]{}\] , \[2ndsw\] where the brackets on the right side of eq. (\[2ndsw\]) denote a suitably defined spacetime average. It can be shown that $\langle - G^{(2)}_{ab} [g^{(1)}] \rangle$ is gauge invariant in a suitably defined sense. We refer to [@Burnett1989] for further details of the derivation and meaning of these equations.
Although eq. (\[2ndsw\]) is quite similar in form to eq. (\[2nd\]), the meaning and range of validity of these equations are quite different. In contrast to eq. (\[2nd\]), it should be possible to use eq. (\[2ndsw\]) to calculate the back-reaction effects of gravitational radiation even when these effects are large. The catch, however, is that eq. (\[2ndsw\]) can be used only when the wavelength of the perturbation is much smaller than the curvature lengthscale of the background spacetime. Thus, if the universe were filled with gravitational radiation of wavelength much smaller than the Hubble radius, then it should be possible to use eq. (\[2ndsw\]) to reliably calculate the back-reaction effects of this radiation, even if this radiation is the dominant form of “matter” in the universe. However, eq. (\[2ndsw\]) manifestly [*cannot*]{} be used to calculate the back-reaction effects of long-wavelength perturbations.
We conclude this section by deriving an explicit formula for the gauge dependence of the second-order “effective stress-energy tensor” arising in ordinary perturbation theory for long-wavelength scalar-type perturbations of a FLRW universe containing a scalar field. By doing so, we will see that one can get essentially any answer one wishes for this effective stress-energy tensor by making appropriate gauge transformations. This graphically shows that one cannot draw any physical conclusions merely by examining the form of the effective stress-energy tensor arising in second-order perturbation theory. [^6]
Consider a background flat FLRW universe ds\^2 = - dt\^2 + a\^2(t) \_[ij]{}dx\^idx\^j , filled with a scalar field whose energy momentum tensor is given by T\_[ab ]{} = \_a \_b - g\_[ab ]{} {\^c\_c + 2U() } . \[stress:scalar\] The unperturbed background equations of motion for $a$ and $\phi$, which are functions of only $t$, are given by && H\^2 ()\^2 = (\^2+ U ) ,\
&& + 3H + = 0 , where in these equations and hereafter the [*dot*]{} denotes the derivative with respect to $t$.
We focus on the scalar-type perturbations. The general form of a scalar-type metric perturbation is d \^2 = - (1 + 2A)dt\^2 - 2 a B\_i dt dx\^i + a\^2 { (1 + 2H\_L)\_[ij]{} + 2H\_T\_[ij]{} } dx\^idx\^j , \[metric:pert\] and the scalar field perturbation is given by = + . Here $\scalar$ denotes a plane wave on flat $3$-space with wavevector ${\bf k}$, and $\scalar_{i}$ and $\scalar_{ij}$ are the divergence-free vector and transverse-traceless tensor defined by \_[i]{} = - D\_i, \_[ij]{} = ( D\_iD\_j - \_[ij]{}\_[(3)]{} ), with $D_i$ being the derivative operator associated with the $3$-space metric $\gamma_{ij}$ as in (\[condi:Newtonian\]), and $k^2={\bf k}\cdot {\bf k}$. Here and in the following, perturbation variables are understood as corresponding Fourier expansion coefficients—hence functions merely of $t$—and we omit the index ${\bf k}$ unless otherwise stated.
Under infinitesimal gauge transformations of the scalar-type; t t + T, x\^i x\^i + L \^i , \[coord\] the perturbation variables, $A,\,B,\,H_L,\,H_T,\, \delta \phi$, change as A && A - T , \[gt:A\]\
B && B + aL + , \[gt:B\]\
H\_L && H\_L - - H [T]{} , \[gt:HL\]\
H\_T && H\_T + k[L]{} , \[gt:HT\]\
&& - . \[gt:phi\] In particular, it follows from eqs. (\[gt:B\]) and (\[gt:HT\]) that the following combinations X\_T ( - B ) , X\_L - H\_T , change as X\_T X\_T - T , X\_L X\_L - L . \[gaugetransf:XTXL\] Hence, by inspection of eqs. (\[gt:A\]), (\[gt:HL\]) and (\[gt:phi\]), one can immediately obtain gauge-invariant perturbation variables [@Bardeen1980; @KS1984]; A - , H\_L - X\_L - HX\_T , - X\_T . Any scalar-type gauge-invariant perturbation quantity can be expressed as a linear combinations of the gauge-invariant variables $\Psi$, $\Phi$, and $\Dphi$, and their time derivatives.
It follows from the linearized Einstein equations that the gauge-invariant variables defined above satisfy the following equations [@KS1984] ( + ) = 0 , \[Ein:ij:tracefree\] k = 0 . \[Ein:0j\] which correspond, respectively, to the trace-free part of the space-space component and the time-space component of the linearized Einstein equations. For $k^2 \neq 0$, we obtain from eq. (\[Ein:0j\]) the following relations between $\Phi$, $\Psi$, and $\Dphi$; = - , = - ( + H ) . \[rel:Psi:Phi:X\] It then also follows from Einstein equations that $\Psi$ is governed by && + ( H - ) + ( 2 - ) H + = 0 . \[eom:Psi\]
We therefore have found that all of the scalar-type perturbations variables are given in terms of the variables $\Psi$, $X_T$, and $X_L$ by A &=& + , \[expr:A\]\
B &=& - a - X\_T , \[expr:B\]\
H\_L &=& - + H X\_T + X\_L , \[expr:HL\]\
H\_T &=& - k X\_L , \[expr:HT\]\
&=& - ( + H ) + X\_T . \[expr:Delta\] The variable $\Psi$ is gauge invariant and satisfies eq. (\[eom:Psi\]). On the other hand, the gauge transformation law (\[gaugetransf:XTXL\]) implies that the functions $X_T$ and $X_L$ are completely arbitrary, i.e., they may be chosen to take any values that one wishes. Thus, the specification of $X_T$ and $X_L$ in terms of $\Psi$ corresponds to fixing the gauge freedom. For example, the choice X\_T=X\_L=0 corresponds to the Poisson gauge (or the longitudinal gauge), in which the metric, eq. (\[metric:pert\]), takes precisely the form of eq. (\[metric:Newton-FLRW\]). Another example is the choice X\_T &=& - \^t\_[t\_\*]{} (t’) dt’+ C\_1(k) , \[X:gauge:synchronous:1\]\
X\_L &=& k\^t\_[t\_\*]{} {\^[t’]{}\_[t’\_\*]{} (t”) dt” } - k C\_1(k) \^t\_[t\_\*]{} + C\_2(k) , \[X:gauge:synchronous:2\] where $t_*$ denotes some reference time and $C_1$ and $C_2$ are arbitrary constants. This choice corresponds to the synchronous gauge, $A=B=0$, in which $C_1$ and $C_2$ parameterize the residual gauge freedom in this gauge. The second-order effective stress-energy tensor for the Einstein-scalar-field system is defined by \^[([eff]{})]{}T\_[ab]{} - G\^[(2)]{}\_[ab]{}\[g\^[(1)]{}\] + T\^[(2)]{}\_[ab]{}\[,g\^[(1)]{}\] , where $G^{(2)}_{ab}$ denotes the second order Einstein tensor and $T^{(2)}_{ab}$ is the similarly defined second order contribution to $T_{ab}$, eq. (\[stress:scalar\]), arising from the first order perturbation $(\delta \phi,g^{(1)})$. We now calculate the second-order effective stress-energy tensor in order to explicitly demonstrate its gauge dependence. It is very convenient to express ${}^{({\rm eff})}\!T_{ab}$ in terms of the variables $\Psi$, $X_T$, and $X_L$, since any dependence of ${}^{({\rm eff})}\!T_{ab}$ on $X_T$, and $X_L$ will explicitly show its gauge dependence. Clearly, since ${}^{({\rm eff})}\!T_{ab}$ is quadratic in the first order perturbation, it must consist of a part, ${}^{({\rm eff:}\Psi)}\!T_{ab}$, that is quadratic in $\Psi$, a part, ${}^{({\rm eff:}X)}\!T_{ab}$, that is quadratic in $(X_T, X_L)$, and a part, ${}^{({\rm eff:}\Psi,X)}\!T_{ab}$, containing the “cross-terms” between $\Psi$ and $X_T, X_L$. The quantity ${}^{({\rm eff:}\Psi)}\!T_{ab}$ is gauge invariant, [^7] but both ${}^{({\rm eff:}X)}\!T_{ab}$ and ${}^{({\rm eff:}\Psi,X)}\!T_{ab}$ are gauge dependent. Thus, ${}^{({\rm eff})}\!T_{ab}$ will be gauge invariant if and only if these latter pieces vanish.
It is easy to verify that both ${}^{({\rm eff:}X)}\!T_{ab}$ and ${}^{({\rm eff:}\Psi,X)}\!T_{ab}$ are nonvanishing. To see explicitly that ${}^{({\rm eff:}X)}\!T_{ab}$ is nonvanishing, it suffices to consider the case where we impose the additional restriction X\_L = - k\^t\_[t\_\*]{} dt’ , which ensures that $B=0$, thereby considerably simplifying the calculation. We also focus attention on the long-wavelength limit. A brute force calculation then yields \^2 \^[([eff:]{}X)]{}T\_[00]{} &=& + O(k\^2), \[T00\]\
\^2 \^[([eff:]{}X)]{}T\_[ij]{} &=& g\_[ij]{} + O(k\^2). \[Tij\] Thus, even in the case of a pure gauge perturbation, $\Psi = 0$, we can obtain a non-vanishing effective stress-energy tensor for long-wavelength perturbations. Indeed, since $X_T$ is entirely arbitrary, we see that we can get essentially any answer one wishes for ${}^{({\rm eff})}\!T_{ab}$. For example, if one wishes to have a pure gauge perturbation in which ${}^{({\rm eff})}\!T_{ab}$ takes the form of a cosmological constant, one would merely have to solve the second-order ordinary differential equation for $X_T$ that results when one equates the right side of eq. (\[T00\]) to minus the right side of eq. (\[Tij\]). This manifestly demonstrates that one cannot derive any physical consequences by merely examining the form of the second-order effective stress-energy tensor.
Finally, we comment that we derived the above “long-wavelength limit” form of ${}^{({\rm eff})}\!T_{ab}$ for scalar-type perturbations by considering perturbations with $k^2 \neq 0$ and then taking the limit as $k^2 \rightarrow 0$. Alternatively, we could have directly considered scalar-type perturbations with $k^2 = 0$. It is not immediately obvious that this would give equivalent results, since when $k^2 = 0$, the quantities $\scalar_{i}$ and $\scalar_{ij}$ do not exist, so $B$ and $H_T$ are not defined and the $x^i$ coordinate freedom in eq. (\[coord\]) does not exist. Furthermore, eqs. (\[Ein:ij:tracefree\]) become trivial, and therefore the relation, eq. (\[rel:Psi:Phi:X\]) need not hold. Thus, it is not entirely straightforward to make a physical correspondence between perturbations with $k^2=0$ and the $k^2 \rightarrow 0$ limit of perturbations with $k^2 \neq 0$. Nevertheless, such a one-to-one, onto correspondence does exist[^8], and can be explicitly achieved by using the gauge freedom available when $k^2 \neq 0$ to set $B = H_T =0$ and using the gauge freedom available when $k^2 = 0$ to set $A = - H_L$. Further discussion of the relationship between perturbations in the long-wavelength limit and exactly homogeneous perturbations can be found in Refs. [@NT1998; @KH1998; @ST1998].
Since scalar-type perturbations with $k^2=0$ manifestly correspond to perturbations to other FLRW spacetimes, it is clear that one cannot find any new physical phenomena that are not already present in FLRW models by studying long-wavelength perturbations and dropping all terms that are $O(k^2)$. For example, consider a FLRW model which contains two matter components, such as dust and radiation or two scalar fields. In such a model, there exist nontrivial gauge-invariant perturbations even in the $k\rightarrow 0$ limit, and implications of such perturbations to the cosmological back-reaction problem have been discussed in [@Nambu05a; @AW01; @GB05]. However, in the $k\rightarrow 0$ limit such perturbations merely correspond to perturbations to other FLRW models; in the above examples, they would correspond to changing the proportion of dust and radiation or changing the initial conditions of the scalar fields. These perturbations cannot give rise to any new phenomena—such as physically measurable acceleration—that are not already present in exact FLRW models.
Summary
=======
In this paper, we have argued that the attempts to explain cosmic acceleration by effects of inhomogeneities, without invoking a cosmological constant or dark energy, are, at best, highly implausible. A Newtonianly perturbed FLRW metric appears to describe our universe very accurately on all scales. In this model, the back-reaction effects of inhomogeneities on the cosmological dynamics are negligible even though the density contrast may be very large on small scales.
We focused much of our attention on exposing the flaws in two types of attempts to explain acceleration by effects of inhomogeneities. (i) Starting from an inhomogeneous model, one can obtain an effective FLRW universe by spatial averaging. This effective FLRW universe may display acceleration. However, we showed explicitly via concrete examples that acceleration of the effective FLRW universe may occur in situations where no physically observable effects of acceleration actually occur. (ii) The back-reaction effects of a perturbation of a FLRW universe are described at second order by an effective stress-energy tensor constructed from the first order perturbation. In particular cases, this effective stress-energy tensor may take the form of a cosmological constant, thereby suggesting that it could produce acceleration. However, we pointed out that (unlike the effective stress-energy tensor arising in the shortwave approximation), the effective stress-energy tensor arising in second order perturbation theory is highly gauge dependent and must be small in order to justify neglecting higher order corrections. We explicitly evaluated the second-order effective stress-energy tensor for pure gauge scalar-type perturbations of an Einstein-scalar field model, and showed that it can take essentially any form one wishes, including the form of a cosmological constant.
[**Acknowledgments**]{}
This research was supported by NSF grant PHY 00-90138 to the University of Chicago.
Appendix {#appendix .unnumbered}
========
Here we provide some details of the construction of the accelerating domain ${{\mathcal D}}$ in Minkowski spacetime that was described below eq. (\[average:acceleration\]). Let $t$ and $r$ be, respectively, the standard time and radial coordinates in Minkowski spacetime. Let $a>0$ and $r_0>\epsilon>0$. Let $f$ be a smooth, monotone decreasing function of one variable such that $f(x) = 1$ for all $x \leq 1/2$, $f(x) = 0$ for all $x \geq 1$. Define (r) = f ( +1 ) . Then $\psi(r) = 1$ whenever $r\leq r_0-\epsilon$ and $\psi(r) = 0$ whenever $r \geq r_0$. Furthermore, there exists a constant $C>0$, independent of $\epsilon$, such that $\epsilon |\psi'| < C $ for all $0<\epsilon<r_0$, where $\psi' \equiv \partial \psi/\partial r$. Define F(r) {(r) - (-r + 2r\_0 ) } ( - ) + . \[embedding\] Then $F$ is smooth and the hypersurface $\Sigma$ defined by $t=F(r)$ also is smooth. For $r < r_0 - \epsilon$, $\Sigma$ is the expanding hyperboloid $t = \sqrt{r^2 + a^2}$, whereas for $r > r_0 + \epsilon$, $\Sigma$ is the contracting hyperboloid $t = - \sqrt{r^2 + a^2} + 2\sqrt{r_0^2 + a^2}$. The local expansion rate $H\equiv \dot a/a$ of $\Sigma$ smoothly changes from $1/a$ to $-1/a$ in the junction interval $(r_0 -\epsilon , \,r_0+\epsilon)$. Nowhere does $\Sigma$ display an accelerated expansion locally.
Now let us take our domain ${{\mathcal D}}$ to be a ball of radius $r_{{\mathcal D}}$ on $\Sigma$, where $r_{{\mathcal D}}$ is chosen so that $\dot a_{{\mathcal D}}$ vanishes. It is always possible to find such an $r_{{\mathcal D}}$ since, by construction, $V_{{\mathcal D}}$, hence $\dot a_{{\mathcal D}}$, is a smooth function of $r$, and the local expansion rate $H=\dot a/a$ is positive when $r< r_0-\epsilon$, whereas it is negative when $r_0+\epsilon < r$. Since the third term of eq. (\[average:acceleration\]) vanishes for Minkowski spacetime, if $\langle {\cal R} \rangle_{{\mathcal D}}$ is negative in ${{\mathcal D}}$, then eq. (\[average:acceleration\]) shows an acceleration $3\ddot a_{{\mathcal D}}/a_{{\mathcal D}}= -\langle {\cal R}\rangle_{{\mathcal D}}>0$. However, apart from the junction region $(r_0 -\epsilon, \, r_0 + \epsilon)$, $\Sigma$ is intrinsically a hyperbolic space with a negative scalar curvature ${\cal R} = -6/a^2$. Furthermore, the following calculation shows that the contribution to $\langle {\cal R} \rangle_{{\mathcal D}}$ from the junction region can be made negligibly small. The induced metric on $\Sigma$ is ds\^2 = ( 1-F’\^2 )dr\^2 + r\^2 d\^2 , so the scalar curvature of the junction region is given by = -. Using the formula, F’ = - (+ ’) + O() , which is obtained from the properties of $\psi$ in the junction interval, one finds \_[r\_0-]{}\^[r\_0+]{} dr r\^2 &=& \_[r\_0-]{}\^[r\_0+]{} dr { 2 - 4 ( ) }\
&=& O() , which can be made arbitrarily small by taking $\epsilon \rightarrow 0$. Thus, the contribution to $\langle {\cal R} \rangle_{{\mathcal D}}$ from the junction region can indeed be made arbitrarily small, so that $-\langle {\cal R}\rangle_{{\mathcal D}}\approx 6/a^2 > 0$. Thus, one obtains an accelerating domain ${{\mathcal D}}\subset \Sigma$ in Minkowski spacetime through the volume averaging process.
[99]{}
Buchert, T., 2000, Gen. Rel. Grav. [**32**]{}, 105.
Buchert, T., 2001, Gen. Rel. Grav. [**33**]{}, 1381.
Buchert, T. and Carfora, M., 2003 Phys. Rev. Lett. [**90**]{}, 031101.
Rasanen, S., 2004, JCAP 0402 003.
Kolb, E.W., Matarrese, S., Notari, A., and Riotto, A., hep-th/0503117.
Kolb, E.W., Matarrese, S., and Riotto, A., astro-ph/0506534.
Barausse, E., Matarrese, S., and Riotto, A., 2005 Phys. Rev. D [**71**]{}, 063537.
Nambu, Y., 2005 Phys. Rev. D [**71**]{}, 084016.
Nambu, Y. and Tanimoto, M., gr-qc/0507057.
Moffat, J.W., astro-ph/0505326.
Mukhanov, V.F. Abramo, L.R.W., and Brandenberger, R.H., 1997 Phys. Rev. Lett. [**78**]{}, 1624.
Abramo, L.R.W., Brandenberger, R.H., and Mukhanov, V.F., 1997 Phys. Rev. D [**56**]{}, 3246.
Nambu, Y., 2002 Phys. Rev. D [**65**]{}, 104013.
Brandenberger, R.H. and Lam, C.S., hep-th/0407048.
Unruh, W., 1998 astro-ph/9802323.
Brill, D. and Hartle, J., 1964 Phys. Rev. [**135**]{}, B271.
Isaacson, R., 1968 Phys. Rev. [**166**]{}, 1272.
Flanagan, E.E., 2005 Phys. Rev. D [**71**]{}, 103521.
Hirata, C.M. and Seljak, U., 2005 astro-ph/0503582.
Geshnizjani, G., Chung, D.J.H., and Afshordi, N., 2005 Phys. Rev. D [**72**]{}, 023517.
Wald, R.M., 1984 [*General Relativity*]{}, University of Chicago Press: Chicago.
Futamase, T. and Schutz, B.F., 1983 Phys. Rev. D [**28**]{}, 2363.
Ehlers, J., 1997 Class. Quant. Grav. [**14**]{}, A119.
Holz, D.E. and Wald, R.M., 1998 Phys. Rev. D [**58**]{}, 063501.
Barrow, J.D., 1988 Quart. J. Roy. astr. Soc., [**30**]{}, 163.
Futamase, T., 1989 Mon. Not. R. astr. Soc. [**237**]{}, 187.
Zalaletdinov, R.M., 1997 Bull. Astron. Soc. India [**25**]{}, 401.
Coley, A.A., Pelavas, N. and Zalaletdinov, R.M., gr-qc/0504115.
Buchert, T., gr-qc/0507028.
Misner, C. W., Thorne, K. S. and Wheeler, J.A., 1973 [*Gravitation*]{} (Freeman, San Francisco).
Burnett, G.A., 1989 J. Math. Phys. [**30**]{}, 90.
Abramo, L.R. and Woodard, R.P., 2002 Phys. Rev. D [**65**]{}, 043507.
Geshnizjani, G. and Brandenberger, R.H., 2005 JCAP [**04**]{}, 006.
Bardeen, J.M., 1980 Phys. Rev. D [**22**]{}, 1882.
Kodama, H. and Sasaki, M., 1984 Prog. Theor. Phys. Supple. [**78**]{}, 1.
Nambu, Y. and Taruya, A, 1998 Class. Quant. Grav. [**15**]{}, 2761.
Kodama, H. and Hamazaki, T., 1998 Phys. Rev. D [**57**]{}, 7177.
Sasaki, M. and Tanaka, T., 1998 Prog. Theor. Phys. [**99**]{}, 763.
[^1]: For example, rotation of the cosmic matter may produce acceleration effects [@KMNR05] but these acceleration effects must be negligible in view of the observed isotropy of our universe [@Flanagan05; @HS05; @GCA05].
[^2]: We take this opportunity to comment upon one misconception related to the validity of the Newtonianly perturbed FLRW metric. It is commonly stated that when $\delta \rho/{\rho} \gg 1$, one enters a “nonlinear regime.” This might suggest that the validity of eqs. (\[metric:Newton-FLRW\]) and/or (\[eq:Poisson\]) would be questionable whenever $\delta \rho/{\rho} \gg 1$. However, this is not the case; the proper criteria for the validity of the metric eq. (\[metric:Newton-FLRW\]) are conditions (\[condi:Newtonian\]), not $\delta \rho/{\rho} \ll 1$. It is true that nonlinear effects become important for the motion of matter when $\delta \rho/{\rho} \gg 1$. This follows simply from the fact that self-gravitation is a nonlinear effect, and self-gravitational effects on the motion of matter cannot be ignored when $\delta \rho/{\rho} > 1$. But this does not mean that one must include nonlinear corrections to the metric form, eq. (\[metric:Newton-FLRW\]), or to eq. (\[eq:Poisson\]) in order to get a good approximation to the spacetime metric. Indeed, if one is trying to describe the solar system in the context of the ordinary Newtonian limit of general relativity, one must include “nonlinear effects” to obtain the correct motion of the planets; they would move on geodesics of the flat metric rather than the Newtonianly perturbed metric if not for these “nonlinear effects.” However, the corrections to the spacetime metric of the solar system arising from nonlinear terms in Einstein’s equation are entirely negligible.
[^3]: If the universe is filled with irrotational dust, then the comoving synchronous gauge defines a natural choice of slicing, namely the slices orthogonal to the world lines of the dust. However, for an inhomogeneous universe, this gauge choice typically will break down on timescales much shorter than cosmological timescales, due to formation of caustics. For example, synchronous coordinates defined in a neighborhood of the Earth would typically break down on a timescale of order the free fall time to the center of the Earth, i.e., $\sim 1$ hour.
[^4]: For a different type of averaging procedure than that given by eq. (\[def:averaging\]) and its application to cosmology, see e.g., [@Zalaletdinov97; @CPZ05].
[^5]: The motivation in [@MAB1997Lett; @ABM1997PRD] was actually to use back-reaction effects to attempt to cancel the presence of a large cosmological constant rather than to use back-reaction to directly produce acceleration.
[^6]: Discussion of the characterization of the back-reaction effects of perturbations in terms of physical variables can be found in [@AW01; @GB05].
[^7]: In fact, ${}^{({\rm eff:}\Psi)}\!T_{ab}$ is precisely the “effective energy-momentum tensor for cosmological perturbations” of [@MAB1997Lett; @ABM1997PRD]. However, contrary to the claims of [@MAB1997Lett; @ABM1997PRD], this effective energy-momentum tensor is gauge-invariant only in the trivial sense that any gauge-dependent quantity can be viewed as gauge invariant once a gauge has been completely fixed. In the variations taken in [@MAB1997Lett; @ABM1997PRD] to obtain their effective energy-momentum tensor, $X_T$ and $X_L$ were implicitly assumed to be independent of $\Psi$, corresponding to the choice of the Poisson (longitudinal) gauge $X_T=X_L=0$. However, different specifications of $X_T$ and $X_L$ in terms of $\Psi$—i.e., different choices of gauge—would lead to [*different*]{} expressions for the effective energy-momentum tensor in terms of $\Psi$.
[^8]: It also is worth pointing out that, although the gauge freedom is different, the effective stress-energy tensor for $k^2=0$ perturbations remains highly gauge dependent.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Extensive Path Integral Monte Carlo simulations of condensed [*para*]{}-Hydrogen in two dimensions at low temperature have been carried out. In the zero temperature limit, the system is a crystal at equilibrium, with a triangular lattice structure. No metastable liquid phase is observed, as the system remains a solid down to the spinodal density, and breaks down into solid clusters at lower densities. The equilibrium crystal is found to melt at a temperature close to 7 K.'
author:
- Massimo Boninsegni
title: 'Low temperature phase diagram of condensed [*para*]{}-Hydrogen in two dimensions'
---
A fluid of [*para*]{}-Hydrogen (2) molecules has long been regarded as a potential superfluid, due to the light mass and the bosonic character of its constituents [@ginzburg72]. In bulk 2, however, superfluidity (SF) is not observed, because, unlike Helium, molecular Hydrogen solidifies at a temperature ($T$ $\approx$ 14 K) significantly higher than that ($\sim$ 4 K) at which phenomena such as Bose Condensation and, possibly, SF might occur. This is due the depth of the attractive well of the potential between two Hydrogen molecules, significantly greater than that between two Helium atoms. Several, attempts have been made [@bretz81; @maris86; @maris87; @schindler96] to supercool bulk liquid 2, but no observation of SF in the bulk phase of 2 has been reported to date.
Reduction of dimensionality is regarded as a plausible avenue to the stabilization of a liquid phase of 2 at temperatures sufficiently low that a superfluid transition may be observed. This has been the primary motivation underlying the experimental investigation of adsorbed films of 2 on different substrates. For example, the phase diagram and structure of monolayer 2 films adsorbed on graphite have been studied by various techniques. [@nielsen80; @lauter90; @wiechert91; @vilches92] One of the most remarkable aspects is that the melting temperature $T_m$ of a solid 2 monolayer can be significantly less than bulk 2. [@vilches92] This motivates our interest in the study of the phase diagram of 2 in two mathematical dimensions, which is still relatively unexplored (a systematic study of the one-dimensional phase diagram at zero temperature has been recently carried out [@boronat00]).
Some information has been provided in previous numerical work by Gordillo and Ceperley [@gordillo97], and by Wagner and Ceperley; [@wagner94; @wagner96] for example, it is known that, in the low temperature limit, the equilibrium phase of the system is a triangular solid, with a two dimensional (2D) density $\theta_\circ$ $\approx$ 0.067 2. On the other hand, little is known about the possible existence of a [*metastable*]{} liquid phase, at densities below $\theta_\circ$ (i.e., at negative pressure), which one may be able to investigate experimentally by “stretching" the equilibrium uniform solid phase. Conceivably, such a liquid phase ought to turn superfluid at sufficiently low temperature.
Microscopic calculations for condensed 2 have recently focused on realistic models of adsorbed films, both on graphite[@nho02; @nho03], as well as on alkali metal substrates. [@shi03; @boninsegni04] Here, we present results of a theoretical study of the phase diagram of condensed 2 in 2D, based on path integral Monte Carlo (PIMC) simulations. The temperature range explored is between 1 and 8 K; we extrapolate the results obtained at low temperature to obtain the $T=0$ thermodynamic equation of state.
In agreement with previous calculations, we find that the equilibrium phase of the system at $T$=0 is a triangular crystal; we estimate the equilibrium density $\theta_\circ$=0.0668$\pm$0.0005 2. We also estimate the [*spinodal*]{} density $\theta_s$, namely the lowest density down to which the uniform phase can be stretched, before becoming unstable against density fluctuations (at which point it breaks down into individual clusters). Our computed value of $\theta_s$ is 0.0585$\pm$0.0010 2. Analysis of our numerical results suggests that no metastable liquid phase exists in this system, at low $T$; that is, the system remains a solid all the way down to $\theta_s$, below which it breaks down into solid clusters. We also study the melting of the 2D triangular solid, and determine its melting temperature at approximately 6.8 K.
Although the PIMC method utilized in this work allows for the sampling of permutations of particles, which is essential in order to reproduce in the simulation any effect due to quantum statistics, permutations are not seen to occur, in the temperature range explored. This is because in the crystal phase, the only one observed here at low $T$, permutations are suppressed (as in most solids) by the localization of 2 molecules; at higher temperature, on the other hand, though the crystal melts and molecules are less localized, they also behave more classically, as their thermal wave length decreases. Consistently with permutations not being important, i.e. 2 molecules obeying Boltzmann statistics in the temperature range explored here, no evidence of SF can be seen. [@note]
Our system of interest is modeled as an ensemble of $N$ 2 molecules, regarded as point particles and whose motion is restricted to two physical dimensions. The quantum-mechanical many-body Hamiltonian is the following: $$\label{one}
\hat H = -{\hbar^2\over 2m}\sum_{i=1}^N \nabla_i^2 + \sum_{i<j} V(r_{ij})$$ The system is enclosed in a simulation cell shaped as a parallelogram of area $A$, with periodic boundary conditions in all directions. The density is $\theta$=$N/A$. In Eq. (\[one\]), $m$ is the 2 molecular mass and $V$ is the potential describing the interaction between two 2 molecules, only depending on their relative distance. The Silvera-Goldman potential[@silvera78] was chosen to model the interaction $V$, mostly for consistency with existing, comparable calculations. However, this potential has also been shown to provide an acceptable quantitative description of bulk condensed 2. [@johnson96; @operetto04]
The PIMC method is a numerical (Quantum Monte Carlo) technique that allows one to obtain accurate estimates of physical averages for quantum many-body systems at finite temperature. The only input of a PIMC calculation is the many-body Hamiltonian (\[one\]) (i.e., the potential energy function $V$). Because thorough descriptions of PIMC exist, [@ceperley95] it will not be reviewed here. The main technical details of this calculation are illustrated in Ref. .
Most of the results provided in this manuscript pertain to a system of $N$=64 molecules. We also obtained results for a system of 144 particles at the $T=0$ equilibrium density $\theta_\circ$ and near the spinodal density $\theta_s$. At the beginning of the simulation, molecules are arranged on a triangular lattice. No significant dependence of the estimates on the size of the system can be observed for the physical quantities studied here, with the exception of the energy (see below).
Fig. \[fig1\] shows computed values of the energy per 2 molecule (in K) vs the 2D density (coverage) $\theta$, expressed in 2, for a system of $N$=64 molecules. Energy estimates are found to be nearly independent on temperature, below $T$ $\approx$ 3 K. Thus, the results shown in Fig. \[fig1\] are essentially ground state estimates. A polynomial fit of the data yields an equilibrium density $\theta_\circ$ (corresponding to the minimum of the $\epsilon(\theta)$ curve) of 0.0668$\pm$0.0005 2. This result is in agreement with a recent, independent Quantum Monte Carlo calculation at zero temperature. [@cazorla04] Also shown in Fig. \[fig1\] are the estimates obtained by Gordillo and Ceperley, who carried out PIMC calculations at low temperature on a system of thirty molecules or less. The two calculations agree, insofar as locating the equilibrium density; there is a numerical discrepancy between our calculation and theirs, which can be attributed to the difficulty of determining quantitatively, on a small size system, the contribution to the potential energy associated to the periodic images of the system outside the simulation cell. For, the H$_2$ intermolecular potential has a long-range attractive tail whose overall contribution to the potential energy is considerably greater than that obtained, for example, for condensed Helium at equilibrium, in spite of the fact that the interparticle potentials decay as 1/$r^6$ at long distance in both cases. All of our energy estimates are obtained by computing the above-mentioned contribution to the potential energy by setting the value of the pair correlation function to one outside the simulation cell. On comparing results obtained on systems with $N$=64 and $N$=144 particles, we estimate the systematic error on the energy values furnished here, due to the finite size of the system, to be less than 0.15 K per particle when $N$=64, and less than 0.03 K per particle for $N$=144.
Fig. \[fig2\] shows a typical many-particle configuration, generated by the PIMC simulation, for a system of 144 2 molecules at a density $\theta$=0.067 2 (i.e., close to $\theta_\circ$), at a temperature $T$=2 K. The arrangement of molecules on a triangular lattice is clearly seen. Each fuzzy “cloud" represents a 2 molecule, the typical size of each cloud being a measure of quantum delocalization. There is essentially no overlap of clouds associated to different molecules, which is qualitatively an indication that quantum exchanges are unimportant, and molecules can be regarded as obeying essentially Boltzmann statistics.
Using the polynomial fit to the energy data shown in Fig. \[fig1\], we computed the low temperature chemical potential $\mu(\theta)$ through $$\label{mu}
\mu(\theta)=\epsilon(\theta)+\theta\frac{d\epsilon}{d\theta}$$ The chemical potential is shown in Fig. \[fig3\] (solid line). The equilibrium density $\theta_\circ$ is identified by the condition $\mu(\theta_\circ)=\epsilon(\theta_\circ)$. A second density of interest is the [*spinodal*]{} ($\theta_s$), corresponding to the condition $d\mu/d\theta=0$. Based on our $\epsilon(\theta)$ data, we obtain $\theta_s=0.0585\pm0.0010$ 2.
The spinodal density is the lowest density down to which the uniform phase can be [*stretched*]{} (at negative pressure), before becoming unstable against density fluctuations. At any density lower than $\theta_s$, the uniform film breaks down into “puddles". We can observe this effect directly, by examining many-particle configurations generated by a PIMC simulation of a system of 144 2 molecules; an example is shown in Fig. \[fig2a\], for 2D 2 at a density $\theta$=0.056 2, i.e., slightly lower than $\theta_s$. Such visual observations strongly suggest that such puddles are not liquid; rather, the system retains therein its triangular crystal structure.
In their 1997 calculation, Gordillo and Ceperley made the suggestion that a metastable liquid phase may exist at densities below $\approx$ 0.059 2; however, they did not attempt to locate the spinodal density in their study. [@gordillo97] As it turns out, $\theta_s$ as obtained in this work lies precisely in correspondence of their proposed location of the solid-liquid transition. By direct observation of configurations such as those shown in Figs. \[fig2\] and \[fig2a\], as well as by examining the pair correlation function and the static structure factor, we have confirmed the conclusion of Ref. , namely that the system is a crystal, for $\theta > \theta_s$. Because the uniform phase breaks down below $\theta_s$, we further conclude that no metastable liquid phase of 2 exists in 2D, in the $T$$\to$0 limit. The system remains a solid all the way down to $\theta_s$, and breaks down into solid clusters at lower densities.
In order to study the melting of the equilibrium 2D crystal, we computed the energy per particle $\epsilon(T)$ as a function of temperature for a system of $N$=144 particles. For simplicity, we have kept the density fixed at its $T$=0 equilibrium value, namely 0.067 2. The results are shown in Fig. \[fig3a\].
At low $T$, $\epsilon(T)$ follows the expected $\sim$ $T^3$ behavior, which is consistent with phonons being the low-lying excitations of the 2D quantum crystal. The extrapolated $T$=0 energy value is -23.25$\pm$0.05 K, which is in satisfactory agreement with the value of -23.4 K recently reported by Cazorla and Boronat, who carried out a $T$=0 calculation. We estimate our uncertainty in the determination of the potential energy to amount to less than 0.03 K per molecule, on a system of this size. There is a difference of approximately 0.1 K between the above, extrapolated energy value, and that obtained on a system of 64 particles, the latter being the higher one.
A sudden change of slope is seen to occur at $T_m$=6.8 K, where the specific heat $c(T)=d\epsilon/dT$ takes on a peak, which corresponds to the melting of the 2D crystal. This value of the melting temperature is comparable to that found in studies of 2 films on alkali metal substrates [@boninsegni04], and slightly higher than that of a 2 surface (computed by PIMC), [@wagner96] and of an adsorbed 2 monolayer .[@vilches92] This suggests that melting may occur at a lower temperature in three dimensions, as zero-point motion of molecules in the direction perpendicular to the substrate promotes evaporation. In any case, these melting temperatures are still too high to expect that a superfluid transition of 2 may be observed.
In conclusion, we have carried out extensive PIMC studies of 2D condensed 2, determining its low temperature equation of state, inferring its equilibrium and spinodal densities, as well as the melting temperature of the equilibrium system. Consistently with previous study, we found the system to be a triangular crystal at equilibrium; this 2D crystal melts at a temperature of approximately 6.8 K.
We have found no evidence of any metastable liquid phase at low $T$. The system is found to remain a solid all the way down to the spinodal density, below which it breaks down into solid clusters. This result, perhaps unexpected, raises some doubts about the long term prospects of observing a superfluid phase of 2 upon supercooling the liquid. Observing novel phases of 2, including a (superfluid) liquid, may require achieving a substantial renormalization of the interaction of 2 molecules, possibly through their interaction with the surface electrons of a metal substrate, or of a nanostructure. An alternate route might be that suggested in Ref. , namely stabilizing a liquid at low temperature by means of an external potential incommensurate with the crystal structure of 2.
This work was supported in part by the Petroleum Research Fund of the American Chemical Society under research grant 36658-AC5, and by the Natural Science and Engineering Research Council of Canada under research grant G121210893.
[99]{} V. L. Ginzburg and A. A. Sobyanin, JETP Letters [**15**]{}, 242 (1972). M. Bretz and A. L. Thomson, Phys. Rev. B [**24**]{}, 467 (1981).
G. M. Seidel, H. J. Maris, F. I. B. Williams and J. G. Cardon, Phys. Rev. Lett. [**56**]{}, 2380 (1986).
H. J. Maris, G. M. Seidel and F. I. B. Williams, Phys. Rev. B [**36**]{}, 6799 (1987).
M. Schindler, A. Dertinger, Y. Kondo and F. Pobell, Phys. Rev. B [**53**]{}, 11451 (1996).
M. Nielsen, J. P. McTague and L. Passell in [*Phase Transitions in Surface Films*]{}, edited by J. Dash and J. Ruvalds (Plenum, New York, 1980).
H. J. Lauter, H. Godfrin, V. L. P. Frank and P. Leiderer in [*Phase Transitions in Surface Films 2*]{}, edited by H. Taub, G. Torzo, H. J. Lauter and S. C. Fain Jr. (Plenum, New York, 1990).
H. Wiechert in [*Excitations in Two-Dimensional and Three-Dimensional Quantum Fluids*]{}, edited by A. F. G. Wyatt and H. J. Lauter (Plenum, New York, 1991).
F. C. Liu, Y. M. Liu and O. E. Vilches, J. Low Temp. Phys. [**89**]{}, 649 (1992). M. C. Gordillo, J. Boronat and J. Casulleras, Phys. Rev. Lett. [**85**]{}, 02348 (2000). M. C. Gordillo and D. M. Ceperley, Phys. Rev. Lett. [**79**]{}, 3010 (1997).
M. Wagner and D. M. Ceperley, J. Low Temp. Phys. [**94**]{}, 161 (1994).
M. Wagner and D. M. Ceperley, J. Low Temp. Phys. [**102**]{}, 275 (1996).
K. Nho and E. Manousakis, [ Phys. Rev. B]{} [**65**]{}, 115409 (2002). K. Nho and E. Manousakis, Phys. Rev. B [**67**]{}, 195411 (2003). W. Shi, J. K. Johnson and M. W. Cole, Phys. Rev. B [**68**]{}, 125401 (2003). M. Boninsegni, Phys. Rev. B [**70**]{}, 125405 (2004). We have not attempted to search for a possible superfluid phase of [*solid*]{} 2. The absence of many-particle permutations observed in this work leads us to believe that, if such a phase exists, it must be at significantly lower temperatures than those considered here. I. F. Silvera and V. V. Goldman, [J. Chem. Phys.]{} [**69**]{}, 4209 (1978). Q. Wang, J. K. Johnson and J. Q. Broughton, [Mol. Phys.]{} [**89**]{}, 1105 (1996). F. Operetto and F. Pederiva, Phys. Rev. B [**69**]{}, 024203 (2004). D. M. Ceperley, [Rev. Mod. Phys.]{} [**67**]{}, 279 (1995).
S. Jang, S. Jang and G. A. Voth, J. Chem. Phys. [**115**]{}, 7832 (2001). E. L. Pollock and D. M. Ceperley, Phys. Rev. B [**30**]{}, 2555 (1984). C. Cazorla and J. Boronat, J. Low Temp. Phys. [**134**]{}, 43 (2004).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $G$ be a connected graph and $S$ be a set of vertices. The $h$-extra connectivity of $G$ is the cardinality of a minimum set $S$ such that $G-S$ is disconnected and each component of $G-S$ has at least $h+1$ vertices. The $h$-extra connectivity is an important parameter to measure the reliability and fault tolerance ability of large interconnection networks. The $h$-extra connectivity for $h=1,2$ of $k$-ary $n$-cube are gotten by Hsieh et al. in \[Theoretical Computer Science, 443 (2012) 63-69\] for $k\geq 4$ and Zhu et al. in \[Theory of Computing Systems, arxiv.org/pdf/1105.0991v1 \[cs.DM\] 5 May 2011\] for $k=3$. In this paper, we show that the $h$-extra connectivity of the $3$-ary $n$-cube networks for $h=3$ is equal to $8n-12$, where $n\geq 3$.'
author:
- |
Meimei Gu, Rongxia Hao$^{\ast}$\
[*Department of Mathematics, Beijing Jiaotong University, 100044, China*]{}
title: '$3$-extra connectivity of $3$-ary $n$-cube networks'
---
Interconnection networks, $3$-ary $n$-cube networks, extra connectivity, conditional connectivity
Introduction
============
It is well known that a topological structure of an interconnection network can be modeled by a loopless undirected graph $G=(V,E)$, where the vertex set $V$ represents the processors and the edge set $E$ represents the communication links. In this paper, we use graphs and networks interchangeably.
Let $G$ be a simple undirected graph. Two vertices $v_{1},v_{2}$ in $V(G)$ are said to be $adjacent$ if and only if $(v_{1},v_{2})\in E(G)$. The $neighborhood$ of a vertex $u$ in $G$ is the set of all vertices adjacent to $u$ in $V(G)$, denoted by $N_{G}(u)$. The cardinality $|N_{G}(u)|$ represents the $degree$ of $u$ in $G$, denoted by $d_{G}(u)$(or simply $d(u)$), $\delta(G)$ the $minimum$ $degree$ of $G$. For a vertex subset $S\subseteq V(G)$, the neighborhood of $S$ in $G$ is $N_{G}(S)=(\bigcup_{u\in S}N_{G}(u))-S$. A $subgraph$ of $G=(V,E)$ is a graph $H=(V^{'},E^{'})$ such that $V^{'}\subseteq V$ and $E^{'}\subseteq E$. For a subgraph $H$ of $G$, $N_{G}(V(H))$ can be simplified as $N_{G}(H)$. For a subset $S$ of $V(G)$, the $induced$ $subgraph$ of $S$, written by $G[S]$, is a subgraph of $G$, whose vertex set is $S$ and an edge $e\in G[S]$ if and only if both end vertices of $e$ are in $S$. $N[S]$ is also used to denote the induced subgraph of $N_{G}(S)\bigcup S$. A subset $S\subseteq V(G)$ is a vertex cut if $G-S$ is disconnected. The components of $G$ are its maximal connected subgraphs.
A $path$ $P_{k}=(v_{1},v_{2},\cdots,v_{k})$ for $k\geq2$ in a graph $G$ is a sequence of distinct vertices such that any two consecutive vertices are adjacent, and $v_{1}$ and $v_{k}$ are the $end$-$vertices$ of the path. For convenience, use $P_{t}$ to denote a path of $t$ vertices. A path of $G$ of length $n$ will be called an $n$-$path$. A $cycle$ $C_{k}=(v_{1},v_{2},\cdots,v_{k},v_{1})$ for $k\geq 3$ is a sequence of vertices in which any two consecutive vertices are adjacent, where $v_{1},v_{2},\cdots,v_{k}$ are all distinct. A cycle of $G$ of length $n$ will be called an $n$-$cycle$. A $complete$ $graph$ of $n$ vertices, denoted by $K_{n}$, is a simple graph whose vertices are pairwise adjacent.
Let $G$ and $H$ be two graphs. $G$ and $H$ are $distinct$ if their vertex sets are different, and $disjoint$ if they have no common vertices. An $isomorphism$ from a graph $G$ to a graph $H$ is a bijection function $\pi: V(G) \rightarrow V(H)$ such that $(u, v)\in E(G)$ if and only if $(\pi(u), \pi(v))\in E(H)$. We write $G\cong H$ if there is an isomorphism from $G$ to $H$.
The $connectivity$ $\kappa(G)$ of a connected graph $G$ is the minimum number of vertices removed to get the graph disconnected or trivial. A graph $G$ is said to be $super$ $connected$, or simply super-$\kappa$, if every minimum vertex cut creates exactly two components, one of which is a singleton. Connectivity as a measure of reliability underestimates the fault tolerance ability of these multiprocessor systems.
Conditional connectivity introduced by Harary [@Harary] can be used to better measure the reliability of multiprocessor systems. If any component of $G-S$ has some property $\mathcal{P}$, where $S$ is a vertex cut of $G$, then $S$ is called a $\mathcal{P}$-$vertex$ $cut$. The $\mathcal{P}$-$conditional$ $connectivity$ of $G$ is defined to be the minimum over all cardinalities of $\mathcal{P}$-vertex cuts. J. F$\grave{a}$brega and M.A. Fiol introduced the $extra$ $connectivity$ of interconnection networks as follows. A vertex set $S\subseteq V(G)$ is called to be an $h$-$extra$ $vertex$ $cut$ if $G-S$ is disconnected and every component of $G-S$ has at least $h+1$ vertices. The $h$-$extra$ $connectivity$ of $G$, denoted by $\kappa_{h}(G)$, is defined as the cardinality of a minimum $h$-extra vertex cut, if exist. An $(h+1)$-extra vertex cut of a graph $G$ is clearly an $h$-extra vertex cut, and thus $\kappa_{h}(G)\leq \kappa_{h+1}(G)$. Extra connectivity is an example of $\mathcal{P}$-conditional connectivity.
It is obvious that $\kappa_{0}(G)=\kappa(G)$ for any graph $G$ that is not a complete graph. In particular, the 1-extra vertex cut is called as the extra vertex cut and the 1-extra connectivity is called as the extra connectivity. The problem of determining the $h$-extra connectivity of numerous networks has received a great deal of attention in recent years. Interested readers may refer to [@Balbuena],[@Esfahanian.S],[@Yang] or others for further details.
The $k$-$ary$ $n$-$cube$ $Q_{n}^{k}$, proposed by Scott and Goodman [@Scott], is one of the most popular interconnection networks. Some properties of the $k$-ary $n$-cube network have been investigated, for example, fault diameter [@Day.], pan-connectivity [@Lin] etc. Moreover, many interconnection networks can be viewed as the subclasses of $Q_{n}^{k}$, including the cycle, the torus and the hypercube. The $h$-extra connectivity for $h=1,2$ of $k$-ary $n$-cube are gotten by Hsieh et al. in [@Hsieh] for $k\geq4$ and Zhu et al. in [@Zhu] for $k=3$. In this paper, we show that the $3$-extra connectivity of the $3$-ary $n$-cube network is $8n-12$ for $n\geq 3$.
Definitions which not been given here are referred to [@Bondy] and [@Xu]. The remainder of this paper is organized as follows. In Section $2$, the $k$-ary $n$-cube and its properties will be given. Section $3$ discusses the $3$-extra connectivity of the $3$-ary $n$-cube. Section 4 concludes the paper. Last is acknowledgements.
The $k$-ary $n$-cube and its properties
=======================================
The $k$-$ary$ $n$-$cube$, denoted by $Q_{n}^{k}$, where $k\geq2$ and $n\geq1$ are integers, is a graph consisting of $k^{n}$ vertices. Each of these vertices has the form $u=u_{n-1}u_{n-2}\cdots u_{0}$ where $u_{i}\in \{0,1,2,\cdots,k-1\}$ for $0\leq i\leq n-1$. Two vertices $u=u_{n-1}u_{n-2}\cdots u_{0}$ and $v=v_{n-1}v_{n-2}\cdots v_{0}$ in $Q_{n}^{k}$ are adjacent if and only if there exists an integer $j$, where $0\leq j\leq n-1$, such that $u_{j}=v_{j}\pm 1(mod$ $k)$, and $u_{i}=v_{i}$ for every $i\in \{0,1,2,\cdots,j-1,j+1,\cdots,n-1\}$. In this case, $(u, v)$ is a $j$-dimensional edge. For clarity of presentation, $``$(mod $k)"$ does not appear in similar expressions in the remainder of the paper. Obviously, $Q_{1}^{k}$ is a cycle of length $k$, $Q_{n}^{2}$ is an $n$-dimensional hypercube, $Q_{2}^{k}$ is a $k\times k$ wrap-around mesh. This study considers $3$-ary $n$-cube, $Q_{2}^{3}$ and $Q_{3}^{3}$ are illustrated in Fig.1.
{height="5cm" width="12cm"}
[**Fig 1.**]{} The illustration of $Q_{2}^{3}$ and $Q_{3}^{3}$
It is possible to partition $Q_{n}^{k}$ over $j$-dimension, for a $j\in \{0,1,2,\cdots,n-1\}$, into $k$ disjoint subcubes, denoted by $Q_{n-1}^{k}[0],Q_{n-1}^{k}[1],\cdots,Q_{n-1}^{k}[k-1]$ by deleting all the $j$-dimensional edges from $Q_{n}^{k}$. For convenience, abbreviate these as $Q[0],Q[1],\cdots, Q[k-1]$ if there is no ambiguity. Moreover, $Q[i]$ for $0\leq i\leq k-1$ is isomorphic to $k$-ary $(n-1)$-cube and there are $k^{n-1}$ edges between $Q[i]$ and $Q[i+1]$. For each vertex $u \in V(Q[i])$, the $right$ $neighbor$ (respectively, $left$ $neighbor$) of $u$, denoted by $u_{R}$ (respectively, $u_{L}$), is the $outer$ $neighbor$ of $u$ in $Q[i+1]$ (respectively, $Q[i-1]$).
The following useful properties of $Q_{n}^{k}$ which will be used later on can be found in [@Bose],[@Ghozati],[@Scott],[@Xu].
([@Scott]) For $n\geq 1$, $Q_{n}^{k}$ is $n$-regular and has $nk^{n-1}$ edges when $k=2$; $Q_{n}^{k}$ is $2n$-regular and has $nk^{n}$ edges when $k\geq3$.
([@Ghozati; @Xu]) For $n\geq 2$, $\kappa(Q_{n}^{k})=\delta(Q_{n}^{k})=2n$ when $k\geq3$; $\kappa(Q_{n}^{k})=\delta(Q_{n}^{k})=n$ when $k=2$. Moreover, $Q_{n}^{k}$ is super-connected for $n\geq 2$.
([@Ghozati]) For $k\geq2, n\geq 2$, $Q_{n}^{k}$ is vertex transitive and edge transitive.
([@Bose]) For $k\geq3, n\geq 2$, $Q_{n}^{k}$ can be divided into $k$ disjoint subgraphs, each subgraph is isomorphic to $Q_{n-1}^{k}$. The two outer neighbors of every vertex in $Q_{n}^{k}$ are in different subgraphs which contained in $\{Q[i]:0\leq i\leq k-1\}$.
Main result
===========
The $h$-extra connectivity for $h=1,2$ of $k$-ary $n$-cube are gotten by Hsieh et al. in [@Hsieh] for $k\geq 4$ and Zhu et al. in [@Zhu] for $k=3$. Nevertheless, the $h$-extra for $h\geq 3$ connectivity of $3$-ary $n$-cube has not been obtained yet. In this section, the $3$-extra connectivity of the $3$-ary $n$-cube will be determined.
([@Zhu]) Any two adjacent vertices in $Q_{n}^{3}$ have exactly one common neighbor for $n\geq 1$; If any two nonadjacent vertices in $Q_{n}^{3}$ have common neighbors, they have exactly two common neighbors for $n\geq 2$.
([@Zhu]) $\kappa_{1}(Q_{n}^{3})=4n-3$ for $n\geq 2$ and $\kappa_{2}(Q_{n}^{3})=6n-7$ for $n\geq 3$.
Suppose that $F\subseteq V(Q_{n}^{3})$ with $|F|\leq 4n-4$ is a vertex cut of $ Q_{n}^{3}$ for $n\geq 2$, then $Q_{n}^{3}-F$ has two components, one of which is a singleton.
This lemma can be proved by using induction on $n$.
For $n=2$, the graph is $Q_{2}^{3}$, shown in Fig.1, in this case, $|F|\leq 4n-4=4$; Since $F$ is a vertex cut of $ Q_{n}^{3}$, then $|F|\geq \kappa(Q_{n}^{3})=2n=4$. Thus $|F|=4$, $F$ is the minimum vertex cut. By Lemma 2.2, $Q_{2}^{3}-F$ has two components, one of which is a singleton.
Assume now $n\geq 3$ and the lemma is true for $Q_{n-1}^{3}$. Recall that $Q[i]$ is $2(n-1)$-regular and is isomorphic to $Q_{n-1}^{3}$ for any $i\in\{0,1,2\}$. Let $F_{i}=F \bigcap V(Q[i])$, so $\sum_{i=0}^{2}|F_{i}|\leq4n-4$. Then at least one of $|F_{0}|$, $|F_{1}|$ and $|F_{2}|$ is strictly less than $2(n-1)$ since $F_{0}\bigcap F_{1}\bigcap F_{2}=\emptyset$ and $|F_{0}|+|F_{1}|+|F_{2}|=|F|\leq 4n-4 <6n-6$ for $n\geq2$. Without loss of generality, suppose $|F_{0}|\leq 2(n-1)-1$. We consider the following cases.
[**Case 1.**]{} For any $i\in \{1,2\}$, $|F_{i}|\leq 2(n-1)-1$.
Since $\kappa(Q_{n-1}^{3})=2(n-1)$, then $Q[i]-F_{i}$ is connected. There are $3^{n-1}$ edges between $Q[i]$ and $Q[j]$ for $0\leq i\neq j\leq2$, and $3^{n-1}>2(n-1)-1+2(n-1)-1=4n-6$ for $n\geq2$, then $Q[i]-F_{i}$ is connected to $Q[j]-F_{j}$. By the arbitrary of $i$ and $j$, $Q_{n}^{3}-F$ is connected which is a contradiction.
[**Case 2.**]{} $|F_{1}|\leq 2(n-1)-1$ and $2(n-1)\leq|F_{2}|\leq 4(n-1)-4$ (The case of $2(n-1)\leq|F_{1}|\leq 4(n-1)-4$, $|F_{2}|\leq 2(n-1)-1$ is the similar discussion).
By the similar argument to Case 1, $Q[0]-F_{0}$ is connected to $Q[1]-F_{1}$. $Q[i]-F_{i}$ for $i\in\{0,1\}$ belong to a same component of $Q_{n}^{3}-F$, denoted by $C$. If $F_{2}$ is not a vertex cut of $Q[2]$, then $Q[2]-F_{2}$ is connected, there are at least $3^{n-1}-(4n-8)-(2n-3)\geq2$ edges between $Q[i]-F_{i}$ for each $i\in\{0, 1\}$ and $Q[2]-F_{2}$ for $n\geq3$, thus $Q[2]-F_{2}$ is contained in $C$, it implies that $Q_{n}^{3}-F$ is connected which is a contradiction. So $F_{2}$ is a vertex cut of $Q[2]$. By inductive hypothesis, $Q[2]-F_{2}$ has two components, one of which is a singleton. Let $B$ be the largest component of $Q[2]-F_{2}$. We will show that $B$ is connected to $C$. Note that $B$ has at least $3^{n-1}-(4n-8)-1$ vertices, and $2[3^{n-1}-(4n-8)-1]$ different outer neighbors. Since there are at most $4n-4-(2n-2)=2n-2$ vertices in $F-F_{2}$, and $2[3^{n-1}-(4n-8)-1]>2n-2$ for $n\geq3$, there must be an edge between $B$ and $C$, then $B$ is contained in $C$. Thus $Q_{n}^{3}-F$ has two components, one of which is a singleton.
[**Case 3.**]{} For some $i\in \{1,2\}$, $|F_{i}|\geq 4n-7$.
Since $2(4n-7)>4n-4$ for $n\geq 3$, there is only one $i\in\{1,2\}$ such that $|F_{i}|\geq 4n-7$. Without loss of generality, let $i=1$. Since $|F|\leq 4n-4$, then $\sum_{j\neq 1}|F_{j}|\leq3$. By the the similar argument as Case 1, $Q[j]-F_{j}$ for $j\in \{0,2\}$ belong to a same component of $Q_{n}^{3}-F$, denoted by $C$. If a component, denoted by $B$, of $Q[1]-F_{1}$ has an edge, then its endpoints have exactly four distinct outer neighbors, so $B$ is contained in $C$. Thus only singletons of $Q[1]-F_{1}$ may not be contained in $C$. Since each singleton of $Q[1]-F_{1}$ has two outer neighbors, which are all different, there can be only one such singleton. Hence $Q_{n}^{3}-F$ has two components, one of which is a singleton.
[**Case 4.**]{} For any $i\in \{1,2\}$, $4n-7\geq |F_{i}|\geq 2(n-1)$.
Since $|F|\leq 4n-4$, then $|F_{0}|\leq 0$, that is $|F_{0}|=0$. So $Q[0]-F_{0}$ is connected, denoted it by $C$. Every vertex of $Q[i]-F_{i}$ has one neighbor outside $Q[1]\bigcup Q[2]$ for each $i\in \{1,2\}$, so any component of $Q[i]-F_{i}$ is connected to $C$. Then $Q_{n}^{3}-F$ is connected which is a contradiction.
The lemma is completed.
Suppose that $F\subseteq V(Q_{n}^{3})$ with $|F|\leq 6n-8$ is a vertex cut of $ Q_{n}^{3}$ for $n\geq 3$, then $Q_{n}^{3}-F$ either has two components, one of which is a singleton, or an edge; or has three components, two of which are singletons.
Let $F_{i}=V(Q[i])\bigcap F$ for $i\in\{0,1,2\}$, so $\sum_{i=0}^{2}|F_{i}|=|F|\leq 6n-8$. Since $F_{0}\bigcap F_{1}\bigcap F_{2}=\emptyset$ and $|F_{0}|+|F_{1}|+|F_{2}|=|F|\leq 6n-8 <6n-6$ for $n\geq3$, then at least one of $|F_{0}|$, $|F_{1}|$ and $|F_{2}|$ is strictly less than $2(n-1)$. Without loss of generality, suppose $|F_{0}|\leq 2(n-1)-1$. We will prove the lemma by using induction on $n$.
[**Case 1.**]{} We prove the result for $Q_{n}^{3}$ with $n=3$.
The graph is $Q_{3}^{3}$, shown in Fig.1. Let $F\subseteq V(Q_{3}^{3})$ with $|F|\leq 6n-8=10$ be a vertex cut of $ Q_{3}^{3}$. We consider the following subcases.
[**Subcase 1.1.**]{} For any $i\in\{1,2\}$, $|F_{i}|\leq3$.
Since $\kappa(Q_{2}^{3})=4$, then $Q[i]-F_{i}$ is connected. There are $3^{2}-|F_{i}|-|F_{j}|\geq 3^{2}-2\times3=3$ edges between $Q[i]-F_{i}$ and $Q[j]-F_{j}$ for $0\leq i\neq j\leq 2$, thus $Q[i]-F_{i}$ is connected to $Q[j]-F_{j}$. By the arbitrary of $i$ and $j$, $Q_{3}^{3}-F$ is connected which is a contradiction.
[**Subcase 1.2.**]{} $|F_{1}|\leq3$ and $4=2(n-1)\leq |F_{2}|\leq6(n-1)-8=4$, that is $|F_{2}|=4$ (The case of $|F_{1}|=4$ and $|F_{2}|\leq3$ is the similar discussion).
By the similar argument to Subcase 1.1, $Q[0]-F_{0}$ is connected to $Q[1]-F_{1}$. $Q[i]-F_{i}$ for $i\in\{0,1\}$ belong to a same component of $Q_{n}^{3}-F$, denoted by $C$. If $Q[2]-F_{2}$ is connected, since there are at least $3^{2}-|F_{i}|-|F_{2}|\geq3^{2}-4-3=2$ edges between $Q[i]-F_{i}$ and $Q[2]-F_{2}$ for $i\in\{0,1\}$, then $Q[2]-F_{2}$ is contained in $C$, thus $Q_{3}^{3}-F$ is connected which is a contradiction. Hence $Q[2]-F_{2}$ is disconnected, since $|F_{2}|=\kappa(Q_{2}^{3})=4$, $F_{2}$ is a minimum vertex cut of $Q[2]$, by Lemma 2.2, $Q[2]-F_{2}$ has two components, one of which is a singleton. Let $B$ be the largest component of $Q[2]-F_{2}$. Note that $B$ has $4$ vertices, which has $8$ different outer neighbors. Since there are at most $10-4=6$ vertices in $F-F_{2}$, there must be two edges between $B$ and $C$, then $B$ is contained in $C$. Thus $Q_{3}^{3}-F$ has two components, one of which is a singleton.
[**Subcase 1.3.**]{} For some $i\in \{1,2\}$, $|F_{i}|\geq 6n-13=5$.
Without loss of generality, suppose $|F_{2}|\geq 5$. Since $|F|\leq 10$ and $|F_{0}|\leq3$, then $\sum_{j\neq 2}|F_{j}|\leq5$, $|F_{1}|\leq2$. Since $\kappa(Q_{2}^{3})=4$, then $Q[i]-F_{i}$ is connected for $i\in\{0,1\}$. By the similar argument as Subcase 1.1, $Q[0]-F_{0}$ is connected to $Q[1]-F_{1}$. $Q[i]-F_{i}$ for $i\in\{0,1\}$ belong to a same component of $Q_{n}^{3}-F$, denoted by $C$. Every vertex of $Q[2]-F_{2}$ has two outer neighbors, any component with more than two vertices of $Q[2]-F_{2}$ is contained in $C$. Thus $Q_{3}^{3}-F$ either has two components, one of which is a singleton, or an edge; or has three components, two of which are singletons.
[**Subcase 1.4.**]{} $|F_{1}|\geq4, |F_{2}|\geq 4$.
Since $|F|\leq 10$, then $|F_{0}|=|F|-|F_{1}|-|F_{2}|\leq 2$. So $Q[0]-F_{0}$ is connected. Let $C$ be a component of $Q_{3}^{3}-F$ which contains $Q[0]-F_{0}$. Every vertex of $Q[i]-F_{i}$ has one neighbor outside $Q[1]\bigcup Q[2]$ for $i\in\{1,2\}$, so any component with more than two vertices of $Q[i]-F_{i}$ is contained in $C$. Thus $Q_{3}^{3}-F$ either has two components, one of which is a singleton, or an edge; or has three components, two of which are singletons.
Then the result holds for $n=3$. In what follows, assume that $n\geq 4$ and the result holds for $ Q_{n-1}^{3}$.
[**Case 2.**]{} We prove the result for $Q_{n}^{3}$ and $n\geq 4$. We consider the following subcases.
[**Subcase 2.1.**]{} For any $i\in\{1,2\}$, $|F_{i}|\leq2(n-1)-1$.
Since $\kappa(Q_{n-1}^{3})=2(n-1)$, then $Q[i]-F_{i}$ is connected. There are $3^{n-1}$ edges between $Q[i]$ and $Q[j]$ for $0\leq i\neq j\leq2$, and $3^{n-1}>2(n-1)-1+2(n-1)-1=4n-6$ for $n\geq3$, then $Q[i]-F_{i}$ is connected to $Q[j]-F_{j}$. By the arbitrary of $i$ and $j$, $Q_{n}^{3}-F$ is connected which is a contradiction.
[**Subcase 2.2.**]{} $|F_{1}|\leq 2(n-1)-1$ and $2(n-1)\leq|F_{2}|\leq 6n-14$ (The case of $2(n-1)\leq|F_{1}|\leq 6n-14$ and $|F_{2}|\leq 2(n-1)-1$ is the similar discussion).
By the similar argument as Subcase 2.1, $Q[0]-F_{0}$ is connected to $Q[1]-F_{1}$. The component which contains $Q[i]-F_{i}$ for $i\in\{0,1\}$ of $Q_{n}^{3}-F$ is denoted by $C$. If $Q[2]-F_{2}$ is connected, there are at least $3^{n-1}-(6n-14)-(2n-3)\geq12$ edges between $Q[i]-F_{i}$ (for $i\in\{0,1\}$) and $Q[2]-F_{2}$ for $n\geq4$, thus $Q[2]-F_{2}$ is connected to $C$. It implies that $Q_{n}^{3}-F$ is connected which is a contradiction. Hence $F_{2}$ is a vertex cut of $Q[2]$. By inductive hypothesis, there are at most three components in $Q[2]-F_{2}$, with two of them having at most two vertices in total. Let $D$ be the largest component of $Q[2]-F_{2}$. $D$ has at least $3^{n-1}-(6n-14)-2$ vertices, by Lemma 2.4, $D$ has at least $2[3^{n-1}-(6n-14)-2]$ distinct outer neighbors. Since $|F_{0}|+|F_{1}|=|F|-|F_{2}|\leq6n-8-(2n-2)=4n-6$ and $2[3^{n-1}-(6n-14)-2]>4n-6$ for $n\geq4$, then there exists at least one edge between $D$ and $C$. Hence $D$ is contained in $C$. Thus the smallest component at most contains two vertices, the result is proved in this case.
[**Subcase 2.3.**]{} For some $i\in \{1,2\}$, $|F_{i}|\geq 6n-13$.
Without loss of generality, suppose $|F_{2}|\geq 6n-13$. Then $\sum_{j\neq 2}|F_{j}|=|F|-|F_{2}|\leq 5\leq2(n-1)-1$ for $n\geq4$. Since $\kappa(Q_{n-1}^{3})=2(n-1)\geq6$ for $n\geq4$, then $Q[i]-F_{i}$ is connected for each $i\in\{0,1\}$. By the similar argument as Subcase 2.1, $Q[0]-F_{0}$ is connected to $Q[1]-F_{1}$. Let $C$ be the component of $Q_{n}^{3}-F$ which contains $Q[i]-F_{i}$ for $i\in\{0,1\}$. Every vertex has two outer neighbors, any component of $Q[2]-F_{2}$ with more than two vertices is contained in $C$. Thus $Q_{n}^{3}-F$ either has two components, one of which is a singleton, or an edge; or has three components, two of which are singletons.
Next we only need to consider the case that $F_{i}$ is greater than $2(n-1)-1$ for each $i\in\{1,2\}$.
[**Subcase 2.4.**]{} $|F_{1}|\geq|F_{2}|\geq 2(n-1)$ (The case of $|F_{2}|\geq|F_{1}|\geq 2(n-1)$ is the similar discussion).
Clearly, $|F_{2}|\leq|F_{1}|\leq 6n-8-2(n-1)=4n-6$. If $|F_{1}|=4n-6$, then $|F_{2}|=2n-2$, $|F_{0}|=0$. So $Q[0]-F_{0}$ is connected, denoted it by $C$. Every vertex of $Q[i]-F_{i}$ has one neighbor outside $Q[1]\bigcup Q[2]$ for each $i\in\{1,2\}$, so any component of $Q[i]-F_{i}$ is contained in $C$. Then $Q_{n}^{3}-F$ is connected which is a contradiction.
Now we consider $|F_{1}|\leq 4n-7$. First consider $|F_{1}|=4n-7$, if $|F_{2}|=2n-1$, we have done by the previous argument. So the left case is $|F_{2}|=2n-2$, $|F_{0}|=1$. Thus $Q[0]-F_{0}$ is connected, denoted it by $C$. Since every vertex in $Q[1]$ and $Q[2]$ has an outer neighbors in $Q[0]$ and $|F_{0}|=1$, then at most one vertex can be disconnected from $C$ in $Q_{n}^{3}-F$, hence $Q_{n}^{3}-F$ has two components, one of which is a singleton.
Finally, we consider those cases where $|F_{2}|\leq|F_{1}|\leq 4n-8=4(n-1)-4$. This case is divided into three subcases.
[**Subcase 2.4.1.**]{} Both $Q[1]-F_{1}$ and $Q[2]-F_{2}$ are connected.
In this case, there are at least $3^{n-1}-|F_{i}|-|F_{0}|\geq 3^{n-1}-(4n-8)-(2n-3)\geq14$ edges between $Q[0]-F_{0}$ and $Q[i]-F_{i}$ for each $i\in \{1,2\}$ and $n\geq4$, thus $Q[i]-F_{i}$ is connected to $Q[0]-F_{0}$. Hence $Q_{n}^{3}-F$ is connected which is a contradiction.
[**Subcase 2.4.2.**]{} Only one of $Q[1]-F_{1}$ and $Q[2]-F_{2}$ is connected.
Without loss of generality, assume that $Q[1]-F_{1}$ is connected and $Q[2]-F_{2}$ is disconnected. By the similar argument as Subcase 2.4.1, $Q[1]-F_{1}$ is connected to $Q[0]-F_{0}$. Let $C$ be the component of $Q_{n}^{3}-F$ which contains $Q[0]-F_{0}$ and $Q[1]-F_{1}$. Since $Q[2]-F_{2}$ is disconnected and $|F_{2}|\leq 4n-8=4(n-1)-4$, by Lemma 3.3, $Q[2]-F_{2}$ has two components, one of which is a singleton. Let $D$ be the largest component of $Q[2]-F_{2}$. Note that $D$ has at least $3^{n-1}-(4n-8)-1$ vertices, and has at least $3^{n-1}-(4n-8)-1$ neighbors in $Q[0]$, since $3^{n-1}-(4n-8)-1>2(n-1)-1\geq|F_{0}|$ for $n\geq 4$, thus $D$ is contained in $C$. Hence $Q_{n}^{3}-F$ has two components, one of which is a singleton.
[**Subcase 2.4.3.**]{} Both $Q[1]-F_{1}$ and $Q[2]-F_{2}$ are disconnected.
By Lemma 3.3, $Q[i]-F_{i}$ for each $i\in\{1,2\}$ has two components, one of which is a singleton, denoted by $x_{i}$. Since $|F_{0}|\leq2(n-1)-1$, then $Q[0]-F_{0}$ is connected. Let $B_{i}$ be the largest component of $Q[i]-F_{i}$ for each $i\in\{1,2\}$. Note that $B_{i}$ has at least $3^{n-1}-(4n-8)-1$ vertices, and has at least $3^{n-1}-(4n-8)-1$ outer neighbors in $Q[0]$, since $3^{n-1}-(4n-8)-1>2(n-1)-1$ for $n\geq4$, thus $B_{i}$ is connected to $Q[0]-F_{0}$. Let $C$ be the component of $Q_{n}^{3}-F$ which contains $B_{i}$ and $Q[0]-F_{0}$.
If both $x_{1}$ and $x_{2}$ are contained in $C$, then $Q_{n}^{3}-F$ is connected which is a contradiction. If only one of $x_{i}$ is contained in $C$, then $Q_{n}^{3}-F$ has two components, one of which is a singleton. Besides, the two singletons in $Q[1]-F_{1}$ and $Q[2]-F_{2}$ may either remain singleton components in $Q_{n}^{3}-F$ or they could belong to one component of $Q_{n}^{3}-F$, forming a $K_{2}$. Hence $Q_{n}^{3}-F$ either has two components, one of which is an edge; or has three components, two of which are singletons. The result holds in this case.
The proof of the lemma is finished.
For $n\geq 3$, $\kappa_{3}(Q_{n}^{3})\leq 8n-12$.
Let $u=(0,0,0,0,\cdots,0)$, $v=(0,1,0,0,\cdots,0)$, $w=(0,1,1,0,\cdots,0)$ and $t=(0,0,1,0,\cdots,0)$ be four vertices in $Q_{n}^{3}$, $P_{4}=uvwt \in Q_{n}^{3}$ be a path of length three. Let $F=N(P_{4})$, obviously, $Q_{n}^{3}-F$ is disconnected. Note that $(u,v,w,t,u)$ is a cycle of length four. By Lemma 3.1 and the structure of $Q_{n}^{3}$, $u$ has a neighbor set $X_{1}$ with the order $2n-2$ in $V(Q_{n}^{3})-V(P_{4})$; Since $u$ and $v$ have one common neighbor $x_{1}=(0,2,0,0,\cdots,0)$, $v$ has a neighbor set $X_{2}$ with the order $2n-2-1=2n-3$ in $V(Q_{n}^{3})-V(P_{4})-X_{1}$; Since $w$ and $u$ have two common neighbors $v$ and $t$, $w$ and $v$ have one common neighbor $x_{2}=(0,1,2,0,\cdots,0)$, $w$ has a neighbor set $X_{3}$ with the order $2n-2-1=2n-3$ in $V(Q_{n}^{3})-V(P_{4})-X_{1}-X_{2}$; Since $t$ and $u$ have one common neighbor $x_{3}=(0,0,2,0,\cdots,0)$, $t$ and $v$ have two common neighbors $w$ and $u$, $t$ and $w$ have one common neighbor $x_{4}=(0,2,1,0,\cdots,0)$, $t$ has a neighbor set $X_{4}$ with the order $2n-2-1-1=2n-4$ in $V(Q_{n}^{3})-V(P_{4})-X_{1}-X_{2}-X_{3}$. Thus $|F|=|X_{1}|+|X_{2}|+|X_{3}|+|X_{4}|=(2n-2)+(2n-3)+(2n-3)+(2n-4)=8n-12$. We will show that $F$ is a $3$-extra vertex cut of $Q_{n}^{3}$ for $n\geq 3$.
For $n=3$, from Fig.1, it is easy to see that $F$ is a $3$-extra vertex cut of $Q_{3}^{3}$. We assume that $n\geq 4$ in the following. Recall that $N[P_{4}]=N(P_{4})\bigcup P_{4}$, we will prove that $Q_{n}^{3}- N[P_{4}]$ is connected for $n\geq 4$.
Without loss of generosity, we partition $Q_{n}^{k}$ over $0$-dimension. Let $F_{i}=N(P_{4})\bigcap Q[i]$, where $i\in\{0,1,2\}$. Note that $P_{4}=uvwt\in Q[0]$, the two outer neighbors of every vertex in $P_{4}$ are in different subgraph $Q[j]$, $j\neq i$, thus $|F_{k}|=4$, $k=1,2$. By Lemma 2.2, $\kappa(Q_{n-1}^{3})=2(n-1)\geq 6>4$, then $Q[k]-F_{k}$ $(k=1,2)$ is connected. Since there are $3^{n-1}\geq 9$ edges between $Q[i]$ and $Q[j]$ for $0\leq i\neq j\leq2$ and $n\geq 3$. Thus $Q_{n}^{3}-Q[0]-N[P_{4}]$ is connected, denoted by $C$.
Now we consider $Q[0]-N[P_{4}]$, for any $ x\in Q[0]-N[P_{4}]$, $x$ has two outer neighbors $x_{L}$ and $x_{R}$, obviously, $x_{L}$($x_{R}$) is not in $N[P_{4}]\bigcap(Q_{n}^{3}-Q[0])$. Hence $x$ is connected to $C$. By the arbitrary of $x$, $C\bigcup(Q[0]-N[P_{4}])=Q_{n}^{3}-N[P_{4}]$ is connected.
Thus $Q_{n}^{3}-F$ has two components, $Q_{n}^{3}-N[P_{4}]$ and $P_{4}$. Then $F$ is a $3$-extra vertex cut of $Q_{n}^{3}$ for $n\geq 3$, thus $\kappa_{3}(Q_{n}^{3})\leq |F|=8n-12$. The theorem is completed.
In the following, suppose $F\subseteq V(Q_{n}^{3})$ is a faulty vertex set of $ Q_{n}^{3}$. For convenience, let $F_{i}=F \bigcap V(Q[i])$, $I=\{i|$ $Q[i]-F_{i}$ is disconnected for $i\in\{0,1,2\}\}$, $J=\{0,1,2\}\backslash I$. $F_{I}=\bigcup_{i\in I}F_{i}$, $F_{J}=\bigcup_{j\in J}F_{j}$, $ Q[I]=\bigcup_{i\in I}Q[i]$, $ Q[J]=\bigcup_{j\in J}Q[j]$.
Let $F\subseteq V(Q_{n}^{3})$ with $|F|\leq 8n-13$ be a faulty vertex set of $Q_{n}^{3}$. Then, $Q[J]-F_{J}$ is connected when $|I|\leq 2$ and $n\geq4$. Furthermore, let $H$ be a component of $Q_{n}^{3}-F$ and $H\bigcap(Q[J]-F_{J})=\emptyset$, then $N_{Q[I]}(H)\subseteq F_{I}$, $N_{Q[J]}(H)\subseteq F_{J}$.
Clearly, $|I|\leq 3$, $|F_{i}|\geq 2n-2$ for any $i\in I$. By the definition of $J$, for any $j\in J$, $Q[j]-F_{j}$ is connected. We consider the following three cases.
[**Case 1.**]{} $|I|=0$.
For any $j\in J$, $Q[j]-F_{j}$ is connected. Since $3^{n-1}-(8n-13)\geq 8$ for $n\geq4$, there must be an edge between $Q[j]-F_{j}$ and $Q[k]-F_{k}$ for $j\neq k$ and $j,k\in \{0,1,2\}$. $Q[J]-F_{J}$ is connected for $n\geq4$.
[**Case 2.**]{} $|I|=1$.
Without loss of generality, we assume that $I=\{0\}$, then $|F_{0}|\geq \kappa(Q_{n-1}^{3})= 2(n-1)$, both $Q[1]-F_{1}$ and $Q[2]-F_{2}$ are connected. There are at least $3^{n-1}-(|F|-|F_{0}|)\geq 3^{n-1}-[8n-13-(2n-2)]\geq3^{3}-(19-6)=14$ edges between $Q[1]-F_{1}$ and $Q[2]-F_{2}$ for $n\geq4$, thus $Q[J]-F_{J}$ is connected.
[**Case 3.**]{} $|I|=2$.
Without loss of generality, we assume that $I=\{0,1\}$. Then $Q[J]-F_{J}=Q[2]-F_{2}$ is also connected.
Hence $Q[J]-F_{J}$ is connected for $|I|\leq 2$ and $n\geq4$.
Suppose there exists a vertex $u\in N_{Q[I]}(H)$, $u\notin F_{I}$. Then $u$ is connected to $H$, hence $u$ belongs to $H$, which leads to a contradiction. Thus $N_{Q[I]}(H)\subseteq F_{I}$. If there exists a vertex $v\in N_{Q[J]}(H)$, $v\notin F_{J}$, then $v$ is connected to $H$, $v$ is contained in $H$, then $H\bigcap (Q[J]-F_{J})=\{v\}$, which is contradict to $H\bigcap (Q(J)-F_J)=\emptyset$. Hence, $N_{Q[J]}(H)\subseteq F_{J}$. The lemma is completed.
Suppose that $F\subseteq V(Q_{n}^{3})$ with $|F|\leq 8n-13$ is a vertex cut of $ Q_{n}^{3}$ for $n\geq 3$, then $Q_{n}^{3}-F$ has one of the following conditions:
$(1)$ two components, one of which is a singleton or an edge or a 2-path or a 3-cycle.
$(2)$ three components, two of which are singletons.
$(3)$ three components, two of which are a singleton and an edge, respectively.
$(4)$ four components, three of which are singletons.
This theorem can be proved by using induction on $n$.
[**Case 1.**]{} Suppose that $n=3$ and $F\subseteq V(Q_{3}^{3})$ with $|F|\leq 8n-13=11$, is a vertex cut of $ Q_{3}^{3}$. We consider the following four subcases.
[**Subcase 1.1.**]{} $|I|=0$.
Since $J=\{0,1,2\}\backslash I=\{0,1,2\}$.
If for any $j\in J$, $|F_{j}|<\kappa(Q_{n-1}^{3})=2(n-1)=4$. Since $3^{3-1}-|F_{j}|-|F_{k}|\geq 3^{3-1}-4(3-1)=1$, there must be an edge between $Q[j]-F_{j}$ and $Q[k]-F_{k}$ for $j\neq k$ and $j,k\in \{0,1,2\}$. In this case $Q[J]-F_{J}$ is connected which is a contradiction.
If there exists only one $j\in J$, $|F_{j}|\geq\kappa(Q_{n-1}^{3})=2(n-1)=4$, suppose $j=0$. $Q[1]-F_{1}$ is connected to $Q[2]-F_{2}$. Since $|V(Q[0])|=3^{2}=9$, then $4\leq|F_{0}|\leq9$. If $Q[0]-F_{0}$ is empty, then $\sum_{j=1}^{2}|F_{j}|\leq 2$, there are at least $9-2=7$ edges between $Q[1]-F_{1}$ and $Q[2]-F_{2}$, thus $Q_{3}^{3}-F$ is connected. If $Q[0]-F_{0}$ has one vertex, then $\sum_{j=1}^{2}|F_{j}|\leq 3$. By Lemma 2.4, $Q_{3}^{3}-F$ is connected or has two components, one of which is a singleton. If $Q[0]-F_{0}$ has two vertices, then $\sum_{j=1}^{2}|F_{j}|\leq 4$. In this case, $Q_{3}^{3}-F$ is connected or has two components, one of which is an edge. If $Q[0]-F_{0}$ has three vertices, then $\sum_{j=1}^{2}|F_{j}|\leq 5$. The three vertices have six outer neighbors, $Q_{3}^{3}-F$ is connected in this case. If $Q[0]-F_{0}$ has four or five vertices, $Q_{3}^{3}-F$ is connected by the similar reason.
If there exists only two $j\in J$, $|F_{j}|\geq\kappa(Q_{n-1}^{3})=2(n-1)=4$, suppose $j\in\{0,1\}$. Then $|F_{2}|\leq 3$. Clearly, $|F_{0}|\leq|F|-|F_{1}|\leq7$, $|F_{1}|\leq|F|-|F_{0}|\leq7$. Suppose $|F_{0}|\leq|F_{1}|$. If $|F_{1}|=7$, then $|F_{0}|=4$, $|F_{2}|=0$, every vertex of $Q[j]-F[j]$ for $j\in\{0,1\}$ has an outer neighbor in $Q[2]$, then $Q_{3}^{3}-F$ is connected. Now we consider $|F_{1}|=6$, if $|F_{0}|=5$, as the similar reason, $Q_{3}^{3}-F$ is connected. The left case is $|F_{0}|=4$, $|F_{2}|\leq1$, note that $|V(Q[0]-F[0])|=5$ and $|V(Q[1]-F[1])|=3$, every vertex of $Q[j]-F[j]$ for each $j\in\{0,1\}$ has an outer neighbor in $Q[2]$. Thus there is an edge between $Q[j]-F[j]$ and $Q[2]-F[2]$, then $Q_{3}^{3}-F$ is connected. Next consider $|F_{1}|=5$, if $|F_{0}|=5$, then $|F_{2}|\leq 1$, $Q_{3}^{3}-F$ is connected; if $|F_{0}|=4$, then $|F_{2}|\leq 2$, $Q_{3}^{3}-F$ is also connected.
If for any $j\in J$, $|F_{j}|\geq\kappa(Q_{n-1}^{3})=2(n-1)=4$, then $|F|\geq12$, this is contradict to $|F|\leq 8n-13=11$.
In summary, $Q_{3}^{3}-F$ has two components, one of which is a singleton or an edge. The result holds.
[**Subcase 1.2.**]{} $|I|=1$.
Without loss of generality, we assume that $I=\{0\}$. Then $Q[0]-F_{0}$ is disconnected. By Lemma 2.2, $|F_{0}|\geq 2n-2=4$. Thus $|F|-|F_{0}|\leq 11-4=7$. There are at least $3^{3-1}-(|F|-|F_{0}|)\geq 3^{2}-(11-4)\geq2$ edges between $Q[1]-F_{1}$ and $Q[2]-F_{2}$, then $Q[J]-F_{J}$ is connected.
Suppose $W$ is the union of all components of $Q_{3}^{3}-F$ and has no vertices in $Q[J]-F_{J}$. Since $Q_{3}^{3}-F$ is disconnected, then $W$ exists. Every vertex in $W$ has two outer neighbors, so $2|W|\leq |F|-|F_{0}|\leq7$, that means $|W|\leq 3$, the result holds.
[**Subcase 1.3.**]{} $|I|=2$.
Without loss of generality, we assume that $I=\{0,1\}$. Then $Q[0]-F_{0}$ and $Q[1]-F_{1}$ are disconnected, $Q[2]-F_{2}$ is connected. By Lemma 2.2, $|F_{0}|\geq 2n-2=4$ and $|F_{1}|\geq 2n-2=4$, so $|F_{2}|=|F|-|F_{0}|-|F_{1}|\leq11-4-4=3$. Suppose $W$ is the union of all components of $Q_{3}^{3}-F$ and has no vertices in $Q[2]-F_{2}$. Since $Q_{3}^{3}-F$ is disconnected, then $W$ exists. Every vertex in $W$ has one outer neighbor in $Q[2]$, so $|W|\leq |F_{2}|\leq 3$. The desired result.
[**Subcase 1.4.**]{} $|I|=3$.
In this case, $I=\{0,1,2\}$. For any $i\in I$, $|F_{i}|\geq 2n-2=4$, then $|F|=|F_{0}|+|F_{1}|+|F_{2}|\geq 12$, this is contradict to $|F|\leq 11$.
In summary, we have proved the result holds for $ Q_{3}^{3}$.
In what follows, assume that $n\geq 4$ and the result holds for $ Q_{n-1}^{3}$.
[**Case 2.**]{} We prove the result for $ Q_{n}^{3}$ and $n\geq 4$. We divide the proof into the following two subcases.
[**Subase 2.1.**]{} For any $i\in \{0,1,2\}$, $|F_{i}|\leq 6n-14$.
Recall that $|F|\leq 8n-13$, by Lemma 3.6, $Q[J]-F_{J}$ is connected when $|I|\leq 2$. We consider the following four subcases.
[**Subcase 2.1.1.**]{} $|I|=0$.
Since $J=\{0,1,2\}\backslash I=\{0,1,2\}$, by Lemma 3.6, $ Q[J]-F_{J}=Q_{n}^{3}-F$ is connected which is a contradiction.
[**Subcase 2.1.2.**]{} $|I|=1$.
Without loss of generality, we assume that $I=\{0\}$. By Lemma 3.6, $Q[J]-F_{J}$ is connected. Since $Q[0]-F_{0}$ is disconnected, and $|F_{i}|\leq 6n-14=6(n-1)-8$, by Lemma 3.4, $Q[0]-F_{0}$ either has two components, one of which is a singleton or an edge, denoted by $X_{0}$; or has three components, two of which are singletons, denoted by $X_{0}=\{u,v\}$. Let $B$ be the largest component of $Q[0]-F_{0}$. Next we show $B$ is connected to $Q[J]-F_{J}$. Note that $B$ has at least $3^{n-1}-(6n-14)-2$ vertices, and has at least $2[3^{n-1}-(6n-14)-2]$ outer neighbors in $Q[J]$. Since $2[3^{n-1}-(6n-14)-2]> 8n-13-(2n-2)=6n-11$ for $n\geq4$, then $B$ is connected to $Q[J]-F_{J}$. Thus $Q_{n}^{3}-F$ either has two components, one of which is a singleton or an edge; or has three components, two of which are singletons. The result holds.
[**Subcase 2.1.3.**]{} $|I|=2$.
Without loss of generality, we assume that $I=\{0,1\}$. By Lemma 3.6, $Q[2]-F_{2}$ is connected. For $i\in I$, since $|F_{i}|\leq 6n-14=6(n-1)-8$, by Lemma 3.4, $Q[i]-F_{i}$ either has two components, one of which is a singleton or an edge, denoted by $X_{i}$; or has three components, two of which are singletons, denoted by $X_{i}=\{u_{i},v_{i}\}$. Let $B_{i}$ be the largest component of $Q[i]-F_{i}$, then $X_{i}=Q[i]-F_{i}-B_{i}$. We claim that $B_{i}$ is connected to $Q[2]-F_{2}$. In fact, $B_{i}$ has at least $3^{n-1}-(6n-14)-2$ vertices, and $3^{n-1}-(6n-14)-2$ outer neighbors in $Q[2]$, since $|F_{2}|=|F|-|F_{0}|-|F_{1}|\leq 8n-13-2(2n-2)=4n-9$ and $3^{n-1}-(6n-14)-2 >4n-9$ for $n\geq4$, then $B_{i}$ is connected to $Q[2]-F_{2}$. Let $C$ be the component of $Q_{n}^{3}-F$ which contains $B_{i}$ and $Q[2]-F_{2}$. Next we consider the following three subcases.
[**Subcase 2.1.3a.**]{} Both $Q[0]-F_{0}$ and $Q[1]-F_{1}$ have two components, one of which is a singleton or an edge.
If both $Q[0]-F_{0}$ and $Q[1]-F_{1}$ have two components, one of which is a singleton. By the similar argument as Subcase 2.4.3 of Lemma 3.4, $Q_{n}^{3}-F$ either has three components, two of which are singletons; or has two components, one of which is a singleton or an edge. The result holds.
If only one of $Q[0]-F_{0}$ and $Q[1]-F_{1}$ has two components, one of which is a singleton. Without loss of generality, assume that $Q[0]-F_{0}$ has two components, and one of which is a singleton which is denoted by $x_{0}$ and $Q[1]-F_{1}$ has two components, and one of which is an edge which is denoted by $X_{1}=u_{1}v_{1}$. If both $x_{0}$ and $X_{1}$ are contained in $C$, then $Q_{n}^{3}-F$ is connected which is a contradiction. If $x_{0}$ is contained in $C$, $X_{1}$ is not contained in $C$, then $Q_{n}^{3}-F$ has two components, one of which is an edge. If $x_{0}$ is not contained in $C$ and $X_1$ is contained in $C$, then $Q_{n}^{3}-F$ has two components, one of which is a singleton. Besides, the singleton $x_{0}$ in $Q[0]-F_{0}$ and the isolated edge $X_{1}$ in $Q[1]-F_{1}$ may either remain singleton and isolated edge in $Q_{n}^{3}-F$; or they could belong to one component of $Q_{n}^{3}-F$, forming a $2$-path. Thus $Q_{n}^{3}-F$ either has three components, two of which are a singleton and an edge, respectively; or has two components, one of which is a $2$-path. The result holds in this case.
If both $Q[0]-F_{0}$ and $Q[1]-F_{1}$ have two components, one of which is an edge, denoted by $X_{i}$ for $i\in I$. By Lemma 3.3, $|F_{i}|\geq 4(n-1)-3=4n-7$, $|F_{2}|=|F|-|F_{0}|-|F_{1}|\leq 8n-13-2(4n-7)=1$. Every vertex in $X_{i}$ has one outer neighbor in $Q[2]$. By Lemma 3.6, $N_{Q[2]}(X_{i})\subseteq F_{2}$, then $2=|N_{Q[2]}(X_{i})|\leq|F_{2}|\leq 1$ which is a contradiction.
[**Subcase 2.1.3b.**]{} Only one of $Q[0]-F_{0}$ and $Q[1]-F_{1}$ has two components.
Without loss of generality, assume that $Q[0]-F_{0}$ has two components and $Q[1]-F_{1}$ has three components.
If $Q[0]-F_{0}$ has two components, one of which is a singleton, denoted by $x_{0}$. $Q[1]-F_{1}$ has three components, two of which are singletons, denoted by $u_{1},v_{1}$. If $x_{0}$, $u_{1}$ and $v_{1}$ are contained in $C$, then $Q_{n}^{3}-F$ is connected which is a contradiction. If $x_{0}$ is contained in $C$, only one of $u_{1}$ and $v_{1}$ is contained in $C$, then $Q_{n}^{3}-F$ has two components, one of which is a singleton. If $x_{0}$ is contained in $C$, both $u_{1}$ and $v_{1}$ are not contained in $C$, then $Q_{n}^{3}-F$ has three components, two of which are singletons. If $x_{0}$ is not contained in $C$, both $u_{1}$ and $v_{1}$ are contained in $C$, $Q_{n}^{3}-F$ has two components, one of which is a singleton. If $x_{0}$ is not contained in $C$, only one of $u_{1}$ and $v_{1}$ is contained $C$. In this case, $Q_{n}^{3}-F$ has two components, one of which is an edge; or $Q_{n}^{3}-F$ has three components, two of which are singletons. Besides, the singletons in $Q[1]-F_{1}$ and $Q[0]-F_{0}$ may remain singleton components in $Q_{n}^{3}-F$; or they could belong to two components of $Q_{n}^{3}-F$, $K_{2}$ and a singleton. Thus $Q_{n}^{3}-F$ either has four components, three of which are singletons; or has three components, two of which are a singleton and an edge, respectively. The result holds.
If $Q[0]-F_{0}$ have two components, one of which is an edge, denoted by $X_{0}$. By Lemma 3.3, $|F_{0}|\geq 4(n-1)-3=4n-7$ and $|F_{1}|\geq 4(n-1)-3=4n-7$, by the similar argument as Subcase 2.1.3a, $2=|N_{Q[2]}(X_{0})|\leq|F_{2}|\leq 1$ which is a contradiction.
[**Subcase 2.1.3c.**]{} Both $Q[0]-F_{0}$ and $Q[1]-F_{1}$ have three components, two of which are singletons, denoted by $X_{i}=\{u_{i},v_{i}\}$ for $i\in I$.
By Lemma 3.3, $|F_{0}|\geq 4(n-1)-3=4n-7$ and $|F_{1}|\geq 4(n-1)-3=4n-7$. Then $|F_{2}|=|F|-|F_{0}|-|F_{1}|\leq 8n-13-2(4n-7)=1$. By Lemma 3.6, $N_{Q[2]}(u_{i})\subseteq F_{2}$, and $N_{Q[2]}(v_{i})\subseteq F_{2}$. Then $2=|N_{Q[2]}(X_{i})|\leq|F_{2}|\leq 1$ which is a contradiction.
$|I|=3$.
We have $I=\{0,1,2\}$, by Lemma 2.2, for any $i\in I$, $|F_{i}|\geq2(n-1)=2n-2$. Since $|F|\leq 8n-13$, then $|F_{i}|\leq 8n-13-2(2n-2)=4n-9\leq 4(n-1)-4$. By Lemma 3.3, $Q[i]-F_{i}$ has two components, one of which is a singleton $x_{i}$. Let $B_{i}$ be the largest component of $Q[i]-F_{i}$ for $i\in I$. Note that $B_{i}$ has at least $3^{n-1}-(4n-9)-1$ vertices, and has at least $3^{n-1}-(4n-9)-1$ neighbors in $Q[k]$ for $0\leq k\neq i\leq2$. $B_{i}$ is connected to $B_{k}$ since $|F_{k}|\leq 4n-9$ and $3^{n-1}-(4n-9)-1>4n-9$ for $n\geq4$. Let $C$ be the component of $Q_{n}^{3}-F$ which contains $B_{i}$ for $i\in I$. We consider the following four subcases.
[**Subcase 2.1.4a.**]{} For any $i\in I$, $x_{i}$ is contained in $C$.
In this case, $Q_{n}^{3}-F$ is connected which is a contradiction.
[**Subcase 2.1.4b.**]{} There exists only two $i\in I$, such that $x_{i}$ is contained in $C$.
In this case, $Q_{n}^{3}-F$ has two components, one of which is a singleton.
[**Subcase 2.1.4c.**]{} Only one $i\in I$, such that $x_{i}$ is contained in $C$.
In this case, $Q_{n}^{3}-F$ either has three components, two of which are singletons; or has two components, one of which is an edge. The result holds.
[**Subcase 2.1.4d.**]{} For any $i\in I$, $x_{i}$ is not contained in $C$.
In this case, the three singletons in $Q[i]-F_{i}$ may remain singleton components of $Q_{n}^{3}-F$; or they could belong to two components, a singleton and an edge; or they could belong to one component forming a $3$-cycle or a $2$-path. Thus $Q_{n}^{3}-F$ either has four components, three of which are singletons; or has three components, two of which are a singleton and an edge, respectively; or has two components, one of which is a $2$-path, or a $3$-cycle. Then the result holds.
[**Subcase 2.2.**]{} There exists some $i\in \{0,1,2\}$, such that $|F_{i}|\geq 6n-13$.
Without loss of generality, we assume that $|F_{0}|\geq 6n-13$. For $j\in\{1,2\}$, $|F_{j}|\leq 8n-13-(6n-13)=2n<4(n-1)-3=4n-7$ for any $n\geq 4$. We consider the following two subcases.
[**Subcase 2.2.1.**]{} For $j\in\{1,2\}$, $Q[j]-F_{j}$ is connected.
Since there are at least $3^{n-1}-|F_{1}|-|F_{2}|\geq3^{n-1}-2n-2n >11$ edges between $Q[1]-F_{1}$ and $Q[2]-F_{2}$ for $n\geq4$, then $Q[1]-F_{1}$ is connected to $Q[2]-F_{2}$. The component which contains $Q[j]-F_{j}$ for $j\in\{1,2\}$ of $Q_{n}^{3}-F$ is denoted by $C$.
If $|F_{0}|\leq 8(n-1)-13=8n-21$, suppose $Q[0]-F_{0}$ is connected. There are at least $3^{n-1}-(8n-21)-(2n-2)\geq 10$ edges between $Q[j]-F_{j}$ and $Q[0]-F_{0}$ for $n\geq4$, then $Q[0]-F_{0}$ is contained in $C$. It implies that $Q_{n}^{3}-F$ is connected which is a contradiction. Hence $F_{0}$ is a vertex cut of $Q[0]$. By induction on $n$, $Q[0]-F_{0}$ satisfies one of the conditions $(1)$-$(4)$. Let $D$ be the largest component of $Q[0]-F_{0}$. We claim that $D$ is contained in $C$. In fact, $D$ has at least $3^{n-1}-(8n-21)-3$ vertices, and has $3^{n-1}-(8n-21)-3$ outer neighbors in $Q[2]$, since $3^{n-1}-(8n-21)-3 > 8n-13-(2n-2)=6n-11$ for $n\geq4$, then $D$ is contained in $C$. Thus $Q_{n}^{3}-F$ satisfies one of the conditions $(1)$-$(4)$. The result holds in this case.
Suppose $|F_{0}|> 8n-21$, then $|F|-|F_{0}|<8n-13-(8n-21)=8$. Suppose $W$ is the union of the components of $Q_{n}^{3}-F$ and has no vertices in $C$. Since $Q_{n}^{3}-F$ is disconnected, then $W$ exists. By Lemma 3.6, the outer neighbors of $W$ is in $F-F_{0}$. By Lemma 2.4, $2|W|\leq|F|-|F_{0}|<8$, then $|W|<4$. Hence $|W|\leq 3$, the desired result.
[**Subcase 2.2.2.**]{} There exists $j\in \{1,2\}$, $Q[j]-F_{j}$ is disconnected.
Without loss of generality, we assume that $j=1$ and $Q[1]-F_{1}$ is disconnected. By Lemma 2.2, $|F_{1}|\geq \kappa(Q_{n-1}^{3})=2n-2$, $|F_{2}|=|F|-|F_{0}|-|F_{1}|\leq 8n-13-(6n-13)-(2n-2)=2$, then $Q[2]-F_{2}$ is connected. Furthermore, $|F_{0}|\leq 8n-13-(2n-2)=6n-11$, $|F_{1}|\leq8n-13-(6n-13)=2n\leq4(n-1)-4$ for $n\geq4$. By Lemma 3.3, $Q[1]-F_{1}$ has two components, one of which is a singleton $x_{1}$. Let $D$ be the largest component of $Q[1]-F_{1}$, then $D$ is connected to $Q[2]-F_{2}$. In fact, $D$ has at least $3^{n-1}-2n-1$ vertices and $3^{n-1}-2n-1$ neighbors in $Q[2]$, since $|F_{2}|\leq2$ and $3^{n-1}-2n-1-2>0$ for $n\geq4$, then $D$ is connected to $Q[2]-F_{2}$. The component which contains $D$ and $Q[2]-F_{2}$ of $Q_{n}^{3}-F$ is denoted by $C$.
If $Q[0]-F_{0}$ is connected, since $|F_{2}|\leq2$, $|F_{0}|\geq 6n-13$ and $3^{n-1}-(6n-13)\geq 17>|F_{2}|$ for $n\geq4$, there is at least one edge between $Q[0]-F_{0}$ and $Q[2]-F_{2}$. Then $Q[0]-F_{0}$ is connected to $Q[2]-F_{2}$, $Q[0]-F_{0}$ is contained in $C$. Thus $Q_{n}^{3}-F$ has two components, one of which is a singleton. Then the result holds.
If $Q[0]-F_{0}$ is disconnected, suppose $X$ is the union of the components of $Q[0]-F_{0}$ and has no neighbors in $C$. We will show $|X|\leq2$. Suppose by the way of contradiction that $|X|\geq3$. Every vertex in $X$ has one outer neighbor in $Q[2]\bigcap F$, by Lemma 3.6 and $|F_{2}|\leq 2$, thus $|X|\leq2$ which is a contradiction. Let $E$ be the largest component of $Q[0]-F_{0}$, then $E$ is connected to $Q[2]-F_{2}$. In fact, there must be an edge between $Q[0]-F_{0}$ and $Q[2]-F_{2}$ since $|F_{2}|\leq2$ and $3^{n-1}-(2n-2)-2>2\geq|F_{2}|$ for $n\geq4$. Thus $E$ is contained in $C$.
Suppose $W$ is the union of the components of $Q_{n}^{3}-F$ and has no vertices in $C$. Since $Q_{n}^{3}-F$ is disconnected, then $W$ exists. Obviously, $W\subseteq\{x_{1}\}\bigcup X$. Since $|X|\leq 2$, then $|W|\leq 3$. The desired result.
This covers all possibilities and the proof of the theorem is complete.
The following theorem about the 3-extra connectivity of the $3$-ary $n$-cube network follows from Theorem 3.5 and Theorem 3.7.
For $n\geq 3$, $\kappa_{3}(Q_{n}^{3})=8n-12$.
By Theorem 3.7, $\kappa_{3}(Q_{n}^{3})\geq8n-12$ for $n\geq 3$. On the other hand, by Theorem 3.5, $\kappa_{3}(Q_{n}^{3})\leq8n-12$ for $n\geq 3$. Hence, $\kappa_{3}(Q_{n}^{3})=8n-12$ for $n\geq 3$. The proof of the theorem is complete.
Conclusion
==========
In this paper, the 3-extra connectivity of the $3$-ary $n$-cube networks is gotten. The result shows that at least $8n-12$ vertices must be moved to disconnect the $3$-ary $n$-cube for $n\geq3$, provided that the removal of these vertices does not isolate either a singleton, an edge, a $2$-path, or a $3$-cycle. We will further study $3$-extra connectivity of the $k$-ary $n$-cube networks for $k\geq 4$, $h$-extra connectivity of the $k$-ary $n$-cube networks for $h\geq 4$ and $h$-extra connectivity of other interconnection networks. Determining the $h$-extra connectivity of various multiprocessor systems requires further research efforts.
This work was supported by the National Natural Science Foundation of China (11371052, 11271012, 11231008, 11171020).
[99]{} C. Balbuena, Extraconnectivity of s-geodetic digraphs and graphs, Discrete Mathematics, 195 (1999) 39-52. C. Balbuena, A. Carmona, J. Fbrega, M.A. Fiol, Extraconnectivity of graphs with large minimum degree and girth, Discrete Mathematics, 167/168 (1997) 85-100. J.A. Bondy, U.S.R. Murty, Graph Theory with Applications, North-Holland, New York, 1976. B. Bose, B. Broeg, Y. Kwon, Y. Ashir, Lee distance and topological properties of $k$-ary $n$-cubes, IEEE Transactions on Computers, 44 (8) (1995) 1021-1030. E. Cheng, L. Lipta’k, F. Sala, Linearly many faults in $2$-tree-generated networks, Networks, 55 (2010) 90-98. K. Day, The conditional node connectivity of the $k$-ary $n$-cube, Journal of Interconnection Networks, 5 (1) (2004) 13-26. Khaled Day, Abdel Elah, Al-Ayyoub, Fault diameter of $k$-ary $n$-cube networks, IEEE Transactions on Parallel and distributed systems, 8 (9) (1997) 903-907. A.H. Esfahanian, Generalized measures of fault tolerance with application to $n$-cube networks, IEEE Transactions on Computers, 38 (11) (1989) 1586-1591. A.H. Esfahanian, S. L. Hakimi, On computing a conditional edge-connectivity of a graph, Information Processing Letters, 27 (4) (1988) 195-199. J. F$\grave{a}$brega, M.A. Fiol, On the extra connectivity graphs, Discrete Mathematics, 155 (1996) 49-57. J. F$\grave{a}$brega, M.A. Fiol, Extraconnectivity of graphs with large girth, Discrete Mathematics, 127 (1994) 163-170. S.A. Ghozati, H.C. Wasserman, The $k$-ary $n$-cube network: modeling, topological properties and routing strategies, Computers and Electrical Engineering, 25 (3) (1999) 155-168. R.X. Hao, Y.Q. Feng, J.X. Zhou, Conditional diagnosability of alternating group graphs, IEEE Transactions on Computers, 62 (4) (2013) 827-831. R.X. Hao, J.X. Zhou, Characterize a kind of fault tolerance of alternating group network, Acta Mathematica Sinica, Chinese Series, 55 (6) (2012) 1055-1066. F. Harary, Conditional connectivity, Networks, 143 (12) (1983) 346-357. Sun-Yuan Hsieh, Ying-Hsuan Chang, Extra connectivity of $k$-ary $n$-cube networks, Theoretical Computer Science, 443 (2012) 63-69. S.Y. Hsieh, T.J. Lin, H.L. Huang, Panconnectivity and edge-pancyclicity of $3$-ary $n$-cubes, The Journal of Supercomputing, 42 (2007) 233-255. S. LatiL, M. Hegde, M. Naraghi-Pour, Conditional connectivity measures for large multiprocessor systems, IEEE Transactions on Computers, 43 (2) (1994) 218-222. S.L. Scott, J.R. Goodman, The impact of pipelined channel on $k$-ary $n$-cube networks. IEEE Transactions on Parallel and Distributed Systems, 5 (1) (1994) 2-16. J.M. Xu, Combination of Network Theory, Science Press, Beijing, 2013. W.H. Yang, J.X. Meng, Extraconnectivity of hypercubes, Applied Mathematics Letters, 22 (6) (2009) 887-891. Q. Zhu, X.K. Wang, J.J. Ren, Extra connectivity measures of $3$-ary $n$-cubes, Theory of Computing Systems, arxiv.org/pdf/1105.0991v1 \[cs.DM\] 5 May 2011.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce a network design game where the objective of the players is to design the interconnections between the nodes of two different networks $G_1$ and $G_2$ in order to maximize certain local utility functions. In this setting, each player is associated with a node in $G_1$ and has functional dependencies on certain nodes in $G_2$. We use a distance-based utility for the players in which the goal of each player is to purchase a set of edges (incident to its associated node) such that the sum of the distances between its associated node and the nodes it depends on in $G_2$ is minimized. We consider a heterogeneous set of players (i.e., players have their own costs and benefits for constructing edges). We show that finding a best response of a player in this game is NP-hard. Despite this, we characterize some properties of the best response actions which are helpful in determining a Nash equilibrium for certain instances of this game. In particular, we prove existence of pure Nash equilibria in this game when $G_2$ contains a star subgraph, and provide an algorithm that outputs such an equilibrium for any set of players. Finally, we show that the price of anarchy in this game can be arbitrarily large.'
author:
- 'Ebrahim Moradi Shahrivar and Shreyas Sundaram [^1]'
bibliography:
- 'refs.bib'
title: 'The Game-Theoretic Formation of Interconnections Between Networks'
---
[Shell : Bare Demo of IEEEtran.cls for IEEE Journals]{}
Interconnected Networks, Network Design, NP-hardness, Nash Equilibria, Price of Anarchy, Hub-and-Spoke.
Introduction
============
demo file is intended to serve as a “starter file” for IEEE journal papers produced under LaTeX using IEEEtran.cls version 1.8b and later. I wish you the best of success.
Conclusion
==========
We introduced the interconnection network design game between two networks $G_1$ and $G_2$. In this game, there is a heterogeneous (in terms of utility function) set of network designers, each associated with a node in the network $G_1$. Each node in $G_1$ is dependent on certain nodes in $G_2$, and these dependencies are captured by a network $G_I$. The utility of the players is defined based on the distance-utility function where the objective of each player is to build a set of edges from its associated node to nodes in the network $G_2$ such that distances between its associated node and the nodes it depends on in $G_2$ are minimized. We showed that finding a best response action of a player is NP-hard. Nevertheless, we showed certain important properties of the best response networks, which enabled us to find a Nash equilibrium for certain instances of the game. Finally, we applied our framework to model the interdependencies between communication and power networks. Our simulations suggest that the social welfare is larger when players are homogeneous in terms of their edge construction costs, compared to players with heterogeneous edge costs.
One interesting avenue for future research is to consider other classes of utility functions for the players. Another important topic for further research on this problem is to address the scenario where players can build different [*types*]{} of edges (e.g., representing different types of relationships between the nodes). Defining appropriate utility functions to capture this scenario, along with a characterization of the resulting Nash equilibria, would be of interest. Finally, proving existence of Nash equilibria when $G_2$ has an arbitrary structure would be of value.
Proof of the First Zonklar Equation
===================================
Appendix one text goes here.
Appendix two text goes here.
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to thank...
[1]{}
H. Kopka and P. W. Daly, *A Guide to LaTeX*, 3rd ed.1em plus 0.5em minus 0.4emHarlow, England: Addison-Wesley, 1999.
[Michael Shell]{} Biography text here.
[John Doe]{} Biography text here.
[Jane Doe]{} Biography text here.
[Ebrahim Moradi Shahrivar]{} is a Postdoctoral Researcher at the University of Waterloo. He received his BSc and MSc degrees in Electrical Engineering from the University of Tehran in 2009 and 2011, and his PhD degree in Electrical and Computer Engineering from the University of Waterloo in 2016. His research interests include network science, game theory, algorithms, large-scale dynamical systems and optimal control.
[Shreyas Sundaram]{} is an Assistant Professor in the School of Electrical and Computer Engineering at Purdue University. He received his MS and PhD degrees in Electrical Engineering from the University of Illinois at Urbana-Champaign in 2005 and 2009, respectively. He was a Postdoctoral Researcher at the University of Pennsylvania from 2009 to 2010, and an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Waterloo from 2010 to 2014. At Waterloo, he received the Department of Electrical and Computer Engineering Research Award in 2014 and the Faculty of Engineering Distinguished Performance Award in 2012. He received the M. E. Van Valkenburg Graduate Research Award and the Robert T. Chien Memorial Award from the University of Illinois, and he was a finalist for the Best Student Paper Award at the 2007 and 2008 American Control Conferences. His research interests include network science, large-scale dynamical systems, fault-tolerant and secure control, linear system and estimation theory, and the application of algebraic graph theory to system analysis.
[^1]: This material is based upon work supported by the Natural Sciences and Engineering Research Council of Canada.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Jon Haël Brenas
- Rachid Echahed
- Martin Strecker
bibliography:
- 'main.bib'
title: On the Verification of Logically Decorated Graph Transformations
---
Introduction {#sec:intro}
============
Preliminaries {#sec:GGT}
=============
Graph Rewriting Systems and Strategies {#sec:GRS}
======================================
Verification {#sec:verification}
============
Assertion Logics {#sec:logic}
================
Conclusion {#sec:conclusion}
==========
Appendix {#appendix .unnumbered}
========
| {
"pile_set_name": "ArXiv"
} |
---
address: 'Institut für Theoretische Physik, Universität Hannover Appelstraße 2, D–30167 Hannover, Germany'
author:
- Olaf Lechtenfeld
---
phyzzx =
I investigate non-perturbative aspects of zero-dimensional matrix models. Subtleties in the large-$N$ limit of the semiclassical picture are pointed out. The tunneling of eigenvalues is seen to correspond to a chaotic sequence of recursion coefficients determining the orthogonal polynomials. =0 =1 [**[Introduction.]{}**]{}
Over the last four years, we have learned how to model two-dimensional euclidean quantum gravity (with topological fluctuations) by hermitean matrix models \[\], in the so-called double-scaling limit \[\]. The latter entails sending the size $N$ of the matrix $M$ to infinity while the couplings $g$ in the matrix potential $V$ approach a critical value $g_c$ in such a way that some combination of $N$ and $(g\-g_c)$ is kept constant. In the pure gravity case, one has $$V(M)\ =\ \shalf M^2 + gM^4 \quad,\qquad g_c=-{\textstyle{1\over48}}\quad.
\eqn\quarticpot$$ Interestingly, the critical matrix potential retains a local minimum but is unbounded from below, and analytic continuation is to be performed into a region where the partition function is ill-defined. The result is either singular or develops an imaginary part, associated with the instability towards the tunneling escape of individual matrix eigenvalues \[\].
The tunneling phenomenon is most transparent in a semiclassical treatment of large-$N$ matrix models. As a first step, the saddle-point analysis was pioneered in ref. \[\]. To work with a well-defined theory, I will bound the potential from below by changing its ‘large-$M$’ behavior through a modification $\d V(M)=\e M^6$, with a small, positive regulator $\e$. This turns out to change substantially the large-$N$ phase structure of the model \[–\]; in particular, the critical line $g_c\=g_3(\e)$ becomes [*metastable*]{} and invisible at finite $N$. Nevertheless, tunneling may now be studied directly, without the need for analytic continuation.
Instrumental to the success of the random matrix model has been the technology of orthogonal polynomials \[,\]. The saddle-point results, however, have not yet been completely understood in this framework. Only in the case of [*degenerate*]{} potential wells, a precise relation between orthogonal polynomial recursion coefficients $R_k$ on one side and the (multi-band) classical eigenvalue density on the other side has been established \[,,\].
The more general connection between the two approaches involves an interpretation of eigenvalue tunneling in terms of orthogonal polynomials. This will be addressed in the third part of my talk, where I shall present a surprising resolution of a puzzle mentioned earlier \[,,\]. The second part, following this introduction, will outline the semiclassical approach. Along the way, I will relate $N{=}\infty$ saddle points to airfoils, and interpret eigenvalue tunneling in terms of the thawing of a frozen Dyson gas. A short summary shall close my talk, which reviews some research of mine conducted over the past three years and published for the most part in refs. \[,,\]. .2in
[**[The Semiclassical Approach.]{}**]{}
I begin by formulating a collective field theory for the zero-dimensional hermitean one-matrix model at [*finite*]{} $N$. My starting point is the partition function $$Z_N \ \propto\ \int\!d^{N^2}\!\! M \;e^{-N\tr V(M)}
\eqn\ZMdef$$ for an $N{\times}N$ random hermitean matrix ensemble, in a potential $V$. Upon diagonalization $M=diag(x_i)$ this reduces to $$Z_N = e^{-N^2 F_N}\ :=\ \biggl[\prod_{i=1}^N \int\! dx_i\biggr] \; \exp\biggl\{
-N\sum_i V(x_i) + \sum_{i<j}\ln (x_i-x_j)^2 \biggr\} \quad,
\eqn\Zxdef$$ the partition function of a two-dimensional Coulomb gas of charges restricted to a line, in an external potential $V(x)$ at temperature $\beta^{-1}{=}1/N^2$. I like to change variables from the matrix eigenvalues $x_i$ to their density distribution $$\r(x)\ :=\ \N\sum_{i=1}^N\d(x-x_i)\quad.
\eqn\rhodef$$ More precisely, I insert $$\eqalign{
1\ &=\ \int\dr\;\prod_x \d\bigl(\r(x)-\N\medsum_i \d(x\-x_i)\bigr)\cr
&=\ \int\!\!\!\!\int\dr\dl\;\exp\Bigl\{i\int\!dx\,\la(x)
\bigl[\r(x)-\N\medsum_i\d(x\-x_i)\bigr]\Bigr\} \cr}
\eqn\insertion$$ into and express the action in terms of the density, $$S_N^0[\r]\ =\ N^2\int\!\!dx\,\r(x)V(x) -\shalf
N^2\int\!\!\!\!\int\!\!dxdy \;\r(x)f(x-y)\r(y)+\shalf N\,f(0)\quad.
\eqn\Srho$$ The self-interaction had to be regulated by replacing $\,\ln z^2 \to
f(z)\,$ in eq. , choosing some suitable, symmetric and bounded, function $f$.
Following ref. \[\] I am able to perform the integration over $x_i$, $$\eqalign{
Z_N\ &=\ \int\!\!\!\!\int\dr\dl\;\;e^{-S_N^0[\r]+i\int\!\la\,\r}\;
\biggl[\prod_{i=1}^N\int\!dx_i\biggr]\; e^{-\iN\sum_i\la(x_i)} \crr
&=\ \int\!\!\!\!\int\dr\dl\;\;e^{-S_N^0[\r]+i\int\!\la\,\r}\;
\biggl[\int\!dx\;e^{-\iN\la(x)}\biggr]^N \quad,\cr}
\eqn\Zrholambda$$ and arrive at an [*exact*]{} effective action $$S_N[\r,\la]\ =\ S_N^0[\r]-i\int\!dx\,\la(x)\r(x)-N\ln\int\!dx\,e^{-\iN\la(x)}
\eqn\Srholambda$$ which is not only nonlocal in the two real fields $\r$ and $\la$ but also non-polynomial in the latter. Interestingly, the constant mode of $\la$ can be integrated out exactly to yield the constraint $\d(\int\!\r-1)$ that was apparent already from the definition . However, I shall keep those modes in the measure for the time being. In principle, another constraint arises from the positivity of $\r$. Perturbation theory about a strictly positive $\rh$, however, is insensitive to this restriction, and I will therefore ignore it in the following.
My goal is to initiate a systematic semiclassical analysis of this peculiar one-dimensional field theory. To leading order in $\hbar$ we must determine the saddle-point configurations $(\rh,\lh)$, where the action is stationary. The first variations yield $$\eqalign{
0\ &=\ N^2 V(x) - N^2\int\!dy\,f(x-y)\,\rh(y) -i\lh(x) \crr
0\ &=\ \rh(x) -e^{-\iN\lh(x)}\Big/\medint\!dy\,e^{-\iN\lh(y)} \cr}
\eqn\SPrholambda$$ where $\rh$ comes out to be properly normalized, $\int\rh=1$. The second equation determines $\lh$ up to a constant, $$\lh(x)\ =\ iN\ln\rh(x)+\la_0 \quad,
\eqn\SPlambda$$ a result which may be inserted into the first equation. Differentiating with respect to $x$ to remove constants and deregulating the Coulomb repulsion I get $$\shalf V'(x)\ +\ {1\over2N}{\rh'\over\rh}(x)\ =
\ -\!\!\!\!\!\!\int \! {dy\over x-y}\;\rh(y)\ \equiv\ \pi\,\h_x[\rh]
\eqn\SPrhoc$$ where $-\!\!\!\!\!\int$ denotes Cauchy’s principal value of the integral. The is known as the Hilbert transform (of $\rh$) which has been thoroughly investigated \[\]. Together with normalization and positivity, this equation describes the classical eigenvalue density for any finite $N$. It is noteworthy that is not homogeneous in $\N$, so its solution cannot be, either. At $N\=\infty$ the equation has been widely studied and solved \[,,\], and it was learned \[,\] that a [*unique*]{} solution extends to $N<\infty$. Unfortunately, the equation is not easily solved for finite $N$. Even in the large-$N$ limit some care is required, as shown by the following. For instance, should I drop the $\rh'/\rh$-term since it is down by $\N$? A little inspection reveals that such a step is in general not consistent with the asymptotic large-$|x|$ behavior of the equation, which demands a ${1\over x}$ fall-off for the due to the normalization of $\rh$. In fact, the formal ‘solution’ $$\rh_\e(x)\ \sim\
\exp\Bigl\{-N\bigl[V(x)-\medint\ln(x\-y)^2\,\rh_\e(y)\bigr]\Bigr\}
\ \mathrel{\mathop\approx^{|x|\to\infty}}\ x^{2N}\,e^{-NV(x)}
\eqn\mock$$ of equation shows that the $\rh'\over\rh$ term dominates the of eq. for $|x|\gg1$ (unless $V\sim\ln x^2$), so that $\rh_\e={\cal O}(e^{-N})$ asymptotically. Note that I have attached a subscript $\e\equiv\N$ to the solution to indicate that it depends on the value of $N$. One can read off that the effective potential seen by eigenvalues $|x|{\gg}1$ approaches $V(x)-2\ln|x|$. The situation is different, however, near the minimum of the potential where $V'$ dominates the of eq. and most of $\rh$ is concentrated. When $\rh$ is ${\cal O}(1)$ the $\rh'\over\rh$-term may safely be neglected or treated as a $\N$ perturbation in , leading to $$\rh_\e(x)\ \approx\ -{1\over\pi^2}-\!\!\!\!\!\!\int\!{dy\over x-y}\;\shalf
V'(y)
\eqn\mock2$$ by simply inverting the Hilbert transform in equation . The result is a modulation of Wigner’s semicircle distribution. When $N\to\infty$, the crossover regions between the interior and exterior of the ‘Fermi sea’ shrink to points $x{=}a,b$, and the saddle-point equation turns into \[\] $$\eqalign{
\shalf V'(x)\ &=\ -\!\!\!\!\!\!\int_a^b \!{dy\over x-y}\,\rh_0(y)
\ \equiv\ \pi\,\T_x[\rh_0] \qquad {\rm for}\ x\in[a,b]\cr
\rh_0(x)\ &=\ 0 \qquad\qquad\qquad\qquad\qquad\qquad\quad
{\rm for}\ x\notin[a,b] \quad.\cr}
\eqn\velocity$$ This relation is known as the [*airfoil equation*]{} and determines the vorticity $\rh_0$ related to a given velocity field $V'$ along the airfoil ($b\-a$ is the span of the wings) \[\]. In our case the location of $a$ and $b$ is determined from the normalization of $\rh_0$. Equation is solved by inverting the [*finite*]{} Hilbert transform , $$\rh_0(x)\ =\ {1\over\sqrt{(b-x)(x-a)}}\biggl[{1\over\pi}\ -\ {1\over\pi^2}
-\!\!\!\!\!\!\int_a^b \!{dy\over x-y}\;\shalf V'(y)\;\sqrt{(b-y)(y-a)}\biggr]
\quad.\eqn\vorticity$$ Again, I have assumed a single-well potential $V$, so that the support of $\rh_0$ is a single, connected interval $[a,b]$.
It is worthwhile to give the form of the saddle-point action. Employing eq. as well as $\,\ln\int e^{-\iN\lh}=-\iN\la_0\,$ I find, in agreement with ref. \[\], that ($\rh\=\rh_\e$) $$\eqalign{
S_N[\rh,\lh]\ &=\
N^2\int\!\rh(x)V(x)-N^2\int\!\!\!\!\int\!\rh(x)\ln|x\-y|\rh(y)
+N\int\!\rh(x)\ln\rh(x)\cr
&=\ {\textstyle{N^2\over2}}\int\!\rh(x)(V(x)\-\ln|x|)
+{\textstyle{N^2\over2}}V(0)+{\textstyle{N\over2}}\int\!\rh(x)\ln\rh(x)
+{\textstyle{N\over2}}\ln\rh(0) \cr}
\eqn\SPaction$$ where I dropped a (singular) term $\shalf Nf(0)$ and made use of the saddle-point equation . The last integral permits an interpretation as the entropy of the distribution $\rh$. Taking the naive large-$N$ limit, I obtain $$S_\infty[\rh_0]\ =\
{\textstyle{N^2\over2}}\int\!dx\;\rh_0(x)\,V(x)-N^2\ga\quad,
\eqn\Sinfty$$ with $$\ga\ =\ \int\!dy\;\ln|x-y|\,\rh_0(y) - \shalf V(x)\ =\
{\rm constant~~for~}x\in[a,b]
\eqn\chempot$$ being the chemical potential (or Lagrange multiplier enforcing $\int\rh_0~=1$).
At this point, I would like to drive home an essential point of my talk. The non-trivial $N$-dependence of the classical backgound $\rh_\e$ implies that the classical limit, $\r\to\rh_\e$, [*differs*]{} from the low-temperature limit, $N\to\infty$, because of $\N$ corrections coming from the integration measure. As a consequence, the semiclassical loop expansion will [*not*]{} be identical to the topological $\N$ expansion; rather, a [*double*]{} expansion arises. Beyond this, non-perturbative (in $\N$) contributions appear already at tree level (in $\hbar$). It is therefore by no means clear that interchanging limits, by [*first*]{} taking $N\to\infty$ in equation and [*then*]{} solving equation to obtain $\rh_0$, provides the large-$N$ limit, $\lim_{\e\to0}\rh_\e$, of a proper solution to equation , as my notation suggests. In fact, I will now demonstrate that in general a solution $\rh_0$ does [*not*]{} correspond to a finite-$N$ saddle point $\rh_\e$.
Non-perturbative effects become tangible when two or more potential wells compete for eigenvalues. For even potentials, the full complexity of the problem appears first in the triple-well potential $$V(x)\ =\ \shalf x^2 + g x^4 + \e x^6 \quad,\qquad\e>0\quad,
\eqn\pot$$ since the $x\leftrightarrow-x$ symmetry trivializes the double-well case. I like to fix $\e$ to some small value and probe the phase diagram by decreasing the quartic coupling $g$ along the negative axis. For $g<g_*=-\sqrt{3\e/2}$ the potential develops three well-separated minima which become degenerate at $g=g_@=-\sqrt{2\e}$. For sufficiently negative $g$ the ‘Fermi sea’ must, therefore, consist of three or two disconnected oceans, called [*arcs*]{} or [*bands*]{}. At $N\to\infty$, the eigenvalue density should then be supported on one, two, or three disjoint intervals. As shown in refs. \[,,\], the general solution of equation reads $$\rh_0(x)\ =\ {\textstyle{3\e\over\pi}}\,(n^2-x^2)\,\sqrt{
(4a^2-x^2)\,(4b^2-x^2)\,(4c^2-x^2)} \quad,
\eqn\rhothree$$ describing a positive three-band density with $0\le2a<n<2b\le2c$ and support on $[-2c,-2b]\cup[-2a,2a]\cup[2b,2c]$. Normalization imposes three conditions on $\{n,a,b,c\}$ which leaves a one-parameter [*family*]{} of solutions. A convenient parameter to label these solutions is the difference $\D\ga=\ga_o-\ga_c$ of the chemical potentials $\ga$ for the outer and the central bands. The chemical potential is nothing but the integration constant appearing when integrating equation . Since it may take different constant values for two eigenvalues $x_c\in[-2a,2a]$ and $x_o\in[2b,2c]$ from two different bands, the difference $$\D\ga\ =\ \int\! dy\;\ln{|x_o\-y|\over|x_c\-y|}\,\rh_0(y)
- \shalf\bigl[V(x_o)\-V(x_c)\bigr]
\eqn\chempotdiff$$ is a genuine property of the solution $\rh_0$. In this situation one must replace $$\ga\longrightarrow\sum_{\rm bands}\ga_i n_i\quad,\qquad
n_i=\int_{i\rm th~band}\!\!\!\!\!\!dx\;\rh_0(x)\quad,\qquad
\sum_{\rm bands}n_i=1\quad,
\eqn\replace$$ in equation . Extremal values of $\D\ga$ occur when the number of bands decreases, for $b\to c$ or $a\to0$, and the density becomes unique. However, those solutions do not exist everywhere in the $(\e,g)$ plane. For sufficiently small $\e$, the range of one-, two- and three-band solutions is given by the sequence $g_3<g_@<g_2<g_*<g_1<0$. Here, [*one-*]{}band densities arise for $g>g_3$, band distributions occur for $g<g_2$, and [*three-*]{}band solutions appear for $g<g_1$. Hence, equation admits a [*unique*]{} solution only for $g>g_1$. The overlapping regions above indicate a coexistence of multiple-band densities elsewhere. Even more astonishing is the discovery that the members $\rh_0(x,\D\ga)$ of such a family are [*not*]{} degenerate in free energy \[\]. This contradiction in terms is resolved by noticing that infinitesimal variations within the family, which correspond to the tunneling of individual eigenvalues, are actually [*singular*]{} at the band edges. One might say that it requires a [*finite*]{} variation to move an individual eigenvalue to another well although it is only an ${\cal O}(\N)$ effect. Since ${\partial\over\partial\D\ga}\rh_0$ is not square-integrable the inclusion of this mode among the density fluctuations in debatable. As the chemical potential drives the tunneling of eigenvalues, one should expect the [*minimal*]{} action to belong to the unique family member with $\D\ga\=0$, which I call [*dominant*]{}. This is indeed what happens and can be checked numerically \[\].
Of course, for $N<\infty$ there is always a [*unique*]{} saddle-point density $\rh_\e$, with a unique limit as $\e\to0$, because the strict positivity of the distribution implies that $\D\ga\=0$ all along. Hence, we have $$\lim_{N\to\infty} \rh_\e(x)\ =\ \rh_0(x,\D\ga\=0)\quad.
\eqn\stable$$ With hindsight it is clear that the ‘sub-dominant’ members of a three-band family could only appear because I took the limit $N\to\infty$ prematurely by going from equation to . ‘Physically’ speaking, the freezing of the Dyson gas of eigenvalues at [*zero*]{} temperature entirely suppresses any tunneling and permits those ficticious saddle-point distributions. At [*finite*]{} temperature, the Dyson solid melts at the edges, and tunneling, although exponentially small, destabilizes all but the dominant solutions. The ‘entropy term’ in equation plays the crucial role. However, I may still employ the incomplete large-$N$ saddle-point equation , if it is complemented by the additional global requirement $\D\ga\=0$.
In this light the correct phase diagram for the potential looks as follows. Again taking $\e$ small but fixed, I find a new sequence $g''<g_@<g'<0$, where a small region of three-band dominance around the degenerate point ($g\=g_@$) separates two-band from one-band densities. Inspecting the shape of the potential at the transition values $g'$ and $g''$, one learns that essentially the number of [*degenerate absolute minima*]{} determines the number of eigenvalue bands. Interestingly, the jump from one to three well-separated bands is smoothed out by an interpolating three-band region occuring when non-degenerate wells are in some sense comparable and can compete for eigenvalues. .2in
[**[Orthogonal Polynomials.]{}**]{}
The standard approach to matrix model calculations, and so far the only one capable of producing the topological expansion and the double-scaling limit, is the method of orthogonal polynomials \[\]. Its starting observation is that the van der Monde determinant in $$e^{-N^2 F_N}\ =\ \biggl[\prod_{i=1}^N \int\! dx_i\biggr] \;
\prod_{i<j} (x_i-x_j)^2 e^{-N\sum_i V(x_i)}
\eqn\Fxdef$$ can be rewritten as $$\prod_{i<j} |x_i-x_j|\ =\ \left|\matrix{
1&1&\ldots&1\cr x_1&x_2&\ldots&x_N\cr x_1^2&x_2^2&\ldots&x_N^2\cr
\vdots&\vdots&\ddots&\vdots\cr x_1^{N{-}1}&x_2^{N{-}1}&\ldots&x_N^{N{-}1}\cr}
\right| \ =\ \left|\matrix{
P_0(x_1)&P_0(x_2)&\ldots&P_0(x_N)\cr P_1(x_1)&P_1(x_2)&\ldots&P_1(x_N)\cr
P_2(x_1)&P_2(x_2)&\ldots&P_2(x_N)\cr \vdots&\vdots&\ddots&\vdots\cr
P_{N{-}1}(x_1)&P_{N{-}1}(x_2)&\ldots&P_{N{-}1}(x_N)\cr} \right| \quad,
\eqn\vandermonde$$ with monic polynomials $P_k(x)=x^k+{\rm (lower~order)}$. If one cleverly chooses the $P_k$ to be mutually orthogonal with respect to the measure $e^{-NV(x)}$, $$h_k\,\d_{kl}\ =\ \int\!dx\;e^{-NV(x)}\;P_k(x)\,P_l(x)\quad,
\eqn\Pdef$$ the change of basis $x^k\to P_k(x)$ in equation exactly diagonalizes the Coulomb interaction and trivializes the free energy to $$F_N\ =\ -{\textstyle{1\over N^2}}\,\ln[N!h_0h_1h_2\ldots h_{N{-}1}]\quad.
\eqn\FN$$ The construction of the polynomials simplifies thanks to the classic recursion relation $$P_{k+1}(x)\ =\ x\,P_k(x) - R_k\;P_{k-1}(x) \quad,\qquad
R_k\ =\ {h_k\over h_{k-1}}\ \ge 0\quad,
\eqn\Prec$$ so that is it sufficient to compute the norms $h_k$, starting with the initial condition $R_0\=0$.
The Stieltjes method of constructing the orthogonal polynomials consists of iterating eq. and the norm computation, : $$P_k(x;R_0,\ldots,R_{k{-}1})\longrightarrow h_k \longrightarrow R_k
\longrightarrow P_{k{+}1}(x;R_0,\ldots,R_k) \quad.
\eqn\stieltjes$$ Unfortunately, it is quite inappropriate for numerical analysis. However, the special form $e^{-NV}$ of the measure allows for a finite recursion relation among the $R_k$ themselves, the ‘string equation’ \[\] $$\eqalign{
{\textstyle{k\over N}}\ =\ R_k\,\Bigl\{1&+4g\bigl(R_{k-1}+R_k+R_{k+1}\bigr)+
6\e\bigl(R_{k-1}+R_k+R_{k+1}\bigr)^2 \cr
&+6\e\bigl(R_{k-2}R_{k-1}-R_{k-1}R_{k+1}+R_{k+1}R_{k+2}\bigr)\Bigr\}\quad,\cr}
\eqn\Rrec$$ displayed here for the potential . After solving for $\ R_k=R_k(R_{k-1},\ldots,R_{k-4})\ $ one still needs the initial values $$R_1\ =\ {h_1\over h_0}\ =\ {\int e^{-NV}\,x^2\over\int e^{-NV}}\quad,\qquad
R_2\ =\ {h_2\over h_1}\ =\
{\int e^{-NV}\,(x^2\-R_1)^2\over\int e^{-NV}\,x^2}
\eqn\Ronetwo$$ besides $\ R_{-1}=0=R_0\ $ to start the iteration. It turns out that this procedure is numerically tractable, but $R_k$ becomes increasingly sensitive to the initial conditions for growing $k$ or $N$ \[\]. Trivial but instructive is the exactly solvable example of purely quadratic potential, i.e. $g\=\e\=0$, or $V\=\shalf x^2$. In this case one simply rediscovers the Hermite polynomials from $R_k=k/N$, as appropriate for the harmonic oscillator.
Let us now investigate the large-$N$ or planar limit, in order to connect up with the semiclassical results. Here, I have to rely on an assumption, namely a continuum approach of the $R_k$ needs to be [*postulated*]{}. Writing $${\textstyle{k\over N}}\ =\ \x\in[0,1]\quad,
\qquad R_k\ =\ r_\e(\x)\quad,\quad \e=\N\quad,
\eqn\condnot$$ the simplest ansatz $$R_{k{+}1}-R_k\ =\ {\cal O}(\N) \qquad {\rm as} \quad N\to\infty
\eqn\oneRansatz$$ implies $r_\e(\x)\to r(\x)$, a smooth positive function with $r(0)\=0$. Numerical studies show that this behavior indeed occurs whenever a single potential well is clearly dominant \[\]. This coincides with the one-band regime of our potential. The string equation is then dramatically simplified to the algebraic relation \[\] $$\x\ =\ r\bigl\{ 1 + 12gr + 60\e r^2 \bigr\} \quad.
\eqn\oner$$ The condition $r(0)\=0$ selects a unique branch of $r$ which monotonically reaches $\x\=1$ provided $g>g_3(\e)$, the BPIZ critical line. Finally, the free energy $F_\infty$ is obtained by naively taking the $N\to\infty$ limit of eq. , $$\eqalign{
F_N\ &=\ -{\textstyle{1\over N^2}}\ln N! - \N\ln h_0
- \N\sum_{k=1}^N\left(1-{\textstyle{k\over N}}\right)\ln R_k \cr
&\rightarrow\ -{\textstyle{1\over N^2}}\ln N! + V_{\rm min}
- \int_0^1\!d\x\;(1-\x)\,\ln\r(\x) \quad,\cr}
\eqn\Finfty$$ to be compared to ${1\over N^2}S_\infty$ from equation .
However, the continuity assumption is clearly violated when $g<g_@$, because an estimate of $R_1$ from equation reveals that it must [*jump*]{} from ${\cal O}(\N)$ to the square of the location of the outer potential minima, $\approx{-}{g_@\over3\e}\={1\over\sqrt{6\e}}$, when $g$ drops below the degenerate point, $g\=g_@$. Such a behavior is known from studies of double-well potentials \[,\], where an [*alternating*]{} sequence $$R_{k{+}2}-R_k\ =\ {\cal O}(\N)\quad,\qquad
r_\e(\x)\ \to\ \cases{r^{(0)}(\x)&for $k$ even\cr r^{(1)}(\x)&for $k$ odd\cr}
\eqn\twoRansatz$$ is observed. Under this modified assumption, the string equation turns into two coupled cubic equations for $r^{(0)}$ and $r^{(1)}$. Their graphical solution exhibits $r^{(0)}(\x)$ as an increasing function starting from $r^{(0)}(0)\=0$, and $r^{(1)}(\x)$ as decreasing from $r^{(1)}(0)=x_{\rm min}^2$ \[\].
Like for the saddle-point method, it is not evident which large-$N$ assumption is to be chosen for given values of $\e$ and $g$. It can be shown, however, that the series of $\sqrt{R_k}$ approximates the sequence of consecutive eigenvalues $x_k$ eventually building up to the distribution $\rh_\e(x)$ in the limit $\e\to0$. Hence, one- and two-band regions in the phase diagram must correspond to the continuum behavior of eqs. and , respectively. An unsettling gap remains, however, in the three-band dominated buffer zone. How can the one-branch ansatz merge with the two-branch ansatz when $r^{(1)}(0)-r^{(0)}(0)\approx x_{\rm min}^2\geq{1\over\sqrt{6\e}}$? The resolution of this paradox was discovered through a numerical analysis, which uncovered a $\x$ interval $[\bar{\x},\tilde{\x}]$ with seemingly [*chaotic*]{} recursion coefficients $R_k$, interpolating between one or two branches for $\x<\bar{\x}$ and a single branch for $\x>\tilde{\x}(>1)$. The three-band region $g''<g<g'$ coincides with $\bar{\x}<1$, i.e. the onset of the unpredictable behavior creeping into the relevant $\x$ interval $[0,1]$. Only exactly at the degenerate point, $g\=g_@$, is a simple three-branch solution realized \[\]. This picture has been confirmed by several groups \[,\]. .2in
[**[Conclusions.]{}**]{}
I have reviewed some non-perturbative aspects of hermitean random matrix models, with an emphasis on the distribution of eigenvalues among several potential wells. It turned out that the classical, quasi-continuous, large-$N$ eigenvalue distribution depends [*non-perturbatively*]{} on $\N$, so that the semiclassical loop expansion must be distinguished from the standard topological (or string loop) expansion.
For multiple-well matrix potentials, interchanging the limits $N{\to}\infty$ and $S_N{\to}$extremum is dangerous due to the $N{<}\infty$ equilibration between different Dyson gas components, mediated by eigenvalue tunneling. As a consequence, the point of critical coupling becomes [*metastable*]{} for pure gravity when $V$ gets bounded from below.
From numerical simulations of the sequence of recursion coefficients $R_k$ for the orthogonal polynomials, I conjecture that their behavior is characterized by the critical points of the potential function $V$ itself. The consecutive equilibrium deposition of eigenvalues on the real line is suggested to be sensitive to features of $V$ at increasing values. For [*degenerate*]{} absolute minima one observes a quasi-periodic series of $R_k$, whereas [*non-degenerate*]{} minima produce [*chaotic*]{} behavior! The unpredictability of the coefficients reflects the competition of incommensurate potential wells for eigenvalues.
I hope to have demonstrated that the continuum limit of matrix models is more complicated than imagined originally. In view of this it would be very beneficial to understand their critical properties in the semiclassical description. An attempt to push the latter beyond the classical limit is currently in progress (see \[\]). I am grateful to the organizers for the charming and stimulating atmosphere of the Symposium, and especially K. Behrndt for his personal efforts to make it all work.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we investigate an opportunistic relaying scheme where the selected relay assists the source-destination (direct) communication. In our study, we consider a regenerative opportunistic relaying scheme in which the direct path can be considered unusable, and takes into account the effect of the possible erroneously detected and transmitted data at the best relay. We first derive statistics based on exact probability density function (PDF) of each hop. Then, the PDFs are used to determine accurate closed form expressions for end-to-end bit-error rate (BER) of binary phase-shift keying (BPSK) modulation. Furthermore, we evaluate the asymptotical performance analysis and the diversity order is deduced. Finally, we validate our analysis by showing that performance simulation results coincide with our analytical results over different network architectures.'
author:
- 'Kamel Tourki, Hong-Chuan Yang, Mohamed-Slim Alouini, .'
bibliography:
- 'IEEEabrv.bib'
- 'refJrlTYA.bib'
title: 'Accurate Performance Analysis of Opportunistic Decode-and-Forward Relaying[^1]'
---
**Keywords**:
Cooperative diversity, Opportunistic Regenerative relaying, Performance analysis.
Introduction {#Intro}
============
In many wireless applications, users may not be able to support multiple antennas due to size, complexity, power, or other constraints. The wireless medium brings along its unique challenges such as fading and multiuser interference. This can be mitigated with cooperative diversity [@LaW; @SEA1; @SEA2], which is becoming very attractive for small-size, antenna-limited wireless devices. Opportunistic relaying (OR) technique has been proposed where only the best relay from a set of $K$ available candidate relays is selected to cooperate [@Bal06; @TAH09; @Kal09; @Bal10]. With this technique, the selection strategy is to choose the relay with the best equivalent end-to-end channel gain which is calculated as the minimum of the channel gains of the first and the second hops under decode-and-forward (DF) protocol or with the best harmonic mean of both channel gains under amplify-and-forward (AF) protocol. However, some works have chosen the best relay-destination link as possible selection criteria [@TAH09; @IA09].\
Previous works have largely focused on information theoretic aspects of OR and derived outage performance results of such systems. Some of these analysis are accurate only at high signal to noise ratio (SNR) [@Bcal06; @Dal08; @Bal08; @ZL09]. Particularly in [@Bcal06], the end-to-end outage probability analysis of opportunistic relaying without direct link between source and destination nodes was presented. In addition, several works have considered the OR scheme under DF protocol in Rayleigh fading environment, where only the upper bound for the statistics of the best relay local SNR[^2] was obtained [@GDC08; @Adal08]. Moreover, performance analysis of single relay selection for DF protocols were proposed in [@MK08; @FU09; @NB09; @Cal10]. In [@MK08], Michalopoulos and Karagiannidis proposed closed-form expressions for the outage and bit error probability (BEP). However, the activated relay is selected from a decoding set, so that the input signal-to-noise ratio (SNR) is compared to a threshold before forwarding, and the diversity order was not derived explicitly. In [@FU09], Fareed and Uysal considered a relay selection method in a DF multi-relay network where the selected relay cooperates only if the SNR of the source-destination (direct) link is less than the minimum of the channel gains of the first and the second hops. The authors proposed an approximated closed-form symbol error rate (SER) expression. Recently, Nikjah and Beaulieu in [@NB09] offered the first exact performance analysis of opportunistic DF relaying. However [@NB09] focused on outage probability and ergodic capacity performance metrics and the end results were expressed in integral forms. However, in [@Cal10] , Chen **e*t. al* derived only approximate symbol error probability (SEP) expression in integral form for opportunistic DF relaying.
Contributions of this Paper {#Contri}
---------------------------
In this paper we consider a half duplex DF-based cooperative two-hop communications where an opportunistic relaying problem is considered. We state that the objective of this paper is not to revisit path selection, but to focus on giving valid accurate analysis over all SNR regimes. In fact, we determine the exact closed-form expressions of the end-to-end bit error rate (BER) where the source may or may not be able to communicate directly with the destination due to the shadowing. In particular, we consider the important effect of the possible erroneously detected and transmitted data at the regenerative relay. Our analytical approach requires that we determine the probability density function (PDF) of the received SNR by and from the selected relay, called $\gamma_{sr_*}$ and $\gamma_{r_*d}$, respectively. To the best of our knowledge, such performance analysis based on exact statistics (explicit form) of each hop has not been considered in the literature, and using the newly derived exact statistics, we investigate the asymptotic error performance and find the diversity order of these systems.
Organization of this Paper
--------------------------
The remainder of this paper is organized as follows. In section \[systmodel\], we introduce the system model and the statistics of each hop. In section \[Perf\], the accurate closed form for the end-to-end BER is derived and the diversity order of each scheme is determined. Finally, the simulation results for symmetric and linear networks are depicted in section \[simulations\] while some concluding remarks are given in section \[conclu\].
System model {#systmodel}
============
In this section, we describe our proposed cooperative diversity scheme in which the source may or may not be able to communicate directly with the destination, and we note that only a selected relay from a cluster is targeted to cooperate. The source, destination, and relays are denoted as S, D and $r_k$ where $k \in \{1,...,K\}$. We assume that each terminal is equipped with one antenna. We denote $h_{sr_k}$, $h_{sd}$ and $h_{r_{k}d}$ as the coefficients of the channels between the source (S) and the $k^{th}$ relay, the source and the destination (D), and the $k^{th}$ relay and the destination, modeled as flat fading and Rayleigh distributed with variances $\sigma_{sr_k}^{2}$, $\sigma_{sd}^{2}$ and $\sigma_{r_{k}d}^{2}$, respectively.
Fixed Selection Cooperative Relaying (FSCR) {#FSCR}
-------------------------------------------
The source broadcasts the symbols $s(n)$ which are received by (D) and each relay $r_k$ as
\[eqbr1\] $$\begin{aligned}
y_{d}(n) &= \sqrt{E_s} h_{sd} s(n) + n_{d}(n) \label{eqbr1_1}\\
y_{r_k}(n) &= \sqrt{E_s} h_{sr_k} s(n) + n_{r_k}(n) \label{eqbr1_2},\end{aligned}$$
respectively, where $n_{d}(n)$ and $n_{r_k}(n)$ are the additive-noise symbols at the destination and the $k^{th}$ relay, respectively, with the same variance $N_{0}$, and $E_s$ is the symbol energy. Hence, we denote $\gamma_{sd} = E_s |h_{sd}|^{2}/N_{0}$ (resp. $\gamma_{sr_k} = E_s |h_{sr_k}|^{2}/N_{0}$) and $\gamma_{r_{k}d} = E_s |h_{r_{k}d}|^{2}/N_{0}$ the instantaneous received SNR at the destination (resp. at the $k^{th}$ relay) from the source and the $k^{th}$ relay, respectively, and $\bar{\gamma}_{sd} = \sigma_{sd}^{2} E_{s}/N_0$, $\bar{\gamma}_{sr_k} = \sigma_{sr_k}^{2} E_{s}/N_0$ are the average received SNR at the destination and the $k^{th}$ relay, respectively. We assume that the relays are close to each other and forming a cluster[^3] and we assume that the relays and the destination receive the same average SNRs $\bar{\gamma}_{sr_k}$ and $\bar{\gamma}_{r_{k}d}$ from the source and the relays, respectively. Thus, we denote $\bar{\gamma}_{rd} = \bar{\gamma}_{r_{k}d}$ and $\bar{\gamma}_{sr} = \bar{\gamma}_{sr_k}$ for all $k$.\
During the second hop, only a selected relay $r_*$ will transmit using the DF protocol, $$\label{eqd}
y_{d}(n+1) = \sqrt{E_s} h_{r_*d} \widetilde{s}(n) + n_{d}(n+1) ,$$ where $\widetilde{s}(n)$ is the decoded and retransmitted signal by the best relay $r_*$ which is selected following the rule $$r_* = \arg\max_k\min\left(\gamma_{sr_{k}},\gamma_{r_{k}d}\right),$$ where $\min\left(\gamma_{sr_{k}},\gamma_{r_{k}d}\right)$ represents a bottleneck in term of end-to-end capacity[@Adal08]. Therefore, the destination combines the received signals from (S) and $r_*$ using a maximum ratio combining (MRC) detector as $$y_{c} = \left(h_{sd}\right)^{*} y_{d}(n) + \left(h_{rd}\right)^{*} y_{d}(n+1).$$ Hence, the combined SNR at the destination, called $\beta$, is the sum of the two independent SNRs $\gamma_{sd}$ and $\gamma_{r_*d}$ with the corresponding PDFs $p_{\gamma_{sd}}(.)$ and $p_{\gamma_{r_*d}}(.)$ where $$\label{pdfSD}
p_{\gamma_{sd}}(\mathrm{y}) = \frac{1}{\bar{\gamma}_{sd}}~e^{-\mathrm{y}/\bar{\gamma}_{sd}},$$ and the PDF of $\gamma_{r_*d}$ may be shown to be given by (see Appendix \[AppA\]), $$\label{eqPDF}
p_{\gamma_{r_*d}}(\mathrm{x}) = \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}} \frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}}\left(e^{-\mathrm{x}/\bar{\gamma}_{rd}} - e^{-i\mathrm{x}/\bar{\gamma}}\right) + \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{rd}}~i~e^{-i\mathrm{x}/\bar{\gamma}}$$ where $\bar{\gamma} = \frac{\bar{\gamma}_{sr}\bar{\gamma}_{rd}}{\bar{\gamma}_{sr}+\bar{\gamma}_{rd}}$ and $i\bar{\gamma}_{rd} \neq \bar{\gamma}$ $\forall i = 1, 2, \cdots, K$. Therefore, the PDF of $\beta = \gamma_{sd} + \gamma_{r_*d}$ can be obtained by the convolution of the PDF of $\gamma_{sd}$ and $p_{\gamma_{r_*d}}$, as $$\label{pBB}
p_{\beta}(\beta) = \int_{0}^\beta p_{\gamma_{sd}}(\mathrm{x})p_{\gamma_{r_*d}}(\beta-\mathrm{x}) d\mathrm{x}$$ which is expressed in (\[PDFbeta\]).
$$\begin{aligned}
\label{PDFbeta}
p_{\beta}(\beta) &=& \sum_{i=1}^{K} \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}}\frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}} \left[\frac{\bar{\gamma}_{rd}}{\bar{\gamma}_{sd}-\bar{\gamma}_{rd}}\left(e^{-\beta/\bar{\gamma}_{sd}}-e^{-\beta/\bar{\gamma}_{rd}}\right) - \frac{\bar{\gamma}}{i\bar{\gamma}_{sd}-\bar{\gamma}} \left(e^{-\beta/\bar{\gamma}_{sd}}-e^{-i\beta/\bar{\gamma}}\right)\right] \nonumber \\
&{+}& \sum_{i=1}^{K} \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{rd}}\frac{i\bar{\gamma}}{i\bar{\gamma}_{sd}-\bar{\gamma}}\left(e^{-\beta/\bar{\gamma}_{sd}} - e^{-i\beta/\bar{\gamma}}\right)\end{aligned}$$
Distributed Selection Combining (DSC) Scheme {#DSC}
--------------------------------------------
In this scheme, the destination chooses whether to receive from the direct link S-D or the relayed branch according to the instantaneous SNRs $\gamma_{sd}$ and $\gamma_{r_*d}$, respectively. Otherwise, the instantaneous SNR at the output of the selection combining (SC) detector is given by: $$\label{eq1SC}
\gamma_{DSC} = \max\left(\gamma_{sd},\gamma_{r_*d}\right).$$ It can be noted that the statistics of $\gamma_{DSC}$ depends on statistics of $\gamma_{sd}$ and $\gamma_{r_*d}$. In particular, the cumulative density function (CDF) of $\gamma_{DSC}$ is given by $$\label{cdfDSC}
F_{\gamma_{DSC}}(\mathrm{x}) = F_{\gamma_{sd}}(\mathrm{x})~F_{\gamma_{r_*d}}(\mathrm{x}),$$ where the CDFs of $\gamma_{sd}$ and $\gamma_{r_*d}$, denoted as $F_{\gamma_{sd}}(.)$ and $F_{\gamma_{r_*d}}(.)$, respectively, can be derived using the PDFs of $\gamma_{sd}$ and $\gamma_{r_*d}$ in (\[pdfSD\]) and (\[eqPDF\]), respectively, as $$\label{eqCDFsd}
F_{\gamma_{sd}}(\mathrm{x}) = 1-e^{-\mathrm{x}/\bar{\gamma}_{sd}},$$ and $$\label{eqCDFrd}
F_{\gamma_{r_*d}}(\mathrm{x}) = \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}}~\frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}}\left[\Psi_{\frac{1}{\bar{\gamma}_{rd}}}(\mathrm{x}) - \Psi_{\frac{i}{\bar{\gamma}}}(\mathrm{x})\right] + \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}i}{\bar{\gamma}_{rd}}~\Psi_{\frac{i}{\bar{\gamma}}}(\mathrm{x}),$$ where $$\label{eq5SCR}
\Psi_a(\mathrm{x}) = \frac{1}{a}~\left(1-e^{-a\mathrm{x}}\right).$$
Selection Relaying (SR) Scheme {#SR}
------------------------------
In this scheme, it is assumed that the direct link is in deep fading. Hence only $r_*$ will receive the information reliably from the source during the first phase given by (\[eqbr1\_2\]), and the destination decodes only the message coming from $r_*$ as given by (\[eqd\]).
BER analysis {#Perf}
============
Fixed Selection Cooperative Relaying {#SCRan}
------------------------------------
The destination combines the received signals such as the relay can retransmit an erroneously decoded message. The end-to-end probability of error can be expressed as $$\label{eqSCR1}
P_{e,FSCR} = P_{prop} P_{sr_*} + \left(1 - P_{sr_*}\right) P_{mrc}$$ where $P_{prop}$ denotes the error propagation probability which can be tightly approximated for a BPSK modulation by $$\label{eqSCR2}
P_{prop} \approx \frac{\bar{\gamma}_{r_*d}}{\bar{\gamma}_{r_*d} + \bar{\gamma}_{sd}},$$ where $\bar{\gamma}_{r_*d}$, the expected value of $\gamma_{r_*d}$, can be easily verified to be expressed as $$\label{eqbar}
\bar{\gamma}_{r_*d} = \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}}\frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}}\left[\bar{\gamma}_{rd}^2 - \left(\frac{\bar{\gamma}}{i}\right)^2\right] + \sum_{i=1}^K \left(_i^K\right)~\frac{(-1)^{i-1}}{i\bar{\gamma}_{rd}}~\bar{\gamma}^2 ,$$ and $P_{sr_*}$ is the probability of error for the communication link between the source and the relay which is derived by $$\label{eqSCR3}
P_{sr_*} = \int_0^{\infty} \frac{1}{2}~\mathrm{erfc}(\sqrt{\mathrm{x}})p_{\gamma_{sr_*}}(\mathrm{x})\mathrm{dx}$$ where $p_{\gamma_{sr_*}}(.)$ can be derived as in (\[eqPDF\]) by replacing $\gamma_{r_*d}$ with $\gamma_{sr_*}$ and $\mathrm{erfc}(.)$ is the complementary error function. Hence performing the integration in (\[eqSCR3\]), and using the following identity $$l(\alpha) \triangleq \frac{1}{2}~\int_0^\infty \mathrm{erfc}(\sqrt{\beta}) e^{-\alpha\beta} \mathrm{d}\beta = \frac{1}{2\alpha}\left[1-\frac{1}{\sqrt{1+\alpha}}\right],$$ $P_{sr_*}$ is found to be expressed as in (\[eqSCR4\]).
$$\label{eqSCR4}
P_{sr_*} = \sum_{i=1}^{K} \left(_i^K\right)\frac{(-1)^{i-1}}{\bar{\gamma}_{rd}} \frac{i\bar{\gamma}}{i\bar{\gamma}_{sr}-\bar{\gamma}}\left[l\left(\frac{1}{\bar{\gamma}_{sr}}\right) - l\left(\frac{i}{\bar{\gamma}}\right)\right] + \sum_{i=1}^{K} \left(_i^K\right)~\frac{i (-1)^{i-1}}{\bar{\gamma}_{sr}}l\left(\frac{i}{\bar{\gamma}}\right).$$
In (\[eqSCR1\]), we need also $P_{mrc}$ which is the error probability of the combined direct and opportunistic paths, given by $$\label{MRCint}
P_{mrc} = \int_{0}^{\infty} \frac{1}{2}~\mathrm{erfc}(\sqrt{\beta})p(\beta) \mathrm{d}\beta ,$$ which is given in (\[eqSCR5\]).
$$\begin{aligned}
\label{eqSCR5}
P_{mrc} &=& \sum_{i=1}^{K} \left(_i^K\right)\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}}\frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}} \left[\frac{\bar{\gamma}_{rd}}{\bar{\gamma}_{sd}-\bar{\gamma}_{rd}} \left(l\left(\frac{1}{\bar{\gamma}_{sd}}\right)-l\left(\frac{1}{\bar{\gamma}_{rd}}\right)\right)
-\frac{\bar{\gamma}}{i\bar{\gamma}_{sd}-\bar{\gamma}}\left(l\left(\frac{1}{\bar{\gamma}_{sd}}\right)-l\left(\frac{i}{\bar{\gamma}}\right)\right)\right] \nonumber \\
&&{+} \sum_{i=1}^{K} \left(_i^K\right)\frac{(-1)^{i-1}}{\bar{\gamma}_{rd}}\frac{i\bar{\gamma}}{i\bar{\gamma}_{sd}-\bar{\gamma}}\left(l\left(\frac{1}{\bar{\gamma}_{sd}}\right)-l\left(\frac{i}{\bar{\gamma}}\right)\right)\end{aligned}$$
Finally, with (\[eqSCR2\]), (\[eqbar\]), (\[eqSCR4\]) and (\[eqSCR5\]), the end-to-end probability of error can be easily evaluated.
\[lemma1\] For a source-destination pair with $K$ potential relays in Rayleigh fading channels, the end-to-end BER of the fixed selection cooperative relaying scheme in the high-SNR regime, is $$\label{eqLemma1}
P_{e,FSCR} \approx P_{prop}\frac{\Gamma(K+\frac{1}{2})}{2\sqrt{\pi}}\frac{1}{\bar{\gamma}_{sr}}\left(\frac{1}{\bar{\gamma}}\right)^{K-1} + \frac{\Gamma(K+\frac{3}{2})}{2\sqrt{\pi}(K+1)}\frac{1}{\bar{\gamma}_{sd}}\frac{1}{\bar{\gamma}_{rd}}\left(\frac{1}{\bar{\gamma}}\right)^{K-1},$$
See Appendix \[AppB\].
Distributed Selection Combining Scheme
--------------------------------------
The destination selects the best coming path and the end-to-end BER is found to be $$\label{eqDSC1}
P_{e,DSC} = P_{prop} P_{sr_*} + \left(1 - P_{sr_*}\right) P_{DSC},$$ where $P_{prop}$ and $P_{sr_*}$ are detailed above and $P_{DSC}$ is the probability of error for the selected link communication to the destination. Based on a general result in [@T04 Eq. 32], we can derive $P_{DSC}$ for a BPSK modulation as $$\label{eqDSC2}
P_{DSC} = \frac{1}{2\sqrt{\pi}}\int_{0}^{\infty} \frac{e^{-\mathrm{z}}}{\sqrt{\mathrm{z}}}~F_{\gamma_{DSC}}(\mathrm{z})d\mathrm{z},$$ which can be rewritten as $$\label{eqDSC3}
P_{DSC} = \underbrace{\frac{1}{2\sqrt{\pi}}\int_{0}^{\infty} \frac{e^{-\mathrm{z}}}{\sqrt{\mathrm{z}}}~F_{\gamma_{r_*d}}(\mathrm{z})d\mathrm{z}}_{I_1} - \underbrace{\frac{1}{2\sqrt{\pi}}\int_{0}^{\infty} \frac{e^{-\mathrm{z}(1+1/\bar{\gamma}_{sd})}}{\sqrt{\mathrm{z}}}~F_{\gamma_{r_*d}}(\mathrm{z})d\mathrm{z}}_{I_2}.$$ It should be noted that $I_1$ defines the probability of error for the communication link between $r_*$ and D, which can be expressed as in (\[eqDSC4\])
$$\label{eqDSC4}
I_1 \triangleq P_{r_*d} = \sum_{i=1}^{K} \left(_i^K\right)\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}} \frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}}\left[l\left(\frac{1}{\bar{\gamma}_{rd}}\right) - l\left(\frac{i}{\bar{\gamma}}\right)\right] + \sum_{i=1}^{K} \left(_i^K\right) \frac{i(-1)^{i-1}}{\bar{\gamma}_{rd}}~l\left(\frac{i}{\bar{\gamma}}\right),$$
and $I_2$ can be derived with the help of the following identity $$\label{eqDSC5}
\Theta_{a} \triangleq \frac{1}{2\sqrt{\pi}} \int_{0}^{\infty} \frac{e^{-\mathrm{z}(1+1/\bar{\gamma}_{sd})}}{\sqrt{\mathrm{z}}}~\Psi_{a}(\mathrm{z})d\mathrm{z} = \frac{1}{2a}\left[\sqrt{\frac{1}{1 + \frac{1}{\bar{\gamma}_{sd}}}} - \sqrt{\frac{1}{1 + a + \frac{1}{\bar{\gamma}_{sd}}}}~\right].$$ and be given by (\[eqDSC6\]).\
$$\label{eqDSC6}
I_{2} = \sum_{i=1}^{K} \left(_i^K\right)\frac{(-1)^{i-1}}{\bar{\gamma}_{sr}} \frac{i\bar{\gamma}}{i\bar{\gamma}_{rd}-\bar{\gamma}}\left[\Theta_{\frac{1}{\bar{\gamma}_{rd}}} - \Theta_{\frac{i}{\bar{\gamma}}}\right] + \sum_{i=1}^{K} \left(_i^K\right) \frac{i(-1)^{i-1}}{\bar{\gamma}_{rd}}~\Theta_{\frac{i}{\bar{\gamma}}},$$
Finally, with (\[eqSCR2\]), (\[eqbar\]), (\[eqSCR4\]), (\[eqDSC4\]) and (\[eqDSC6\]), the end-to-end probability of error can be easily evaluated.
\[lemma2\] For a source-destination pair with $K$ potential relays in Rayleigh fading channels, the end-to-end BER of the distributed selection combining scheme in the high-SNR regime, is $$\label{eqLemma2}
P_{e,DSC} \approx P_{prop}\frac{\Gamma(K+\frac{1}{2})}{2\sqrt{\pi}}\frac{1}{\bar{\gamma}_{sr}}\left(\frac{1}{\bar{\gamma}}\right)^{K-1} + \frac{\Gamma(K+\frac{3}{2})}{2\sqrt{\pi}}\frac{1}{\bar{\gamma}_{sd}}\frac{1}{\bar{\gamma}_{rd}}\left(\frac{1}{\bar{\gamma}}\right)^{K-1}.$$
See Appendix \[AppC\].
Selection Relaying {#SRan}
------------------
In this form of relaying it is assumed that the direct path is unusable due to the deep fade instances or heavy shadowing [@Adal08]. Therefore the end-to-end BER is found to be $$\label{eqSR1}
P_{e,SR} = P_{sr_*} + P_{r_*d} - P_{sr_*}P_{r_*d},$$ where $P_{sr_*}$ was already defined by (\[eqSCR4\]), and $P_{r_*d}$ is the probability of error for the communication link between $r_*$ and D which can be expressed by (\[eqDSC4\]).
\[lemma3\] For a source-destination pair with $K$ potential relays in Rayleigh fading channels, the end-to-end BER of the selection relaying scheme in the high-SNR regime, is $$\label{eqLemma3}
P_{e,SR} \approx \frac{\Gamma(K+\frac{1}{2})}{2\sqrt{\pi}}~\left(\frac{1}{\bar{\gamma}}\right)^{K}.$$
Performance results {#simulations}
===================
Network Geometry
----------------
We anticipate that cooperation will perform differently as function of the positions of the mobiles with respect to the destination. Hence we study an [*asymmetric*]{} or [*linear*]{} network (LN) where we model the path-loss, i.e. the mean channel powers $\sigma_{ij}^2$, as a function of the relays cluster position $d$ by $$\label{eqsigma}
\sigma_{sd}^2 = 1,~ \sigma_{sr_k}^2 = d^{-\nu},~ \sigma_{r_{k}d}^2 = (1-d)^{-\nu},$$ where $\nu$ is the path loss exponent and $0 < d (= \textrm{distance}_{s-cluster}) < 1$. The distances are normalized by the distance $d_{sd}$. In these coordinates, the source can be located at (0,0), the destination can be located at (1,0), without loss of generality, and the relays are located at ($d$,0).
Simulation Results {#SimRes}
------------------
In this section, we evaluate the performance of our schemes in terms of the end-to-end BER at the destination as function of the SNR $= E_b/N_0$ for a number ($K$) of potential relays in phase II. All schemes were simulated assuming BPSK modulation. It is also assumed that the amplitudes of the fading from each transmit antenna to each receive antenna are uncorrelated in the case of cooperative selection relaying scheme and Rayleigh distributed. Furthermore, we assumed that all receivers have the same noise properties. This implies that in all depicted figures, the noise power of all paths is the same. Further, we assumed that the receiver has perfect knowledge of the channels.\
Figures \[fig:fig1\]-\[fig:fig3\] depict the end-to-end error-rate performance and the corresponding asymptotic curves in LN networks as function of SNR for FSCR, DSC and SR schemes, respectively, where a relays cluster is located at different distances $d$ from the source. Figure \[fig:fig4\] shows performance comparison between FSCR and DSC schemes. All figures compare the analytical and simulation results for $K = 2$ and $K = 4$, respectively.\
Figures \[fig:fig1\] and \[fig:fig2\] depict the end-to-end BER as function of the SNR for the FSCR and DSC schemes, respectively, where the relays cluster is located at $d = 0.1$ and $d = 0.5$, respectively. It could be noted that the diversity order is $K$ when the relay cluster is located at the mid-distance to the destination. This is due to the fact that the error propagation probability, $P_{prop}$, is becoming close to 1. The full diversity order is recovered when $d = 0.1$. In addition, It may be noted that the gap, between FSCR and DSC cures, is shrinked when $d = 0.5$. Therefore, DSC scheme could be considered as appropriate since its BER penalty is minor as shown in figure \[fig:fig4\], and it is considered as the less complicated than MRC [@AlouiniBook].\
Figure \[fig:fig3\] depicts the end-to-end BER as function of the SNR for the SR scheme for the same network architectures. We note that the SR scheme do better when the relay cluster is located in the middle between the the source and the destination, and the full diversity order is achieved as expected by (\[eqLemma3\]). Our proposed analysis is well confirmed by the simulation results.
Conclusions {#conclu}
===========
In this work, we studied three opportunistic cooperation protocols namely fixed selection cooperative relaying, distributed selection combining and selection relaying, based on DF transmission in a Rayleigh fading environment. We provided exact statistics, and as result, we presented the BER performance analysis as well as the asymptotic analysis. We performed several simulations to confirm our theoretical analysis.
Derivation of Eq. (\[eqPDF\]) {#AppA}
-----------------------------
Based on our problem formulation we can write the PDF $p_{\gamma_{r_*d}}(\mathrm{x})$ as follows $$\label{A1eq1}
p_{\gamma_{r_*d}}(\mathrm{x}) = \int_0^{\infty} p_{\gamma_{r_id}/Z_i = \mathrm{z}}(\mathrm{x}) . p_{\max(Z_i)}(\mathrm{z}) d\mathrm{z},$$ where $Z_i = \min(\gamma_{sr_i},\gamma_{r_id})$.\
Using the Bayes rule, it is well known that the conditional probability density function can be expressed as $$\label{A1eq2}
p_{\gamma_{r_id}/Z_i = \mathrm{z}}(\mathrm{x}) = \frac{p_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z})}{p_{Z_i}(\mathrm{z})}.$$ Now, we can show that the cumulative density function (CDF) of $Z_i$ can be expressed as $$\label{A1eq3}
F_{Z_i}(\mathrm{z}) = 1 - Pr[\gamma_{sr_i} \geq \mathrm{z}]Pr[\gamma_{r_id} \geq \mathrm{z}] = 1 - e^{-\mathrm{z}/\bar{\gamma}},$$ where $\bar{\gamma} = \frac{\bar{\gamma}_{sr}\bar{\gamma}_{rd}}{\bar{\gamma}_{sr}+\bar{\gamma}_{rd}}$, and the joint CDF $F_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z})$ can be expressed as $$\begin{aligned}
\label{A1eq4}
&&{} F_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z}) = Pr[\gamma_{r_id}<\mathrm{x},\min(\gamma_{sr_i},\gamma_{r_id})<\mathrm{z}] \nonumber \\
&&{=} \left\{\begin{array}{ll}
F_{\gamma_{r_id}}(\mathrm{x})-\left(F_{\gamma_{r_id}}(\mathrm{x})-F_{\gamma_{r_id}}(\mathrm{z})\right)\left(1-F_{\gamma_{sr_i}}(\mathrm{z})\right), & \mathrm{x} \geq \mathrm{z} \\
F_{\gamma_{r_id}}(\mathrm{x}), & \mathrm{x} < \mathrm{z}
\end{array}
\right.\end{aligned}$$ We note that the joint CDF $F_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z})$ is not continuous along the $\mathrm{x}$ direction at $\mathrm{x} = \mathrm{z}$. Therefore, the result of the derivative (i.e the joint PDF $p_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z})$) involves an impulse at the position $\mathrm{x} = \mathrm{z}$. Specially, the joint PDF $p_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z})$ is given by $$p_{\gamma_{r_id},Z_i}(\mathrm{x},\mathrm{z}) = \left\{\begin{array}{ll}
p_{\gamma_{r_id}}(\mathrm{x})p_{\gamma_{sr_i}}(\mathrm{z}) + p_{\gamma_{r_id}}(\mathrm{x})\left(1-F_{\gamma_{sr_i}}(\mathrm{z})\right)\delta(\mathrm{x}-\mathrm{z}) , & \mathrm{x} \geq \mathrm{z} \\
0, & \mathrm{x} < \mathrm{z}
\end{array}
\right.$$ It follows the PDF of $\gamma_{r_*d}$ is given by $$\label{A1eq6}
p_{\gamma_{r_*d}}(\mathrm{x}) = \int_0^{\mathrm{x}} \frac{p_{\gamma_{r_id}}(\mathrm{x})p_{\gamma_{sr_i}}(\mathrm{z})}{p_{Z_i}(\mathrm{z})}~p_{\max(Z_i)}(\mathrm{z}) d\mathrm{z} + \frac{p_{\gamma_{r_id}}(\mathrm{x}) \left(1-F_{\gamma_{sr_i}}(\mathrm{x})\right)}{p_{Z_i}(\mathrm{x})}~p_{\max(Z_i)}(\mathrm{x})$$ where $$\label{A1eq7}
p_{Z_i}(\mathrm{z}) = \frac{1}{\bar{\gamma}} e^{-\mathrm{z}/\bar{\gamma}},$$ It can be easily shown that $F_{\max(Z_i)}(\mathrm{z})$ be expressed as $$\label{A1eq8}
F_{\max(Z_i)}(\mathrm{z}) = \prod_{i=1}^{K} Pr\left[Z_i < \mathrm{z}\right] = \left(1 - e^{-\mathrm{z}/\bar{\gamma}}\right)^K$$ Therefore after taking the derivative, we have $$\label{A1eq9}
p_{\max(Z_i)}(\mathrm{z}) = \frac{K}{\bar{\gamma}}e^{-\mathrm{z}/\bar{\gamma}}\left(1 - e^{-\mathrm{z}/\bar{\gamma}}\right)^{K-1} = \sum_{i=1}^{K} \left(_i^K\right)~(-1)^{i-1}~\frac{i}{\bar{\gamma}}~e^{-i\mathrm{z}/\bar{\gamma}}.$$ where the second equality holds from the binomial expansion.\
Substituting (\[A1eq7\]) and (\[A1eq9\]) in (\[A1eq1\]), Eq. (\[eqPDF\]) is derived by performing the integration.
Derivation of Eq. (\[eqLemma1\]) {#AppB}
--------------------------------
It is easy to note that $P_{e,FSCR}$ could be approximated by $$\label{Beq1}
P_{e,FSCR}^{\infty} = P_{prop} P_{sr_*}^{\infty} + P_{mrc}^{\infty}$$ where $P_{sr_*}^{\infty}$ and $P_{mrc}^{\infty}$ are the approximated expressions of $P_{sr_*}$ and $P_{mrc}$, respectively.\
To this end, we needed the approximated expressions of the PDFS $p_{\gamma_{sr_*}}(.)$ and $p_{\beta}(.)$ which are given by $$\label{Beq2}
p_{\gamma_{sr_*}}(\mathrm{x}) \approx \frac{K}{\bar{\gamma}_{sr}} \left(\frac{1}{\bar{\gamma}}\right)^{K-1} \mathrm{y}^{K-1}, ~\mathrm{y} > 0,$$ and $$\label{Beq3}
p_{\beta}(\beta) \approx \frac{1}{\gamma_{sd}}\frac{1}{\bar{\gamma}_{rd}}\left(\frac{1}{\bar{\gamma}}\right)^{K-1} \beta^{K}, ~\mathrm{\beta} > 0$$ Based on (\[Beq2\]) and (\[Beq3\]), it becomes easy to derive $P_{sr_*}^{\infty}$ and $P_{mrc}^{\infty}$ by using integrations in (\[eqSCR3\]) and (\[MRCint\]).
Derivation of Eq. (\[eqLemma2\]) {#AppC}
--------------------------------
It is easy to note that $P_{e,DSC}$ could be approximated by $$\label{Ceq1}
P_{e,DSC}^{\infty} = P_{prop} P_{sr_*}^{\infty} + P_{DSC}^{\infty}$$ where $P_prop$ and $P_{sr_*}^{\infty}$ are already given, and $P_{DSC}^{\infty}$ is the approximated expressions of $P_{DSC}$ which can be derived by integrating the approximated expression of $F_{\gamma_{DSC}}(.)$. For this end, let us start by approximating $F_{\gamma_{sd}}(.)$ and $F_{\gamma_{r_*d}}(.)$, given by $$\label{Ceq2}
F_{\gamma_{sd}} \approx \frac{\mathrm{z}}{\bar{\gamma}_{sd}},~\mathrm{z} > 0,$$ and $$\label{Ceq3}
F_{\gamma_{r_*d}}(\mathrm{z}) \approx \frac{1}{\bar{\gamma}_{rd}}~\frac{1}{\bar{\gamma}}~\mathrm{z}^{K},~\mathrm{z} > 0.$$ Using the following identity $$\label{Ceq4}
\int_0^{\infty} \mathrm{z}^{K+\frac{1}{2}} e^{-\mathrm{z}} d\mathrm{z} = \Gamma\left(K + \frac{3}{2}\right),$$ and substituting (\[Ceq2\]) and (\[Ceq3\]) in (\[eqDSC2\]), Eq. (\[eqLemma2\]) is derived by performing the integration.
![End-to-end bit error rate versus SNR of the FSCR scheme using a DF transmission in the LN case for $K = 2$ and $K = 4$.[]{data-label="fig:fig1"}](fig1.eps)
![End-to-end bit error rate versus SNR of the DSC scheme using a DF transmission in the LN case when $K = 2$ and $K = 4$.[]{data-label="fig:fig2"}](fig2.eps)
![End-to-end bit error-rate versus SNR of the SR scheme using a DF transmission in the LN case when $K = 2$ and $K = 4$.[]{data-label="fig:fig3"}](fig3.eps)
![Comparisons of error performance versus SNR of the FSCR and DSC schemes using a DF transmission in the LN case when $K = 2$ and $K = 4$..[]{data-label="fig:fig4"}](fig4.eps)
[^1]: This work was supported by the Qatar National Research Fund (A member of Qatar Foundation). Kamel Tourki is with the Electrical and Computer Engineering Program, Texas A&M University at Qatar, Education City, Doha, Qatar. Hong-Chuan Yang is with the Department of Electrical and Computer Engineering, University of Victoria, BC, Canada. Mohamed-Slim Alouini is with the Electrical Engineering Program, Division of Physical Sciences and Engineering, King Abdullah University of Science and Technology (KAUST), Thuwal, Mekkah Province, Kingdom of Saudi Arabia. The corresponding author is Kamel Tourki ([email protected]).
[^2]: The statistic refers to the probability density function (PDF) of the received SNR at the destination, called $\gamma_{r_*d}$, from the best relay $r_*$.
[^3]: We assume short distances between the relays compared to the distances (S)-cluster, and cluster-(D), respectively.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give an explicit axiomatic formulation of the quantum measurement theory which is free of the projection postulate. It is based on the generalized nondemolition principle applicable also to the unsharp, continuous–spectrum and continuous-in-time observations. The collapsed state–vector after the objectification is simply treated as a random vector of the *a posteriori* state given by the quantum filtering, i.e., the conditioning of the *a priori* induced state on the corresponding reduced algebra. The nonlinear phenomenological equation of continuous spontaneous localization has been derived from the Schrödinger equation as a case of the quantum filtering equation for the diffusive nondemolition measurement. The quantum theory of measurement and filtering suggests also another type of the stochastic equation for the dynamical theory of continuous reduction, corresponding to the counting nondemolition measurement, which is more relevant for the quantum experiments.'
address: 'Philipps Universität, Fachbereich Physik D–3550, Marburg, Germany and University of Nottingham, NG7 2RD, UK'
author:
- 'V.P. Belavkin'
date: 'Received August 31, 1992 '
nocite:
- '[@bib:1; @2; @3]'
- '[@bib:7; @8; @9; @10]'
- '[@bib:24; @25; @26]'
- '[@bib:27; @28]'
- '[@bib:21; @22; @23]'
- '[@bib:24; @25]'
title: Nondemolition Principle of Quantum Measurement Theory
---
[^1]
The status of quantum measurement theory
========================================
Quantum measurement theory, based on the ordinary von Neumann or a generalized reduction postulate, was never an essential part of quantum physics but rather of metaphysics. First, this was because the orthodox quantum theory had always dealt with a closed quantum system while the object of measurement is an open system due to the interaction with the measurement apparatus. Second, the superposition principle of quantum mechanics, having dealt with simple algebras of observables, is in contradiction with the von Neumann projection postulate while it may be not so in the algebraic quantum theory with the corresponding superselection rules. Third, due to the dynamical tradition in quantum theory going on from the deterministic mechanics, the process of the measurement was always considered by theoretical physicists as simply just an ordinary interaction between two objects while any experimentalist or statistician knows that this is a stochastic process, giving rise to the essential difference between a *priori* and a *posteriori* description of the states.
The last and most essential reason for such an unsatisfactory status of the quantum measurement theory was the limitations of the projection postulate applicable only to the instantaneous measurement of the observables with the discrete spectra, while the real experiments always have a finite duration and the most important observation is the measurement of the position having the continuous spectrum.
There are many approaches to the theory of quantum measurement ranging from purely philosophical to qualitative and even quantitative theories in which the projection postulate apparently is not needed or is generalized to meet the indirect, or unsharp, measurements \[1–10\].[bib:4,5,6]{}
The most general, the philosophical level, of the discussion of these problems is of course the simplest and the appropriate one for the largest audience. But it provides room for unprofessional applications of the more sophisticated theoretical arguments, giving rise to different kinds of the speculations and paradoxes. I believe that the professional standard of quantum measurement theory ought to be an axiomatic and rigorous one and the quantum measurement problems must be formulated within it and solved properly instead of making speculations.
In order to examine the quantum paradoxes of Zeno type related to the continuous measurements, the study must be based on advanced mathematical methods of the quantum theory of compound systems with not regular but rather singular interaction, and this has recently received a stochastic treatment in the quantum theory of open systems and noise. It must use the tools of the quantum algebraic theory for the calculus of input fields of the apparatus, i.e., the quantum noises which usually have an infinite number of degrees of freedom, and for the superselection of output fields, i.e., commutative (classical) pointer processes which are usually the stochastic processes in continuous time.
Perhaps some philosophers and physicists would not like such a treatment of quantum measurement theory; the more mathematical a theory is the less philosophical it is, and the more rigorous it is, the less alive it is. But this is just an objective process of the development of any scientific theory and has already happened with the classical information and measurement theory.
The corresponding classical dynamical measurement theory, called the stochastic filtering theory, was developed in the beginning of the 60’s by Stratonovich \[11\] and for the particular linear case by Kalman \[12\]. This theory, based on the notion of the partial (unsharp) observation and the stochastic calculus method, is optional for the classical deterministic physics, having dealt with the complete (sharp) observations of the phase trajectories and ordinary differential calculus, and is usually regarded as a part of the stochastic systems theory or, more precisely, the classical information and control theory. The main task of the filtering theory is to derive and solve a stochastic reduction equation for the present posterior state of the object of incomplete measurement, giving a means to calculate the conditional probabilities of the future observations with respect to the results of the past measurements. The corresponding filtering equation describes, for example, the continuous spontaneous localization of the classical Brownian particle under an unsharp observation as the result of the dynamical reduction of the statistical posterior state given by the classical conditional expectations under the continuous increase of the interval of the observation. The stochasticity of this nonlinear equation is generated either by the Wiener process or by the Poisson process, or by mixture of them, corresponding to the diffusive, counting, or mixed type of continuous measurement on the fixed output. It can be also written in the linear form in terms of the classical renormalized state vector (probability density), and is sometimes called “the Schrödinger equation of the classical systems theory” to emphasize its importance and the probabilistic interpretation.
Recently the corresponding quantum filtering theory was developed for the different types of continuous observations, \[13,14\], although the particular linear case of quantum Kalman filter was proposed by the author much earlier \[6,7\]. This gives rise to an axiomatic quantum measurement theory based on the new quantum calculus method to handle rigorously the singular interactions of the quantum object and input fields, and based on the generalized nondemolition principle to select properly the output observable processes. The mathematical quantum measurement theory plays the same central role in the general quantum theory of compound systems containing the information and control channels. as in the classical systems theory. But in distinction to the classical case it is not optional for the quantum physics due to the irreducible probabilistic nature of quantum mechanics which results in the absence of the phase trajectories. There is no need in this theory to use the projection or any other reduction postulate. But it does not contradict the quantum theory, as claimed in Ref. \[15\], and its application can be derived in the relevant cases simply as the result of state vector filtering by means of which the conditional probabilities of the future observation with respect to the results of the past measurements are calculated.
There is no need to postulate different nonlinear stochastic modifications of the Schrödinger equation in the phenomenological theories of spontaneous localization or of the nonstandard quantum theories of dynamical reduction and continuous collapse, \[16–20\] and to argue which type is more universal. They all are given as particular cases \[21–24\] of the general diffusive type quantum filtering equation, \[25\], rigorously derived by conditioning the corresponding Schrödinger equation for the uniquely determined minimal compound quantum system in Fock–Hilbert space.
The quantum filtering theory gives also a new type of phenomenological stochastic equations which are relevant to the quantum mechanics with spontaneous localization, \[19,20\], corresponding to the random quantum jumps, \[26–28\]. This pure discontinuous type is also rigorously derived from the Schrödinger equation \[29\] by conditioning the continuous-in-time counting measurement which contains the diffusive type as the central limit case \[30\].
Thus, the stochastic nature of measurement processes is reconciled with unitarity and deterministic interaction on the level of the compound system. But to account for the unavoidable noise in the continuous observation the unitary model necessarily involves a quantum system with infinitely many degrees of freedom and a singular interaction.
The purpose of this paper is to describe explicitly a new universal nondemolition principle for quantum measurement theory which makes possible the derivations of the reduction postulates from the quantum interactions. We show on simple examples what it means to derive rigorously the quantum filtering equation (thus the Hilbert stochastic process) by conditioning a Schrödinger equation for a compound system. Here, we demonstrate these derivations from the corresponding unitary interactions with the apparatus for the particular cases of the measurement of a single observable with the trivial Hamiltonian $H=0$ of the object using the operator quantum calculus method instead of the quantum stochastic one \[21–23\]. But if one wants to obtain such results in nontrivial cases related to the dynamical observables that are continuous in time and continuous in spectra and that do not commute with $H\neq 0$, one needs to use the appropriate mathematical tools, such as quantum differential calculus and quantum conditional expectations, recently developed within the algebraic approach in quantum probability theory. Otherwise, one would be in the same situation as trying to study the Newton mechanics in nontrivial cases without using the ordinary differential calculus.
Note that the quantum filtering equation was first obtained in a global form \[9\] and then in the differential form \[30\] within the operational approach, \[1,2\], giving the reduced description of the open quantum systems and quantum continuous measurements. This was done by the stochastic representation of the continuous instrument, described by the semigroup of the operational valued measures which are absolutely continuous with respect to the standard Wiener or Poisson process. The most general approach \[31\] to these problems is based on the quantum stochastic calculus of nondemolition measurements and quantum conditional expectations. It clearly shows that the operational semigroup approach is restricted to only the Markovian case of the quantum stochastic object as an open system and to the conditionally independent nondemolition observations describing the output of the compound system.
Causality and nondemolition principle
=====================================
Let us begin with the discussion of the quantum nondemolition principle which forms the basis of the axiomatic formulation of the quantum measurement theory without the projection postulate, and which has been implicitly explored also in other approaches \[1–10\]. The term “nondemolition measurement” was first introduced into the theory of ultrasensitive gravitational experiments by Braginski and others \[32–34\] to describe the sequential observations in a quantum Weber antenna as a simultaneous measurement of some quantum observables. But the property of nondemolition has never been formalized or even carefully described other than by requiring the commutativity of the sequential observables in the Heisenberg picture, which simply means that the measurement process can be represented as a classical stochastic one by the Gelfand transformation. Therefore no essentially quantum, noncommutative results have been obtained, and no theorems showing the existence of such measurements in nontrivial time continuous models have been proved.
An operator $X$ in a Hilbert space $\mathcal{H}$ is said to be demolished by an observable $Y=Y^{\dagger }$ in $\mathcal{H}$ if the expectation $\langle
X\rangle $ is changed for $\langle \tilde{X}\rangle \neq \langle X\rangle $ in an initial state when $Y$ has been measured, although without reading. According to the projection postulate the demolished observable $\tilde{X}%
=\delta \lbrack X]$ is described by the reduction operation $\delta \lbrack
X]=\sum P_{i}XP_{i}$ for a discrete observable $Y=\sum y_{i}P_{i}$ given by the orthoprojectors $P_{i}^{2}=P_{i}=P_{i}^{\dagger }$, $\sum P_{i}=I$ and eigenvalues $\{y_{i}\}$. The observable $Y$ is nondemolition with respect to $X$ if $\delta \lbrack X]$ is compatible, $\langle \delta \lbrack X]\rangle
=\langle X\rangle $, with respect to each initial state, i.e., iff $\delta
\lbrack X]=X$. It follows immediately in this discrete case that the nondemolition condition is $XY=YX$, as the main filtering theorem says \[30\] even in the general case. Moreover, for each demolition observable $Y$ there exists a nondemolition representation $\tilde{Y}=\varrho \lbrack Y]$ in an extended Hilbert space $\mathcal{H}\otimes \mathcal{F}$, which is statistically equivalent to $Y$ in the sense that $\langle \tilde{X}Y\rangle
=\langle X\tilde{Y}\rangle $ for each input state in $\mathcal{H}$ and corresponding output state in $\mathcal{H}\otimes \mathcal{F}$. This follows from the reconstruction theorem \[35\] for quantum measurements giving the existence of the nondemolition representation for any kind of observations, which might be even continuously distributed in the relativistic space-times $\mathbb{R}^{1+d}$. In the case of a single discrete observable $Y$ it proves the unitary reconstruction of the projection postulate, which is given in section 3.
Now we give several equivalent formulations of the dynamical nondemolition considered not just as a *possible* property for the quantum measurements but rather as the *universal* condition to handle such problems as the modeling of the unsharp measurements, the generalized reduction and instantaneous collapse for the continuous spectrum observables, the quantum sequential measurements, and the dynamical reduction and spontaneous localization under the continuous-in-time observation. This condition, based on the reconstruction theorem, was discovered in Ref. \[7\] and consists of a new principle of quantum axiomatic measurement theory for the proper representation of the observable process in a Hilbert space, such as the interaction representation of the object with the measurement apparatus.
On the philosophical level, one can say that the nondemolition principle is equivalent to the quantum causality principle of the statistical predictability for the present and all possible future observations and for all possible initial states from the a posteriori probability distributions which are conditioned by the results of the past measurements. This should be regarded rather as the physical content and purpose of this principle and not as a definition.
On the mathematical level the nondemolition principle must be formulated as a necessary and sufficient condition for the existence of the conditional expectations on the algebras generated by the present and future Heisenberg operators of the object of the measurement and all the output observables with respect to the subalgebras of the past measurements and arbitrary input states.
In the most general algebraic approach this formulation was first obtained in Ref. \[7\], (see also Refs. \[13\] and \[14\]) as the condition $$[X(t),Y(s)]:=X(t)Y(s)-Y(s)X(t)=0\ ,\qquad \forall s\geq t \label{eq:1.1}$$ of compatibility of all system operators $X(t)$ considered as the possible observables at a time instant $t$ with all past observables $Y(s)$, $s\leq t$, which have been measured up to $t$. It says that the Heisenberg operators $%
X(t)$ of the quantum object of the measurement given, say, in the interaction representation with the apparatus must commute with all past output observables $Y(s)$, $s\leq t$, of the pointer for any instant $t$. And according to the causality principle there is no restriction on the choice of the future observables $Y(r)$, $r\geq t$, with respect to the present operators $X(t)$ except the self-nondemolition $[Y(r),Y(s)]=0$ for the compatibility of the family $\{Y(t)\}$. Generalized then in [bib:21,22,23]{} \[21–28\] for arbitrary $X$ and $Y$, these conditions define a stochastic process $Y(t)$ which is nondemolition with respect to a given quantum process $X(t)$. Note that the condition (\[eq:1.1\]) for clearly distinguished object and pointer observables does not reduce completely the algebra of the compound system to the commutative one as it does in the case of the direct observations $Y=X$ when it reads as the self-nondemolition condition $[X(t),X(s)]=0$, $\forall
t, s$. The nondemolition measurements considered in Refs. \[32–34\] were defined only by the self-nondemolition condition, corresponding to this trivial (Abelian) case $X(t)=Y(t)$.
In the operational approach \[1,2\], applicable for the reduced description of the quantum Markov open system, one might prefer to have a condition that is equivalent to the nondemolition principle in that case. It can be given in terms of the induced states on the reduced algebra, i.e., of the states given by the expectations $\phi (Z)=\langle \psi ,Z(t)\psi \rangle $ on the algebra of observables $Z$ generated in the Heisenberg picture $%
Z(t)=U^{\dagger }(t)ZU(t)$ by all $X(t)$ and $Y(t)$ for a given initial state vector $\psi $. The nondemolition principle simply means that the induced current quantum state of the object coincides with the *a priori* one, as a statistical mixture of a posteriori states with respect to the past, but not the future, observations \[30\]. The a posteriori state as a quantum state of the object after the measurement, when a result has been read mathematically, will be defined in the next section. Here we only point out that the coincidence means that the induced state is not demolished by the measurement if the results have not been read. This justifies the use of the word nondemolition in the generalized sense.
One can call this coincidence the generalized reduction principle because it does not restrict the consideration to the projection valued operations only, corresponding to the von Neumann reduction of the quantum states, which is not applicable even for the relatively simple case of instantaneous measurements of the quantum observables with the continuous spectrum.
The equivalence of these two formulations in the quantum Markovian case and their relation to the projection postulate (see the next section) can be illustrated even in the case of the single operation corresponding to an instantaneous measurement, or a measurement with fixed duration.
Let $\mathcal{H}$ and $\mathcal{F}$ be the Hilbert spaces of state vectors $%
\eta \in \mathcal{H}$, and $\varphi \in \mathcal{F}$ for the quantum object and the measurement apparatus, respectively, and let $R$ be a self-adjoint operator in $\mathcal{H}$, representing a dynamical variable with the spectral values $x\in \mathbb{R}$ to be measured by means of the measurement apparatus with a given observable $\hat{y}$, representing the pointer of the apparatus as a self-adjoint operator in $\mathcal{F}$ with either discrete or continuous spectrum $\Lambda \subseteq \mathbb{R}$. The measurement apparatus has the fixed initial state $\varphi _{0}\in \mathcal{F}$, $\Vert
\varphi _{0}\Vert =1$ and is coupled to the object by an interaction operator $S^{\dagger }=V_{0}U^{\dagger }V_{1}$, where $U$ is a unitary evolution operator of the system in the product space $\mathcal{G}=\mathcal{H%
}\otimes \mathcal{F}$, $U^{\dagger }=U^{-1}$, and $V_{0}=V\otimes \hat{1}$, $%
V_{1}=I\otimes \hat{v}$ are the unitarities given by the free evolution operators $V:\mathcal{H}\rightarrow \mathcal{H}$, $\hat{v}:\mathcal{F}%
\rightarrow \mathcal{F}$ of the object and the apparatus, respectively, during the fixed measurement interval $[0,t]$. It is natural to suppose that the interaction does not disturb the variable $R$ in the sense $%
R_{0}:=R\otimes \hat{1}=S^{\dagger }R_{0}S$, or equivalently, $\langle x|S=%
\hat{s}_{x}\langle x|$, i.e., $$S:|x\rangle \otimes \varphi _{0}\mapsto |x\rangle \otimes \varphi _{x}\
,\quad \forall x\in \mathbb{R} \label{eq:1.2}$$in terms of (generalized) eigenvectors $|x\rangle $ of $R$, where $\varphi
_{x}=\hat{s}_{x}\varphi _{0}$. But it must disturb the input observable $%
\hat{q}=\hat{v}^{\dagger }\hat{y}\hat{v}$ in order to get the distinguishable probability densities $f_{x}(y)=|\varphi _{x}(y)|^{2}$ of the output observable $Y=S^{\dagger }(\kappa I\otimes \hat{q})S$, corresponding to the different eigen values $x\in \mathbb{R}$ of the input states $|x\rangle $ to be tested by the usual methods of mathematical statistics. Here $\kappa >0$ is a scaling parameter and we have assumed, for simplicity that the observable $\hat{y}$ and hence $\hat{q}$ has the nondegenerate spectral values $y\in \Lambda $, so that $\varphi \in \mathcal{%
F}$ in the input representation is described by the (generalized) eigenvectors $|y\rangle $ of $\hat{q}:|y\rangle \mapsto y|y\rangle $ as a square integrable function $\varphi (y)=\langle y|\varphi $, $\Vert \varphi
\Vert ^{2}=\int |\varphi (y)|^{2}\mathrm{d}\nu <\infty $ with respect to a given measure $\nu $ on $\Lambda $.
The positive measure $\nu $ is either discrete or continuous or can even be of mixed type normalizing the probability densities $g(y)=\langle \psi
(y),\psi (y)\rangle $ for the state vectors $\psi \in \mathcal{G}$: $$\Vert \psi \Vert ^{2}=\int_{\Lambda }\langle \psi (y),\ \psi (y)\rangle
\mathrm{d}\nu =\int_{\Lambda }g(y)\mathrm{d}\nu =1 \label{eq:1.4}$$where $\psi (y)=\langle y|\psi $ are the $\mathcal{H}$-valued wavefunctions of the system quantum object plus measurement apparatus. One can consider, for example, the standard Lebesgue measures $\mathrm{d}\nu =\mathrm{d}\lambda $ on $\Lambda =\mathbf{Z}
$, $\mathrm{d}\lambda =1$ and on $\Lambda =\mathbb{R}$, $\mathrm{d}\lambda =%
\mathrm{d}y$: $$\Vert \psi \Vert ^{2}=\sum \langle \psi (k),\psi (k)\rangle \;(\mathrm{d}%
\lambda =1)\ ;\quad \Vert \psi \Vert ^{2}=\int \langle \psi (y),\psi
(y)\rangle \mathrm{d}y\;(\mathrm{d}\lambda =\mathrm{d}y)$$respectively for the discrete spectrum $y\in \mathbf{Z}$ and for the continuous one $y\in \mathbb{R}$, given by the distributions $f(y)=\sum
\delta (y-k)$ and $f(y)=1$ as $\mathrm{d}\lambda =f(y)\mathrm{d}y$.
The output state vectors $\chi =S(\xi \otimes \varphi _{0})\in \mathcal{G}$, corresponding to the arbitrary input ones $\xi =V\eta $, $\langle \xi ,\xi
\rangle =1$, are given by the vector-functions $\chi :y\mapsto \chi (y)\in
\mathcal{H}$ of $y\in \Lambda $ with values$$\chi (y)=\langle y|S(\xi \otimes \varphi _{0})=\langle y|\chi .$$The operators $\langle y|S:\mathcal{G}\rightarrow \mathcal{H}$ correspond to the adjoint ones $S^{\dagger }|y\rangle :\eta \mapsto S^{\dagger }(\eta
\otimes |y\rangle )$, $$\langle \eta ,\langle y|S(\xi \otimes \varphi )\rangle =\langle S^{\dagger
}(\eta \otimes |y\rangle ),\xi \otimes \varphi \rangle \label{eq:1.3}$$defining the (generalized) vector-functions $S^{\dagger }|y\rangle \eta $ by $$\int S^{\dagger }|y\rangle \eta \varphi _{0}(y)\mathrm{d}\nu =S^{\dagger
}(\eta \otimes \varphi _{0})\;\;\;\;\forall \eta ,\varphi .$$The operator $(R_{0}\chi )(y)=R\chi (y)$ commutes with $Q=\kappa I\otimes
\hat{q}$ as well as with any other operator $C_{0}=C\otimes \hat{1}$ representing an object variable $C:\mathcal{H}\rightarrow \mathcal{H}$ in $%
\mathcal{H}\otimes \mathcal{F}$ as the constant function $Z(y)=C$. This is because the general operator $Z$ in $\mathcal{H}\otimes \mathcal{F}$ commuting with $Q$ corresponds to an operator–valued function $Z(y):%
\mathcal{H}\rightarrow \mathcal{H}$, which is defined by the operator $Z$ as $$\langle y|Z\psi =Z(y)\langle y|\psi \ ,\quad \forall \psi \in \mathcal{G}\
,\quad y\in \Lambda \label{eq:1.5}$$in the case $Z=Q$ it corresponds to $Z(y)=\kappa yI$: $\ \langle y|Q\psi
=\kappa y\langle y|\psi $. It is trivial in this case that the Heisenberg operators $X=S^{\dagger }ZS$ satisfy the nondemolition condition $[X,Y]=0$ with respect to the output observable $Y=S^{\dagger }QS$, but not the initial operators $Z:[Z,Y]\not=0$ if $[Z(y),R]\not=0$. This makes it possible to condition, by the observation of $Y$, the future measurements of any dynamical variable of the quantum object, but not the potential measurements of $Z$ in the past with respect to the present observation of $Y
$ if they have not been done initially.
Indeed, let $P_{\Delta }=S^{\dagger }I_{\Delta }S$ be the spectral orthoprojector of $Y$, given for a measurable $\Delta \subseteq \Lambda $ by $I_{\Delta }=I\otimes \hat{1}_{\Delta }$ as $$\langle y|I_{\Delta }\chi =1_{\Delta }(y)\chi (y)=1_{\Delta }(y)\langle
y|\chi \ ,\quad 1_{\Delta }(y)=\{%
\begin{array}{cc}
1 & y\in \Delta \\
0 & y\neq \Delta%
\end{array}
\label{eq:1.6}$$and $p_{\Delta }=\langle \eta \otimes \varphi ,P_{\Delta }(\eta \otimes
\varphi )\rangle \not=0$. Then the formula $$\varepsilon _{\Delta }[X]=\langle \eta ,\omega \lbrack XP_{\Delta }]\eta
\rangle /\langle \eta ,\omega \lbrack P_{\Delta }]\eta \rangle \,
\label{eq:1.7}$$where $\langle \eta ,\omega \lbrack X]\eta \rangle =\langle \eta \otimes
\varphi ,X(\eta \otimes \varphi )\rangle $, $\forall \eta \in \mathcal{H}$, defines the conditional expectation of $X=S^{\dagger }ZS$ with respect to $Y$. It gives the conditional probability $\varepsilon _{\Delta }[X]\in \lbrack
0,1]$ for any orthoprojector $X=O$, while $\varepsilon _{\Delta }[Z]$ defined by the same formula for $Z=\{Z(y)\}$ may not be the conditional expectation due to the lack of positivity $\omega \lbrack EP_{\Delta }]\geq
0 $, for all $\varphi \in \mathcal{F}$ if the orthoprojector$Z=E$ does not commute with $P_{\Delta }$. The necessity of the nondemolition principle for the existence of the conditional probabilities is the consequence of the main filtering theorem consistent with the causality principle according to which the conditioning with respect to the current observation has the sense of preparation for future measurements but not for past ones.
This theorem proved in the general algebraic form in Ref. \[30\] reads in the simplest formulation as
<span style="font-variant:small-caps;">Main Measurement Theorem.</span> Let $O$ be an orthoprojector in $\mathcal{G}=\mathcal{H}\otimes \mathcal{F}$. Then for each state vector $%
\psi =\xi \otimes \varphi $ there exists the conditional probability $%
\varepsilon _{\Delta }[O]\in \lbrack 0,1]$, defined by the compatibility condition $$\varepsilon _{\Delta }[O]\langle \xi \otimes \varphi ,P_{\Delta }(\xi
\otimes \varphi )\rangle =\langle \xi \otimes \varphi ,\ OP_{\Delta }(\xi
\otimes \varphi )\rangle \label{eq:1.8}$$if and only if $OP_{\Delta }=P_{\Delta }O$. It is uniquely defined for any measurable $\Delta \subset \Lambda $ with respect to $P_{\Delta }=S^{\dagger
}I_{\Delta }S$, $\varphi =\varphi _{0}$ as $$\varepsilon _{\Delta }[O]={\frac{1}{\mu _{\Delta }}}\int_{\Delta }\langle
\chi _{y},E(y)\chi _{y}\rangle \mathrm{d}\mu \label{eq:1.9}$$Here $E(y):\mathcal{H}\rightarrow \mathcal{H}$ is the orthoprojector valued function, describing $O$, commuting with all $P_{\Delta }$ in the Schrödinger picture as $O=S^{\dagger }ES$, $\mu _{\Delta }=\int_{\Delta }g_{\xi
}(y)\mathrm{d}\nu $ is the absolutely continuous with respect to $\nu $ probability distribution of $y\in \Lambda $, $g_{\xi }(y)=\Vert \chi
(y)\Vert ^{2}$, $\chi (y)=\langle y|S(\xi \otimes \varphi _{0})$, and $%
y\mapsto \chi _{y}$ is the random state vector $\chi _{y}\in \mathcal{H}$ of the object uniquely defined for almost all $y:g_{\xi }(y)\not=0$ up to the random phase $\theta (y)=\mathrm{arg}c_{\xi }(y)$ by the normalization $$\chi _{y}=\chi (y)/c_{\xi }(y)\ ,\quad |c_{\xi }(y)|^{2}=g_{\xi }(y)
\label{eq:1.10}$$
The generalized *a posteriori* reduction
========================================
It follows immediately from the main theorem that the input state vector $%
\xi :\Vert \xi \Vert =1$ of the object of measurement has to be changed for $%
\chi _{y}\in \mathcal{H}$ due to the preparation $\xi \mapsto \{\chi
(y):y\in \Lambda \}$ of the *a priori* state vector $\chi =S(\eta
\otimes \varphi _{0})$ of the meter and the object after the objectification $\hat{q}=y$. The former is given by the dynamical interaction in the pointer representation $\chi (y)=\langle y\mid \chi $ due to the choice of the measurement apparatus and the output observables, and the latter is caused by statistical filtering $\chi \mapsto \chi (y)$ due to the registration of the measurement result $y\in \Lambda $ and the normalization $\chi _{y}=\chi
(y)/\Vert \chi (y)\Vert $.
While the process of preparation described by a unitary operator applied to a fixed initial state of the meter encounters no objection among physicists, the process of objectification encounters objection because of the nonunitarity of the filtering and nonlinearity of the normalization. But the main theorem shows clearly that there is nothing mysterious in the objectification. It is not a physical process but only a mathematical operation to evaluate the *conditional state* $$\pi _{y}[Z]=\varepsilon _{y}[S^{\dagger }ZS]=\langle \chi _{y},Z(y){\chi _{y}%
}\rangle \label{eq:2.1}$$which are defined by the conditional expectations $\varepsilon
_{y}[X]=\lim_{\Delta \downarrow y}\varepsilon _{\Delta }[X]$ of the Heisenberg operators $X$ for $Z=\{Z(y)\}$. The linear random operator $$G(y):\xi \in {\mathcal{H}}\mapsto \langle y|S(\xi \otimes \varphi _{0})\
,\quad y\in \Lambda \label{eq:2.2}$$defines the reduction transformations $G(y)$ as the partial matrix elements $%
\langle y|S\varphi _{0}$ of the unitary operator $S$. They map the normalized vectors $\xi \in \mathcal{H}$ into the *a posteriori* ones $\chi (y)=G(y)\xi $, renormalized to the probability density $$g_{\xi }(y)=\Vert G(y)\xi \Vert ^{2}=\langle \xi ,E(y)\xi \rangle ,\quad
E=G^{\dagger }G\ .$$If the condition (\[eq:1.2\]) holds, then the only eigen vectors $%
|x\rangle $ of $R$ remain unchanged up to a phase multiplier: $$G(y)|x\rangle =|x\rangle \varphi _{x}(y),\ \varphi _{x}(y)=\langle y|\hat{s}%
_{x}\varphi _{0}=\langle y|\varphi _{x} \label{eq:2.3}$$and hence $\chi _{y}=e^{\mathrm{i}\theta _{x}(y)}|x\rangle $, where $\theta
_{x}(y)=\arg \,\varphi _{x}(y)$. The superpositions $\xi =\int |x\rangle \xi
(x)\mathrm{d}\lambda $ change their amplitudes $\xi (x)=\langle x|\xi $ for $%
\chi _{y}(x)=\langle x|\chi _{y}$ $$\langle x|\chi _{y}=c_{\xi }^{-1}(y)\chi (x,y)\ ,\quad \chi (x,y)=\langle
x|G(y)\xi =\chi _{x}(y)\xi (x) \label{eq:2.4}$$where $c_{\xi }(y)=(\int |\varphi _{x}(y)|^{2}h(x)\mathrm{d}\lambda )^{1/2}$, $h(x)=|\xi (x)|^{2}$.
In the case of a purely continuous spectrum of $R$ there are no invariant state vectors at all because the generalized eigenvectors cannot be considered as input ones due to $|x\rangle \notin \mathcal{H}$ as $\langle
x|x\rangle=\infty$ in that case.
The generalized reduction (\[eq:2.1\]) of the state-vector corresponds to the limit case $\Delta\downarrow y$ when the accuracy of the instrument $%
\Delta\ni y$ tends to the single-point subset $\{y\}\subset\Lambda$. It is not even the mathematical idealization of the real physical experiment if the observable $\hat q$ has the discrete spectrum $\Lambda=\{y_i\}$.
Prior to discussing why the generalized reduction does not contradict the main postulates of the quantum theory, let us show how to derive the von Neumann projection postulate in this way, corresponding to the orthogonal transformations $G(y_{i})=F_{i}$ given by a partition $\sum A_{i}=\mathbb{R}$ of the spectrum of $R$ as $F_{i}=E_{A_{i}}$. Here $A\mapsto E_{A}$, $%
E_{A}^{\dagger }E_{A^{\prime }}=E_{A\cap A^{\prime }}$, $\sum E_{A_{i}}=I$ is the spectral measure of $R=\int x\mathrm{d}E$ which might be either of discrete or of continuous type as in the cases $$E_{A}=\sum_{x\in A}|x\rangle \langle x|\ ,\quad E_{A}=\int_{A}|x\rangle
\langle x|\mathrm{d}x\ ,$$corresponding to the nondegenerate spectrum of $R:\mathrm{d}E=|x\rangle
\langle x|\mathrm{d}\lambda $.
Considering the indices $i$ of $y_{i}$ in $\mathbf{Z}$ it is always possible to find the unitary interaction in the Hilbert space ${\mathcal{G}=\mathcal{H%
}}\otimes l^{2}(\mathbf{Z})$ of the two–sided sequences $\psi =\{\eta
^{k}|k=0,\pm 1,\pm 2,\dots \}$ with $\eta ^{k}\in \mathcal{H}$ such that $%
\Vert \psi \Vert ^{2}=\sum_{-\infty }^{\infty }\langle \eta ^{k},\eta
^{k}\rangle <\infty $. Indeed, we can define the interaction as the block-matrix $S^{\dagger }=[W_{k}^{i}]$ acting in $\mathcal{G}$ as $%
W^{i}\psi =\sum_{k=-\infty }^{\infty }W_{k}^{i}\eta ^{k}$, by $%
W_{k}^{i}=F_{k-i}$, where $F_{k}=0$ if there is no point $y_{k}$ in $\Lambda
$ numbered by a $k\in \mathbf{Z}$. It is the unitary one because $S=[F_{i-k}]
$ is inverse to $S^{\dagger }=[F_{k-i}]$ as $$\sum_{j=-\infty }^{\infty }F_{i-j}F_{k-j}=\delta _{k-j}^{i-j}\sum_{j=-\infty
}^{\infty }F_{-j}=\delta _{k}^{i}\sum F_{i}=\delta _{i}^{k}I$$due to the orthogonality $F_{i}F_{k}=0$, $i\not=k$, and completeness $\sum
F_{i}=I$ of $\{F_{i}\}$.
Let us fix the initial sequence $\varphi _{0}\in l^{2}(\mathbf{Z})$ as the eigenstate $\varphi _{0}=\{\delta _{0}^{k}\}=|0\rangle $ of the input observable $\hat{k}$ in $l^{2}(\mathbf{Z})$ as the counting operator $$\hat{k}=\sum_{k=-\infty }^{\infty }k|k\rangle \langle k|\ ,\quad |i\rangle
=\{\delta _{i}^{k}\}\in l^{2}(\mathbf{Z}) \label{eq:2.5}$$with the spectrum $\mathbf{Z}$. Then we obtain the conditional states ([eq:2.1]{}) defined as $$\pi _{i}[Z]={\frac{1}{p_{i}}}\langle F_{i}\eta ,Z_{i}F_{i}\eta \rangle
=\langle \eta _{i},Z_{i}\eta _{i}\rangle ,\;\eta _{i}=F_{i}\eta /p_{i}^{1/2}$$up to the normalizations $p_{i}=\langle F_{i}\eta ,F_{i}\eta \rangle \not=0$ by the linear operations $\sigma \mapsto W_{i}^{0}\sigma W_{i}^{0}$, $$W_{i}^{0}\eta =\langle i|S(\varphi _{0}\otimes \eta )=\sum_{k=-\infty
}^{\infty }F_{i-k}\delta _{0}^{k}\eta =F_{i}\eta \ . \label{eq:2.6}$$It is only in that case that the *a posteriori* state always remains unchanged under the repetitions of the measurement. Such an interaction satisfies the condition (\[eq:1.2\]) with $\varphi _{x}=\hat{s}_{x}\varphi
_{0}$ given by the sequences $\varphi _{x}=\{\delta
_{i(x)}^{k}\}=|i(x)\rangle $ because $$F_{i-k}|x\rangle =|x\rangle \delta _{i-k}^{i(x)}=W_{i}^{k}|x\rangle \quad
(=|x\rangle \,,\quad \forall x\in A_{i-k})\,,$$where $i(x)=i$ if $x\in A_{i}$ is the index map of the coarse-graining $%
\{A_{i}\}$ of the spectrum of $R$. Hence in the $x$-representation $\psi
=\int |x\rangle \psi (x)\mathrm{d}\lambda $, $\psi (x)=\langle x|\psi $ it can be described by the shifts $\hat{s}_{x}^{\dagger }=[\delta _{k-i}^{i(x)}]
$ in $l^{2}(\mathbf{Z})$ $$\hat{s}_{x}^{\dagger }:\psi (x)=\{\eta ^{k}(x)\}\mapsto \{\langle x|\eta
^{i(x)+k}\}\,\quad \eta ^{k}(x)=\langle x|\eta ^{k} \label{eq:2.7}$$replacing the initial state $\varphi _{0}=|0\rangle $ of the meter for each $%
x$ by another eigenstate $|i(x)\rangle =\hat{s}_{x}|0\rangle $ if $x\notin
A_{0}$.
This realizes the coarse-grained measurement of $R$ by means of the nondemolition observation of the output $$Y=S^{\dagger }(I\otimes \hat{k})S=i(R)\otimes \hat{1}+I\otimes \hat{k}\,,
\label{eq:2.8}$$where $i(R)=\int i(x)\mathrm{d}E=\sum iF_{i}$. If $q(R)=\hbar i(R)$ is the quantized operator $R$ given, say, by the integer $i(x)=\lfloor x/\hbar
\rfloor $, then the rescaled model $\hat{y}_{x}=\hbar \hat{s}_{x}^{\dagger }%
\hat{k}\hat{s}_{x}=q(x)\hat{1}+\hbar \hat{k}$ of the nondemolition measurement has the classical limit $\lim \hat{y}_{x}=x\hat{1}$ if $\hbar
\rightarrow 0$, corresponding to the direct observation of a continuous variable $R$ by means of $\lim \hbar Y=R\otimes \hat{1}$.
Note that the observable $Y$ commutes with the arbitrary Heisenberg operator $A=S^{\dagger }(C\otimes \hat{1})S$ of the object, but not with the initial operators $C_{0}=C\otimes \hat{1}$ if $[C,i(R)]\not=0$.
The unitary operator $S^{\dagger }$ is given by the interaction potential $%
q(R)\otimes \hat{p}$ as $S^{\dagger }=\exp \{(\mathrm{i}/\hbar )q(R)\otimes
\hat{p}\}$, where $\mathrm{i}=\sqrt{-1}$, and $\hat{p}=[\langle i|\hat{p}%
|k\rangle ]$, $\langle i|\hat{p}|k\rangle =(1/2\pi )\int_{-\pi }^{\pi }pe^{-%
\mathrm{i}(i-k)p}\mathrm{d}p$ is the matrix of the momentum operator in $%
l^{2}(\mathbf{Z})$, generating the shifts $\hat{s}_{x}^{\dagger }=[\langle i|%
\hat{s}_{x}^{\dagger }|k\rangle ]$ as $\hat{s}_{x}^{\dagger }=e^{i(x)\mathrm{%
i}\hat{p}} $: $$\langle i|\hat{s}_{x}^{\dagger }|k\rangle ={\frac{1}{2\pi }}\int_{-\pi
}^{\pi }e^{i(x)\mathrm{i}p}e^{-\mathrm{i}(i-k)p}\mathrm{d}p=\delta
_{i-k}^{i(x)}\,.$$The nondemolition observation reproduces the statistics of the demolition measurement of $R$ by the direct observation of $q(R)$ because the output observable $Y$ has the same characteristic function with respect to the state vector $\xi \otimes
\varphi _{0}$ as $i(R)$ with respect to $\xi $: $$\begin{aligned}
&\langle \xi \otimes \varphi _{0},\exp \{\mathrm{i}pY\}(\xi \otimes \varphi
_{0})\rangle =&\langle S(\xi \otimes \varphi _{0}),e^{\mathrm{i}pQ}S(\xi
\otimes \varphi _{0})\rangle \\
&&\qquad =\sum \langle F_{i}\xi ,e^{i\mathrm{i}p}F_{i}\xi \rangle =\langle
\xi ,\exp \{\mathrm{i}pi(R)\}\xi \rangle \,.\end{aligned}$$Here $p$ is the parameter of the characteristic function, $Q=I\otimes \hat{k}
$, and $F_{i}=\langle i|S\varphi _{0}=F_{i}^{\dagger }$ are the orthoprojectors, such that $\sum_{i}F_{i}^{\dagger }F_{i}=\int i(x)\mathrm{d}%
E=i(R)$. If the observable $R$ is discrete, then the nondemolition observation (\[eq:2.8\]) realizes the precise measurement of $R$, if the partition $\{A_{i}\}$ separates all the eigenvalues $\{x_{i}\}$ as in the case $x_{i}\in A_{i}$, $\forall i$, corresponding to $x_{i}=\hbar i$, $%
A_{i}=[\hbar i,\hbar (i+1)[$, $i=0,1,2,\ldots $.
The nondemolition principle helps not only to derive the projection postulate as a reduced description of the shift interaction in the enlarged Hilbert space $\mathcal{G}$ with respect to the initial eigenvector $\varphi
_{0}=|0\rangle $ of the discrete meter $\hat{q}$, but also extends it to the generalized reductions under the unsharp measurements with arbitrary spectrum $\Lambda $, corresponding to the nonrepeatable instruments \[1,2\] $$\Pi _{\Delta }[C]=\int_{\Delta }\Psi \lbrack C](y)\mathrm{d}\nu \,,\quad
\Psi \lbrack C](y)=G(y)^{\dagger }CG(y)\,. \label{eq:2.9}$$The density $\Psi (y)$ of the instrument defines completely positive but not necessarily orthoprojective operations $E(y)=\Psi \lbrack I](y)$, called the effects for the probability densities $g(y)=\sigma \lbrack E(y)]$, and also the nonlinear operation $\sigma \mapsto \sigma \circ \Psi (y)/\sigma \lbrack
E(y)]$ of the generalized reduction, mapping the pure input states $\sigma
_{\xi }[C]=\langle \xi |C|\xi \rangle $ into the *a posteriori* ones $$\rho _{y}[C]={\frac{1}{g_{\xi }(y)}}\rho \lbrack C](y)=\pi _{y}[C_{0}]\
,\quad \rho \lbrack C](y)=\langle \chi (y),\ C\chi (y)\rangle \,.
\label{eq:2.10}$$They are also pure because of the completeness of the nondemolition measurement, i.e., nondegeneracy of the spectrum of the observable $\hat{q}$ in $\mathcal{F}$. Thus, the reduction of the state-vector is simply the way of representing in the form (\[eq:2.1\]) the *a posteriori* pure states (\[eq:2.10\]) given at the limit $\Delta \rightarrow 0$ by the usual (in the statistics) Bayesian formula (\[eq:1.7\]) for $X=S^{\dagger
}C_{0}S=A$, which is applicable due to the commutativity of $A$ and $%
P_{\Delta }$.
The reduction $\sigma _{1}\rightarrow \rho _{y}$ of the prepared state $%
\sigma _{1}=\sigma \circ \Psi $ for the object measurement is given as the evaluation of the conditional expectations which are the standard attributes of any statistical theory. All the attempts to derive the reduction as a result of deterministic interaction only are essentially the doomed attempts to derive the probabilistic interpretation of quantum theory. There is no physical explanation of the stochasticity of the measurement process as there is no adequate explanation of the randomness of an observable in a pure quantum state.
It is not a dynamical but a purely statistical effect because the input and output state-vectors of this process are not the observables of the individual object of the statistical ensemble but only the means for calculating the *a priori* and the *a posteriori* probabilities of the observables of this object. Hence there is no observation involving just a single quantum object which can confirm the reduction of its state. The reduction of the state-vector can be treated as an observable process only for an infinite ensemble of similar object plus meter systems. But the measurements for the corresponding collective observables also involves preparation and objectification procedures, this time for the ensemble, i.e., for a second quantized compound system. So the desirable treatment of all the reductions as some objective stochastic process can never be reached in this way. They are secondary stochastic since they are dependent on the random information that has been gained up to a given time instant $t$.
The reduction of the state-vector is not at variance with the coherent superposition principle, because a vector $\eta \in \mathcal{H}$ is not yet a pure quantum state but defines it rather up to a constant $c\in \mathbb{C}$ as the one-dimensional subspace $\{c\eta |c\in \mathbb{C}\}\subset \mathcal{H%
}$ which is a point of the projective space over $\mathcal{H}$. For every reduced state-vector $\chi _{y}$ there exists an equivalent one, namely $%
\chi (y)=\sqrt{g_{\xi }(y)}\chi _{y}$, defining the same quantum pure state, given by the linear transformation $G(y):\xi \mapsto \chi (y)$, so that the superposition principle holds: $\chi (y)=\sum c_{i}\chi ^{i}(y)$ if $\xi
=\sum c_{i}\xi ^{i}$. The pure state transformation $G(y)$ does not need to be unitary, but as an operator $G:\mathcal{H}\rightarrow \mathcal{G}$ with $$G^{\dagger }G=\int G(y)^{\dagger }G(y)\mathrm{d}\nu =\int \varphi
_{0}^{\dagger }S^{\dagger }|y\rangle \langle y|S\varphi _{0}\mathrm{d}\nu
=\varphi _{0}^{\dagger }S^{\dagger }S\varphi _{0}=I$$it preserves the total probability by mapping the normalized $\xi \in
\mathcal{H}$ into the $\chi (y)=G(y)\xi $, normalized to the probability density $g_{\xi }(y)$.
According to the nondemolition principle it makes sense to apply the vector $%
\chi=\{\chi(y)\}$ of the system after the measurement preparation only against the reduced observables $Z=\{Z(y)\}$ which commute with $Q=\kappa
I\otimes \hat q$. Otherwise according to the main theorem the conditional probabilities of the future observations may not exist for an initial state-vector $\chi_0=\eta\otimes\varphi$ and a given result $y\in\Lambda$ of the measurement. It is against the physical causality to consider the unreduced operators as the observables for the future measurements since the causality means that the future observations must be statistically predictable from the data of a measurement and such prediction can be given only by the conditional probabilities (\[eq:1.9\]). Once the output observables are selected as a part of a preparation, the algebra of the actual observables is reduced and there is no way to measure an observable $%
Z $ which is not compatible with $Q$. It could be measured in the past if another preparation had been made but the irreversibility of the time arrow does not give this possibility. Thus, the quantum measurement theory implies a kind of time-dependent superselection rule for algebras such as those of the observables $Z$ chosen as the actual observable at the moment $t$. But it does not prevent one from considering other operators as the virtual observables defining super operators, i.e., the subsidiary operators for the description of some meaningful operations, although an evaluation of their expectations does not make any sense as it does for the differential operators in the classical theory.
The *a priori* states are the induced ones $$\sigma _{1}(C)=\int \langle \chi _{y},C\chi _{y}\rangle \mathrm{d}\mu
=\langle \chi ,C_{0}\chi \rangle \ ,\quad C_{0}=C\otimes \hat{1}$$on the algebra generated by the operators in $\mathcal{H}$ of the object only. They are given as the statistical mixtures of the *a posteriori* pure states (\[eq:2.10\]) of the object even if the initial state $\sigma $ was pure. But it does not contradict quantum mechanics because the prepared state $\phi (Z)=\langle \chi ,Z\chi \rangle $ of the quantum system after the measurement is reduced to the object plus pointer but is still given uniquely by the state-vector $\chi \in \mathcal{G}$, up to a random phase. Namely, the vector $\chi $ is a coherent superposition $$\chi =\sum \chi _{i}\otimes |y_{i}\rangle c_{i}\ ,\quad \chi _{i}=\chi
(y_{i})/c_{i}\ ,\quad |c_{i}|^{2}=p_{i}$$of the *a posteriori* states $\chi _{i}\otimes |y_{i}\rangle
$ of the system, if $\hat{q}$ has the spectral decomposition $\hat{q}=\sum
y_{i}|y_{i}\rangle \langle y_{i}|$ and $p_{i}$ are the probabilities of $%
y_{i}$.
This uniqueness does not hold for the density-matrix representations $\phi
\lbrack Z]=\mathrm{Tr}\{\hat{\phi}Z\}$; among the equivalent density matrices $\hat{\phi}\geq 0$ there exists always the projector $\hat{\phi}%
=|\chi \rangle \langle \chi |$, but there are also mixtures such as the diagonal one $$\hat{\phi}_{1}=\sum p_{i}|\eta _{i}\rangle \langle \eta _{i}|\ ,\quad |\eta
_{i}\rangle =\eta _{i}\otimes |y_{i}\rangle$$in the discrete case $\Lambda =\{y_{i}\}$. Hence, the diagonalization $\hat{%
\phi}\mapsto \hat{\phi}_{1}$ of the density matrix due to the measurement of $\hat{q}$ is only the rule to choose the most mixed one $\hat{\phi}_{1}$ which is equivalent to the coherent choice $\hat{\phi}$ due to $$\mathrm{Tr}\{\hat{\phi}Z\}=\sum p_{i}\langle \eta _{i},Z_{i}\eta _{i}\rangle
=\mathrm{Tr}\{\hat{\phi}_{1}Z\}$$for all reduced operators $Z=\sum Z_{i}\otimes |y_{i}\rangle \langle y_{i}|$. There is no special need to fix such a choice, which is even impossible in the continuous spectrum case. This is because the continuous observable $%
\hat{q}$ has no ordinary eigenvectors, $\langle y|y\rangle =\infty $ and hence $\chi _{y}\otimes |y\rangle \notin \mathcal{G}$, but there exist the eigenstates $\omega _{y}[\hat{z}]=z(y)$ on the algebra of complex functions $%
z(y)$, defining the conditional expectations $\varepsilon _{y}[X]$ for $%
X=S^{\dagger }ZS$ as $$\varepsilon _{y}[X]=\pi _{y}[SXS^{\dagger }]\ ,\quad \pi _{y}=\rho
_{y}\otimes \omega _{y}\ ,\quad \forall y\in \Lambda \,.$$Thus, the nondemolition principle abandons the collapse problem, reducing it to the evaluation of the *a posteriori* state. The decrease of the observable algebra is the only reason for the irreversibility of the linear transformation $\phi _{0}\mapsto \phi $ of the initial states $\phi
_{0}(X)=\langle \chi _{0},X\chi _{0}\rangle $, which are pure on the algebra of all operators $X$ into the prepared (mixed) ones on the algebra of the reduced operators $Z$.
The main measurement problem
============================
As was shown using an instantaneous measurement as an example, the nondemolition principle leads to the notion of the instrument, described by the operational-valued measure (\[eq:2.9\]), and gives rise to the generalized reduction (\[eq:2.10\]) of the quantum statistical states. In the operational approach \[1,2\] one starts from the instrumental description $%
\sigma\mapsto\sigma\circ\Phi(y)=\rho(y)$ of the measurement, which is equivalent to postulating the generalized reduction (20) given up to the probabilistic normalization $g(y)=\rho[I](y)$ by the linear map $%
\sigma\mapsto\sigma\circ\Psi(y)$ due to $\Psi_y(\sigma)=(1/g(y))\sigma\circ%
\Psi(y)=\rho_ y$.
The main measurement problem is the reconstruction of an interaction representation of the quantum measurement, that is, finding a proper dilation $\mathcal{G}$ of the Hilbert space $\mathcal{H}$ and the output process $Y $, satisfying the nondemolition (and self-nondemolition) condition (\[eq:1.1\]) with respect to the Heisenberg operators $X$ of the object of measurement in order to derive the same reduction as the result of conditional expectation.
The minimal dilation giving, in principle, the solution of this problem even for non-Markovian relativistic cases was constructed in \[35\], but it is worth finding also more realistic, nonminimal dilations defining the object of measurement as a quantum stochastic process in the strong sense for the particular Markovian cases.
In the case of a single instantaneous measurement described by an instrument $\Pi _{\Delta }$, this can be formulated as the problem of finding the unitary dilation $U\varphi _{0}:\eta \in \mathcal{H}\mapsto U(\eta \otimes
\varphi _{0})$ in a tensor product $\mathcal{G}=\mathcal{H}\otimes \mathcal{F%
}$ and an observable $\hat{y}=\int y\mathrm{d}\hat{1}$ in $\mathcal{F}$, giving $\Pi _{\Delta }$ as the conditional expectation $$\Pi _{\Delta }[C]=\omega _{0}[AE_{\Delta }]\ ,\quad \langle \eta ,\omega
_{0}[X]\eta \rangle =\langle \eta \otimes \varphi _{0},X\eta \otimes \varphi
_{0}\rangle$$of $AE_{\Delta }=U^{\dagger }(C\otimes \hat{1}_{\Delta })U$. In principle, such a quadruple $(\mathcal{F},\varphi _{0},\hat{y},U)$ was constructed in \[36\] and \[37\] for the normal completely positive $\Pi _{\Delta }$, giving a justification of the general reduction postulate as described above for the case of the projective $\Pi _{\Delta }$. For the continuous observation this problem was solved \[39\] on the infinitesimal level in terms of the quantum stochastic unitary dilation of a differential evolution equation for characteristic operations $$\tilde{\Psi}(t,q)=\int e^{\mathrm{i}qy}\Psi (t,y)\mathrm{d}\nu \ ,\quad \Psi
(t,y)=\lim_{\Delta \downarrow y}{\frac{1}{\nu _{\Delta }}}\Pi _{\Delta
}^{t}\ ,$$where $\mathrm{d}\pi $ is a standard probability measure of $\mathrm{d}%
y\subset \Lambda $. This corresponds to the stationary Markovian evolution of the convolutional instrumental semigroups $\{\Pi _{\Delta }^{t}|t\geq 0\}$ giving the reduced description of the continuous measurement, with the data $%
y(t)$ having the values in an additive group.
Unfortunately the characteristic operational description of the quantum measurement is not relevant to the sample-paths representation. It is not suitable for the conditioning of the quantum evolution under the given data of the observations and hence does not allow one to obtain explicitly the corresponding dynamical reduction. Moreover, the continuous measurements have the data $y$ not necessary in a group, and in the nonstationary cases they cannot be described by the convolution instrumental semigroups and the corresponding evolution equations.
Recently a new differential description of continual nondemolition measurements was developed within the noncommutative stochastic calculus method \[13,14,31\]. A general stochastic filtering equation was derived for the infinitesimal sample-paths representation of the quantum conditional expectations, giving the continuous generalized reduction of the *a posteriori* states \[25,26,29\].
Simultaneously, some particular cases of the filtering equation for the stochastic state-vector $\varphi (t,\omega )=\chi _{y^{t}}(\omega )$, corresponding to the functional spectrum $\Lambda ^{t}$ of the diffusion trajectories $y^{t}(\omega )=\{y(s,\omega )|s\leq t\}$, were discovered within the phenomenological theories of the dynamical reduction and spontaneous localization \[16–18\]. As was shown in \[21,27\] and \[29\], the nonlinearity of such equations is related only to the normalization $\Vert
\varphi (t,\omega )\Vert =1$ and after the proper renormalization $\chi
_{t}(\omega )=\sqrt{g_{t}(\omega )}\varphi (t,\omega )$, where $g_{t}(\omega
)$ is the probability density of the process $$y(s,\omega )={\frac{1}{s}}\int_{0}^{s}\langle \varphi (t,\omega ),R\varphi
(t,\omega )\rangle \mathrm{d}t+s^{-1}w_{s}\ ,\quad s\in \lbrack 0,t)$$generated by the standard Wiener process $\omega =\{w_{t}\}$ with respect to the Wiener probability measure $\mathrm{d}\pi $ on the continuous trajectories $\omega \in \Omega $, they become the linear ones $$\mathrm{d}\chi _{t}+\left( {\frac{\mathrm{i}}{\hbar }}H+{\frac{1}{2}}%
L^{\dagger }L\right) \chi _{t}\mathrm{d}t=L\chi _{t}\mathrm{d}w\ .
\label{eq:3.1}$$Here $H$ is the Hamiltonian of the object, $L$ is an arbitrary operator in $%
\mathcal{H}$ defining the variable $R=\sqrt{\hbar }(L+L^{\dagger })$, under the continuous measurement, and $\mathrm{d}w=w_{t+\mathrm{d}t}-w_{t}$ is the forward increment, such that the stochastic equation (\[eq:3.1\]) has to be solved in the Itô sense. This solution can be explicitly written as $$\chi _{t}(\omega )=T_{t}(\omega )\xi ,\quad T_{t}(\omega )=\exp
\{w_{t}L-tL^{2}\} \label{eq:3.2}$$in the case $L=\sqrt{\pi /2h}\,R$, $(h=2\pi \hbar )$, $H=0$, corresponding to the unsharp measurement of the self-adjoint operator $R$ during the time interval $[0,t)$ with the trivial free Hamiltonian evolution of the object. In the case $H\not=0$ this can be used for the approximate solution of ([eq:3.1]{}) with $L^{\dagger }=L$, $\chi (0)=\eta $ as $\chi _{t}(\omega
)\simeq T_{t}(\omega )\xi (t)$, where $\xi (t)=V(t)\eta $ is the unitary evolution $V(t)=\exp \left\{ -{\frac{\mathrm{i}}{\hbar }}Ht\right\} $ without the measurement.
The stochastic transformation (\[eq:3.2\]) defines the operational density $$\Theta _{t}[C](\omega )=T_{t}^{\dagger }(\omega )CT_{t}(\omega )$$of an instrument as in (\[eq:2.9\]) with respect to the standard Wiener probability measure $\mathrm{d}\pi $ on $\omega ^{t}=\{w_{s}\}_{s\leq t}\in
\Omega ^{t}$ having the Gaussian marginal distribution of $q_{t}=\sqrt{\hbar
}w_{t}$ $$\mathrm{d}\nu :=\int_{q_{t}\in \mathrm{d}q}\mathrm{d}\pi =(ht)^{-1/2}\exp
[-\pi q^{2}/ht\}\mathrm{d}q\,.$$Hence $\Psi (t,q)\mathrm{d}\nu :=\int\limits_{q_{t}\in \mathrm{d}q}\Theta
_{t}(\omega )\mathrm{d}\pi =\Phi (t,y)\mathrm{d}y$, where $y={\frac{1}{t}}q$, $$\Phi \lbrack C](t,y)=\sqrt{\frac{t}{h}}\exp \left\{ -{\frac{\pi t}{2h}}%
(y-R)^{2}\right\} C\exp \left\{ -{\frac{\pi t}{2h}}(y-R)^{2}\right\} \,,
\label{eq:3.3}$$because $\Theta _{t}(\omega )$ depends only on $w_{t}$: $\Theta _{t}(\omega
)=\Psi (t,\sqrt{\hbar }w_{t})$, and $$\Psi \lbrack C](t,q)=G(t,q)^{\dagger }CG(t,q)\,,\quad G(t,q)=\exp \left\{ -{%
\frac{\pi }{h}}\left( qR-{\frac{t}{2}}R^{2}\right) \right\} \,.$$The operator $E(t,y)=\Phi \lbrack I](t,y)=f_{R}(t,y)$, $$f_{R}(t,y)=\sqrt{\frac{t}{h}}\exp \left\{ -\pi {\frac{t}{h}}%
(y-R)^{2}\right\} =F(t,y)^{\dagger }F(t,y)$$defines the probability density of the unsharp measurement of $R$ with respect to the ordinary Lebesgue measure $\mathrm{d}y$ as the convolution $$g_{\xi }(t,y)=\int \sqrt{\frac{t}{h}}\exp \left\{ -\pi {\frac{t}{h}}%
(y-x)^{2}\right\} h_{\xi }(x)\mathrm{d}\lambda =(f_{0}\ast h_{\xi })(y)\,,$$where $h_{\xi }(x)=|\xi (x)|^{2}$, $\xi (x)=\langle x|\xi $, $\mathrm{d}%
\lambda =\sum \delta (x-x_{i})\mathrm{d}x$ in the case of discrete spectrum $%
\{x_{i}\}$ of $R$, and $\mathrm{d}\lambda =\mathrm{d}x$ in the case of purely continuous spectrum of $R$.
This means that the continuous unsharp measurement of $R$ can be described by the observation model $y_x(t)=x+(1/t)q_t$ of signal $x$ plus Gaussian error $e(t)=(1/t)q_t$ with independent increments as $$y_R(t)=R+e(t)I\,,\quad e(t)={\frac{\sqrt\hbar}{t}}w_t\,. \label{eq:3.4}$$ The noise $e(t)$ with the mean value $\langle e(t)\rangle=0$ gives a decreasing unsharpness $\langle e(t)^2\rangle=\hbar/t$ of the measurement from infinity to zero that is inversely proportional to the duration of the observation interval $t>0$.
In general, such a model can be realized \[21\]–\[25\] as the nondemolition observation within the quantum stochastic theory of unitary evolution of the compound system on the product ${\mathcal{G}=\mathcal{H}}\otimes \mathcal{F}$ with the Fock space $\mathcal{%
F}$ over the one-particle space $L^{2}(\mathbb{R}_{+})$ for a one-dimensional bosonic field, modeling the measurement apparatus of the continuous observation.
Let us illustrate this general construction for our particular case $H=0$, $%
L=L^{\dagger }$. The unitary interaction $S(t)$ in $\mathcal{G}$, defining the transformations (\[eq:3.2\]) as (\[eq:2.2\]) with respect to the vacuum state-vector $\varphi _{0}\in \mathcal{F}$, is generated by the field momenta operators $$\hat{p}_{s}={\frac{\mathrm{i}}{2}}\sqrt{\hbar }(\hat{a}_{s}^{\dagger }-\hat{a%
}_{s})\,,\quad s\in \mathbb{R}_{+} \label{eq:3.5}$$as $S(t)=\exp \left\{ -{\frac{\mathrm{i}}{\hbar }}R\otimes \hat{p}%
_{t}\right\} $.
Here $\hat{a}_{s}$ and $\hat{a}_{s}^{\dagger }$ are the canonical annihilation and creation operators in $\mathcal{F}$, localized on the intervals $[0,s]$ according to the commutation relations $$\lbrack \hat{a}_{r},\hat{a}_{s}]=0,\quad \lbrack \hat{a}_{r},\hat{a}%
_{s}^{\dagger }]=s\hat{1}\ ,\quad \forall r\geq s\,,$$The pointer of the apparatus for the measurement of $R$ is defined by the field coordinate observables $$\hat{q}_{s}=\sqrt{\hbar }(\hat{a}_{s}+\hat{a}_{s}^{\dagger })\,,\quad s\in
\mathbb{R}_{+} \label{eq:3.6}$$which are compatible with $[\hat{q}_{r},\hat{q}_{s}]=0$ as well as with $[%
\hat{p}_{r},\hat{p}_{s}]=0$, but incompatible with (\[eq:3.5\]): $$\lbrack \hat{p}_{r},\hat{q}_{s}]=s{\frac{\hbar }{\mathrm{i}}}\hat{1}\ ,\quad
\forall r\geq s\,.$$The operators $S^{\dagger }(t)$ satisfy the condition (\[eq:1.2\]): $%
\langle x|S(t)=\hat{s}_{x}(t)\langle x|$, where the unitary operators $\hat{s%
}_{x}^{\dagger }(t):\mathcal{F}\rightarrow \mathcal{F}$ can be described by the shifts $$\hat{s}_{x}^{\dagger }(t):|q,t\rangle \mapsto |xt+q,t\rangle \,,\quad
\forall x,q,t$$similarly to (\[eq:2.7\]). Here $|q,t\rangle $ is the (generalized) marginal eigenvector of the self-adjoint operator $$\hat{e}(t)=t^{-1}\hat{q}_{t}\ ,\quad \hat{q}_{t}|q,t\rangle =q|q,t\rangle
\,,$$uniquely defined up to the phase by an eigenvalue $q\in \mathbb{R}$ as the Dirac $\delta $-function $\delta _{q}$ in the $\hat{q}_{t}$-representation $%
L^{2}(\mathbb{R})$ of the Hilbert subspace $\mathcal{A}(t)\varphi
_{0}\subset \mathcal{F}$, where $\mathcal{A}(t)$ is the Abelian algebra generated by $\hat{q}_{t}$, and $\varphi _{0}\in \mathcal{F}$ is the vacuum–vector of the Fock space $\mathcal{F}$. Due to this, $$\hat{y}_{x}(t)=\hat{s}_{x}^{\dagger }(t)\hat{e}(t)\hat{s}_{x}(t)=x\hat{1}+%
\hat{e}(t)\,,$$which gives the quantum stochastic realization of the observation model ([eq:3.4]{}) in terms of the output nondemolition process $\hat{y}_{R}(t)={%
\frac{1}{t}}Y(t)$, $$Y(t)=S^{\dagger }(t)(I\otimes \hat{q}_{t})S(t)=tR\otimes \hat{1}+I\otimes
\hat{q}_{t} \label{eq:3.8}$$similarly to (\[eq:2.8\]) with $\hat{q}_{t}$ represented by the operator $%
\sqrt{\hbar }(\hat{a}_{t}+\hat{a}_{t}^{\dagger })$. Indeed, the classical noise $q_{t}=\sqrt{\hbar }w_{t}$ is statistically equivalent to the quantum one $\hat{q}_{t}=\sqrt{\hbar }(\hat{a}_{t}+\hat{a}_{t}^{\dagger })$ with respect to the vacuum state, as can be seen by a comparison of their characteristic functionals:$$\begin{aligned}
\langle e^{\mathrm{i}\int f(s)\mathrm{d}q}\rangle &:&=\int \exp \{\mathrm{i}%
\sqrt{\hbar }\int_{0}^{\infty }f(s)\mathrm{d}w\}\mathrm{d}\pi =\exp \left\{ -%
{\frac{\hbar }{2}}\int_{0}^{\infty }f(s)^{2}\mathrm{d}s\right\} \\
&=&\langle \varphi _{0},e^{\mathrm{i}\int f(s)\mathrm{d}\hat{a}^{\dagger
}}e^{-{\frac{\hbar }{2}}\int_{0}^{\infty }f(s)^{2}\mathrm{d}s}e^{\mathrm{i}%
\int f(s)\mathrm{d}\hat{a}}\varphi _{0}\rangle =\langle \varphi _{0},e^{%
\mathrm{i}\int f(s)\mathrm{d}\hat{q}}\varphi _{0}\rangle \,.\end{aligned}$$Here we used the annihilation property $\hat{a}_{s}\varphi _{0}=0$ and the Wick ordering formula $$\exp \{z^{\prime }\hat{a}_{s}+\hat{a}_{s}^{\dagger }z\}=e^{z\hat{a}%
_{s}^{\dagger }}\exp \left\{ z^{\prime }{\frac{s}{2}}z\right\} e^{z^{\prime }%
\hat{a}_{s}}\ . \label{eq:3.9}$$The observable process (\[eq:3.8\]) satisfies the nondemolition condition (\[eq:1.1\]) (and self-nondemolition) with respect to any quantum process $%
X(t)=\left( S^{\dagger }ZS\right) (t)$ given by the operators $Z(t)$, commuting with all $Q(s)=I\otimes \hat{q}(s)$, $s\leq t$, because $$Y(s)=S^{\dagger }(t)(I\otimes \hat{q}(s))S(t)\,,\quad \forall s\leq t\ ,$$as follows from the commutation relations $$\hat{s}_{x}^{\dagger }(t)\hat{q}_{s}=(sx\hat{1}+\hat{q}_{s})\hat{s}%
_{x}^{\dagger }(t)\ ,\quad \forall s\leq t$$for $\hat{s}_{x}^{\dagger }(t)=\exp \left\{ {\frac{\mathrm{i}}{\hbar }}x\hat{%
p}_{t}\right\} $. Indeed, due to this $$\lbrack X(t),Y(s)]=W(t)[Z(t),Q(s)]W^{\dagger }(t)=0\,,$$if $t>s$ and $[Z(t),Q(s)]=0$, as in the cases $Z(t)=C\otimes \hat{1}$ and $%
Z(t)=Q(t)$, where $Q(t)=I\otimes \hat{q}_{t}$.
Now we can find the transform $$\langle q,t|S\varphi _{0}=G(t,q)\varphi _{0}(t,q)={\frac{1}{\sqrt{t}}}%
T\left( t,{\frac{1}{t}}q\right) \,,$$where $\varphi _{0}(t,q)=\langle q,t|\varphi _{0}$ is the vacuum-vector $%
\varphi _{0}\in \mathcal{F}$ in the marginal $\hat{q}_{t}=q$ representation $$\varphi _{0}(t,q)=(ht)^{-1/4}\exp \{-\pi q^{2}/2ht\}\,,\quad q\in \mathbb{R}$$normalized with respect to the Lebesgue measure $\mathrm{d}q$ on $\mathbb{R}$. To this end, let us apply the formula (\[eq:3.9\]) to $S^{\dagger
}(t)=\exp \left\{ {\frac{\mathrm{i}}{\hbar }}R\otimes \hat{p}_{t}\right\} $: $$\exp \{-L\otimes \hat{a}_{t}+L\otimes \hat{a}_{t}^{\dagger }\}=e^{L\otimes
\hat{a}_{t}^{\dagger }}\exp \left\{ -{\frac{t}{2}}L^{2}\right\} e^{-L\otimes
\hat{a}_{t}}\,,$$where $L=R/2\sqrt{\hbar }$. Using the annihilation property $\exp \{\pm
L\otimes \hat{a}_{t}\}\varphi _{0}=\varphi _{0}$, we obtain $$\begin{aligned}
W(t)^{\dagger }\varphi _{0} &=&e^{L\otimes \hat{a}_{t}^{\dagger }}\exp
\left\{ -{\frac{t}{2}}L^{2}\right\} e^{-L\otimes \hat{a}_{t}}\varphi _{0} \\
&=&e^{L\otimes \hat{a}_{t}^{\dagger }}\exp \left\{ -{\frac{t}{2}}%
L^{2}\right\} e^{L\otimes \hat{a}_{t}}\varphi _{0}=e^{L\otimes \hat{w}%
_{t}-tL^{2}}\varphi _{0}\,.\end{aligned}$$This is equivalent to (\[eq:3.2\]) because of the Segal isometry of the vectors $\exp \{x\hat{w}_{t}\}\varphi _{0}\in \mathcal{F}$, where $x\in
\mathbb{R} $, $\hat{w}_{t}=\hat{a}_{t}+\hat{a}_{t}^{\dagger }$, and the stochastic functions $\exp \{xw_{t}\}\in L_{\pi }^{2}(\Omega )$ in the Hilbert space of the Wiener measure $\pi $ on $\Omega $. Hence the transform $F\left( t,{\frac{1}{t}}q\right) =\sqrt{t}G(t,q)\varphi _{0}(t,q)$ defining the density $\Phi (t,y)=F(t,y)^{\dagger }[\cdot ]F(t,y)$ of the instrument (\[eq:2.9\]) with respect to $\mathrm{d}y$ has the same form, as in ([eq:3.3]{}): $$F(t,y)=(t/h)^{1/4}\exp \left\{ -{\frac{\pi t}{2h}}(y-R)^{2}\right\} \,.
\label{eq:3.10}$$
A Hamiltonian model for continuous reduction
============================================
As we have shown in the previous section the continuous reduction equation (\[eq:3.1\]) for the non-normalized stochastic state-vector $\chi(t,\omega)$ can be obtained from an interaction model of the object of measurement with a bosonic field. This can be done by conditioning with respect to a nondemolition continuous observation of field coordinate observables ([eq:3.6]{}) in the vacuum state.
The unitary evolution $\psi (t)=U(t)\psi _{0}$ in the tensor product $%
\mathcal{G}=\mathcal{H}\otimes \mathcal{F}$ with the Fock space $\mathcal{F}$ corresponding to (\[eq:3.1\]) can be written as the generalized Schrödinger equation $$\mathrm{d}\psi (t)+K_{0}\psi (t)\mathrm{d}t=(L\otimes \mathrm{d}\hat{a}%
_{t}^{\dagger }-L^{\dagger }\otimes \mathrm{d}\hat{a}_{t})\psi (t)
\label{eq:4.1}$$in terms of the annihilation and creation canonical field operators $\hat{a}%
_{s}$, $\hat{a}_{s}^{\dagger }$. This is a singular differential equation which has to be treated as a quantum stochastic one \[29\] in terms of the forward increments $\mathrm{d}\psi (t)=\psi (t+\mathrm{d}t)-\psi (t)$ with $%
K_{0}=K\otimes \hat{1}$, $K=(\mathrm{i}/\hbar )H+{\frac{1}{2}}L^{\dagger }L$. In the particular case $L=R/2\sqrt{\hbar }=L^{\dagger }$ of interest, eq. (\[eq:4.1\]) can be written simply as a classical stochastic one, $\mathrm{d%
}\psi +K\psi \mathrm{d}t=(\mathrm{i}/\hbar )R\mathrm{d}p$, in Itô sense with respect to a Wiener process $p_{t}$ of the same intensity $(\mathrm{d}%
p_{t})^{2}=\hbar \mathrm{d}t/4$ as the field momenta operators (\[eq:3.5\]) with respect to the vacuum state. But the standard Wiener process $%
v_{t}=2p_{t}/\sqrt{\hbar }$ cannot be identified with the Wiener process $%
w_{t}$ in the reduction equation (\[eq:3.1\]) because of the nondemolition principle. Moreover, there is no way to get the nondemolition property ([eq:1.1]{}) for $$X(t)=U(t)^{\dagger }X_{0}U(t)\ ,\quad Y(s)=U(s)^{\dagger }Y_{0}(s)U(s)$$with the independent or if only commuting $v_{t}$ and $w_{t}$, as one can see in the simplest case $H=0$, $X_{0}={\frac{\hbar }{\mathrm{i}}}{\frac{%
\mathrm{d}}{\mathrm{d}x}}\otimes \hat{1}$, $R=x$, $Y_{0}(s)=I\otimes \hat{q}%
_{s}$.
Indeed, the error process $q_{t}=\sqrt{\hbar }w_{t}$ is appearing in ([eq:3.4]{}) as a classical representation of the field coordinate observables (\[eq:3.6\]) which do not commute with (\[eq:3.5\]). In this case, eq. (\[eq:4.1\]) gives the unitary operator $U(t)=\exp \left\{ -{\frac{\mathrm{i%
}}{\hbar }}x\otimes \hat{p}_{t}\right\} $ and the Heisenberg operators $$X(t)={\frac{\hbar }{\mathrm{i}}}{\frac{\mathrm{d}}{\mathrm{d}x}}\otimes \hat{%
1}-I\otimes \hat{p}_{t}\ ,\quad Y(s)=sx\otimes \hat{1}+I\otimes \hat{q}_{s}$$commute for all $t\geq s$ only because $$\left[ {\frac{\hbar }{\mathrm{i}}}{\frac{\mathrm{d}}{\mathrm{d}x}},sx\right]
\otimes \hat{1}=s{\frac{\hbar }{\mathrm{i}}}I\otimes \hat{1}=[\hat{p}_{t},%
\hat{q}_{s}]\ ,\quad \forall t\geq s\,.$$Hence, there is no way to obtain (\[eq:1.1\]) for the classical stochastic processes $p_{t}$, $q_{s}$ by replacing simultaneously $\hat{p}_{t}$ and $%
\hat{q}_{s}$ for commuting $\sqrt{\hbar }v_{t}/2$ and $\sqrt{\hbar }w_{t}$ even though $p_{t}$ is statistically identical to $\hat{p}_{t}$ and separately $q_{s}$ to $\hat{q}_{s}$.
Let us show now how one can get a completely different type of the reduction equation than postulated in \[16\]–\[20\] simply by fixing an another nondemolition process for the same interaction, corresponding to the Schrödinger stochastic equation (\[eq:4.1\]) with $L=L^{\dagger }$ and $%
H=0 $.
We fix the discrete pointer of the measurement apparatus, which is described by the observable $\hat{n}_{s}={\frac{1}{s}}\hat{a}_{s}^{\dagger }\hat{a}_{s}
$, by counting the quanta of the Bosonic field in the mode $1_{s}(r)=1$ if $%
r\in \lbrack 0,s)$ and $1_{s}(r)=0$ if $r\notin \lbrack 0,s)$. The operators $\hat{n}_{t}$ have the integer eigenvalues $0,1,2,\dots $ corresponding to the eigen-vectors $$|n,t\rangle =e^{t/2}(\hat{a}_{t}^{\dagger }/t)^{n}\varphi _{0}\ ,\quad \hat{a%
}_{t}\varphi _{0}=0$$which we have normalized with respect to the standard Poissonian distribution $$\nu _{n}=e^{-t}t^{n}/n!\ ,\quad n=0,1,\dots \ \label{eq:4.2}$$as $\langle n,t|n,t\rangle =1/\nu _{n}$. Let us find the matrix elements $$\langle n,t|S(t)\varphi _{0}=G(t,n)$$for the unitary evolution operators $$S(t)=\exp \{-L\otimes \hat{a}_{t}+L\otimes \hat{a}_{t}^{\dagger }\}\,,
\label{eq;4.3}$$by resolving eq. (\[eq:4.1\]) in the considered case. This can be done again by representing $S(t)$ in the form (\[eq:3.9\]) for $z^{\prime }=L$, $z=-L$ and the commutation rule $$(I\otimes \hat{a}_{t})e^{L\otimes \hat{a}_{t}^{\dagger }}=e^{L\otimes \hat{a}%
_{t}^{\dagger }}(tL\otimes \hat{1}+I\otimes \hat{a}_{t})\ .$$Due to the annihilation property, this gives $$\varphi _{0}^{\dagger }(\hat{a}_{t}/t)^{n}e^{L\otimes \hat{a}_{t}^{\dagger
}}\exp \left\{ {\frac{t}{2}}(1-L^{2})\right\} e^{-L\otimes \hat{a}%
_{t}}\varphi _{0}=L^{n}\exp \left\{ {\frac{t}{2}}(1-L^{2})\right\} =G(t,n)\ .
\label{eq:4.4}$$The obtained reduction transformations are not unitary and not projective for any $n=0,1,2,\dots $, but they define the nonorthogonal identity resolution $$\sum_{n=0}^{\infty }G(t,n)^{\dagger }G(t,n)e^{-t}t^{n}/n!=I$$corresponding to the operational density $$\Psi \lbrack C](t,n)=e^{t}L^{n}e^{-L^{2}/2}CL^{n}e^{-L^{2}/2} \label{eq:4.5}$$with respect to the measure (\[eq:4.2\]). Now we can easily obtain the stochastic reduction equation for $\chi (t,\omega )=T(t,\omega )\eta $ if we replace the eigenvalue $n$ of $\hat{n}_{t}$ by the standard Poissonian process $n_{t}(\omega )$ with the marginal distributions (\[eq:4.2\]). Such a process $n_{t}$ describes the trajectories $t\mapsto n_{t}(\omega )$ that spontaneously increase by $\mathrm{d}n_{t}(\omega )=1$ at random time instants $\omega =\{t_{1}<t_{2}<\dots \}$ as the spectral functions $%
\{n_{t}(\omega )\}$ for the commutative family $\{\hat{n}_{t}\}$. The corresponding equation for the stochastic state-vector $\chi (t,\omega
)=\chi (t,n_{t}(\omega ))$ can be written in the Itô sense as $$\mathrm{d}\chi (t)+{\frac{1}{2}}(L^{2}-I)\chi (t)\mathrm{d}t=(L-1)\chi (t)%
\mathrm{d}n_{t}\,. \label{eq:4.6}$$Obviously Eq. (\[eq:4.6\]) has the unique solution $\chi (t)=F(t)\eta $ written for a given $\eta \in \mathcal{H}$ as $$\chi (t)=L^{n_{t}}\exp \left\{ {\frac{t}{2}}(1-L^{2})\right\} \eta
=G(t,n_{t})\eta \label{eq:4.7}$$because of $\mathrm{d}\chi (t)=(L-1)\chi (t)$ when $\mathrm{d}n_{t}=1$, otherwise $\mathrm{d}\chi (t)={\frac{1}{2}}(1-L^{2})\chi (t)\mathrm{d}t$ in terms of the forward differential $\mathrm{d}\chi (t)=\chi (t+\mathrm{d}%
t)-\chi (t)$.
Such an equation was derived in \[26\]–\[30\] also for the general quantum stochastic equation (\[eq:4.1\]) on the basis of quantum stochastic calculus and filtering theory \[31\]. Moreover, it was proved that any other stochastic reduction equation can be obtained as a mixture of Eq. ([eq:3.1]{}) and (\[eq:4.4\]) which are of fundamentally different types.
Finally let us write down a Hamiltonian interaction model corresponding to the quantum stochastic Schrödinger equation (\[eq:4.1\]). Using the notion of chronologically ordered exponential $$U(t)=\exp ^{\leftarrow }\left\{ -{\frac{\mathrm{i}}{\hbar }}\int_{0}^{t}H(r)%
\mathrm{d}r\right\} \label{eq:4.8}$$one can extend its solutions $\psi (t)=\exp \left\{ -{\frac{\mathrm{i}}{%
\hbar }}R\otimes \hat{p}_{t}\right\} \psi _{0}$ also to the general case, $%
H\not=0$, $L^{\dagger }\not=L$ in terms of the generalized Hamiltonian $$H(t)=H_{0}+{\frac{\hbar }{\mathrm{i}}}(L^{\dagger }\otimes \hat{a}%
(t)-L\otimes \hat{a}(t)^{\dagger })\,,$$where $\hat{a}(t)=\mathrm{d}\hat{a}_{t}/\mathrm{d}t$, $\hat{a}^{\dagger }(t)=%
\mathrm{d}\hat{a}_{t}^{\dagger }/\mathrm{d}t$, $H_{0}=H\otimes \hat{1}$. The time-dependent Hamiltonian $H(t)$ can be treated as the object interaction Hamiltonian $$H(t)=H_{0}+{\frac{\hbar }{\mathrm{i}}}e^{{\frac{\mathrm{i}}{\hbar }}%
H_{1}t}(L^{\dagger }\otimes \hat{a}(0)-L\otimes \hat{a}(0)^{\dagger })e^{-{%
\frac{\mathrm{i}}{\hbar }}H_{1}t}$$for a special free evolution Hamiltonian $H_{1}=I\otimes \hat{h}$ of the quantum bosonic field $\hat{a}(r)$, $r\in \mathbb{R}$ described by the canonical commutation relations $$\lbrack \hat{a}(r),\hat{a}(s)]=0,\quad \lbrack \hat{a}(r),\hat{a}%
(s)^{\dagger }]=\delta (r-s)\hat{1}\,,\quad \forall \,r,s\in \mathbb{R}\,.$$This free evolution in the Fock space $\mathcal{F}$ over one particle space $%
L^{2}(\mathbb{R})$ is simply given by the shifts $$e^{{\frac{\mathrm{i}}{\hbar }}\hat{h}t}\hat{a}(r)e^{-{\frac{\mathrm{i}}{%
\hbar }}\hat{h}t}=\hat{a}(r+t)\,,\quad \forall \,r,t\in \mathbb{R}\,,$$corresponding to the second quantization $\hat{h}=\hat{a}^{\dagger }\hat{%
\varepsilon}\hat{a}$ of the one-particle Hamiltonian $\hat{\varepsilon}={%
\frac{\hbar }{\mathrm{i}}}{\frac{\partial }{\partial r}}$ in $L^{2}(\mathbb{R%
})$. Hence, the total Hamiltonian of the system object plus measurement apparatus can be written as $$H_{s}=H\otimes \hat{1}+{\frac{\hbar }{\mathrm{i}}}(L^{\dagger }\otimes \hat{a%
}(0)-L\otimes \hat{a}(0)^{\dagger }+I\otimes \hat{a}^{\dagger }\hat{a}%
^{\prime })\,, \label{eq:4.9}$$where $a^{\dagger }a^{\prime }=\int\limits_{-\infty }^{\infty }\hat{a}%
(r)^{\dagger }\hat{a}(r)^{\prime }$$r$, $\hat{a}(r)^{\prime }=%
\mathrm{d}\hat{a}(r)/\mathrm{d}r$. Of course, the free field Hamiltonian $%
\hat{h}=\hbar \hat{a}^{\dagger }\hat{a}^{\prime }/\mathrm{i}$ is rather unusual as with respect to the single-particle energy $\varepsilon (p)=p$ in the momentum representation giving the unbounded (from below) spectrum of $%
\hat{\varepsilon}$.
But one can consider such an energy as an approximation $$\varepsilon (p)=\lim_{p_{0}\rightarrow \infty }c\left( \sqrt{%
(p+p_{0})^{2}+(m_{0}c)^{2}}-\sqrt{p_{0}^{2}+(m_{0}c)^{2}}\right) =v_{0}p
\label{eq:4.10}$$in the velocity units $v_{0}=c/\sqrt{1+(m_{0}c/p_{0})^{2}}=1$ for the shift $%
\varepsilon _{0}(p)-\varepsilon _{0}(0)$ of the standard relativistic energy $\varepsilon _{0}(p)=c\sqrt{(p+p_{0})^{2}+(m_{0}c)^{2}}$ as the function of small deviations $|p|\ll p_{0}$ from the initially fixed momentum $p_{0}>0$. This corresponds to the treatment of the measurement apparatus as a beam of bosons with mean momentum $p_{0}\rightarrow \infty $ given in an initial coherent state by a plane wave $$f_{0}(r)=c\exp \{\mathrm{i}p_{0}r/\hbar \}\,.$$This input beam of bosons illuminate the position $R=\sqrt{\hbar }%
(L+L^{\dagger })$ of the object of measurement via the observation of the commuting position operators $Y(t)$, $t\in \mathbb{R}$ of the output field given by the generalized Heisenberg operator-process, $$\begin{aligned}
\dot{Y}(t) &=&e^{{\frac{\mathrm{i}}{\hbar }}H_{s}t}(I\otimes \hat{q}(0))e^{-{%
\frac{\mathrm{i}}{\hbar }}H_{s}t} \\
&=&U(t)^{\dagger }(I\otimes \hat{q}(t))U(t)=R(t)+I\otimes \hat{q}(t)\end{aligned}$$This is the simplest quantum Hamiltonian model for the continuous nondemolition measurement of the physical quantity $R$ of a quantum object.
Thus the unitary evolution group $U_{s}(t)=e^{{\frac{-\mathrm{i}}{\hbar }}%
H_{s}t}$ of the compound system is defined on the product $\mathcal{H}%
\otimes \mathcal{F}$ with the two–sided Fock space $\mathcal{F}=\Gamma
(L^{2}(\mathbb{R}))$ by $U_{s}(t)=V_{1}(t)U(t)$, where $V_{1}=I\otimes \hat{v%
}$ is the free evolution group $\hat{v}(t)=e^{{\frac{-\mathrm{i}}{\hbar }}%
\hat{h}t}$ of the field, corresponding to the shifts $$f\in L^{2}(\mathbb{R})\mapsto f^{t}(s)=f(s-t)$$of the one-particle space $L^{2}(\mathbb{R})$. To obtain such an evolution from a realistic Hamiltonian of a system of atoms interacting with an electromagnetic field one has to use a Markovian approximation, corresponding to the weak-coupling or low density limits \[39\].
Thus, the problem of unitary dilation of the continuous reduction and spontaneous collapse was solved in \[25\] even for infinite-dimensional Wiener noise in a stochastic equation of type (\[eq:3.2\]).
Conclusion {#conclusion .unnumbered}
==========
Analysis \[1\] of the quantum measurement notion shows that it is a complex process, consisting of the stage of preparation \[15\] and the stage of registration, i.e., fixing of the pointer and its output state and the objectification \[40\].
The dynamical process of the interaction is properly treated within the quantum theory of singular coupling to get the nontrivial models of continuous nondemolition observation while the statistical process of the objectification is properly treated within the quantum theory of stochastic filtering to get the nonlinear models of continuous spontaneous localization \[21–31\].
The nondemolition principle plays the role of superselection for the observable processes provided the quantum dynamics is given and restricts the dynamics provided the observation is given. It is a necessary and sufficient condition for the statistical interpretation of quantum causality, giving rise to the quantum noise environment but not to the classical noise environment of the phenomenological continuous reduction and spontaneous localization theories \[16–20\].
The axiomatic quantum measurement theory based on the nondemolition principle abandons the projection postulate as the redundancy given by a unitary interaction with a meter in the initial eigen-state. It treats the reduction of the wave packet not as a real dynamical process but as the statistical evaluation of the *a posteriori* states for the prediction of the probabilities of the future measurements conditioned by the past observation.
There is no need to postulate a nonstandard, nonunitary, and nonlinear evolution for the continuous state-vector reduction in the phenomenological quantum theories of spontaneous localization, and there is no universal reduction modification of the fundamental Schrödinger equation. The nonunitary stochastic evolution giving the continuous reduction and the spontaneous localization of the state-vector can be and has been rigorously derived within the quantum stochastic theory of unitary evolution of the corresponding compound system, the object of the measurement and an input Bose field in the vacuum state.
The statistical treatment of the quantum measurement as nondemolition observation is possible only in the framework of open systems theory in the spirit of the modern astrophysical theory of the spreading universe. The open systems theory assumes the possibility of producing for each quantum object an arbitrary time series of its copies and enlarges these objects into an environment, a quantum field, innovating the measurement apparatus by means of a singular interaction for a continuous observation.
It is nonsense to consider seriously a complete observation in the closed universe; there is no universal quantum observation, no universal reduction and spontaneous localization for the wave function of the world. Nobody can prepare an* a priori* state compatible with a complete world observation and reduce the *a posteriori* state, except God. But acceptance of God as an external subject of the physical world is at variance with the closeness assumption of the universe. Thus, the world state-vector has no statistical interpretation, and the humanitarian validity of these interpretations would, in any case, be zero. The probabilistic interpretation of the state-vector is relevant to only the induced states of the quantum open objects being prepared by experimentalists in an appropriate compound system for the nondemolition observation to produce the reduced states after the registration.
Acknowledgment {#acknowledgment .unnumbered}
==============
This work was supported by Deutsche Forschung Gemeinschaft at Philipps Universität, Marburg. I am deeply grateful to Professors L. Accardi, O. Melsheimer, and H. Neumann for stimulating discussions and encouragement.
References {#references .unnumbered}
==========
[\[1\]]{} <span style="font-variant:small-caps;">G. Ludwig</span>, *Math. Phys.*, 4:331 (1967); 9, 1 (1968).
[\[2\]]{} <span style="font-variant:small-caps;">E.B. Davies, J. Lewis</span>, *Commun. Math. Phys.*, 17:239–260, (1970).
[\[3\]]{} <span style="font-variant:small-caps;">L.E. Ballentine</span>, *Rev. Mod. Phys.*, 42:358–381, (1970).
[\[4\]]{} <span style="font-variant:small-caps;">A. Shimony</span>, *Phys. Rev. D*, 9:2321–2323, (1974).
[\[5\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Optimal linear random filtration of quantum Boson signals. Problems of Control and Inform. Theory, 3:47–62, (1974).
[\[6\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Optimal quantum filtration of Markovian singals. Problems of Control and Inform. Theory", 7(5):345–360, (1978).
[\[7\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Optimal filtering of Markov signals with quantum noise, *Radio Eng. Electron. Physics*, 25:1445–1453, (1980).
[\[8\]]{} <span style="font-variant:small-caps;">A. Barchielli, L. Lanz, G.M. Prosperi</span>, *Nuovo Cimento*, 72B:79, (1982).
[\[9\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Theory of Control of Observable Quantum Systems, *Automatica and Remote Control*, 44(2):178–188, (1983).
[\[10\]]{} <span style="font-variant:small-caps;">A. Peres</span>, *Am. J. Phys.*, 52:644, (1984).
[\[11\]]{} <span style="font-variant:small-caps;">R.L. Stratonovich</span>, Conditional Markov processes and their applications to optimal control, *MGU*, Moscow 1966.
[\[12\]]{} <span style="font-variant:small-caps;">R.E. Kalman, R.S. Bucy</span>, New results in linear filtering theory and prediction problems, *J. Basic Engineering, Trans. ASME*, 83:95–108, (1961).
[\[13\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Nondemolition measurement and control in quantum dynamical systems. In: Proc. of CISM seminar on "*Inform. Compl. and Control in Quantum Physics*, A. Blaquiere, ed., Udine 1985, 311–239, Springer–Verlag, Wien, 1987.
[\[14\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Nondemolition measurements, nonlinear filtering and dynamical programming of quantum stochastic processes. In: Proc. of Bellmann Continuum Workshop *Modelling and Control of Systems*, A. Blaquiere, ed., Sophia–Antipolis 1988, 245–265, *Lect. Not. Cont. Inf. Sci.*, 121, Springer–Verlag, Berlin, 1988.
[\[15\]]{} <span style="font-variant:small-caps;">L.E. Ballentine</span>, *Int. J. Theor. Phys.*, 27:211–218, (1987).
[\[16\]]{} <span style="font-variant:small-caps;">P. Pearle</span>, *Phys. Rev.*, D29:235, (1984).
[\[17\]]{} <span style="font-variant:small-caps;">N. Gisen</span>, *J. Phys. A.: Math. Gen.*, 19:205–210, (1986).
[\[18\]]{} <span style="font-variant:small-caps;">L. Diosi</span>, *Phys. Rev.*, A40:1165–1174, (1988).
[\[19\]]{} <span style="font-variant:small-caps;">G.C. Ghirardi, A. Rimini, T. Weber</span>, *Phys. Rev.*. D34(2):470–491, (1986).
[\[20\]]{} <span style="font-variant:small-caps;">G.C. Ghirardi, P. Pearle, A. Rimini</span>, *Phys. Rev.*, A42:478–89, (1990).
[\[21\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, A new wave equation for a continuous non–demolition measurement", *Phys. Lett.*, A140:355–358, (1989).
[\[22\]]{} <span style="font-variant:small-caps;">V.P. Belavkin, P. Staszewski</span>, A quantum particle undergoing continuous observation", *Phys. Lett.*, A140:359–362, (1989).
[\[23\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, A posterior Schrödinger equation for continuous non–demolition measurement, *J. Math. Phys*, 31(12):2930–2934, (1990).
[\[24\]]{} <span style="font-variant:small-caps;">V.P. Belavkin, P. Staszewski</span>, Nondemolition observation of a free quantum particle, *Phys. Rev. A.*, 45(3):1347–1356, (1992).
[\[25\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Quantum continual measurements and a posteriori collapse on CCR, *Commun. Math. Phys.*, 146, 611–635, (1992).
[\[26\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, A continuous counting observation and posterior quantum dynamics, *J. Phys. A, Math. Gen.*, 22: L 1109–1114, (1989).
[\[27\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, A stochastic posterior Schrödinger equation for counting non–demolition measurement, *Letters in Math. Phys.*, 20"85–89, (1990).
[\[28\]]{} <span style="font-variant:small-caps;">V.P. Belavkin, P. Staszewski</span>, *Rep. Math. Phys.*, 29:213–225, (1991).
[\[29\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Stochastic posterior equations for quantum nonlinear filtering. Probab., *Theory and Math. Stat.*, ed. B. Grigelionis, 1:91–109, VSP/Mokslas 1990.
[\[30\]]{} <span style="font-variant:small-caps;">A. Barchielli, V.P. Belavkin</span>, Measurements continuous in time and a posteriori states in quantum mechanics, *J. Phys. A, Math. Gen.*, 24:1495–1514, (1991).
[\[31\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Quantum stochastic calculus and quantum nonlinear filtering, *J. of Multivar. Analysis*, 42(2):171–201, (1992).
[\[32\]]{} <span style="font-variant:small-caps;">V.B. Braginski, Y.I. Vorontzov, F.J. Halili</span>, *Sov. Phys.–JETP*, 46(2):765, 4:171–201, (1977).
[\[33\]]{} <span style="font-variant:small-caps;">K.S. Thorne, R.W.P. Drever, C.M. Caves, M. Zimmermann, V.D. Sandberg</span>, *Phys. Rev. Lett.*, 40:667, (1978).
[\[34\]]{} <span style="font-variant:small-caps;">A.S. Holevo</span>, Quantum estimation. In Advances in statistical signal processing, 1:157–202, (1987).
[\[35\]]{} <span style="font-variant:small-caps;">V.P. Belavkin</span>, Reconstruction theorem for quantum stochastic processes, *Theoret. Math. Phys.*, 3:409–431, (1985).
[\[36\]]{} <span style="font-variant:small-caps;">K. Kraus</span>, States, Effects and operations, Springer–Verlag, Berlin 1983.
[\[37\]]{} <span style="font-variant:small-caps;">E.B. Ozawa</span>, *J. Math. Phys.*, 25:79–87, (1984).
[\[38\]]{} <span style="font-variant:small-caps;">A. Barchielli, G. Lupieri</span>, *J. Math. Phys.*, 26:2222–2230, (1985).
[\[39\]]{} <span style="font-variant:small-caps;">L. Accardi, R. Alicki, A. Frigerio, Y.G. Lu</span>, An invitation to weak coupling and low density limits, Quantum probability and re. topics VI, ed. L. Accardi, 3–62, World Scientific, Singapore 1991.
[\[40\]]{} <span style="font-variant:small-caps;">P. Busch, P.J. Lahti, P. Mittelstaedt</span>, The quantum theory of measurement, *Lecture Notes in Physics*, Springer–Verlag, Berlin 1991.
[^1]: Published in:* Foundations of Physics*, **24** (1994) No 5, 685–714
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The energy budget of all known optical jets is discussed. It is found that to power the extended radio lobes of radio galaxies, the jet bulk motion on kpc scales must be relativistic, on average. Based on various constraints, a “most probable” region centered around $\Gamma_{bulk} \sim 7.5$ and $\theta \sim 20^\circ$ is found. Because of the consequent relativistic beaming, the rest frame magnetic field is lower and electron lifetimes longer. Combining the effects of time dilation and lower emission rate, the electron diffusion length becomes fully consistent with the deprojected jet length, without the need for reacceleration.'
address: 'Space Telescope Science Institute, 3700 San Martin Dr., Baltimore, MD 21218'
author:
- Riccardo Scarpa
- 'C. Megan Urry'
title: |
On the Physical Conditions in\
AGN Optical Jets
---
,
Introduction {#intro}
============
To emit at optical-UV frequencies in a field of $B\sim
10^{-4}$ G, typical of the magnetic fields found in AGN jets on kpc scales, electrons must have Lorentz factors $\gamma {\raisebox{-3.8pt}{$\;\stackrel{\textstyle > }{\sim }\;$}}10^6$ and have diffusion times of a few hundred years. Despite the shortness of the expected synchrotron cooling times, the jets are long and there is no indication of strong steepening of the radio-to-optical spectral index as the distance from the nucleus increases (Sparks 1996; Scarpa 1999). Therefore, either electrons are continuously reaccelerated or the magnetic field is weaker than the equipartition value, as may be the case if relativistic beaming is important.\
To shed light on this issue we analyze the energy budget of all known optical jets, requiring they transport on average as much energy as needed to explain the existence of extended radio lobes. Our approach relies on the hypothesis that the extended radio structures are powered by jets (Blandford & Königl 1979), and on the fact that, even if rare, optical jets are discovered in all kinds of radio sources (of both FRI and FRII morphologies and powers).
---------------- ------ ------- --------- ------- --------- ------- ------ ------ --------------- ---------- ----------------------------------------------------------------- --------- ---------
Name Cl. z $\nu_R$ $F_R$ $\nu_O$ $F_O$ l Vol $\alpha_{RO}$ ref. $J$ B$_{o}$ B$_{r}$
(b) (c) (c) (d) (d) (e) (f) (g) (h) (i) (j) (k)
M87 I 0.004 15 4.2 6.88 1960 2.1 62.7 $ -0.71$ 2,24 $ >450 $ 31 12
3C 15 I/II 0.073 2.78 0.21 5.50 5.9 3.2 65.4 $ -0.86$ 21 $ >50 $ 8.7 3.3
3C 66B I/II 0.021 15 0.04 8.17 6.5 3.0 65.1 $ -0.80$ 2,4,6,25 $ {\raisebox{-3.8pt}{$\;\stackrel{\textstyle > }{\sim }\;$}}40$ 4.5 1.7
3C 78 I 0.029 1.66 0.8 4.34 33 1.1 62.2 $ -0.81$ 12,26 $ >40 $ 53 20
3C 120 I 0.033 5.0 0.05 6.88 14 0.32 64.6 $ -0.69$ 11,13,27 $ >1000 $ 6.3 2.4
3C 200 II 0.458 5.0 0.076 4.16 18 0.32 65.9 $ -0.74$ 9,10 $ >37.5 $ 15 5.8
3C 212 II 1.049 8.33 0.04 9.09 0.19 0.85 67.1 $ -1.06$ 7 $ \dots $ 56 21
3C 245 II 1.029 5.0 0.045 9.09 0.17 31.7 67.2 $ -1.03$ 7,8,22 $ \dots $ 84 32
3C 264 I 0.020 15 0.2 4.16 33 0.32 62.1 $ -0.85$ 1,2 $ >37 $ 89 34
3C 273 II 0.158 15 2.41 5.50 57 4.8 64.8 $ -1.07$ 2,5,18 $ >5300 $ 5.8 2.2
3C 346$^{(l)}$ II 0.161 15 0.08 4.16 11 11.0 62.8 $ -0.87$ 14,15 $ \dots $ 5.6 2.1
3C 371 II 0.051 1.6 0.18 5.50 15 28.9 65.9 $ -0.74$ 17,19,23 $ >700 $ 17 6.4
0521–365 I 0.055 15 0.1 4.16 60 21.9 65.9 $ -0.73$ 3,16,23 $ >10 $ 14 5.1
2201+044 I 0.027 5.0 0.009 4.16 0.57 2.05 63.9 $ -0.85$ 20,23 $ \dots $ 7.9 3.0
---------------- ------ ------- --------- ------- --------- ------- ------ ------ --------------- ---------- ----------------------------------------------------------------- --------- ---------
All values are in the observer frame. [**(b)**]{} Fanaroff & Riley morphology class. [**(c)**]{} Radio frequency in GHz and radio flux in Jy. [**(d)**]{} Optical frequency in $10^{14}$ Hz and optical flux in $\mu$Jy. [**(e)**]{} Projected jet length in kpc. [**(f)**]{} Logarithm of jet volume in cm$^3$, estimated assuming cylindrical symmetry. [**(g)**]{} Radio to optical spectral index assuming $F_\nu \propto \nu^\alpha$.\
[**(h)**]{} References for fluxes, size, and $J$: 1) Baum 1997; 2) Crane 1993; 3) Keel 1986; 4) Fraix-Burnet 1989; 5) Bahcall 1995; 6) Macchetto 1991; 7) Ridgway & Stockton 1997; 8) Laing 1989; 9) Burns 1984; 10) Bogers 1994; 11) Hjorth 1995; 12) Sparks 1995; 13) Fraix-Burnet 1991; 14) van Breugel 1992; 15) Dey & van Breugel 1994; 16) Macchetto 1991; 17) Wrobel & Lind 1990; 18) Davis 1991; 19) Akujor 1994; 20) Laurent-Muehleisen 1993; 21) Leahy 1997; 22) Saikia 1990; 23) Scarpa 1999.; 24) Stiavelli 1992; 25) Fraix-Burnet 1997; 26) Saikia 1986; 27) Walker 1987.\
[**(i)**]{} Jet/counter-jet luminosity ratio as published or derived by us from published radio maps. [**(j)**]{} “Observed” equipartition magnetic field in units of $10^{-5}$ G, computed without transforming to the rest frame and without beaming. [**(k)**]{} “Rest frame” equipartition magnetic field in units of $10^{-5}$ G, computed assuming $\Gamma=7.5$ and $\theta = 20^\circ$ for all jets. [**(l)**]{} Only knot “c” as defined in van Breugel (1992) considered.
Theoretical Considerations and Available Data
=============================================
Standard formulae for synchrotron emission are used (Pacholczyk 1970). The synchrotron emission is assumed to extend, in the observer frame, from $\nu_1=10^7$ to $\nu_2=10^{15}$ Hz, following a single power law ($F_\nu \propto \nu^\alpha$) of constant spectral index $\alpha$. The electron distribution in energy space is $N(E)=N_0 E^p$, where $p=2\alpha - 1$. Each electron emits in a narrow range of frequencies centered on $\nu=c_1BE^2$, where $c_1=\frac{3e}{4\pi m^3c^5}$, at a rate of $dE/dt=c_2B^2E^2$, where $c_2=\frac{2e^4}{3m^4c^7}$. In the presence of relativistic beaming, rest frame ($L_R$) and observed ($L$) jet luminosities are related by $L=L_R\delta^3$ (as appropriate for a continuous jet; it would be $L=L_R\delta^4$ for a moving sphere), where $\delta=[\Gamma(1-\beta \cos \theta)]^{-1}$ is the Doppler beaming factor. The rest-frame luminosity is $L_R = \int_{E_1}^{E_2}{\frac{dE}{dt} N(E)dE}$, where $E_1$ and $E_2$ are the cutoffs of the electron energy distribution corresponding to $\nu_1$ and $\nu_2$. After integrating, $L_R$ depends on $B$ and $N_0$, so that a second equation is necessary to solve completely for jet properties. This can be obtained by imposing equipartition of energy, in which case we have $\frac{B^2\phi V}{8\pi} = (1+k)E_e$, where $\phi$ is the magnetic field filling factor, V is the source volume, and $E_e$ is the total electron energy. The proton energy, which remains unconstrained, is accounted for by the proton-to-electron energy ratio $k$. For consistency with previous works, we set $k=0$ (assigning no energy to the protons). Solving for $B$ and $N_0$ allows the calculation of the rest-frame number density $n=\frac{L}{c_2 B^2 <E^2> V \delta^3}$ of the emitting electrons, which is used to derive the kinetic energy transported in the jet, which is $L_k = \pi R^2 \Gamma^2 \beta c n <E>(1+k)$ (Celotti & Fabian 1993). Here $\Gamma$ is the Lorentz factor of the bulk motion, $\beta c$ the plasma speed, and $R$ the jet radius. As before, the term $(1+k)$ accounts for the proton energy. Under these quite standard assumptions, the flux at 2 frequencies and the bulk speed of the plasma suffice to evaluate the jet kinetic energy. Relevant data for all known optical jets are summarized in Table 1. Interestingly, optical jets are discovered in radio sources of all types (Column 2), and have remarkably similar radio-to-optical spectral indexes ($\sim -0.8$), consistent with radio observations where 90% of the jets have $-0.5<\alpha <-0.9$ (Bridle & Perley 1984). This indicates a high degree of homogeneity of all jets, so that it is reasonable to compare their energy budgets with those for radio lobes observed in a large sample of radio sources.
Bulk Jet Speed
==============
Relativistic beaming elegantly explains both jet one-sidedness and superluminal motion, observed routinely on parsec scales (Zensus & Pearson 1987; Readhead 1993), and also on kpc scales in the jet of M87 (Biretta 1999). Assuming the two sides of the jet are intrinsically identical (Rees 1978; Shklovsky 1970; Saslaw & Whittle 1988; Laing 1988), the Lorentz factor of the bulk speed can then be derived (Scheuer & Readhead 1979) from the jet/counterjet luminosity ratio $ J=\left(\frac{1+\beta \cos \theta}{1-\beta \cos \theta}\right)^{(2-\alpha)}$, where $\theta$ is the angle between jet speed and line of sight, and the exponent $(2-\alpha)$ is appropriate for a continuum jet (it would be $3-\alpha$ for discrete emitting blobs; see Lind & Blandford 1985).\
The dependence of $J$ on jet inclination $\theta$ is shown in Figure 1. It is seen that $J$ easily reaches very large values even for modest bulk speed, but for large inclinations it remains finite and quite small even for $\Gamma
\rightarrow \infty$, so that $J$ can be effectively used to constrain $\theta$. From Table 1 the median lower limit of $J$ for optical jets is 40, implying median inclinations ${\raisebox{-3.8pt}{$\;\stackrel{\textstyle < }{\sim }\;$}}55^\circ$ and $\Gamma>1.1$.
![ [**Left:**]{} Jet/counterjet luminosity ratio $J$ for an intrinsically symmetric jet, as function of jet inclination, for different Lorentz factors for the bulk motion. Observed lower limits to $J$ give upper limits on $\theta$ and/or lower limits on $\Gamma$. For example the dashed line correspond to the available lower limits on $J$ for 3C 120. This implies a jet inclination $<35^\circ$ and $\Gamma > 1.7$. [**Right:**]{} Median kinetic power in equipartition for the 14 optical jets versus jet inclination $\theta$ and for several values of $\Gamma$. The horizontal line at $L_{kin}=10^{45}$ erg/s indicates the median power estimated for radio lobes (Rawlings & Saunders 1991). The dashed line is the locus of beaming $\delta = 1$, separating beamed (below) from de-beamed (above) jets. Combining these two constraints, allowed $\theta$ and $\Gamma$ are restricted to the shaded area. ](G_J.ps "fig:"){width="0.48\linewidth"} ![ [**Left:**]{} Jet/counterjet luminosity ratio $J$ for an intrinsically symmetric jet, as function of jet inclination, for different Lorentz factors for the bulk motion. Observed lower limits to $J$ give upper limits on $\theta$ and/or lower limits on $\Gamma$. For example the dashed line correspond to the available lower limits on $J$ for 3C 120. This implies a jet inclination $<35^\circ$ and $\Gamma > 1.7$. [**Right:**]{} Median kinetic power in equipartition for the 14 optical jets versus jet inclination $\theta$ and for several values of $\Gamma$. The horizontal line at $L_{kin}=10^{45}$ erg/s indicates the median power estimated for radio lobes (Rawlings & Saunders 1991). The dashed line is the locus of beaming $\delta = 1$, separating beamed (below) from de-beamed (above) jets. Combining these two constraints, allowed $\theta$ and $\Gamma$ are restricted to the shaded area. ](g_theta.ps "fig:"){width="0.49\linewidth"}
Jet Versus Lobe Kinetic Energy
==============================
The kinetic luminosity depends on the density and average energy of the emitting electrons, as well as the bulk speed of the plasma. These quantities are therefore constrained if the kinetic luminosity can be estimated in an independent way. We require jets to transport enough energy for powering an average radio lobe. Setting all relevant parameters as before (i.e., same low frequency cutoff, $k=0$, and $\phi=1$), a median kinetic power of $<L_{kin}>=10^{45}$ erg/s was found for a large sample of radio galaxies including both weak and powerful radio sources (Rawlings & Saunders 1991). The high frequency cutoff is higher for optical jets than radio lobes, but this should have no effect on the energy budget because for $p<-2$ ($\alpha<-0.5$) the energetics are dominated by the low energy particles. Comparing the [*median*]{} kinetic power of all optical jets with that of radio lobes (Figure 1), it is found that for very low bulk speeds ($\Gamma{\raisebox{-3.8pt}{$\;\stackrel{\textstyle < }{\sim }\;$}}2$), the median kinetic energy for this sample of jets is at least one order of magnitude smaller than needed to power average lobes, independently of inclination angle. Very small inclinations are also excluded, because for small $\theta$ the beaming is so strong that the rest frame luminosity of the jet is very small, and the kinetic energy significantly reduced.\
The maximum speed of the plasma can be loosely constrained if jets with enhanced rather than dimmed emission are preferentially discovered. For any given value of $\theta$, as soon as $\Gamma$ becomes larger than the value defined by $\Gamma-\sqrt{\Gamma^2-1} \cos \theta=1$, the emission is de-amplified, severely reducing our probability of discovering the jet. In this way we derive a “most probable” region, centered on $\theta \sim 20^\circ$ and $\Gamma \sim 7.5$, where the average kinetic energy carried by (optical) jets is fully consistent with the requirement imposed by radio lobes. The conclusion is that jets should be relativistic at kpc scales.
![ [**Left:**]{} Average equipartition magnetic field for the whole sample versus jet inclination for several values of $\Gamma$. The horizontal line at $B=2.8 \times10^{-4}$ G indicates the median magnetic field derived directly from values given in Table 1 without including beaming. [**Upper right**]{}: Kinetic energy distribution of jets (dashed histogram) compared to radio lobes power (open histogram), evaluated assuming equipartition and $\theta=20^\circ$ and $\Gamma=7.5$ for all jets. The median jet kinetic energy is $10^{45}$ erg/s as for radio lobes (Rawlings & Saunders 1991). [**Lower Right**]{}: Distribution of estimated electron diffusion length $ct_{1/2}\Gamma$ for electrons emitting at $10^{15}$ Hz in the observer frame (dashed histogram), compared with the observed deprojected jet lengths (open histogram). The diffusion length was computed assuming equipartition and the same values of $\Gamma$ and $\theta$ as before. ](b_theta.ps "fig:"){width="0.5\linewidth"} ![ [**Left:**]{} Average equipartition magnetic field for the whole sample versus jet inclination for several values of $\Gamma$. The horizontal line at $B=2.8 \times10^{-4}$ G indicates the median magnetic field derived directly from values given in Table 1 without including beaming. [**Upper right**]{}: Kinetic energy distribution of jets (dashed histogram) compared to radio lobes power (open histogram), evaluated assuming equipartition and $\theta=20^\circ$ and $\Gamma=7.5$ for all jets. The median jet kinetic energy is $10^{45}$ erg/s as for radio lobes (Rawlings & Saunders 1991). [**Lower Right**]{}: Distribution of estimated electron diffusion length $ct_{1/2}\Gamma$ for electrons emitting at $10^{15}$ Hz in the observer frame (dashed histogram), compared with the observed deprojected jet lengths (open histogram). The diffusion length was computed assuming equipartition and the same values of $\Gamma$ and $\theta$ as before. ](figure5.ps "fig:"){width="0.245\linewidth"}
Conclusions
===========
Comparing the kinetic power (as derived assuming equipartition) for all known optical jets with that for typical radio lobes suggests the presence of relativistic bulk motion of the emitting plasma at $kiloparsec$ scales. In a non-relativistic scenario the estimated kinetic jet luminosity is at least one order of magnitude less than needed to power the lobes.\
This has important consequences. Indeed, at face value, the data in Table 1 imply strong magnetic fields and short electron lifetimes, leading to the conclusion that electron reacceleration is necessary to explain the optical emission on kpc scales (e.g., Meisenheimer 1996). The analysis here points to an opposite possibility. Indeed, the constraints on the kinetic luminosity allow $\Gamma$ and $\theta$ to lie within a “most probable” region (Fig. 1), centered near $\Gamma \sim 7.5$, corresponding to a highly relativistic bulk speed. Because of relativistic beaming, the rest-frame magnetic field is reduced (Table 1), and the electron lifetimes are lengthened because of the lower energy losses and time dilation. In these conditions the de-projected length $l/\sin
\theta$ of the jets is fully consistent with the electron diffusion length $ct_{1/2}\Gamma$, without the need for reacceleration (Figure 2), explaining the very nearly uniform $\alpha_{RO}$ observed in the jets of M87 (Sparks 1996) and PKS 0521-365 (Scarpa 1999).
It is a pleasure to thanks G. Ghisellini, L. Maraschi, and F. Macchetto for helpful and encouraging comments. Support for this work was provided by NASA through grant GO-06363.01-95A from the Space Telescope Science Institute, which is operated by AURA, Inc., under NASA contract NAS 5-26555.
Akujor C.E., Lüdke E., Browne I.W.A., 1994 A&AS 105, 247 Bachall J.N., Kirhakos S., Schneider D.P., 1995, ApJ 452, L91 Baum S.A., O’Dea C.P., Giovannini G. 1997, ApJ 483, 178 Biretta J.A., Perlman E., Sparks W.B. & Macchetto F. 1999, AAS meeting N. 193, 07.09 Blandford R.D. & Königl A. 1979 ApJ 332, 34 Bogers W.J., Hes R., Barthel P.D. & Zensus J.A. 1994, A&AS 105, 91 Bridle A.H. & Perley R.A. 1984, ARA&A, 22, 319 Burns J.O., Basart J.P., De Young D.S. & Ghiglia D.C. 1984, ApJ 283, 515 Celotti A. & Fabian A.C. 1993, MNRAS 264, 228 Crane P., Peletier R., Baxter D. 1993, ApJ 402, L37 Davis R.J., Unwin S.C., Muxlow T.W.B. 1991, Nature 354, 374 Dey A. & van Breugel 1994, AJ 107, 1977 Fraix-Burnet D. 1997, MNRAS 284, 911 Fraix-Burnet D., Nieto J.L., Leliévre G., 1989 ApJ 336, 121 Fraix-Burnet D., Golombek D. & Macchetto D. 1991, AJ 102, 562 Hjorth J., Vestergaard M., So rensen A.N. & Grundahl F. 1995, ApJ 452, L17 Keel W.C. 1986 ApJ 302, 296 Laing R.A. 1988, Nature 331, 149 Laing R. 1989, in “Hot spots in extragalactic radio sources”, Springer-Verlag, p. 27 Laurent-Muehleisen S.A., Kollgaard R.I., Moellenbrock G.A. & Feigelson E.D. 1993, AJ 106, 875 Leahy J.P., Black A.R.S. & Dennet-Thorpe J. 1997, MNRAS 291, 20L Lind, K.R. & Blandford R.D. 1985, ApJ 295, 358 Macchetto F., Albrecht R., Barbieri C., 1991 ApJ 369, L55 Martel A.R., Sparks W.B., Macchetto F., 1998 ApJ 496, 203 Meisenheimer K., Röser H.-J. & Schlötelburg M. 1996, A&A 307, 61 Nilsson K., Heidt J., Pursimo T. 1997 ApJ 484, L107 Pacholcsyk A.G. 1970, “Radio Astrophysics”, printed by W.H. Freeman and Company, San Francisco. Rawlings S & Saunders R. 1991, Nature 349, 138 Readhead A.C.S. 1993, in [*Sub-Arcsecond Radio Astronomy*]{}, Cambridge: Cambridge Univ. Press, pp. 173 Rees M. J 1978 MNRAS 184, 61 Ridgway S.E. & Stockton A. 1997, AJ 114, 511 Saikia D.J., Subrahmanya C.R., Patnaik A.R. 1986, MNRAS 219, 545 Saikia D.J., Junor W., Cornwell T.J. 1990, MNRAS 245, 408 Saslaw W.C. & Whittle M. 1988, ApJ 325, 142 Scarpa R., Urry C.M., Falomo R. & Treves A. 1999, ApJ 526, in press Scarrot S.M. & Warren-Smith R.F. 1987, MNRAS 228, 35p Scheuer P.A.G. & Readhead A.C.S. 1979, Nature 277, 182 Scheuer P.A.G. 1974, MNRAS 166, 513 Shklovsky J. 1970, Nature 228, 1174 Sparks W.B., Biretta J.A. & Macchetto F. 1996, ApJ 473, 254 Sparks W.B., Golombek D., Baum S.A. 1995, ApJ 450, L55 Stiavelli M., Biretta J., Møller P. & Zeilinger W.W. 1992 Nature 355, 802 van Breugel W.J.M., Fanti C., Fanti R., 1992, A&A 256, 56 Walker R.C., Bensons J.M. & Unwin S.C. 1978, in [ *Superluminal Radio Sources*]{}, ed. J.A. Zensus & T.J. Pearson, pg. 48 ApJ 278, 521 Wrobel J.M. & Lind K.R. 1990, ApJ 348, 135 Zensus J.A. & Pearson T.J. 1987, in [*Superluminal Radio Sources*]{}, Cambridge Univ. Press, pp. 1
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Direct imaging of extra-solar planets is now a reality, especially with the deployment and commissioning of the first generation of specialized ground-based instruments such as the Gemini Planet Imager and SPHERE. These systems will allow detection of Jupiter-like planets $10^7$ times fainter than their host star. Obtaining this contrast level and beyond requires the combination of a coronagraph to suppress light coming from the host star and a wavefront control system including a deformable mirror (DM) to remove residual starlight (speckles) created by the imperfections of telescope. However, all these current and future systems focus on detecting faint planets around single host stars, while several targets or planet candidates are located around nearby binary stars such as our neighboring star Alpha Centauri. Here, we present a method to simultaneously correct aberrations and diffraction of light coming from the target star as well as its companion star in order to reveal planets orbiting the target star. This method works even if the companion star is outside the control region of the DM (beyond its half-Nyquist frequency), by taking advantage of aliasing effects.'
author:
- |
Sandrine J. Thomas and Ruslan Belikov and Eduardo Bendek Large Synoptic Survey Telescope, 950N Cherry Av, Tucson AZ 85719, USA;\
Nasa Ames Research Center, Moffett Field, CA 94035, USA
bibliography:
- 'MSSNWC\_spie.bib'
title: 'A method to directly image exoplanets in multi-star systems such as Alpha-Centauri'
---
INTRODUCTION {#sec:intro}
============
The exoplanets field is rapidly expanding with the success of the Kepler mission (Burke et al. 2014 [@Burke14] and references therein) and the emergence of direct imaging ground based instruments (GPI [@Macintosh14], SPHERE [@Beuzit08], SCExAO [@Guyon10], P1640 [@Hinkley08]). One of the most exciting prospects for future telescopes is finding other Earth analogues in our galaxy or solar neighborhood and ultimately detect life on them. The Kepler space telescope has already revealed that roughly 22% of stars have planets between 1 and 2 Earth radii in their habitable zone [@Batalha14].
However, most Sun-like stars are found in multi-star systems or visual binaries and conventional coronagraphs and wavefront control systems are designed for single-star systems, which means that they only suppress the light of the one on-axis star. Because of the perceived challenges of suppressing starlight of more than one star, multi-star systems are usually excluded from mission target lists, despite being more numerous (at least among Sun-like or earlier star systems). Therefore, there is a need for a technique enabling the direct imaging of planetary systems and disks around multi-star systems, which we present here. In particular, our new technique potentially allows the detection of biomarkers on Earth-like planets around our nearest-neighbor star, Alpha Centauri, with a small and cheap space telescope, potentially decades sooner than a large space coronagraph could do the same around a single star system (see Belikov et al. 2015 [@Belikov15] and Bendek et al. 2015 [@Bendek15]).
![\[fig:issues1\] Illustration of the challenges involved with creating a dark zone around a target belonging to a double star system. Left: simplified case where the on-axis star (close to the red region of interest) has been assumed to be completely suppressed, and only the light from the off-axis star (on the opposite end) is considered. Oftentimes, the region of interest around the on-axis star lies beyond the Nyquist-limited outer working angle of the deformable mirror. Right: zoomed in version of the left image, centered on the on-axis star. ](Figure7.jpg "fig:") ![\[fig:issues1\] Illustration of the challenges involved with creating a dark zone around a target belonging to a double star system. Left: simplified case where the on-axis star (close to the red region of interest) has been assumed to be completely suppressed, and only the light from the off-axis star (on the opposite end) is considered. Oftentimes, the region of interest around the on-axis star lies beyond the Nyquist-limited outer working angle of the deformable mirror. Right: zoomed in version of the left image, centered on the on-axis star. ](Figure9-eps-converted-to.pdf "fig:")
As described in previous papers [@Thomas14], [@Thomas15], we identified three main challenges associated with observing double-star (or multi-star) systems –illustrated in Figures \[fig:issues1\] and \[fig:issues2\] . (a) Even if we assume the on-axis star is perfectly suppressed or otherwise ignored, and only consider the off-axis star, the off-axis star will often lie beyond the outer working angle (Half-Nyquist frequency) of the deformable mirror (especially for nearby systems such as Alpha Centauri). (b) Light from different stars is mutually incoherent and must be suppressed independently but simultaneously for each star. (c) Demonstrating the combination of the first two challenges, and doing so in broadband light.
![\[fig:issues2\] Illustration of the challenges involved with creating a dark zone around a target belonging to a double star system. It shows a combination of light from both stars, requiring independent suppression of each, using conventional (sub-Nyquist) wavefront control for the on-axis star (A) and super-Nyquist wavefornt control for the off-axis star (B).](Figure5.jpg "fig:"){height="7cm"}\
In Thomas et al. 2015 [@Thomas15], we presented a technique called “Super-Nyquist Wavefront Control” (SNWC) which is a solution to the first challenge (limited field of view, or FOV, correctable by a given deformable mirror), and demonstrated computer simulations of SNWC. We also outlined proposed solutions to the other two challenges, which we call, respectively, Multi-Star Wavefront Control (MSWC), and Multi-Star Super-Nyquist Wavefront Control (MSSNWC).
In the present paper, we go a step further and demonstrate MSWC and MSSNWC with computer simulations. In particular, we present the results of the double star simulation for two different configurations: (a) the separation of the two binaries is in within the correctable FOV of the deformable mirror (b) the separation of the two binaries is larger than the correctable FOV of the deformable mirror. Both configurations come with different challenges. If the binary is tight, the deformable mirror does not have to be exercised in a non-standard way but the flux of the secondary target is high and requires more stroke to cancel. If the separation of the binaries is large, the flux contaminating the dark region is lower but the dark region is in the non-standard controllable zone of the mirror. Section \[sec:MSWC\] describes the Multi-Star Wavefront Control (MSWC) theory and results. Section \[sec:MSSNWC\] describes the Multi-star Super-Nyquist Wavefront Control (MSSNWC) theory and results. Both configurations are presented in monochromatic light in this paper.
Multi-Star Wavefront Control (MSWC) {#sec:MSWC}
===================================
Theory
------
In a well-baffled coronagraphic system, the light from a star on the focal plane can be thought of as consisting of two separate components: (a) diffraction such as Airy rings from the telescope and all other optical components (which are predictable a priori, fixed in time in space, and are fundamentally due to the physics of light); (b) aberrations, or “speckles” from these components (which are random and/or varying in time, and are due to engineering limitations rather than physics). In general, coronagraphs suppress diffraction but not aberrations. An active control system is necessary to suppress aberrations. This same active control is also capable of suppressing diffraction to some degree. Diffraction usually dominates aberrations very close to the star (Airy rings are brighter than speckles), which implies that a coronagraph is useful in that case. However, beyond a certain angle from the star (usually about $10 \lambda/D$), diffraction becomes faint enough that wavefront control systems can usually suppress it without the need for a coronagraph. Furthermore, aberrations often decay with angle slower than diffraction does and thus become dominant at larger angles. In other words, a wavefront control system is both necessary and sufficient to suppress starlight at larger angles from the star. Therefore, our solution is based on wavefront control, rather than coronagraphic suppression of the off-axis star. Note that a simple mask blocking a second star can still be useful as a baffle if the rest of the system is insufficiently baffled.
![\[fig:MSWC\] Principle of Multi-Star Wavefront Control, which uses conventional single-star wavefront control hardware (for this example one 32$\times$32 DM), and a special algorithm. Two stars (black A and grey B) are separated by $16 \lambda/D$. Sub-Nyquist wavefront control regions are shown with respect to star A (left), star B (center), and both stars together (A and B). Vertical dashed lines divide regions showing which DM modes control which star in that region. In regions I or II, independent suppression of both stars can be achieved. In general, one can always find such regions regardless of star separation, the only limitation is the controllable region in a binary system mush be half of that in the single-star system.](Figure13.pdf){height="5cm"}
The key principle of most high-contrast wavefront control methods existing today is to use a Deformable Mirror to diffract a little bit of starlight from the main beam into the science plane region of interest, in such a way as to cause destructive interference with the unwanted speckles initially present there. This of course requires that starlight interfere coherently with the initial speckles in the region of interest. This will always be the case for a single-star target (at least assuming a point-like star and a well-performing system with no technical complications such as poor baffling, unresolved jitter, etc.) However, in the presence of two or more stars, light in any region of interest in general consists of an incoherent combination (both diffraction and aberrations) from all stars. Therefore, in order to remove light from such a region of interest, we must independently and simultaneously suppress each star’s speckles by destructively interfering them with only that star’s light. In other words, any multi-star wavefront control system must have the property that it removes light for each star independently from the other star(s).
It turns out that this is possible with existing wavefront control hardware and only requires a modification of the software algorithms, albeit a cost of reducing the size of the maximum controllable region of interest. The basic principle is illustrated in Figure \[fig:MSWC\] and is based on separating DM modes into two independent sets in a way that each set primarily affects the light of one star in the region of interest but not the other. For this example, we consider the case of two stars (black dot A and grey dot B) separated by 16 $\lambda/D$ in the x direction. Figure \[fig:MSWC\] (left) shows the full sub-Nyquist control region of a 32 $\times$ 32 deformable mirror out to +/- 16 $\lambda/D$ in both coordinates with respect to star A. Consider two sets of DM modes: the “low-order x” and “high-order x” modes, corresponding to ones whose x-direction spatial frequency content is limited to 0-8 and 8-16 cycles per aperture, respectively. The low order modes primarily affect the regions between 0-8 $\lambda/D$ in the x-direction with respect to each star. Similarly, the high order modes primarily affect the 8-16 $\lambda/D$ region with respect to each star, as shown in Figure \[fig:MSWC\], (left and center, showing the situation with respect to the A and B stars, respectively). Figure \[fig:MSWC\] (right) shows a diagram of the superposition of the left and center images. In the region labeled “I”, the low-order modes primarily affect speckles of the B star and the high order modes primarily affect the speckles of the A star, so that in that region speckles from both stars are independently controllable with a single DM. A similar situation occurs in region II. Therefore, as long as we limit our region of interest to either I or II, we can achieve independent control of the speckles from both stars.
It the stars are separated by a distance other than 16 $\lambda/D$, similar regions can be found (as long as the separation is sub-Nyquist – the super-Nyquist case is treated below). Such regions will have more complicated shapes but will always have maximum total area equal to half of the maximum single-star independently controllable region. This is because the number of degrees of freedom on the DM is fixed, so that doubling the number of stars must necessarily halve the controllable region. We can also generalize this to systems with more than 2 stars – in that case we can also find regions where light from all stars can be independently controlled with a single DM (or 2 if control of both phase and amplitude errors is desired, just as in the case of single-star wavefront control). In general, the shape of each such region may be complicated and sometimes not even convex or connected, but its total area will in general be 1/N less than the area of a single-star region for that wavefront control system. (This is the price for multi-star wavefront control.)
So far we have only talked about being able to independently control speckles from multiple stars, but mere control does not necessarily imply the ability to find a DM setting where the speckles are actually suppressed. We can demonstrate this solution exists as follows. Most single-star wavefront control algorithms are based on solving a system of linear equations relating DM modes and starlight electric field in the region of interest. As long as we can measure the electric field of each star separately, we can then combine the linear sets of equations for each star into a larger set of linear equations, and simply solve that set. If each star’s system of equations is solvable by itself, and each star uses different DM modes, then the combined system must also be solvable.
A remaining component is the “estimation problem”, i.e. the problem of estimating each star’s electric field in the region of interest separately (which is one of the necessary parameters in the system of equations for control in the above paragraph). This can be done in a manner analogous to the above paragraph: single-star wavefront control methods usually use the deformable mirror to diffract some trial “probe” light in the region of interest. This modulates the initial speckles there, and this information can be used to estimate their electric fiend by solving another system of linear equations relating the electric field and DM probe shapes. Thus, we have a system of linear equations for each star, which we can combine to yield a solution to the electric field of each star in the region of interest. As long as each of the single-star systems is solvable by itself, and as long as different DM modes affect primarily different stars (i.e. are not degenerate), then the combined system must also be solvable.
Results
-------
The simulation process is similar to the one described in Thomas et al, 2015 [@Thomas15]. In order to create a dark zone around one of the two components of the binary, we consider the correction part of an Electric Field Conjugation algorithm (EFC) [@Giveon07] coded in Matlab. We assume that we know the wavefront to be corrected and therefore do not consider the estimation part of EFC as mentioned in the previous section. We added 10nm rms worth of aberrations, following a power law with a coefficient equal to -2. The aberrations were introduced in the pupil plane. In addition, we focus in this paper on monochromatic light. However, the goal is to run the simulations with a 10% bandwidth, which is computationally expensive. The separation of the star is set to 10 $\lambda$/d (sub-Nyquist) and the region of interest is 2-6 $\lambda$/D.
![\[fig:MSWC\_res\] Preliminary proof-of-principle simulation of Multi-Star Wavefront Control, before (left) and after (right) correction. The deformable mirror is a 32 $\times$ 32. The contrast achieved over a 4 $\times$ 4 area is 5.2$\times 10^{-8}$. A colored version is available electronically.](MSWC1.pdf "fig:"){height="6cm"} ![\[fig:MSWC\_res\] Preliminary proof-of-principle simulation of Multi-Star Wavefront Control, before (left) and after (right) correction. The deformable mirror is a 32 $\times$ 32. The contrast achieved over a 4 $\times$ 4 area is 5.2$\times 10^{-8}$. A colored version is available electronically.](MSWC2.pdf "fig:"){height="6cm"}
Figure \[fig:MSWC\_res\] shows the results. The contrast was improved from 1.3$\times 10^{-4}$ to 5.2$\times 10^{-8}$. Note that the performance here is limited by the on-axis diffraction rings and better results are achieved when using a coronagraph to block the on-axis star, which we did for the next section.
Combining SNWC and MSWC: MSSNWC {#sec:MSSNWC}
================================
Theory
------
In the previous section, we described how MSWC can remove light for the case that the separation between two stars is sub-Nyquist. In Thomas et al 2014 [@Thomas14] and Thomas et al 2015 [@Thomas15], we demonstrated SNWC, a method to suppress starlight from a star beyond the DM Nyquist frequency. In this section we demonstrate how these two techniques can be combined theoretically and then demonstrate an example by a computer simulation.
Figure \[fig:MSSNWC\_solution\] shows a field of view containing two stars: a coronagraphically suppressed star A (centered inside the black square on the right), and star B (centered in black square on the left). The back squares represent the DM sub-Nyquist region, and the small “DZ” rectangle represents the region of interest, or “dark zone” in the planetary system around star A. The dark zone consists of contribution of light from both stars. If star B was in the sub-Nyquist region of star A, we could just apply MSWC to remove both stars’ light from the DZ, but star B is too far away for MSWC. In Thomas et al. 2014 and 2015 [@Thomas14], [@Thomas15], we described how SNWC is essentially equivalent to conventional (sub-Nyquist) wavefront control, but applied with respect to one of the diffraction orders of the star caused by a mild grating in the pupil plane. The segmentation of segmented telescopes or segmented DMs already acts as an appropriate grating, or the “quilting” pattern left on most DMs as part of their fabrication process can suffice. If neither of those are present, a mild grating [@Guyon12], [@Bendek13] can be deliberately installed in the system. In any case, if this grating generates a (faint) diffracted copy of star B inside the sub-Nyquist region of star A, then we can now use this diffracted copy of star B as its “proxy” and apply MSWC on this proxy and star A. This essentially reduces the super-Nyquist multi-star problem to the already solved MSWC, in exactly the same way as SNWC reduces the single-star super-Nyquist problem to the single-star sub-Nyquist problem.
Note that Figure \[fig:MSSNWC\_solution\] shows a monochromatic rather than broadband simulation. In broadband light, the diffracted copy of star B will be chromatically elongated along a line pointing to star B, which would seem to pose a complication and unique challenge in broadband. However, all other light from star B (speckles, other DM modes) will also be chromatically elongated in a similar fashion. Thus, the elongation of the proxy star should help, rather than hinder, broadband control in that region. Indeed, as we already showed in the computer simulations in [@Thomas15], SNWC works fairly fell in broadband, at least under certain assumptions. Although we have not yet demonstrated broadband operation in general, the above arguments suggest that broadband challenges are entirely a function of stellocentric angle and not whether super-Nyquist or sub-Nyquist control is used. For example, SNWC at 24 $\lambda/D$ with a 32 $\times$ 32 DM (where the sub-Nyquist boundary is 16 $\lambda/D$) should not cause any unique broadband challenges beyond those already associated with performing conventional (sub-Nyquist) wavefront control at 24 $\lambda$/D with a 64 $\times$ 64 DM. In the same way, we expect that SNMSWC will not cause any unique broadband challenges.
![\[fig:MSSNWC\_solution\] Schematic of the MSSNWC scenario: 2 stars separated by 100 $\lambda$/d. The on-axis star A is now blocked by a simple apodized Lyot coronagraph. The star B is replicated due to the grid off the DM. The grid is due to the 32 $\times$ 32 actuators from the DM.](Figure14.pdf){height="12cm"}
Results
-------
For this section, we simulated a portion of an Airy pattern resulting from a normal circular aperture at 25 $\lambda / d$. For this preliminary simulation, we use an Apodized Lyot Coronagraph to block the light coming from the center star and we used a grid of dots with a frequency equal to 32 $\lambda / d$, which corresponds to the regular pattern that MEMS (Micro-Electro-Mechanical system) create. Because they have a fixed number of actuators the frequency that the MEMS can create has to be a multiple of the number of actuators. The simulated dots are the size of a pixel and therefore depend on the resolution of the simulation. The amount of aberrations included in the pupil plane was 10nm with a power law of -2. The preliminary results presented here are in monochromatic only. The ultimate goal is to repeat the experiment in 10% band but the computation was very expensive in order to get a goal sample of the bandwidth. We moved the code over to the super computer facilities at NASA Ames and will present results in a future paper.
![\[fig:MSSNWC\] Results of the MSSNWC method, combining the SNWC and the MSWC. The separation of the two stars is 25 $lambda$/d and is located in the upper left corner in the above image. The on-axis star is blocked by an apodized Lyot coronagraph and the grid has a dots frequency of 32. A colored version is available electronically.](MSSNWC.jpg){height="7cm"}
The contrast achieved here is $2 \times10^{-8}$ with a 6 $\lambda/d \times 6 \lambda/d$ region. If we adjust the dark zone region to with a 4 $\lambda/d \times 4 \lambda/d$ region, the contrast reached is less than $10^{-8}$. Note that these results are preliminary.
Conclusion and Applications
===========================
We presented the basic principles and first computer simulations of methods that enable high-contrast imaging of exoplanets and disks in multi-star systems. This is particularly important because most Sun-like and earlier stars are in multi-star systems. Of special note is Alpha Centauri, which is not only the closest star system to Earth, but also 2.5x closer than the next closest non-M-dwarf star. It would be the most favorable target (by a large margin) for any direct imaging telescope capable of suppressing light from a binary.
Our methods are called Multi-Star Wavefront Control (MSWC) and Multi-Star Super Nyquist Wavefront Control (MSSNWC), depending on whether the separation between the two stars is sub-Nyquist or super-Nyquist. Both of them rely on a single-star coronagraph and suppress the other star(s) with wavefront control only because the dominant source of noise from the other stars is aberrations and not diffraction. Both methods use existing, high technology readiness single-star imaging systems with no required changes, and only use special software algorithms. Suppression of multiple stars is achieved simply by using different modes of the deformable mirror on different stars. This technique requires reducing the maximum area of the high-contrast region of interest, which can be enlarged by using a DM with more actuators.
This work was supported in part by the National Aeronautics and Space Administration’s APRA (Astrophysics Research and Analysis) program through solicitation NNH13ZDA001N-APRA at NASA’s Science Mission Directorate, and NASA Ames Research Center. It was carried out at the NASA Ames Research Center. Any opinions, findings, and conclusions or recommendations expressed in this article are those of the authors and do not necessarily reflect the views of the National Aeronautics and Space Administration.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We use a statistical approach to determine the relationship between the stellar masses of galaxies and the masses of the dark matter halos in which they reside. We obtain a parameterized stellar-to-halo mass (SHM) relation by populating halos and subhalos in an $N$-body simulation with galaxies and requiring that the observed stellar mass function be reproduced. We find good agreement with constraints from galaxy-galaxy lensing and predictions of semi-analytic models. Using this mapping, and the positions of the halos and subhalos obtained from the simulation, we find that our model predictions for the galaxy two-point correlation function (CF) as a function of stellar mass are in excellent agreement with the observed clustering properties in the SDSS at $z=0$. We show that the clustering data do not provide additional strong constraints on the SHM function and conclude that our model can therefore predict clustering as a function of stellar mass. We compute the conditional mass function, which yields the average number of galaxies with stellar masses in the range $m\pm{\rm d}m/2$ that reside in a halo of mass $M$. We study the redshift dependence of the SHM relation and show that, for low mass halos, the SHM ratio is lower at higher redshift. The derived SHM relation is used to predict the stellar mass dependent galaxy CF and bias at high redshift. Our model predicts that not only are massive galaxies more biased than low mass ones at all redshifts, but the bias increases more rapidly with increasing redshift for massive galaxies than for low mass ones. We present convenient fitting functions for the SHM relation as a function of redshift, the conditional mass function, and the bias as a function of stellar mass and redshift.'
author:
- 'Benjamin P. Moster, Rachel S. Somerville, Christian Maulbetsch, Frank C. van den Bosch, Andrea V. Macciò, Thorsten Naab, and Ludwig Oser'
title: Constraints on the relationship between stellar mass and halo mass at low and high redshift
---
Introduction {#s:introduction}
============
In the standard Cold Dark Matter (CDM) paradigm, the formation of galaxies is driven by the growth of the large-scale structure of the Universe and the formation of dark matter halos. Galaxies form by the cooling and condensation of gas in the centers of the potential wells of extended virialized dark matter halos [@whiterees78; @fallefstathiou1980; @blumenthal1984]. In this picture, galaxy properties, such as luminosity or stellar mass, are expected to be tightly coupled to the depth of the halo potential and thus to the halo mass.
There are various different approaches to link the properties of galaxies to those of their halos. A first method attempts to derive the halo properties from the properties of its galaxy population using e.g. galaxy kinematics [@erickson1987; @zaritsky1993; @carlberg1996; @more2009a; @more2009b], gravitational lensing [@mandelbaum05; @mandelbaum06; @cacciato2008], or X-ray studies [@lin2003; @linmohr2004].
A second approach is to attempt to model the physics that shapes galaxy formation *ab initio* using either large numerical simulations including both gas and dark matter [@katz1996; @springel2003] or semi-analytic models (SAMs) of galaxy formation [e.g. @kauffmann1993; @cole1994; @somerville1999]. In “hybrid” SAMs [e.g. @croton06; @bower2006], dark matter “merger trees” are extracted from a dark matter only N-body simulation, and gas processes are treated with semi-analytic recipes. An advantage of this method is that high-resolution N-body simulations can track the evolution of individual subhalos [@klypin1999; @springel01] and thus provide the precise positions and velocities of galaxies within a halo. However, many of the physical processes involved in galaxy formation (such as star formation and various kinds of feedback) are still not well understood, and in many cases simulations are not able to reproduce observed quantities with high accuracy.
With the accumulation of data from large galaxy surveys over the last decade, a third method has been developed, which links galaxies to halos using a statistical approach. The Halo Occupation Distribution (HOD) formalism specifies the probability distribution for a halo of mass $M$ to harbour $N$ galaxies with certain intrinsic properties, such as luminosity, color, or type [e.g. @peacock2000; @seljak2000; @white2001; @berlind2002]. More complex formulations of this kind of modelling, such as the conditional luminosity function (CLF) formalism [@yang03; @vdb2003; @yang04] have extended the HOD approach. These methods have the advantage that they do not rely on assumptions about the (poorly understood) physical processes that drive galaxy formation. In this way, it is possible to constrain the relationship between galaxy and halo properties (and thus, indirectly, the underlying physics), and to construct mock catalogs that reproduce in detail a desired observational quantity (such as the luminosity function). One disadvantage of the classical HOD approach was that one had to make assumptions about the distribution of positions and velocities of galaxies within their host halos. In addition, the results of the HOD modelling can be difficult to interpret in terms of the underlying physics of galaxy formation.
In recent years, HOD models have been introduced that make use of information about the positions, velocities and masses of halos and subhalos extracted from a dissipationless N-body simulation. The (sub)halo mass is then empirically linked to galaxy properties by requiring that a statistical observational quantity (e.g. galaxy luminosity function and/or galaxy two-point-correlation-function) is reproduced. This is either done by assuming parameterized functions to relate galaxy properties (such as luminosity) to halo mass or by assuming a non-parametric monotonic relation. It has been shown that these simple models reproduce galaxy clustering as a function of luminosity over a wide range in redshift [@kravtsov2004; @tasitsiomi2004; @tinker2005; @valeostriker06; @conroy06; @shankar2006; @wang06; @marin2008].
Observationally, it is well known that galaxy clustering is a function of spatial scale, galaxy properties (such as luminosity and type), and redshift. Luminous (massive) galaxies are more strongly clustered than less luminous (less massive) galaxies [@norberg2001; @norberg2002; @zehavi2002; @zehavi2005; @li06]. One can split the galaxy two-point correlation function (2PCF) into two separate parts: the one-halo and the two-halo terms. The one-halo term, which dominates on small scales, depends strongly on the galaxy distribution within the halo as well as the details of the HOD. The two-halo term, which dominates on scales that are much larger than a typical halo, is proportional to the auto-correlation of the halo population. In general the two terms are not expected to combine to produce a featureless power-law, but generally show a break or dip at the scale where the transition from the one-halo to the two-halo term occurs [@zehavi2004].
The extensive multi-wavelength spectrophotometric information that is now available for large numbers of galaxies allows us to estimate physical parameters of galaxies, such as stellar masses, instead of relying on observational properties such as magnitudes [@bell2001; @kauffmann03; @panter2004]. These estimates can even be obtained — with a proper measure of caution — for high redshift galaxies. Stellar mass estimates have been presented in the literature for galaxies up to redshifts as high as $z\sim 6$ [@yan2006; @eyles2007], and stellar mass function estimates have been presented up to $z\sim 5$ [@drory2005; @fontana2006; @elsner2008]. The goal of our paper is to develop a “Conditional Stellar Mass Function” (CMF) formalism, which is the stellar mass analog of the CLF. The CMF yields the average number of galaxies with stellar masses in the range $m\pm{\rm d}m$ as a function of the host halo mass $M$ and can be regarded as the stellar mass function (SMF) for halos of mass $M$. We apply this formalism at low redshift and up to the highest redshifts where reliable observational stellar mass estimates are available ($0.1 \lesssim z \lesssim 4$). In this way, we derive a parameterized relationship between dark matter halo mass and galaxy mass as a function of redshift.
Using a parameterized relationship has several advantages. First, it provides a convenient way for other researchers to make use of our results and obtain an expression for stellar mass as a function of halo mass. Second, it is straightforward to include scatter in the relation, which is physically more realistic: one just has to choose a number drawn from an assumed random distribution and add that to the average relation. Finally, it is straightforward to treat central and satellite galaxies separately and assume different relations between stellar and halo mass for those populations. However, here we make the assumption that both populations follow the same relation, which has consequences for the clustering predictions of our model.
Using the CMF derived *only* from constraints from the observed SMF, we compute the predicted (projected) galaxy CF at $z\sim 0$ as a function of stellar mass, and find good agreement with the observational results of @li06. Furthermore, we show that assuming central and satellite galaxies follow the same relation between stellar and halo mass, adding the clustering constraints does not tighten the constraints on our model parameters; i.e., any model that satisfies the mass function constraints will produce the correct clustering. Based on this result, we use our redshift-dependent CMF results to *predict* the clustering as a function of stellar mass and redshift. To date, observational measurements of clustering as a function of stellar mass have only been published for $z\lta 1$ [@meneux2008; @meneux2009]. We show that our model predictions agree very well with these measurements. Very soon it will be possible to test our predictions for redshifts beyond $z=1$ with the results from deep wide-field surveys (e.g. MUSYC, UKIDDS, etc). We again present convenient fitting functions for the galaxy bias as a function of both stellar mass and redshift. In a companion paper we will employ our estimates of galaxy bias in order to compute the “cosmic variance”, the uncertainty in observational estimates of the volume density of galaxies arising from the underlying large-scale density fluctuations.
This paper is organized as follows: in section \[s:simulation\] we describe the $N$-body simulation, the halo finding algorithm that was used to obtain a halo catalogue and the treatment of ‘orphaned’ galaxies. Section \[s:galaxies\] specifies our model: we motivate the form of the stellar-to-halo mass (SHM) relation and constrain it by requiring that the observed SMF is reproduced. The clustering properties of galaxies are then inferred from those of the halo population. We discuss the meaning of the parameters of the SHM relation and demonstrate that clustering puts only weak constraints on them. In section \[s:cmf\] we introduce the CMF, which describes how halos are occupied by galaxies, and compute the occupation numbers. Section \[s:comparison\] gives a comparison between our results and several other models and observations. In section \[s:redshift\] we apply our method to higher redshifts and determine the redshift dependence of the SHM relation. We make predictions of the stellar mass dependent galaxy CF at higher redshift which we use to compute the galaxy bias. Finally, we summarize our methods and conclusions in section \[s:conclusions\].
Throughout this paper we assume a $\Lambda$CDM cosmology with ($\Omega_m$,$\Omega_{\Lambda}$,$h$,$\sigma_8$,$n$) = ($0.26$,$0.74$,$0.72$,$0.77$,$0.95$). We employ a @kroupa01 initial mass function (IMF) and convert all stellar masses to this IMF. 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.
The simulation and halo catalogs {#s:simulation}
================================
High-resolution dissipationless N-body simulations have shown that distinct halos[^1] contain subhalos which orbit within the potential of their host halo. These subhalos were distinct halos in the past, and entered the larger halo via merging during the process of hierarchical assembly. We will refer to the galaxy at the center of a distinct halo as a central galaxy, and the galaxies within subhalos as “satellites”, and we will use the term ‘halo’ to refer to the distinct halo for central galaxies and to the subhalo in which the galaxy originally formed for satellite galaxies.
[*A*b initio]{} models of galaxy formation predict that the stellar mass of a galaxy is tightly correlated with the depth of the potential well of the halo in which it formed. For distinct halos, the relevant mass is the virial mass at the time of observation. Subhalos, however, lose mass while orbiting in a larger system as their outer regions are tidally stripped. Stars are centrally concentrated and more tightly bound than the dark matter, however, and so the stellar mass of a galaxy which is accreted by a larger system probably changes only slightly until most of the dark matter has been stripped off. Therefore the subhalo mass at the time of observation is probably not a good tracer for the potential well that shaped the galaxy properties. A better tracer is the subhalo mass at the time that it was accreted by the host halo, or its maxmimum mass over its history[^2]. This was first proposed by @conroy06.
The population of dark matter halos used in this work is drawn from an $N$-body simulation run with the simulation code [GADGET]{}-2 [@springel05a] on a SGI AltixII at the University Observatory Munich. The cosmological parameters of the simulation are chosen to match results from [WMAP]{}-3 [@spergel06] for a flat $\Lambda$CDM cosmological model: $\Omega_m=0.26$, $\Omega_{\Lambda}=0.74$, $h=H_0/(100$ km s$^{-1}$ Mpc$^{-1})=0.72$, $\sigma_8=0.77$ and $n=0.95$. The initial conditions were generated using the GRAFIC software package [@bertschinger2001]. The simulation was done in a periodic box with side length $100$ Mpc, and contains $512^3$ particles with a particle mass of $2.8\times 10^8{\!\,{\rm M}_\odot}$ and a force softening of $3.5$ kpc.
Dark matter halos are identified in the simulation using a friends-of-friends (FoF) halo finder. Substructures inside the FoF groups are then identified using the [SUBFIND]{} code described in [@springel01]. For the most massive subgroup in a FoF group the virial radius and mass are determined with a spherical overdensity criterion: the density inside a sphere centered on the most bound particle is required to be greater than or equal to the value predicted by the spherical collapse model for a tophat perturbation in a $\Lambda$CDM cosmology [@bryan98]. As discussed above, for subhalos we use the maximum mass over its past history, which is typically the mass when the halo was last a distinct halo and did not yet overlap with its later host. Merger trees were constructed out of the halo catalogs at 94 time-steps, equally spaced in expansion factor ($\Delta a=0.01$), based on the particle overlap of halos at different time-steps.
Due to the finite mass resolution of the simulation ($M_{\rm min,halo}\simeq 10^{10} {\!\,{\rm M}_\odot}$), subhalos can no longer be identified when their mass has dropped below this limit due to tidal stripping. Since mass loss can be substantial (>90%) this is important even for fairly massive subhalos. A special treatment of these so-called “orphans” is 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 recipe of [@boylan08], which is applicable at radii $r<r_{vir}$. We also tried an alternate recipe in which we make no explicit use of the subhalo information, but apply the dynamical friction formula from the time when the satellites first enter the host halo. We obtained very similar subhalo mass functions and radial distributions with the alternate recipe, confirming the self-consistency of the approach.
For the halo positions in the determination of CFs, we use the coordinates of the most bound particle for distinct and subhalos. For orphans, by definition, the position is not known, so we follow the position of the most bound particle from the last time-step when a subhalo was identified. Since the dynamical friction force vanishes in the dark matter only simulation after a subhalo is dissolved, yet not in reality when a galaxy is present at the center of the subhalo, the distance to the center of the host halo might be slightly overestimated with this prescription.
Connecting galaxies and halos {#s:galaxies}
=============================
In this section we describe how we derive the relationship connecting the stellar mass of a galaxy to the mass of its dark matter halo. In the standard picture of galaxy formation, gas can only cool and form stars if it is in a virialized gravitationally bound dark matter halo [@whiterees78]. In this model the gas cooling rate, the star formation rate and thus the properties of the galaxy depend mainly on the virial mass of the host halo. Thus we expect the stellar mass of a central galaxy to be strongly correlated with the virial mass of the halo in which the galaxy formed. As we discussed in the last section, this corresponds to the virial mass for central galaxies, and to the maximum mass over the halo’s history for satellite galaxies. In the rest of this work, unless noted otherwise, the halo mass $M$ will represent: $$M = \begin{cases}
M_{\rm vir} & \text{for host halos}\\
M_{\rm max} & \text{for subhalos}
\end{cases}
\label{eqnmmax}$$
Note that we have also experimented with instead using the present mass for subhalos, and found that we were not able to reproduce the galaxy clustering properties [see also @conroy06].
The stellar-to-halo mass relation {#s:galaxymap}
---------------------------------
In order to link the stellar mass of a galaxy $m$ to the mass of its dark matter halo $M$ we need to specify the SHM ratio. A direct comparison of the halo mass function $n(M)$ and the galaxy mass function $\phi(m)$ helps to constrain the stellar-to-halo mass function. If we assume that every host (sub) halo contains exactly one central (satellite) galaxy and that each system has exactly the same SHM ratio $m/M$, the galaxy stellar mass function can be derived trivially from the halo mass function and has the same features. The galaxy mass function derived for $m/M =
0.05$ is compared to the observed SDSS galaxy mass function in Figure \[f:fig1\]. The observed galaxy mass function is steeper for high masses and shallower for low masses than the one derived from the halo mass function. Thus, for a constant SHM ratio there will inevitably be too many galaxies at the low and high mass end.
This implies that the actual SHM ratio $m/M$ is not constant, but increases with increasing mass, reaches a maximum around $m^*$ and then decreases again. Hence we adopt the following parametrization, similar to the one used in @yang03: $$\frac{m(M)}{M} = 2 \left( \frac{m}{M}\right)_0 \left[\left(\frac{M}{M_1}\right)^{-\beta} + \left(\frac{M}{M_1}\right)^{\gamma}\right]^{-1}
\label{eqnmmap}$$ It has four free parameters: the normalization of the stellar-to-halo mass ratio $(m/M)_0$, a characteristic mass $M_1$, where the SHM ratio is equal to $(m/M)_0$, and two slopes $\beta$ and $\gamma$ which indicate the behavior of $m/M$ at the low and high mass ends respectively. We use the same parameters for the central and satellite populations, since – unlike luminosity – the stellar mass of satellites changes only slightly after they are accreted by the host halo.
Note that though both $\beta$ and $\gamma$ are expected to be positive, they are not restricted to be so. The SHM relation is therefore not necessarily monotonic.
Constraining the free parameters {#constraining}
--------------------------------
Having set up the model we now need to constrain the four free parameters $M_1$, $(m/M)_0$, $\beta$ and $\gamma$. To do this, we populate the halos in the simulation with galaxies. The stellar masses of the galaxies depend on the mass of the halo and are derived according to our prescription (equation \[eqnmmap\]). The positions of the galaxies are given by the halo positions in the $N$-body simulation.
Once the simulation box is filled with galaxies, it is straightforward to compute the SMF $\Phi_{mod}(m)$. As we want to fit this model mass function to the observed mass function $\Phi_{obs}(m)$ by @panter2007, we choose the same stellar mass range ($10^{8.5}-10^{11.85}~{\!\,{\rm M}_\odot}$) and the same binsize. The observed SMF was derived using spectra from the Sloan Digital Sky Survey Data Release 3 (SDSS DR3); see @panter2004 for a description of the method.
Furthermore it is possible to determine the stellar mass dependent clustering of galaxies. For this we compute projected galaxy CFs $w_{p,mod}(r_p,m_i)$ in several stellar mass bins which we choose to be the same as in the observed projected galaxy CFs of @li06. These were derived using a sample of galaxies from the SDSS DR2 with stellar masses estimated from spectra by @kauffmann03.
We first calculate the real space CF $\xi(r)$. In a simulation this can be done by simply counting pairs in distance bins: $$\label{corrfunc}
\xi(r_i) = \frac{dd(r_i)}{N_p(r_i)}-1$$ where $dd(r_i)$ is the number of pairs counted in a distance bin and $N_p(r_i) = 2 \pi N^2 r_i^2 \Delta r_i / L_{\rm box}^3$ where $N$ is the total number of galaxies in the box. The projected CF $w_p(r_p)$ can be derived by integrating the real space correlation function $\xi(r)$ along the line of sight: $$w_p(r_p) = 2 \int_{0}^{\infty} {\rm d}r_{||} \xi(\sqrt{r_{||}^2+r_p^2}) = 2 \int_{r_p}^{\infty} {\rm d}r \frac{r~\xi(r)}{\sqrt{r^2-r_p^2}} \;,$$ where the comoving distance ($r$) has been decomposed into components parallel ($r_{||}$) and perpendicular ($r_p$) to the line of sight. The integration is truncated at $45$ Mpc. Due to the finite size of the simulation box ($L_{\rm box}=100$ Mpc) the model correlation function is not reliable beyond scales of $r\sim0.1~L_{\rm box}\sim10$ Mpc.
In order to fit the model to the observations we use Powell’s directions set method in multidimensions (e.g. Press et al. 1992) to find the values of $M_1$, $(m/M)_0$, $\beta$ and $\gamma$ that minimize either $$\chi_r^2 = \chi_r^2 (\Phi)= \frac{\chi^2(\Phi)}{N_{\Phi}}$$ (mass function fit) or $$\chi_r^2 = \chi_r^2 (\Phi)+\chi_r^2 (w_p) = \frac{\chi^2(\Phi)}{N_{\Phi}} +
\frac{\chi^2(w_p)}{N_r\;N_m}$$ (mass function and projected CF fit) with $N_{\Phi}$ and $N_{r}$ the number of data points for the SMF and projected CFs, respectively, and $N_{m}$ the number of mass bins for the projected CFs.
In this context $\chi^2(\Phi)$ and $\chi^2(w_p)$ are defined as: $$\begin{aligned}
\chi^2(\Phi) &=& \sum_{i=1}^{N_{\Phi}} \left[ \frac{\Phi_{\rm mod}(m_i)-\Phi_{\rm obs}(m_i)}{\sigma_{\Phi_{\rm obs}(m_i)}} \right]^2\notag\\
\chi^2(w_p) &=& \sum_{i=1}^{N_m} \sum_{j=1}^{N_r} \left[ \frac{w_{p,\rm mod}(r_{p,j},m_i)-w_{p,\rm obs}(r_{p,j},m_i)}{\sigma_{w_{p,\rm obs}(r_{p,j},m_i)}} \right]^2\notag\;,\end{aligned}$$ with $\sigma_{\Phi_{\rm obs}}$ and $\sigma_{w_{p,\rm obs}}$ the errors for the SMF and projected CFs, respectively. Note that for the simultaneous fit, by adding the reduced $\chi_r^2$, we give the same weight to both data sets.
Estimation of parameter errors {#s:probdis}
------------------------------
In order to obtain estimates of the errors on the parameters, we need their probability distribution prob($A\vert I$), where $A$ is the parameter under consideration and $I$ is the given background information. The most likely value of $A$ is then given by: $A_{\rm best}
={\rm max(prob(}A\vert I))$.
As we have to assume that all our parameters are coupled, we can only compute the probability for a given set of parameters. This probability is given by: $${\rm prob(}M_1,(m/M)_0,\beta,\gamma\vert I) \propto \exp(-\chi^2)$$ In a system with four free parameters $A,B,C$ and $D$ one can calculate the probability distribution of one parameter (e.g. $A$) if the probability distribution for the set of parameters is known, using marginalization: $$\begin{aligned}
{\rm prob(}A\vert I) &=& \int_{-\infty}^{\infty}{\rm prob(}A,B\vert I) {\rm d}B\notag\\
&=& \int_{-\infty}^{\infty}{\rm prob(}A,B,C,D\vert I) {\rm d}B {\rm d}C {\rm d}D\notag\end{aligned}$$
Once the probability distribution for a parameter is determined, one can assign errors based on the confidence intervals. This is the shortest interval that encloses a certain percentage $X$ of the area under the posterior probability distribution. For the 1-sigma error $X = 68\%$ while for the 2-sigma error $X = 95\%$. Assuming that the probability distribution has been normalized to have unit area we seek $A_1$ and $A_2$ such that
$$\int_{0}^{A_1} {\rm prob(}A\vert I) {\rm d}A = \int_{A_2}^{\infty}
{\rm prob(}A\vert I) {\rm d}A = \frac{1-X}{2} .$$
Finally the parameter $A$ is given as $A = A_{\rm best}\text{
}_{-\sigma_{-}}^{+\sigma_{+}}$ with $\sigma_{+} = A_2 - A_{\rm best} $ and $\sigma_{-} = A_{\rm best} - A_1$. The errors derived in this way only include sources that have been considered when computing $\chi^2$. The calculation of the errors applies for uncorrelated data points. Since in our case the data points are correlated the values of the errors are slightly modified. Also errors caused by cosmic variance are not included.
Fitting results {#s:results}
===============
Here we present the results we obtain by fitting to the stellar mass function only, and for the combined fit to the SMF and the projected CF.
The stellar mass function fit
-----------------------------
[lllllll]{}\[hb!\] best fit & 11.884 & 0.02820 & 1.057 & 0.556 & 1.56 & 3.83\
$\sigma^+$ & 0.030 & 0.00061 & 0.054 & 0.010 & &\
$\sigma^-$ & 0.023 & 0.00053 & 0.046 & 0.004 & &\
\[t:mfresults\]
First we fit to the SDSS SMF and use the derived best-fit parameters to calculate the model projected correlation functions. Note that for now, we do not take into account any possible scatter in the $m(M)$ relation. We will consider scatter in §\[s:scattersec\].
We see in Figure \[f:fig1\] that our fit produces excellent agreement with the observed SMF. Using the approach described above we also compute the errors on the parameters. The results are summarized in Table \[t:mfresults\].
Having derived the best-fit parameters, we can predict the projected CFs. We present the results both including and not including orphan galaxies, where we have fitted to the SMF for each case.
Figure \[f:fig2\] shows a comparison between our model and the SDSS projected correlation functions in five stellar mass bins ranging from $\log m/{\!\,{\rm M}_\odot}=9.0$ to $\log m/{\!\,{\rm M}_\odot}=11.5$ with a binsize of $0.5 {\rm~dex}$. The correlation function that has been derived without orphans is too low at small scales and can be regarded as a lower limit. Neglecting these galaxies results in an underprediction of satellite galaxy clustering. As on small scales the projected CF depends mainly on the one-halo term this results in the underprediction of $w_p(r_p)$. This effect weakens for the clustering of more massive galaxies as they are more likely to be central galaxies and thus not effected by tidal stripping at all.
The agreement with the observationally derived $w_p(r_p)$ for the catalogue including orphaned galaxies is very good, which is also reflected in the low value of $\chi_r^2(w_p) = 3.83$. Note that this value has been calculated with the parameters from the mass function fit given above and does not correspond to a fit to the projected CFs.
Note that we plot the projected CFs only up to $20\, {\rm
Mpc}$. Because of the finite box size, the clustering of host halos and thus central galaxies is underpredicted at large scales independent of mass. Additionally, due to the lack of long-wavelength modes, massive halos and galaxies can be underproduced leading to an underprediction of $w_p$ for the massive objects, independent of scale. However, the latter effect is very small, since the abundance of the massive halos in our simulation agrees very well with the predicted average [@sheth1999].
As a test we also used the present mass instead of the maximum mass for subhalos. We then found that the projected CF was underpredicted particularly on small scales. This effect is due to tidal stripping of subhalos and is thus strongest at small scales where the subhalo contribution dominates.
The combined fit {#combinedfit}
----------------
We now investigate whether we can improve the agreement between the model and the observed projected CFs by performing a combined fit as described above. We obtain the same parameters as those we derived from the fit to the SMF alone. This seems surprising, but on further inspection we find that this is due to $\chi^2(m)$ being a lot more sensitive to changes of the parameters than $\chi^2(w_p)$. This means that if one changes the parameters a little in order to improve the fit to the projected correlation functions, one can get a slightly better agreement between the model and the observed projected CFs only at the cost of a large disagreement between the model and the observed stellar mass functions. In other words: $\chi^2(w_p)$ is much flatter around its minimum than $\chi^2(w_p)$, as shown in Figure \[f:fig3\].
This means that, assuming that both central and satellite galaxies follow the same SHM relation, the model that matches the SMF can reproduce the correct clustering. However, if subhalos have a different SHM ratio there is an infinite number of solutions that match the SMF but produce very different correlation functions. The only way to constrain the SHM relations then is to take the clustering data into account. By adopting different SHM relations for central and satellite populations it is even possible to produce a slightly better fit to the correlation functions [@wang06].
On the other hand, if one wants to [*predict*]{} clustering as a function of stellar mass (e.g. at higher redshift) then one has to make an assumption about how the SHM ratios of central and satellite galaxies are related. We made the very simple assumption, that the relation between the stellar mass of central galaxies and the virial mass of their host halo and the relation between the stellar mass of satellite galaxies and the mass of the subhalo at the time of accretion is the same, and have shown that this leads to very good predictions for the mass dependent clustering. We conclude that under this simple assumption we can use our model to predict clustering as a function of stellar mass.
The resulting stellar-to-halo mass relation {#masstomass}
-------------------------------------------
\[f:fig4\]
The upper panel of Figure \[f:fig4\] shows the derived stellar mass as a function of halo mass. The light shaded area gives the 68% confidence interval while the dark and light shaded areas together give the 95% confidence interval. These have been derived using a set of different models computed on a mesh, as described in §\[s:probdis\].
For the SHM ratio we apply the same procedure. The result is shown in the lower panel of Figure \[f:fig4\]. We see that the SHM ratio has the form we expected: it increases with increasing halo mass, reaches its maximum value around $M_1$ and then decreases again.
Meaning of parameters and correlations {#s:parameters}
--------------------------------------
\[f:fig5\]
We now explore the effects of changing each parameter in order to understand how they affect the SMF. \[ratio\] If we keep $M_1$, $\beta$ and $\gamma$ fixed and only vary $(m/M)_0$, this corresponds to changing the stellar mass of the galaxy that lives inside each halo by a constant factor. This has no impact on the form of the SMF. Its shape stays the same, while only the position on the stellar mass axis changes. Due to the monotonic form of the SMF this directly determines the value of the normalization $\phi^*$. For a larger value of $(m/M)_0$ we get a larger value of $\phi^*$.
\[m1\] Varying only $M_1$ we find that the shape of the SMF changes drastically. For a higher $M_1$ than our best fit value, we get too many massive galaxies and too few low mass galaxies, while for a lower value of $M_1$ we get too few massive galaxies and too many low mass galaxies. This is because $M_1$ is the characteristic mass corresponding to the highest SHM ratio. In the SMF, this corresponds to the knee and we get a SMF which has its knee at the stellar mass corresponding to $M_1$. For a larger $M_1$ the knee is shifted to a higher stellar mass. Together, $M_1$ and the maximum stellar-to-halo mass ratio $(m/M)_0$ determine the normalization of the stellar mass function $\phi$ and the characteristic mass $m^*$.
\[beta\] Changing $\beta$ affects mainly the low mass slope of the stellar mass function. For larger values of $\beta$ the slope becomes shallower. As $\beta$ influences mainly the slope of the low mass end of the SMF, it is strongly related to the parameter $\alpha$ of the Schechter function. A small value of $\beta$ corresponds to a high value of $\alpha$.
\[gamma\] If we change $\gamma$, this mainly impacts the slope of the massive end of the SMF. For larger values of $\gamma$ than for its best-fit value the slope of the massive end becomes steeper. As $\gamma$ affects mainly the slope of the massive end of the SMF it is not coupled to a parameter of the Schechter function though it is related to the high-mass cutoff, assumed to be exponential in a Schechter function.
Figure \[f:fig5\] shows the contours of the two-dimensional probability distributions for the parameters pairs. We see a correlation between the parameters \[$M_1,\gamma$\] and \[$(m/M)_0,\gamma$\] and an anti-correlation between \[$\beta,\gamma$\], \[$\beta,M_1$\] and \[$(m/M)_0,M_1$\]. There does not seem to be a correlation between \[$\beta,(m/M)_0$\].
Introducing scatter {#s:scattersec}
-------------------
Up until now we have assumed that there is a one-to-one, deterministic relationship between halo mass and stellar mass. However, in nature, we expect that two halos of the same mass $M$ may harbor galaxies with different stellar masses, since they can have different halo concentrations, spin parameters and merger histories.
For each halo of mass $M$, we now assign a stellar mass $m$ drawn from a log-normal distribution with a mean value given by our previous expression for $m(M)$ (Equation ), with a variance of $\sigma_m^2$. We assume that the variance is a constant for all halo masses, which means that the percent deviation from $m$ is the same for every galaxy. This is consistent with other halo occupation models, semi-analytic models and satellite kinematics [@cooray2006; @vdb2007; @more2009b].
Assuming a value of $\sigma_m = 0.15{\rm~dex}$ and fitting the stellar mass function only, we find the values given in Table \[t:mfscatresults\]. These values lie within the (2$\sigma$) error bars of the best-fit values that we obtained with no scatter. The largest change is on the value of $\gamma$, which controls the slope of the SHM relation at large halo masses. The SMF and the projected CFs for the model including scatter are shown in Figures \[f:fig1\] and \[f:fig2\], respectively, and show very good agreement with the observed data.
[lllllll]{}\[hb!\] best fit & 11.899 & 0.02817 & 1.068 & 0.611 & 1.42 & 4.21\
$\sigma^+$ & 0.026 & 0.00063 & 0.051 & 0.012 & &\
$\sigma^-$ & 0.024 & 0.00057 & 0.044 & 0.010 & &\
\[t:mfscatresults\]
In Figure \[f:fig6\] we compare our model without scatter with the model including scatter. We have also included the relation between halo mass and the average stellar mass. Especially at the massive end scatter can influence the slope of the SMF, since there are few massive galaxies. This has an impact on $\gamma$ and as all parameters are correlated scatter also affects the other parameters. We thus see a difference between the model without scatter and the most likely stellar mass in the model with scatter in Figure \[f:fig6\].
The conditional mass function {#s:cmf}
=============================
In the previous section we derived a model that specifies the stellar mass of a central galaxy as a function of the virial mass of its host halo and the stellar mass of a satellite galaxy as a function of the maximum mass of the subhalo in which it lives. It has become common to represent the population of host halos by the Halo Occupation Distribution (HOD). This includes the halo occupation function $P(N\vert M)$ which is the probability distribution that a halo of mass $M$ contains $N$ galaxies (of a specific type). A close relative of the HOD is the “conditional luminosity function” [CLF; e.g. @yang03; @vdb2007; @yang04]. It extends the halo occupation function $P(N\vert M)$ (which gives only information about the total number of galaxies per halo in a given luminosity range) and yields the average number of galaxies with luminosities in the range $L \pm {\rm d}L/2$ as a function of the virial mass $M$ of their host halo.
We define its analog, the “conditional mass function” (CMF), or the average number of galaxies with stellar masses in the range $m \pm
{\rm d}m/2$ as a function of the virial mass $M$ of their host halo. This provides a direct link between the SMF $\Phi(m)$ and the host halo mass function ${\rm d}n(M)/{\rm d}M$: $$\Phi(m) = \int_0^\infty \Phi(m\vert M) \frac{{\rm d}n(M)}{{\rm d}M} \text{d}M$$ A host halo of mass $M$ can contain a whole population of galaxies with different stellar masses $m$. If we count the number of galaxies living in host halos with a virial mass in the range $M \in [M_1,M_2]$ we can compute the SMF of the halo bin $[M_1,M_2]$: $$\label{cmftosmf}
\tilde{\Phi}(m) = \int_{M_1}^{M_2} \Phi(m\vert M) \frac{{\rm d}n(M)}{{\rm d}M}
{\rm d}M \approx \Phi(m\vert \bar{M}) \Delta n$$ The tilde over a function represents the fact that it is computed in a halo mass bin. We have replaced the integral by a “tophat” with a width of $\Delta n$ (number of host halos in the bin) and a height of $\Phi(m\vert M_m)$, where $\bar{M}$ is the geometric mean of the minimum and maximum halo masses bracketing the bin.
This equation allows us to put constraints on $\Phi(m\vert M)$ by calculating $\tilde{\Phi}(m)/\Delta n$. We can then choose an adequate parameterization of $\Phi(m\vert M)$ and fit these parameters to $\tilde{\Phi}(m)/\Delta n$ in every halo mass bin. Finally we can investigate the halo mass dependence of the parameters.
Parameterization
----------------
In order to specify the CMF $\Phi(m \vert M)$ we divide the galaxy population into a central and a satellite part, as in the updated CLF formalism [@zheng2005; @zehavi2005; @cooray2006; @yang08; @cacciato2008]. The central part is $\Phi_c(m \vert M)$ and the satellite part is $\Phi_s(m \vert M)$. Then the total CMF is the sum of both parts: $$\Phi(m \vert M) = \Phi_c(m \vert M) + \Phi_s(m \vert M)$$
Note that both $\Phi_c(m \vert M)$ and $\Phi_s(m \vert M)$ are statistical functions and should not be regarded as the mass functions of galaxies living in a given individual halo.
For the central population we expect the CMF to have a peak around the stellar mass $m_c$ that corresponds to the host halo’s virial mass $M$ in the SHM relation (equation \[eqnmmap\]). Due to the halo mass bin size this distribution gets smeared out, because halos in the interval $[M_1,M_2]$ contain central galaxies of stellar masses $m \in [m_1(M_1),m_2(M_2)]$. Thus $\tilde{\Phi}(m)/\Delta n$ will be finite inside the interval $[m_1(M_1),m_2(M_2)]$ and zero elsewhere with a normalization such that the number of central galaxies per halo equals one. This can be regarded as scatter $\sigma_{\rm bin}$ due to the binning. If we add intrinsic scatter $\sigma_m$ to relation , we expect $\Phi_c(m\vert M)$ to be a lognormal with a maximum around $m_c(M)$ and a variance of $\sigma_m^2$. To this scatter the binning scatter $\sigma_{\rm bin}$ adds in quadrature (assuming that $\sigma_{\rm bin}$ and $\sigma_m$ are uncorrelated), resulting in a total scatter of $\sigma_c^2=\sigma_m^2+\sigma_{\rm bin}^2$. For both cases ($\sigma_m=0$ and $\sigma_m\neq0$) we use a lognormal distribution: $$\Phi_c(m \vert M) = \frac{1}{\sqrt{2\pi}\ln10\;m\;\sigma_c}
\exp\left[-\frac{\log^2(m/m_c)}{2\sigma_c^2}\right]\;,$$ where the mean $m_c(M)$ and width $\sigma_c^2(M)$ are parameterized functions of the halo mass M.
For the satellite population we adopt a Schechter function with a steeper slope for the massive end. This is done by squaring the argument of the exponential function in the Schechter function: $$\Phi_s(m \vert M) = \frac{\Phi_s^*}{m_s} \left( \frac{m}{m_s}\right)^{\alpha_s}
\exp \left[ -\left( \frac{m}{m_s}\right)^2 \right]\;.$$ Also here the parameters $\Phi_s^*(M)$, $m_s(M)$ and $\alpha_s(M)$ are functions of the host halo mass $M$. They are the normalization, the characteristic mass and the low mass slope of the satellite population of host halos of mass $M$.
Constraining the conditional mass function {#constraincmf}
------------------------------------------
We populate the halos and subhalos in our simulation with central and satellite galaxies according to the prescription in section \[s:galaxies\]. Then we choose halo mass bins between $\log M/{\!\,{\rm M}_\odot}= 10.2$ and $\log M/{\!\,{\rm M}_\odot}= 15.0$ with a bin size of $\Delta M =
0.4{\rm~dex}$. In every halo mass bin we seek all galaxies which live in a host halo with a mass in that bin, which we divide between central and satellite galaxies. For these populations we then compute two seperate SMFs which we normalize such that the number of central galaxies per host halo equals one. This procedure then yields for every halo mass bin a central and a satellite distribution $({\rm d}\tilde{n}_g / {\rm d}\log M) \Delta
n_h$.
Using equation we can now relate the stellar mass function in a halo mass bin to the CMF: $$\begin{aligned}
\frac{{\rm d}\tilde{n}_g(m)}{{\rm d}\log M}\frac{1}{\Delta n_h} &=&
\frac{\ln{10}}{\Delta n_h} \; M \; \frac{{\rm d}\tilde{n}_g(m)}{{\rm d}M} \nonumber\\
&=& \ln{10} \; M \; \frac{\tilde{\Phi}(m)}{\Delta n_h} \nonumber\\
&\approx& \ln{10} \; M \; \Phi(m\vert M)\end{aligned}$$ Now we can fit the five parameters $m_c(M)$, $\sigma_c(M)$, $m_s(M)$, $\Phi_s^*(M)$ and $\alpha_s(M)$ to the SMFs in each halo bin. We compute and fit the central and the satellite parts seperately.
The left panels of Figure \[f:fig7\] show the CMF in a subsample of halo mass bins running from $\log M/{\!\,{\rm M}_\odot}= 10.2 \pm 0.2 \text{ to }
15.0 \pm 0.2$, where we have not included intrinsic scatter in the SHM relation. For the satellite part, only galaxies with a mass above the completness limits for each halo mass bin (as indicated in Figure \[f:fig7\]) have been used in the fit.
In low-mass halos ($\log M/{\!\,{\rm M}_\odot}<11.0$) the contribution from satellite galaxies is very small and the central contribution dominates until $\log M/{\!\,{\rm M}_\odot}=12.0$. For massive halos ($\log
M/{\!\,{\rm M}_\odot}>13.0$) the satellite contibution dominates by number. The mean of the lognormal fit to the central contribution also increases with halo mass as stipulated by the model derived in Section \[s:galaxies\]. The characteristic mass scale of the satellite contribution also increases with halo mass meaning that the most massive satellite galaxies have a mass which is comparable to the mass of the central galaxy.
The scatter of the central contribution $\sigma_c(M)$ decreases with halo mass. As we did not include any scatter in the model, this scatter reflects the width ($0.4{\rm~dex}$) of the halo mass bins ($\sigma_{\rm bin}$). The halo mass dependence of $\sigma_c(M)$ arises because a fixed halo mass bin is mapped to a smaller galaxy mass bin for larger halo mass due to the shape of the SHM relation. Another feature of the CMF is the slope for low mass satellite galaxies $\alpha_s(M)$ which becomes shallower with increasing halo mass.
The parameters of the conditional mass function {#s:cmfpars}
-----------------------------------------------
In this section we investigate the halo mass dependence of the five parameters of the CMF: $m_c(M)$, $\sigma_c(M)$, $m_s(M)$, $\Phi_s^*(M)$ and $\alpha_s(M)$. They have been fixed by fitting to the stellar mass functions in each halo mass bin. We introduce a parameterization in order to describe the dependence on halo mass and constrain these by a fit to each parameter. The results are presented in Table \[t:cmfparameters\]. This provides a complete description of the CMF.
[lllll]{} $\log M_{1c}$ & 11.9347 & $\pm$ 0.0257 & 11.9008 & $\pm$ 0.0119\
$(m_c/M)_0$ & 0.0267 & $\pm$ 0.0006 & 0.0297& $\pm$ 0.0004\
$\beta_c$ & 1.0059 & $\pm$ 0.0332 & 1.0757& $\pm$ 0.0097\
$\gamma_c$ & 0.5611 & $\pm$ 0.0065 & 0.6310 & $\pm$ 0.0121\
$\log M_2$ & 11.9652 & $\pm$ 0.1118 & 11.8045 & $\pm$ 0.0458\
$\sigma_{\infty}$ & 0.0569 & $\pm$ 0.0052 & 0.1592& $\pm$ 0.0030\
$\sigma_1$ & 0.1204 & $\pm$ 0.0191 & 0.0460& $\pm$ 0.0029\
$\xi$ & 6.3020 & $\pm$ 3.0720 & 4.2503& $\pm$ 0.9945\
$\log M_{1s}$ & 12.1988 & $\pm$ 0.0878 & 12.0640& $\pm$ 0.0931\
$(m_s/M)_0$ & 0.0186 & $\pm$ 0.0012 & 0.0198& $\pm$ 0.0015\
$\beta_s$ & 0.7817 & $\pm$ 0.0629 & 0.8097 & $\pm$ 0.0971\
$\gamma_s$ & 0.7334 & $\pm$ 0.0452 & 0.6910& $\pm$ 0.0390\
$-\log\Phi_0$ & 11.1622 & $\pm$ 0.2874 & 10.8924& $\pm$ 0.4615\
$\lambda$ & 0.8285 & $\pm$ 0.0215 & 0.8032 & $\pm$ 0.0367\
$\log M_3$ & 12.5730 & $\pm$ 0.1351 & 12.3646 & $\pm$ 0.0260\
$-\alpha_{\infty}$ & 1.3740 & $\pm$ 0.0066 & 1.3676 & $\pm$ 0.0043\
$-\alpha_1$ & 0.0309 & $\pm$ 0.0076 & 0.0524& $\pm$ 0.0051\
$\zeta$ & 4.3629 & $\pm$ 2.6810 & 9.5727 & $\pm$ 6.8240\
\[t:cmfparameters\]
As we have already determined the mean relation between the stellar mass of a galaxy and the mass of its halo, the form of $m_c(M)$ has to be the same and can thus be decribed by equation : $$\label{mcmap}
m_c(M) = 2 \; M \; \left( \frac{m_c}{M}\right)_0 \left[\left(\frac{M}{M_{1c}}\right)^{-\beta_c} + \left(\frac{M}{M_{1c}}\right)^{\gamma_c}\right]^{-1}$$ This yields four parameters $(m_c/M)_0$, $M_{1c}$, $\beta_c$ and $\gamma_c$.
In the upper left panel of Figure \[f:fig8\] $m_c(M)$ is plotted as a function of halo mass. Note that by construction, it has the same form as the SHM relation.
\[sigmac\] The scatter of the central galaxy contribution is high for low halo masses and decreases for more massive halos. The middle left panel of Figure \[f:fig8\] shows $\sigma_c(M)$ as a function of halo mass. As one can see, $\sigma_c(M)$ goes to a constant value both for low and high halo masses while it decreases with halo mass. We therefore choose the following parameterization:
$$\label{scmap}
\sigma_c(M) = \sigma_{\infty} + \sigma_1 \left[ 1-\frac{2}{\pi} \arctan\left( \xi \log \frac{M}{M_2} \right) \right]$$
This yields four more parameters $\sigma_{\infty}$, $\sigma_1$, $\xi$ and $M_2$. Here, $\sigma_{\infty}$ sets the high mass limit of $\sigma_c(M)$ while $\sigma_1$ sets the difference between the low and high mass limits of $\sigma_c(M)$. The parameter $M_2$ determines the mass scale at which the transition occurs and $\xi$ sets the strength. For a large (small) value of $\xi$ the transition occurs in a small (large) interval around $M_2$.
The specific shape of $\sigma_c(M)$ can be explained by the form of the SHM relation (equation \[eqnmmap\]). As we have not included any scatter in this relation ($\sigma_m=0$), the width of the lognormal function of the central galaxy distribution arises from the width of the halo mass bin ($\sigma_c=\sigma_{\rm bin}$). A halo mass interval $[M_1,M_2]$ contains only central galaxies with stellar masses of $m
\in [m_1(M_1),m_2(M_2)]$. The lower left panel of Figure \[f:fig8\] illustrates this by showing how halo mass bins affect the bin size of the stellar mass. If we choose the same bin size for low and high mass halos, we get different bin sizes for low and high mass galaxies, due to the changing slope of $m(M)$. Therefore the transition occurs where the slope of $m(M)$ changes which is around $M_1$, so the value of $M_2$ is very close to that value.
As Figure \[f:fig7\] shows that the satellite contribution falls off around the mean mass of the central galaxy, we expect the characteristic mass of the modified Schechter function $m_s(M)$ to follow $m_c(M)$. We therefore describe $m_s(M)$ with the same function we used for the parametrisation of $m_c(M)$: $$\label{msmap}
m_s(M) = 2 \; M \; \left( \frac{m_s}{M}\right)_0 \left[\left(\frac{M}{M_{1s}}\right)^{-\beta_s} + \left(\frac{M}{M_{1s}}\right)^{\gamma_s}\right]^{-1}$$ This function yields four parameters $(m_s/M)_0$, $M_{1s}$, $\beta_s$ and $\gamma_s$.
The upper right panel of Figure \[f:fig8\] plots $m_s(M)$ as a function of halo mass. We see that the shape is similar to that of $m_c(M)$. Note that $m_s(M)$ is always lower than $m_c(M)$, while the deviation increases with increasing halo mass. This implies that for high halo masses the satellite contribution to the CMF falls off before the mean mass of the central galaxy.
The normalization of the modified Schechter function is small for low halo masses and increases with the mass of the host halo. The middle right panel of Figure \[f:fig8\] shows $\Phi_s^*(M)$ as a function of halo mass. We see that $\Phi_s^*(M)$ can be described by a power law and choose the following parametrisation: $$\label{psmap}
\Phi_s^*(M) = \Phi_0 \left( \frac{M}{{\!\,{\rm M}_\odot}}\right)^{\lambda}$$ We get two more parameters, $\Phi_0$ and $\lambda$. The normalization of $\Phi_s^*(M)$ is given by $\Phi_0$ and the slope by $\lambda$. The shape of $\Phi_s^*(M)$ implies that the probability for a host halo to harbor satellite galaxies (in a given stellar mass range) increases with increasing halo mass.
\[alphas\] The slope of the modified Schechter function for the satellite contribution becomes shallower for more massive halos. The lower right panel of Figure \[f:fig8\] shows $\alpha_s(M)$ as a function of halo mass and shows that $\alpha_s(M)$ goes to a constant value for both low and high halo masses. Similar to $\sigma_c(M)$, we choose the parameterization: $$\label{asmap}
\alpha_s(M) = \alpha_{\infty} + \alpha_1 \left[ 1-\frac{2}{\pi} \arctan\left( \zeta \log \frac{M}{M_3} \right) \right]$$ This yields four more parameters $\alpha_{\infty}$, $\alpha_1$, $\zeta$ and $M_3$. Here, $\alpha_{\infty}$ sets the high mass limit of $\alpha_c(M)$ while $\alpha_1$ sets the difference between the low and high mass limits of $\alpha_c(M)$. The mass scale at which this transition occurs is determined by $M_3$ and $\zeta$ sets its strength. The transition occurs in a small (large) interval around $M_3$ for a large (small) value of $\zeta$.
The impact of scatter
---------------------
Until now, we have used the SHM relation without any intrinsic scatter. In this section we investigate how the CMF and the parameters change if we include a scatter $\sigma_m$ as described in section \[s:scattersec\]. This scatter is again assumed to be constant with host halo mass.
The right panels of Figure \[f:fig7\] show the resulting CMF in a subsample of halo mass bins for an intrinsic scatter of $\sigma_m =
0.15$. The central part is now no longer near-constant in the interval $[m(M-\Delta M/2),m(M+\Delta M/2)]$ as in the left panels of Figure \[f:fig7\] (where $\sigma_m = 0.0$) but has the form of a lognormal with a broader distribution for bigger $\sigma_m$. As the scatter has been taken from a lognormal distribution, the central galaxy contribution to the CMF is distributed in the same way. Hence, $\sigma_c(M)$ changes with respect to the model that does not include artificial scatter. We notice that at the massive end the binning scatter $\sigma_{\rm bin}^2$ and the intrinsic scatter $\sigma_m^2$ add to the total scatter $\sigma_{\rm tot}^2$. At the low mass end, however, the total scatter is less than what has been obtained by using no intrinsic scatter. This shows that the two forms of scatter do not add in quadrature and indicates that they are correlated.
We compare $m_c(M)$, $\sigma_c(M)$, $m_s(M)$, $\Phi_s^*(M)$ and $\alpha_s(M)$ for $\sigma_m=0$ and $\sigma_m=0.15$ and show the resulting parameters in Table \[t:cmfparameters\] (columns four and five) and in Figure \[f:fig8\]. The mean mass of the central galaxy $m_c(M)$ does not change much if artificial scatter is introduced. The most likely stellar mass of a central galaxy is still given by the SHM relation, so the mean of the gaussian in logarithmic space stays the same. Also the parameters of the satellite population \[$m_s(M)$,$\Phi_s^*(M)$ and $\alpha_s(M)$\] do not change significantly.
The occupation numbers
----------------------
In order to compare our results to other HOD models it is useful to compute the average number of galaxies per halo $\langle N\rangle$, as this is the main prediction of the HOD approach. To compute $\langle
N\rangle(M)$ from the CMF we simply integrate $\Phi(m\vert M)$ over the desired stellar mass range: $$\langle N\rangle(M) = \int_{m_1}^{m_2} \Phi(m\vert M) {\rm d}m$$ As we have divided $\Phi(m\vert M)$ into a central galaxy contribution $\Phi_c(m\vert M)$ and a satellite galaxy contribution $\Phi_s(m\vert
M)$, we can compute seperate occupation numbers for central and satellite galaxies: $$\begin{aligned}
\langle N\rangle(M) &=& \int_{m_1}^{m_2} \Phi_c(m\vert M) {\rm d}m + \int_{m_1}^{m_2} \Phi_s(m\vert M) {\rm d}m\notag\\
&=& \langle N_c\rangle(M) + \langle N_s\rangle(M)\end{aligned}$$ The average number of central galaxies per halo $\langle
N_c\rangle(M)$ is given by $$\label{eqnocccen}
\langle N_c\rangle(M) = \frac{1}{2}\left[{\rm erf}(\eta_2)-{\rm erf}(\eta_1)\right] \;,$$ with the error-function ${\rm erf}(x)$ and the integration boundaries $$\eta_1 = \frac{\log(m_1/m_c)}{\sqrt{2}\sigma_c} \quad \text{and} \quad \eta_2 = \frac{\log(m_2/m_c)}{\sqrt{2}\sigma_c} \;.$$ The average number of satellite galaxies per halo $\langle N_s\rangle(M)$ is $$\label{eqnoccsat}
\langle N_s\rangle(M) = \frac{\Phi_s}{2}\left[\Gamma\left(\frac{\alpha_s}{2}+\frac{1}{2},\kappa_1\right)-\Gamma\left(\frac{\alpha_s}{2}+\frac{1}{2},\kappa_2\right)\right] \;,$$ with the upper incomplete gamma function $\Gamma(a,x)$ and the integration boundaries $$\kappa_1 = (m_1/m_s)^2 \quad \text{and} \quad \kappa_2 = (m_2/m_s)^2 \;.$$
Figure \[f:fig9\] shows the resulting occupation numbers for the values of the CMF parameters that were derived in section \[s:cmfpars\] (using a scatter of $\sigma_m=0.15$). The five lines in each panel correspond to different stellar mass bins.
The left panel shows the average number of central galaxies per halo $\langle N_c\rangle(M)$ as a function of halo mass. In the middle panel, the average number of satellite galaxies per halo $\langle
N_s\rangle(M)$ as a function of halo mass is shown. The right panel plots the average number of all galaxies per halo $\langle
N_{\rm tot}\rangle(M)$ as a function of halo mass. A galaxy of a low stellar mass can thus either be a central galaxy of a low mass halo, or a satellite galaxy of a massive halo. It is not likely to live in a halo of intermediate mass.
As it is common in the literature to plot occupation numbers not for stellar mass intervals, but for galaxy samples with a mass above a given threshold, we need to adjust equations and . The stellar mass threshold is then given by $m_1$ while $m_2 \to \infty$. This yields for the average number of central galaxies $$\langle N_c\rangle(M, m_1) = \frac{1}{2}\left[1 - {\rm erf}\left(\frac{\log(m_1/m_c)}{\sqrt{2}\sigma_c}\right)\right]\;,$$ since ${\rm erf}(x\to\infty)\to 1$, and for the average number of satellite galaxies $$\langle N_s\rangle(M, m_1) = \frac{\Phi_s}{2} \; \Gamma\left[\frac{\alpha_s}{2}+\frac{1}{2},\left( \frac{m_1}{m_s}\right)^2\right]\;$$ since $\Gamma(a,x\to\infty)\to 0$.
Figure \[f:fig10\] shows occupation numbers for different stellar mass thresholds. The left panel shows the average number of central galaxies per halo $\langle N_c\rangle(M)$ as a function of halo mass. The middle panel plots the average number of satellite galaxies per halo $\langle N_s\rangle(M)$ as a function of halo mass. It is similar to the middle panel of Figure \[f:fig9\] while it is larger at a given halo mass. In the right panel the average number of all galaxies per halo $\langle N_{tot}\rangle(M)$ as a function of halo mass is shown.
Comparison {#s:comparison}
==========
Other HOD models
----------------
[lllll]{} Our model & 11.884 & 0.0282 & 1.06 & 0.556\
Non-Parametric & 11.766 & 0.0324 & 1.43 & 0.565\
@wang06 & 11.845 & 0.0319 & 1.42 & 0.710\
Somerville SAM & 11.888 & 0.0276 & 0.98 & 0.629\
Croton SAM & 11.742 & 0.0405 & 0.92 & 0.610\
Yang GC & 12.067 & 0.0384 & 0.71 & 0.698\
\[t:paracomparison\]
Numerous variations on halo occupation models have been presented in the literature. In this section we describe some of the most popular ones and compare them to our model. As many authors use different initial mass functions and definitions of halo masses, we convert all results to the conventions that we have used in this work (Kroupa IMF and virial overdensity).
In the Non-Parametric model [@valeostriker06; @conroy06; @shankar2006], galaxy properties, such as luminosity and stellar mass, are monotonically related to the mass of dark matter halos. Using the observed galaxy SMF, the most massive halo is matched to the most massive galaxy: $$n_g(>m_i) = n_h(>M_i)$$ In this way, the observed SMF is automatically reproduced. Applying this procedure and fitting the parameters of the SHM relation to the result, we have derived the values given in Table \[t:paracomparison\]. These are in good agreement with the parameters of our model, except for $\beta$. We find that this is due to the shape of the SHM ratio for low masses. For the Non-Parametric model, $m(M<M_1)$ can not be perfectly described by a single power law, as is assumed in our model.
Adding an additional parameter and assuming a fitting function with five free parameters, we are able to fit the SHM relation predicted by the non-parametric model quite precisely. The fifth parameter accounts for the deviation from the power-law at high and low masses. Using the parameterization $$\label{eqnnonpar}
m(M)=m_0 \; \frac{(M/M_1)^{\gamma_1}}{\left[1+(M/M_1)^{\beta}\right]^{(\gamma_1-\gamma_2)/\beta}}$$ we determine the values given in Table \[t:paranonpar\]. Figure \[f:fig11\] shows the results of four- and five-parameter fits to the SHM relation derived via the non-parametric method, compared with our usual model. In the range where we applied the mass function fit, the non-parametric model lies within our error-bars.
[llllll]{} & 10.864 & 10.456 & 7.17 & 0.201 & 0.557\
$\pm$ & 0.043 & 0.211 & 1.16 & 0.018 & 0.031\
\[t:paranonpar\]
In @wang06 a model similar to ours is used to constrain the SHM ratio. The halo catalogue is taken from the Millennium simulation [@springel05b]; halos are identified using a friends-of-friends group finder while substructure is found using the [SUBFIND]{} algorithm of @springel01. As observational constraints, the authors use a SMF which they compute from the SDSS DR2 data using the mass estimates of @kauffmann03 and the projected CFs of @li06.
The parameterization they use is similar to ours, with four free parameters that can easily be converted to $M_1$, $(m/M)_0$, $\beta$ and $\gamma$ and an unconstrained scatter. These are fixed by generating a grid of models and the best-fit model is defined as the one for which $\chi^2=\chi^2(\Phi)+\chi^2(w_p)$ is minimal. They find that their fit improves if they take a different set of parameters for central and satellite galaxies. In Table \[t:paracomparison\] we compare our best-fit parameters with their central galaxy best-fit parameters which have been updated in @wang07. We show these results in Figure \[f:fig11\].
The values of $M_1$ and $(m/M)_0$ are in very good agreement with our values, but the slopes are both higher, resulting in fewer massive and fewer low mass galaxies. The reason for the difference in the low mass end is the different simulation used. As the resolution of the simulation in our model is higher, the low mass end can be constrained more tightly. For the massive end the difference in $\gamma$ can be explained by the additional unconstrained scatter that is used in @wang06. As the mass function is steep at high masses and shallow for low masses, a change in the scatter will influence the number of massive galaxies strongly, while it will have only a small effect on the low mass end. As the other three parameters $M_1$, $(m/M)_0$ and $\beta$ are coupled to the Schechter function parameters, there are two parameters to constrain the slope of the massive end of the SMF. This degeneracy can cause the difference in $\gamma$ between the two models. The fact that in the Millennium simulation the cosmology is different to that of our simulation also affects the value of the parameters.
Gravitational lensing
---------------------
The relation between stellar mass and halo mass can be constrained observationally using galaxy-galaxy lensing. Gravitational lensing induces shear distortions of background objects around foreground galaxies, allowing the mass of the dark matter halo to be estimated. @mandelbaum05 [@mandelbaum06] have used SDSS data to calibrate the predicted signal from a halo model which has been derived from a dissipationless simulation. They have extracted the mean halo mass as a function of stellar mass. The lensing data for combined early and late-type galaxies (Mandelbaum, private communication) are shown in Figure \[f:fig11\] and are in excellent agreement with our model.
Semi-analytic models {#sams}
--------------------
As we discussed in the introduction, semi-analytic models (SAMs) of galaxy formation attempt to predict the relationship between dark halo mass and stellar mass by a priori modelling of physical processes, such as the growth of structure, cooling, star formation, and stellar and AGN feedback. We compare our results with predictions from the latest version of the semi-analytic models of @somerville1999; see @somerville2008. For this we compute the mean stellar mass of central galaxies as a function of the mass of the host halo in halo mass bins. The results are shown in Figure \[f:fig11\] and are in good agreement with our model. This is not surprising, as the physical parameters in the model of @somerville2008 have been tuned to match the observed stellar mass function at $z=0$.
In @wang06 the authors use the semi-analytic model of @croton06 and link galaxy properties, such as the stellar mass, to the mass of the halo in which the galaxy was last a central object $M_{\rm infall}$. They fit the same four-parameter function that they used for their empirical model (described above) to obtain the parameter estimates from the SAM. We summarize these results in Table \[t:paracomparison\], and show them in Figure \[f:fig11\].
The two slopes are in very good agreement with our results. However, the normalization in the @croton06 SAM is $\sim25\%$ higher and the characteristic mass is $\sim25\%$ lower than what we found and what @wang06 find for their model. This is because the SAM of @croton06 does not produce a perfect fit to the observed SMF.
SDSS group catalogue {#groupcat}
--------------------
Another direct way of studying galaxy properties as a function of halo mass is using the SDSS group catalogue presented in @yang07. In this approach, galaxies are first linked together into “groups” using a friends-of-friends algorithm. Each group is then assigned a total halo mass by matching to the theoretical dark matter halo mass function. @yang08 present the relation between the mean stellar mass of the central galaxy and the host halo mass. We fit the parameters of equation \[eqnmmap\] to their relation and present the results in Table \[t:paracomparison\].
We note that the characteristic mass and the normalization derived from the group catalogue are both higher than our model parameters. The high mass slope of the SHM relation in the group catalogue is shallower than that of our model. The low mass slope is also shallower, however, the constraints on the low mass slope in the group catalogue are weak, since the lowest halo masses are $log(M/{\!\,{\rm M}_\odot})\sim11.7$. This can also be seen in Figure \[f:fig11\] where we show the SHM relation of the group catalogue for comparison.
High Redshift {#s:redshift}
=============
The discussion in the previous sections has focussed solely on the present day universe. In this section we extend our analysis to higher redshifts and derive the redshift dependence of the stellar-to-halo mass relation. Having chosen a particular observed stellar mass function at a given redshift, we can investigate how the parameters of the SHM ratio change with time. This allows us to learn about the evolution of galaxies. Also, with this information, we can populate the $N$-body simulation snapshots with galaxies at different redshifts using the appropriate redshift dependent SHM relation, and then use the spatial information from the simulation to compute the stellar mass dependent correlation functions.
Since at the present time there are no high redshift ($z\gta 1$) clustering data as a function of stellar mass available, we fit the four parameters of equation to the observed SMFs at a given redshift. We argued in section \[combinedfit\] that, under the assumption that central and satellite galaxies follow the same SHM relation, the SMFs provide much stronger constraints on the SHM ratio than the clustering data. Thus we should be able to use our model to predict clustering as a function of stellar mass at any redshift.
Which survey for which redshift
-------------------------------
In order to constrain the SHM relation we have to first select observational stellar mass functions at the redshifts we want to investigate. Because of the trade-off between surveying large areas and obtaining deep samples, measurements of the SMF at high redshift tend to suffer from limited dynamic range. Therefore it is important to think about how the constraints on our four SHM function parameters arise from the observations.
The characteristic mass $M_1$ and the maximum SHM ratio $(m/M)_0$ mostly depend on galaxies and halos of intermediate mass. The high mass slope $\gamma$ is fixed by the number of massive galaxies since these live in the massive halos. On the other hand, the low mass slope $\beta$ is set by the number of low mass galaxies since these live in the low mass halos.
For a survey with a fixed area on the sky, the observed volume is smaller for low redshifts $(z\lesssim1)$ than for high redshifts. In order to compute the SMF at high galaxy masses, the observed volume has to be relatively large, as massive galaxies are rare. Thus for low redshifts one has to choose a wide survey (large area) to determine the SMF for massive galaxies and properly constrain $\gamma$. Constraining the SMF at the low mass end requires a high level of completeness for low mass galaxies, which are very faint objects. Hence we have to choose a deep survey that can detect faint galaxies in order to constrain $\beta$.
Taking these considerations into account, we choose the stellar mass functions presented in @drory2004 to constrain the parameters $M_1$, $(m/M)_0$ and $\gamma$ at low redshifts. The authors derive the SMFs using MUNICS which is a wide area, medium-deep survey selected in the $K$ band. The detection limit is $K\approx19.5$ and the subsample the authors use covers $0.28 {\rm~deg}^2$. We apply our method using these mass functions and take the three parameters from that analysis.
However, the MUNICS survey is not deep enough to detect galaxies that are fainter than the characteristic mass of the SMF (the knee) and thus is not sufficient to constrain the parameter $\beta$. To constrain $\beta$ we choose the SMFs derived in @fontana2006. This work is based on the GOODS-MUSIC sample, a multicolor catalogue extracted from the survey conducted over the Chandra Deep Field South. The catalogue is selected in the $z_{850}$ and $K$ bands, covers an area of $143.2 {\rm~arcmin}^2$, and is complete to a typical magnitude of $K\approx23.5$. We apply our method using the SMFs computed with the $z_{850}$ band selected sample and take the parameter $\beta$ from that analysis.
For high redshift $(z\gtrsim1)$ we use the SMFs presented in @fontana2006 to constrain all four parameters. For high redshifts, the volume of a redshift bin becomes large enough to sample massive galaxies, and therefore the GOODS-MUSIC sample is sufficient to constrain $\gamma$.
We convert all SMFs which use a Salpeter initial mass function to the Kroupa/Chabrier initial mass function.
Evolution of the parameters {#s:parevo}
---------------------------
[c cc cc ccc cc]{} 0.0 & 11.88 & 0.02 & 0.0282 & 0.0005 & 1.06 & 0.05 & 0.05 & 0.56 & 0.00\
0.5 & 11.95 & 0.24 & 0.0254 & 0.0047 & 1.37 & 0.22 & 0.27 & 0.55 & 0.17\
0.7 & 11.93 & 0.23 & 0.0215 & 0.0048 & 1.18 & 0.23 & 0.28 & 0.48 & 0.16\
0.9 & 11.98 & 0.24 & 0.0142 & 0.0034 & 0.91 & 0.16 & 0.19 & 0.43 & 0.12\
1.1 & 12.05 & 0.18 & 0.0175 & 0.0060 & 1.66 & 0.26 & 0.31 & 0.52 & 0.40\
1.5 & 12.15 & 0.30 & 0.0110 & 0.0044 & 1.29 & 0.25 & 0.32 & 0.41 & 0.41\
1.8 & 12.28 & 0.27 & 0.0116 & 0.0051 & 1.53 & 0.33 & 0.41 & 0.41 & 0.41\
2.5 & 12.22 & 0.38 & 0.0130 & 0.0037 & 0.90 & 0.20 & 0.24 & 0.30 & 0.30\
3.5 & 12.21 & 0.19 & 0.0101 & 0.0020 & 0.82 & 0.72 & 1.16 & 0.46 & 0.21\
\[t:pararedshifttab\]
Having selected the observational SMFs for a set of different redshifts, we fit the four free parameters $M_1$, $(m/M)_0$, $\beta$ and $\gamma$ to the observations. The errors on the parameters are derived in a similar way as explained in section \[s:probdis\], but instead of using confidence intervals we have fitted a Gaussian to the probability distributions of $M_1$, $(m/M)_0$ and $\gamma$ and a lognormal to the probability distribution of $\beta$.
Figure \[f:fig12\] shows the observed and the model stellar mass functions for different redshifts (indicated at the top of each panel). The values of the resulting four parameters for the different redshifts are shown in Table \[t:pararedshifttab\] and the redshift evolution is plotted in Figure \[f:fig13\]. The characteristic mass $M_1$ grows with increasing redshift, while the normalization of the SHM ratio $(m/M)_0$ becomes smaller with increasing redshift. This means that there is less stellar content in a halo of a given mass at a higher redshift.
The high mass slope $\gamma$ can be constrained only weakly. This is due to the limitation of the available galaxy surveys. As the area of the survey is small, the volume in which galaxies are detected is limited, and thus massive galaxies are very rare. This results in large error bars for the SMF for massive galaxies which propagate into the error bars of $\gamma$. The situation improves slightly for higher redshifts as the volume of higher redshift bins is larger and thus more massive galaxies can be observed. The value of $\gamma$ decreases with increasing redshift. For higher redshifts ($z>1$) the error bars on $\gamma$ become very large because of the limited area covered by the available deep surveys (in this case, GOODS). We leave it up to the reader to assess the reliability of our results at $z>1$ based on our quoted error bars.
The low mass slope $\beta$ seems to increase with redshift until $z\approx2$ and then drops to a low value. However, as the redshift increases it becomes more and more difficult to observe low mass galaxies which are very faint. Thus the high redshift values for $\beta$ are not very well constrained and perhaps not to be fully trusted. We therefore assume that $\beta$ grows with increasing redshift.
As we explained in Section \[beta\], $\beta$ is strongly related to the parameter $\alpha$ of the Schechter function. A small value of $\beta$ corresponds to a large absolute value of $\alpha$ while a large value of $\beta$ results in a low absolute value of $\alpha$. This would mean that for higher redshifts the stellar mass function would become shallower, in contradiction with observations (e.g. @fontana2006 show that the absolute value of $\alpha$ increases with redshift). However, one has to remember that the halo mass function also changes with redshift and becomes steeper. Thus the halo mass function steepens more than the SMF, so $\beta$ has to increase in order to compensate.
With the derived parameter values it becomes possible to interpolate and find the SHM ratio at any redshift. This is done by choosing a redshift-parameterization for each of the parameters.
As $M_1$ and $(m/M)_0$ do not change much above a redshift of $z>1.5$ we choose power laws for the redshift dependence: $$\log M_1(z) = \log M_1\vert_{z=0} \cdot(z+1)^{\mu} \;.$$ and $$\left( \frac{m}{M}\right)_0(z) = \left( \frac{m}{M}\right)_{z=0} \cdot(z+1)^{\nu} \;.$$ with the normalizations $M_0$ and $(m/M)_{z=0}$ and the slopes $\mu$ and $\nu$.
To parameterize $\gamma$ over redshift, a linear dependence would lead to a negative $\gamma$ at a certain redshift. Though this is not forbidden, it leads to a SHM ratio which would be increasing monotonically with halo mass which is inconsistent with feedback processes at the massive end. Hence we also choose a power-law parameterization for $\gamma$: $$\gamma(z) = \gamma_0 \cdot(z+1)^{\gamma_1} \;.$$ with the normalization $\gamma_0$ and the slope $\gamma_1$.
From Figure \[f:fig13\] we are not able to infer whether $\beta$ converges to a constant value. Thus we adopt a simple linear parameterization: $$\beta(z) = \beta_1 \cdot z + \beta_0 \;.$$ Note that we have also tried other parameterizations (constant $\beta$, decreasing $\beta$) but could not reproduce the observed stellar mass functions. Using the linear parameterization for $\beta$ and the power laws for the other parameters we were able to compute stellar mass functions that are in good agreement with the observed ones.
[l l l l l l l l l]{} & 11.88 & 0.019 & 0.0282 & -0.72 & 0.556 & -0.26 & 1.06 & 0.17\
$\pm$ & 0.01 & 0.002 & 0.0003 & 0.06 & 0.001 & 0.05 & 0.06 & 0.12\
\[t:redpars\]
A fit to the derived values presented in Table \[t:pararedshifttab\] yields the parameters given in Table \[t:redpars\]. As we do not fully trust the derived values of $\beta$ for $z\gtrsim2$ we neglect these two values and fit a line to the remaining values of $\beta$.
The stellar-to-halo mass relation for different redshifts {#s:diffred}
---------------------------------------------------------
Having developed a redshift dependent model of the stellar-to-halo mass relation we now test this model by computing interpolated stellar mass functions for different redshifts. For this we use the method described in section \[s:galaxies\]. However, now we do not use the parameters that have been derived at each redshift by fitting the model to the observations but we use the eight parameters of the redshift dependent SHM relation that have been derived in the previous section.
The resulting interpolated SMFs are compared to the observations (and the fitted mass functions) in Figure \[f:fig12\]. For $z\lesssim2$ we see excellent overall agreement, the interpolated mass functions mostly overlap with the error bars of the observations.
The SMFs for the high redshifts $z\gtrsim2$ are too low. The deviations are largest at the low mass end. However, if we look at Figure \[f:fig12\], we see that $\beta$ is higher than the derived value for the two highest redshifts which results in a low mass slope that is too shallow.
To compare the relation at different redshifts, we use the redshift dependent SHM relation with the eight parameters that have been derived in the previous section. Figure \[f:fig14\] plots stellar mass versus halo mass for different redshifts. The plot shows that at a fixed low halo mass (e.g. $M=10^{11}{\!\,{\rm M}_\odot}$), galaxies that live in such halos are more massive at low redshift ($m\sim10^9{\!\,{\rm M}_\odot}$ for $z=0$) than galaxies that live in a halo of the same mass at a higher redshift ($m\sim10^8{\!\,{\rm M}_\odot}$ for $z=2$). In contrast, massive halos contain more massive galaxies at high redshift, while at low redshifts the galaxies in massive halos have less mass. However, as halos also become more massive over time, one cannot identify a halo of a certain mass at high redshifts with a halo of the same mass at low redshifts. Thus the fact that at a given (high) halo mass the mass of the central galaxy is lower at present than at an earlier epoch does not imply that individual galaxies lose mass during their evolution. This only means that large halos accrete dark matter faster than large galaxies grow in stellar mass, while the growth of low mass halos is slower than that of the central galaxies they harbor [see also @conroy2008]. Because of its statistical nature, our model is not suitable for following the evolution of an individual galaxy through cosmic time. We also note that the SHM relation at the massive end ($M\gta 10^{13}{\!\,{\rm M}_\odot}$) undergoes very little evolution, which has also been found by @brown2008.
Clustering at higher redshift {#s:clustering}
-----------------------------
Having determined the SHM relation as a function of redshift we are now able to populate halos with galaxies at any redshift. We choose a set of redshifts and populate the halos with galaxies, deriving the stellar masses from the redshift dependent SHM relation. We divide these galaxies into six samples of different stellar mass between $\log m/{\!\,{\rm M}_\odot}=8.5$ and $11.5$. For each of these samples we compute the real space CF $\xi(r)$ by counting pairs in distance bins (equation \[corrfunc\]). This leads to six CFs for every selected redshift.
Figure \[f:fig15\] shows the CFs for six different redshifts as a function of stellar mass. We also plot the correlation function of dark matter at the respective redshifts for comparison. For all redshifts we see that massive galaxies are clustered more strongly than low mass galaxies. The higher the redshift, the more the CFs for different stellar masses differ. For high redshift, there are very few massive galaxies in our limited volume simulation box, and so the error bars become larger.
At low redshift ($z\lta 1$), observational measurements of stellar mass dependent galaxy clustering have recently been published using the VIMOS-VLT Deep Survey (VVDS) and the zCOSMOS Survey [@meneux2008; @meneux2009]. In order to compare our model predictions to these data, we compute correlation functions for the same stellar mass bins and thresholds and convert these to projected correlation functions as described in section \[constraining\]. Figure \[f:fig16\] plots the observed projected correlation functions (symbols) and the model predictions (lines) for different stellar mass bins or thresholds in three redshift bins for the zCOSMOS Survey and one redshift bin for the VVDS. There is good general agreement between the model and observations. The zCOSMOS clustering amplitude agrees very well with the model for $r_p<1{\rm~Mpc}$, but for $z<0.8$ deviates at larger distances and becomes higher than the prediction. As suggested by @meneux2009, this may be because the COSMOS field represents an overdense volume at these redshifts. In contrast, the VVDS clustering amplitudes are lower than those predicted by our model, leading to the speculation that perhaps the VVDS represents an underdense region.
The galaxy bias {#bias}
---------------
The bias of any object may be defined as the square root of the ratio between the CF of the object $\xi_o(r)$ and the CF of dark matter particles $\xi_{dm}(r)$: $$\label{bias1}
b(r)=\sqrt{\frac{ \xi_o(r) }{ \xi_{dm}(r) }}\quad.$$ Here we focus on the galaxy two-point CF $\xi_{gg}(r,m,z)$, which in addition to the distance between the galaxies also depends on the redshift and the stellar mass of the galaxies: $$\label{bias2}
b(r,m,z)=\sqrt{\frac{ \xi_{gg}(r,m,z) }{ \xi_{dm}(r,z) }}\quad.$$
From our predicted galaxy CFs, we compute the bias for every redshift and stellar mass by averaging between $r=2{\rm~Mpc}$ and $10{\rm~Mpc}$, where $b(r)$ is roughly a constant (as one can see from Figure \[f:fig15\], the scale dependence of the bias is quite weak). Figure \[f:fig17\] shows the redshift dependence of the bias. The symbols represent the averaged value of the bias while the solid lines correspond to a fit to the symbols. For this we have used a power law form: $$b(z)=b_0 (z+1)^{b_1}+b_2 \quad$$ where the parameters $b_0$, $b_1$, and $b_2$ are functions of stellar mass. The fit parameters are given in Table \[t:biasfit\].
[lccc]{} 8.5 - 9.0 & 0.062 $\pm$ 0.017 & 2.59 $\pm$ 0.18 & 1.025 $\pm$ 0.062\
9.0 - 9.5 & 0.074 $\pm$ 0.008 & 2.58 $\pm$ 0.26 & 1.039 $\pm$ 0.028\
9.5 - 10.0 & 0.042 $\pm$ 0.003 & 3.17 $\pm$ 0.05 & 1.147 $\pm$ 0.021\
10.0 - 10.5 & 0.053 $\pm$ 0.014 & 3.07 $\pm$ 0.17 & 1.225 $\pm$ 0.077\
10.5 - 11.0 & 0.069 $\pm$ 0.014 & 3.19 $\pm$ 0.13 & 1.269 $\pm$ 0.087\
11.0 - 11.5 & 0.173 $\pm$ 0.035 & 2.89 $\pm$ 0.20 & 1.438 $\pm$ 0.061\
\[t:biasfit\]
This shows that the bias at a fixed stellar mass increases with increasing redshift. Massive galaxies are biased more strongly than galaxies of lower mass at any redshift. We find that the bias of massive galaxies evolves more rapidly than that of low mass ones [cf. @white2007; @brown2008]. Since the bias of massive halos evolves more rapidly than that of low mass galaxies, this seems to be a feature of any model in which the SHM relation is monotonically increasing (i.e. the most massive galaxies reside in the most massive halos).
Conclusions {#s:conclusions}
===========
The goal of this paper is to characterize the relationship between the stellar masses of galaxies and the masses of the dark matter halos in which they live at low and high redshift, and to make predictions of stellar mass dependent galaxy clustering at high redshift.
We used a high-resolution $N$-body simulation and identified halos and subhalos. Halos and subhalos were populated with central and satellite galaxies using a parameterized SHM relation. For host halos the mass was given by the virial mass $M_{\rm vir}$ while for subhalos we used the maximum mass of the halo over its history $M_{\rm max}$ since we expect the stellar mass of the satellite galaxy to be more tightly linked to this quantity.
We described the ratio between stellar and halo mass by a function with four free parameters, a low-mass slope $\beta$, a characteristic mass $M_1$, a high-mass slope $\gamma$, and a normalization $(m/M)_0$. We fit for the values of these parameters by requiring that the observed galaxy SMF is reproduced. We find that the SHM function has a characteristic peak at $M_1\sim 10^{12}{\!\,{\rm M}_\odot}$, and declines steeply towards both smaller mass ($\beta \sim 1$) and less steeply towards larger mass halos ($\gamma \sim
0.6$). The physical interpretation of this behavior is the interplay between the various feedback processes that impact the star formation efficiency. Supernova feedback is more effective at reheating and expelling gas in low mass halos, while AGN feedback is more effective in high mass halos [e.g. @shankar2006; @croton06; @bower2006; @somerville2008]. In this picture, the characteristic mass $M_1$ is the halo mass where the efficiency of these two processes crosses.
We have thoroughly discussed the meaning of the parameters. We have also investigated the effects on the SHM relation that arise from introducing scatter to the relation. To do this we have added scatter drawn from a lognormal distribution with a typical variance of $\sigma_m=0.15$ to the SHM function. We showed that the impact of such a scatter on three of the four parameters is negligible, with a small but significant impact on the high-mass slope $\gamma$.
We showed that adding constraints from stellar mass dependent galaxy clustering did not change the values of our best-fit parameters. Put another way, the likelihood (here $\chi^2$) function for the clustering constraint is much “flatter” than that for the mass constraint, so adding the clustering constraint does not significantly change the distribution for the most likely (best-fit) parameter values. Fitting to the SMF only, we found that the observed projected CFs of galaxies for five samples of different stellar mass were reproduced well. This means that the clustering properties of galaxies are predominantly driven by the clustering of the halos and subhalos in which they reside. From this we concluded that our model can predict clustering as a function of stellar mass at any redshift.
In order to describe how galaxies of different masses populate host halos, we introduced the conditional mass function $\Phi(m\vert M)$, which yields the average number of galaxies with stellar masses in the range $m\pm{\rm d}m/2$ that live in a distinct halo of mass $M$. It is described by five parameters which are functions of halo mass. We divided the conditional mass function into a contribution from central galaxies (described by a lognormal distribution) and a contribution from satellite galaxies (described by a modified Schechter function). We computed the SMF in different halo mass bins and fitted the five parameters in each bin. Introducing halo mass dependent functions for every parameter and fitting these to the derived values of the parameters in the halo mass bins, we determined the halo mass dependence of the five parameters and thus fully described the conditional mass function. We also computed the occupation numbers of halos which give the average number of galaxies of a given stellar mass that live inside a halo of mass $M$.
We compared the results for our SHM function with those that have been derived using other approaches. These include other halo occupation type models, gravitational lensing and semi-analytic models. We showed that all methods yield consistent SHM relations.
Using SMFs at higher redshifts, we applied our model at earlier epochs of the universe. We thus constrained the SHM relation at a given set of redshifts between $z=0$ and $z\sim4$. This allowed us to study how the four parameters of the SHM function depend on redshift. For each parameter we introduced a redshift dependent function. We found that the characteristic mass increases with redshift while the normalization decreases with redshift. This indicates that there is less stellar content in halos at higher redshifts. As the halo mass function steepens more with redshift than the stellar mass function, the low mass slope increases with redshift. We present an eight parameter fitting function describing the redshift dependent SHM relation.
Using the SHM relation that we derived in this way, along with spatial information for halos from the $N$-body simulation, we predicted the high-redshift real space CFs for five stellar mass intervals. We find that for all redshifts, massive galaxies are more clustered than galaxies of lower mass. Using the real space CF of dark matter we calculated the galaxy bias as a function of distance, redshift and stellar mass. Averaging over spatial scale in an interval around $r\approx6\,\,{\rm Mpc}$, we demonstrated that the galaxy bias increases with redshift, and presented fitting formulae for the galaxy bias as a function of stellar mass and redshift. In a companion paper [@mostercv] we will use these bias results to present predictions for the cosmic variance $\sigma_c$ for galaxies of different stellar mass.
, E. F. & [de Jong]{}, R. S. 2001, , 550, 212
, A. A. & [Weinberg]{}, D. H. 2002, , 575, 587
Bertschinger, E. 2001, , 137, 1
, G. R., [Faber]{}, S. M., [Primack]{}, J. R., & [Rees]{}, M. J. 1984, , 311, 517
Bower, R. G., Benson, A. J., Malbon, R., Helly, J. C., Frenk, C. S., Baugh, C. M., Cole, S., & Lacey, C. G. 2006, , 370, 645
Boylan-Kolchin, M., Ma, C.-P., & Quataert, E. 2008, , 383, 93
Brown, M. J. I., et al. 2008, , 682, 937
Bryan, G. L., & Norman, M. L. 1998, , 495, 80
, M., [van den Bosch]{}, F. C., [More]{}, S., [Li]{}, R., [Mo]{}, H. J., & [Yang]{}, X. 2008, ArXiv e-prints
, R. G., [Yee]{}, H. K. C., [Ellingson]{}, E., [Abraham]{}, R., [Gravel]{}, P., [Morris]{}, S., & [Pritchet]{}, C. J. 1996, , 462, 32
, S., [Aragon-Salamanca]{}, A., [Frenk]{}, C. S., [Navarro]{}, J. F., & [Zepf]{}, S. E. 1994, , 271, 781
, C., [Wechsler]{}, R. H., & [Kravtsov]{}, A. V. 2006, , 647, 201
, C. & [Wechsler]{}, R. H. 2008, ArXiv e-prints
, A. 2006, , 365, 842
Croton, D. J., et al. 2006, , 365, 11
Drory, N., Bender, R., Feulner, G., Hopp, U., Maraston, C., Snigula, J., & Hill, G. J. 2004, , 608, 742
Drory, N., Salvato, M., Gabasch, A., Bender, R., Hopp, U., Feulner, G., & Pannella, M. 2005, , 619, L131
Elsner, F., Feulner, G., & Hopp, U. 2008, , 477, 503
, L. K., [Gottesman]{}, S. T., & [Hunter]{}, Jr., J. H. 1987, , 325, 779
Eyles, L. P., Bunker, A. J., Ellis, R. S., Lacy, M., Stanway, E. R., Stark, D. P., & Chiu, K. 2007, , 374, 910
, S. M. & [Efstathiou]{}, G. 1980, , 193, 189
Fontana, A., et al. 2006, , 459, 745
, N., [Weinberg]{}, D. H., & [Hernquist]{}, L. 1996, , 105, 19
Kauffmann, G., et al. 2003, , 341, 33
, G., [White]{}, S. D. M., & [Guiderdoni]{}, B. 1993, , 264, 201
Klypin, A., Gottl[ö]{}ber, S., Kravtsov, A. V., & Khokhlov, A. M. 1999, , 516, 530
Kravtsov, A. V., Berlind, A. A., Wechsler, R. H., Klypin, A. A., Gottl[ö]{}ber, S., Allgood, B., & Primack, J. R. 2004, , 609, 35
, P. 2001, , 322, 231
, R., [Seljak]{}, U., [Kauffmann]{}, G., [Hirata]{}, C. M., & [Brinkmann]{}, J. 2006, , 368, 715
, R., [Tasitsiomi]{}, A., [Seljak]{}, U., [Kravtsov]{}, A. V., & [Wechsler]{}, R. H. 2005, , 362, 1451
Mar[í]{}n, F. A., Wechsler, R. H., Frieman, J. A., & Nichol, R. C. 2008, , 672, 849
Meneux, B., et al. 2008, , 478, 299
Meneux, B., et al. 2009, arXiv:0906.1807
More, S., van den Bosch, F. C., & Cacciato, M. 2009a, , 392, 917
More, S., van den Bosch, F. C., Cacciato, M., Mo, H. J., Yang, X., & Li, R. 2009b, , 392, 801
Moster, B. P., Somerville, R. S., & Newman, J. A. 2009, in prep
, C., [Kauffmann]{}, G., [Jing]{}, Y. P., [White]{}, S. D. M., [B[ö]{}rner]{}, G., & [Cheng]{}, F. Z. 2006, , 368, 21
Lin, Y.-T., Mohr, J. J., & Stanford, S. A. 2003, , 591, 749
Lin, Y.-T., & Mohr, J. J. 2004, , 617, 879
, P., et al. 2001, , 328, 64
, P., et al. 2002, , 332, 827
, B., [Heavens]{}, A. F., & [Jimenez]{}, R. 2004, , 355, 764
, B., [Jimenez]{}, R., [Heavens]{}, A. F., & [Charlot]{}, S. 2007, , 378, 1550
, J. A. & [Smith]{}, R. E. 2000, , 318, 1144
, U. 2000, , 318, 203
Shankar, F., Lapi, A., Salucci, P., De Zotti, G., & Danese, L. 2006, , 643, 14
Sheth, R. K., & Tormen, G. 1999, , 308, 119
, R. S. & [Primack]{}, J. R. 1999, , 310, 1087
Somerville, R. S., Hopkins, P. F., Cox, T. J., Robertson, B. E., & Hernquist, L. 2008, , 391, 481
Spergel, D. N., et al. 2007, , 170, 377
, V., [White]{}, S. D. M., [Tormen]{}, G., & [Kauffmann]{}, G. 2001, , 328, 726
Springel, V., & Hernquist, L. 2003, , 339, 289
Springel, V. 2005, , 364, 1105
Springel, V., et al. 2005, , 435, 629
Tasitsiomi, A., Kravtsov, A. V., Wechsler, R. H., & Primack, J. R. 2004, , 614, 533
Tinker, J. L., Weinberg, D. H., Zheng, Z., & Zehavi, I. 2005, , 631, 41
, A. & [Ostriker]{}, J. P. 2006, , 371, 1173
van den Bosch, F. C., Yang, X., & Mo, H. J. 2003, , 340, 771
van den Bosch, F. C., et al. 2007, , 376, 841
, L., [Li]{}, C., [Kauffmann]{}, G., & [De Lucia]{}, G. 2006, , 371, 537
Wang, L., Li, C., Kauffmann, G., & De Lucia, G. 2007, , 377, 1419
, S. D. M. & [Rees]{}, M. J. 1978, , 183, 341
, M. 2001, , 321, 1
White, M., Zheng, Z., Brown, M. J. I., Dey, A., & Jannuzi, B. T. 2007, , 655, L69
Yan, H., Dickinson, M., Giavalisco, M., Stern, D., Eisenhardt, P. R. M., & Ferguson, H. C. 2006, , 651, 24
, X., [Mo]{}, H. J., & [van den Bosch]{}, F. C. 2003, , 339, 1057
Yang, X., Mo, H. J., Jing, Y. P., van den Bosch, F. C., & Chu, Y. 2004, , 350, 1153
Yang, X., Mo, H. J., van den Bosch, F. C., Pasquali, A., Li, C., & Barden, M. 2007, , 671, 153
Yang, X., Mo, H. J., & van den Bosch, F. C. 2008, , 676, 248
, D., [Smith]{}, R., [Frenk]{}, C., & [White]{}, S. D. M. 1993, , 405, 464
, I., et al. 2002, , 571, 172
Zehavi, I., et al. 2004, , 608, 16
, I., et al. 2005, , 630, 1
Zheng, Z., et al. 2005, , 633, 791
[^1]: We refer to virialized halos that are not subhalos of another halo as “distinct”.
[^2]: In an idealized situation, halo mass should increase monotonically with time until the halo becomes a subhalo, at which point the mass begins to decrease due to tidal stripping.
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'PSGM.bib'
title: 'Stochastic Proximal Methods for Non-Smooth Non-Convex Constrained Sparse Optimization: Supplementary Material'
---
Applications of non-smooth non-convex constrained sparse optimization
=====================================================================
All applications presented here optimize over a closed convex feasible region $C$. We take $h(w)$ as the indicator function $\delta_{C}(w)$, $$\delta_{C}(w)=
\begin{cases}
0 & \text{ if } w\in C\\
\infty & \text{ otherwise. }
\end{cases}$$ For a nonempty set $C$, ${\operatorname{prox}}_{\delta_C}(w)={\operatornamewithlimits{argmin}}_{u\in C}||u-w||_2$ is the projection onto the set, see for example [@beck2017 Theorem 6.24].
[*Sparse binary classification with outlier detection and fairness constraints:*]{}
We are given training data $\{x,y\}$ where $y=\{y^1,y^2,...,y^n\}$, $y^j\in\{-1,1\}$ is the label set, and $x=\{x^1,x^2,...,x^n\}$, $x^j\in \RR^{d'}$ is the feature set. In the application of classifying people, there may be sensitive attributes such as race or sex. Even if a sensitive attribute $x_a\in
x$ is removed from the feature set, our predictions may still be correlated to it, resulting in our model disproportionally treating a subset of the population unfairly. This is remedied by bounding the covariance between the sensitive attribute $x_a$ and the model output as done in [@zafar2017], $$\frac{1}{n}\left|\sum_{j=1}^n(x_a^j-\bar{x}_a)v^Tx_{-a}^j\right|\leq c,$$
where $\bar{x}_a$ is the mean of $x_a$, $x^j_{-a}$ is the $j^{th}$ feature vector with the sensitive attribute removed, $v^Tx_{-a}^j$ is our model output using decision variables $v\in \RR^{d'}$, and $c>0$ determines the maximum covariance tolerated.
We consider the smoothed 0-1 loss of [@zhao2010] as our loss function, $$\begin{aligned}
{6}\label{eq:lossf}
\LL(u)=\begin{cases}
0 & \text{ if } u>1 \\
\frac{1}{4}u^3-\frac{3}{4}u+\frac{1}{2} & \text{ if } -1\leq u\leq 1 \\
1 & \text{ otherwise }.\\
\end{cases}\end{aligned}$$ We implement outlier detection by the mean-shift method, modifying our prediction to $v^Tx_{-a}^j+z^j$, using decision variables $z\in \RR^n$ to reduce the loss incurred by outliers. It was shown in [@she2011] that the $l_1$ norm is not effective as a penalizer of $z$ when multiple outliers are present, which motivates the use of a non-convex regularizer. As all of the regularizers considered for $g(w)$ in Section 1 of the main text are separable, we are able to take $g(w)=g^1(v)+g^2(z)$, and are free to use different regularizers for $v$ and $z$. The classification problem is then solved by the following minimization, $$\begin{aligned}
{6}\label{eq:ap3}
&\min\limits_{v,z}&&\text{ }\frac{1}{n}\sum_{j=1}^n\LL(y^j(v^Tx_{-a}^j+z^j))+g^1(v)+g^2(z)\\
&\text{s.t.}&&\frac{1}{n}\left|\sum_{j=1}^n(x_a^j-\bar{x}_a)v^Tx_{-a}^j\right|\leq c.\nonumber\end{aligned}$$ The feasible region of decision variables $v$ can be rewritten as $$C=\left\{v :
\begin{matrix}
\hat{x}^Tv\leq c\\
-\hat{x}^Tv\leq c
\end{matrix}
\right\},$$ where $\hat{x}=\frac{1}{n}\sum_{j=1}^n(x_a^j-\bar{x}_a)x_{-a}^j$. The projection onto $C$ can be computed as $$P_C(v)=\begin{cases}
v-\frac{\hat{x}^Tv-c}{||\hat{x}||^2_2}\hat{x} & \text{ if } \hat{x}^Tv > c \\
v-\frac{\hat{x}^Tv+c}{||\hat{x}||^2_2}\hat{x} & \text{ if } -\hat{x}^Tv > c \\
v & \text{ else, } \\
\end{cases}$$ which uses the projection onto a halfspace [@beck2017 Lemma 6.26]. We take $g^1(v)$ equal to MCP. This function is separable with $g^1(v):=\sum_{i=1}^{d'}g^1_i(v_i)$, where for $\kappa_1, \nu_1>0$, $$g^1_i(v_i)=\kappa_1\int_0^{|v_i|}\max\left(0,1-\frac{u}{\nu_1\kappa_1}\right)du=
\begin{cases}
\kappa_1|v_i| - \frac{v^2_i}{2\nu_1} & \text{ if } |v_i|\leq \nu_1\kappa_1 \\
\nu_1\kappa_1^2/2 & \text{ if } |v_i|>\nu_1\kappa_1
\end{cases}$$
\[firstprop\] $g^1(v)$ is $\kappa_1\sqrt{d'}$-Lipschitz continuous.
Assume $v_i\geq 0$, over which $g^1_i(v_i)$ is differentiable and $\left|\frac{d g^1_i}{v_i}(v_i)\right|\leq \kappa_1$. Using the mean value theorem, for $u_i\geq
0$, $|g^1_i(u_i)-g^1_i(v_i)|\leq \kappa_1|u_i-w_i|$. Given the symmetry of $g^1_i(\cdot)$, this bound holds for all $u_i,w_i$, and $$\begin{aligned}
{6}
&&|g^1(u)-g^1(w)|&=\left|\sum_{i=1}^{d'}(g^1_i(u_i)-g^1_i(w_i))\right|\nonumber\\
&&&\leq \sum_{i=1}^{d'}|g^1_i(u_i)-g^1_i(w_i)|\nonumber\\
&&&\leq \kappa_1\sum_{i=1}^{d'}|u_i-w_i|\nonumber\\
&&&\leq \kappa_1\sqrt{d'}||u-w||_2\nonumber
\end{aligned}$$
As considered in [@she2011], we set $g^2(z)$ equal to SCAD, which is also separable. For $\kappa_2>0$ and $\nu_2>2$, $$g^2_i(z_i)=\kappa_2\int_0^{|z_i|}\min\left(1,\frac{\max(0,\nu_2\kappa_2-u)}{(\nu_2-1)\kappa_2}\right)du=
\begin{cases}
\kappa_2|z_i| & \text{ if } |z_i|\leq \kappa_2 \\
\frac{-z^2_i+2\nu_2\kappa_2|z_i|-\kappa_2^2}{2(\nu_2-1)} & \text{ if } \kappa_2<|z_i|\leq
\nu_2\kappa_2\\
(\nu_2+1)\kappa_2^2/2 & \text{ if } |z_i|>\nu_2\kappa_2.
\end{cases}$$ Similarly to MCP, SCAD is symmetric and $|\operatorname*{dist}(0,\partial g^2_i(z_i))|\leq \kappa_2$. Using the same reasoning as in the proof of Property \[firstprop\], we get the following property.
\[thirprop\] $g^2(z)$ is $\kappa_2\sqrt{n}$-Lipschitz continuous.
For the closed form solutions of the proximal operators of MCP and SCAD see [@gong2013].
\[secprop\] The function $\frac{1}{n}\sum_{j=1}^n\LL(y^j(v^Tx_{-a}^j+z^j))$, with $\LL(\cdot)$ as defined in , is $\frac{3}{2n}\sum_{j=1}^n||[(x_{-a}^j)^T,1]||^2_2$-smooth in $v$ and $z$.
We see that $|\LL''(u)|\leq \frac{3}{2}$. Using the mean value theorem, $|\LL'(u)-\LL'(t)|\leq \frac{3}{2}|u-t|$. Composing it with the affine function $y^j(v^Tx_{-a}^j+z^j)$, the resulting function is $\frac{3}{2}||[(x_{-a}^j)^T,1]||^2_2$-smooth [@shalev2014 Claim 12.9]. We conclude that $\frac{1}{n}\sum_{j=1}^n\LL(y^j(v^Tx_{-a}^j+z^j))$ is $\frac{3}{2n}\sum_{j=1}^n||[(x_{-a}^j)^T,1]||^2_2$-smooth.
In the following two applications, it is assumed that $g(w)$ is taken as either MCP or SCAD.
[*Sparse portfolio optimization using S-shaped utility with loss aversion:*]{}
We assume there are $d$ risky assets with stochastic returns $r_i$, $i=1,...,d$, and an investor desires to place a fraction $w_i$ of their wealth into each asset $i$. Finding a sparse portfolio is desirable as trading fewer assets results in fewer transaction costs. Motivated by prospect theory [@kahneman1979], we assume the investor is risk adverse in gains (concave utility) and risk seeking in losses (convex utility). Our objective is to maximize the following exponential utility function $$\begin{aligned}
{6}\label{eq:eu}
F(w,r)=
\begin{cases}
\frac{1-e^{-\psi^1(\sum_{i=1}^dw_ir_i)}}{\psi^1} & \text{ if } \sum_{i=1}^dw_ir_i\geq 0\\
\frac{e^{\psi^2(\sum_{i=1}^dw_ir_i)}-1}{\psi^2} & \text{ otherwise, }
\end{cases}\end{aligned}$$ where $\psi^1,\psi^2>0$. This utility function has been considered in [@kobberling2005; @pirvu2012]. Choosing $\psi^1>\psi^2$ models loss aversion, where the investor has increased sensitivity to losses than to gains. Our optimization problem is then $$\begin{aligned}
{6}\label{eq:ap2}
&\max&&\text{ }\EE_r[F(w,r)]+g(w)\\
&\text{s.t.}&&\sum_{i=1}^dw_i\leq 1, w\geq 0,\nonumber\end{aligned}$$ where we assume there should be no short selling. In order to project onto the constraints let $w_{d+1}=1-\sum_{i=1}^d w_i$ and $C=\{w:\sum_{i=1}^{d+1}w_i=1, w\geq 0\}$. The projection onto the probability simplex can be achieved using a simple non-iterative algorithm such as found in [@wang2013].
We assume we have access to $n$ historical observations of $r$, $r^j$ for $j=1,...,n$. We take a distribution-free approach, optimizing directly over the observations, $$\begin{aligned}
{6}
f(w)=\frac{1}{n}\sum_{j=1}^n F(w,r^j)\label{ef}\end{aligned}$$
\[thirdprop\] The function $f(w)=\frac{1}{n}\sum_{j=1}^n F(w,r^j)$, where $F(w,r^j)$ is as defined in is $\frac{\max(\psi^1,\psi^2)}{n}\sum_{j=1}^n||r^j||^2_2$-smooth.
We first consider the univariate function $$\begin{aligned}
{6}
\hat{F}(u)=
\begin{cases}
\hat{F}_1(u)=\frac{1-e^{-\psi^1u}}{\psi^1} & \text{ if } u\geq 0\\
\hat{F}_2(u)=\frac{e^{\psi^2u}-1}{\psi^2} & \text{ otherwise }
\end{cases}
\end{aligned}$$ The first and second derivatives are $$\hat{F}'_1(u)=e^{-\psi^1u}\quad\hat{F}''_1(u)=-\psi^1e^{-\psi^1u}\quad
\hat{F}'_2(u)=e^{\psi^2u}\quad\hat{F}''_2(u)=\psi^2e^{\psi^2u}$$ We can see that $|\hat{F}''_1(u)|\leq \psi^1$ and $|\hat{F}''_2(u)|\leq \psi^2$ over their domains. Assume that $w,x\geq 0$ and $v,u\leq 0$. Using the mean value theorem, $$\begin{aligned}
{6}
|\hat{F}'_1(w)-F_1'(x)|\leq \psi^1|w-x|\label{l1}\\
|F_2'(v)-F_2'(u)|\leq \psi^2|v-u|\label{l2}
\end{aligned}$$ Given $w$ and $v$, we can take $u=\frac{-\psi^1}{\psi^2}w$ with $\hat{F}'_2(u)=\hat{F}'_1(w)$ and $x=\frac{-\psi^2}{\psi^1}v$ with $\hat{F}'_1(x)=\hat{F}'_2(v)$. Assume $\psi^1\geq\psi^2$, then $$\begin{aligned}
{6}
|\hat{F}'_1(w)-\hat{F}_2'(v)|&=|\hat{F}'_1(w)-\hat{F}_1'(x)|\nonumber\\
&\leq \psi^1|w-x|\nonumber\\
&\leq \psi^1|w-v|.\label{l3}
\end{aligned}$$ Assuming now $\psi^2\geq\psi^1$, then $$\begin{aligned}
{6}
|\hat{F}'_1(w)-\hat{F}_2'(v)|&=|\hat{F}_2'(u)-\hat{F}_2'(v)|\nonumber\\
&\leq \psi^2|u-v|\nonumber\\
&\leq \psi^2|w-v|,\label{l4}
\end{aligned}$$ From -, we conclude that $\hat{F}(u)$ is $\max(\psi^1,\psi^2)$-smooth. As shown in the proof of Property \[secprop\], since $F(w,r^j)$ is $\hat{F}(u)$ composed with the affine function $\sum_{i=1}^dw_ir^j_i$, it is $||r^j||^2_2\max(\psi^1,\psi^2)$-smooth and $f(w)$ is $\frac{\max(\psi^1,\psi^2)}{n}\sum_{j=1}^n||r^j||^2_2$-smooth.
[*Non-negative sparse principal component analysis:*]{}
The projection onto $C=\{w: ||w||_2\leq 1, w\geq 0\}$ has the explicit solution [@bauschke2018 Theorem 7.1] $$P_C(w)=\frac{\max(w,0)}{\max(||\max(w,0)||_2,1)}.$$
\[fourthprop\] Given a dataset $x\in \RR^{d\times n}$, the function $f(w)=-\frac{1}{2n}\sum_{j=1}^n(w^Tx_j)^2$ is $\frac{1}{n}||xx^T||_2$-smooth, where $||\cdot||_2$ is the spectral norm.
We can rewrite the function $f(w)$, $$f(w)=-\frac{1}{2n}\sum_{j=1}^n(w^Tx_j)^2=-\frac{1}{2n}w^T\left(\sum_{j=1}^nx_jx_j^T\right)w=-\frac{1}{2n}w^Txx^Tw.$$ $$\begin{aligned}
{6}
&\left\lVert\nabla f(w)-\nabla f(w')\right\lVert_2&=&\bigg\lVert
-\frac{1}{n}xx^Tw+\frac{1}{n}xx^Tw'\bigg\lVert_2\nonumber\\
&&\leq&\frac{1}{n}\big\lVert xx^T\big\lVert_2\big\lVert w-w'\big\lVert_2.\nonumber\end{aligned}$$
VRSPA requires that each $f_j(w)$ be $L$-smooth, so in our numerical experiments we took $L=\max\limits_j\big\lVert x_jx^T_j\big\lVert_2$. A similar approach can be taken when implementing the other applications.
Proof of Property 1
===================
[1]{} The following holds for $E^k_{\lambda}(w)$. $$\begin{aligned}
{6}
&E_{\lambda}^k(w)+h(w)\geq \tPhi_{\lambda}(w)\hspace{0 mm} \text{ for all
}w\in\RR^d\label{eq:e1}\\
&E_{\lambda}^k(w^k)+h(w^k)=\tPhi_{\lambda}(w^k)\label{eq:e3}\\
&E_{\lambda}^k(w) \text{ is }
L_{\lambda}:=\left(L+\frac{1}{\lambda}\right)-\text{smooth.}\label{eq:e2}
\end{aligned}$$
The Moreau envelope can be written as a difference of convex functions, $$\begin{aligned}
{6}
&e_{\lambda}g(w)&&=\frac{1}{2\lambda}||w||^2_2-D^{\lambda}(w).\label{fsg}
\end{aligned}$$ As the supremum of a set of affine functions, $D^{\lambda}(w)$ (14) is convex, and $\zeta^{\lambda}(w)$ attains the supremum of $D^{\lambda}(w)$ from (8). The difference between $E_{\lambda}^k(w)+h(w)$ and $\tPhi_{\lambda}(w)$ is the difference between $U^k_{\lambda}(w)$ and $e_{\lambda}g(w)$, so we only focus on showing that and hold between these two terms.
: As found in [@liu2017], for any $w,z\in \RR^d$, $$\begin{aligned}
{6}
&D^{\lambda}(w)-D^{\lambda}(z)&=&\sup_{x\in
\RR^d}\left(\frac{1}{\lambda}w^Tx-\frac{1}{2\lambda}||x||^2_2-g(x)\right)-\sup_{x\in
\RR^d}\left(\frac{1}{\lambda}z^Tx-\frac{1}{2\lambda}||x||^2_2-g(x)\right)\nonumber\\
&&\geq&\frac{1}{\lambda}w^T\zeta^{\lambda}(z)-\frac{1}{2\lambda}||\zeta^{\lambda}(z)||^2_2
-g(\zeta^{\lambda}(z))\nonumber\\
&&&-\left(\frac{1}{\lambda}z^T\zeta^{\lambda}(z)-\frac{1}{2\lambda}||\zeta^{\lambda}(z)||^2_2-g(\zeta^{\lambda}(z))\right)\nonumber\\
&&=&\frac{1}{\lambda}\zeta^{\lambda}(z)^T(w-z).\nonumber
\end{aligned}$$ Setting $z=w^k$, $$\begin{aligned}
{6}
&e_{\lambda}g(w)&&=\frac{1}{2\lambda}||w||^2_2-D^{\lambda}(w)\nonumber\\
&&&\leq\frac{1}{2\lambda}||w||^2_2-(D^{\lambda}(w^k)+\frac{1}{\lambda}\zeta^{\lambda}(w^k)^T(w-w^k))\nonumber\\
&&&=U^k_{\lambda}(w).\nonumber
\end{aligned}$$
: $U^k_{\lambda}(w^k)=\frac{1}{2\lambda}||w^k||^2_2-D^{\lambda}(w^k)=e_{\lambda}g(w^k)$ from .\
: $$\begin{aligned}
{6}
&\left\lVert\nabla E_{\lambda}^k(w)-\nabla E_{\lambda}^k(w')\right\lVert_2&=&\bigg\lVert\nabla
f(w)+\frac{1}{\lambda}\left(w-\zeta^{\lambda}(w^k)\right)-\left(\nabla
f(w')+\frac{1}{\lambda}\left(w'-\zeta^{\lambda}(w^k)\right)\right)\bigg\lVert_2\nonumber\\
&&\leq&(L+\frac{1}{\lambda})\lVert w-w'\rVert_2.\nonumber
\end{aligned}$$
Proof of Lemma 2
================
[2]{} For an initial value $w_1\in \RR^d$, $N\in\ZZ_{>0}$, and $\alpha,\theta \in \RR$, MBSPA generates $w^R$ satisfying the following bound. $$\begin{aligned}
{6}
&&\EE||\G^R_{\gamma, E}(w^R)||^2_2&\leq
\frac{(L+N^{\theta})}{N}\tilde{\DD}+\frac{6}{\lceil
N^{\alpha}\rceil}\sigma^2\nonumber
\end{aligned}$$ where $\tilde{\DD}=4(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*_{\lambda}))$ and $w^*_{\lambda}$ is a global minimizer of $\tPhi_{\lambda}(\cdot)$.
In order to prove this result, we require the following properties. The proof of Property \[eq:minib\] can be found in [@metel2019], which we include here for completeness.
\[eq:minib\] $$\begin{aligned}
{6}
&&\EE||\nabla A^k_{\lambda, M}(w^k,\xi^k)-\nabla
E^k_{\lambda}(w^k)||^2_2&\leq\frac{\sigma^2}{M}\nonumber
\end{aligned}$$
From (15) and (16), $\nabla A^k_{\lambda, M}(w^k,\xi^k)-\nabla
E^k_{\lambda}(w^k)=\frac{1}{M}\sum_{j=1}^{M}\nabla F(w^k,\xi^k_j)-\nabla
f(w^k)$. Taking the expectation of its squared norm, $$\begin{aligned}
{6}
&&\EE||\nabla A^k_{\lambda, M}(w^k,\xi^k)-\nabla
E^k_{\lambda}(w^k)||^2_2&=\EE||\frac{1}{M}\sum_{j=1}^{M}(\nabla F(w^k,\xi^k_j)-\nabla
f(w^k))||^2_2\nonumber\\
&&&=\frac{1}{M^2}\EE\sum_{i=1}^n\left(\sum_{j=1}^M\nabla F(w^k,\xi^k_j)_i-\nabla
f(w^k)_i\right)^2.\nonumber
\end{aligned}$$ For $j\neq l$, $\nabla F(w^k,\xi^k_j)_i-\nabla f(w^k)_i$ and $\nabla F(w^k,\xi^k_l)_i-\nabla
f(w^k)_i$ are independent random variables with zero mean. It follows that $$\begin{aligned}
{6}
&\EE[(\nabla F(w^k,\xi^k_j)_i-\nabla f(w^k)_i)(\nabla F(w^k,\xi^k_l)_i-\nabla
f(w^k)_i)]=\nonumber\\
&\EE[(\nabla F(w^k,\xi^k_j)_i-\nabla f(w^k)_i)]\EE[(\nabla F(w^k,\xi^k_l)_i-\nabla
f(w^k)_i)]=0,&\nonumber
\end{aligned}$$ and $$\begin{aligned}
{6}
&&\frac{1}{M^2}\EE\sum_{i=1}^n\left(\sum_{j=1}^M\nabla F(w^k,\xi^k_j)_i-\nabla
f(w^k)_i\right)^2&=\frac{1}{M^2}\EE\sum_{i=1}^n\sum_{j=1}^{M}(\nabla F(w^k,\xi^k_j)_i-\nabla
f(w^k)_i)^2\nonumber\\
&&&=\frac{1}{M^2}\sum_{j=1}^{M}\EE||\nabla F(w^k,\xi^k_j)-\nabla
f(w^k)||^2_2\leq\frac{\sigma^2}{M}\nonumber
\end{aligned}$$ using (6).
The proof of Property \[eq:psize\] can be found in [@ghadimi2016], which we include in our notation here for clarity.
\[eq:psize\] Let $w,s\in\RR^d$ and $\gamma>0$, then $$\begin{aligned}
{6}
&-\langle s,\P_{\gamma}(w,s)\rangle\leq
\frac{1}{\gamma}\left(h(w)-h({\operatorname{prox}}_{\gamma h}(w-\gamma
s))\right)-||\P_{\gamma}(w,s)||^2_2.\nonumber
\end{aligned}$$
By the optimality of ${\operatorname{prox}}_{\gamma h}(w-\gamma s)$ in (7), $$0\in -\P_{\gamma}(w,s)+s+\partial h({\operatorname{prox}}_{\gamma h}(w-\gamma s)).$$ Taking $p\in \partial h({\operatorname{prox}}_{\gamma h}(w-\gamma s))$ such that $0=-\P_{\gamma}(w,s)+s+p$, it follows that $$\begin{aligned}
{6}
&0&&=\langle-\P_{\gamma}(w,s)+s+p,\P_{\gamma}(w,s)\rangle\nonumber\\
&&&=\langle s+p,\P_{\gamma}(w,s)\rangle-||\P_{\gamma}(w,s)||^2_2\nonumber\\
&&&\leq \langle s,\P_{\gamma}(w,s)\rangle +\frac{1}{\gamma}\left(h(w)-h({\operatorname{prox}}_{\gamma
h}(w-\gamma
s))\right)-||\P_{\gamma}(w,s)||^2_2,\nonumber
\end{aligned}$$ where the inequality uses the convexity of $h$.
\[eq:none\] $$\begin{aligned}
{6}
&\||\G^k_{\gamma, A}(w^k)-\G^k_{\gamma, E}(w^k)||_2\leq||\nabla A^k_{\lambda,
M}(w^k,\xi^k)-\nabla
E^k_{\lambda}(w^k)||_2\nonumber
\end{aligned}$$
$$\begin{aligned}
{6}
&\||\G^k_{\gamma, A}(w^k)-\G^k_{\gamma, E}(w^k)||_2&=&||\frac{1}{\gamma}\left(w^k-{\operatorname{prox}}_{\gamma
h}(w^k-\gamma
\nabla A^k_{\lambda, M}(w^k,\xi^k))\right)\nonumber\\
&&&-\frac{1}{\gamma}\left(w^k-{\operatorname{prox}}_{\gamma h}(w^k-\gamma \nabla E^k_{\lambda
}(w^k))\right)||_2\nonumber\\
&&=&\frac{1}{\gamma}||{\operatorname{prox}}_{\gamma h}(w^k-\gamma \nabla E^k_{\lambda
}(w^k))-{\operatorname{prox}}_{\gamma h}(w^k-\gamma \nabla A^k_{\lambda, M}(w^k,\xi^k))||_2\nonumber\\
&&\leq&\frac{1}{\gamma}||w^k-\gamma\nabla E^k_{\lambda
}(w^k)-w^k+\gamma\nabla A^k_{\lambda, M}(w^k,\xi^k)||_2\nonumber\\
&&=&||\nabla A^k_{\lambda, M}(w^k,\xi^k))- \nabla E^k_{\lambda}(w^k))||_2,\nonumber\\
\end{aligned}$$
where the inequality holds due to the nonexpansivity of the proximal operator of proper closed convex functions [@beck2017 Theorem 6.42].
Given the smoothness of $E^k_{\lambda}(w)$ as shown in Property 1, $$\begin{aligned}
{6}
&E^k_{\lambda}(w^{k+1})&&\leq E^k_{\lambda}(w^k)+\langle\nabla
E^k_{\lambda}(w^k),w^{k+1}-w^k\rangle+\frac{L_{\lambda}}{2}||w^{k+1}-w^k||^2_2\nonumber\\
&&&=E^k_{\lambda}(w^k)+\langle\nabla
E^k_{\lambda}(w^k),-\gamma\G^k_{\gamma,
A}(w^k)\rangle+\frac{L_{\lambda}}{2}||\gamma\G^k_{\gamma, A}(w^k)||^2_2,\nonumber
\end{aligned}$$ from (17). Let $\delta_k:=\nabla A^k_{\lambda, M}(w^k,\xi^k)-\nabla
E^k_{\lambda}(w^k)$, then $$\begin{aligned}
{6}
&E^k_{\lambda}(w^{k+1})&&\leq E^k_{\lambda}(w^k)-\gamma\langle\nabla A^k_{\lambda,
M}(w^k,\xi^k),\G^k_{\gamma, A}(w^k)\rangle+\gamma\langle \delta_k,\G^k_{\gamma,
A}(w^k)\rangle+\frac{L_{\lambda}}{2}||\gamma\G^k_{\gamma,A}(w^k)||^2_2.\nonumber
\end{aligned}$$ Using Property \[eq:psize\] with $w=w^k$ and $s=\nabla A^k_{\lambda, M}(w^k,\xi^k)$, $$\begin{aligned}
{6}
&E^k_{\lambda}(w^{k+1})&&\leq
E^k_{\lambda}(w^k)+h(w^k)-h(w^{k+1})-\gamma||\G^k_{\gamma, A}(w^k)||^2_2+\gamma\langle
\delta_k,\G^k_{\gamma, A}(w^k)\rangle+\frac{L_{\lambda}\gamma^2}{2}||\G^k_{\gamma,
A}(w^k)||^2_2.\nonumber
\end{aligned}$$ Applying and , $$\begin{aligned}
{6}
&\tPhi_{\lambda}(w^{k+1})&&\leq
\tPhi_{\lambda}(w^k)-\gamma||\G^k_{\gamma, A}(w^k)||^2_2+\gamma\langle
\delta_k,\G^k_{\gamma, A}(w^k)\rangle+\frac{L_{\lambda}\gamma^2}{2}||\G^k_{\gamma,
A}(w^k)||^2_2\nonumber\\
&&&=
\tPhi_{\lambda}(w^k)+\left(\frac{L_{\lambda}\gamma^2}{2}-\gamma\right)||\G^k_{\gamma,
A}(w^k)||^2_2+\gamma\langle
\delta_k,\G^k_{\gamma, E}(w^k)\rangle+\gamma\langle \delta_k,\G^k_{\gamma,
A}(w^k)-\G^k_{\gamma,
E}(w^k)\rangle\nonumber\\
&&&\leq
\tPhi_{\lambda}(w^k)+\left(\frac{L_{\lambda}\gamma^2}{2}-\gamma\right)||\G^k_{\gamma,
A}(w^k)||^2_2+\gamma\langle
\delta_k,\G^k_{\gamma, E}(w^k)\rangle+\gamma||\delta_k||_2||\G^k_{\gamma, A}(w^k)-\G^k_{\gamma,
E}(w^k)||_2\nonumber\\
&&&\leq
\tPhi_{\lambda}(w^k)+\left(\frac{L_{\lambda}\gamma^2}{2}-\gamma\right)||\G^k_{\gamma,
A}(w^k)||^2_2+\gamma\langle
\delta_k,\G^k_{\gamma, E}(w^k)\rangle+\gamma||\delta_k||^2_2,\nonumber
\end{aligned}$$ where the last inequality uses Property \[eq:none\]. After $N$ iterations, $$\begin{aligned}
{6}
&&\left(\gamma-\frac{L_{\lambda}\gamma^2}{2}\right)\sum_{k=1}^N ||\G^k_{\gamma,
A}(w^k)||^2_2&\leq
\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^{N+1})+\gamma\sum_{k=1}^N\left(\langle
\delta_k,\G^k_{\gamma, E}(w^k)\rangle+||\delta_k||^2_2\right)\nonumber\\
&&&\leq \tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)+\gamma\sum_{k=1}^N\left(\langle
\delta_k,\G^k_{\gamma, E}(w^k)\rangle+||\delta_k||^2_2\right)\nonumber
\end{aligned}$$ It follows from (5) that for $w$ independent of $\xi^k$, $\EE[\nabla
A^k_{\lambda, M}(w,\xi^k)]=\nabla E^k_{\lambda}(w)$, and so $\EE[\delta_k]=0$. Taking the expectation of both sides, $$\begin{aligned}
{6}
&& \left(\gamma-\frac{L_{\lambda}}{2}\gamma^2\right)\sum_{k=1}^N\EE||\G^k_{\gamma,
A}(w^k)||^2_2\leq&
\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)+\gamma\sum_{k=1}^N\EE||\delta_k||^2_2\nonumber\\
&&\leq&
\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)+\gamma\frac{N}{M}\sigma^2,\nonumber
\end{aligned}$$ where the second inequality uses Property \[eq:minib\]. As we choose $R$ uniformly over $\{1,...,N\}$, $$\begin{aligned}
{6}
&&\EE||\G^R_{\gamma, A}(w^R)||^2_2&=\frac{1}{N}\sum_{k=1}^N\EE||\G^k_{\gamma,
A}(w^k)||^2_2\nonumber\\
&&&\leq\frac{1}{N\left(\gamma-\frac{L_{\lambda}}{2}\gamma^2\right)}\left(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)+\gamma\frac{N}{M}\sigma^2\right)\nonumber\\
&&&=\frac{2L_{\lambda}}{N}\left(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)\right)+\frac{2}{M}\sigma^2,\nonumber
\end{aligned}$$ where the final equality holds since $\gamma=\frac{1}{L_{\lambda}}$. $$\begin{aligned}
{6}
&&\EE||\G^R_{\gamma, E}(w^R)||^2_2&=\EE||\G^R_{\gamma, A}(w^R)+\G^R_{\gamma,
E}(w^R)-\G^R_{\gamma, A}(w^R)||^2_2 \nonumber\\
&&&\leq \EE\left(||\G^R_{\gamma, A}(w^R)||^2_2+2\langle
\G^R_{\gamma, A}(w^R),\G^R_{\gamma,E}(w^R)-\G^R_{\gamma, A}(w^R)\rangle+||\G^R_{\gamma,
E}(w^R)-\G^R_{\gamma, A}(w^R)||^2_2\right) \nonumber\\
&&&\leq2\EE||\G^R_{\gamma, A}(w^R)||^2_2+2\EE||\G^R_{\gamma,
E}(w^R)-\G^R_{\gamma, A}(w^R)||^2_2 \nonumber\\
&&&\leq
\frac{4L_{\lambda}}{N}\left(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)\right)+\frac{4}{M}\sigma^2+2\EE||\nabla
A^R_{\lambda, M}(w^R,\xi^R)-\nabla E^R_{\lambda}(w^R)||^2_2\nonumber\\
&&&\leq
\frac{4L_{\lambda}}{N}\left(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)\right)+\frac{4}{M}\sigma^2+\frac{2}{M}\sigma^2\nonumber\\
&&&=
\frac{4L_{\lambda}}{N}\left(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*)\right)+\frac{6}{M}\sigma^2\nonumber\\
&&&=\frac{(L+N^{\theta})}{N}\tilde{\DD}+\frac{6}{\lceil
N^{\alpha}\rceil}\sigma^2,\nonumber
\end{aligned}$$ where the second inequality uses Young’s inequality on the middle term.
Proofs of Properties 3 and 4
============================
[3]{} Assume that $g(w)$ is Lipschitz continuous with parameter $l$ and $\bar{\gamma}\geq \gamma$, then $$\begin{aligned}
{6}
&&\operatorname*{dist}(0,\G_{\bar{\gamma}}(\zeta^{\lambda}(w^k)))&\leq
&||\G^k_{\gamma, E}(w^k)||_2+2l\lambda\left(\frac{2}{\bar{\gamma}}+L\right).\nonumber
\end{aligned}$$
In order to prove Property 3 we require the following property. Its proof can be found in [@beck2017 Theorem 10.9] under a slightly different setting, so we present it here for clarity.
\[eq:mono\] For $\gamma^1\geq \gamma^2>0$ and any $w,s\in \RR^d$, $$||\P_{\gamma^1}(w,s)||_2\leq ||\P_{\gamma^2}(w,s)||_2.$$
For an arbitrary $v\in \RR^d$ and $\gamma>0$, ${\operatorname{prox}}_{\gamma h}(v)$ is the minimizer of $\frac{1}{2\gamma}||v-x||^2_2+h(x)$ and so $$\begin{aligned}
{6}
&\frac{1}{\gamma}(v-{\operatorname{prox}}_{\gamma h}(v))\in \partial h({\operatorname{prox}}_{\gamma h}(v)).\label{eq:subg}
\end{aligned}$$ By the definition of a subgradient of a convex function, for any $y\in \RR^d$, $$h(y)-h({\operatorname{prox}}_{\gamma h}(v))\geq\frac{1}{\gamma}\left<v-{\operatorname{prox}}_{\gamma h}(v),y-{\operatorname{prox}}_{\gamma
h}(v)\right>.$$ First let $\gamma=\gamma^1$, $v=w-\gamma^1 s$ and $y={\operatorname{prox}}_{\gamma^2h}(w-\gamma^2s)$, $$\begin{aligned}
{6}
&&&h({\operatorname{prox}}_{\gamma^2h}(w-\gamma^2s))-h({\operatorname{prox}}_{\gamma^1 h}(w-\gamma^1s))\label{eq:one}\\
&&\geq&\frac{1}{\gamma^1}\left<w-\gamma^1 s-{\operatorname{prox}}_{\gamma^1 h}(w-\gamma^1
s),{\operatorname{prox}}_{\gamma^2h}(w-\gamma^2s)-{\operatorname{prox}}_{\gamma^1
h}(w-\gamma^1 s)\right>\nonumber\\
&&=&\left<\P_{\gamma^1}(w,s)-
s,\gamma^1\P_{\gamma^1}(w,s)-\gamma^2\P_{\gamma^2}(w,s)\right>.\nonumber
\end{aligned}$$ Exchanging $\gamma^1$ and $\gamma^2$, letting $\gamma=\gamma^2$, $v=w-\gamma^2 s$ and $y={\operatorname{prox}}_{\gamma^1h}(w-\gamma^1s)$, $$\begin{aligned}
{6}
&h({\operatorname{prox}}_{\gamma^1h}(w-\gamma^1s))-h({\operatorname{prox}}_{\gamma^2
h}(w-\gamma^2s))\geq\left<\P_{\gamma^2}(w,s)-s,\gamma^2\P_{\gamma^2}(w,s)-\gamma^1\P_{\gamma^1}(w,s)\right>.\label{eq:two}
\end{aligned}$$ Adding inequalities and , $$\begin{aligned}
{6}
&0&\geq&\left<\P_{\gamma^1}(w,s)-
\P_{\gamma^2}(w,s),\gamma^1\P_{\gamma^1}(w,s)-\gamma^2\P_{\gamma^2}(w,s)\right>.\label{eq:thr}
\end{aligned}$$ Expanding and rearranging , $$\begin{aligned}
{6}
&\gamma^1||\P_{\gamma^1}(w,s)||^2_2+\gamma^2||\P_{\gamma^2}(w,s)||^2_2&\leq&
(\gamma^1+\gamma^2)\left<\P_{\gamma^1}(w,s),\P_{\gamma^2}(w,s)\right>\label{eq:csin}\\
&&\leq&(\gamma^1+\gamma^2)||\P_{\gamma^1}(w,s)||_2||\P_{\gamma^2}(w,s)||_2,\nonumber
\end{aligned}$$ using the Cauchy-Schwarz inequality. Assume $||\P_{\gamma^1}(w,s)||_2>0$, otherwise the property trivially holds, and set $t=||\P_{\gamma^2}(w,s)||_2/||\P_{\gamma^1}(w,s)||_2$. Inequality can now be written as $$\begin{aligned}
{6}
&\gamma^1+\gamma^2t^2-(\gamma^1+\gamma^2)t&\leq&0.\nonumber
\end{aligned}$$ The roots of the left hand side function occur at $t=1$ and $t=\frac{\gamma^1}{\gamma^2}$, so for the inequality to hold, $$1\leq t\leq \frac{\gamma^1}{\gamma^2},$$ which includes the desired inequality, $$||\P_{\gamma^1}(w,s)||_2\leq ||\P_{\gamma^2}(w,s)||_2.$$
Given that $\zeta^{\lambda}(w)$ is a minimizer of $\frac{1}{2\lambda}||w-x||^2_2+g(x)$ from (8), $$\begin{aligned}
{6}
&\frac{1}{\lambda}(w-\zeta^{\lambda}(w))\in \partial g(\zeta^{\lambda}(w)).
\end{aligned}$$ It follows that $$\begin{aligned}
{6}
&\operatorname*{dist}(0,\G_{\bar{\gamma}}(\zeta^{\lambda}(w^k)))&\leq&
||\P_{\bar{\gamma}}(\zeta^{\lambda}(w^k),\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))||_2\nonumber\\
&&=&||\P_{\bar{\gamma}}(\zeta^{\lambda}(w^k),\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))\nonumber\\
&&&+\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))-\P_{\bar{\gamma}}(w^k,\nabla E^k_{\lambda}(w^k))||_2\nonumber\\
&&\leq&||\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))||_2\nonumber\\
&&&+||\P_{\bar{\gamma}}(\zeta^{\lambda}(w^k),\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))-\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))||_2.\nonumber
\end{aligned}$$ Given that $||\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))||_2\leq ||\P_{\gamma}(w^k,\nabla
E^k_{\lambda}(w^k))||_2=||\G^k_{\gamma, E}(w^k)||_2$ from Property \[eq:mono\], $$\begin{aligned}
{6}
&\operatorname*{dist}(0,\G_{\bar{\gamma}}(\zeta^{\lambda}(w^k)))&\leq&||\G^k_{\gamma, E}(w^k)||_2\nonumber\\
&&&+||\P_{\bar{\gamma}}(\zeta^{\lambda}(w^k),\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))-\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))||_2.\label{comb1}
\end{aligned}$$ Focusing on the second term, $$\begin{aligned}
{6}
&||\P_{\bar{\gamma}}(\zeta^{\lambda}(w^k),\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))-\P_{\bar{\gamma}}(w^k,\nabla
E^k_{\lambda}(w^k))||_2\nonumber\\
=&\left\Vert\frac{1}{\bar{\gamma}}\left(\zeta^{\lambda}(w^k)-{\operatorname{prox}}_{\bar{\gamma}
h}(\zeta^{\lambda}(w^k)-\bar{\gamma}(\nabla
f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k))))\right)\right.\nonumber\\
&\left.-\frac{1}{\bar{\gamma}}\left(w^k-{\operatorname{prox}}_{\bar{\gamma}
h}(w^k-\bar{\gamma} (\nabla
f(w^k)+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k))))\right)\right\Vert_2\nonumber\\
\leq&\frac{1}{\bar{\gamma}}||\zeta^{\lambda}(w^k)-w^k||_2+\frac{1}{\bar{\gamma}}||{\operatorname{prox}}_{\bar{\gamma}h}
(w^k-\bar{\gamma}
(\nabla f(w^k)+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k))))\nonumber\\
&-{\operatorname{prox}}_{\bar{\gamma}
h}(\zeta^{\lambda}(w^k)-\bar{\gamma}
(\nabla f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k))))||_2\nonumber\\
\leq&\frac{1}{\bar{\gamma}}||\zeta^{\lambda}(w^k)-w^k||_2+\frac{1}{\bar{\gamma}}||w^k-\bar{\gamma}
(\nabla f(w^k)+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k)))\nonumber\\
&-(\zeta^{\lambda}(w^k)-\bar{\gamma}
(\nabla f(\zeta^{\lambda}(w^k))+\frac{1}{\lambda}(w^k-\zeta^{\lambda}(w^k))))||_2\nonumber\\
\leq&\frac{2}{\bar{\gamma}}||\zeta^{\lambda}(w^k)-w^k||_2+||
\nabla f(\zeta^{\lambda}(w^k))-\nabla f(w^k)||_2\nonumber\\
\leq&\frac{2}{\bar{\gamma}}||\zeta^{\lambda}(w^k)-w^k||_2+L||\zeta^{\lambda}(w^k)-w^k||_2,\label{comb2}
\end{aligned}$$ where the second inequality follows from the nonexpansivity of the proximal operator. In order to bound $||\zeta^{\lambda}(w^k)-w^k||_2$, recall from (12) that $$\begin{aligned}
{6}
&&g(w)&\geq e_{\lambda}g(w)\nonumber\\
&&&=\frac{1}{2\lambda}||w-\zeta^{\lambda}(w)||_2^2+g(\zeta^{\lambda}(w)).\nonumber
\end{aligned}$$ Rearranging and using the Lipschitz continuity of $g(w)$, $$\begin{aligned}
{6}
&&\frac{1}{2\lambda}||w-\zeta^{\lambda}(w)||_2^2&\leq
g(w)-g(\zeta^{\lambda}(w))\nonumber\\
&&&\leq l||w-\zeta^{\lambda}(w)||_2\nonumber\\
&&||w-\zeta^{\lambda}(w)||_2&\leq 2l\lambda.\label{comb3}
\end{aligned}$$ Using -, $$\begin{aligned}
{6}
&\operatorname*{dist}(0,\G_{\bar{\gamma}}(\zeta^{\lambda}(w^k)))&\leq&||\G^k_{\gamma,
E}(w^k)||_2+\frac{2}{\bar{\gamma}}||\zeta^{\lambda}(w^k)-w^k||_2+L||\zeta^{\lambda}(w^k)-w^k||_2\nonumber\\
&&\leq&||\G^k_{\gamma, E}(w^k)||_2+2l\lambda\left(\frac{2}{\bar{\gamma}}+L\right).\nonumber
\end{aligned}$$
[4]{} Let $w^*$ be a global minimizer of $\Phi(\cdot)$ and let $w^*_{\lambda}$ be a global minimizer of $\tPhi_{\lambda}(\cdot)$. Assume that $g(w)$ is Lipschitz continuous with parameter $l$, then $$\tPhi_{\lambda}(w)-\tPhi_{\lambda}(w^*_{\lambda})\leq
\Phi(w)-\Phi(w^*) + \frac{l^2\lambda}{2}.$$
$$\begin{aligned}
{6}
&\tPhi_{\lambda}(w)-\tPhi_{\lambda}(w^*_{\lambda})-\Phi(w)+\Phi(w^*)&=&e_{\lambda}g(w)-
f(w_{\lambda}^*) -e_{\lambda}g(w_{\lambda}^*)-h(w_{\lambda}^*)\nonumber\\
&&&-g(w)+f(w^*)+g(w^*)+h(w^*)\nonumber\\
&&\leq&- f(w_{\lambda}^*) -e_{\lambda}g(w_{\lambda}^*)-h(w_{\lambda}^*)+f(w^*)
+g(w^*)+h(w^*)\nonumber\\
&&\leq&- f(w_{\lambda}^*) -e_{\lambda}g(w_{\lambda}^*)-h(w_{\lambda}^*)+f(w_{\lambda}^*)
+g(w_{\lambda}^*)+h(w_{\lambda}^*)\nonumber\\
&&=&g(w_{\lambda}^*)-e_{\lambda}g(w_{\lambda}^*),\nonumber
\end{aligned}$$
where the first inequality follows from (12). For any $w$, $$\begin{aligned}
{6}
&&e_{\lambda}g(w)&=\frac{1}{2\lambda}||w-\zeta^{\lambda}(w)||_2^2+g(\zeta^{\lambda}(w))\nonumber\\
&&g(w)-e_{\lambda}g(w)&=g(w)-g(\zeta^{\lambda}(w))-\frac{1}{2\lambda}||w-\zeta^{\lambda}(w)||_2^2\nonumber\\
&&&\leq l||w-\zeta^{\lambda}(w)||_2-\frac{1}{2\lambda}||w-\zeta^{\lambda}(w)||_2^2.\nonumber
\end{aligned}$$ The right-hand side is maximized when $||w-\zeta^{\lambda}(w)||_2=l\lambda$, giving the desired result, $$\begin{aligned}
{6}\label{eq:propK1} &g(w)-e_{\lambda}g(w)\leq \frac{l^2\lambda}{2}.
\end{aligned}$$
Proof of Lemma 7
================
[7]{} For an initial value $\tilde{w}_1$, $N\in\ZZ_{>0}$, and $\alpha,\theta\in\RR$, VRSGA generates $w^R_T$ satisfying the following bound. $$\begin{aligned}
{6}
&&\EE\left[||\G^{R,T}_{\gamma,
E}(w^R_T)||^2_2\right]&\leq\tilde{\DD}\frac{L+(Sm)^{\theta}}{Sm}\nonumber
\end{aligned}$$ where $\tilde{\DD}=36(\tPhi_{\lambda}(\tilde{w}^1)-\tPhi_{\lambda}(w^*_{\lambda}))$ and $w^*_{\lambda}$ is a global minimizer of $\tPhi_{\lambda}(\cdot)$.
In order to prove this result, we require the following properties. The proof of Property \[vrprop1\] can be found in [@li2018], which we include here in our notation for clarity.
\[vrprop1\] Consider arbitrary $w,s,z\in \RR^d$, and $w^+={\operatorname{prox}}_{\gamma h}(w-\gamma s)$, $$\begin{aligned}
{6}
&E^{k,t}_{\lambda}(w^+)+h(w^+)&\leq& E^{k,t}_{\lambda}(z)+h(z)+\langle\nabla
E^{k,t}_{\lambda}(w)-s,w^+-z\rangle+\frac{L_{\lambda}}{2}||w^+-w||^2_2+\frac{L_{\lambda}}{2}||z-w||^2_2\nonumber\\
&&&-\frac{1}{\gamma}\langle w^+-w,w^+-z\rangle.\nonumber
\end{aligned}$$
As was done in the proof of Property \[eq:psize\], let us take $p\in \partial h({\operatorname{prox}}_{\gamma
h}(w-\gamma s))$ such that $0=-\P_{\gamma}(w,s)+s+p=\frac{1}{\gamma}(w^+-w)+s+p$. It follows by the convexity of $h(\cdot)$ that $$\begin{aligned}
{6}\label{conin}
&h(w^+)&&\leq h(z)+\langle p,w^+-z\rangle\nonumber\\
&&&=h(z)-\bigg\langle\frac{1}{\gamma}(w^+-w)+s,w^+-z\bigg\rangle.
\end{aligned}$$ Adding with the following two inequalities, which come from the smoothness of $E^{k,t}_{\lambda}(w)$ and $-E^{k,t}_{\lambda}(w)$, see Property 1, proves the result. $$\begin{aligned}
{6} &&E^{k,t}_{\lambda}(w^+)&\leq E^{k,t}_{\lambda}(w)+\langle\nabla
E^{k,t}_{\lambda}(w),w^+-w\rangle+\frac{L_{\lambda}}{2}||w^+-w||^2_2\nonumber\\
&&-E^{k,t}_{\lambda}(z)&\leq -E^{k,t}_{\lambda}(w)+\langle-\nabla
E^{k,t}_{\lambda}(w),z-w\rangle+\frac{L_{\lambda}}{2}||z-w||^2_2\nonumber
\end{aligned}$$
\[yineq\] For vectors $w$, $x$, $z$, and $\beta>0$, $$\begin{aligned}
{6}
&&||w-x||^2_2&\leq
(1+\beta)||w-z||^2_2+\left(1+\frac{1}{\beta}\right)||z-x||^2_2.\nonumber
\end{aligned}$$
$$\begin{aligned}
{6}
&&||w-x||^2_2&=||w-z+z-x||^2_2\nonumber\\
&&&\leq \left(||w-z||_2+||z-x||_2\right)^2\nonumber\\
&&&=||w-z||^2_2+2||w-z||_2||z-x||_2+||z-x||^2_2\nonumber\\
&&&\leq||w-z||^2_2+\left(\beta||w-z||^2_2+\frac{1}{\beta}||z-x||^2_2\right)+||z-x||^2_2\nonumber\\
&&&=(1+\beta)||w-z||^2_2+\left(1+\frac{1}{\beta}\right)||z-x||^2_2,\nonumber
\end{aligned}$$
where the second inequality uses Young’s inequality.
Let $\hat{w}^k_{t+1}={\operatorname{prox}}_{\gamma h}(w^k_t-\gamma \nabla E^{k,t}_{\lambda}(w^k_t))$, with $w^+=w^k_{t+1}$, $w=w^k_t$, $s=V^k_t$, and $z=\hat{w}^k_{t+1}$ in Property \[vrprop1\] to get the inequality $$\begin{aligned}
{6}
&E^{k,t}_{\lambda}(w^k_{t+1})+h(w^k_{t+1})&\leq&
E^{k,t}_{\lambda}(\hat{w}^k_{t+1})+h(\hat{w}^k_{t+1})+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\nonumber\\
&&&\frac{L_{\lambda}}{2}||w^k_{t+1}-w^k_t||^2_2+\frac{L_{\lambda}}{2}||\hat{w}^k_{t+1}-w^k_t||^2_2-\frac{1}{\gamma}\langle
w^k_{t+1}-w^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle.\label{vr:ineq2}
\end{aligned}$$ In addition, let $w^+=\hat{w}^k_{t+1}$, $w=w^k_t$, $s=\nabla E^{k,t}_{\lambda}(w^k_t)$, and $z=w^k_t$ in Property \[vrprop1\] to get $$\begin{aligned}
{6}
&E^{k,t}_{\lambda}(\hat{w}^k_{t+1})+h(\hat{w}^k_{t+1})&\leq&
E^{k,t}_{\lambda}(w^k_t)+h(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-\nabla
E^{k,t}_{\lambda}(w^k_t),\hat{w}^k_{t+1}-w^k_{t+1}\rangle+\frac{L_{\lambda}}{2}||\hat{w}^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&+\frac{L_{\lambda}}{2}||w^k_t-w^k_t||^2_2-\frac{1}{\gamma}\langle
\hat{w}^k_{t+1}-w^k_t,\hat{w}^k_{t+1}-w^k_t\rangle\nonumber\\
&&=&E^{k,t}_{\lambda}(w^k_t)+h(w^k_t)+\left(\frac{L_{\lambda}}{2}-\frac{1}{\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2.\label{vr:ineq3}
\end{aligned}$$ Adding and , $$\begin{aligned}
{6}
&E^{k,t}_{\lambda}(w^k_{t+1})+h(w^k_{t+1})&\leq&
E^{k,t}_{\lambda}(w^k_t)+h(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\frac{L_{\lambda}}{2}||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&-\frac{1}{\gamma}\langle
w^k_{t+1}-w^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+
\left(L_{\lambda}-\frac{1}{\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2.\nonumber
\end{aligned}$$ From and , $$\begin{aligned}
{6}
&\tPhi_{\lambda}(w^k_{t+1})&\leq& \tPhi_{\lambda}(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\frac{L_{\lambda}}{2}||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&-\frac{1}{\gamma}\langle
w^k_{t+1}-w^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+
\left(L_{\lambda}-\frac{1}{\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2.\label{keyineq}
\end{aligned}$$ Plugging $\langle w^k_{t+1}-w^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle=\frac{1}{2}\left(||
w^k_{t+1}-w^k_t||^2_2+||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2-||\hat{w}^k_{t+1}-w^k_t||^2_2\right)$ into and rearranging, $$\begin{aligned}
{6}
&\tPhi_{\lambda}(w^k_{t+1})&\leq& \tPhi_{\lambda}(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\left(\frac{L_{\lambda}}{2}-\frac{1}{2\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&-\frac{1}{2\gamma}||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2+
\left(L_{\lambda}-\frac{1}{2\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2.\label{byi}
\end{aligned}$$ Focusing on the term $-\frac{1}{2\gamma}||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2$, we apply Property \[yineq\] with $w=w^k_{t+1}$, $x=w^k_t$, and $z=\hat{w}^k_{t+1}$. After rearranging, $$\begin{aligned}
{6}
&&-(1+\beta)||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2&\leq
-||w^k_{t+1}-w^k_t||^2_2+\left(1+\frac{1}{\beta}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2\nonumber\\
&&-\frac{1}{2\gamma}||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2&\leq
-\frac{1}{(1+\beta)2\gamma}
||w^k_{t+1}-w^k_t||^2_2+\frac{\left(1+\frac{1}{\beta}\right)}{(1+\beta)2\gamma}||\hat{w}^k_{t+1}-w^k_t||^2_2.\nonumber
\end{aligned}$$ Choosing $\beta=3$, $$\begin{aligned}
{6}
&&-\frac{1}{2\gamma}||w^k_{t+1}-\hat{w}^k_{t+1}||^2_2&\leq
-\frac{1}{8\gamma}
||w^k_{t+1}-w^k_t||^2_2+\frac{1}{6\gamma}||\hat{w}^k_{t+1}-w^k_t||^2_2.\nonumber
\end{aligned}$$ Using this inequality in , $$\begin{aligned}
{6}
&\tPhi_{\lambda}(w^k_{t+1})&\leq& \tPhi_{\lambda}(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\left(\frac{L_{\lambda}}{2}-\frac{1}{2\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&-\frac{1}{8\gamma} ||w^k_{t+1}-w^k_t||^2_2+\frac{1}{6\gamma}||\hat{w}^k_{t+1}-w^k_t||^2_2+
\left(L_{\lambda}-\frac{1}{2\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2\nonumber\\
&&=& \tPhi_{\lambda}(w^k_t)+\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle+\left(\frac{L_{\lambda}}{2}-\frac{5}{8\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&+\left(L_{\lambda}-\frac{1}{3\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2\nonumber\\
&&\leq& \tPhi_{\lambda}(w^k_t)+\gamma ||\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t||^2_2+\left(\frac{L_{\lambda}}{2}-\frac{5}{8\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\nonumber\\
&&&+\left(L_{\lambda}-\frac{1}{3\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2,\label{nonexp}
\end{aligned}$$ where the last inequality holds since $$\begin{aligned}
{6}
&\langle\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t,w^k_{t+1}-\hat{w}^k_{t+1}\rangle&\leq& ||\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t||_2||w^k_{t+1}-\hat{w}^k_{t+1}||_2\nonumber\\
&&=& ||\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t||_2||{\operatorname{prox}}_{\gamma h}(w^k_t-\gamma
V^k_t)-{\operatorname{prox}}_{\gamma h}(w^k_t-\gamma \nabla E^{k,t}_{\lambda}(w^k_t))||_2\nonumber\\
&&\leq& \gamma||\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t||^2_2\nonumber
\end{aligned}$$ using the Cauchy-Schwarz inequality and the nonexpansivity of the proximal operator of $h$. Taking the expectation of both sides of , $$\begin{aligned}
{6}
&\EE\tPhi_{\lambda}(w^k_{t+1})&\leq& \EE\left[\tPhi_{\lambda}(w^k_t)+\gamma
||\nabla E^{k,t}_{\lambda}(w^k_t)-V^k_t||^2_2
+\left(\frac{L_{\lambda}}{2}-\frac{5}{8\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\right.\nonumber\\
&&&\left.+\left(L_{\lambda}-\frac{1}{3\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2\right].\label{bvb}
\end{aligned}$$ Focusing on $\EE\left[||\nabla E^{k,t}_{\lambda}(w^k_t)-V^k_t||^2_2\right]$, from (15) and the definition of $V^k_t$ found in Algorithm 2, $\nabla
E^{k,t}_{\lambda}(w^k_t)-V^k_t=\nabla f(w)-(\frac{1}{b}\sum_{j\in
I}\left(\nabla f_j(w^k_t)-\nabla f_j(\tilde{w}^k)\right)+G^k)$. After rearranging, $$\begin{aligned}
{6}
&&\EE||\nabla E^{k,t}_{\lambda}(w^k_t)-V^k_t||^2_2&=\EE||\frac{1}{b}\sum_{j\in I}\left(\nabla
f_j(\tilde{w}^k)-\nabla f_j(w^k_t)\right)-\left(G^k-\nabla f(w^k_t)\right)||^2_2\nonumber\\
&&&=\frac{1}{b^2}\EE\sum_{j\in I}||\nabla f_j(\tilde{w}^k)-\nabla f_j(w^k_t)-\left(G^k-\nabla
f(w^k_t)\right)||^2_2\nonumber\\
&&&\leq\frac{1}{b^2}\EE\sum_{j\in I}||\nabla f_j(\tilde{w}^k)-\nabla
f_j(w^k_t)||^2_2\nonumber\\
&&&\leq\frac{L^2}{b}\EE||\tilde{w}^k-w^k_t||^2_2.\nonumber
\end{aligned}$$ As the expectation of the squared norm of a sum of independent random variables with zero mean, the second equality holds using the same reasoning found in Property \[eq:minib\], and the first inequality holds since $\EE||x-\EE[x]||^2_2\leq \EE||x||^2_2$ for any random variable $x$. Using this bound in , $$\begin{aligned}
{6}
&\EE\tPhi_{\lambda}(w^k_{t+1})&\leq& \EE\left[\tPhi_{\lambda}(w^k_t)+\gamma
\frac{L^2}{b}||\tilde{w}^k-w^k_t||^2_2+
\left(\frac{L_{\lambda}}{2}-\frac{5}{8\gamma}\right)||w^k_{t+1}-w^k_t||^2_2\right.\nonumber\\
&&&\left.+\left(L_{\lambda}-\frac{1}{3\gamma}\right)||\hat{w}^k_{t+1}-w^k_t||^2_2\right]\nonumber\\
&&\leq&\EE\left[\tPhi_{\lambda}(w^k_t)+\frac{L_{\lambda}}{6b}||\tilde{w}^k-w^k_t||^2_2-
\frac{13L_{\lambda}}{4}||w^k_{t+1}-w^k_t||^2_2-L_{\lambda}||\hat{w}^k_{t+1}-w^k_t||^2_2\right]\nonumber\\
&&=&\EE\left[\tPhi_{\lambda}(w^k_t)+\frac{L_{\lambda}}{6b}||\tilde{w}^k-w^k_t||^2_2-
\frac{13L_{\lambda}}{4}||w^k_{t+1}-w^k_t||^2_2-\frac{1}{36L_{\lambda}}||\G^{k,t}_{\gamma
E}(w^k_t)||^2_2\right],\label{byi2}
\end{aligned}$$ where the last two lines use $\gamma=\frac{1}{6L_{\lambda}}$. Focusing on $-\frac{13L_{\lambda}}{4}||w^k_{t+1}-w^k_t||^2_2$, we apply Property \[yineq\] with $w=w^k_{t+1}$, $x=\tilde{w}^k$, and $z=w^k_{t}$, $$\begin{aligned}
{6}
&&(1+\beta)||w^k_{t+1}-w^k_{t}||^2_2&\geq
||w^k_{t+1}-\tilde{w}^k||^2_2-\left(1+\frac{1}{\beta}\right)||w^k_{t}-\tilde{w}^k||^2_2\nonumber\\
&&-\frac{13L_{\lambda}}{4}||w^k_{t+1}-w^k_{t}||^2_2&\leq
-\frac{13L_{\lambda}}{4(1+\beta)}||w^k_{t+1}-\tilde{w}^k||^2_2+\frac{13L_{\lambda}\left(1+\frac{1}{\beta}\right)}{4(1+\beta)}||w^k_{t}-\tilde{w}^k||^2_2.\nonumber
\end{aligned}$$ Setting $\beta=2t-1$, $$-\frac{13L_{\lambda}}{4}||w^k_{t+1}-w^k_{t}||^2_2\leq
-\frac{13L_{\lambda}}{8t}||w^k_{t+1}-\tilde{w}^k||^2_2+\frac{13L_{\lambda}}{8t-4}||w^k_{t}-\tilde{w}^k||^2_2.$$ Applying this bound in , $$\begin{aligned}
{6}
&\EE\tPhi_{\lambda}(w^k_{t+1})&\leq&
\EE\left[\tPhi_{\lambda}(w^k_t)+\left(\frac{L_{\lambda}}{6b}+\frac{13L_{\lambda}}{8t-4}\right)||\tilde{w}^k-w^k_t||^2_2-
\frac{13L_{\lambda}}{8t}||w^k_{t+1}-\tilde{w}^k||^2_2\right.\nonumber\\
&&&\left.-\frac{1}{36L_{\lambda}}||\G^{k,t}_{\gamma, E}(w^k_t)||^2_2\right].
\end{aligned}$$ Summing over $t$, $$\begin{aligned}
{6}
&\EE\tPhi_{\lambda}(w^k_{m+1})&\leq&
\EE\left[\tPhi_{\lambda}(w^k_1)+\sum_{t=1}^m\left(\frac{L_{\lambda}}{6b}+\frac{13L_{\lambda}}{8t-4}\right)||\tilde{w}^k-w^k_t||^2_2\right.\nonumber\\
&&&\left.-\sum_{t=1}^{m}\frac{13L_{\lambda}}{8t}||w^k_{t+1}-\tilde{w}^k||^2_2-\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right].\nonumber
\end{aligned}$$ As $\tilde{w}^k=w^k_1$ and $||w^k_{m+1}-\tilde{w}^k||^2_2\geq 0$, $$\begin{aligned}
{6}
&\EE\tPhi_{\lambda}(w^k_{m+1})&\leq&\EE\left[
\tPhi_{\lambda}(w^k_1)+\sum_{t=2}^m\left(\frac{L_{\lambda}}{6b}+\frac{13L_{\lambda}}{8t-4}\right)||\tilde{w}^k-w^k_t||^2_2\right.\nonumber\\
&&&\left.-\sum_{t=1}^{m-1}\frac{13L_{\lambda}}{8t}||w^k_{t+1}-\tilde{w}^k||^2_2-\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right]\nonumber\\
&&=&
\EE\left[\tPhi_{\lambda}(w^k_1)+\sum_{t=1}^{m-1}\left(\frac{L_{\lambda}}{6b}+\frac{13L_{\lambda}}{8t+4}-\frac{13L_{\lambda}}{8t}\right)||w^k_{t+1}-\tilde{w}^k||^2_2\right.\nonumber\\
&&&\left.-\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,E}(w^k_t)||^2_2\right]\nonumber\\
&&\leq &
\EE\left[\tPhi_{\lambda}(w^k_1)+\sum_{t=1}^{m-1}\left(\frac{L_{\lambda}}{6b}-\frac{L_{\lambda}}{2t^2}\right)||w^k_{t+1}-\tilde{w}^k||^2_2-\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right]\nonumber\\
&&\leq &
\EE\left[\tPhi_{\lambda}(w^k_1)-\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right],\nonumber
\end{aligned}$$ where the last inequality holds since $6b=6m^2>2(m-1)^2\geq 2t^2$ for $t=1,...,m-1$. This summation can be equivalently written as $$\begin{aligned}
{6}
&&\EE\tPhi_{\lambda}(\tilde{w}^{k+1})&\leq
\EE\tPhi_{\lambda}(\tilde{w}^k)-\EE\left[\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right]\nonumber\\
&&\EE\left[\frac{1}{36L_{\lambda}}\sum_{t=1}^m||\G^{k,t}_{\gamma, E}(w^k_t)||^2_2\right]&\leq
\EE\tPhi_{\lambda}(\tilde{w}^k)-\EE\tPhi_{\lambda}(\tilde{w}^{k+1})\nonumber\\
&&\EE\left[\frac{1}{36L_{\lambda}}\sum_{k=1}^S\sum_{t=1}^m||\G^{k,t}_{\gamma,
E}(w^k_t)||^2_2\right]&\leq
\tPhi_{\lambda}(\tilde{w}^1)-\EE\tPhi_{\lambda}(\tilde{w}^{S+1})\nonumber\\
&&&\leq\tPhi_{\lambda}(\tilde{w}^1)-\tPhi_{\lambda}(w^*_{\lambda})\nonumber\\
&&\EE\left[||\G^{R,T}_{\gamma,
E}(w^R_T)||^2_2\right]&\leq\frac{36L_{\lambda}\left(\tPhi_{\lambda}(\tilde{w}^1)-\tPhi_{\lambda}(w^*_{\lambda})\right)}{Sm}\nonumber\\
&&&=\tilde{\DD}\frac{L+(Sm)^{\theta}}{Sm}.\nonumber
\end{aligned}$$
Proofs of Theorem 8 and Corollary 9
===================================
[8]{} Assume that $g(w)$ is Lipschitz continuous with parameter $l$ and $\bar{\gamma}=\frac{1}{N^{\tau}}$ for $\tau\leq\theta$. The output $\bar{w}^R_T$ of VRSPA satisfies the following inequality. $$\begin{aligned}
{6}
&\EE\left[\operatorname*{dist}(0,\G_{\bar{\gamma}}(\bar{w}^R_T))\right]&\leq&\sqrt{\frac{\left(L+(Sm)^{\theta}\right)
\left(\DD+18l^2(Sm)^{-\theta}\right)}{Sm}}+\frac{2l}{(Sm)^{\theta}}\left(2N^{\tau}+L\right),\nonumber
\end{aligned}$$ where $\DD=36(\Phi(w^1)-\Phi(w^*))$ and $w^*$ is a global minimizer of $\Phi(\cdot)$.
The proof follows what was done to prove Theorem 5. We first verify that $\bar{\gamma}=\frac{1}{N^{\tau}}\geq\frac{1}{(Sm)^{\tau}}\geq
\frac{1}{(Sm)^{\theta}}\geq\frac{1}{6(L+(Sm)^{\theta})}=\gamma$. From Property 3, $$\operatorname*{dist}(0,\G_{\bar{\gamma}}(\bar{w}^R_T))\leq
||\G^{R,T}_{\gamma, E}(w^R_T)||_2+2l\lambda\left(\frac{2}{\bar{\gamma}}+L\right).$$ Taking its expectation, $$\begin{aligned}
{6}
&\EE\left[\operatorname*{dist}(0,\G_{\bar{\gamma}}(\bar{w}^R_T)))\right]&\leq&
\EE[||\G^{R,T}_{\gamma,E}(w^R_T)||_2]+2l\lambda\left(\frac{2}{\bar{\gamma}}+L\right)\nonumber\\
&&\leq&\sqrt{\EE\left[||\G^{R,T}_{\gamma,
E}(w^R_T)||^2_2\right]}+\frac{2l}{(Sm)^{\theta}}\left(2N^{\tau}+L\right)\nonumber\\
&&\leq&\sqrt{\frac{\left(L+(Sm)^{\theta}\right)
\left(\DD+18l^2(Sm)^{-\theta}\right)}{Sm}}+\frac{2l}{(Sm)^{\theta}}\left(2N^{\tau}+L\right),\nonumber
\end{aligned}$$ where the third inequality follows from Lemma 7 and using Property 4, $$\begin{aligned}
{6}
&&\tilde{\DD}=36(\tPhi_{\lambda}(w^1)-\tPhi_{\lambda}(w^*_{\lambda}))&\leq 36(\Phi(w^1)-\Phi(w^*))
+18l^2\lambda\nonumber\\
&&&=\DD+\frac{18l^2}{(Sm)^{\theta}}\nonumber
\end{aligned}$$
[9]{} Assume that $g(w)$ is Lipschitz continuous with parameter $l$. To obtain an $\epsilon$-accurate solution (9) using VRSPA, the gradient call complexity is $O(n^{\frac{2}{3}}\epsilon^{-3})$ and the proximal operator complexity is $O(\epsilon^{-3})$ choosing $\alpha=\theta=\frac{1}{3}$, and $\tau=0$.
From Theorem 8, $$\begin{aligned}
{6}
&\EE\left[\operatorname*{dist}(0,\G_{\bar{\gamma}}(\bar{w}^R_T))\right]&\leq&\sqrt{\frac{\left(L+(Sm)^{\frac{1}{3}}\right)
\left(\DD+18l^2(Sm)^{\frac{-1}{3}}\right)}{Sm}}+\frac{2l}{(Sm)^{\frac{1}{3}}}\left(2+L\right)\nonumber\\
&&&=O((Sm)^{-\frac{1}{3}}).\nonumber
\end{aligned}$$ An $\epsilon$-accurate solution will require at most $Sm=O(\epsilon^{-3})$ iterations. Two proximal operations are required each iteration, so the proximal operator complexity is $O(\epsilon^{-3})$. The number of gradient calls after $Sm$ iterations is $$Sn+Smb=Sm\frac{n}{\lceil n^{\frac{1}{3}}\rceil}+Sm\lceil
n^{\frac{1}{3}}\rceil^2=O(n^{\frac{2}{3}}\epsilon^{-3}).$$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Three observations on commutators of Singular Integral Operators with BMO functions are exposed, namely
The already known subgaussian local decay for the commutator, namely $$\frac{1}{|Q|}\left|\left\{x\in Q\, : \, |[b,T](f\chi_Q)(x)|>M^2f(x)t\right\}\right|\leq c e^{-\sqrt{ct\|b\|_{BMO}}}$$ is sharp, since it cannot be better than subgaussian.
It is not possible to obtain a pointwise control of the commutator by a finite sum of sparse operators defined by $L\log L$ averages.
Motivated by the conjugation method for commutators, it is shown the failure of the following endpoint estimate, if $w\in A_p\setminus A_1$ then $$\left\| wM\left(\frac{f}{w}\right)\right\|_{L^1(\mathbb{R}^n)\rightarrow L^{1,\infty}(\mathbb{R}^n)}=\infty.$$
address:
- 'Carlos Pérez, Department of Mathematics, University of the Basque Country UPV/EHU, IKERBASQUE, Basque Foundation for Science, Bilbao and BCAM, Basque Center for Applied Mathematics, Bilbao, Spain. '
- 'Israel P. Rivera-Ríos, IMUS & Departamento de Análisis Matemático, Universidad de Sevilla, Sevilla, Spain'
author:
- Carlos Pérez
- 'Israel P. Rivera-Ríos'
bibliography:
- 'refs.bib'
title: Three observations on commutators of Singular Integral operators with BMO functions
---
[^1]
Introduction
============
The purpose of this paper is to present some observations concerning commutators of singular integral operators with BMO functions. These operators were introduced by Coifman, Rochberg and Weiss in [@MR0412721] as a tool to extend the classical factorization theorem for Hardy spaces in the unit circle to $\mathbb R^n$. These operators are defined by the expression $$T_b f(x)=\int_{\mathbb R^n} (b(x)-b(y))K(x,y)f(y)\,dy,$$ where $K$ is a kernel satisfying the standard Calderón-Zygmund estimates and where $b$, the “symbol” of the operator, is a locally integrable function. Of course, these are special cases of the more general commutators given by the expression $$T_b= [b,T]=M_{b}\circ T -T \circ M_{b}$$ where $T$ is any operator and $M_{b}$ is the multiplication operator $M_{b}f=b \cdot f$.
The classical well known result from [@MR0412721] establishes that $[b, T]$ is a bounded operator on $L^{p}( \mathbb R^{n} )$, $1<p<\infty$, when the symbol $b$ is a $BMO$ function. We state this result.
\[Thm:LpBddness\] Let $T$ be a singular integral operator and $b$ a $BMO$ function. The commutator $T_b$ is bounded on $L^{p}(\mathbb{R}^{n})$ for every $1<p<\infty$.
In the same paper it is shown that $b\in BMO$ is also a necessary condition namely, if the commutators $[b,R_{j}]$, $j=1,\cdots,n$ of $b$ with the Riesz transforms $R_{j}$ are bounded on $L^{p}(\mathbb{R}^{n})$ for some $p\in(1,\infty)$ and every $j\in\{1,2,\dots,n\}$ then $b\in BMO$.
None of the different proofs of this result follows the usual scheme of the classical Calderón-Zygmund theory for proving the $L^{p}(\mathbb{R}^{n})$ boundedness of singular integral operators $T$. Two proofs of Theorem \[Thm:LpBddness\] can be found in [@MR0412721]. The first and main one in that paper is based on methods involving techniques similar to those used in [@MR0380244] to understand the Calderón commutator. As far as we know this approach has not been so influential. However, the second proof, based on the so called conjugation method from operator theory, has been widely used. In fact, it is quite surprising that this proof was postponed to the end of the paper since it turns out to be highly interesting. Indeed, the method shows the intimate connection between these commutators and the $A_{p}$ theory of weights. Furthermore, this proof can be applied to general linear operators, not only for Singular Integral Operators. As a sample we will point out the following particular $L^{2}$ case:
\[Thm:Alt\] Suppose that $T$ is a linear operator such that $$T:L^{2}(w)\longrightarrow L^{2}(w)$$ for every $w\in A_{2}$. Then for every $b\in BMO$, $$[b,T]:L^{2}(\mathbb{R}^{n})\longrightarrow L^{2}(\mathbb{R}^{n}).$$
The method of proof can be carried out in more generality as shown in [@MR1211818]. The key initial argument of the proof is that we can write $[b,T]$ as a complex integral operator using the Cauchy integral theorem as follows $$[b,T]f=\left.\frac{d}{dz}e^{zb}T(fe^{-zb})\right|_{z=0}=
\frac{1}{2\pi i}\int_{|z|=\varepsilon}
\frac{T_z(f)}{z^2}\,dz\, , \quad \varepsilon>0$$ where $$z\to T_z(f) := e^{zb} T\left(\frac{f}{e^{zb}}\right) \qquad z \in \mathbb{C}.$$ This is called the “conjugation” of $T$ by $e^{zb}$ and the terminology comes most probably from group theory. Now, if $\|\cdot\|$ is a norm we can apply Minkowski inequality: $$\left\| [b,T]f \right\| \leq \frac{1}{2\pi\,\varepsilon} \,\sup_{|z|=\varepsilon}
\left\| T_z(f)\right\| \qquad \varepsilon>0.$$ The effectiveness of the method can be checked in the modern context of weighted $L^{p}$ estimates. Indeed, the method produces very optimal bounds of the operator norm as shown in [@MR2869172] (see also [@MR3092729]).
This method reveals the role played by the following operation: $$f
\to T_w(f) := w\,T\left(\frac{f}{w}\right)$$ where $w$ is a weight which, in this context, is an $A_p$ weight. Indeed, this is the case by the well known key property of the BMO class, if $p>1$ and $b\in BMO$ then there is a small $\varepsilon_0$, such that $e^{tb}\in A_p$, for any real number $t$ such that $|t|<\varepsilon_0$. These operators were already studied by B. Muckenhoupt and R. Wheeden in the 70’s and by E. Sawyer in the 80’s. Some of the problems they left open were solved in [@MR2172941]. A consequence of the main result of [@MR2172941] is that if $w\in A_1$ then $T_w$ is of weak type $(1,1)$, namely $$\left\| T_w \right\|_{L^{1}(\mathbb{R}^{n}) \to L^{1, \infty}(\mathbb{R}^{n})}<\infty$$ with bound depending upon the $A_1$ constant of $w$. However, we will exhibit examples of weights $w\in A_{p}\setminus A_{1}$ in Section \[negativeEstimateConjugationMethod\] for which $T_w$ is [**not**]{} of weak type $(1,1)$, namely $$\left\| T_w \right\|_{L^{1}(\mathbb{R}^{n}) \to L^{1, \infty}(\mathbb{R}^{n})} =\infty.$$ This shows that the case $w \in A_{1}$ is specially relevant. Perhaps, this phenomenon can be explained by the fact that the conjugation method is closely attached to commutators with BMO functions which are [**not**]{} of weak type $(1,1)$ as observed in [@MR1317714]. Indeed, the conjugation method works due to the property, already mentioned, that if $p>1$ and $b\in BMO$ then $e^{tb}\in A_p$ for small values of $t$. However, this property turns out to be false in the case $p=1$. The lack of the weak type $(1,1)$ property for commutators is replaced by a $L\log L$ inequality like (\[LogL\]) below and not better.
There is another proof of Theorem \[Thm:LpBddness\] based on the use of the sharp maximal function of C. Fefferman and E. Stein which has also been very influential. It seems that it was first discovered by J. O. Strömberg as mentioned by S. Janson in [@MR524754] (see also [@T] pp. 417-419) The proof relies on combining the following key pointwise estimate $$\label{eq:St}
M^{\sharp}([b,T]f)\leq c\|b\|_{BMO}\left(M_{r}\,\left(Tf\right)+M_{s}\left(f\right)\right)$$ where $1<r,s<\infty$ and $M_r(f)= M(|f|^r)^{1/r}$ together with the classical Fefferman-Stein inequality: $$\|M(f)\|_{L^{p}}\leq c\|M^{\sharp}(f)\|_{L^{p}}.$$ Here we use standard notation, $M$ is the Hardy-Littlewood maximal function and $M^{\sharp}$ is the sharp maximal function. The $L^{p}$ boundedness of $M$ and $T$ yields the alternative proof of Theorem \[Thm:LpBddness\]. Proceeding in the same way we obtain the corresponding estimates for $A_{p}$ weights.
This approach was considered by S. Bloom in ([@B]) extending in an interesting way Theorem \[Thm:LpBddness\] but only on the real line.
Let $\mu,\lambda\in A_{p}$ and let $H$ be the Hilbert transform: $$[b,H]\,:\,L^{p}(\mu)\longrightarrow L^{p}(\lambda)$$ where $\nu=\mu^{\frac{1}{p}}\lambda^{-\frac{1}{p}}$ if and only if $$\label{newBMO}
\|b\|_{BMO(\nu)}=\sup_{Q}\frac{1}{\nu(Q)}\int_{Q}\left|b-b_{Q}\right|<\infty.$$
The power of the pointwise estimate (\[eq:St\]) is reflected in many situations, for instance in [@MR1142721], where similar results were derived for commutators of strongly singular integral with symbol in the new $BMO$ class (\[newBMO\]) (see also [@HLW; @2015arXiv150903769H] for an alternative approach based on dyadic shifts). However, estimate (\[eq:St\]) is not sharp enough for many purposes and much better results can be obtained with the following variation: $$\label{EstimateJFA}
M_{\delta}^{\sharp}([b,T]f) \leq c\|b\|_{BMO}\left(M_{\varepsilon}\left(Tf\right)+M^{2}\left(f\right)\right) \quad 0<\delta<\varepsilon<1$$ where $M^{2}$ stands for $M\circ M$ (see [@MR1317714]). Here, the key difference is that we are considering small parameters $\delta$ and $\varepsilon$. The estimate is sharp since $M^2$ cannot be replaced by the (pointwise) smaller operator $M$. Indeed, otherwise these commutators would be of weak type $(1,1)$ but, as we mentioned above, this is not the case [@MR1317714] where it is shown that commutators satisfy the following “$L\log L$” type estimate, $$\label{LogL}
w\left(\left\{ x\in\mathbb{R}^{n}\,:\,\left|\left[b,T\right]f(x)\right|>\lambda\right\} \right)\leq c\,\int_{\mathbb{R}^{n}}\Phi\left(\frac{|f|}{\lambda}\|b\|_{BMO}\right)wdx \quad \lambda>0,$$ where $w\in A_{1}$, $\Phi(t)=t\log\left(e+t\right)$ and where $c>0$ depends upon the $A_1$ constant. This shows that these commutators are “more singular” than Calderón-Zygmund operators. The original proof of (\[LogL\]) follows from the key pointwise (\[EstimateJFA\]) combined with a good-$\lambda$ type argument, but an alternative proof was obtained by the first author and G. Pradolini in [@MR1827073] with the bonus that non $A_{\infty}$ weights can be considered. This argument is based on a variation of the classical scheme used to prove the weak type $(1,1)$ for Calderón-Zygmund operators. The statement of the result is the following.
\[Thm:EndPCual\]Let $T$ be a Calderón-Zygmund operator and $b\in BMO$. If $w$ is an arbitrary weight the following inequality holds $$w\left(\left\{ x\in\mathbb{R}^{n}\,:\,\left|[b,T]f(x)\right|>\lambda\right\} \right)\leq C_{\varepsilon,T}\int_{\mathbb{R}^{n}} \Phi\left(\|b\|_{BMO}\frac{|f(x)|}{\lambda}\right)M_{L\left(\log L\right)^{1+\varepsilon}}w(x)dx$$ for every $\varepsilon>0$.
Very recently (c.f. [@2015arXiv150708568P]) the authors have obtained a quantitative version of the endpoint estimate for arbitrary weights, namely Theorem \[Thm:EndPCual\]. This result is analogous to the one obtained by the first author and T. Hytönen for singular integrals in [@MR3092729].
Let $T$ be a Calderón-Zygmund operator and $b\in BMO$. If $w\geq0$ is a weight then, for every $\varepsilon>0$ $$w\left(\left\{ x\in\mathbb{R}^{n}\,:\,|[b,T]f(x)|>\lambda\right\} \right)\leq \frac{c}{\varepsilon^{2}}\int_{\mathbb{R}^{n}}\Phi\left(\|b\|_{BMO}\frac{|f|}{\lambda}\right)M_{L\left(\log L\right)^{1+\varepsilon}}wdx.$$
The main novelty here is the appearance of the sharp factor $ \frac{1}{\varepsilon^{2}}$ reflecting again the higher singularity of the operator. As a corollary of this result we can derive the following result obtained previously by C. Ortiz-Caraballo in [@MR3008263], $$w\left(\left\{ x\in\mathbb{R}^{n}\,:\,|[b,T]f(x)|>\lambda\right\} \right)\leq C\Phi\left([w]_{A_{1}}\right)^{2}
\int_{\mathbb{R}^{n}}
\Phi\left (\|b\|_{BMO}\frac{|f|}{\lambda} \right) \,wdx.$$
We remark that it seems that the conjugation method cannot be applied to prove this estimate. Therefore, estimate (\[LogL\]) or Theorem \[Thm:EndPCual\] works, so far, for Calderón-Zygmund operators not for general linear operators assuming a minimal appropriate weighted weak type estimate.
Another interesting difference between Calderón-Zygmund operators and commutators concerns their local behavior. A very nice way of expressing this is by means of the following estimate due to Karagulyan [@MR1913610]: there exists a constant $c>0$ such that for each cube $Q$ and for each function $f$ supported on the cube $Q$ $$\label{Subgausiandecay}
\frac{1}{|Q|} |\{x\in Q: |Tf(x)|> tMf(x)\}|\le c\, e^{-c\, t}\qquad t>0.$$ This result can be seen as an improvement of Buckley’s exponential decay theorem [@MR1124164] which is a very useful result. For instance, it allows to improve in a quantitative way the classical good-$\lambda$ inequality between $T$ and $M$: if $p\in (0,\infty)$ and $w\in A_{\infty}$ $$\|Tf\|_{L^{p}(w)}\leq c_T\,p\,[w]_{{A_{\infty}}} \|M(f)\|_{L^{p}(w)}.$$
Motivated by this result of Karagulyan, Ortiz-Caraballo, Rela and the first author developed a new method for proving (\[Subgausiandecay\]) in [@MR3124931]. This method is flexible enough to deal with other operators including the commutators. In particular, we have the following sub-gaussian estimate.
\[OPRthm\] Let $T$ be a Calderón-Zygmund operator and $b\in BMO$, then there exists a constant $c>0$ such that for each $f$ $$\sup_{Q}\frac{1}{|Q|}|\{x\in Q:|[b,T](f\chi_Q)(x)|>tM^{2}f(x)\}|\leq c\,e^{-\sqrt{c\,t\|b\|_{BMO}}}\qquad t>0.\label{OPRestimate}$$
We will show in Section \[decay\] that this subexponential decay is fully sharp. In Section \[sparse\], we will provide a new proof of (\[Subgausiandecay\]) based on the pointwise domination: if $T$ is a Calderón-Zygmund operator, then it is possible to find a finite set of $\eta$-sparse families $\left\{ \mathcal{S}_{j}\right\} _{j=1}^{3^n}$ (see Section \[sparse\] for the definitions) contained in the same or in different dyadic lattices $\mathcal{D}_{j}$ and depending on $f$ such that $$\label{DominationTheorem}
|Tf(x)|\leq c_{T}\sum_{j=1}^{3^n}A_{\mathcal{S}_{j}}f(x)$$ where $$A_{\mathcal{S}_{j}}f(x)=\sum_{Q\in\mathcal{S}_{j}} \frac{1}{|Q|} \int_{Q} |f| \,\chi_{Q}(x).$$ See Section \[sparse\] for details, in particular Theorem \[DominationThm\].
In view of the interest of an estimate like (\[DominationTheorem\]) it would be relevant to produce a counterpart for commutators. The “natural” sparse operator for these commutators would be $$B_{\mathcal{S}}f(x)=\sum_{Q\in\mathcal{S}}\, \|f\|_{{\rule{0pt}{1.7ex}} L\log L,Q}\, \chi_{Q}(x).$$ The reason that leads to consider this sparse operator in terms of the average\
$\| \cdot \|_{{\rule{0pt}{1.7ex}} L\log L,Q}$ is due to the intimate relationship of commutators and $M^2$ which is an operator pointwise equivalent to $M_{{\rule{0pt}{1.5ex}} L\log L}$. In Section \[sparse\] we prove the impossibility of having a domination theorem for commutators by these “sparse” operators.
First observation: Sharpness of the subexponential local decay {#decay}
==============================================================
We prove in this section that Theorem \[OPRthm\] is sharp, i.e., we can find a Calderón-Zygmund operator $T$, a symbol $b\in BMO$ a function $f$ and a cube $Q$ such that $$\frac{1}{|Q|}|\{x\in Q:|[b,T]f(x)|>tM^{2}f(x)\}|\geq c\,e^{-\sqrt{c\,t\|b\|_{BMO}}}$$ for some constant $c>0$. More precisely we have the following.
\[Thm:SharpLocal\]Let $b(x)=\log|x|$, then we can find a constant $c>0$ such that $$|\{x\in(0,1):|[b ,H](\chi_{(0,1) })(x) |>t\}|\geq e^{-\sqrt{ct}}$$ where $H$ stands for the Hilbert transform.
Let $f(x)=\chi_{(0,1)}(x).$ We are going to show that $$|\{x\in(0,1):|[b,H]f(x)|>tM^{2}f(x)\}|=|\{x\in(0,1):|[b,H]f(x)|>t\}|\geq c\,e^{-\sqrt{\alpha\,t}}\qquad t>0.$$ For $x\in(0,1)$ we have that $$[b,H]f(x)=\int_{0}^{1}\frac{\log(x)-\log(y)}{x-y}\,dy=\int_{0}^{1}\frac{\log(\frac{x}{y})}{x-y}\,dy=\int_{0}^{1/x} \frac{\log(\frac{1}{t})}{1-t}\,dt.$$ Now we observe that $$\int_{0}^{1/x}\frac{\log(\frac{1}{t})}{1-t}\,dt=\int_{0}^{1}\frac{\log(\frac{1}{t})}{1-t}\,dt+\int_{1}^{1/x}\frac{\log(\frac{1}{t})}{1-t}\,dt$$ and since $\frac{\log(\frac{1}{t})}{1-t}$ is positive for $(0,1)\cup(1,\infty)$ we have for $0<x<1$ that $$|[b,H]f(x)|>\int_{1}^{1/x}\frac{\log(\frac{1}{t})}{1-t}\,dt.$$ Finally, a computation shows that $$\int_{1}^{1/x}\frac{\log(\frac{1}{t})}{1-t}\,dt \approx \left(\log\frac{1}{x}\right)^{2} \qquad x\to 0.$$ Consequently, we have that for some $x_{0}<1$ $$|[b,H]f(x)|>c\,\left(\log\frac{1}{x}\right)^{2}\qquad0<x<x_{0}.$$ and then for some $t_0>0$, $$\label{cotafinal}
\left| \{x\in(0,1):|[b,H]f(x)|>t\} \right| \geq \left|\left\{x\in(0,x_{0}):c\,\left(\log\frac{1}{x}\right)^{2}>t\right\}\right|=e^{-\sqrt{t/c}} \qquad t>t_{0}$$ as we wanted to prove.
Second observation: a “natural” but false sparse domination result for commutators {#sparse}
==================================================================================
Before stating the result we are going to prove in this section we need some notation. We borrow it from [@2015arXiv150805639L].
\[Dyadic child\]Let $Q$ be a cube (with sides parallel to the axis). We call dyadic child any of the $2^{n}$ cubes obtained by partitioning $Q$ by $n$ “median hyperplanes” (planes parallel to the faces of $Q$ and dividing each edge into 2 equal parts).
If we iterate the partition process of the preceding definition we obtain a standard dyadic lattice $\mathcal{D}(Q)$ of subcubes of $Q$ which has the usual properties:
1. For each $k=0,1,2,\dots$ cubes in the $k$-th generation have sidelength $2^{-k}$ and tile $Q$ in a regular way.
2. Each $Q'$ in the $k$-th generation has $2^{n}$ children in the in the $(k+1)$-th generation contained in it and one and only one parent in the $(k-1)$-th generation containing it (unless it is $Q$ itself).
3. If $Q',Q''\in\mathcal{D}(Q)$, then $Q'\cap Q''=\emptyset$ or $Q'\subseteq Q''$ or $Q''\subseteq Q'$.
4. If $Q'\in\mathcal{D}(Q)$, then $\mathcal{D}(Q')\subseteq\mathcal{D}(Q)$.
\[Dyadic lattice\]A dyadic lattice $\mathcal{D}$ in $\mathbb{R}^{n}$ is any collection of cubes such that
[00.00.0000]{}
If $Q\in\mathcal{D}$ then each dyadic child of $Q$ is in $\mathcal{D}$ as well.
If $Q',Q''\in\mathcal{D}$ there exists $Q\in\mathcal{D}$ such that $Q',Q''\in\mathcal{D}(Q)$.
If $K$ is a compact set of $\mathbb{R}^{n}$ there exists $Q\in\mathcal{D}$ such that $K\subseteq Q$.
There is an easy way to build a dyadic lattice by considering a increasing sequence of dyadic cubes $Q_{j}$ such that $\cup_{j=1}^{\infty}Q_{j}=\mathbb{R}^{n}$. Then $$\mathcal{D}=\bigcup_{j=1}^{\infty}\mathcal{D}(Q_{j})$$ is a dyadic lattice.
\[DefE(Q\] Let $\eta\in(0,1)$. We say that a family of cubes $\mathcal{S}\subseteq\mathcal{D}$ is $\eta$-sparse if for each $Q\in\mathcal{S}$ we can find a measurable subset $E(Q)\subset Q$ such that:
1. $E(Q)$’s are pairwise disjoint.
2. $\eta|Q|\leq|E(Q)|$
Let $\Lambda>1$. We say a family of cubes $\mathcal{S}$ is $\Lambda$-Carleson if for every cube $Q\in\mathcal{D}$ we have $$\sum_{P\in\mathcal{S},P\subset Q}|P|\leq\Lambda|Q|.$$
There is an interesting relation between Carleson and sparse families that we summarize in the following lemma
\[lem:SparseCarleson\] If $\mathcal{S}$ is a $\Lambda$-Carleson family of cubes then it is $\frac{1}{\Lambda}$-sparse. Conversely if $\mathcal{S}$ is a $\eta$-sparse family of cubes then it is a $\frac{1}{\eta}$-Carleson family of cubes.
Armed with all these definitions we can state the following pointwise domination theorem.
\[DominationThm\] Let $T$ be a Calderón-Zygmund operator. There is a finite set of $\eta$-sparse families $\left\{ \mathcal{S}_{j}\right\} _{j=1}^{3^n}$ contained in the same or in different dyadic lattices $\mathcal{D}_{j}$ and depending on $f$ such that $$T^{*}f(x)\leq c_{T,n}\sum_{j=1}^{3^n}A_{\mathcal{S}_{j}}f(x)$$ where $A_{\mathcal{S}_{j}}f(x)=\sum_{Q\in\mathcal{S}_{j}} \frac{1}{|Q|}\int_Q |f| \, \chi_{Q}(x)$.
The proof of this result can be found in [@2015arXiv150805639L] and [@2014arXiv1409.4351C]. In [@2015arXiv150105818L] M. Lacey obtains the same estimate for Calderón-Zygmund operators that satisfy a Dini condition. Recently a fully quantitative version of Lacey’s result was obtained in [@2015arXiv151005789H] and even more recently this quantitative version has been simplified in [@2015arXiv151207247L].
As a sample of the interest of this result we give a different proof of the exponential estimate (\[Subgausiandecay\]): there exists a constant $c>0$ such that for each cube $Q$ and for each $f$ supported on the cube $Q$ $$\label{Subgausiandecay2}
\frac{1}{|Q|}\left |\left\{x\in Q: T^*f(x)> tMf(x)\right\}\right| \leq c\, e^{-c\, t}\qquad t>0,$$ To prove this result we will use the classical *vector-valued extension of the maximal function* introduced by Fefferman and Stein in [@MR0284802] that can be written as follows: $$\overline{M}_qf(x)=\Big( \sum_{j=1}^{\infty} (Mf_j(x))^q \Big)^{1/q}=|Mf(x)|_q,$$ where ${f}=\{f_j\}_{j=1}^{\infty}$ is a vector–valued function.
Taking into account that $T^*$ is controlled by a finite sum of sparse operators it suffices to establish (\[Subgausiandecay2\]) for those operators.
Assume that $\operatorname{supp}f \subseteq Q$ for a certain cube. It is clear that we can find $c_n$ pairwise disjoint cubes $Q_j\in\mathcal{D}$ which union covers $Q$ and such that $|Q_j| \simeq |Q|$ We can assume those cubes to belong to any sparse family $\mathcal{S}\subset\mathcal{D}$, since it’s easy to check, taking into account Lemma \[lem:SparseCarleson\], that adding a finite number of pairwise disjoints cubes to a sparse family the resulting family is again a sparse family.
First we are going to prove that if $\mathcal{S}$ is and $Q_j\in\mathcal{S}$ with $|Q_j|\simeq |Q|$, as we have just showed that we can assume, then $$\frac{1}{|Q|}\left|\left\{x\in Q: \sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \chi_{P}(x)> t \right\}\right| \leq ce^{-ct}\label{ClaimLocal}$$
We begin observing that $$\begin{aligned}
&&\frac{1}{|Q|}\left|\left\{x\in Q: \sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \chi_{P}(x)> t \right\}\right| \\
&=&\frac{1}{|Q|}\left|\left\{x\in Q\cap Q_j: \sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \chi_{P}(x)> t \right\}\right| \\
& \le & c\frac{1}{|Q_j|}\left|\left\{x\in Q_j: \sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \chi_{P}(x)> t \right\}\right| =C_Q\end{aligned}$$
We use now one of the key estimates from [@MR3124931]. Indeed, let $\{E(P)\}_{{P\in\mathcal{S}, \, P\subseteq{Q_j}}}$ be the family of sets from Definition \[DefE(Q\]. We have then for some $c>0$ that $$\begin{aligned}
\sum_{P\in\mathcal{S}} \chi_{P}(x)&=&\sum_{Q\in\mathcal{S}} \left({\frac{1}{|P|}}\,|P|\right)^q\chi_{P}(x)\\
&\leq& c\,\sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \left({\frac{1}{|P|}}\,|E(P)|\right)^q\chi_{P}(x)\\
&\leq& c\,\sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}} \left({\frac{1}{|P|}}\,\int_{P}\chi_{E(P)}(y)\,dy\right)^q\chi_{P}(x)\\
&\leq& c\, \left(\overline{M}_q\left(\left\{\chi_{E(P)}\right\}_{P\in\mathcal{S}, \, P\subseteq{Q_j}}\right)(x)\right)^q\\
&\leq& c\, \left(\overline{M}_qg_j(x)\right)^q,\end{aligned}$$ where $g_j=\left\{\chi_{E(P)}\right\}_{P\in\mathcal{S}, \, P\subseteq{Q_j}}$ is supported in $Q_j$. Now, since $\{E(Q)\}_{P\in\mathcal{S}, \, P\subseteq{Q_j}}$ is a pairwise disjoint family of subsets, we have that for any $j$ $$\|g_j(x)\|_{\ell^{q}}=\left(\sum_{P\in\mathcal{S}, \, P\subseteq{Q_j}}\left(\chi_{E(Q)}(x)\right)^q\right)^{1/q}\leq 1.$$ We finish the proof of (\[ClaimLocal\]) recalling that if $|g_j|_{\ell^{q}}\in L^{\infty}$, then $\left(\overline{M}_qg_j(x)\right)^q\in Exp L$ (see [@MR0284802]) from which we conclude that: $$C_Q\leq c e^{-c t},\qquad t>0.$$ Now we go back to the proof of the estimate. We first observe that $$\begin{split}&\frac{1}{|Q|}\left |\left\{x\in Q: A_\mathcal{S}f(x)> tMf(x)\right\}\right| \\ \leq & \sum_{j=1}^{c_n}\frac{1}{|Q|}\left |\left\{x\in Q: A_{\mathcal{S}}\left(f\chi_{Q_j}\right)(x)> \frac{t}{c_n}M(f\chi_{Q_j})(x)\right\}\right|. \end{split}$$
Hence it suffices to obtain an estimate for each term of the sum. First we may assume that $|Q_j\cap Q|\neq0$ since otherwise, $\int_E f\chi_{Q_j}=0$ for every measurable set and the corresponding term in the sum equals zero. Now we split $A_\mathcal{S}(f\chi_{Q_j}) $ as follows
$$\label{SplitSparse}
A_\mathcal{S}(f\chi_{Q_j}) (x)=\sum_{P\in\mathcal{S},\, P\subsetneq Q_j} \frac{1}{|P|}\int_{P}f\chi_{Q_j}\chi_P(x) + \sum_{P\in\mathcal{S},\, P\supseteq Q_j} \frac{1}{|P|}\int_{P}f\chi_{Q_j}\chi_P(x) .$$
We observe that for the first term
$$\frac{\sum_{P\in\mathcal{S},\, P\subsetneq Q_j} \frac{1}{|P|}\int_{P}f\chi_{Q_j}\chi_P(x)}{M\left(f\chi_{Q_j}\right)(x)}\leq \sum_{P\in\mathcal{S},\, P\subseteq Q_j} \chi_P(x)$$
For the second term, we have that that since $Q_j\cap Q\neq\emptyset$ and $|Q|\simeq |Q_j|$ then $Q\subset 5Q_j$. Consequently, $$\begin{aligned}
& \frac{\sum_{P\in\mathcal{S},\, P\supseteq Q_j} \frac{1}{|P|}\int_{P}f\chi_{Q_j}\chi_P(x)}{M(f\chi_{Q_j})(x)}
&\leq \frac{\sum_{P\in\mathcal{S},\, P\supseteq Q_j} \frac{1}{|P|}\int_{Q_j}f\chi_P(x)}{\frac{1}{|5Q_j|}\int_{Q_j} f}\\
& &\leq c_n\sum_{P\in\mathcal{S},\, P\supseteq Q_j} \frac{|Q_j|}{|P|}\chi_P(x)\\
& &\leq c_n\sum_{k=0}^\infty \frac{1}{2^{nk}}\end{aligned}$$
Then, combining the estimates obtained for each of the terms of (\[SplitSparse\]), we have that
$$\begin{split}&\frac{1}{|Q|}\left |\left\{x\in Q: A_{\mathcal{S}}\left(f\chi_{Q_j}\right)(x)> \frac{t}{c_n}M(f\chi_{Q_j})(x)\right\}\right|\\
\leq&\frac{1}{|Q|}\left |\left\{x\in Q: \sum_{P\in\mathcal{S},\, P\subseteq Q_j} \chi_P(x) > \frac{t}{c_n}-c_n\sum_{k=0}^\infty \frac{1}{2^{nk}}\right\}\right|\end{split}$$
and the desired conclusion, namely (\[Subgausiandecay2\]), follows from (\[ClaimLocal\]).
Let $T$ be a Calderón-Zygmund operator and $b\in BMO$. It is not possible to find a finite set of $\eta$-sparse families $\left\{ \mathcal{S}_{j}\right\} _{j=1}^{N}$, with $N$ dimensional, contained in the same or in different dyadic lattices $\mathcal{D}_{j}$ and depending on $f$ such that $$\label{eq:FakeSparse}
|[b,T]f(x)|\leq c_{b,T}\sum_{j=1}^{N}B_{\mathcal{S}_{j}}f(x) \qquad a.e. \,\,x \in \mathbb{R}^{n}$$ where $B_{\mathcal{S}_{j}}f(x)=\sum_{Q\in\mathcal{S}_{j}} \|f\|_{{\rule{0pt}{1.7ex}} L\log L,Q}\, \chi_{Q}(x)$.
We are going to give two proofs of this result. The first one is based on the Rubio de Francia algorithm.
\[Proof 1\] Suppose that (\[eq:FakeSparse\]) holds, then we can prove the following $L^{1}$ inequality $$\|[b,T]f\|_{L^{1}(w)}\leq c[w]_{A_{1}}\|M^{2}f\|_{L^{1}(w)}.\label{eq:A1fake}$$ Indeed, $$\begin{split}\|[b,T]f\|_{L^{1}(w)} & \leq c_{b,T}\sum_{j=1}^{N}\|B_{\mathcal{S}_{j}}f\|_{L^{1}(w)}\\
& \leq c_{b,T}\sum_{j=1}^{N}\sum_{Q\in\mathcal{S}_{j}} \|f\|_{{\rule{0pt}{1.7ex}} L\log L,Q} \frac{w(Q)}{|Q|}|Q|\\
& \leq\frac{c_{b,T}}{\eta}\sum_{j=1}^{N}\sum_{Q\in\mathcal{S}_{j}} \|f\|_{{\rule{0pt}{1.7ex}} L\log L,Q} \frac{w(Q)}{|Q|}|E(Q)|\\
& \leq\frac{c_{b,T}}{\eta}\sum_{j=1}^{N}\sum_{Q\in\mathcal{S}_{j}}\int_{E(Q)}M_{{\rule{0pt}{1.7ex}} L\log L}f(x)Mw(x)dx\\
& \leq N\frac{c_{b,T}}{\eta}[w]_{A_{1}}\|M^{2}f\|_{L^{1}(w)},
\end{split}$$ since $M^2 \approx M_{{\rule{0pt}{1.5ex}} L\log L}$. We claim now the $L^p$ version, $$\|[b,T]f\|_{L^{p}(\mathbb{R}^{n})}\leq c_{n}p\|M^{2}f\|_{L^{p}(\mathbb{R}^{n})}\qquad p>1.\label{eq:int}$$ Indeed, by duality we can find $g\geq0$ in $L^{p'}(\mathbb{R}^{n})$ with unit norm such that $$\|[b,T]f\|_{L^{p}(\mathbb{R}^{n})}=\int_{\mathbb{R}^{n}}|[b,T]f(x)|g(x)dx.$$ We consider the Rubio de Francia algorithm $$Rg=\sum_{k=0}^{\infty}\frac{M^{k}(g)}{\|M\|_{L^{p'}(\mathbb{R}^{n})}^{k}}.$$ It’s a straightforward computation that $R(g)$ is an $A_{1}$ weight with constant $$\left[Rg\right]_{A_{1}}\leq2\|M\|_{L^{p'}}\leq c_{n}p$$ and also that $g\leq Rg$ and $\|Rg\|_{L^{p'}}\leq2\|g\|_{L^{p'}(\mathbb{R}^{n})}=2$. Then have that $$\int_{\mathbb{R}^{n}}|[b,T]f(x)|g(x)dx\leq\int_{\mathbb{R}^{n}}|[b,T]f(x)|Rg(x)dx$$ and using and Hölder inequality $$\begin{split} & \int_{\mathbb{R}^{n}}|[b,T]f(x)|Rg(x)dx\leq c\left[Rg\right]_{A_{1}}\int_{\mathbb{R}^{n}}M^{2}f(x)Rg(x)dx\\
& \leq cp\int_{\mathbb{R}^{n}}M^{2}f(x)Rg(x)dx\leq cp\|M^{2}f\|_{L^{p}(\mathbb{R}^{n})}\|Rg\|_{L^{p'}(\mathbb{R}^{n})}\\
& \leq cp\|M^{2}f\|_{L^{p}(\mathbb{R}^{n})}.
\end{split}$$ Hence is established. Now since $$\|M^{2}\|_{L^{p}(\mathbb{R}^{n})}\leq c_n\,\left(p'\right)^{2}\qquad p>1$$ we have that $$\label{eq:BadUpperBound}
\|[b,T]\|_{L^{p}(\mathbb{R}^{n})}\leq cp\left(p'\right)^{2} \qquad p>1$$ Now let us observe that if we take $[b,H]f$ with $b(x)=\log|x|$ and $f(x)=\chi_{(0,1)}(x)$ then $$\|[b,H]f\|_{L^{p}(\mathbb{R})}\geq cp^{2} \qquad p>1,$$ and this leads to a contradiction when $p\rightarrow\infty$. To prove this lower estimate we use estimate (\[cotafinal\]) from Theorem \[Thm:SharpLocal\]. Indeed, for some $t_0>0$ $$\begin{split}\|[b,H]f\|_{L^{p}(\mathbb{R})} & \geq\|\left[b,H\right]f\|_{L^{p,\infty}(\mathbb{R})}=\sup_{t>0}t|\{x\in\mathbb{R}:|[b,H]f(x)|>t\}|^{\frac{1}{p}}\\
& \geq\sup_{t>t_{0}}t\left|\left\{ x\in(0,x_{0}):\,c\left(\log\frac{1}{x}\right)^{2}>t\right\} \right|^{\frac{1}{p}}\\
& \geq\sup_{t>t_{0}}tce^{\frac{-\sqrt{t}}{p}} \geq c\,p^2\,t_0 e^{-\sqrt{t_0} }
\end{split}$$ and this concludes the first proof.
For the second proof we will rely on the sharpness result that was settled in the previous section, namely, we are going to prove that if a pointwise sparse control as the one in (\[eq:FakeSparse\]) holds, then the commutator would have a local exponential decay, which as we established before, is not the case.
\[Proof 2\] Assume again that (\[eq:FakeSparse\]) holds. Then, for some $c>1$ $$\begin{split} & \left|\left\{ x\in Q:|[b,T]f(x)|>tM^{2}f(x)\right\} \right|\leq
\left|\left\{ x\in Q: \sum_{j=1}^{N} \sum_{P\in\mathcal{S}_j }\|f\|_{{\rule{0pt}{1.7ex}} L\log L,P}\chi_{P}(x)>\frac{t}{c}\,M^{2}f(x)\right\} \right|\end{split}$$
It will be enough for our purposes to work on each term of the inner sum, namely to control $$\left|\left\{ x\in Q: \sum_{P\in\mathcal{S}_j }\|f\|_{{\rule{0pt}{1.7ex}} L\log L,P}\chi_{P}(x)>t M^{2}f(x)\right\}\right|$$ Now, recalling that $M^{2}f\simeq M_{{\rule{0pt}{1.7ex}} L\log L}f$, is not hard to see that essentially the same argument we used to prove (\[Subgausiandecay2\]) yields that $$\frac{1}{|Q|}\left|\left\{ x\in Q: \sum_{P\in\mathcal{S}_j }\|f\|_{{\rule{0pt}{1.7ex}} L\log L,P}\chi_{P}(x)>t M^{2}f(x)\right\}\right| \leq ce^{-\alpha t}$$ Combining the preceding estimates we arrive to $$\frac{1}{|Q|}\left|\left\{ x\in Q:|[b,T]f(x)|>tM^{2}f(x)\right\} \right|\leq ce^{-\alpha t} \qquad t>0$$ which is a contradiction by Observation \[Thm:SharpLocal\].
The correct pointwise control for the commutator seems to be the following one
Let $T$ be a Calderón-Zygmund operator and $b\in BMO$. Then $$|[b,T]f(x)|\leq C(n,T)\|b\|_{BMO}\sum_{i,j=1}^{N}A_{\mathcal{S}_{i}}\left(A_{\mathcal{S}_{j}}f\right)(x)$$ where $A_{\mathcal{S}_{j}}f(x)=\sum_{Q\in\mathcal{S}_{j}}\frac{1}{|Q|}\int_{Q}\left|f(y)\right|dy\chi_{Q}(x)$ and the sparse families $\mathcal{S}_{j}$ are not necessarily subfamilies of the same dyadic lattice.
If this conjecture holds it would be very easy to recover the main theorem from [@MR2869172] since it suffices to iterate the following estimate: $$\left\Vert A_{\mathcal{S}_{j}}f\right\Vert _{L^{p}(w)}\leq C_{n,p}[w]_{A_{p}}^{\max\left\{ 1,\frac{1}{p-1}\right\} }\|f\|_{L^{p}}.$$ which was studied in [@MR2628851; @MR2854179] (see also [@2015arXiv150805639L]).
Third observation: The failure of a endpoint estimate motivated by the conjugation method {#negativeEstimateConjugationMethod}
=========================================================================================
In this section we consider the following family of operators: $$f
\to T_w(f) := w\,T\left(\frac{f}{w}\right)$$ where $w$ is a weight and $T$ is a Calderón-Zygmund operator. We already mentioned in the introduction that these operators are of interest since they are very much related to commutators due to the conjugation method. We emphasized that the case $w\in A_1$ is special since $T_w$ is of weak type $(1,1)$ as a consequence of the main results from [@MR2172941]. Understanding the case $w\in A_p$ would be more interesting due to its connection with the conjugation method. However, $T_w$ is [**not**]{} of weak type $(1,1)$ in general since there are weights $w\in A_{p}\setminus A_{1}$ for which $$\left\| T_w \right\|_{L^{1}(\mathbb{R}^{n}) \to L^{1, \infty}(\mathbb{R}^{n})} =\infty,$$ being the purpose of this section to show the existence of such weights. In fact we are going to show something worst replacing $T$ by the less singular operator $M$.
\[Thm19\] Let $1<p<\infty$, then there is $w\in A_{p}\setminus A_{1}$ such that $$\left\Vert M_w \right\Vert_{L^{1}(\mathbb{R}^{n}) \to L^{1, \infty}(\mathbb{R}^{n})} =\infty.$$
In dimension $1$ we choose the $A_p$ weight $w(x)=|x|^{-\delta(1-p)}$ with $\delta\in\left(0,\min\left\{ 1,\frac{1}{p-1}\right\} \right)$ and $f=\chi_{[0,1]}$ so that $f\in L^{1}(w)$. We prove that $$\left\Vert wM\left(\frac{f}{w}\right)\right\Vert _{L^{1,\infty}\left(\mathbb{R}\right)}=\infty.$$ Indeed, a computation shows that for $x>1$ $$M\left(\frac{\chi_{(0,1)}}{w}\right)(x)\geq \frac{1}{x}\frac{1}{\beta}$$ with $\beta=1+\delta(1-p)$ and then $$\begin{split}\left\Vert wM\left(\frac{f}{w}\right)\right\Vert _{L^{1,\infty}\left(\mathbb{R} \right)}
& \geq \frac{1}{\beta}\, \sup_{t>0}t\left|\left\{ x>1\,: \,x^{-\delta(1-p)-1}>t \right\} \right| = \frac{1}{\beta}\, \sup_{1>t>0}t\left(\left(\frac{1}{t}\right)^{\frac{1}{\beta}}-1\right)=\infty
\end{split}$$ since $\beta \in (0,1)$.
An interesting question is to find a necessary and sufficient condition for the boundedness of this operator, namely, characterize the weights $w$ for which $$\left\Vert M_w \right\Vert_{L^{1}(\mathbb{R}^{n}) \to L^{1, \infty}(\mathbb{R}^{n})} < \infty.$$ In [@MR0447956] Muckenhoupt and Wheeden proved that this inequality holds for $w\in A_{1}$ in the real line and also obtained a necessary condition on the weights, namely $$\left\Vert \frac{w\chi_{Q}}{|\cdot-x|^{n}}\right\Vert _{L^{1,\infty}\left(\mathbb{R}^{n}\right)}\leq cw(x)\qquad\text{a.e. }x\in\mathbb{R}^{n}$$ but we don’t know whether is sufficient or not.
To end this section we show that can go further and prove a negative result for possible $L\log L$ type estimates.
Let $1<p<\infty$, and let $\Phi(t)=t\,\log(e+t)^{\alpha} $, $\alpha>0$. Then we can find $w\in A_{p}\setminus A_{1}$ and $f$ such that there’s no $c>0$ for which $$\label{LlogL-false}
\left|\left\{ x\in\mathbb{R}^{n}\,:\,wM\left(\frac{f}{w}\right)>t\right\} \right|\leq c\int_{\mathbb{R}^{n}}\Phi\left(\frac{|f(x)|}{t}\right)dx.$$
As above we do it for the case $n=1$. We assume the contrary, namely there is a finite constant $c>0$ such that (\[LlogL-false\]) holds for any nonnegative $f$. Let $f=\chi_{(0,1)}$. For this choice of $f$ the right hand side of (\[LlogL-false\]) equals $\Phi\left(\frac{1}{t}\right)$ and we have that $$\sup_{t>0} \frac{1}{\Phi\left(\frac{1}{t}\right)}\left|\left\{ x\in\mathbb{R}\,:\,wM\left(\frac{\chi_{(0,1)}}{w}\right)>t\right\} \right| <\infty.$$ Choose again the $A_p$ weight $w(x)=|x|^{-\delta(1-p)}$ with $\delta\in\left(0,\min\left\{ 1,\frac{1}{p-1}\right\} \right)$. Proceeding and using the same notation as in the proof of Observation \[Thm19\] we have that $$\begin{split}\sup_{t>0}\frac{1}{\Phi\left(\frac{1}{t}\right)}\left|\left\{ x\in\mathbb{R}\,:\,wM\left(\frac{\chi_{(0,1)}}{w}\right)>t\right\} \right| & \geq c\,\sup_{0<t<1}\frac{1}{\Phi\left(\frac{1}{t}\right)}\left[\left(\frac{1}{t}\right)^{\frac{1}{\beta}}-1\right]\\
& =c\,\sup_{0<t<1}\frac{t}{\log\left(e+\frac{1}{t}\right)^{\alpha}}\left[\left(\frac{1}{t}\right)^{\frac{1}{\beta}}-1\right] =\infty.
\end{split}$$ since $\beta \in (0,1)$.
[^1]: The first author was supported by Severo Ochoa Excellence Programme and the Spanish Government grant MTM2014-53850-P and the second author was supported by Grant MTM2012-30748, Spanish Government
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Voltage control plays an important role in the operation of electricity distribution networks, especially with high penetration of distributed energy resources. These resources introduces significant and fast varying uncertainties. In this paper, we focus on reactive power compensation to control voltage in the presence of uncertainties. We adopt a probabilistic approach that accounts for arbitrary correlations between renewable resources at each of the buses and we use the linearized DistFlow equations to model the distribution network. We then show that this optimization problem is convex for a wide variety of probabilistic distributions. Compared to conventional per-bus chance constraints, our formulation is more robust to uncertainty and more computationally tractable. We illustrate the results using standard IEEE distribution test feeders.'
author:
- 'Pan Li and Baosen Zhang [^1]'
bibliography:
- 'ref.bib'
title: '**Distribution System Voltage Control under Uncertainties** '
---
Introduction
============
Voltage control is crucial to stable operations of power distribution systems, where it is used to maintain acceptable voltages at all buses under different operating conditions [@LietAl2014]. To control voltage, reactive power is traditionally regulated through tap-changing transformers and switched capacitors [@ZhangetAl2013]. With recent advances in cyber-infrastructure for communication and control, it is also possible to utilize distributed energy resources (DERs, i.e., electric vehicles [@WangEtAl2016], PV panels [@KanchevEtAl2011; @Zhang2015]) to provide voltage regulation. There exists an extensive literature in controlling voltage in a distribution network, some of them focus on centralized control [@FarivarEtAl2012pes; @ValvEtAl2013], while the others address distributed algorithm [@ZhuEtAl2016; @LietAl2014; @ZhangetAl2013; @SulcEtAl2016].
In this paper, we focus on centralized control frameworks to regulate voltage through DERs, that a central controller constantly sends out regulation signals to local DERs in order to control voltage. One important issue to deploy voltage control via DERs is to deal with the uncertainty they bring to the power distribution network [@TuritsynetAl2011]. Since most distribution networks still do not have real-time communication capabilities, decisions made have to be valid for a set of conditions. For example, assume that information is exchanged between a central controller and the local DERs every five minutes. Then the control signal that sets the reactive power on the DERs must be valid for the next five minutes, subject to the fast variabilities and randomness in the renewables.
One way to handle uncertainty is through stochastic programming [@Dantzig2010]. Stochastic formulation takes into account the probabilistic nature of the uncertainty. Chance constraint, which bounds the probability of a certain event, is introduced in these optimization problems. However, it is not obvious to characterize the feasible region of the chance constraint. Monte Carlo simulation is therefore adopted by many researchers, for example in [@ZhangetAl2013], to approximate the optimal solution.
In this paper, we assume that the distribution of the uncertainty is known, and we adopt a stochastic approach to bound the probability that voltage stays within prescribed bounds. Unlike most of the existing literature, we propose to impose a single chance constraint on the whole system. This is different from the standard practice in literature where chance constraints are placed on every bus of the network [@ZhangEtAl2011; @WuEtAl2014]. Putting constraints on each single bus simplifies the problem, but suggests that the uncertainty at each bus is unrelated. However, the randomness across the buses can be well correlated in practice. In this work, we show how a single constraint can be used to capture uncertainties from all buses in the system using the linearized DistFlow model introduced in [@Wu89]. Another approach is to adopt a robust optimization framework, but it is often nontrivial to set the budget of uncertainty and solving the optimization problem maybe computationally challenging [@Ding13; @BertsimasEtAl2013].
We show that our proposed voltage control problem is tractable without sampling techniques. We validate this statement by proving that the proposed chance constraint depicts a convex feasible region and adapts a variety of distributions. The exact optimal solution can therefore be found using standard gradient descent techniques.
In all, we make the following contributions to voltage control in distribution systems:
- We consider voltage control problem with uncertainties in the system. The uncertainty is correlated across the system and is captured by a single constraint imposed onto the whole system.
- The proposed framework is convex, that it only adds one additional convex constraint to the various existing voltage control problems. Therefore there is no need to adopt an approximate algorithm such as Monte Carlo sampling.
The rest of the paper is organized as follows. Section \[sec:distFlow\] presents the modeling of the distribution network. Section \[sec:prob\] proceeds with the formulation of the voltage control problem and demonstrate the robustness of the proposed framework with an illustrating example. Section \[sec:optprob\] states that the problem can be efficiently solved, by proving that the introduced chance constraint is convex. Section \[sec:simu\] validates the statement by simulation results. Finally Section \[sec:conc\] concludes the paper.
Preliminary: Distribution network {#sec:distFlow}
=================================
In this section we present the modeling of components in a radial distribution network in power systems. For interested readers, please refer to [@FarivarEtAl2012; @GanetAl2012] for more details.
Power flow model for radial networks
------------------------------------
We consider a radial distribution network with $N + 1$ buses collected in the set $ \{0, 1, . . . , N\}$. We also denote a line in the network by the pair $(i, k)$ of buses it connects. Bus $0$ is the reference bus.
For each line $(i, k)$ in the network, its impedance is denoted by $z_{ik} = r_{ik} + \text{j}x_{ik}$, where $r_{ik}$ and $x_{ik}$ is its resistance and reactance.
For each bus $i \in \{0, 1, . . . , N\}$, let $V_i$ be the voltage magnitude at bus $i$ and $s_i = p_i + \text{j}q_i$ be the complex power injection, i.e., the generation minus consumption. In addition, the subset $\mathcal{N}_{k}$ denotes bus $k$’s neighboring buses that are further down from the feeder head. The DistFlow equations [@Wu89] model the distribution network flow for every line $(i,k) $ as:
$$\begin{aligned}
P_{ik} - \sum_{l \in \mathcal{N}_k}P_{kl} & = - p_k + r_{ik} \frac{P_{ik}^2 + Q_{ik}^2}{V_i^2},\\
Q_{ik} - \sum_{l \in \mathcal{N}_k}Q_{kl} & = - q_k + x_{ik} \frac{P_{ik}^2 + Q_{ik}^2}{V_i^2},\\
V_i^2 - V_k^2 & = 2(r_{ik}P_{ik} + x_{ik}Q_{ik}) - (r_{ik}^2 + x_{ik}^2) \frac{P_{ik}^2 + Q_{ik}^2}{V_i^2},
\end{aligned}$$
where $P_{ik},Q_{ik}$ are respectively the real and reactive power flow on line $(i,k)$. We let $V_0$ denote the voltage at the reference bus. In addition, the term $\frac{P_{ik}^2 + Q_{ik}^2}{V_i^2}$ represents the loss in the power flow in line $(i,k)$.
Linear approximation of the flow model
--------------------------------------
Assuming the loss is negligible compared to line flow, a linear approximation of can be constructed. Following [@Wu89], we assume that the losses is negligible. We then assume that the voltage at each bus is close to 1. This enables us to approximate $V_i^2 - V_k^2 $ by $2(V_i - V_k)$ [@ZhuEtAl2016]. We then obtain the linearized DistFlow model :
\[lindistflow\] $$\begin{aligned}
P_{ik} - \sum_{l \in \mathcal{N}_k}P_{kl} & = - p_k ,\\
Q_{ik} - \sum_{l \in \mathcal{N}_k}Q_{kl} & = - q_k ,\\
V_i - V_k & = r_{ik}P_{ik} + x_{ik}Q_{ik}.
\end{aligned}$$
From , we can write the voltage magnitude $\bm{V} = [V_1, \dots, V_N ]^{\top}$ in terms of reactive power injection $\bm{Q} = [Q_1, \dots, Q_N]^{\top}$ and real power injection $\bm{P} = [P_1, \dots, P_N]^{\top}$, and the reference voltage $V_0$: $$\label{system}
\bm{V} = \bm{R}\bm{P} + \bm{X}\bm{Q} + V_0$$ where $\bm{R}, \bm{X} \in \mathbb{R}^{N \times N}$ are matrices with $R_{ik}$ and $X_{ik}$ as the element in $i^{th}$ row and $k^{th}$ column, respectively. The voltage profile at bus $1,..., N$ is denoted by $\bm{V} \in \mathbb{R}^N$.
Following the findings in [@LietAl2014], we give the expressions of $R_{ik}$ and $X_{ik}$ in terms of line resistance $r_{ik}$ and reactance $x_{ik}$[^2]:
\[RXelement\] $$\begin{aligned}
R_{ik} & = \sum_{(h,l) \in \mathcal{P}_i \cap \mathcal{P}_k} r_{hl}, \\
X_{ik} & = \sum_{(h,l) \in \mathcal{P}_i \cap \mathcal{P}_k} x_{hl},
\end{aligned}$$
where $\mathcal{P}_i$ is the set of lines on the unique path from bus 0 to bus $i$ [@LietAl2014]. Note that $\bm{R}$ and $\bm{X}$ are positive definite matrices [@LietAl2014].
Voltage regulation with reactive power injection {#sec:prob}
================================================
Suppose that bus $0$ is assumed to be operating at nominal voltage level, i.e., $V_0 = 1$ p.u. (per unit). Rewrite $\bm{V}$ as the difference between the bus voltage and reference voltage $V_0$, then is reduced to the following form: $$\bm{V} = \bm{R}\bm{P} + \bm{X}\bm{Q}.$$
As renewables introduce uncertainty in the bus voltages, the voltage profile is reformulated into the following form: $$\label{noisyV}
\bm{V} = \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon},$$ where $\bm{\epsilon}$ is the uncertainty with zero mean and covariance matrix $\bm{\Sigma}$. The covariance matrix $\bm{\Sigma}$ is not necessarily a diagonal matrix since the uncertainty can be highly correlated across buses. In Section \[sec:optprob\] we illustrate a variety of distributions that $\bm{\epsilon}$ can possibly follow.
In order to maintain the voltage at each bus close to the nominal level, we propose to bound the probability that the voltage profile is within some bounds: $$\label{max00}
\Pr \{ \underline{\bm{V}} \leq \bm{V} \leq \overline{\bm{V}} \} \geq \alpha,$$ which is equivalent to write as: $$\label{max0}
\Pr \{ \underline{\bm{V}} \leq \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon} \leq \overline{\bm{V}} \} \geq \alpha,$$ where $\underline{\bm{V}} $ and $\overline{\bm{V}}$ are the bounds prescribed to the random voltage profile. They indicate how far the voltage profile can be away from a nominal level. The value of $\alpha$ indicates the probability that event $\underline{\bm{V}} \leq \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon} \leq \overline{\bm{V}}$ occurs.
Main Optimizaton Problem
------------------------
In this paper we only consider reactive power regulation and assume that active load injection $\bm{P}$ is determined exogenously and the controllable variable is the reactive power injection $\bm{Q}$. Denote $\Pr \{ \underline{\bm{V}} \leq \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon} \leq \overline{\bm{V}} \}$ by $g(\bm{Q})$, for a given tolerance level $\alpha$, the centralized voltage regulation problem is then captured as the following:
\[prob:main\] $$\begin{aligned}
& \min_{\bm{Q}} \ ||\bm{Q}||_2 \label{eqn:main_obj}\\
s.t.\ & g(\bm{Q}) \geq \alpha, \label{eqn:alpha} \\
& \underline{\bm{Q}} \leq \bm{Q} \leq \overline{\bm{Q}},
\end{aligned}$$
where the box constraint on $\bm{Q}$ represents the limits of reactive power injection at each bus. These bounds can be interpreted as the capacity or availability of devices at each bus. In addition, note that looser bounds on $\underline{\bm{V}} $ and $\overline{\bm{V}}$ in $g(\bm{Q})$ correspond to a larger $\alpha$, whereas a tighter bound corresponds to a smaller $\alpha$. In practice, the value for $\overline{\bm{V}}$ is usually taken as 0.05 p.u. or 0.1 p.u. The objective function in minimize the total action of the DERs, although it can be easily replaced by any time of convex function of $\bm{Q}$.
Per-Bus Constraints
-------------------
Our approach is inherently different from the existing literature when dealing with chance constraints. In most existing literature with randomness in the distribution network, chance constraints are introduced as [@ZhangEtAl2011]: $$\begin{aligned}
&\Pr \{ \underline{V}_i \leq V_i \leq \overline{V}_i \}\\
= & \Pr \{\underline{V}_i \leq \bm{R}_i^{\top}\bm{P} + \bm{X}_i^{\top}\bm{Q} + {\epsilon}_i \leq \overline{V}_i \} \geq \eta_i,
\end{aligned}$$ where $\bm{R}_i^{\top}$ and $\bm{X}_i^{\top}$ extracts the $i$th row in respective matrices. The chance constraint at bus $i$ is associated with prescribed tolerance $\eta_i$. We assume that each bus has the same tolerance, i.e., $\eta_i = \eta$. The optimization problem that incorporates per-bus chance constraint is in the following form:
\[prob:per\_bus\] $$\begin{aligned}
& \min_{\bm{Q}} \ ||\bm{Q}||_2 \\
s.t.\ & g_i(\bm{Q}) \geq \eta, \forall i \label{eqn:perbus}\\
& \underline{\bm{Q}} \leq \bm{Q} \leq \overline{\bm{Q}},
\end{aligned}$$
where $g_i(\bm{Q}) \overset{\Delta}{=} \Pr \{\underline{V}_i \leq \bm{R}_i^{\top}\bm{P} + \bm{X}_i^{\top}\bm{Q} + {\epsilon}_i \leq \overline{V}_i \}$.
The solution to problems described by with chance constraints on each bus $i$ is discussed widely in literature, for example in [@ZhangEtAl2011; @WuEtAl2014]. This per-bus framework is not the same as having a single constraint on the whole system, i.e., is different from . Those two different frameworks do not return the same feasible set and the proposed framework, which places a single constraint on the whole system, captures the coupling between buses and is therefore more realistic and applicable. In addition, as we show in Section \[sec:optprob\], our framework is tractable, in the sense that the chance constraint in is (after some transformations) convex.
Four-bus Toy Example
--------------------
Let us take an illustrating example with a line network with four buses, shown in Fig. \[4bus\].
![An illustration of 4 bus line network.[]{data-label="4bus"}](4bus_1){width="0.3\columnwidth"}
Suppose that the reactive power injections at bus 2 and 3 are limited to 0.2 p.u., we have a linear constraint as $-0.2 \leq Q_2 \leq 0.2$. Then our proposed framework is solving the following optimization problem: $$\label{eq:4bus1}
\begin{aligned}
& \min_{\bm{Q}} \ ||\bm{Q}||_2 \\
s.t. \
& g(\bm{Q}) \geq \alpha, \\
\ -0.2 &\leq Q_j \leq 0.2, \forall j \in \{2,3\}\\
\end{aligned}$$ and its optimal solution is denoted by $\bm{Q}^*_a$. Here we take $\alpha$ to be $0.88$. The per-bus formulation is the following optimization problem: $$\label{eq:4bus2}
\begin{aligned}
& \min_{\bm{Q}} \ ||\bm{Q}||_2 \\
s.t. \
& g_i(\bm{Q}) \geq \eta, \forall i \in \{1,2,3\}\\
\ -0.2 &\leq Q_j \leq 0.2, \forall j \in \{2,3\}\\
\end{aligned}$$ and its optimal solution of by $\bm{Q}^*_p$. Detailed configuration of the line network is left in appendices for interested readers. Unlike the problem in , setting a “right” $\eta$ in is not straightforward. Suppose we want to achieve the same level confidence as where the system operates within the prescribed bounds with probability at least $0.88$, then what is the right $\eta$ to take?
As suggested by previous studies [@ZhangetAl2013], a natural candidate for $\eta$ is to set it equal to $\sqrt[3]{0.88}=0.958$ by thinking of each bus as independent to each other. A second candidate is simply set it at $0.88$, the same as $\alpha$.
The main results of the four-bus line network are shown in Table \[table:4bus\], where $g(\bm{Q})$ denotes the probability $\Pr \{ \underline{\bm{V}} \leq \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon} \leq \overline{\bm{V}} \}$ and is the figure of merit we compare the solutions with. For the proposed framework $\bm{Q} = \bm{Q}^*_a$, and $\bm{Q} = \bm{Q}^*_p$ for the per-bus framework. The bound on the voltage deviation is denoted by $\overline{\bm{V}} = [0.1 \ 0.1 \ 0.1]^{\top}$, which means that voltage deviates no more than 10% of the nominal level. The uncertainty is assume to be multivariate Gaussian and its covariance matrix is shown in the appendices.
------------------------------------------------ --
**Framework & **$g(\bm{Q})$\
$Q^*_a$, $\alpha = 0.88$ & 0.88\
$Q^*_p$, $\eta = \sqrt[3]{0.88}$ & Infeasible\
$Q^*_p$, $\eta = 0.88$ & 0.78\
****
------------------------------------------------ --
: The value of $g(\bm{Q}^*)$ under different frameworks.
\[table:4bus\]
It turns out that assuming each bus as independent drives the optimization problem with the per-bus constraint infeasible. This is due to the fact that in order to achieve $0.88$ as the joint probability, the per bus constraint requires a very small $\eta$ that is too strict. The second value of $\eta=0.88$ makes the problem feasible, but at the cost of lowering the joint probability $g(\bm{Q})$ to be 0.78. This result shows the difficult of using the per bus constraint, where it is difficult to set the correct tolerance level in order to be not overly pessimistic or optimistic. Of course, one could vary $\eta$ and resolve , but the procedure is cumbersome (especially for large networks) and the best $\eta$ does not guarantee that the overall probability of violation is the lowest.
In the following section, we elaborate on the framework based on and show that it can be casted into a convex optimization problem.
Solving the optimization problem {#sec:optprob}
================================
The proposed optimization problem in has a linear objective with a box constraint and a chance constraint. In this section, we present our main statement in Theorem \[theorem0\] that the proposed problem is convex.
\[theorem0\] If the uncertainty $\bm{\epsilon}$ has log-concave probabilistic distribution, then the optimization problem in is convex.
Theorem \[theorem0\] states that even with a chance constraint whose feasible region is not obvious at first sight, it does not complicate the voltage control problem. On the contrary, the chance constraint preserves convexity of the optimization problem and is tractable.
In Theorem \[theorem0\], we require that the uncertainty has a PDF that belongs to a certain class of functions, i.e., log concave functions. We now introduce the definition of log-concavity.
A non-negative function $f:\mathbb{R}^N \rightarrow R^+$ is logarithmically concave (or log-concave for short) if its domain is a convex set, and if it satisfies the inequality: $$\begin{aligned}
f(\theta x+(1-\theta )y) &\geq f(x)^{\theta}f(y)^{1-\theta } f(\theta x+(1-\theta )y) \\
& \geq f(x)^{\theta }f(y)^{1-\theta},
\end{aligned}$$ for all $x, y \in \text{dom}(f)$ and $0 < \theta < 1$. In short, if $\log(f(x))$ is concave, then $f(x)$ is log-concave.
Log concave functions enjoy many properties that lead to convexity. For example, the level set of a log-concave function is convex, which means that if $g(\bm{Q})$ is log concave, then its level set $ \{\bm{Q}:g(\bm{Q}) \geq \alpha\}$ is a convex set [@SaumardEtAl2014]. Therefore, the chance constraint $ g(\bm{Q}) \geq \alpha$ can be easily carried over to other existing deterministic optimization frameworks and preserves the convexity of the problem.
What is more, to maximize over a log concave function, it is sufficient to take the logarithmic of the original function and apply gradient descent techniques. The local optimum is guaranteed to be the global optimum. This suggests that we can easily maximize over $g(\bm{Q})$ without using approximate algorithms such as sampling.
Therefore, to prove Theorem \[theorem0\], $g(\bm{Q})$ is log concave and we introduce two lemmas that discuss log-concavity of certain functions. Lemma \[theorem2\] states that the accumulated mass of a log concave probabilistic function over a convex set is log concave. Lemma \[theorem3\] states that applying linear transformation to the variable in a log-concave function yields another log-concave function, given that the linear transformation has full row rank.
\[theorem2\] Denote $F(\bm{z}) = \int_{\bm{z}-1}^{\bm{z}} f(\epsilon'_1,\dots,\epsilon'_N) \,d\epsilon'_1 \dots d\epsilon'_N$ . If the distribution $f(\epsilon'_1,\dots,\epsilon'_N)$ is log concave in $\bm{\epsilon}'$, then $F(\bm{z})$ is log concave in $\bm{z}$.
Lemma \[theorem2\] only requires the log concave distribution of $\bm{\epsilon}'$ to ensure that the function $F(\bm{z})$ is log concave. A lot of commonly known distributions fall within the log-concave probabilistic distributions, for example, Gaussian distribution and Weibull distribution which is usually adopted to generate intermittent wind energy. Detailed proof is shown in appendices.
Lemma \[theorem2\] states that the objective function $F(\bm{z})$ is log concave in $\bm{z}$. We further show that ${g}(\bm{Q}) = F(\bm{Y}'(\bm{Q}))$ where $\bm{Y}'(\bm{Q}) = A\bm{Q} + \bm{b}$ is log concave in $\bm{Q}$. This is shown in Lemma \[theorem3\] and the proof is in the appendices.
\[theorem3\] Assume that $F(\bm{z})$ is a log concave function, $\bm{z} \in \mathbb{R}^{N}$ and that $\bm{z} = A\bm{y} + \bm{b}$ with $\bm{y} \in \mathbb{R}^{M}$, $A \in \mathbb{R}^{N \times M}$, $\bm{b} \in \mathbb{R}^{N}$. If $A$ has rank $N$, then ${g}(\bm{y}) = F(A\bm{y} + \bm{b})$ is also a log concave function.
With Lemma \[theorem2\] and Lemma \[theorem3\], we now prove the statement in Theorem \[theorem0\].
We reformulate $g(\bm{Q})$ into:
\[formulation2\] $$\begin{aligned}
& g(\bm{Q}) \\
= & \Pr \{ \bm{0} \leq - \underline{\bm{V}} + \bm{R}\bm{P} + \bm{X}\bm{Q} + \bm{\epsilon} \leq - \underline{\bm{V}} + \overline{\bm{V}} \} \\
= & \Pr \{ \bm{0} \leq \bm{Y}(\bm{Q}) + \bm{\epsilon} \leq \bm{u} \}, \label{eqn:YQ}
\end{aligned}$$
where $\bm{Y}(\bm{Q}) = - \underline{\bm{V}} + \bm{R}\bm{P} + \bm{X}\bm{Q}$ and $- \underline{\bm{V}} + \overline{\bm{V}} = \bm{u}$. We can scale row of the inequality inside to make the expression easier to work with. Defining $\bm U$ as $\operatorname{diag}(\bm{u})$ and scale the inequality in by $\bm{U}^{-1}$. Then is equivalent to : $$\label{eq:transform}
\begin{aligned}
&\Pr \{ \bm{0} \leq \bm{Y}'(\bm{Q}) + \bm{\epsilon}' \leq \bm{1} \} \\
= & \Pr \{ - \bm{Y}'(\bm{Q}) \leq \bm{\epsilon}' \leq \bm{1}- \bm{Y}'(\bm{Q}) \} \\
\overset{(a)} = & \Pr \{ \bm{Y}'(\bm{Q}) - \bm{1} \leq \bm{\epsilon}' \leq \bm{Y}'(\bm{Q}) \}
\end{aligned}$$ where $\bm{Y}'(\bm{Q}) = \bm U^{-1} \bm{Y}(\bm{Q})$ and $\bm{\epsilon}' \sim \mathcal{N}(\bm{0}, \bm{\Sigma}')$ and $\bm{\Sigma}' = \bm U^{-1} \bm \Sigma \bm U^{-1}$. The last equality $(a)$ is due to the symmetry of $\bm{\epsilon}'$ with respect to $\bm{0}$.
Denote the probability density function (PDF) of multivariate Gaussian variable as $f(\bm{z}) = f(z_1, ..., z_N)$, $g(\bm{Q})$ is finalized as the integral of the PDF of a multivariate Gaussian variable over an interval $ [\bm{Y}'(\bm{Q}) - \bm{1}, \bm{Y}'(\bm{Q}) ]$: $$\begin{aligned}
g(\bm{Q}) & \overset{\Delta}{=} F(\bm{z}) |_{ \bm{z} = \bm{Y}'(\bm{Q})}\\
& = \int_{\bm{z}-1}^{\bm{z}} f(\epsilon'_1,\dots,\epsilon'_N) \,d\epsilon'_1 \dots d\epsilon'_N |_{ \bm{z} = \bm{Y}'(\bm{Q})}.
\end{aligned}$$
More generally, $F(\bm{z}) $ denotes the probability mass in $[\bm{z-1}, \bm{z}]^N$ from $\bm{\epsilon}'$ where $\bm{\epsilon}'$ has density function $f(\cdot)$. From Lemma \[theorem2\], we know that $F(\bm{z}) $ is log concave, when $\bm{\epsilon}'$ has log-concave distribution.
Since $\bm{Y}'(\bm{Q}) = \bm{U}^{-1}\bm{Y}(\bm{Q}) = \bm{A} \bm{Q} + \bm{b}$, where $\bm{A}$ is full rank and $\bm{b}$ is a constant, Lemma \[theorem3\] implies ${g}(\bm{Q})$ is log concave function.
Simulation {#sec:simu}
==========
In this section, we validate the statement by IEEE standard test feeder. Here we use IEEE 13 bus feeder [@IEEE13busref] as an example. The test feeder is shown in Fig. \[13bus\]. In Fig. \[13bus\], we assume bus 1 is the reference bus.
![Schematic diagram of IEEE 13 bus test feeder.[]{data-label="13bus"}](IEEE13bus.eps){width="0.75\columnwidth"}
Besides, in this distribution test feeder, we assume there are no distributed generation, = so the active power injection on each bus is negative. The line impedance is retrieved from [@IEEE13busref]. In addition, we restrict the available reactive power regulation at the buses to be no more than 0.1 (p.u.).
The randomness presented in this system is assumed to be a multivariate Gaussian distribution with high correlation, i.e., most of the buses have some degree of correlation with the other buses. We solve for $\alpha=0.92$ and for various $\eta$’s. The results are shown in Table \[table:13\_bus\_pr\].
[|c|c|c|c|c|]{} **Framework &**
-----------------
$Q^*_a$
$\alpha = 0.92$
-----------------
: The value of $g(\bm{Q}^*)$ under the proposed framework and per-bus framework.
&
-------------------------
$Q^*_p$
$\eta = \sqrt[N]{0.92}$
-------------------------
: The value of $g(\bm{Q}^*)$ under the proposed framework and per-bus framework.
&
---------------
$Q^*_p$
$\eta = 0.92$
---------------
: The value of $g(\bm{Q}^*)$ under the proposed framework and per-bus framework.
&
---------------
$Q^*_p$
$\eta = 0.98$
---------------
: The value of $g(\bm{Q}^*)$ under the proposed framework and per-bus framework.
\
$g(\bm{Q}^*) $ & 0.92 & Infeasible & 0.66 & 0.89\
\[table:13\_bus\_pr\]
As can be seen from \[table:13\_bus\_pr\], a value of $\sqrt[N]{0.92}=0.992$ is too strict as an requirement to each bus, therefore the optimization with per-bus constraints is infeasible. We therefore adopt two other values, i.e., $\eta = 0.92 = \alpha$ and $\eta = 0.98 >\alpha$, to ensure feasibility. However, neither of these values achieves a satisfactory probably of valid system operation. This highlights are the fact that as the number of buses grows in the network, the per-bus framework becomes difficult to use and extremely sensitive to the exact parameter choices. In contrast, the single constraint formulation in this paper is much more robust and stable.
Conclusion {#sec:conc}
==========
In this paper we adopt a stochastic framework to formulate voltage control problems with uncertainty. Compared to existing literature, we use a single chance constraint to capture the uncertainty in the distribution system. We show that this formulation is more realistic and less conservative than placing constraints on every bus in the distribution network. We also show that the proposed problem is tractable and the feasible region of the constraints is convex, therefore there is no need to adopt approximate algorithms such as Monte Carlo sampling to solve for the problem. Simulation results validate our statement by standard IEEE test feeders.
Appendices {#appendices .unnumbered}
==========
Details of the four-bus line network
------------------------------------
The parameters of the line impedance in the four-bus line network in Fig.\[4bus\] are given as follows: $$R = \begin{bmatrix}
0.1 & 0.1 & 0.1\\
0.1 & 0.3 & 0.3 \\
0.1 & 0.3 & 0.5
\end{bmatrix},
X = \begin{bmatrix}
0.2 & 0.2 & 0.2\\
0.2 & 0.39 & 0.39 \\
0.2 & 0.39 & 0.7
\end{bmatrix}$$
In addition, suppose that there are only active loads at each bus, therefore the active power injection is negative, i.e., $P = [-0.1 \ -0.2 \ -0.3]^{\top}$.
The uncertainty is assume to be multivariate Gaussian, i.e., $\bm{\epsilon} \sim \mathcal{N}(\bm{0}, \bm{\Sigma})$. The covariance matrix $\bm{\Sigma}$ is given as: $$\bm{\Sigma} = 0.002\begin{bmatrix}
1 & 0.7 & 0 \\
0.7 & 1 & 0 \\
0 & 0 & 1
\end{bmatrix}.$$
Proof of Lemma \[theorem2\]
---------------------------
Before proving Lemma \[theorem2\], we introduce Lemma \[convolution\] that is necessary in the proof.
\[convolution\] A convolution over two log-concave probabilistic functions is log concave [@An1996].
Using Lemma \[convolution\], we now proceed to prove Lemma \[theorem2\].
Let $f(\cdot)$ be the PDF of $\bm{\epsilon}'$ that is defined on $\mathbb{R}^N$. Let $u(\cdot)$ be the PDF of a uniform distribution that is defined on $[0,1]^N$. It is easy to check that $u(\cdot)$ is log concave.
Let us use $\star$ to denote convolution operation. The convolution of $f$ and $u$ is: $$\begin{aligned}
(f\star u) (\bm{z}) & = \int_{[0,1]^N}f(\bm{z}-\bm{y})u(\bm{y})d\bm{y} \\
& = \int_{[0,1]^N}f(\bm{z}-\bm{y})d\bm{y} \\
& = - \int_{[\bm{z},\bm{z} - 1]^N}f(\bm{w})d\bm{w} \\
& = \int_{[\bm{z-1},\bm{z}]^N}f(\bm{w})d\bm{w} \\
& = F(\bm{z}) \\
\end{aligned}$$
From Lemma \[convolution\] and that $f$ and $g$ are both log concave, we conclude that $F(\bm{z})$ is log concave in $\bm{z}$.
Proof of Lemma \[theorem3\]
---------------------------
Since log-concavity of a function $F(\bm{z})$ is equivalent to $\log(F(\bm{z}))$ is concave, we have the second order condition on concavity: $$\nabla^2 \log(F(\bm{z})) \preccurlyeq 0,$$ which after some simplification yielding the following form: $$F(\bm{z})\nabla^2 F(\bm{z}) - \nabla F(\bm{z})\nabla F(\bm{z})^{\top} \preccurlyeq 0 .$$
Let us denote $F(\bm{z})\nabla^2 F(\bm{z}) - \nabla F(\bm{z})\nabla F(\bm{z})^{\top}$ by $H_F(\bm{z})$.
To show that ${g}(\bm{y}) = F(A\bm{y}+\bm{b})$ is log-concave, it is equivalent to show that ${g}(\bm{y})\nabla^2 {g}(\bm{y}) - \nabla {g}(\bm{y})\nabla {g}(\bm{y})^{\top} \preccurlyeq 0$.
We first compute $\nabla {g}(\bm{y})$ as the following: $$\nabla {g}(\bm{y}) = A^{\top} \nabla F(\bm{z})|_{\bm{z} = A\bm{y} + b},$$ and $\nabla^2 {g}(\bm{y})$ as the following, with the fact that $\nabla^2 (A\bm{y} + b) = 0$: $$\nabla^2 {g}(\bm{y}) = A^{\top} (\nabla^2 F(\bm{z})|_{\bm{z} = A\bm{y} + b}) A.$$
So ${g}(\bm{y})\nabla^2 {g}(\bm{y}) - \nabla {g}(\bm{y})\nabla {g}(\bm{y})^{\top}$ is equivalent to $A^{\top} H_F( A\bm{y} + b) A$. Let us denote this new quantity as $H_{{g}}(\bm{y}) = A^{\top} H_F( A\bm{y} + b) A$.
Since $H_F(\bm{z}) \preccurlyeq 0, \forall \bm{z} \in \mathbb{R}^{N}$, it suggests that $\forall \bm{w} \in \mathbb{R}^{N}$ and $A\bm{y} + b \in \mathbb{R}^{N}$, we know: $$\bm{w}^{\top} H_F(A\bm{y} + b) \bm{w} \leq 0.$$
Since $A$ has rank $N$, we know that $\bm{w} = A\bm{z'}$ spans the whole subspace $\mathbb{R}^{N}$, $\forall \bm{z'} \in \mathbb{R}^{M}$. Therefore: $$\bm{z'}^{\top}A^{\top} H_F(A\bm{y} + b) A^{\top}\bm{z'} \leq 0,$$ which is equivalent to : $$\bm{z'}^{\top}H_{{g}}(\bm{y})\bm{z'} \leq 0, \forall \bm{z'} \in \mathbb{R}^{M}.$$
This suggests that $H_{{g}}$ is negative semi definite, and therefore concludes the proof.
Log-concavity of multivariate Gaussian variable
-----------------------------------------------
We use a two dimensional multivariate Gaussian to validate Lemma \[theorem2\], that $F(\bm{z})$ is a log concave function.
Suppose that $\bm{\epsilon'} \sim \mathcal{N} (0, \Sigma)$ and $\Sigma = \begin{bmatrix}
0.9 & 0.5 \\ 0.5 & 0.6
\end{bmatrix} $. The visualization of $\log(F(\bm{z}))$ is shown in Fig. \[loggx\].
![plot of $\log F(\bm{z})$ in 2-D, , where $\bm{\epsilon'} \sim \mathcal{N} (0, \Sigma)$.[]{data-label="loggx"}](loggx.eps){width="0.8\columnwidth"}
From Fig. \[loggx\], we observe that $F(\bm{z})$, after taking its log, is concave in $\bm{z}$. Therefore we can apply various gradient descent algorithms to search for the global optimum.
[^1]: Pan Li and Baosen Zhang are with the Electrical Engineering Department, University of Washington, Seattle, WA 98195 [@uw.edu]{}
[^2]: Here we do not have a factor 2 as in [@LietAl2014] because we approximate $V_i^2 - V_k^2$ by $2V_i - 2V_k$.
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'paper.bib'
title: |
Efficient and Effective Volume Visualization with\
Enhanced Isosurface Rendering
---
;
Acknowledgements {#acknowledgements .unnumbered}
================
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The LTB model is studied as the Caushy problem for the equations defined two metrical functions $\lambda(\mu,\tau)$ and $\omega(\mu,\tau)$. The initial conditions throught metrical functions are presented. The rules of calculating three undetermined functions $f(\mu)$, $F(\mu)$ and ${\bf F}(\mu)$ are obtained. The general expressions for the density and Habble function in the LTB model are written by the metrical functions.\
\
PACS number(s):98.80
author:
- |
*by\
\
[**Alexander Gromov**]{}\
\
*St. Petersburg State Technical University\
*Faculty of Technical Cybernetics, Dept. of Computer Science\
*29, Polytechnicheskaya str. St.-Petersburg, 195251, Russia\
and\
*Istituto per la Ricerca di Base\
*Castello Principe Pignatelli del Comune di Monteroduni\
*I-86075 Monteroduni(IS), Molise, Italia\
*e-mail: [email protected]********
title: ' Report IRB-StP-GR-120696 '
---
The Introduction {#introd}
================
The LTB model is one of the most known spherical symmetry model in general relativity. It was created by Lemaitre [@Lemaitre:33], Tolman [@Tolman:34], and Bondy [@Bondy:47] during the period of time from 1933 to 1947. The exact solution have been obtained by Bonnor [@Bonnor:72] in 1972 and [@Bonnor:74] in 1974. The LTB model represented one of the simplest nonhomogeneous nonstationary cosmological models and due to this fact is used to study some new ideas in the cosmology.
The present interest to LTB was risen by the observations shown the fractal structure of the Universe in the large scale [@Coleman], [@Pietronero:87], and discass in the set of article (the modern review is presented in [@Bar]). In a set of papers the LTB model is used to study the observational datas and redshift as a main cosmological test [@Ribeiro:92a] - [@Moffat:94b].
The central problem of using the LTB model is in calculation of three undetermined functions which defined the solution. There is a set of ways how to solve this problem from the physical point of view in the mentioned papers. This article is devoted to the mathematical point of view on this matter [@A.G.].
The LTB Model {#T_sol}
=============
This section is devoted to presentation the Lemaitre-Tolman-Bondy model and section [2]{} of the paper [@Tolman:34] is cited. The co-moving system of coordinate is used in this model where the interval has the form $$ds^2(r,t) = -e^{\lambda(r,t)}dr^2 - e^{\omega(r,t)}
\left(
d\theta^2 + sin^2\theta d\phi
\right) + dt^2.
\label{ds2}$$ $\lambda(r,t)$ and $\omega(r,t)$ are metrical functions [@Tolman:34] definding the solution. Following the [@Tolman:34] we will omit the arguments in the functions $\omega$ and $\lambda$ in this section. In the co-moving system of coordinate with the line element (\[ds2\]) the energy-momentum tensor $$\begin{aligned}
T^{\alpha,\beta} = \rho \frac{dx^{\alpha}}{ds}\frac{dx^{\beta}}{ds}
\nonumber
%\label{T_ik}\end{aligned}$$ has only one non zero component $$T^{4}_{4} = \rho \qquad
T^{\alpha}_{\beta} = 0, \qquad \alpha \quad \mbox{or} \quad \beta =
4.
\label{T_ik:RO}$$ Using them together with Dingle results [@Dingle] we obtain the system of equations of the LTB model: $$8 \pi T^{1}_{1} = e^{-\omega} -
e^{-\lambda}\frac{{\omega^{\prime}}^2}{4} + \ddot \omega +
\frac{3}{4} \dot \omega^2 - \Lambda = 0
\label{T:4}$$ $$\begin{aligned}
8 \pi T^{2}_{2} = 8 \pi T^{3}_{3} = \nonumber\\
- e^{- \lambda}
\left(
\frac{\omega^{\prime\prime}}{2} +
\frac{\omega^{\prime}{}^{2}}{4} -
\frac{\lambda^{\prime}\omega^{\prime}}{}
\right)
+\frac{\ddot \lambda}{4}
+\frac{\dot\lambda^2}{4}
+ \frac{\ddot \omega}{2}
+\frac{\dot \omega^2}{4}
+\frac{\dot \lambda \dot \omega}{4}
- \Lambda = 0
\label{T:5}\end{aligned}$$ $$8 \pi T^{4}_{4} =e^{-\omega} - e^{-\lambda}
\left(
\ddot \omega +\frac{3}{4}\dot \omega^2 - \frac{\lambda^{\prime}
\omega^{\prime}}{2}
\right)
+ \frac{\dot \omega^2}{2}
+ \frac{\dot \lambda \dot \omega}{2}
- \Lambda = 8 \pi \rho
\label{T:6}$$ $$8 \pi e^{\lambda} T^{1}_{4} = - 8 \pi T^{4}_{1} =
\frac{\omega^{\prime}\dot \omega}{2}
- \frac{\dot \lambda \omega^{\prime}}{2}
+\dot \omega^{\prime} = 0,
\label{T:7}$$ where $$\begin{aligned}
{}^{\prime} = \frac{\partial}{\partial r} \qquad
\dot{} = \frac{\partial}{\partial t}
\nonumber
%\label{T:dot_prime}\end{aligned}$$ The equation (\[T:7\]) has the solution $$e^{\lambda} = e^{\omega} \frac{\omega^{\prime \,2}}{4 f^2(r)}
,
\label{T:8}$$ where $f(r)$ is undetermined function. Substituting (\[T:8\]) into (\[T:4\]) we obtain $$e^{\omega}
\left(
\ddot \omega +\frac{3}{4} \dot \omega^2 - \Lambda
\right)
+
\left[
1 - f^2(r)
\right] = 0.
\label{T:10}$$ This equation is integrated twice. First integral gives the equation $$e^{3 \,\omega / 2}
\left(
\frac{\dot \omega^2}{2} - \frac{2}{3} \Lambda
\right)
+2 e^{\omega /2} \left[1 - f^2(r)\right] = F(r),
\label{T:11}$$ and the second one gives the equation $$\int
\frac
{{\rm d e^{\omega / 2}}}
{
\sqrt{{\rm
f^2(r) - 1 +\frac{1}{2} } F {\rm (r) e^{-\omega/2} + \frac{\Lambda}{3}
e^{\omega}}
}
}
= t + {\bf F} {\rm (r)
}
\label{T:12}$$ The equations (\[T:11\]) and (\[T:12\]) hold undetermined functions $\it F(r)$ и $\bf F\it(r)$. The substitution of (\[T:8\]) into (\[T:6\]) together with (\[T:11\]) gives the equation for density $$8 \pi \rho = \frac{1}{\omega^{\prime}e^{3\omega/2}}
\frac{
\partial
F{\rm (r)}}{{\rm \partial r}}
\label{T:15}$$
The Cauchy Problem for the LTB Model
====================================
Before we study the LTB model let us introduce the follow characteristic values: a velocity of light $c$, an observational meaning of the Habble constant $\tilde H$, a characteristic time $1/\tilde H$ and characteristic length $c/ \tilde H$. We use the co-moving system of coordinates in the LTB model, so the radial coordinate $r$ has the sense of Lagrangian mass coordinate , . Two dimensionless variables $\mu$ and $\tau$ are definded by the rules $$\begin{aligned}
\mu = \frac{r}{m} \quad \quad \quad
\tau = \tilde H t,
\nonumber
%\label{new_var}\end{aligned}$$ where $m$ is a full mass of the “gas”. The dimensionless Habble function $h(\mu,\tau)$ and density $\delta(\mu,\tau)$ will be also used: $$\begin{aligned}
h(\mu,\tau) = \frac{H(\mu,\tau)}{\tilde H}, \qquad
\delta(\mu,\tau) = \frac{\rho(r,t)}{\rho(0,0)},
\nonumber
%\label{h}\end{aligned}$$ where $H(0,0) = \tilde{H}$.
Let us write the interval (\[ds2\]) as $$ds^2(r,t) = -A e^{\lambda(r,t)}dr^2 - B e^{\omega(r,t)}
\left(
d\theta^2 + sin^2\theta d\phi
\right) +c^2dt^2,
\label{ds2_1}$$ where two constants $A$ and $B$ are introduced to take into account the fact that (\[ds2\_1\]) is dimension equation.
The dimension of $\left[ ds^2 \right]$ is $L^2$, dimension of $\left[ A \right]$ is $L^2 M^{-2}$ and dimension of $\left[ B \right]$ is $L^2$, so $$\begin{aligned}
A = \left(\frac{c}{\tilde H m}\right)^2, \qquad B =
\left(\frac{c}{\tilde H}\right)^2.
\nonumber
%\label{ds2_coeff}\end{aligned}$$ The interval (\[ds2\_1\]) has now the form $$\left(\frac{\tilde H}{c}\right)^2ds^2(r,t) =
-e^{\lambda(r,t)}dr^2 - e^{\omega(r,t)}
\left(
d\theta^2 + sin^2\theta d\phi
\right) +d\tau^2,
\label{ds2_D-less}$$ Together with metrical functions $\omega(\mu,\tau)$ and $\lambda(\mu,\tau)$, introduced by Tolman, it is conveniently to use the Bonnor’s function [@Bonnor:72] $$R(\mu,\tau) = e^{\omega(\mu,\tau)/2}.
\label{Bonnor}$$ In the Bonnor’s notation the interval (\[ds2\_D-less\]) takes the form $$\begin{aligned}
\left(\frac{\tilde H}{c}\right)^2 ds^2(\mu,\tau) =
-\frac{\left[R^{\prime}(\mu,\tau)\right]^2}{f^2} d\mu^2 -
R^2(\mu,\tau)
\left(
d\theta^2 + sin^2\theta d\phi
\right) + d\tau^2
\nonumber
%\label{ds2_new}\end{aligned}$$ As it is shown in and , the Bonnor’s coordinate $R(\mu,\tau)$ has a sense of Euler coordinate, so the equation (\[Bonnor\]) correlates geometrical radius of the sphere $R(\mu,\tau)$ where the particle is located, and the Lagrangian coordinate $\mu$ of this sphere.
To describe the radial motion we will use the Habble function connected with variation of the radial length $d l$: $$h = \frac{d \dot l}{d l},
\label{habble_def}$$ where, according to the (\[ds2\_D-less\]), for $dl^2$ we read: $$dl^2 = e^{\lambda(\mu,\tau)} d \mu^2 =
\left(
\frac{R^{\prime}(\mu,\tau)}{f(\mu)} d\mu
\right)^2.
\label{d l}$$ By the substitution (\[d l\]) into the definition of the Habble function (\[habble\_def\]) we obtain $$h(\mu,\tau) = \frac{\dot \lambda(\mu,\tau)}{2} = \frac{\dot R
^{\prime}
(\mu,\tau)
}{R^{\prime}(\mu,\tau)}
= \frac{\partial \ln{R^{\prime}(\mu,\tau)}}{\partial \tau}
\label{Habble_def}$$ By the integration of the equation (\[Habble\_def\]) we obtain the formulum for metrical function $\lambda(\mu,\tau)$: $$\begin{aligned}
\lambda(\mu,\tau) = 2 \int\limits_{0}^{\tau}h(\mu,\tau)d\tau +
\lambda(\mu,0)
\nonumber
%\label{lambda_def_Habbl}\end{aligned}$$ A solution in the LTB model is defined by the functions $f(r)$, $\it F(r)$ and $\bf F\it(r)$. These functions are obtained in the process of solution of the system of PDE, so to define them the initial/boundary conditions should definitely be used. The metrical function $\omega(\mu,\tau)$ is the solution of the equation (\[T:12\]). The equations of the LTB model are obtained in [@Tolman:34] and solved in the parametric form for the three cases $f^2(\mu) < 1$, $f^2(\mu) = 1$ and $f^2(\mu) >
1$ in [@Bonnor:72] and [@Bonnor:74].
The equations (\[T:10\]) - (\[T:12\]) are valid for every $\tau$, and due to this fact in the Cauchy problem they [*definde*]{} the functions $f^2(\mu)$, $\it F(\mu)$ and $\bf F\it(\mu)$ at the moment of time $\tau = 0$: $${\rm
f^2(\mu) - 1 =
e^{\omega_0(\mu)}
\left(
\ddot \omega_0(\mu) + \frac{3}{4} \dot \omega^2_0(\mu) - \Lambda
\right)
}
\label{DEF:f^2-1}$$ $$F{\rm(\mu)} =
{\rm
e^{3 \,\omega_0(\mu) / 2}
\left(
\frac{\dot \omega^2_0(\mu)}{2} - \frac{2}{3} \Lambda
\right)
+2
e^{\omega_0(\mu) /2}
\left[1 - f^2(\mu)\right]}
\label{DEF:itF}$$ $${\bf F} {\rm (\mu)} =
{\rm
\int \limits_{
e^{\omega_0(0)/2}
}^{
e^{\omega_0(\mu)/2}
} }
\frac{
d x
%e^{\tilde\omega_0(\mu) / 2}
}
{
\sqrt{
f^2(\mu) - 1 +\frac{F(\mu)}{2 x}
%e^{-\tilde\omega_0(\mu)/2}
%
+
\frac{\Lambda}{3}
%e^{\tilde\omega_0(\mu)}
x^2
}
}
\label{DEF:F}$$ At the time $\tau = 0$ the equation (\[T:8\]) [*defindes*]{} the function $\lambda_0(\mu)$: $$e^{\lambda_0(\mu)} =
e^{\omega_0(\mu)}
\frac{[\omega_0(\mu)]^{\prime \,2}}{4f^2(\mu)}.
\label{DEF:lambda_0_new)}$$ Substituting (\[DEF:f\^2-1\]) into (\[DEF:itF\]), we obtain $$F
{\rm
(\mu) = e^{3\omega_0(\mu)/2}\left(
-2 \ddot\omega_0(\mu) - \dot\omega^2_0(\mu) +
\frac{4}{3}\Lambda
\right)
}.
\label{DEF:itF_new}$$ Comparing (\[T:8\]) - (\[T:11\]) with (\[DEF:f\^2-1\]) - (\[DEF:itF\]), we find out that $$\begin{aligned}
f^2(\mu) - 1 =
e^{\omega_0(\mu)}
\left(
\ddot \omega_0(\mu) + \frac{3}{4} \dot \omega^2_0(\mu) - \Lambda
\right) &=& \nonumber\\
e^{\omega(\mu,\tau)}
\left(
\ddot \omega(\mu,\tau) + \frac{3}{4} \dot \omega^2(\mu,\tau) - \Lambda
\right)
\label{DEF:f^2-1_n}\end{aligned}$$ and $$\begin{aligned}
F(\mu) =
e^{3\omega_0(\mu)/2}\left(
-2 \ddot\omega_0(\mu) - \dot\omega^2_0(\mu) +
\frac{4}{3}\Lambda
\right) &=& \nonumber\\
e^{3\omega(\mu,\tau)/2}\left(
-2 \ddot\omega(\mu,\tau) - \dot\omega^2(\mu,\tau) +
\frac{4}{3}\Lambda
\right)
\label{DEF:itF_new_n}\end{aligned}$$ are not dependent on time. Let’s use the previous results to calculate the functions ${\bf F}(\mu)$ and integral in the equation (\[T:12\]). Substituting the definitions (\[DEF:f\^2-1\]) and (\[DEF:itF\_new\]) into (\[DEF:F\]) we obtain: $${\bf F}(\mu) =
\pm \int \limits_{\omega_0(0)}^{\omega_0(\mu)}
\frac{d \tilde\omega}{\dot{\tilde\omega}}.
\nonumber
%\label{DEF:F_new}$$ The function ${\bf F}(\mu)$ is equal to zero at the moment of time $\tau = 0$ according the definition. Substituting the right part of the equations (\[DEF:f\^2-1\_n\]) and (\[DEF:itF\_new\_n\]) into the (\[DEF:F\]), we obtain the equation $$%{\bf F}(\mu) =
\pm\int \limits^{\omega(\mu,\tau)}_{\omega(\mu,0)}
\frac{d \tilde\omega}{\dot{\tilde\omega}} =
\pm \int \limits_{\omega_0(0)}^{\omega_0(\mu)}
\frac{d \tilde\omega}{\dot{\tilde\omega}} + \tau.
\nonumber
%\label{DEF:F_new_nn}$$
This analysis of the LTB model shows that the functions $$\left.
\begin{array}{c}
%
\left.\omega(\mu,\tau)\right|_{\tau=0} = \omega_0(\mu) \quad
%
\left.\dot\omega(\mu,\tau)\right|_{\tau=0} = \dot\omega_0(\mu) \quad \\ \\
%
\left.\ddot\omega(\mu,\tau)\right|_{\tau=0} = \ddot\omega_0(\mu),
%
\end{array}
\right\}
\label{init}$$ and constants $$\left.
\begin{array}{c}
%
\left.\omega(\mu,0)\right|_{\mu=0} = \omega_0(0) \quad
%
\left.\dot\omega(\mu,0)\right|_{\mu=0} = \dot\omega_0(0) \quad \\ \\
%
\left.\ddot\omega(\mu,0)\right|_{\mu=0} = \ddot\omega_0(0), \quad
%
\Lambda,
\end{array}
\right\}
\label{init-c}$$ are included into the definitions (\[DEF:f\^2-1\]) - (\[DEF:F\]) and they form the initial conditions of the Cauchy problem for the equations (\[T:4\]) - (\[T:7\]). In accordance with (\[DEF:lambda\_0\_new)\]) the function $\lambda_0(\mu)$ is not include in the set of initial conditions.
Substituting (\[DEF:itF\_new\]) into the (\[T:15\]), we obtain the general expression for the density of “gas” in the LTB model: $$\begin{aligned}
{\rm
8 \pi \delta(\mu,\tau) =
\frac{e^{
\frac{3}{2}
[\omega_0(\mu) - \omega(\mu,\tau)]
}
}
{
\omega^{\prime}(\mu,\tau)
}
} \times
\nonumber\\
{\rm
\left\{
3\left[\omega_0(\mu)\right]^{\prime}
\left[
-\ddot\omega_0(\mu) - \frac{1}{2}\dot\omega_0^2(\mu) +
\frac{\Lambda}{6}
\right]
- 2\left[\ddot\omega_0(\mu)\right]^{\prime} -
2\dot\omega_0(\mu)\left[\dot\omega_0(\mu)\right]^{\prime}
\right\}
}
\label{DEF:rho_new}\end{aligned}$$ We obtain the formulum for Habble’s function using (\[T:12\]), (\[Bonnor\]) and (\[Habble\_def\]): $$h(\mu,\tau) =
\frac{
\frac{\partial K(\mu,\tau)}{\partial \mu}
}
{
\frac{\partial}{\partial \mu}\int\limits_{0}^{\tau} K(\mu,\tau)d \tau
+R^{\prime}(\mu,0)
},
\label{Habbl-ini}$$ where $$\begin{aligned}
K(\mu,\tau) = \sqrt{f^2(\mu) - 1 + \frac{F(\mu)}{2 R(\mu,\tau)} +
\frac{\Lambda}{3}R^3(\mu,\tau)}.
\nonumber
%\label{Habbl-K}\end{aligned}$$ The dependence of cosmologycal parameter on the initial conditions of the LTB model is represented by the follow formula: $$\begin{aligned}
\Omega(\mu,\tau) = \frac{\delta(\mu,\tau)}{\delta_c(\mu,\tau)},
\nonumber
%\label{c-p}\end{aligned}$$ where the critical density defined by formulum $$\begin{aligned}
\delta_{c}(\mu,\tau) = \frac{3 \tilde{H}^2}{8 \pi G \rho_0} h^2(\mu,\tau).
\nonumber
%\label{ro-cr-dl_u}\end{aligned}$$ The function $\omega(\mu,\tau)$ from the equation (\[DEF:rho\_new\]) is the solution of the equation (\[T:12\]).
Results
=======
The LTB model is the Cauchy problem for the PDE (\[T:4\]) - (\[T:7\]). Traditionally, three functions $f(\mu)$, $F(\mu)$ and ${\bf F}(\mu)$ are used to chose some physical propetrys of the problem solving in the LTB model. These functions play a role of the initial conditions in the LTB model. But the input equations (\[T:4\]) - (\[T:7\]) are wrtitten throught the metrical functions $\lambda(\mu,\tau)$ and $\omega(\mu,\tau)$. So, should definitly be studied the dependence functions $f(\mu)$, $F(\mu)$ and ${\bf F}(\mu)$ on metrical functions. The present article learns this rpoblem and three initial conditions (\[init\]) - (\[init-c\]) defined three function by the rules (\[DEF:f\^2-1\]) - (\[DEF:F\]). The general expressions for the density (\[DEF:rho\_new\]) and Habble fuction (\[Habbl-ini\]) show the dependence on the initial conditions (\[init\]).
Acknowledgements
================
I’m grateful to Prof. Arthur D. Chernin for encouragement and discussion. Dr. Yurij Barishev has initiated my interest to the modern Cosmology as fractal structure of the Universe and interpretation of the observations by calculation redshift in the LTB model. This paper was financially supported by “COSMION” Ltd., Moscow.
[20]{} Lema$\hat{i}$tre, Ann. de la Soc.Scient. de Bruxelles [**A53**]{}, 51 (1933). R.C.Tolman, Proc.Nat.Acad.Sci (Wash), [**20**]{},(1934). H.Bondy, MNRAS [**107**]{}, p.p. 410 - 425 (1947). W.B.Bonnor, MNRAS [**159**]{}, 261 - 268 (1972) W.B.Bonnor, MNRAS [**167**]{}, 55 (1974). P.H.Coleman and L.Pietronero, Physics Reports, [**213**]{}, 6, (1992). L.Pietronero, Physica, [**144A**]{}, 257 (1987). Yu.V.Baryshev, F.Sylos Labini, M.Montuori, and L.Pietronero, Vistas in Astronomy, [**38**]{}, 4 (1994). M.B.Ribeiro, ApJ.[**388**]{},1 (1992). M.B.Ribeiro, ApJ.[**395**]{},29 (1992). M.B.Ribeiro, ApJ.[**415**]{} (1993). J.W.Moffat and D.S.Tatarski, Phys.Rev.D [**45**]{},10 (1992) J.W.Moffat and D.S.Tatarski, Proceeding of the XXIXth RENCONTRE DE MORIOND, Series:Moriond Workshops, Villars sur Ollon, Switzerland,January 22-29, 1994. Editions Frontieres. J.W.Moffat and D.S.Tatarski, preprint UTPT-94-19 (1994). A.Gromov, astro-ph/9605201 in [email protected] Dingle, Proc.Nat.Acad. [**19**]{}, 559, (1933). L.D.Landau, E.M.Lifshits, “The Field Theory”, Moscow, “Nauka”, (1973). L.E.Gurevich and A.D.Chernin, “The Introduction into Cosmology”, Moscow,“Mir”, (1978). H.B. Dwight, “Tables of Integrals”, NY, (1961). G.A.Corn, T.M.Corn, “Mathematical Handbook”, McGraw-Hill Company, NY, (1968).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'One of the outstanding problems in modern physics is the origin is of accelerated expansion of the universe. High-precision determinations of the Hubble parameter $\mbox{H}_0$ at different redshifts provide direct insight into the Universe expansion: equation of state of the Dark Energy, geometry and curvature of space, sum of neutrino masses and isotropy of the Universe. We investigate the effect of chemical composition on the classical distance ladder based on Cepheids stars and Supernovae type Ia. Cepheids belonging to Magellanic Clouds represent a natural anchor for the ladder, due to their proximity and the very well known geometric distance. Accordingly, the LACES collaboration collected the largest spectroscopic sample of MCs Cepheids, containing more than 300 stars and more than 1200 medium-high resolution spectra. Abundance measurements, performed by equivalent width and curve of growth analysis, have been tested on a wide grid of synthetic spectra to quantify the systematic arising from this procedure. We focus on the errors arising from temperature measurements, using the Line Depth Ratio method. We find that in order to not bias the final results many aspects (e.g. degeneracy of the atmospheric parameters, instrumental set up etc.) should be explicitly taken into account. In particular, there is a not negligible metallicity dependence.'
author:
- 'Sara Mancino,$^{1,2,3}$ Martino Romaniello,$^1$ Richard I. Anderson,$^1$ and Rolf-Peter Kudritzki$^{2,4}$'
bibliography:
- 'BibliographySHORT.bib'
title: |
Towards precise and accurate Cepheid chemical abundances\
for 1$\%$ $\mbox{H}_0$ measurement: temperature determination
---
Rationale: $\mbox{H}_0$ late- and early-time measurements
=========================================================
$\mbox{H}_0$ marks the present universe expansion rate, and its precise value helps to constrain the other cosmological parameters, ultimately the dark energy equation of state. Classical Cepheids are fundamental test bench for stellar physics, optimal tracers of galactic structure, and more importantly they are primary distance indicators and historically the first rung of the local distance ladder that leads to the Hubble Constant. The latest measurement using Cepheieds is from the S$\mbox{H}_0$ES team and yields the most precise among the late-time[^1] values: $\mbox{H}_{0}=74.0\pm 1.4 \, \mbox{km s$^{-1}$ Mpc$^{-1}$}$ [@riess19]. This latter $~4\sigma$ is in tension with the Planck result, $\mbox{H}_{0}=67.4\pm 0.5 \, \mbox{km s$^{-1}$ Mpc$^{-1}$}$ [@planck18], that is inferred from Cosmic Microwave Background, assuming $\Lambda\mbox{CDM}$. The tension increases over $5\sigma$ combining the independent late probes. Thus, while it becomes compelling to search for new physics beyond the standard cosmological model, reduce and control systematic errors on the late universe measurements is mandatory.
In this frame the S$\mbox{H}_0$ES value is the result of a huge effort to simplify the ladder and collect a larger and photometrically homogeneous sample, thus reducing dramatically the systematic errors along this ladder. Nevertheless the classical ladder it is still affected by the uncertainties coming from the zero-point of the period-luminosity relation. One of these comes from the metallicity dependence of the relation that is still highly debated.
From the theoretical point of view @fiorentino13, using non-linear convective pulsation models, predicted a metallicy effect of $\sim 0.4\, \mbox{mag/dex}$ with more metal rich Cepheids being fainter. Instead, @anderson16 found an opposite trend, with more metal poor Cepheids being fainter, using linear non-adiabatic models.
Observationally no appreciable effect was found by @wielgorski17 and @groenewegen18. On the contrary @romaniello08, @freedman11 and @gieren18 found a photometric-band dependent effect. Nevertheless these studies found different magnitude of the effect and more notably a different sign. To be noted that only one study [@romaniello08] was conducted on direct Cepheids chemical abundances, although in a small sample, that limits the obtainable accuracy.
The LACES project
=================
Locking the Abundances of CEpheids for SH$_0$ES (LACES) is a project aimed of assessing the metallicity effect on Leavitt law in optical and NIR bands, through the direct spectrocopic measurements of chemical abundances over a large sample. Moreover it aims to link the Cepheid’s abundance scale to the SNe Ia host galaxies in SH$_0$ES sample. The project is divided in three steps.
Magellanic Clouds Cepheids are a very suitable anchor since MCs geometric distance is very well known [@pietrzynski19] and their metallicities span a wide range $-1 \lesssim [\mbox{Fe}/\mbox{H}] \lesssim 0.0 \, \mbox{dex}$. In these galaxies the LACES team collected spectra of more than 300 Cepheids with the FLAMES/GIRAFFE multi-object spectrograph at VLT. A sub-sample of these stars has been also observed with KMOS at VLT. To achieve $1\%$ error in distance, individual chemical abundances are required to have an error $<0.1 \, \mbox{dex}$.
The calibrated period-luminosity-metallicity relations will be used to calibrate the SNe Ia maximum brightness in SNe Ia host galaxies located at $\sim 40 \, \mbox{Mpc}$. Unfortunately, obtaining direct reliable chemical abundances for Cepheids at these distances is out of reach for current facilities, therefore Red Super Giants are used as metallicity proxy. An intermediate rung is thus needed to tie together Red Super Giant and Cepheid metallicity.
We concentrate here on the first rung of this chemical ladder, in particular how errors on temperature measurements can bias abundance measurements.
Systematic abundance errors due to temperature measurements
===========================================================
The degeneracy of the atmospheric a parameter can affect the chemical abundances measurements. The main one is the effective temperature, that can lead to a bias of $\sim 0.1\, \mbox{dex}$ every $100 \,\mbox{K}$ of error, the sign being dependent on the true effective temperature. This is even more compelling for stars that can vary up to $1000 \, \mbox{K}$ their effective temperature during one pulsation cycle.
In order to avoid bias due to reddening, which affect color-temperature relations, we decided to implement the Line Depth Ratio (LDR) method, that relies on the sensitivity to temperature of metallic lines [@gray89], empirically calibrated by @proxauf18 using Galactic Cepheids, that encompass more than 200 line doublets in the wavelength coverage of the LACES optical spectra.
To test how the estimated temperature is biased by the other atmospheric parameters and the instrumental set up, we built a synthetic spectra grid with Atlas9 (Kurucz) models. The grid spans the temperature range $3500-7800 \, \mbox{K}$, three surface gravity values: $log(\mbox{g})=0.0, \, 1.0, \, 2.0$, three metallicity values: $[\mbox{Fe}/\mbox{H}]=0.0, \, -0.5, \, -1.0$ and two spectral resolutions: $\mbox{R}=20000, \, 50000$. The difference between the LDR temperature and the input temperature of the synthetic spectra, as a function of this latter, is shown on Figure \[p18Synt\]. These residuals can be as high as hundreds of Kelvin at the edge of the grid and also the gravity and resolution dependencies are noticeable.
On Figure \[pair65\], the relation between the input temperature and the measured line depth ratios is shown for an S I - Ni I doublet. For a measured depths ratio, different metallicity models yield different temperatures.
To address this problem, we decided to calibrate the LDRs self-consistently over the synthetic spectra in order to minimize these dependencies. We used Chebyshev polynomials to fit temperature-ratio relations for each of the line pairs, minimizing the temperature residuals. We repeated the calibration over the three metallicity sets, and we applied them to the synthetic spectra grid to cross-check for metallicity dependence. This is shown in Figure \[syntcalib\]. The residuals are minimized by construction for the calibration at solar metallicity, but they show a trend with the input temperature when a different metallicity calibration is applied. At the colder edge of the grid the temperatures are overestimated of $\sim150\, \mbox{K}$ and $\gtrsim 300\, \mbox{K}$ applying respectively $[\mbox{Fe}/\mbox{H}]=-0.5$ and $[\mbox{Fe}/\mbox{H}]=-1.0$ LDR calibrations to solar metallicity spectra.
This not only affects the chemical abundance measurements but it also changes the inferred properties of the star. For example, applying the LDR calibration made on different metallicity with respect to the observed Cepheid up to $\sim 3\%$ difference in mean temperature and up to $\sim20\%$ difference temperature amplitude.
We also analyzed the UVES spectra of LMC Cepheids, already studied by @romaniello08. In Figure \[uves\], the temperatures measured applying the synthetic and empirical calibrations show striking differences. In order to minimize such effects, it is necessary to employ a calibration consistent with the models used for the abundance analysis.
Concluding remarks
==================
Classical Cepheids have a primary role as distance indicators and they provide the most precise anchor for the local H$_0$ measurements. In order to understand and solve the Hubble controversy, tight control of systematic errors is mandatory. In the classical distance ladder metallicity effects are not yet understood to the required level, the LACES team thus collected the largest spectroscopic sample of MCs Cepheids to have direct chemical abundance measurements. Through a synthetic spectra grid we investigated how systematic errors on LDR temperatures bias the abundance measurements although caveats on these measurements must be more carefully inquired.
[^1]: For a recent review about late- and early-time measurements see @verde19.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We perform two high resolution direct numerical simulations of stratified turbulence for Reynolds number equal to $Re\approx25000$ and Froude number respectively of $Fr\approx0.1$ and $Fr\approx0.03$. The flows are forced at large scale and discretized on an isotropic grid of $2048^3$ points. Stratification makes the flow anisotropic and introduces two extra characteristic scales with respect to homogeneous isotropic turbulence: the buoyancy scale, $L_B$, and the Ozmidov scale, $\ell_{oz}$. The former is related to the number of layers that the flow develops in the direction of gravity, the latter is regarded as the scale at which isotropy is recovered. The values of $L_B$ and $\ell_{oz}$ depend on the Froude number and their absolute and relative size affect the repartition of energy among Fourier modes in non easily predictable ways. By contrasting the behavior of the two simulated flows we identify some surprising similarities: after an initial transient the two flows evolve towards comparable values of the kinetic and potential enstrophy, and energy dissipation rate. This is the result of the Reynolds number being large enough in both flows for the Ozmidov scale to be resolved. When properly dimensionalized, the energy dissipation rate is compatible with atmospheric observations. Further similarities emerge at large scales: the same ratio between potential and total energy ($\approx 0.1$) is spontaneously selected by the flows, and slow modes grow monotonically in both regimes causing a slow increase of the total energy in time. The axisymmetric total energy spectrum shows a wide variety of spectral slopes as a function of the angle between the imposed stratification and the wave vector. One-dimensional energy spectra computed in the direction parallel to gravity are flat from the forcing up to buoyancy scale. At intermediate scales a $\sim k^{-3}$ parallel spectrum develops for the $Fr\approx 0.03$ run, whereas for weaker stratification, the saturation spectrum does not have enough scales to develop and instead one observes a power law compatible with Kolmogorov scaling. Finally, the spectrum of helicity is flat until $L_B$, as observed in the nocturnal planetary boundary layer.'
author:
- 'C. Rorai$^{1}$, P.D. Mininni$^{2}$ and A. Pouquet$^{3,4}$'
bibliography:
- 'ms\_14\_v2.bib'
title: 'Stably stratified turbulence in the presence of large-scale forcing'
---
Introduction {#sec:intro}
============
Geophysical fluid dynamics, as encountered in the atmosphere and the oceans, is at the center of our understanding and predicting capabilities in weather and climate. The modes that prevail in such systems are a mixture of nonlinear eddies and waves (for example, inertial waves when a solid-body rotation is considered, or internal gravity waves in a stratified flow). The nonlinear coupling between these modes leads to extreme events which are both sporadic and spatially localized, with steep gradients in the velocity and temperature or density fields, a phenomenon observed both in the stable planetary boundary layer [@lenschow_12], as well as in high-resolution direct numerical simulations (DNS) of the Bousinesq equations [@rorai_14]. Nonlinear interactions are also associated with other phenomena, such as wave steepening and breaking, instabilities (as the Kelvin-Helmoltz instability, a familiar phenomenon in a turbulent stratified atmosphere), and turbulent cascades. As these interactions take place in a wide range of scales, and as timescales in geophysical flows are not homogeneous in scale, universality (as is the case in homogeneous isotropic turbulence) is not necessarily obtained in such complex flows. Numerous observations point to a variety of regimes, for example in the case of surface waves in the ocean, which steepen into well-observed rogue waves [@choi_05b; @fedele_08; @dysthe_08].
Whether the flow is dominated by (strong) waves or by eddies depends a priori on the relative values of the parameters that characterize a given flow. In the case of stratification, these are the period of gravity waves $\tau_\omega\approx 1/N$, and the turnover time of eddies $\tau_\textrm{NL}\approx L_F/u_\textrm{rms}$ (with $N$, $u_\textrm{rms}$, and $L_F$ respectively the Brunt-Väisälä frequency and the characteristic velocity and lengthscale of the flow). The fastest timescale is expected to dominate the dynamics. However, as already mentioned, these timescales are not homogeneous with the lengthscale, and even if the waves dominate over the eddies at large scales, the two timescales may become comparable at a smaller scale. For stratification, the scale at which these two characteristic times are equal is called the Ozmidov scale $\ell_\textrm{oz}=2 \pi/k_\textrm{oz}$, with $k_\textrm{oz} = (N^3/\epsilon)^{1/2}$ and with $\epsilon$ the energy dissipation rate. Beyond $\ell_\textrm{oz}$, isotropy and a classical Kolmogorov range is expected. The scale $\eta$ at which dissipation sets in (of the order of the meter or several centimeters in the atmosphere) marks the end of the turbulence regime. We thus propose in this paper to study forced stratified turbulence using DNS at a sufficiently high resolution of $2048^3$ points to be able to contrast the evolution of such flows in two cases, varying only the ratio $\tau_\omega/\tau_\textrm{NL}$ by a factor 3.
One can find many reviews concerning stratified turbulence (see, e.g., [@riley_rev_00; @staquet_rev_02; @mcwilliams_04; @sagaut_cambon_08; @waite_14]). Concentrating on more recent numerical work, a few concepts seem to emerge, among a variety of possible settings (two-dimensional or three-dimensional forcing, acting at large scale or at small scale, balanced or not, etc.) [@bartello_13; @brethouwer_07; @waite2011; @kimura_12; @almalkie_12; @debruynkops_13; @debruynkops_14]. Several large DNS also consider the stratified case in the presence of shear [@fritts_09a; @chung_12] or of rotation [@smith_02; @4096_LS].
In stratified turbulence, the kinetic energy undergoes a direct cascade to small scales, and its spectrum follows at sufficiently small scales a Kolmogorov-like law in terms of $k_{\perp}$ (i.e., of wavevectors perpendicular to gravity) [@lindborg_07]. At small scales the buoyancy field is believed to follow an equivalent law, similar to that of a passive scalar, again in terms of $k_{\perp}$. However, the spectra in terms of $k_{\parallel}$ seem to follow a steeper $\sim k_\parallel^{-3}$ law, often called the saturation spectrum. Flat spectra at large scale, presumably larger than the buoyancy scale $L_B$, are also reported. Moreover, different simulations with varying configurations and parameters present different behavior.
In [@brethouwer_07], a large-scale two-dimensional forcing is used, with grids up to $1024^2\times 320$ points. Computations are performed at high Reynolds number $Re$ and small Froude number $Fr$, varying the buoyancy Reynolds number ${\cal R}_B=ReFr^2$. Two regimes are identified, for low or high ${\cal R}_B$, with steep spectra and laminar layers in the former case, and the $k_{\perp}^{-5/3}$ spectra for kinetic and potential energy and turbulent layers in the latter case. These findings confirm previous works (see [@brethouwer_07] for a detailed review), and are often put in the context of atmospheric observations. Similarly, oceanic measurements of eddy diffusivity have identified two regimes of mixing, in terms of the same parameter [@shih_05; @ivey_08]. Using larger grid resolution and hyper-viscosity but similar forcing, it is shown in [@waite2011] that resolving or not the buoyancy scale may affect the outcome as far as energy distribution among Fourier modes is concerned, with steeper spectra when $L_B$ is well resolved, and that there is a sharp spectral break at the buoyancy scale as already predicted by [@weinstock_78]. Note that steep spectra mean that non-local interactions between widely separated modes are dominant. Moreover, when energy spectra are steeper than $k^{-2}$, dissipation takes place predominantly at large scale, and one cannot properly talk of an energy cascade phenomenon in the sense that dissipation acts over the entire spectrum.
In [@kimura_12], the choice is made of a cubic grid of $1024^3$ points, and the spectral data is also analyzed in terms of the wave-vortical decomposition introduced in [@craya_58; @herring_74]. The spectra are found to be flat at large scale, a feature explained through the accumulation of sharp layers in the vertical direction. In [@almalkie_12], a set of large numerical simulations on grids of up to $4096^2\times 2048$ points are performed, and in these runs the Ozmidov scale is resolved. The horizontal spectra appear to follow again a $k_{\perp}^{-5/3}$ law, and it is noted that the direct cascade in the vertical direction provides a pathway to dissipation and is consistent with the generation of layers in the flow. This is also in agreement with the idea that the flow evolves towards the generation of layers such that the Froude number based on the vertical scale is of order unity [@billant_01], a feature already observed empirically in [@metais2]. These results are confirmed by yet higher resolution runs [@debruynkops_13; @debruynkops_14] on grids of up to $8192^3$ points (in the homogeneous isotropic case), at unit Prandtl number and with buoyancy Reynolds numbers of up to 220. Such a high resolution allows for a detailed investigation of intermittency. Finally, in [@bartello_13], the role of the buoyancy scale is confirmed; more importantly, the critical parameter to determine what scaling exponents prevail for the spectra seems to be the buoyancy Reynolds number ${\cal R}_B$: at large Reynolds number, the spectra are found to be independent of stratification.
What can be concluded from these past studies is that a consensus has not yet been reached as to whether there will be a universal description of such flows. In the present paper, we show that, as suggested already in [@cambon_94], some of the ambiguities found in preceding studies may well be linked to a competition between several phenomena, namely on one hand the growth of slow modes with $k_{\perp} \approx 0$, and the dynamics of fast modes with $k_{\perp} \neq 0$ on the other hand.
Methods {#sec:method}
=======
Equations
---------
The dynamics of a turbulent flow in a stably stratified environment can be described by the incompressible Navier-Stokes equations under the Boussinesq approximation. According to this model the three-dimensional velocity field ${\mathbf u({\bf x},t)}$ of components $(u,v,w)$, and the temperature fluctuations (or buoyancy field) $\theta({\bf x},t)$ obey the set of equations $$\begin{aligned}
\partial_t {\mathbf u} +{\mathbf u} \cdot \nabla {\mathbf u} &=& -\nabla P - N \theta\ e_z + \nu \Delta {\mathbf u}+{\bf f}_V, \label{eq:mom} \\
\partial _t \theta +{\mathbf u} \cdot \nabla \theta &=& N w + \kappa \Delta \theta , \\ \label{eq:temp}
\nabla \cdot {\bf u} &=&0;\end{aligned}$$ where $P$ is the pressure, $N$ the Brunt-Väisälä frequency, $\nu$ the viscosity, ${\bf f}_V$ a velocity forcing term, and $\kappa$ the thermal diffusivity. As customary, the Brunt-Väisälä frequency is defined by $N=\sqrt{-(g/\theta ) (d\bar \theta /dz)}$, where $g$ is the gravitational acceleration and $\bar \theta$ is a linear temperature profile. Equations (\[eq:mom\]) to (\[eq:temp\]) are solved with the pseudo-spectral Geophysical High-Order Suite for Turbulence (GHOST) code, which is parallelized with hybrid MPI/OpenMP programming, and has been tested on over 98,000 compute cores [@hybrid2011]. The code is based on a $2^{nd}$–order explicit Runge-Kutta temporal scheme, and uses a standard 2/3 de-aliasing rule in Fourier space.
The system can be characterized in terms of its energy, helicity, enstrophy, and dissipation rate, expressed either as a function of time or of Fourier space wavenumbers. The mean total energy $E_T$ is defined as the sum of the kinetic $E_V$ and potential $E_P$ energy $$\frac{1}{2} \left<|{\bf u}|^2 + \theta\ ^2\right>=E_V+E_P =E_T,$$ and is a conserved quantity in the ideal limit. The brackets indicate the spatial mean. We also define helicity, the velocity-vorticity correlation, as $$H_V = \left<{\mathbf u} \cdot {\mathbf \omega} \right>.$$ Helicity is an invariant of the inviscid non-stratified equations, it affects the cascade of energy in the presence of rotation [@helirot], and has been observed to considerably slow-down the decay of turbulent energy in the presence of stratification [@rorai_13].
The kinetic enstrophy, proportional to the kinetic energy dissipation, is given by $Z_V = \left<\omega^2 \right>$. Similarly, the potential enstrophy is $Z_P =\left<|\nabla \theta | ^2 \right>$ and is associated with the dissipation of potential energy. As only the velocity field is forced, the total injection rate is simply given by $$\varepsilon_V = \left< {\bf u} \cdot {\bf f}_V \right>. \label{epsilon}$$ In the turbulent steady state, this quantity is expected to be equal (on the time average) to the total dissipation rate $\nu \left<\omega^2 \right> + \kappa \left<|\nabla \theta | ^2 \right>$.
Finally, we define the reduced energy and helicity spectra. In Fourier space the velocity autocorrelation function is noted $U_{ij}(k_x, k_y, k_z)$ and its trace is $U({\bf k})$. Hence the axisymmetric kinetic energy spectrum is $$e_V(\mathbf{|k|}, \Theta)=\int U({\bf k}) |{\bf k}| \sin \Theta d\phi,$$ where $\phi$ is the longitude with respect to the $k_x$ axis, and $\Theta$ is the co-latitude. By defining parallel $k_{\parallel}=k_z$, perpendicular $k_{\perp}=|{\bf k_{\perp}}|=|{\bf k}| \sin \Theta$, and isotropic $k=|\bf k|$ wavenumbers, we can calculate parallel, perpendicular, and isotropic reduced kinetic energy spectra as follows [@3072] $$\begin{aligned}
E_V(k_{\parallel})&=&\int e_V(\mathbf{|k_{\perp}|}, k_{\parallel})dk_{\perp} \label{E2} , \\
E_V(k_{\perp})&=&\int e_V(\mathbf{|k_{\perp}|}, k_{\parallel})dk_{\parallel} \label{E1} , \\
E_V(k)&=&\int e_V(\mathbf{|k|},\Theta)|{\bf k}|d\Theta \label{E3} .
\label{aniso}\end{aligned}$$ Similar definitions hold for the potential and total energy, and for the helicity spectrum $h(|{\bf k}|,\Theta)$, which is related to the antisymmetric part of the velocity correlation tensor [@3072].
We can also distinguish between slow and fast mode spectra, namely: $$\begin{aligned}
&E_{V, \textrm{slow}}&= e_V(\mathbf{|k_{\perp}|}=0, k_{\parallel}) \label{E4} , \\
&E_{V, \textrm{fast}}&= \int_{k_{\parallel}=0}^{k_{{\parallel}_{\max}}} \int_{|k_{\perp}|=1}^{|k_{\perp}|_{\max}} e_V(\mathbf{|k_{\perp}|}, k_{\parallel})dk_{\perp} dk_{\parallel} \label{E5}.
\label{aniso2}\end{aligned}$$ Equivalent definitions hold for $E_{P, \textrm{slow}}$, $E_{P, \textrm{fast}}$, $E_{T, \textrm{slow}}$, and $E_{T, \textrm{fast}}$. The slow modes satisfy the condition $\omega=0$, where $\omega$ is the frequency of gravity waves given by the dispersion relation $\omega=\sqrt{N^2k_{\perp}^2}/k$. These modes correspond to “pure” eddies (vortical motions), and their characteristic time scale is the eddy turnover time. When $Fr<1$, the waves at large scales are faster than the eddies, or in other words, the wave period is faster than the turnover time. This is why the remaining modes, which correspond to a combination of eddies and waves, are often called “fast” modes.
The fluxes of kinetic and potential energy are respectively given by: $$\begin{aligned}
&\Pi_{V}(k)&= \int_0^k \Gamma(k')dk' , \\
&\Pi_{P}(k)&= \int_0^k \mathrm{P}(k')dk' \label{pi},
\label{aniso2}\end{aligned}$$ with $\Gamma(k)={\bf u}^{\star}(k)\cdot\mathcal{F}({\bf u}\cdot\nabla {\bf u})_k$ and $\mathrm{P}(k)={\bf u}^{\star}(k)\cdot\mathcal{F}(\rho\nabla \rho)_k$, where $\mathcal{F}$ denotes the Fourier transform.
Initial conditions and forcing
------------------------------
Equations (\[eq:mom\]) to (\[eq:temp\]) are integrated numerically in a cubic domain of volume $V=(2\pi)^3$, discretized by an isotropic grid of 2048$^3$ points. The initial condition and the velocity forcing, ${\bf f}_V$, consist of randomly generated isotropic three-dimensional flows [@patterson] with injection wave number $k_F$ between $2$ and $3$. The forcing has amplitude $f_\textrm{rms}=0.22$, chosen to yield an approximately unitary r.m.s. velocity ($u_\textrm{rms}=1$) in the turbulent steady state. We impose $\nu=10^{-4}$, which guarantees that the Kolmogorov scale for a homogeneous isotropic flow with the same parameters and discretization is well resolved [@mininni_nonlocal]. Note this is a conservative choice, since in wave turbulence the energy spectrum is expected to be steeper, and therefore the small scales are expected to be less energetic. As a result of these choices, the ratio between the smallest and largest scales resolved in our calculations is about $\approx 700$, and the Reynolds number is $Re\approx 25000$.
We perform two runs with different values of the Brunt-Väisälä frequency, resulting in $Fr\approx 0.1$ (for $N=4$), and $Fr\approx 0.03$ (for $N=12$). The buoyancy Reynolds number ${\cal R}_B=ReFr^2$ is correspondingly ${\cal R}_B=250$ and ${\cal R}_B=27$. The calculations are carried out for respectively 30 and 20 time units. Finally, in both runs we consider a unitary Prandtl number $Pr=\nu/\kappa=1$.
Characteristic scales
---------------------
For our simulations of stratified flows, the relevant length scales are:
1. The overall size of the periodic domain, equal to $L_0=2\pi$ in dimensionless units, and with associated wavenumber $k_0 = 2\pi /L_0$.
2. The scale at which energy is injected into the system, $L_F=2\pi/k_F$.
3. The buoyancy scale, $L_B=2\pi / k_B$, with $k_B = N/u_\textrm{rms}$, characteristic of the vertical shear.
4. The scale at which isotropy (and presumably a Kolmogorov energy spectrum) is recovered, namely the Ozmidov scale $\ell_\textrm{oz}=2\pi/k_\textrm{oz}$, with $k_\textrm{oz} = (N^3/\varepsilon)^{1/2}$.
5. The dissipation scale, $\eta=2\pi/k_\eta$, with $k_\eta = (\varepsilon/\nu^3)^{1/4}$.
6. The smallest scale resolved in the DNS, namely $\ell_\textrm{min}=2\pi/k_\textrm{max}$. Because of the Fourier transform the pseudospectral code is based upon, and the 2/3-rule for removing aliasing $k_\textrm{max}=n/3 \approx 700$ where $n$ is the number of grid points per dimension.
In Table \[scales\] we report the values of the wavenumbers associated with these characteristic scales.
The Ozmidov and the dissipation scales are usually evaluated by estimating $\varepsilon \approx u_\textrm{rms}^3/L_F$. However, this estimation is valid for isotropic and homogeneous turbulence, while in a stratified flow the energy injection rate, flux, and dissipation rate can be strongly affected by the waves. We estimate then $\varepsilon \approx \varepsilon_V$, following the definition in Eq. (\[epsilon\]) which corresponds to the effective rate of transfer in the flow. Using $\varepsilon_V$ yields a value one order of magnitude smaller for the injection and dissipation rates (see Sec. \[sec:temporal\]). The estimates obtained in this way are marked by a star in Table \[scales\]: $k_{oz}^*$ and $k_{\eta}^*$. As will be shown later, these quantities give a better estimation of at what scales transitions occur in the flow. In practice, a well-resolved run requires $k^*_\eta<k_\textrm{max}$ as dissipation starts to dominate the dynamics at this wavenumber; observe that this is satisfied by our simulations. Also, it can be easily shown that the Ozmidov scale is resolved (e.g., $k^*_\textrm{oz} < k^*_\eta$), when ${\cal R}_B=ReFr^2 \ge 1$; at the scale at which $k^*_\textrm{oz} = k^*_\eta$ the wave period equals the eddy turnover time.
A second effective estimate of the vertical characteristic scale, which can be associated with the bouyancy scale, is given by the integral scale based on the parallel potential energy spectrum, as layers tend to develop more clearly in the temperature: $$L_{B}^*=2\pi\frac{\int E_P(k_{\parallel})/k_\parallel dk_{||}}{\int E_P(k_{\parallel}) dk_{||}} .$$ The corresponding wavenumber $k_B^*=2 \pi /L_{B}^*=2$ is also reported in Table \[scales\].
Runs $N=4$ $N=12$
------------------- ------- --------
$k_0$ 1 1
$k_F$ 2-3 2-3
$k_\textrm{max}$ 683 683
$k_B$ 4 12
$k_B^*$ 7 8
$k_\textrm{oz}$ 13 66
$k_\textrm{oz}^*$ 36 186
$k_\eta$ 795 795
$k_\eta^*$ 472 472
: Wavenumbers corresponding to the box size ($k_0$), injection scale ($k_F$), grid resolution ($k_\textrm{max}$), bouyancy scale ($k_B$ and $k_B^*$, where wavenumbers without a star are computed using $\epsilon \approx u_\textrm{rms}^3/L_F$, and wavenumbers with stars are computed using the measured injection rate $\epsilon_V$), Ozmidov scale ($k_\textrm{oz}$ and $k_\textrm{oz}^*$), and dissipation scale ($k_\eta$ and $k_\eta^*$).[]{data-label="scales"}
Finally, we attempt to assign physical values, characteristic of the atmosphere and the oceans, to the run at the smallest Froude number, with $Fr=0.03$ and $Re \approx 2.5 \times 10^4$. For the atmosphere we assume $u_\textrm{rms}=1$ ms$^{-1}$ and $L_0=1000$ m (roughly the size of a small convective cell). Hence it is readily found that $N \approx 3.3 \times 10^{-2}$ s$^{-1}$, $\varepsilon_V \approx 4\times 10^{-5} $m$^2$s$^{-3}$ (per unit mass), and $\nu=0.04$ m$^2$s$^{-1}$, clearly too large for the atmosphere, as expected, given the limited grid resolution. However, the energy injection rate is close to atmospheric values, which yield $\varepsilon_V \approx 10^{-6}$–$10^{-5} $m$^2$s$^{-3}$ from data analysis of aircraft measurements [@Lindborg99] and of satellite images [@Heas12]. From these values it also follows that $L_B \approx 190$ m and $\ell_\textrm{oz} \approx 33$ m, to be compared to the Kolmogorov dissipation scale of $ \approx 3$ m and to the grid resolution of $\ell_\textrm{min} \approx 1.4$ m. If we consider the ocean instead, the typical velocity is ten times smaller. Hence, given the same Reynolds and Froude numbers, $L_B$ and $\ell_\textrm{oz}$ remain the same, while the Brunt-Väisälä frequency and the viscosity are reduced by an order of magnitude. Yet, $\nu=0.004$ m$^2$s$^{-1}$ is three orders of magnitude larger than realistic values.
Results {#sec:numerics}
=======
We present first in Sec. \[sec:temporal\] the temporal behavior of small and large scale quantities, integrated over the entire domain. Quantities associated with the dynamics of small scales ($Z_V$, $Z_P$ and $\varepsilon_V$) reach a steady turbulent regime at an early stage. In contrast, quantities associated with the energetics of the large scales ($E_T$, $E_{T, fast}$ and $E_{T, slow}$) do not converge to a statistical steady state by the end of our calculations. In this case, stationarity is primarily prevented by the monotonic growth of slow modes, as also found for example in [@smith_02]. Then, in Sec. \[sec:spectra\] we present the energy and helicity distribution among Fourier modes, including a study of spectral anisotropy. Energy and helicity spectra are averaged in time to obtain a representative statistical behavior, and early times are excluded from the average as the turbulent regime is established only after an initial transient in which the flow adapts to the forcing and develops small-scale structures. Finally, in Sec. \[sec:fluxes\] we comment on the energy fluxes.
Temporal evolution of global quantities\[sec:temporal\]
-------------------------------------------------------
Figure \[PE\_KE\_t\] shows the ratio of the potential to the total energy as a function of time. The eddy turn-over time is $\tau_{NL}\approx 2.5 t$. After an initial transient, both curves, independently of the stratification strength, reach a value of $E_P/E_T\approx 0.1$ about $t\approx14$, followed by a slow monotonic decrease. This ratio is comparable with that found in [@brethouwer_07; @kimura_12] for similar values of ${\cal R}_B$.
In Fig. \[eta\_t\](a) the temporal evolution of the kinetic and potential enstrophy is shown. The initial transient is characterized by the development of small-scales through non-linear mode coupling. This mechanism is less efficient at low $Fr$ as testified by the smaller values of the two enstrophies at early times, resulting also in a smaller value for the dissipation of total energy $\left< \nu |\omega|^2 + \kappa |\nabla \theta|^2 \right>$ for the $N=12$ run. The enstrophy maxima occur earlier in terms of the eddy turn-over time, but not in terms of the buoyancy period, for the less stratified flow. Turbulence can be said fully developed beyond the peak of enstrophy where dissipation reaches the smallest scales. Interestingly, at later times, the curves of the two runs merge and undergo a slow decay. This behavior can be identified with the achievement of a turbulent steady state, at least at small scales and as long as the slow modes are not dominant. Note that as in the case of the ratio $E_P/E_T$, the enstrophies (and as a result, the energy dissipation rates) also tend to converge to similar values independently of the two stratification strengths considered.
As a comparison, in Fig. \[eta\_t\](b) we show the temporal behavior of the energy injection rate computed using Eq. (\[epsilon\]). The similarity of the two runs is remarkable. A dimensional Kolmogorov-like evaluation of the energy injection rate for fully developed turbulence $\varepsilon \approx u_{rms}^3/L_F$, yields, for our r.m.s. velocity and forcing scale, $\varepsilon \approx 0.4$, an estimate one order of magnitude larger than the numerical value $\varepsilon_{V}\approx 0.04$. This latter value is also compatible with the dissipation rate at late times obtained from $\left< \nu |\omega|^2 + \kappa |\nabla \theta|^2 \right>$. The order of magnitude difference between the Kolmogorov-like estimation and the actual values of injection and dissipation can be understood as in wave turbulence the energy transfer rate is expected to be smaller than $\approx u_{rms}^3/L_F$ by a factor $Fr$, as indicated by numerous studies [@newell_01; @nazar]. However, it should be noted that this argument fails to explain why $\varepsilon_{V}$ in the two simulations has similar values independently of the value of $Fr$. One possibility is that much of the dissipation occurs in the strong gradients that develop in the vertical in order to insure that the Froude number based on a characteristic vertical scale is of order unity, in which case the weak turbulence argument can only apply to the horizontal dynamics. In other words, the lesser dissipation in the horizontal for smaller Froude number is compensated almost exactly by the increased dissipation in the vertical. This phenomenon is related to the distribution of energy between the potential and kinetic modes on the one hand, and between the vertical and horizontal kinetic energy on the other hand.
We then conclude, from Fig. \[eta\_t\], that small scales have saturated and we remark that the two runs have a tendency towards identical dissipation, a surprising result since the buoyancy Reynolds numbers differ by almost an order of magnitude (but are in both cases above a critical ${\cal R}_B\approx 10$). This result is consistent with the recent finding in [@bartello_13] that the energy spectra are independent of stratification at sufficiently high Reynolds number. However, note that while this is indeed the case for the spectrum of small scale fluctuations, it is not the case at large scales as will emerge from the analysis of the spectra.
As a first indication of differences at large scales, the time evolution of the total energy is shown in Fig. \[L\_t\](a). As turbulence develops, the total energy grows until it reaches a peak. At a later stage a fluctuating behavior, characteristic of the turbulent steady state, is expected. However, after the peak, we observe a new monotonic increase with a timescale which is larger than characteristic times such as $1/N$ or $\tau_{NL}$. It is evident from Fig. \[L\_t\](b) that the energy increase is due to the growth of the slow modes, while the energy in fast modes gradually decreases. We verified on a lower-resolution run and with forcing at substantially smaller scale that the growth of the fast modes saturates after thirty turn-over times, and a steady state is indeed reached [@marino_14].
Several remarks follow from considering the dispersion relation of internal gravity waves $$\omega= \pm k^{-1}\sqrt{N^2k^2_{\perp}} \ .$$ A review of various theoretical approaches to stratified turbulence viewed as a superposition of internal waves can be found for example in [@muller_86; @staquet_rev_02; @sagaut_cambon_08; @polzin]. Here it is of interest to recall that three-wave interactions at resonance play a central role in closing the cumulant expansion; in the present case, for a usual triad of modes ${\bf k}$, ${\bf p}$ and ${\bf q}$ satisfying ${\bf k} = {\bf p} + {\bf q}$ , they read $$s_k \frac{k_{\perp}}{k} = s_p \frac{p_{\perp}}{p} + s_q \frac{q_{\perp}}{q} ,$$ where $s_k$, $s_q$ and $s_q=\pm 1$ depending on the branch of the dispersion relation used. As remarked in [@waleffe_93; @smith_99], this resonance condition is readily satisfied for $k_{\perp} \approx p_{\perp} \approx q_{\perp} \approx 0$. Hence, it can be inferred that the wave-wave interactions lead to a build-up of energy at $k_{\perp} \approx 0$, i.e., in the so-called slow modes. The disctintion between energy in fast modes and in slow modes in Fig. \[eta\_t\] is compatible with this build-up of energy in modes with $k_{\perp} \approx 0$.
The accumulation of energy for $k_{\perp}=0$ was already noticed in [@cambon_94], using a two-point closure of turbulence, the so-called EDQNM2 (Eddy Damped Quasi-Normal Markovian 2 closure). It was also found in [@smith_02; @brethouwer_07] in DNS, and attributed to the growth of the so-called vertically sheared horizontal winds. It is to be noted that such winds, which are off-diagonal elements of the velocity gradient matrix, constitute the vorticity field for negligible vertical velocity and as such they make the velocity field non potential. In the presence of rotation, these are the so-called thermal winds which are a-geostrophic corrections to geostrophic balance. It is interesting (and may be viewed as somewhat paradoxical) that resonant interactions of gravity waves can lead to the growth of vortical modes which eventually come to dominate the flow.
We finally comment on the temporal dynamics of helicity and relative helicity, $\sigma_V=H_V (E_VZ_V)^{-1/2}$. Although we do not use a helical forcing, we are not imposing the forcing to be completely non-helical. As a result, there is a small amount of helicity in the flow at late times. The time evolution is at follows: helicity starts from zero and it undergoes an oscillatory transient, longer for stronger stratification. In both runs, $\sigma_V$ then grows slowly in time until it reaches a final value of $\approx 0.12$ for the $N=4$ run. Similarly, and as will be shown later, the relative spectral density $H_V(k)/[kE(k)]$ remains low, of the order of $0.05$, except in the vicinity of the forcing scales. These values are too small to affect the energetics of the flows for either runs. In [@rorai_13], the dynamics of helicity in freely decaying stratified turbulence was examined in terms of a possible balance between its production and dissipation. It was found that only when the initial condition was an ABC (maximally helical) flow, the energy spectrum was modified by the slowed-down dynamics inherent to the helical case, a situation also found in rotating flows [@teitelbaum].
Energy distribution among modes\[sec:spectra\]
----------------------------------------------
### The development of anisotropy
We now consider the energy and helicity spectra. We first show, in Fig. \[cont\], the isocontours of the axisymmetric kinetic energy spectrum $e_V({k_{\perp},k_{\parallel}})$ (normalized by $\sin \Theta$ to obtain circular isocontours in the case of an isotropic flow). When stratification is stronger, the flow is more anisotropic, i.e., the contours are more stretched along the vertical axis. In other words, there is more energy in the slow modes with $k_\perp \approx 0$, as described before. However, the anisotropy is not the same for all wavenumbers; small wavenumbers (large scales) tend to be more anisotropic than large wavenumbers (small scales). This is more clear in the less stratified flow. Indeed, a rough estimate of the recovery of isotropy can be made by considering the wavenumber for which the contour lines approach a circular shape. This approximately occurs at large wavenumbers for $N=4$, whereas for $N=12$ isotropy is only barely recovered at the smallest resolved scales.
The axisymmetric total energy spectra $e(k_{\perp}, k_{\parallel})=e_V(k_{\perp}, k_{\parallel})+ e_P(k_{\perp}, k_{\parallel})$ for various values of the co-latitude $\Theta$ are shown in Fig. \[2Dspec\]. The dominance of the slow modes (i.e., modes with $k_{\perp}=0$) is evident for both runs. However, the modes with $k_{\perp}=0$ have different spectra at large scales depending on the value of $N$. While in the run with $N=4$ the spectrum is steep, in the run with $N=12$ the spectrum is shallower and almost flat (see, e.g., [@kimura_12] for previous observations of the flat spectrum). We recall that the isotropic spectra for a given $k$ is obtained by summing over $\Theta$ \[see Eq. (\[E3\])\]. Since the energy distribution depends on the stratification intensity, the isotropic spectrum for stratified fluids will be different for different stratifications. Moreover, some of the spectra in Fig. \[2Dspec\] are rather steep \[note that a spectrum $\sim k^{-4}$ in $e(k_{\perp}, k_{\parallel})$ corresponds, after integration, to a power law $\sim k^{-3}$ in the isotropic spectrum\]. For flows whose isotropic spectrum has an inertial index $\alpha$ which falls outside the range $-1<\alpha<-3$, arguments for locality of interactions do not hold, and interactions between modes can become non-local. Both these effects can give rise to non-universality of the spectrum (i.e., as a result of the existence of different physical regimes, or as the result of non-locality). This has been discussed often in the context of numerous and detailed observations of oceanic and atmospheric flows [@polzin], and also noted for example in the framework of internal waves in the ocean in the hydrostatic (and irrotational) limit [@lvov_10].
As already mentioned, in the weakly stratified run, the dominant modes seem to follow a $k_{\parallel}^{-4}$ spectrum, the so-called saturated spectrum (in these units, the isotropic spectra being recovered after one integration over wavenumber), whereas a short range of wavenumbers at small scales is compatible with a $k_{\perp}^{-8/3}$ law (corresponding to a Kolmogorov spectrum after integration). Moreover, note that $k_\textrm{oz}^* \approx 36$ in this run; for smaller wavenumbers all angular spectra start to collapse indicating a return to isotropy. On the other hand, in the run with $N=12$, $k_\textrm{oz}^* \approx 186$ and the spectra only collapse in the dissipative range. It is interesting and significant to note that the wavenumbers based on the dynamical injection rate obtained from $\epsilon_V$ ($k_\textrm{oz}^*$) give a much better estimation of the return of isotropy than those obtained from the Kolmogorov estimate $\epsilon \sim u_\textrm{rms}^3/L_F$ ($k_\textrm{oz}$).
In Ref. [@chung_12] it is suggested to analyze the data once the slow modes, corresponding to the $k_{\perp}=0$ modes, are removed. The analysis of the flow in terms of two-dimensional spectra allows for such a reduction, since the $k_{\perp}=0$ modes are confined to the $\Theta=0$ angle (black solid line in Fig. \[2Dspec\]). Remarkably, the data seems to separate into two ranges of scales with different behaviors, one range similar to the $\Theta=0$ spectrum, and another similar to the $\Theta=\pi/2$ spectrum.
### The resulting one-dimensional energy spectra
We now consider isotropic, perpendicular and parallel reduced spectra in the light of the spectra studied above. The one-dimensional isotropic total and potential energy spectra, and the parallel and perpendicular spectra, are shown in Fig. \[Ek\]. The spectra are averaged in time from the peak of enstrophy until the final time. It was remarked in [@cambon_94] that two-dimensional spectra (Figs. \[cont\] and \[2Dspec\]) may represent a more realistic diagnostics of anisotropic flows given the wide variety of spectral slopes they display as a function of the angle and as a function of the imposed stratification. While this is clear from the previous analysis, the Ozmidov scale and the buoyancy scale will show up more clearly in some of the reduced spectra \[in particular, for the buoyancy scale, in the potential energy spectrum $E_P(k_{\parallel})$\].
Figure \[Ek\](a) shows the isotropic total energy spectrum for both runs. The spectrum displays a peak associated with the forcing wavenumber, followed by a flat range (specially in the run with $N=12$). This flat range extends until a wavenumber close to $k_B$ (see Table \[scales\]). Then, the run with $N=12$ shows a steep spectrum compatible with $\sim k^{-3}$, while the run with $N=4$ has a short steep range followed by an incipient range compatible with $\sim k^{-5/3}$ after $k_\textrm{oz}^*$. The flat spectrum at large scales is also visible in the parallel spectrum $E_T(k_{\parallel})$ in Fig. \[Ek\](b), but is more evident in the parallel spectrum of potential energy in Fig. \[Ek\](d). On the other hand, the perpendicular spectra in Fig. \[Ek\](c) and Fig. \[Ek\](d) are consistent with $\sim k^{-5/3}$ at all scales and independently of the stratification, as observed before in [@lindborg_07].
The flatness of the spectra at large scales is due to the combination of two related factors: (i) the dominance of the $k_{\perp}=0$ modes observed in Fig. \[2Dspec\], and (ii) the organization of the flow in the vertical direction in well-defined strata with strong vertical gradients both in the velocity and in the buoyancy. It was shown, for example in [@kimura_12], that a superposition of such strata can indeed lead to a flat spectrum since, at large scale, these layers can be interpreted as quasi-discontinuities.
Moreover, the scale at which this flat spectrum ends seems to depend linearly with the Froude number, at least for our two runs. The buoyancy scale is generally understood in the context of theoretical studies (see, e.g., [@billant_01]) by advocating that the development of turbulence in the vertical direction leads to an effective vertical Froude number $Fr_z=u_\textrm{rms}/[L_B^* N]$ of order unity. In a different context, the buoyancy wavenumber was introduced before in [@weinstock_78] to take into account the fact that, in the Lagrangian framework, the buoyancy field is advected by the velocity (although not as a passive scalar) and thus should depend on the total kinetic energy. This leads to the prediction of a sharp break in the buoyancy flux spectrum $\left<w\theta\right>$ at $k_B$, a break that should not develop in the kinetic energy spectrum.
As one moves in the spectra in Fig. \[Ek\] to larger wavenumbers, the layers begin to be resolved and their intrinsic dynamics arises. There, the so-called saturation develops; it corresponds to a balance in the vertical between nonlinear advection and buoyancy and leads to the $\sim k^{-3}$ spectrum which is more clear for the strongly stratified run ($N=12$). For the weaker stratification the saturation spectrum does not have enough scales to develop and instead one observes a spectrum steeper than the Kolmogorov spectrum in a short range of wavenumbers. Then the collapse of the anisotropic spectra for different $\Theta$ explains the shallower and Kolmogorov-like spectrum for wavenumbers larger than $k_\textrm{oz}^*$ in $E(k)$ for the run with $N=4$. Note that using large-eddy simulations, the transition from a steep (saturated) large-scale spectrum to a Kolmogorov isotropic spectrum was observed before in [@carnevale_01] but only sporadically, when breaking events occurred and the turbulence was thus more vigorous.
Finally, Fig. \[Hk\] shows the spectrum of the absolute value of helicity for both simulations. Although helicity here is rather small and not important for the flow dynamics, the spectra display, as in the decay runs in Ref. [@rorai_13], a flat region at large scale followed by a decay at smaller scales with a break near the buoyancy scale. There are rapid changes of sign in the small scales, manifesting as large fluctuations. Interestingly, flat helicity spectra in the planetary boundary layer have been observed at night when the flow is more stably stratified [@koprov_05].
Energy fluxes\[sec:fluxes\]
---------------------------
An examination of energy fluxes confirms the analysis presented in the preceding section. The two simulations, at buoyancy Reynolds numbers of $\approx 27$ and $\approx 220$ respectively, behave differently as to how the energy is being transferred among scales, as can be seen in Fig. \[PIVPT\] which displays the kinetic, potential, and total energy fluxes for both runs. Note that the flux constructed from taking only the dot product of Eq. (\[eq:mom\]) with the velocity (i.e., the “kinetic energy flux”) is not a flux, in the sense that its divergence is not zero (i.e., kinetic energy is not conserved alone). Instead, this “flux” should be interpreted as energy flux plus power: when it is larger than zero, kinetic energy is transferred towards smaller scales by the velocity field, or injected per unit of time by work done by the temperature. The same applies to the “potential energy flux” constructed from dotting Eq. (\[eq:temp\]) with the temperature fluctuations: when it is positive, potential energy is transferred towards smaller scales or injected by work done by the velocity, while when it is negative potential energy may be removed by work done by the velocity. Only the total energy defines a proper flux, in the sense that its sign is solely associated with direction of transfer across scales, and in the sense that it goes to zero for $k\to \infty$ (i.e., the total energy is conserved).
In the less stratified run with $N=4$ (high buoyancy Reynolds number ${\cal R}_B$), the total energy flux is approximately constant in a range of wavenumbers that in fact defines the inertial range, with amplitude $\approx 2.7 \times 10^{-2}$. The potential and kinetic energy flux, in the light of the total flux, then indicate how energy is exchanged between the velocity field and the temperature. The potential energy flux is zero at large scale and rather small ($\approx 5 \times 10^{-4}$, or roughly 2% of the kinetic energy flux) in the same inertial range. It becomes negative and progressively larger (in absolute value) at small scale (after $k\approx 40$), at the end of the inertial range and for $k=k_\textrm{max}$, it reaches $\approx -7 \times 10^{-3}$, a value compensating the kinetic flux at that wavenumber, a condition necessary for energy conservation. The negative value of this flux at small scales indicates that energy is transferred from the small scale temperature fluctuations to the velocity field fluctuations (or in other words, that the small scale temperature gradients exert work on the velocity field, exciting small scale motions). This is in good agreement with the evolution of the enstrophies observed in Fig. \[eta\_t\]: more energy is dissipated by small scale velocity fluctuations (i.e., by the kinetic enstrophy) than by temperature fluctuations (whose dissipation is associated with the potential enstrophy). Energy at small scales then is transferred from the temperature to the velocity, where it is finally dissipated.
The dynamics of energetic exchanges is rather different at low ${\cal R}_B$. Although the same trends are observed, there is barely a range where the total energy flux is constant; furthermore, all three fluxes are larger in amplitude, but the ratio of kinetic to potential flux is now only roughly equal to 5 at large scales and the potential flux becomes negative at a higher wavenumber $(\approx 100$). One is led to the conclusion that, at that Froude number (and buoyancy Reynolds number), the flow is not sufficiently turbulent even though it produces strong gradients in the vertical.
Discussion and conclusion {#s:conclu}
=========================
We performed high resolution direct numerical simulations of stratified turbulence for Reynolds number equal to $Re\approx25000$ and two different Froude numbers: $Fr\approx0.1$ and $Fr\approx0.03$, corresponding to different stratification strengths. Stratified turbulence is modeled through the Boussinesq equations integrated numerically in a three-periodic cubical domain of volume $V = (2\pi)^3$, and discretized with an isotropic grid of $2048^3$ points. The flow is forced at large scale ($k_F = 2$ and $3$) by a three-dimensional randomly generated forcing. By contrasting the behavior of the two simulations we identify some similarities despite the fact that the buoyancy Reynolds number differs by almost an order of magnitude: after an initial transient the two runs have comparable values of the kinetic and potential enstrophy ($Z_V$, $Z_P$), and energy injection rates ($\varepsilon_V$). The same ratio between potential and total energy ($E_P/E_T\approx 0.1$) is spontaneously selected by the flows. For both values of $Fr$, slow modes grow monotonically as a consequence of nonlinear interactions and cause a slow increase of the total energy ($E_T$) in time. The axisymmetric kinetic energy spectrum, $e_V(k_\perp, k_\parallel)$, clearly shows the anisotropy of the flow, which survives at small scales for the $Fr\approx0.03$ run but not for the simulation at weaker stratification. The axisymmetric total energy spectrum, $e_V+e_\theta$, shows a wide variety of spectral slopes as a function of the angle between the imposed stratification and the wavevector, and with a clear dominance of the slow modes. As a result, the isotropic total energy spectrum is ambiguous because of the superposition of these different dynamical regimes. One-dimensional energy spectra computed in the direction parallel to gravity are flat from the forcing until the buoyancy scale $k_B$. At intermediate scales, a $k_\parallel^{-3}$ parallel spectrum, consistent with the simple 1D model presented in [@rorai_14], develops for the $Fr\approx 0.03$ run, whereas for $Fr\approx 0.1$ the saturation spectrum does not have enough scales to develop and instead one observes a larger slope compatible with a Kolmogorov spectrum $k_\parallel^{-5/3}$. Finally, the spectrum of helicity (velocity-vorticity correlations) is rather weak, but behaves as observed in decaying simulations in [@rorai_13], with its distribution among scales being flat until $L_B$, as also observed in the nocturnal planetary boundary layer.
As observed before in the literature, the dynamics of stratified turbulence proves to be more complex than the homogeneous isotropic case, specially at values of the buoyancy Reynolds number that are intermediate, when waves and eddies strongly interact. Further studies are needed, in particular because there are several relevant scales that must be separately resolved. One issue concerns the effect that the choice of forcing can have on the outcome of the simulations which is far from evident (see for example the discussion in [@carnevale_01]). The anisotropic development of large-scales and the ensuing lack of inverse cascade was analyzed in [@marino_14], whereas in this paper we deal with the small-scale anisotropy. However, in these studies isotropic forcing (and initial conditions) were used not to bias the development of angular variations. Differences may arise when other forcings, or when correlations between the temperature and the velocity field, are imposed.
Another set of issues is related to the difficulty to perform experiments, in the laboratory as well as numerically, for a set of parameters that accommodates the vast range of physical conditions found in geophysical and astrophysical flows. Part of the difficulty in reaching a full understanding of the behavior of stratified turbulence is the fact that there are different regimes in competition, and that for realistic parameter values for flows in geophysical and astrophysical fluid dynamics, the buoyancy Reynolds number must be sufficiently high, a feature difficult to realize numerically at small Froude number [@bartello_13]. But how much more resolution is needed? Spanning the whole range of multi-scale interactions from the largest scale to the dissipative scale, to cover a potential inverse transfer feeding the slow modes [@waleffe_93; @smith_99; @smith_02; @marino_14], a range where energy is fed into the system, a range dominated by wave interactions, followed by a range dominated by nonlinear eddies, and finally a dissipative range is currently impossible, and various choices have been made in the past even when computing at high resolutions up to $8192^3$ grid points [@debruynkops_14] (see also [@brethouwer_09; @3072; @bartello_13]). Resort to high-performance computing at higher resolution, as well as to modeling, will be some of the avenues to be followed in the near future.
[*This work was supported by NSF/CMG 1025183. C. Rorai acknowledges support from two RSVP/CISL grants. Computer time on Yellowstone through an ASD allocation was provided by NCAR under sponsorship of NSF.*]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We construct a new family of simple $\mathfrak{gl}_{2n}$-modules which depends on $n^2$ generic parameters. Each such module is isomorphic to the regular ${\mathcal{U}}(\frak{gl}_{n})$-module when restricted the $\mathfrak{gl}_{n}$-subalgebra naturally embedded into the top-left corner.'
author:
- Jonathan Nilsson
title: 'New family of simple $\mathfrak{gl}_{2n}({\mathbb{C}})$-modules'
---
Introduction
============
Classification of simple modules is one of the first natural questions which arises when studying the representation theory of some (Lie) algebra. Simple modules are, in some sense, “building blocks” for all other modules, and hence understanding simple modules is important. In some cases, for example for finite dimensional associative algebras, classification of simple modules is an easy problem. However, in most of the cases, the problem of classification of all simple modules is very difficult. Thus, if we consider simple, finite dimensional, complex Lie algebras, then the only algebra for which some kind of classification exists is the Lie algebra $\frak{sl}_{2}$. This was obtained by R. Block in [@Bl], see also a detailed explanation in [@Ma Chapter 6]. However, even in this case the “answer” only reduces the problem to classification of equivalence classes of irreducible elements in a certain non-commutative Euclidean ring.
At the moment, the problem of [*classification*]{} of simple modules over simple Lie algebras seems too hard. However, because of its importance, the problem of [*construction*]{} of new families of modules attracted a lot of attention over the years. The most studied case seem to be the one of the Virasoro Lie algebras, where many different multi-parameter families of simple modules were constructed by various authors, see, for example, [@OW; @GLZ; @LZ; @LLZ; @MZ1; @MZ2; @MW] and references therein.
In contrast to the Virasoro case, the “easier” case of simple, complex, finite dimensional Lie algebras does not yet have an equally large variety of families of simple modules. So, let $\mathfrak{g}$ be a complex, finite dimensional, simple Lie algebra. Some classes of simple $\mathfrak{g}$-modules are, of course, well-understood. For example:
- simple [*finite dimensional*]{} modules are classified already by Cartan in 1913, see [@Ca];
- simple [*highest weight*]{} modules related to a fixed triangular decomposition $\frak{n}_{-} \oplus \frak{h} \oplus \frak{n}_{+}$ of $\mathfrak{g}$ are classified by their highest weights and are extensively studied during last 50 years, see, for example, [@Dix; @Hum2; @BGG];
- simple Whittaker modules in the sense of [@Ko], see also [@AP; @McD1; @McD2];
- simple [*Gelfand-Zeitlin*]{} modules, see [@DFO; @DFO2; @Ma0; @FGR];
- simple weight modules with [*finite dimensional weight spaces*]{} were classified in [@Mathieu] extending the previous work in [@Fe; @Fu];
- simple $\mathfrak{g}$-modules which are free of rank one over the universal enveloping algebra of the Cartan subalgebra were constructed and studied in [@Ni1; @Ni2] (see also [@TZ1; @TZ2] for similar modules over infinite dimensional Lie algebras).
Some further classes of simple modules can be found in [@FOS]. We note that the largest known family of simple $\mathfrak{gl}_{n}$-modules is the one of Gelfand-Zeitlin-modules. It depends on $\frac{n(n+1)}{2}$ generic complex parameters, see [@DFO; @DFO2] for details.
Based on the above, it seems natural to look for new families of simple $\mathfrak{g}$-modules. The present paper contributes with a new large family of simple $\mathfrak{gl}_{2n}$-modules. This family is parameterized by invertible $n\times n$ complex matrices. Let ${\mathcal{A,B,C,D}}$ be the four Lie subalgebras of $\frak{gl}_{2n}$ of dimension $n^2$ as indicated in the following figure: $$\left( \begin{array}{cc}
{\mathcal{A}} & {\mathcal{B}}\\
{\mathcal{C}} & {\mathcal{D}}
\end{array} \right).$$ Then ${\mathcal{B}}$ is nilpotent (and even commutative), and the adjoint action of ${\mathcal{B}}$ on $\frak{gl}_{2n} / {\mathcal{B}}$ is nilpotent, so $({\mathcal{B}},\frak{gl}_{2n})$ is a *Whittaker pair* in the sense of [@BM]. The original motivation for this paper was an attempt to describe generalized Whittaker modules (i.e. modules on which the action of ${\mathcal{B}}$ is locally finite) for this Whittaker pair. Our main result can be summarized as follows:
\[introthm\] For each non-degenerate complex $n \times n$-matrix $Q$, there exists a simple $\mathfrak{gl}_{2n}$ module $M$ with the following properties:
- $M$ has Gelfand-Kirillov dimension $n^{2}$;
- $Res_{{\mathcal{A}}}^{\frak{gl}_{2n}}M$ is isomorphic to the left regular ${\mathcal{U(A)}}$-module;
- $Res_{{\mathcal{B}}}^{\frak{gl}_{2n}}M$ is locally finite. In other words, $M$ is a generalized Whittaker module for the Whittaker pair $({\mathcal{B}},\frak{gl}_{2n})$;
- With respect to a fixed PBW basis in ${\mathcal{U(A)}}$, the action of each fixed element from ${\mathcal{A,B,C,D}}$ can be written explicitly as maps ${\mathcal{U(A)}} \rightarrow {\mathcal{U(A)}}$ of degrees $1,0,2,1$, respectively.
Moreover, different matrices $Q$ give non-isomorphic modules.
The paper is organized as follows. Section 2 introduces notation and lays down some motivation for the construction of our modules. In the same section, for each non-degenerate complex $n \times n$-matrix $Q$, we construct an ${\mathcal{A+B}}$-module having the first three properties listed in Theorem \[introthm\]. We show that there must exist a simple quotient of the corresponding induced $\frak{gl}_{2n}$ module that also has the fourth property. In Section 3 we explicitly construct such a module for $Q$ being the identity matrix $I$ and show that every other module in our family can be obtained by twisting this module by an explicit automorphism. Finally, we give explicit formulas for the $\frak{gl}_{2n}$-action in all cases.
[**Acknowledgements**]{} I am very grateful to Volodymyr Mazorchuk for his ideas and comments.
Motivation and existence
========================
Setup
-----
Let $\mathfrak{g}:=\mathfrak{gl}_{2n}({\mathbb{C}})$. All Lie algebras and vector spaces are over the complex numbers. ${\mathbb{N}}$ denotes the set of nonnegative integers.
First we observe that the subalgebras ${\mathcal{A}}$ and ${\mathcal{D}}$ defined above are both isomorphic to $\frak{gl}_{n}$ while the subalgebras ${\mathcal{B}}$ and ${\mathcal{C}}$ are commutative. Let $e_{i,j}$ be the $2n \times 2n$-matrix with a single $1$ in position $(i,j)$ and zeros elsewhere. By convention, most indices $i,j$ etc. can be assumed to lie between $1$ and $n$; in particular our canonical basis for $\frak{gl}_{2n}$ will be written $$\bigcup_{1 \leq i,j \leq n} \{e_{i,j},e_{n+i,j},e_{i,n+j},e_{n+i,n+j}\}.$$ We denote the identity matrix by $I$, its size ($n$ or $2n$) should be apparent by the context. The transpose of a matrix $A$ is denoted $A^{T}$ and if $A$ is invertible we abbreviate $(A^{-1})^{T}$ by $A^{-T}$.
We also recall how to construct *twisted modules*. For every Lie algebra automorphism $\varphi \in Aut(\frak{g})$ we have a twisting functor $F_{\varphi}: \frak{g}\text{-mod} \rightarrow \frak{g}\text{-mod}$ which is an auto-equivalence. It maps a module $M$ to ${}^{\varphi}M$ which is isomorphic to $M$ as a vector space but has modified action: $x \bullet v := \varphi(x) \cdot v$ for all $x \in \frak{g}$ and $v \in {}^{\varphi}M$.
Existence of simple generalized Whittaker Modules for $\mathfrak{gl}_{2n}$
--------------------------------------------------------------------------
Following Kostant’s idea in [@Ko] we try to construct some modules on which the action of ${\mathcal{B}}$ is locally finite.
Fix Lie algebra homomorphisms $\lambda_{A}: {\mathcal{A}} \rightarrow {\mathbb{C}}$ and $\lambda_{D}: {\mathcal{D}} \rightarrow {\mathbb{C}}$. Let ${\mathbb{C}}_{\lambda_{A}\lambda_{D}}$ be the one dimensional $({\mathcal{A}} + {\mathcal{C}} + {\mathcal{D}})$-module where ${\mathcal{A}}$ acts by $\lambda_{A}$, ${\mathcal{D}}$ acts by $\lambda_{D}$ and ${\mathcal{C}}$ acts trivially. Now define a generalized Verma module $$M_{\lambda_{A}\lambda_{D}} := U(\mathfrak{gl}_{2n}) \bigotimes_{U({\mathcal{A}} + {\mathcal{C}} + {\mathcal{D}})}{\mathbb{C}}_{\lambda_{A}\lambda_{A}}.$$ Denote by $M_{\lambda_{A}\lambda_{D}}^{*}$ the full dual of $M_{\lambda_{A}\lambda_{D}}$. This is a $\mathfrak{gl}_{2n}$ module where the action is given $(x \cdot f)(m) = -f(x\cdot m)$ as usual.
For every $\theta: {\mathcal{B}} \rightarrow {\mathbb{C}}$, there is a unique (up to multiple) eigenvector $w$ in $M_{\lambda_{A}\lambda_{D}}^{*}$ with eigenvalue $\theta$ for ${\mathcal{B}}$.
Note that $M_{\lambda_{A}\lambda_{D}} \simeq U({\mathcal{B}})$ as a left and right $U({\mathcal{B}})$-module. Let ${\mathbb{C}}(\theta)$ be the $1$-dimensional ${\mathcal{B}}$-module where the action is given by $\theta$. By the tensor-hom adjunction we have $$\begin{aligned}
Hom_{U{\mathcal{(B)}}}({\mathbb{C}}(\theta),M_{\lambda_{A}\lambda_{D}}^{*}) &= Hom_{U{\mathcal{(B)}}}({\mathbb{C}}(\theta),Hom_{{\mathbb{C}}}\big( M_{\lambda_{A}\lambda_{D}},{\mathbb{C}}) \big)\\
&\simeq Hom_{U{\mathcal{(B)}}}({\mathbb{C}}(\theta),Hom_{{\mathbb{C}}}\big( U{\mathcal{(B)}},{\mathbb{C}}) \big)\\
& \simeq Hom_{{\mathbb{C}}}(U{\mathcal{(B)}} \otimes_{U{\mathcal{(B)}}}{\mathbb{C}}(\theta),{\mathbb{C}})\\
& \simeq Hom_{{\mathbb{C}}}({\mathbb{C}}(\theta),{\mathbb{C}}) \simeq {\mathbb{C}}.\\\end{aligned}$$ Thus there is a unique $1$-dimensional subspace of $M_{\lambda_{A}\lambda_{D}}^{*}$ isomorphic to ${\mathbb{C}}(\theta)$ in ${\mathcal{B}}$-mod, which is equivalent to the statement in the proposition.
The submodule generated by such an eigenvector must be simple (see [@BM]), so we get the following result.
There exist simple generalized Whittaker modules for the pair $({\mathcal{B}},\frak{gl}_{2n})$ and they can be realized as simple submodules in the dual of the generalized Verma module $M_{\lambda_{A}\lambda_{D}}^{*}$.
The drawback with this approach in our case is that it is difficult to say anything more explicit about the resulting modules as $M_{\lambda_{A}\lambda_{D}}^{*}$ is very big and inconvenient to work in.
An ${\mathcal{A+B}}$-module
---------------------------
### Construction and a formula for the action
We now turn to a more explicit construction. Note that ${\mathcal{B}}$ is commutative. Let $Q=(q_{ij})$ be a nonsingular $n\times n$ matrix and define $L_{Q}$ to be the $1$-dimensional $U({\mathcal{B}})$-module with generator $v$ where the action of ${\mathcal{B}}$ is given by $Q$: $$e_{i,n+j} \cdot v := q_{i,j}v \qquad 1 \leq i,j \leq n.$$
Define an induced module $$M_{Q}:= Ind_{{\mathcal{B}}}^{{\mathcal{A+B}}} L_{Q} = U({\mathcal{A}}+{\mathcal{B}}) \bigotimes_{U({\mathcal{B}})} L_{Q}.$$ Then $M_{Q}$ is clearly isomorphic to $U{\mathcal{(A)}}$ as a left ${\mathcal{A}}$-module, and for $a \in U{\mathcal{(A)}}$ we shall write just $av$ or just $a$ for $a \otimes v$. To explicitly see how ${\mathcal{B}}$ acts on $M_{Q}$, we introduce some more notation. Consider $U{\mathcal{(A)}} \otimes_{{\mathbb{C}}} {\mathcal{A}}$ as a tensor product in the category of unital associative algebras. This becomes an infinite dimensional Lie algebra under the commutator bracket. Note that $U{\mathcal{(A)}} \otimes {\mathcal{A}} \simeq Mat_{n \times n}({\mathcal{U(A)}})$ in a natural way and we shall even extend the trace function to $U{\mathcal{(A)}} \otimes {\mathcal{A}}$ by defining $tr(a \otimes B):=a \:tr(B)$. Note also that ${\mathcal{A}}$ embeds into $U{\mathcal{(A)}} \otimes {\mathcal{A}}$ (as both associative algebra and Lie algebra) by the map $A \mapsto 1 \otimes A$, and we shall sometimes need to identify elements of ${\mathcal{A}}$ with their images under this map. To resolve some ambiguity in our notation, for $A,B \in {\mathcal{A}}$ we shall write $AB$ for the product in ${\mathcal{U(A)}}$ and $A.B$ for the product in the associative algebra ${\mathcal{A}}$ or $U{\mathcal{(A)}} \otimes {\mathcal{A}}$.
Let $\psi': {\mathcal{A}} \rightarrow U{\mathcal{(A)}} \otimes_{{\mathbb{C}}} {\mathcal{A}}$ be the Lie algebra homomorphism defined by $$\psi': A \mapsto A \otimes I -1 \otimes A^{T}.$$ This extends to an algebra homomorphism $\psi: U{\mathcal{(A)}} \rightarrow U{\mathcal{(A)}} \otimes_{{\mathbb{C}}} {\mathcal{A}}$.
The action of ${\mathcal{B}}$ on $M_{Q}$ is given by $$\left( \begin{array}{cc}0 & B\\0 & 0\end{array} \right) av = tr(\psi(a).Q.B^{T})v.$$
This follows by induction on the degree of $a$ as follows. The lemma clearly holds for $a=1$ by the definition of the action of ${\mathcal{B}}$ on $L_{Q}$: we have $tr(Q.B^{T}) = \sum_{ij}q_{ij}b_{ij}$. Suppose the lemma holds for all monomials $a$ of a fixed degree (with respect to any fixed PBW basis). We then have $$\begin{aligned}
\left( \begin{array}{cc}0 & B\\0 & 0\end{array} \right) (Aa)v &= A \left( \begin{array}{cc}0 & B\\0 & 0\end{array} \right) av
+ \big[ \left( \begin{array}{cc}0 & B\\0 & 0\end{array} \right), \left( \begin{array}{cc}A & 0\\0 & 0\end{array} \right) \big]av\\
&=A \left( \begin{array}{cc}0 & B\\0 & 0\end{array} \right) av - \left( \begin{array}{cc}0 & A.B\\0 & 0\end{array} \right)av\\
&=A \: tr(\psi(a).Q.B^{T})v - tr(\psi(a).Q.(A.B)^{T})v\\
&=tr((A \otimes I).\psi(a).Q.B^{T})v - tr(A^{T}.\psi(a).Q.B^{T})v\\
&=tr(((A \otimes I)-1 \otimes A^{T}).\psi(a).Q.B^{T})v\\
&=tr(\psi(A).\psi(a).Q.B^{T})v\\
&=tr(\psi(Aa).Q.B^{T})v.\\\end{aligned}$$ This shows that the lemma holds for all monomials in ${\mathcal{U(A)}}$ by induction. Since $\psi$ is linear it holds for all of ${\mathcal{U(A)}}$.
### Proof of simplicity
We proceed to prove that $M_{Q}$ is simple by first proving it for $Q=I$.
\[rel\] The following relations hold in $U{\mathcal{(A+B)}}$. $$[e_{j,k+n},e_{i,j}^{m}] = \begin{cases}
-m \: e_{i,j}^{m-1} e_{i,k+n} & \text{ for } i \neq j\\
((e_{i,j}-1)^{m} -e_{i,j}^{m})e_{i,k+n} & \text{ for } i = j. \\
\end{cases}$$
This follows easily by induction on $m$.
Fix a PBW basis of $U({\mathcal{A}})$ of form $$\{ e_{11}^{l_{11}}e_{12}^{l_{12}} \cdots e_{1n}^{l_{1n}} e_{21}^{l_{21}} \cdots \cdots e_{n1}^{l_{n1}} \cdots e_{nn}^{l_{nn}} | l_{ij} \in {\mathbb{N}}\},$$
Then $U({\mathcal{A}}) \simeq M_{I}$ has a filtration: $$M_{I}^{(0)} \subset M_{I}^{(1)} \subset M_{I}^{(2)} \subset \cdots$$ where $M_{I}^{(m)}$ is the span of all monomials $f$ with $\deg f := \sum_{ij}l_{ij} \leq m$.
\[mod\] For each $1 \leq j,k \leq n$, the element $(e_{j,k+n} - \delta_{j,k}) \in U({\mathcal{B}})$ has degree $-1$ with respect to the filtration of $M_{I}$. Moreover, the action on an arbitrary monomial in $M_{I}^{(d)}$ is given by $$(e_{j,k+n} - \delta_{j,k}) \cdot e_{11}^{l_{11}} \cdots e_{kj}^{l_{kj}} \cdots e_{nn}^{l_{nn}} = -l_{kj} \: e_{11}^{l_{11}} \cdots e_{kj}^{l_{kj}-1} \cdots e_{nn}^{l_{nn}} \mod M_{I}^{(d-2)}.$$
We have $$(e_{j,k+n} - \delta_{j,k}) \cdot f = f (e_{j,k+n} - \delta_{j,k}) + [e_{j,k+n} - \delta_{j,k},f] = [e_{j,k+n},f],$$ so the fact that $(e_{j,k+n} - \delta_{j,k})$ has degree $ \leq -1$ follows from the previous lemma and the fact that $ad_{e_{j,k+n}}$ is a derivation.
For the second more precise statement, let $f$ be an arbitrary monomial of degree $d$. For each $i$ let $P_{i}$,$Q_{i}$ be the monomial factors of $f$ such that $f= P_{i} e_{ij}^{l_{ij}} Q_{i}$ and $e_{ij} \not| P_{i},Q_{i}$. We now calculate $$\begin{aligned}
(e_{j,k+n} &- \delta_{j,k}) \cdot f = [e_{j,k+n},f] = \sum_{i}P_{i} [e_{j,k+n},e_{ij}^{l_{ij}}] Q_{i}\\
&= P_{j} ((e_{jj}-1)^{l_{jj}}-e_{jj}^{l_{jj}}) e_{j,k+n}\cdot Q_{j} + \sum_{i \neq j}-l_{ij} \: P_{i} e_{ij}^{l_{ij}-1} e_{i,k+n}\cdot Q_{i}\end{aligned}$$
By writing $e_{i,k+n} = (e_{i,k+n}-\delta_{ik})+\delta_{ik}$ and using the fact that the first term has negative degree, we see that $$\begin{aligned}
(e_{j,k+n} - \delta_{j,k}) \cdot f &= \delta_{j,k} P_{j} ((e_{jj}-1)^{l_{jj}}-e_{jj}^{l_{jj}}) Q_{j} + \sum_{i \neq j}- \delta_{ik} l_{ij} \: P_{i} e_{ij}^{l_{ij}-1} Q_{i} \mod M_{I}^{(d-2)}\\
&= -\delta_{j,k}l_{jj} \: P_{j} e_{ij}^{l_{ij}-1} Q_{j} + \sum_{i \neq j}- \delta_{ik} l_{ij} \: P_{i} e_{ij}^{l_{ij}-1} Q_{i} \mod M_{I}^{(d-2)}\\
&= -\sum_{i}\delta_{ik}l_{ij} \: P_{i} e_{ij}^{l_{ij}-1} Q_{i} \mod M_{I}^{(d-2)}\\
&= -l_{kj} \: P_{k} e_{kj}^{l_{kj}-1} Q_{k} \mod M_{I}^{(d-2)}.\end{aligned}$$ The lemma follows.
For each $1 \leq i,j \leq n$, the action of $(e_{i,j}-\delta_{i,j})$ on $M_{I}$ is surjective. Its kernel is spanned by all monomials not divisible by $e_{ij}$.
\[sprop\] The module $M_{I}$ is simple in $U({\mathcal{A+B}})$-mod.
It suffices to show that any $f \in M_{I}$ can be reduced to $1 \in M_{I}^{0}$ via the ${\mathcal{B}}$-action. Fix $f \in M_{I}$ and let $p \in M_{I}^{(d)}$ be a nonzero monomial occurring in $f$ with maximal degree $d$. If $p= \prod_{ij}e_{ij}^{l_{ij}}$ (in the PBW order), it is clear by the previous lemma that $B_{p}:=\prod_{ij}(e_{j,n+i}-\delta_{ij})^{l_{ij}} \in U({\mathcal{B}})$ maps $p$ to a nonzero constant. By the maximality of $d$, $B_{p}$ annihilates all other monomials occurring in $f$ so in fact $B_{p} \cdot f \in M_{I}^{(0)}$ is a nonzero constant as desired.
The module $M_{Q}$ is simple if and only if $Q$ is nonsingular.
For each nonsingular $S \in {\mathcal{A}}$, define $\varphi_{S}: {\mathcal{A+B}} \rightarrow {\mathcal{A+B}}$ by $$\varphi_{S}: \left( \begin{array}{cc} A & B \\ 0 & 0 \end{array}\right) \mapsto \left( \begin{array}{cc} A & B.S^{-1} \\ 0 & 0 \end{array}\right).$$ It is easy to verify that $\varphi_{S}$ is a Lie algebra automorphism and that $\varphi_{S} \circ \varphi_{T} = \varphi_{ST}$. It is also clear that the twisted module ${}^{\varphi_{Q^{-T}}}M_{I}$ is isomorphic to $M_{Q}$. Since $M_{I}$ is simple by Proposition \[sprop\], and since twisting by automorphisms defines an auto-equivalence on $\mathfrak{gl}_{2n}$-Mod, $M_{Q}$ is also simple for nonsingular $Q$.
Conversely, assume that $Q$ is singular and let $A$ be a nonzero matrix such that $Q^{T}A=0$. We shall show that $U{\mathcal{(A)}}Av$ is a proper ${\mathcal{A+B}}$-submodule of $M_{Q}$. The subspace $U{\mathcal{(A)}}Av$ is clearly ${\mathcal{A}}$-stable. For $a \in U{\mathcal{(A)}}$ we compute $$\begin{aligned}
\left( \begin{array}{cc} 0 & B \\ 0 & 0\end{array} \right) \cdot aAv &= tr(\psi(aA).Q.B^{T})v\\
&= tr(\psi(a).\psi(A).Q.B^{T})v = tr(\psi(a).(A \otimes I -1\otimes A^{T}).Q.B^{T})v\\
&=tr(Q.B^{T}.\psi(a).(A \otimes I))v- tr(\psi(a).A^{T}.Q.B^{T})v \\
&= tr(Q.B^{T}.\psi(a))Av-tr(\psi(a).(Q^{T}.A)^{T}.B^{T})v \\
&= tr(Q.B^{T}.\psi(a))Av.\end{aligned}$$ Thus $U{\mathcal{(A)}}Av$ is also ${\mathcal{B}}$-stable, and is thus a proper submodule of $M_{Q}$.
### Injectivity and an existence theorem
Our next goal is to prove that for most $Q$’s, the module $M_{Q}$ is injective when restricted to ${\mathcal{U(B)}}$. We begin by recalling a result about injective envelopes for the trivial module over polynomial rings. For a proof, see for example [@L $\S 3$J].
Let $k$ be a field, let $R=k[x_{1}, \ldots x_{n}]$ and let $L$ be the trivial $R$-module. Let $E$ be the $R$-module $k[x_{1}^{-1}, \ldots x_{n}^{-1}]$ where $x_{i}$ acts by $$x_{i} \cdot (x_{1}^{-k_{1}} \cdots x_{n}^{-k_{n}}) = \begin{cases}
x_{1}^{-k_{1}} \cdots x_{i}^{-k_{i}+1} \cdots x_{n}^{-k_{n}} & \text{if } k_{i} > 0 \\
0 & \text{otherwise.}
\end{cases}$$ Then $E=E(L)$ is the injective envelope of $L$.
By twisting $E$ by automorphisms we obtain injective envelopes of all $1$-dimensional $R$-modules as follows:
With notation as in the previous lemma, for scalars $q_{i} \in k$, let $L_{q_{1}, \ldots q_{n}}$ be the $1$-dimensional $R$-module with action $x_{i} \cdot v = q_{i} v$. Then $E(L_{q_{1}, \ldots q_{n}}) \simeq {}^\varphi E(L)$ where $\varphi$ is the $R$-automorphism mapping $x_{i} \mapsto x_{i}-q_{i}$.
We have $L_{q_{1}, \ldots q_{n}} \simeq {}^\varphi L$ and since twisting by an automorphism is an auto-equivalence on $R$-mod, the corollary follows.
For nonsingular matrices $Q$, the module $Res_{U({\mathcal{B}})}^{U({\mathcal{A+B}})} \; M_{Q}$ is injective.
Let $I(L_{Q})$ be the injective envelope of $L_{Q}$. Applying the exact functor $Hom_{{\mathcal{B}}}( -,I(L_{Q}))$ to the exact sequence $$0 {\rightarrow}L_{Q} {\rightarrow}M_{Q} {\rightarrow}Coker {\rightarrow}0$$ we obtain the exact sequence $$0 {\rightarrow}Hom_{{\mathcal{B}}}( Coker,I(L_{Q})) {\rightarrow}Hom_{{\mathcal{B}}}( M_{Q},I(L_{Q})) {\rightarrow}Hom_{{\mathcal{B}}}(L_{Q},I(L_{Q})) {\rightarrow}0.$$ Hence the morphism $L_{Q} {\rightarrow}I(L_{Q})$ mapping $L_{Q}$ into its injective envelope is the image of some morphism $f:M_{Q} {\rightarrow}I(L_{Q})$. Since $f$ is nonzero on $span(v)=soc(M_{Q})$, $f$ is injective. Moreover, for all $k \in {\mathbb{N}}$ we have $$\dim soc_{k}(M_{Q}) = \binom{n^{2}+k-2}{k-1}= \dim soc_{k}(I(L_{Q})),$$ which shows that $f$ is surjective. This shows that $f$ is an isomorphism and in particular that $M_{Q}$ is the injective envelope of $L_{Q}$.
Indecomposable injectives over noetherian rings R correspond to Spec(R) via $\frak{p} \mapsto $injective envelope of $(R/ \frak{p} )$. Moreover $L_{Q} = U({\mathcal{B}}) / \mathfrak{m}$ where $\frak{m}$ is the maximal ideal generated by $(e_{i,n+j}-q_{i,j})$, so if $M_{Q}$ is injective, it must be the injective envelope of $U({\mathcal{B}}) / \mathfrak{m}$.
For each nonsingular matrix $n \times n$-matrix $Q$ there exists a $\mathfrak{gl}_{2n}$-module $M$ such that
- $M$ is generated by a single ${\mathcal{B}}$-eigenvector with eigenvalues corresponding to the entries of $Q$.
- $Res_{U({\mathcal{B}})}^{U({\mathfrak{gl}_{2n}})} M \simeq U({\mathcal{A}}) \simeq U({\mathfrak{gl}_{n}})$.
As we’ve seen before, we take $L_{Q}$ as the $1$-dimensional ${\mathcal{B}}$-module corresponding to $Q$ and we let $M_{Q} = U({\mathcal{A}}+{\mathcal{B}}) \bigotimes_{U({\mathcal{B}})} L_{Q}$. Then $M_{Q}$ is injective in ${\mathcal{B}}$-mod. Next we define $$W:= U({\mathcal{A}}+{\mathcal{B}}+{\mathcal{D}}) \bigotimes_{U({\mathcal{A}}+{\mathcal{B}})} M_{Q}.$$ Fixing $d \in {\mathcal{D}}$ we note that $span(v,d\cdot v)$ is a two-dimensional ${\mathcal{B}}$-submodule of $W$, and moreover it is a non-split self-extension of $L_{Q}$ with itself. Now by the injectivity of $M_{Q}$ there exists a morphism $\varphi$ such that the following diagram commutes in ${\mathcal{B}}$-mod: $$\xymatrix@C=2cm@R=2cm{ & span(v,d \cdot v) \ar@{.>}[dl]^{\varphi} \\ M_{Q} & L_{Q} \ar@{^{(}->}[u] \ar@{^{(}->}[l]}$$
Thus there exists $a_{d} \cdot v \in soc_{2}(M_{Q})={\mathcal{A}} \cdot v$ such that $a_{d}\cdot v-d\cdot v$ spans a $1$-dimensional $B$-submodule $S_{d}$ of $W$. The module $W' := W / \sum_{d \in {\mathcal{D}}} U({\mathcal{A+B+D}})S_{d}$ is then isomorphic to $M_{Q}$ when restricted to $U({\mathcal{A+B}})$.
Next, let $W'' := U({\mathcal{A+B+C+D}}) \bigotimes_{U({\mathcal{A+B+D}})} W'$. For a fixed $c \in {\mathcal{C}}$ we have a ${\mathcal{B}}$-submodule ${\mathcal{B}}^{2} (c\cdot v)$ with simple top and simple socle, both isomorphic to $L_{Q}$. By similar arguments, there exists $x \in soc_{3}(M_{Q}) = {\mathcal{A}}^{2} \cdot v$ such that $x-c \cdot v$ spans a ${\mathcal{B}}$-submodule of $W''$. Forming the quotient of all these subs we get the module required by the theorem.
In the next section we shall give explicit formulas for the elements $a_{d}$ and $x$ of the proof above in order to write down the action on the simple $\frak{gl}_{2n}$-modules explicitly.
Explicit formulas for the $\mathfrak{gl}_{2n}$-modules
======================================================
Preliminaries
-------------
The following formula will be particularly useful for $m=2$.
\[glemma\] Let $F:=(e_{j,i})_{i,j}=\sum_{i,j}e_{j,i} \otimes e_{i,j} \in {\mathcal{U(A)}} \otimes {\mathcal{A}}$. For any $A,B \in \mathfrak{gl}_{n}$ and for all $m \in {\mathbb{N}}$ we have $$[A,tr(B.F^{m})] = tr ([A,B].F^{m})$$ in $U(\mathfrak{gl}_{n})$
We proceed by induction on $m$. Since $tr(X.F)=X$ the equality clearly holds for $m=1$. The equation above is linear in both $A$ and $B$ so it suffices to verify it for $A=e_{ij}$, $B=e_{kl}$. Note that we explicitly have $$tr(e_{ij}.F^{m+1}) = \sum_{1 \leq r_{1}, \ldots, r_{m} \leq n} e_{ir_{1}}e_{r_{1}r_{2}} \cdots e_{r_{m}j}.$$ Assume that the equality holds for some fixed $m$. We now compute
$$\begin{aligned}
[e_{ij},&tr(e_{kl}.F^{m+1})] = [e_{ij},\sum_{r_{1}, \ldots, r_{m}} e_{kr_{1}}e_{r_{1}r_{2}} \cdots e_{r_{m}l}]\\
&=\sum_{r_{1}, \ldots, r_{m}}([e_{ij}, e_{kr_{1}}]e_{r_{1}r_{2}} \cdots e_{r_{m}l} + e_{kr_{1}}[e_{ij}, e_{r_{1}r_{2}} \cdots e_{r_{m}l}])\\
&=\sum_{r_{1}, \ldots, r_{m}}(\delta_{jk}e_{ir_{1}} - \delta_{r_{1}i}e_{kj})e_{r_{1}r_{2}} \cdots e_{r_{m}l} + \sum_{r_{1}}e_{kr_{1}}[e_{ij}, \sum_{r_{2}, \ldots r_{m}}e_{r_{1}r_{2}} \cdots e_{r_{m}l}]\\
&=\delta_{jk}tr(e_{il}.F^{m+1}) - e_{kj}\sum_{r_{2}, \ldots, r_{m}}e_{ir_{2}} \cdots e_{r_{m}l} + \sum_{r_{1}}e_{kr_{1}}[e_{ij},tr(e_{r_{1}l}.F^{m})]\\
&=\delta_{jk}tr(e_{il}.F^{m+1}) - e_{kj}tr(e_{il}.F^{m}) + \sum_{r_{1}}e_{kr_{1}}tr([e_{ij},e_{r_{1}l}].F^{m})\\
&=\delta_{jk}tr(e_{il}.F^{m+1}) - e_{kj}tr(e_{il}.F^{m}) + \sum_{r_{1}}e_{kr_{1}}(\delta_{jr_{1}}tr(e_{il}.F^{m})- \delta_{il}tr(e_{r_{1}j}.F^{m}))\\
&=\delta_{jk}tr(e_{il}.F^{m+1}) - e_{kj}tr(e_{il}.F^{m}) + e_{kj}tr(e_{il}.F^{m}) -\delta_{il}\sum_{r_{1}}e_{kr_{1}}tr(e_{r_{1}j}.F^{m}))\\
&=\delta_{jk}tr(e_{il}.F^{m+1}) - \delta_{il}tr(e_{kj}.F^{m+1})\\
&=tr([e_{ij},e_{kl}].F^{m+1}).\end{aligned}$$
By induction the lemma holds.
Fixing $B$ as the identity matrix above we obtain $[A,tr(F^{k})] = 0$ for all $A$ in $\mathfrak{gl}_{n}$ which shows that $tr(F^{k})$ is central in $U(\mathfrak{gl}_{n})$. In fact, $Z(\mathfrak{gl}_{n}) = {\mathbb{C}}[tr(F),tr(F^{2}), \ldots , tr(F^{n})]$. The elements $tr(F^{k})$ are called Gelfand invariants.
The main result
---------------
We are now ready to state our main result. Define $\varphi': {\mathcal{A}} \rightarrow {\mathcal{U(A)}} \otimes {\mathcal{A}}$ by $$\varphi': A \mapsto A \otimes I +1 \otimes A.$$ This is a Lie algebra homomorphism and it extends to an algebra homomorphism $\varphi: {\mathcal{U(A)}} \rightarrow {\mathcal{U(A)}} \otimes {\mathcal{A}}$. Also recall that we previously have defined $\psi: {\mathcal{U(A)}} \rightarrow {\mathcal{U(A)}} \otimes {\mathcal{A}}$ which satisfied $\psi: A \mapsto A \otimes I - 1 \otimes A^{T}$ for $A \in {\mathcal{A}}$. Using these two homomorphisms we now state our main theorem.
\[mainthm\] Define an action of $\mathfrak{gl}_{2n}$ on $M_{I} \simeq U{\mathcal{(A)}}$ as follows: for any $a \in U{\mathcal{(A)}}$, let $$\left( \begin{array}{cc} A & B \\ C & D \end{array} \right) \cdot a = Aa - aD +tr(\psi(a).B^{T}) -tr(\varphi(a).F^{2}.C) -tr(\varphi(a).C)tr(F). \tag{1}$$ This is a $\mathfrak{gl}_{2n}$-module structure.
First, for all $X,Y \in \frak{gl}_{2n}$, $A \in {\mathcal{A}}$ and $a \in {\mathcal{U(A)}}$ we have
$$\begin{aligned}
X &\cdot Y \cdot Aa - Y \cdot X \cdot Aa=\\
&=A (X \cdot Y \cdot a) + [XY,A]a - A (Y \cdot X \cdot a) - [YX,A]a\\
&=A (X \cdot Y \cdot a - Y \cdot X \cdot a) + X \cdot [Y,A]a - [X,A]\cdot Ya\\
& \qquad \qquad - Y \cdot [X,A]a - [Y,A] \cdot Xa\\
&=A \cdot [X,Y] a + [X , [Y,A]]a + [Y,[A,X]]a\\
&= A \cdot [X,Y] a - [A , [X,Y]]a\\
&=[X,Y] \cdot Aa.\end{aligned}$$
This shows that it suffices to check that $$X \cdot Y \cdot 1 - Y \cdot X \cdot 1 = [X,Y]\cdot 1$$ for all $X,Y \in \frak{gl}_{2n}$ in order to prove that the formula in the theorem gives a module structure.
We first consider the case $Y:=A_{0} \in {\mathcal{A}}$. We compute $$\begin{aligned}
&\left( \begin{array}{cc} A & B \\ C & D \end{array} \right) \cdot \left( \begin{array}{cc} A_{0} & 0 \\ 0 & 0 \end{array} \right) \cdot 1
- \left( \begin{array}{cc} A_{0} & 0 \\ 0 & 0 \end{array} \right) \cdot \left( \begin{array}{cc} A & B \\ C & D \end{array} \right) \cdot 1\\
&= AA_{0}- A_{0}D+tr((A_{0} \otimes I - 1 \otimes A_{0}^{T}).B^{T})-tr((A_{0}\otimes I+1\otimes A_{0}).F^{2}.C)\\
& \qquad -tr((A_{0}\otimes I+1\otimes A_{0}).C)tr(F)\\
& \qquad - \big( A_{0}A-A_{0}D+A_{0}tr(B^{T})-A_{0}tr(F^{2}.C)-A_{0}tr(C)tr(F) \big)\\
&=AA_{0}- A_{0}D+A_{0}tr(B^{T})+tr(A_{0}^{T}.B^{T})-A_{0}tr(F^{2}.C) -tr(A_{0}.F^{2}.C)\\
& \qquad -A_{0}tr(C)tr(F)-tr(A_{0}.C)tr(F)\\
& \qquad - A_{0}A+A_{0}D-A_{0}tr(B^{T})+A_{0}tr(F^{2}.C)+A_{0}tr(C)tr(F)\\
&=[A,A_{0}]+tr(A_{0}^{T}.B^{T}) -tr(A_{0}.F^{2}.C)-tr(A_{0}.C)tr(F)\\
&=[A,A_{0}]+tr((A_{0}.B)^{T}) -tr(F^{2}.C.A_{0})-tr(C.A_{0})tr(F)\\
&=\left( \begin{array}{cc} [A,A_{0}] & A_{0}.B \\ C.A_{0} & 0 \end{array} \right) \cdot 1\\
&=\Big[ \left( \begin{array}{cc} A & B \\ C & D \end{array} \right) , \left( \begin{array}{cc} A_{0} & 0 \\ 0 & 0 \end{array} \right) \Big] \cdot 1.\end{aligned}$$
It remains to check that $X \cdot Y \cdot 1 - Y \cdot X \cdot 1 = [X,Y] \cdot v$ for $X,Y \in {\mathcal{B,C,D}}$. Moreover, since the right side of $(1)$ is linear in $A,B,C,$ and $D$ it suffices to check $(1)$ it for the standard basis elements of $\mathfrak{gl}_{2n}$.\
When $X,Y \in {\mathcal{B}}$ the calculation is easy: $$\begin{aligned}
\left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) &\cdot \left( \begin{array}{cc} 0 & B' \\ 0 & 0 \end{array}\right)\cdot 1
- \left( \begin{array}{cc} 0 & B' \\ 0 & 0 \end{array}\right) \cdot \left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) \cdot 1\\
&= \left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) \cdot tr(B')
- \left( \begin{array}{cc} 0 & B' \\ 0 & 0 \end{array}\right) \cdot tr(B)\\
&=tr(B)tr(B') - tr(B')tr(B) = 0 = \Big[\left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) ,\left( \begin{array}{cc} 0 & B' \\ 0 & 0 \end{array}\right) \Big] \cdot 1.\end{aligned}$$
Similarly, for $X,Y \in {\mathcal{D}}$ we have $$\begin{aligned}
\left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) &\cdot \left( \begin{array}{cc} 0 & 0 \\ 0 & D' \end{array}\right) \cdot 1
- \left( \begin{array}{cc} 0 & 0 \\ 0 & D' \end{array}\right) \cdot \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot 1\\
&= -\left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot D'
+ \left( \begin{array}{cc} 0 & 0 \\ 0 & D' \end{array}\right) \cdot D\\
&= D'D - DD' = [D',D] = \left( \begin{array}{cc} 0 & 0 \\ 0 & [D,D'] \end{array}\right) \cdot 1 = \Big[\left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) ,\left( \begin{array}{cc} 0 & 0 \\ 0 & D' \end{array}\right) \Big] \cdot 1. \end{aligned}$$
For $X \in {\mathcal{B}}, Y \in {\mathcal{D}}$ we get $$\begin{aligned}
\left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) &\cdot \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot 1
- \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot \left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right)\cdot 1\\
&= -\left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) \cdot D - tr(B^{T})\left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot 1\\
&= -tr((D \otimes I - 1 \otimes D^{T}).B^{T}) +tr(B^{T})D \\
&= -Dtr(B^{T})+tr(D^{T}.B^{T}) +tr(B^{T})D = tr((D.B)^{T})\\
&= \left( \begin{array}{cc} 0 & D.B \\ 0 & 0 \end{array}\right) \cdot 1 = \Big[\left( \begin{array}{cc} 0 & B \\ 0 & 0 \end{array}\right) , \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \Big] \cdot 1.\end{aligned}$$
For $X \in {\mathcal{C}}, Y \in {\mathcal{D}}$ we apply Lemma \[glemma\] for $m=1,2$ to obtain $$\begin{aligned}
\left( \begin{array}{cc} 0 & 0 \\ C & 0 \end{array}\right) &\cdot \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot 1
- \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot \left( \begin{array}{cc} 0 & 0 \\ C & 0 \end{array}\right)v \cdot 1 \\
&= -\left( \begin{array}{cc} 0 & 0 \\ C & 0 \end{array}\right) \cdot D + \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \cdot (tr(C.F^{2})+tr(C)tr(F))\\
&= tr((D \otimes I + 1 \otimes D).F^{2}.C) + tr((D \otimes I + 1 \otimes D).C)tr(F)\\
&\qquad- (tr(C.F^{2})+tr(C)tr(F))D\\
&=D \: tr(F^{2}.C) + tr(D.F^{2}.C) + D \: tr(C)tr(F) + tr(D.C)tr(F)\\
&\qquad- (tr(C.F^{2})+tr(C)tr(F))D\\
&=[D,tr(C.F^{2})] + tr(C)[D,tr(F)] + tr(D.F^{2}.C) + tr(D.C)tr(F)\\
&=tr([D,C].F^{2})] + tr(C.D.F^{2}) + tr(D.C)tr(F)\\
&=tr(F^{2}.D.C)] + tr(D.C)tr(F)\\
&=\left( \begin{array}{cc} 0 & 0 \\ -D.C & 0 \end{array}\right) \cdot 1\\
&= \Big[ \left( \begin{array}{cc} 0 & 0 \\ C & 0 \end{array}\right) , \left( \begin{array}{cc} 0 & 0 \\ 0 & D \end{array}\right) \Big] \cdot 1.\end{aligned}$$
Next, for $X \in {\mathcal{B}}, Y \in {\mathcal{C}}$, take $X=e_{i,n+j}$ and $Y=e_{n+k,l}$. We then have $$\begin{aligned}
&e_{i,n+j} \cdot e_{n+k,l}\cdot 1 - e_{n+k,l}\cdot e_{i,n+j} \cdot 1\\
&= -e_{i,n+j} \cdot (tr(e_{kl}.F^{2})+tr(e_{kl})tr(F)) - e_{n+k,l}\cdot tr(e_{ij}^{T})\\
&= -e_{i,n+j} \cdot ((\sum_{r=1}^{n}e_{kr}e_{rl}) + \delta_{kl}tr(F)) +\delta_{ij}(tr(e_{kl}.F^{2})+tr(e_{kl})tr(F))\\
&= -\sum_{r=1}^{n} \big( tr(\psi(e_{kr}e_{rl}).e_{ji}) \big) - \delta_{kl}tr(\psi(tr(F)).e_{ji}) +\delta_{ij}(tr(e_{kl}.F^{2})+tr(e_{kl})tr(F)) \\
&= -\sum_{r=1}^{n} tr\big((e_{kr}\otimes I - 1 \otimes e_{rk}).(e_{rl} \otimes I - 1 \otimes e_{lr}).e_{ji} \big) - \delta_{kl}tr((tr(F) \otimes I - 1 \otimes tr(F)).e_{ji})\\
& \qquad +\delta_{ij}(tr(e_{kl}.F^{2})+tr(e_{kl})tr(F)) \\
&= \sum_{r=1}^{n} \Big( -tr(e_{ji}.e_{rk}.e_{lr}) + e_{kr}tr(e_{ji}.e_{lr}) + e_{rl}tr(e_{ji}.e_{rk}) - e_{kr}e_{rl}tr(e_{ji})\Big)\\
&\qquad + \delta_{kl}\big( tr(e_{ji}tr(F)) - tr(F)tr(e_{ji}) \big) +\delta_{ij}(tr(e_{kl}F^{2})+tr(e_{kl})tr(F))\\
&= \big(-\delta_{kl}tr(e_{ji}tr(F))+ \delta_{li}e_{kj} + \delta_{jk}e_{il} -\delta_{ji}tr(e_{kl}F^{2})\big)\\
& \qquad +\delta_{kl}\delta_{ji} - \delta_{kl}\delta_{ji}tr(F) +\delta_{ij}tr(e_{kl}F^{2})+\delta_{ij}\delta_{kl}tr(F)\\
&= -\delta_{kl}\delta_{ji}+ \delta_{li}e_{kj} + \delta_{jk}e_{il} +\delta_{kl}\delta_{ji}\\
&= \delta_{li}e_{kj} + \delta_{jk}e_{il} = \delta_{jk}e_{il} -\delta_{li}e_{n+k,n+j} = [e_{i,n+j},e_{n+k,l}] \cdot 1\\\end{aligned}$$
It remains only to show that $(1)$ holds for $X,Y \in {\mathcal{C}}$. Let $X=e_{n+i,j}$ and $Y=e_{n+k,l}$. In this case we have $$\begin{aligned}
e_{n+i,j} &\cdot e_{n+k,l}\cdot 1 - e_{n+k,l}\cdot e_{n+i,j} \cdot 1\\
=& -e_{n+i,j} \cdot \big( tr(e_{kl}.F^{2}) +tr(e_{kl})tr(F)\big) + e_{n+k,l}\cdot \big( tr(e_{ij}.F^{2}) +tr(e_{ij})tr(F)\big)\\
=& -e_{n+i,j} \cdot \big( \sum_{r=1}^{n}e_{kr}e_{rl} +\delta_{kl}tr(F)\big) + e_{n+k,l}\cdot \big( \sum_{r=1}^{n}e_{ir}e_{rj} +\delta_{ij}tr(F)\big)\\
=& \Bigg(\sum_{r=1}^{n} \Big( tr(e_{ij}.e_{kr}.e_{rl}.F^{2}) + e_{kr}tr(e_{ij}.e_{rl}.F^{2}) + e_{rl}tr(e_{ij}.e_{kr}.F^{2}) + e_{kr}e_{rl}tr(e_{ij}.F^{2})\\
&+ \big(tr(e_{ij}.e_{kr}.e_{rl}) + e_{kr}tr(e_{ij}.e_{rl}) + e_{rl}tr(e_{ij}.e_{kr}) + e_{kr}e_{rl}tr(e_{ij})\big) tr(F)\Big)\\
&+ \delta_{kl}\big( tr(e_{ij}.tr(F).F^{2}) + tr(F)tr(e_{ij}.F^{2}) + tr(e_{ij}.tr(F))tr(F) + tr(F)tr(e_{ij})tr(F) \big) \Bigg)\\
&- \Bigg(\sum_{r=1}^{n} \Big( tr(e_{kl}.e_{ir}.e_{rj}.F^{2}) + e_{ir}tr(e_{kl}.e_{rj}.F^{2}) + e_{rj}tr(e_{kl}.e_{ir}.F^{2}) + e_{ir}e_{rj}tr(e_{kl}.F^{2})\\
&+ \big(tr(e_{kl}.e_{ir}.e_{rj}) + e_{ir}tr(e_{kl}.e_{rj}) + e_{rj}tr(e_{kl}.e_{ir}) + e_{ir}e_{rj}tr(e_{kl})\big) tr(F)\Big)\\
&+ \delta_{ij}\big( tr(e_{kl}.tr(F).F^{2}) + tr(F)tr(e_{kl}.F^{2}) + tr(e_{kl}.tr(F))tr(F) + tr(F)tr(e_{kl})tr(F) \big) \Bigg)\\
=& n \: tr(e_{ij}.e_{kl}.F^{2}) + e_{kj}tr(e_{il}.F^{2}) +\sum_{r}e_{rl}tr(e_{ij}.e_{kr}.F^{2}) + tr(e_{kl}.F^{2})tr(e_{ij}.F^{2})\\
&+ \big( n \:tr(e_{ij}.e_{kl}) + e_{kj}tr(e_{il}) + \sum_{r}e_{rl}tr(e_{ij}.e_{kr}) + \delta_{ij}tr(e_{kl}.F^{2})\big) tr(F)\\
&+ \delta_{kl}\big( tr(e_{ij}.F^{2}) + tr(F)tr(e_{ij}.F^{2}) + \delta_{ij}tr(F) + \delta_{ij}tr(F)tr(F) \big) \\
&- n \: tr(e_{kl}.e_{ij}.F^{2}) - e_{il}tr(e_{kj}.F^{2}) -\sum_{r}e_{rj}tr(e_{kl}.e_{ir}.F^{2}) - tr(e_{ij}.F^{2})tr(e_{kl}.F^{2})\\
&+ \big( -n \:tr(e_{kl}.e_{ij}) - e_{il}tr(e_{kj}) - \sum_{r}e_{rj}tr(e_{kl}.e_{ir}) - \delta_{kl}tr(e_{ij}.F^{2})\big) tr(F)\\
&+ \delta_{ij}\big( -tr(e_{kl}.F^{2}) - tr(F)tr(e_{kl}.F^{2}) - \delta_{kl}tr(F) - \delta_{kl}tr(F)tr(F) \big) \\
=& n \delta_{jk} tr(e_{il}.F^{2}) + e_{kj}tr(e_{il}.F^{2}) +\delta_{jk}\sum_{r}e_{rl}tr(e_{ir}.F^{2}) + tr(e_{kl}.F^{2})tr(e_{ij}.F^{2})\\
&+ n\delta_{jk}\delta_{il}tr(F) + e_{kj}\delta_{il}tr(F) + \delta_{jk}e_{il}tr(F) + \delta_{ij}tr(e_{kl}.F^{2})tr(F)\\
&+ \delta_{kl}tr(e_{ij}.F^{2}) + \delta_{kl}tr(F)tr(e_{ij}.F^{2}) + \delta_{kl}\delta_{ij}tr(F) + \delta_{kl}\delta_{ij}tr(F)^{2}\\
&- n \delta_{li} tr(e_{kj}.F^{2}) - e_{il}tr(e_{kj}.F^{2}) -\delta_{li}\sum_{r}e_{rj}tr(e_{kr}.F^{2}) - tr(e_{ij}.F^{2})tr(e_{kl}.F^{2})\\
&- n\delta_{li}\delta_{kj}tr(F) - e_{il}\delta_{kj}tr(F) - \delta_{li}e_{kj}tr(F) - \delta_{kl}tr(e_{ij}.F^{2})tr(F)\\
&- \delta_{ij}tr(e_{kl}.F^{2}) - \delta_{ij}tr(F)tr(e_{kl}.F^{2}) - \delta_{ij}\delta_{kl}tr(F) - \delta_{ij}\delta_{kl}tr(F)^{2}\\
=& \delta_{jk}\sum_{r}e_{rl}tr(e_{ir}.F^{2}) -\delta_{li}\sum_{r}e_{rj}tr(e_{kr}.F^{2}) + e_{kj}tr(e_{il}.F^{2})\\
&+ n \delta_{jk} tr(e_{il}.F^{2}) + \delta_{kl}tr(e_{ij}.F^{2}) - \delta_{ij}tr(e_{kl}.F^{2}) - n \delta_{li} tr(e_{kj}.F^{2}) - e_{il}tr(e_{kj}.F^{2}) \\
&+[tr(e_{kl}.F^{2}),tr(e_{ij}.F^{2})] \\\end{aligned}$$ We proceed to compute $[tr(e_{kl}.F^{2}),tr(e_{ij}.F^{2})]$ separately.
$$\begin{aligned}
[tr(e_{kl}.F^{2})&,tr(e_{ij}.F^{2})] = \sum_{r}[e_{kr}e_{rl},tr(e_{ij}.F^{2})]\\
&= \sum_{r} \Big( e_{kr}[e_{rl},tr(e_{ij}.F^{2})] + [e_{kr},tr(e_{ij}.F^{2})]e_{rl} \Big)\\
&= \sum_{r}\Big( e_{kr}tr([e_{rl},e_{ij}].F^{2}) + tr([e_{kr},e_{ij}].F^{2})e_{rl} \Big)\\
&= \sum_{r}\Big( \delta_{li}e_{kr}tr(e_{rj}.F^{2}) - \delta_{jr}e_{kr}tr(e_{il}.F^{2}) +\delta_{ri}tr(e_{kj}.F^{2})e_{rl} - \delta_{kj}tr(e_{ir}.F^{2})e_{rl}\Big)\\
&= \delta_{li} tr(e_{kj}.F^{3})- e_{kj}tr(e_{il}.F^{2}) + tr(e_{kj}.F^{2})e_{il} - \delta_{kj}tr(e_{il}.F^{3})\end{aligned}$$
Inserting this into the previous expression gives $$\begin{aligned}
=& \delta_{jk}\sum_{r}e_{rl}tr(e_{ir}.F^{2}) -\delta_{li}\sum_{r}e_{rj}tr(e_{kr}.F^{2}) + e_{kj}tr(e_{il}.F^{2})\\
&+ n \delta_{jk} tr(e_{il}.F^{2}) + \delta_{kl}tr(e_{ij}.F^{2}) - \delta_{ij}tr(e_{kl}.F^{2}) - n \delta_{li} tr(e_{kj}.F^{2}) - e_{il}tr(e_{kj}.F^{2}) \\
&+ \delta_{li} tr(e_{kj}.F^{3})- e_{kj}tr(e_{il}.F^{2}) + tr(e_{kj}.F^{2})e_{il} - \delta_{kj}tr(e_{il}.F^{3})\\
&= \delta_{jk}\sum_{r}\big(tr(e_{ir}.F^{2})e_{rl} + [e_{rl},tr(e_{ir}.F^{2})]\big) -\delta_{li}\sum_{r} \big(tr(e_{kr}.F^{2})e_{rj} + [e_{rj},tr(e_{kr}.F^{2})] \big)\\
&+ n \delta_{jk} tr(e_{il}.F^{2}) + \delta_{kl}tr(e_{ij}.F^{2}) - \delta_{ij}tr(e_{kl}.F^{2}) - n \delta_{li} tr(e_{kj}.F^{2}) \\
&+ \delta_{li} tr(e_{kj}.F^{3}) + [tr(e_{kj}.F^{2}),e_{il}] - \delta_{kj}tr(e_{il}.F^{3})\\
&= \delta_{jk}tr(e_{il}.F^{3}) + \delta_{jk}\sum_{r}tr([e_{rl},e_{ir}].F^{2}) -\delta_{li}tr(e_{kj}.F^{3}) -\delta_{li} \sum_{r}tr([e_{rj},e_{kr}].F^{2})\\
&+ n \delta_{jk} tr(e_{il}.F^{2}) + \delta_{kl}tr(e_{ij}.F^{2}) - \delta_{ij}tr(e_{kl}.F^{2}) - n \delta_{li} tr(e_{kj}.F^{2}) \\
&+ \delta_{li} tr(e_{kj}.F^{3}) + tr([e_{kj},e_{il}].F^{2}) - \delta_{kj}tr(e_{il}.F^{3})\\
&= \delta_{jk}\big( \delta_{li}tr(tr(F).F^{2}) - n\: tr(e_{il}.F^{2}) \big) -\delta_{li} \big( \delta_{jk}tr(tr(F).F^{2})-n \: tr(e_{kj}.F^{2}) \big)\\
&+ n \delta_{jk} tr(e_{il}.F^{2}) + \delta_{kl}tr(e_{ij}.F^{2}) - \delta_{ij}tr(e_{kl}.F^{2}) - n \delta_{li} tr(e_{kj}.F^{2}) \\
&+ \delta_{ij}tr(e_{kl}.F^{2}) - \delta_{lk}tr(e_{ij}.F^{2})\\
&= \delta_{jk}\delta_{li}tr(F^{2}) -\delta_{li}\delta_{jk}tr(F^{2})=0 = [e_{n+i,j},e_{n+k,l}] \cdot 1\\\end{aligned}$$ This completes the proof.
\[corthm\] Define an action of $\mathfrak{gl}_{2n}$ on $M_{Q} \simeq U{\mathcal{(A)}}$ as follows: for any $a \in U{\mathcal{(A)}}$, let $$\left( \begin{array}{cc} A & B \\ C & D \end{array} \right) \cdot a = Aa - aD +tr(\psi(a).Q.B^{T}) -tr(\varphi(a).F^{2}.Q^{-T}.C) -tr(\varphi(a).Q^{-T}.C)tr(F).$$ This is a $\mathfrak{gl}_{2n}$-module structure.
For each nonsingular $S \in Mat_{n \times n}$, define $\varphi_{S}: \mathfrak{gl}_{2n} \rightarrow \mathfrak{gl}_{2n} $ by $$\varphi_{S}: \left( \begin{array}{cc} A & B \\ C & D \end{array}\right) \mapsto \left( \begin{array}{cc} A & B.S^{-1} \\ S.C & S.D.S^{-1} \end{array}\right).$$ It is easy to verify that $\varphi_{S}$ is a Lie algebra automorphism and that $\varphi_{S} \circ \varphi_{T} = \varphi_{S.T}$, so the map $\Xi: Mat_{n \times n}({\mathbb{C}})^{*} \rightarrow Aut(\mathfrak{gl}_{2n})$ with $S \mapsto \varphi_{S}$ is an injective algebra homomorphism. Let $V$ be the $\frak{gl}_{2n}$ module from in Theorem \[mainthm\]. Now by the action of $\mathfrak{gl}_{2n}$ on the twisted module $V_{Q}:={}^{\varphi_{Q^{-T}}}V$ is precisely as in the statement of this theorem.
The modules $V_{Q}$ now satisfy the conditions of Theorem \[introthm\] in the introduction:
(of Theorem \[introthm\]) The module $V_{Q}$ is simple since $Res_{{\mathcal{A+B}}}^{\frak{gl}_{2n}}V_{Q} \simeq M_{Q}$ is. That the GK-dimension is $n^{2}$ and that $Res_{{\mathcal{A}}}^{\frak{gl}_{2n}} V_{Q} \simeq {\mathcal{U(A)}}$ follows directly from the definition in Theorem \[corthm\]. Since the linear maps $tr(\psi(-).B^{T}): {\mathcal{U(A)}} \rightarrow {\mathcal{U(A)}}$ never increases the degree of a monomial, the module $Res_{{\mathcal{B}}}^{\frak{gl}_{2n}}V_{Q}$ is locally finite. The fourth point follows from similar arguments: the maps $tr(\psi(-).F^{2}.C): {\mathcal{U(A)}} \rightarrow {\mathcal{U(A)}}$ have degree $2$ and the maps $A(-)$ and $(-)D$ clearly have degree $1$ (compare with Theorem \[corthm\]). Finally, we note that any isomorphism $\varphi:V_{Q} \rightarrow V_{Q'}$ must map the generator of $V_{Q}$ to a multiple of the generator of $V_{Q'}$. But then $q_{ij}'\varphi(1) = e_{i,n+j} \varphi(1) = \varphi(e_{i,n+j} \cdot 1) = q_{ij}\varphi(1)$, showing that $Q=Q'$ whenever such an isomorphism exists.
Alternative formula
-------------------
Since the automorphisms $\varphi$ and $\psi$ themselves are not very explicit, we present another formula for how elements of $\frak{gl}_{2n}$ act on monomials of ${\mathcal{U(A)}}$. We need some more conventions in notation for this formula.
In the argument of the trace functions, any product is by convention to be taken in $Mat_{n \times n}(U(\mathfrak{gl}_{n}))$ (in particular we identify ${\mathcal{A}}$ with $Mat_{n \times n}({\mathbb{C}})$ here). Outside the trace function all products are in $U(\mathfrak{gl}_{n})$. When $S \subset {\mathbb{Z}}$, the product $\prod_{i \in S}A_{i}$ means that the product is to be taken in order inherited from ${\mathbb{Z}}$. For example, $\prod_{i \in \{3,2,5\}}A_{i} = A_{2}A_{3}A_{5}$. For $S \subset \{1, \ldots, k\}$, we denote by $S^{*}$ the complement $\{1, \ldots, k\} \setminus S$ and by $|S|$ the cardinality of $S$.
Let $a = \prod_{i=1}^{k}A_{i}$ be a monomial in $V_{Q}$ (see Theorem \[corthm\]). The action of $\frak{gl}_{2n}$ on the monomial $a$ can be written explicitly as follows. $$\begin{aligned}
&\left( \begin{array}{cc}
A & B \\
C & D
\end{array} \right) \cdot \prod_{i=1}^{k}A_{i} := A \prod_{i=1}^{k}A_{i} - \prod_{i=1}^{k}A_{i}(Q^{-T}.D.Q^{T}) \\
&+ \sum_{S \subset \{1, \ldots, k\}}\Big( \prod_{i \in S^{*}} A_{i} \Big) \Big( (-1)^{|S|}tr(B^{T}. \prod_{i \in S} A_{i}^{T}.Q) - tr( Q^{-T}.C.\prod_{i \in S} A_{i}. F^{2}) - tr( Q^{-T}.C.\prod_{i \in S} A_{i}) tr(F) \Big)\end{aligned}$$
This follows by induction on $k$ by comparing with the formula in Theorem \[corthm\]. The verification is omitted here.
[9999]{} D. Arnal, G. Pinczon. *On algebraically irreducible representations of the Lie algebra $\mathfrak{sl}(2)$.* J. Math. Phys. [**15**]{} (1974) 350–359. I. N. Bernstein, I. M. Gelfand, S. I. Gelfand. *A certain category of $\mathfrak{g}$-modules*. Funkcional. Anal. i Prilozen. [**10**]{}, 1–8. R. Block. *The irreducible representations of the Lie algebra $\mathfrak{sl}(2)$ and of the Weyl algebra*. Adv. in Math. [**139**]{} (1981), no. 1, 69–110. P. Batra, V. Mazorchuk. *Blocks and modules for Whittaker pairs*. J. Pure Appl. Algebra [**215**]{} (2011), no. 7, 1552–1568. E. Cartan. *Les groupes projectifs qui ne laissent invariante aucune multiplicit[é]{} planet.* Bull. Soc. Math. France vol. [**41**]{} (1913) pp. 53–96. J. Dixmier. *Enveloping Algebras.* American Mathematical Society, 1977. Yu. Drozd, S. Ovsienko, V. Futorny. *On Gelfand-Zetlin modules*. Proceedings of the Winter School on Geometry and Physics (Srni, 1990). Rend. Circ. Mat. Palermo (2) Suppl. No. [**26**]{} (1991), 143–147. Yu. Drozd, S. Ovsienko, V. Futorny. *Harish-Chandra subalgebras and Gel’fand–Zetlin modules.* In: Finite-Dimensional Algebras and Related Topics (Ottawa, ON, 1992), NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., Vol. [**424**]{}, Kluwer Acad. Publ., Dordrecht, 1994, 79–93. S. Fernando. *Lie algebra modules with finite dimensional weight spaces, I.* Trans. Amer. Mas. Soc., [**322**]{} (1990), 757–781. V. Futorny, D. Grantcharov, L.E. Ramirez. *Irreducible Generic Gelfand???Tsetlin Modules of $\frak{gl}(n)$*. SIGMA Symmetry Integrability Geom. Methods Appl. [**11**]{} (2015), Paper 018, 13 pp. V. Futorny, S. Ovsienko, M. Saorín. *Torsion theories induced from commutative subalgebras*. J. Pure Appl. Algebra [**215**]{} (2011), no. 12, 2937–2948. V. Futorny. *Weight representations of semisimple finite dimensional Lie algebras*. Ph. D. Thesis, Kiev University, 1987. X.Guo, R. Lu, K. Zhao. *Irreducible modules over the Virasoro algebra.* Doc. Math. [**16**]{} (2011), 709–721. J. E. Humphreys. *Representations of Semisimple Lie Algebras in the BGG Category $\mathcal{O}$.* American Mathematical Society, 2008. B. Kostant. *On Whittaker vectors and representation theory*. Invent. Math. [**48**]{} 1978, no. 2, 101–184. T.Y. Lam. *Lectures on Modules and Rings.* Graduate Texts in Math., Vol. 189, Springer-Verlag, 1999. G. Liu, R. Lu, K. Zhao. *A class of simple weight Virasoro modules.* J. Algebra [**424**]{} (2015), 506–521. R. Lu, K. Zhao. *Irreducible Virasoro modules from irreducible Weyl modules.* J. Algebra [**414**]{} (2014), 271–287. O. Mathieu. *Classification of irreducible weight modules*, Ann. Inst. Fourier (Grenoble) [**50**]{} (2000), no. 2, 537–592. V. Mazorchuk. *On Gelfand-Zetlin modules over orthogonal Lie algebras.* Algebra Colloq. [**8**]{} (2001), no. 3, 345–360. V. Mazorchuk. *Lectures on $\mathfrak{sl}_2(\mathbb{C})$-modules*. Imperial College Press, London, 2010. V. Mazorchuk, E. Wiesner. *Simple Virasoro modules induced from codimension one subalgebras of the positive part.* Proc. Amer. Math. Soc. [**142**]{} (2014), no. 11, 3695–3703. V. Mazorchuk, K. Zhao. *Classification of simple weight Virasoro modules with a finite-dimensional weight space.* J. Algebra [**307**]{} (2007), no. 1, 209–214. V. Mazorchuk, K. Zhao. *Simple Virasoro modules which are locally finite over a positive part.* Selecta Math. (N.S.) [**20**]{} (2014), no. 3, 839–854. E. McDowell. *On modules induced from Whittaker modules.* J. Algebra [**96**]{} (1985), 161–177. E. McDowell. *A module induced from a Whittaker module.* Proc. Amer. Math. Soc. [**118**]{} (1993), 349–354. J. Nilsson. *Simple $\mathfrak{sl}_{n+1}$-module structures on ${\mathcal{U}(\mathfrak{h})}$.* J. Algebra [**424**]{} (2015), 294–329. J. Nilsson. *${\mathcal{U}(\mathfrak{h})}$-free modules and coherent families.* Preprint arXiv:1501.03091. M. Ondrus, E. Wiesner. *Whittaker modules for the Virasoro algebra.* J. Algebra Appl. [**8**]{} (2009), no. 3, 363–377. H. Tan, K. Zhao. *Irreducible modules over Witt algebras ${\mathcal{W}}_{n}$ and over $\mathfrak{sl}_{n+1}({\mathbb{C}})$*. Preprint arXiv:1312.5539. H. Tan, K. Zhao. *$\mathcal{W}_n^+$ and $\mathcal{W}_n$-module structures on $U(h)$.* J. Algebra, [**424**]{} (2015), 357-375.
Department of Mathematics, Uppsala University, Box 480, SE-751 06, Uppsala, Sweden, email: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Motion planning and control are key problems in a collection of robotic applications including the design of autonomous agile vehicles and of minimalist manipulators. These problems can be accurately formalized within the language of affine connections and of geometric control theory. In this paper we overview recent results on kinematic controllability and on oscillatory controls. Furthermore, we discuss theoretical and practical open problems as well as we suggest control theoretical approaches to them.'
author:
- |
Sonia Mart[í]{}nez\
Escuela Universitaria Politécnica\
Universidad Politécnica de Cataluña\
Av. V. Balaguer s/n\
Vilanova i la Geltrú 08800, Spain
- |
Jorge Cortés, Francesco Bullo\
Coordinated Science Laboratory\
University of Illinois at Urbana-Champaign\
1308 W. Main St\
Urbana, IL 61801, USA
title: |
\
\[3ex\] Motion planning and control problems for underactuated robots
---
Motivating problems from a variety of robotic applications
==========================================================
The research in Robotics is continuously exploring the design of novel, more reliable and agile systems that can provide more efficient tools in current applications such as factory automation systems, material handling, and autonomous robotic applications, and can make possible their progressive use in areas such as medical and social assistance applications.
Mobile Robotics, primarily motivated by the development of tasks in unreachable environments, is giving way to new generations of autonomous robots in its search for new and “better adapted” systems of locomotion. For example, traditional wheeled platforms have evolved into articulated devices endowed with various types of wheels and suspension systems that maximize their traction and the robot’s ability to move over rough terrain or even climb obstacles. The types of wheels that are being employed include passive and powered castors, ball-wheels or omni-directional wheels that allow a high accuracy in positioning and yet retain the versatility, flexibility and other properties of wheels. A rich and active literature includes (i) various vehicle designs [@SO-JA-RS:00; @FGP-SMK:94; @SS-JA-JD:95; @MW-HA:97], (ii) the automated guided vehicle “OmniMate” [@JBo:00], (iii) the roller-walker [@GE-SH:99] and other dexterous systems [@SH:00] that change their internal shape and constraints in response to the required motion sequence, and (iv) the omni-directional platform in [@RH-OK:00].
Other types of remotely controlled autonomous vehicles that are increasingly being employed in space, air and underwater applications include submersibles, blimps, helicopters, and other crafts. More often than not they rely on innovative ideas to affect their motion instead of on classic design ideas. For example, in underwater vehicle applications, innovative propulsion systems such as shape changes, internal masses, and momentum wheels are being investigated. Fault tolerance, agility, and maneuverability in low velocity regimes, as in the previous example systems, are some of the desired capabilities.
![Underactuated robots appear in a variety of environments. From left to right, a planar vertical take-off and landing (PVTOL) aircraft model, a horizontal model of a blimp and the snakeboard.[]{data-label="fig:examples"}](pvtol-oscillatory "fig:"){width=".35\textwidth"} ![Underactuated robots appear in a variety of environments. From left to right, a planar vertical take-off and landing (PVTOL) aircraft model, a horizontal model of a blimp and the snakeboard.[]{data-label="fig:examples"}](blimp "fig:"){width=".25\textwidth"} ![Underactuated robots appear in a variety of environments. From left to right, a planar vertical take-off and landing (PVTOL) aircraft model, a horizontal model of a blimp and the snakeboard.[]{data-label="fig:examples"}](snakeboard "fig:"){width=".35\textwidth"}
A growing field in Mobile Robotics is that of *biomimetics*. The idea of this approach is to obtain some of the robustness and adaptability that biological systems have refined through evolution. In particular, biomimetic locomotion studies the periodic movement patterns or *gaits* that biological systems undergo during locomotion and then takes it as reference for the design of the mechanical counterpart. In other cases, the design of to robots without physical counterpart is inspired by similar principles. Robotic locomotion systems include the classic *bipeds and multi-legged robots* as well as swimming snake-like robots and flying robots. These systems find potential applications in harsh or hazardous environments, such as under deep or shallow water, on rough terrain (with stairs), along vertical walls or pipes and other environments difficult to access for wheeled robots. Specific examples in the literature include hyper-redundant robots [@GC-JWB:94; @SH:93], the snakeboard [@ADL-JPO-RMM-JWB:94; @JPO-JWB:98], the $G$-snakes and roller racer models in [@PSK-DPT:94; @PSK-DPT:01], fish robots [@NK-TI:98; @SDK-RJM-CTA-RMM-JWB:98], eel robots [@JCJ-SK-JA:95; @KAM-JPO:99a], and passive and hopping robots [@JKH-MHR:90; @TM:90; @MHR:86].
All this set of emerging robotic applications have special characteristics that pose new challenges in motion planning. Among them, we highlight:
#### Underactuation.
This could be owned to a design choice: nowadays low weight and fewer actuators must perform the task of former more expensive systems. For example, consider a manufacturing environment where robotic devices perform material handling and manipulation tasks: automatic planning algorithms might be able to cope with failures without interrupting the manufacturing process. Another reason why these systems are underactuated is because of an unavoidable limited control authority: in some locomotion systems it is not possible to actuate all the directions of motion. For example, consider a robot operating in a hazardous or remote environment (e.g., aerospace or underwater), an important concern is its ability to operate faced with a component failure, since retrieval or repair is not always possible.
#### Complex dynamics.
In these control systems, the drift plays a key role. Dynamic effects must necessarily be taken into account, since kinematic models are no longer available in a wide range of current applications. Examples include lift and drag effects in underwater vehicles, the generation of momentum by means of the coupling of internal shape changes with the environment in the eel robot and the snakeboard, the dynamic stability properties of walking machines and nonholonomic wheeled platforms, etc.
#### Current limitations of motion algorithms.
Most of the work on motion planning has relied on assumptions that are no longer valid in the present applications. For example, one of these is that (wheeled) robots are kinematic systems and, therefore, controlled by velocity inputs. This type of models allows one to design a control to reach a desired point and then immediately stop by setting the inputs to zero. This is obviously not the case when dealing with complex dynamic models.
Another common assumption is the one of fully actuation that allows to decouple the motion planning problem into path planning (computational geometry) and then tracking. For underactuated systems, this may be not possible because we may be obtaining motions in the path planning stage that the system can not perform in the tracking step because of its dynamic limitations.
![Vertical view of an omni-directional robotic platform with 6 degrees of freedom and 3 nonholonomic constraints [@GC-GB-BDAN:96; @RH-OK:00]. This device is capable of highly accurate positioning, high payloads, and high speed motion. In its fully actuated configuration, the robot is endowed with 6 motors at the three wheels and at the three joints $(\beta_1,\beta_2,\beta_3)$. However, underactuated configurations can arise because of failures or intentional design.[]{data-label="fig:omni"}](figure-omni){width=".42\textwidth"}
Furthermore, motion planning and optimization problems for these systems are nonlinear, non-convex problems with exponential complexity in the dimension of the model. These issues have become increasingly important due to the high dimensionality of many current mechanical systems, including flexible structures, compliant manipulators and multibody systems undergoing reconfiguration in space.
#### Benefits that would result from better motion planning algorithms for underactuated systems.
From a practical perspective, there are at least two advantages to designing controllers for underactuated robotic manipulators and vehicles. First, a fully actuated system requires more control inputs than an underactuated system, which means there will have to be more devices to generate the necessary forces. The additional controlling devices add to the cost and weight of the system. Finding a way to control an underactuated version of the system would improve the overall performance or reduce the cost. The second practical reason for studying underactuated vehicles is that underactuation provides a backup control technique for a fully actuated system. If a fully actuated system is damaged and a controller for an underactuated system is available, then we may be able to recover gracefully from the failure. The underactuated controller may be able to salvage a system that would otherwise be uncontrollable.
Mathematical unifying approach to the modeling of robotic systems {#sec:modeling}
=================================================================
Most of the robotic devices we have mentioned so far can be characterized by their special Lagrangian structure. They usually exhibit symmetries and their motion is constrained by the environment where they operate. In the following, we introduce a general modeling language for underactuated robotic systems.
Let $q=(q^1,\ldots,q^n)\in Q$ be the configuration of the mechanical system and consider the control equations: $$\begin{gathered}
\label{eq:mechsys}
\ddot{q}^i + \Gamma_{jk}^i(q)\dot{q}^j\dot{q}^k = - M^{ij}{\frac{\partial V}{\partial q^j}}
+ k^i_j(q) \dot{q}^j + Y_1^i(q) u_1 + \ldots + Y_m^i(q) u_m \, ,\end{gathered}$$ where the summation convention is in place for the indices $j,k$ that run from $1$ to $n$, and
1. $V:Q \rightarrow {{\mathbb{R}}}$ corresponds to potential energy, and $k^i_j(q) \dot{q}^j$ corresponds to damping forces,
2. $\{\Gamma_{jk}^i: i,j,k=1,\ldots,n\}$ are $n^3$ Christoffel symbols, derived from $M(q)$, the inertia matrix defining the kinetic energy, according to $$\Gamma_{{i}{j}}^k = \frac{1}{2} M^{mk}\left(
{\frac{\partial M_{m{j}}}{\partial q^{i}}} +
{\frac{\partial M_{m{i}}}{\partial q^{j}}}-{\frac{\partial M_{{i}{j}}}{\partial q^m}} \right) \, ,$$ where $M^{mk}$ is the $(m,k)$ component of $M^{-1}$, and,
3. $\{F_a: a=1,\ldots,m\}$ are the $m$ input co-vector fields, and $\{Y_a =
M^{-1} F_a: a=1,\ldots,m\}$ are the $m$ input vector fields.
Underactuated systems have fewer control actuators, $m$, than degrees of freedom $n>m$. Other limitations on the control signals $u_a$ might be present, e.g., actuators might have magnitude and rate limits, or they might only generate unilateral or binary signals (e.g., thrusters in satellites).
The notion of affine connection provides a coordinate-free means of describing the dynamics of robotic systems. Given two vector fields $X,Y$, the *covariant derivative* of $Y$ with respect to $X$ is the third vector field $\nabla_XY$ defined via $$\label{cov:der:Christoffels}
(\nabla_X Y)^i = \frac{\partial Y^i}{\partial q^j} X^j + \Gamma^i_{jk} X^j
Y^k .$$ The operator $\nabla$ is called the *affine connection* for the mechanical system in equation . We write the Euler-Lagrange equations for a system subject to a conservative force $Y_0$, a damping force $k(q)\dot{q}$ and $m$ input forces as: $$\label{eq:mechsys:affine}
\nabla_{\dot{q}}\dot{q} = Y_0(q) + k(q)(\dot{q}) + \sum_{a=1}^{m} Y_a (q)
u_a(t).$$ Equation is a coordinate-free version of equation . A crucial observation is the fact that systems subject nonholonomic constraints can also be modeled by means of affine connections. In the interest of brevity, we refer to [@FB-MZ:01d; @ADL:97a] for the exposition of this result and the explicit expression of the Christoffel symbols corresponding to the Lagrange-d’Alembert equations.
### The homogeneous structure of mechanical systems. {#subsec:Lie .unnumbered}
The fundamental structure of the control system in equation is the polynomial dependence of the various vector fields on the velocity variable $\dot{q}$. This structure affects the Lie bracket computations involving input and drift vector fields. The system is written in first order differential equation form as $$\begin{aligned}
\frac{d}{dt} \begin{bmatrix} q \\ \dot{q} \end{bmatrix} =
\begin{bmatrix} \dot{q} \\ -\Gamma(q,\dot{q}) +Y_0(q)
+k(q)(\dot{q})\end{bmatrix}
+ \sum_{a=1}^m \begin{bmatrix}0\\Y_a\end{bmatrix}u_a(t)\end{aligned}$$ where $\Gamma(q,\dot{q})$ is the vector with $i$th component $\Gamma^i_{jk}(q)\dot{q}^j\dot{q}^k$. Also, if $x=(q,\dot{q})$, $$\begin{aligned}
Z(x) = \begin{bmatrix} \dot{q} \\ -\Gamma(q,\dot{q}) \end{bmatrix} ,
\quad
Y_a{^{\text{lift}}}(x) {\triangleq}\begin{bmatrix} 0 \\ Y_a(q) \end{bmatrix} ,
\quad \text{and} \quad
k{^{\text{lift}}}(x) {\triangleq}\begin{bmatrix} 0 \\ k(q)(\dot{q}) \end{bmatrix} ,\end{aligned}$$ the control system is rewritten as $$\dot{x} = Z(x) + Y_0{^{\text{lift}}}(x) + k{^{\text{lift}}}(x) + \sum_{a=1}^m Y_a{^{\text{lift}}}(x)
u_a(t) \, .$$
Let $h_i(q,\dot{q})$ be the set of scalar functions on ${{\mathbb{R}}}^{2n}$ which are arbitrary functions of $q$ and homogeneous polynomials in $\{\dot{q}^1,\dots,\dot{q}^n\}$ of degree $i$. Let $\mathcal{P}_i$ be the set of vector fields on ${{\mathbb{R}}}^{2n}$ whose first $n$ components belong to $h_i$ and whose second $n$ components belong to $h_{i+1}$. We note that these notions can also be defined on a general manifold, see [@FB-ADL:00d].
We are now ready to present two simple ideas. First, all the previous vector fields are homogeneous polynomial vector fields for some specific value of $i$. Indeed, $Z \in \mathcal{P}_1$, $k{^{\text{lift}}}\in \mathcal{P}_{0}$, and $Y_a{^{\text{lift}}}\in \mathcal{P}_{-1}$. Second, since the Lie bracket between a vector field in $\mathcal{P}_i$ and a vector field in $\mathcal{P}_j$ belongs to $\mathcal{P}_{i+j}$, any Lie bracket of the given relevant vector fields remains a homogeneous polynomial. In other words, the set of homogeneous vector fields is closed under the operation of Lie bracket.
A consequence of this analysis is the definition of symmetric product of vector fields. We define the *symmetric product* between $Y_b$ and $Y_a$ as the vector field ${\langle Y_a: Y_b\rangle} = {\langle Y_b: Y_a\rangle}$ given by $$\begin{aligned}
{\langle Y_b: Y_a\rangle}^i = {\langle Y_a: Y_b\rangle}^i
&= \frac{\partial Y_a^i}{\partial q^j}Y_b^j
+\frac{\partial Y_b^i}{\partial q^j}Y_a^j + \Gamma^i_{jk}
\left(Y_a^jY_b^k + Y_a^kY_b^j\right).\end{aligned}$$ Straightforward computations show that ${\langle Y_a: Y_b\rangle}^{\text{lift}} =
[Y_b^{\text{lift}},[Z_g,Y_a^{\text{lift}}]]$. This operation plays a key role in nearly all the control problems associated with this class of systems: nonlinear controllability [@JC-SM-FB:01m; @ADL-RMM:95c], optimal control [@MC-FSL-PEC:95; @ADL:99a], dynamic feedback linearization [@MR-RMM:96b], algorithms for motion planning and stabilization [@FB-NEL-ADL:00; @SM-JC:02; @JPO:00a], etc.
### A series expansion for the forced evolution starting from rest. {#sec:series-expansion .unnumbered}
The homogeneous structure of the mechanical control system , together with the symmetric product, set the basis to establish the following description of the evolution of the system trajectories starting with zero initial velocity [@FB:99a; @JC-SM-FB:01m]. Assume no potential or damping forces are present in the system. Let $Y(q,t) = \sum_{a=1}^m Y_a(q) u_a(t)$. Define recursively the vector fields $V_k$ by $$\begin{aligned}
V_1(q,t) &= \int_0^tY(q,s)ds \, , \; \;
V_k(q,t) = - \frac{1}{2} \sum_{j=1}^{k-1} \;
\int_0^t\!{\Big\langle V_j(q,s) : \null\; V_{k-j}(q,s)\Big\rangle} ds.\end{aligned}$$ Then, the solution $q(t)$ of equation satisfies $$\label{eq:thm3}
\dot{q}(t) = \sum_{k=1}^{+\infty} {V}_k(q(t),t),$$ where the series converges absolutely and uniformly in a neighborhood of $q_0$ and over a fixed time interval $t\in [0,T]$. This series expansion provides a means of describing the open-loop response of the system to any specific forcing. As we will see below, it plays a key role in several motion planning and control strategies for underactuated robots.
Existing results on planning for underactuated systems
======================================================
To design planning algorithms for underactuated robotic systems, we advocate an integrated approach based on modeling, system design, controllability analysis, dexterity, manipulability, and singularities. These analysis concepts are fundamental for robust planning algorithms that do not solely rely on randomization or nonlinear programming. We do not suggest closed-form planning algorithms, rather we envision methods that combine the best features of formal analysis and of numerical algorithms.
For reasons of space, we cannot present a detailed account of all existing results on motion planning for underactuated systems, and not even of the results obtained within the modeling approach proposed in Section \[sec:modeling\]. Therefore, we focus on two specific control methodologies for motion planning: decoupled planning algorithms for kinematically controllable systems, and approximate inversion algorithms based on oscillatory controls.
Section \[subsec:kinematic-controllability\] reviews *decoupled planning algorithms* that exploit certain differential geometric properties to reduce the complexity of the motion planning problem (still to be solved via numerical algorithms). The notion of *kinematic controllability* is extremely effective: trajectory planning decouples from being a problem on a $2n$ dimensional space to an $n$ dimensional space. Furthermore, various state constraints can be neglected in the reduced space. For systems that are not kinematically controllable and that require oscillatory controls to locomote, Section \[subsec:approximate-inversion\] presents motion planning algorithms based on *approximate inversion*. Both design methods are closely related to recent results on nonlinear controllability [@FB-KML:01a; @ADL-RMM:95c], power series expansions [@FB:99a; @JC-SM-FB:01m], two time-scales coordinate-free averaging [@FB:99b; @SM-JC-FB:01e], and nonlinear inversion algorithms [@FB-NEL-ADL:00; @SM-JC:02].
The strengths of this methodology are as follows. Both methodologies provide solutions to the corresponding problems, i.e., point to point and trajectory planning. These analytic results do not rely on non-generic assumptions such as feedback linearization, nilpotency or flatness. The results are coordinate-free and hence widely applicable, e.g., to aerospace or underwater robotics settings. Both methods are *consistent*, *complete* and constructive (consistent planners recover the known solutions available for linear and nilpotent systems, and complete planners are guaranteed to find a local solution for any nonlinearly controllable system).
Kinematic controllability for underactuated robots {#subsec:kinematic-controllability}
--------------------------------------------------
The following decoupling methodology was proposed in [@FB-KML:01a] to reduce the complexity of the motion planning problem. The method is constructive (only quadratic equations and no PDEs are involved) and physically intuitive.
![A three-revolute-joints device. It can be proven [@FB-KML:01a] that any two-actuator configuration of this system is kinematically controllable, i.e., one can always find two decoupling vector fields whose involutive closure is full-rank.[]{data-label="fig:three-links"}](three-links){width=".4\textwidth"}
We consider as a motivating example a common pick-&-place manipulator: Fig. \[fig:three-links\] shows a vertical view of a three-revolute-joints device. We investigate planning schemes for this system when one of its three motors is either failed or missing. We present a decoupling idea to reduce the complexity of the problem: instead of searching for feasible trajectories of a dynamic system in ${{\mathbb{R}}}^6$, we show how it suffices to search for paths of a simpler, kinematic (i.e., driftless) system in ${{\mathbb{R}}}^3$.
A curve $\gamma:[0,T]\mapsto Q$ is a controlled solution to equation if there exist inputs $u_a:[0,T]\to{{\mathbb{R}}}$ for which $\gamma$ solves . To avoid the difficult task of characterizing all controlled solutions of the system , we focus on curves satisfying $\dot{\gamma}=\dot{s}(t)X(\gamma)$, where $X$ is a vector field on $Q$, and where the map $s:[0,T]\to [0,1]$ is a “time-scaling” parameterization of $\gamma$. Such curves are called *kinematic motions*.
We call $V$ a *decoupling vector field* if all curves $\gamma$ satisfying $\dot{\gamma}=\dot{s}(t)V(\gamma)$ for any time scaling $s$, are kinematic motions. This definition is useful for three reasons. First, $V$ is decoupling if and only if $V$ and $\nabla_VV$ are linear combinations in $\{Y_1,\dots,Y_m\}$. Second, decoupling vector fields can be computed by solving $(n-m)$ quadratic equations. Third, if enough decoupling vector fields, say $V_1,\dots,V_p$, are available to satisfy the LARC, we call the system *kinematically controllable*. In the latter case, we can plan motions for the kinematic system $\dot{q} = \sum_{a=1}^p w_a(t) V_a(q)$, and they will automatically be controlled curves for the original system .
Approximate inversion via small amplitude and oscillatory controls {#subsec:approximate-inversion}
------------------------------------------------------------------
As in the previous section, the objective is to design motion planning and stabilization schemes for underactuated systems. We propose perturbation and inversion methods as widely applicable approaches to solve point to point and trajectory planning problems. Let us regard the flow map $\Phi$ of equation over a finite time interval as a map from the input functions $u_i:[0,T]\to{{\mathbb{R}}}$ to the target state $x(T)$. The ideal algorithm for point-to-point planning computes an exact (right) inverse $\Phi^{-1}$ of $\Phi$. Unfortunately, closed form expressions for $\Phi^{-1}$ are available only assuming *non-generic* differential geometric conditions (e.g., the system needs to be feedback linearizable, differentially flat, or nilpotent). Instead of aiming at “exact” solutions, we focus on computing an *approximate inverse map* using perturbation methods such as power series expansions and averaging theory. Although these tools are only approximate, the resulting algorithms are consistent and complete.
### Oscillatory (high frequency, high amplitude) controls for trajectory planning. {#oscillatory-high-frequency-high-amplitude-controls-for-trajectory-planning. .unnumbered}
We present the approach in three steps and refer to [@SM-JC-FB:01e] for all the details. As first step, we present a recent coordinate-free averaging result. Let $0<{\epsilon}\ll 1$. Assume the control inputs are of the form $$u_i = \frac{1}{{\epsilon}} u_i\left(\frac{t}{{\epsilon}},t\right),$$ and assume they are $T$-periodic and zero-mean in the first variable. Define the averaged multinomial iterated integrals of $u_1,\dots,u_m$ as $${\overline}{{{\mathcal{U}}}}_{k_1, \dots, k_m}(t) = \frac{T^{-1}}{k_1! \dots k_{m}!}
\int_0^T\!\! \left(\int_0^s\!\!u_1(\tau,t) d\tau \right)^{k_1}
\hspace*{-5pt} \dots \left(\int_0^s\!\! u_m(\tau,t) d\tau \right)^{k_m}
\hspace*{-5pt} ds \, .$$ Let $a,b,c$ take value in $\{1,\dots,m\}$. Let $\vec{k}_a$ (resp. $\vec{k}_{ab}$) denote the tuple $(k_1,\dots,k_m)$ with $k_c = \delta_{ca}$ (resp. $k_c =\delta_{ca} + \delta_{cb}$). Then, over a finite time $q(t) =
r(t) + O({\epsilon})$, as ${\epsilon}\rightarrow 0$, where $r(t)$ satisfies $$\begin{aligned}
\label{eq:mechsys:averaged}
& \nabla_{\dot{r}}\dot{r} = Y_{0}(r) + k(r)(\dot{r}) + \sum_{a=1}^m \left(
\frac{1}{2} {\overline}{{{\mathcal{U}}}}_{\vec{k}_a}^2 (t) - {\overline}{{{\mathcal{U}}}}_{\vec{k}_{aa}}(t)\right)
{\langle Y_a: Y_a\rangle}(r) \\
& \hspace*{1.5cm} + \sum_{a<b}\left( {\overline}{{{\mathcal{U}}}}_{\vec{k}_a}(t) {\overline}{{{\mathcal{U}}}}_{\vec{k}_b}(t)
-{\overline}{{{\mathcal{U}}}}_{\vec{k}_{ab}}(t)\right){\langle Y_a: Y_b\rangle}(r) \,. \nonumber\end{aligned}$$
As a second step, given $z_a(t)$, $z_{bc}(t)$ arbitrary functions of time, we propose the following inversion procedure
1. take the functions $\psi_{N(a,b)}(t) = \sqrt{2} \, N(a,b) \cos
(N(a,b)\,t)$, where $(a,b)\mapsto N(a,b)\in\{1,\ldots,N\}$ is an enumeration of the pairs of integers $(a,b)$, $a<b$.
2. select the following controls in , $$\begin{aligned}
u_a(t,q) &= v_a(t,q) + \frac{1}{{\epsilon}} w_a\left(\frac{t}{{\epsilon}},t\right) \,
, \\
w_a(\tau,t) &=
-\sum_{c=1}^{a-1} \psi_{N(c,a)} (\tau) + \sum_{c=a+1}^{m} z_{ac}(t)
\psi_{N(a,c)} (\tau) \, ,
\end{aligned}$$ where $v_a(t,q)$ are still to be chosen.
![Approximate trajectory tracking for an underactuated PVTOL model by means of oscillatory controls. The curve to be tracked is shown solid, and the various oscillating curves correspond to different values of the parameter ${\epsilon}$[]{data-label="fig:pvtol"}](pvtol-comparison.eps){width=".55\textwidth"}
After computing the averaged iterated integrals of the oscillatory inputs $w_a(t/{\epsilon},t)$, equation for the averaged system becomes $$\begin{gathered}
\nabla_{\dot{r}}\dot{r} = Y_0(r)+ k(r)(\dot{r}) +
\sum_{a=1}^m v_a(t,r) Y_a(r) \\ -
\sum_{a=1}^m {\overline}{{{\mathcal{U}}}}_{\vec{k}_{aa}}(t) {\langle Y_a: Y_a\rangle}(r)
+ \sum_{a<b} z_{ab}(t) {\langle Y_a: Y_b\rangle}(r)\,.
\end{gathered}$$
As a third and final step, assume that all the vector fields of the form ${\langle Y_b: Y_b\rangle}$ belong to ${\operatorname{span}}\{Y_a\}$. Let $\alpha_{ab} :
Q\rightarrow {{\mathbb{R}}}$ be such that ${\langle Y_a: Y_a\rangle}(q) = \sum_{b}
\alpha_{ab}(q)Y_b(q)$, $q \in Q$. Select $$v_a(t,q) = z_a(t) + \frac{1}{2} \sum_{b=1}^m \alpha_{ba}(q) \left( b-1 +
\sum_{c=b+1}^m (z_{bc}(t))^2 \right) \, .$$ Then, we have $$\begin{aligned}
\sum_{a=1}^m v_a(t,r)Y_a(r) = \sum_{a=1}^m z_a^d(t) Y_a(r) + \sum_{a=1}^m
{\overline}{{{\mathcal{U}}}}_{\vec{k}_{aa}}(t){\langle Y_a: Y_a\rangle}(r)\,,\end{aligned}$$ which implies that eq. takes the final form, $$\begin{aligned}
\nabla_{\dot{r}}\dot{r} &= Y_0(r)+ k(r)(\dot{r}) + \sum_{a=1}^m z_a(t) Y_a(r) +
\sum_{b<c} z_{bc}(t) {\langle Y_b: Y_c\rangle}(r)\,,
\end{aligned}$$ The averaged system now has more available control inputs than the original one. If the input distribution ${{\mathcal{I}}}= {\operatorname{span}}\{Y_a\,, {\langle Y_b: Y_c\rangle}\}$is full rank, then the latter system is fully actuated (i.e., one control input is available for each degree of freedom). If the input distribution ${{\mathcal{I}}}$ contains a sufficient number of decoupling vector fields, then the system is kinematically controllable. In both cases, we have reduced the complexity of the motion planning problem.
[A related approach to motion planning relies on small amplitude periodic forcing; see [@FB-NEL-ADL:00; @SM-JC:02]. The planning problem is solved by approximately inverting the series expansion describing the evolution of the control system (cf. Section \[sec:series-expansion\]). This inversion procedure is very similar to the one presented above. Based on it, one can establish two simple primitives of motion to change and maintain velocity, while keeping track of the changes in the configuration. These primitives can then be used as the building blocks to design high-level motion algorithms that solve the point-to-point reconfiguration problem, the static interpolation problem and the local exponential stabilization problem. Fig. \[fig:small-amplitude\] shows two examples of the execution of these algorithms. ]{}
![Illustration of the motion planning algorithms via small amplitude periodic forcing for a simple planar body (left) and the blimp model (right). The errors in the final configuration are within the same order of magnitude of the input employed[]{data-label="fig:small-amplitude"}](c-1.eps "fig:"){width=".58\textwidth"} ![Illustration of the motion planning algorithms via small amplitude periodic forcing for a simple planar body (left) and the blimp model (right). The errors in the final configuration are within the same order of magnitude of the input employed[]{data-label="fig:small-amplitude"}](ptp1.ps "fig:"){width=".4\textwidth"}
Open problems and possible approaches
=====================================
Immediate open questions arising from the above-presented results are the following:
#### Kinematic modeling and control.
The current limitations are as follows: the design problem is now reduced to planning for a kinematic system with the additional constraint of zero-velocity transitions between feasible motions. This additional constraint leads to poor performance when coupled with current randomized planners [@DH-JCL-RM:99a; @LEK-PS-JCL-MHO:96; @SML-JJK:00; @SML-JJK:01] that switch frequently between the available motions. The zero-velocity switches also create problems for trajectory tracking controllers based on linearization, since the system loses linear controllability at zero-velocity. Finally, there is no notion of time-optimality for these kinematic motions and there is no way of dealing with systems where oscillatory inputs are needed for locomotion (see below for a discussion on this point). Motivated by this analysis, we identify the following open issues:
1. Develop a catalog of kinematically controllable systems, including planar manipulators with revolute as well as prismatic joints, parallel manipulator, manipulators in three dimensional space and in aerospace and underwater environments (accounting for the different dynamics in such settings). Some preliminary work in this direction can be found in [@FB-ADL-KML:02a]. Analyze and classify the singularities that these vector fields possess as a prerequisite step for planning purposes.
2. A (left) group action is a map $\psi: G \times Q \rightarrow Q$ such that $\psi (e,q)=q$, for all $q \in Q$, where $e$ denotes the identity element in $G$, and $\psi(g,\psi(h,q))=\psi(gh,q)$, for all $g,h \in G$, $q
\in Q$. Usually $G \subset SE(n)$, and then the action describes a rigid displacement of some components of the robot. An interesting problem would be to identify conditions under which decoupling vector fields can be found which are invariant under such group actions. When this is the case, motion plans can be designed exploiting established “inverse kinematics” methods; see [@RMM-ZXL-SSS:94 Chapter 3]. This simplification eliminates the need for any numerical procedure if the robot moves in an un-obstructed environment, or further reduces the dimensionality and complexity of the resulting search problem in complex environments.
3. To tackle the difficulties inherent with zero-velocity transitions, it would be appropriate to develop randomized planners which require as few switches between decoupling vector fields as possible, and to develop trajectory tracking controllers for these systems able to adequately perform through the singularities.
4. Another interesting idea would consist of switching between decoupling vector fields without stopping. In some sense, this is also related to the problem of developing transitions between relative equilibria. *Relative equilibria* are “steady trajectories” that the system admits as feasible solutions. This family of trajectories is of great interest in theory and applications as they provide a rich family of motions with the simplifying property of having constant body-fixed velocity. Relative equilibria for systems in three dimensional Euclidean space include straight lines, circles, and helices. Despite partial results, no method is currently available to design provably stable switching maneuvers from one relative equilibrium to another (or from one decoupling vector field to another without stopping). A necessary preliminary step toward this objective is to analyze the controllability properties of underactuated systems moving along a relative equilibrium or along a decoupling vector field.
#### Small-amplitude and high-frequency controls.
The current limitations are as follows. The implementation of the small amplitude approach requires the computation and manipulation of high order tensors, and the approach has a limited region of convergence. The implementation of the oscillatory control approach presents difficulties in most physical settings because of the required high frequency, high amplitude inputs. Motivated by this analysis, we think that the following are interesting issues to explore:
1. For the small amplitude controls formulation, open questions include (a) investigate tight estimates for the region of validity of the truncations (simulation studies suggest that there are better bounds than the conservative ones currently available), (b) design base functions optimal with regards to region of convergence and appropriate cost criteria, (c) design inversion algorithms for systems that are not linearly controllable. The latter setting is equivalent to a non-definite quadratic programming problem, i.e., to the problem of finding sufficient conditions for a vector-valued quadratic form to be surjective (see [@FB-JC-ADL-SM:02d] for a discussion on this subject).
2. For the oscillatory controls formulation, standing problems are (a) investigate the use of high-frequency bounded amplitude controls, (b) characterize approximate kinematic controllability and differential flatness via oscillations, (c) investigate physical settings in which oscillatory controls are natural control means, e.g., micro-electromechanical robots, (d) investigate extensions of this coordinate-free perturbation theory to discrete-time nonlinear systems, and to distributed parameter systems and partial differential equations.
3. An ambitious program would consist of developing schemes that combine the proposed analytic methods with iterative numerical algorithms. One approach is via homotopy and level set methods [@ELA-KG:90; @JAS:96] as schemes that overcome the limitations induced by the small parameter (small convergence region or high amplitude high frequency). A second direction is to use the planner based on small amplitude controls as a local planner inside a global search algorithm based on randomization; see [@TK-FB:01h] for some preliminary results on local/global planners.
Acknowledgments {#acknowledgments .unnumbered}
===============
This research was partially funded by NSF grants CMS-0100162 and IIS-0118146. The authors would like to thank Andrew Lewis, Kevin Lynch, Miloš Žefran, Todd Cerven, and Timur Karatas.
[10]{}
Allgower EL, Georg K (1990) . Springer Verlag, New York.
Borenstein J (2000) The [OmniMate]{}: a guidewire- and beacon-free [AGV]{} for highly reconfigurable applications. 38(9):1993–2010.
Bullo F (2001) Series expansions for the evolution of mechanical control systems. 40(1):166–190.
Bullo F (2002) Averaging and vibrational control of mechanical systems. 41(2):542–562.
Bullo F, Cort[é]{}s J, Lewis AD, Mart[í]{}nez S (2002) Vector-valued quadratic forms in control theory. In: [*Mathematical Theory of Networks and Systems (MTNS)*]{}, Notre Dame, Indiana, August 2002. Workshop on Open Problems in Mathematical Systems and Control Theory.
Bullo F, Leonard NE, Lewis AD (2000) Controllability and motion algorithms for underactuated [L]{}agrangian systems on [L]{}ie groups. 45(8):1437–1454.
Bullo F, Lewis AD (2000) On the homogeneity of the affine connection model for mechanical control systems. In: [*[IEEE]{} Int Conf on Decision and Control*]{}, Sydney, Australia, December 2000, pp 1260–1265.
Bullo F, Lewis AD, Lynch KM (2002) Controllable kinematic reductions for mechanical systems: concepts, computational tools, and examples. In: [*Mathematical Theory of Networks and Systems (MTNS)*]{}, Notre Dame, Indiana, August 2002.
Bullo F, Lynch KM (2001) Kinematic controllability for decoupled trajectory planning in underactuated mechanical systems. 17(4):402–412.
Bullo F, [Ž]{}efran M (2002) On mechanical control systems with nonholonomic constraints and symmetries. 45(2):133–143.
Camari[ñ]{}a M, Silva Leite F, Crouch PE (1995) Splines of class $C^k$ on non-[E]{}uclidean spaces. 12:399–410.
Campion G, Bastin G, D’Andrea-Novel B (1996) Structural properties and classification of kinematic and dynamic models of wheeled mobile robots. 12(1):47–62.
Chirikjian G, Burdick JW (1994) Kinematics of hyperredundant locomotion. 11(6):781–793.
Cort[é]{}s J, Mart[í]{}nez S, Bullo F (2002) On nonlinear controllability and series expansions for [L]{}agrangian systems with dissipative forces. 47(8):1396–1401.
Endo G, Hirose S (1999) Study on roller-walker (system integration and basic experiments). In: [*[IEEE]{} Int Conf on Robotics and Automation*]{}, Detroit, Michigan, May 1999, pp 2032–2037.
Hirose S (1993) . Oxford University Press, Oxford.
Hirose S (2000) Variable constraint mechanism and its application for design of mobile robots. 19(11):1126–1138.
Hodgins JK, Raibert MH (1990) Biped gymnastics. 9(2):115–132.
Holmberg R, Khatib O (2000) Development and control of a holonomic mobile robot for mobile manipulation tasks. 19(11):1066–1074.
Hsu D, Latombe JC, Motwani R (1999) Path planning in expansive configuration spaces. 9(4):495–512.
Jalbert JC, Kashin S, Ayers J (1995) Design considerations and experiments of a biologically based undulatory lamprey [AUV]{}. In: [*9th Int Symposium on Unmanned Untethered Submersible Technology*]{}, Durham, New Hampshire.
Karatas T, Bullo F (2001) Randomized searches and nonlinear programming in trajectory planning. In: [*[IEEE]{} Conf on Decision and Control*]{}, Orlando, Florida, December 2001, pp 5032–5037.
Kato N, Inaba T (1998) Guidance and control of fish robot with apparatus of pectoral fin motion. In: [*[IEEE]{} Int Conf on Robotics and Automation*]{}, Leuven, Belgium, May 1998, pp 446–451.
Kavraki LE, [Š]{}vestka P, Latombe JC, Overmars MH (1996) Probabilistic roadmaps for path planning in high-dimensional space. 12(4):566–580.
Kelly SD, Mason RJ, Anhalt CT, Murray RM, Burdick JW (1998) Modelling and experimental investigation of carangiform locomotion for control. In: [*[IEEE]{} [A]{}merican [C]{}ontrol [C]{}onference*]{}, Philadelphia, PA, 1998, pp 1271–1276.
Krishnaprasad PS, Tsakiris DP (1994) -snakes: Nonholonomic kinetic chains on [Lie]{} groups. In: [*[IEEE]{} Int Conf on Decision and Control*]{}, Lake Buena Vista, Florida, December 1994, pp 2955–2960.
Krishnaprasad PS, Tsakiris DP (2001) Oscillations, ${SE}(2)$-snakes and motion control: a study of the roller racer. 16(4):347–397.
LaValle SM, Kuffner JJ Rapidly-exploring random trees: Progress and prospects. In: [*Workshop on Algorithmic Foundations of Robotics*]{}, Dartmouth, New Hampshire, March 2000, pp 293–308.
LaValle SM, Kuffner JJ. (2001) Randomized kinodynamic planning. 20(5):378–400.
Lewis AD (2000) The geometry of the maximum principle for affine connection control systems. Preprint available at: `http://penelope.mast.queensu.ca/~andrew`.
Lewis AD (2000) Simple mechanical control systems with constraints. 45(8):1420–1436.
Lewis AD, Murray RM (1997) Configuration controllability of simple mechanical control systems. 35(3):766–790.
Lewis AD, Ostrowski JP, Murray RM, Burdick JW (1994) Nonholonomic mechanics and locomotion: the snakeboard example. In: [*[IEEE]{} Int Conf on Robotics and Automation*]{}, San Diego, California, May 1994, pp 2391–2400.
Mart[í]{}nez S, Cortés J (2002) Motion control algorithms for simple mechanical systems with symmetry. , to appear.
Mart[í]{}nez S, Cort[é]{}s J, Bullo F (2001) Analysis and design of oscillatory controls systems. , submitted. Available electronically at `http://motion.csl.uiuc.edu`.
McGeer T (1990) Passive dynamic walking. 9(2):62–82.
Mclsaac KA, Ostrowski JP (1999) A geometric approach to anguilliform locomotion: modeling of an underwater eel robot. In: [*[IEEE]{} Int Conf on Robotics and Automation*]{}, Detroit, Michigan, May 1999, pages 2843–2848.
Murray RM, Li ZX, Sastry SS (1994) . CRC Press, Boca Raton, Florida.
Ostrovskaya S, Angeles J, Spiteri R (2000) Dynamics of a mobile robot with three ball-wheels. 19(4):383–393.
Ostrowski JP (2000) Steering for a class of dynamic nonholonomic systems. 45(8):1492–1497.
Ostrowski JP, Burdick JW (1998) The geometric mechanics of undulatory robotic locomotion. 17(7):683–701.
Pin FG, Killough SM (1994) A new family of omnidirectional and holonomic wheeled platforms for mobile robots. 10(4):480–489.
Raibert MH (1986) . MIT Press, Cambridge, Massachusetts.
Rathinam M, Murray RM (1998) Configuration flatness of [L]{}agrangian systems underactuated by one control. 36(1):164–179.
Saha S, Angeles J, Darcovich J (1995) The design of kinematically isotropic rolling robots with omnidirectional wheels. 30(8):1127–1137.
Sethian JA (1996) . Cambridge University Press, New York.
West M, Asada H (1997) Design of ball wheel mechanisms for omnidirectional vehicles with full mobility and invariant kinematics. 119(2):153–161.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Sreetama Das$^{1,2}$, Titas Chanda$^{1,2}$, Maciej Lewenstein$^{3, 4}$, Anna Sanpera$^{4, 5}$, Aditi Sen(De)$^{1,2}$, and Ujjwal Sen$^{1,2}$'
title: |
Lectures on Quantum Information\
Chapter 1: The separability versus entanglement problem
---
Introduction
============
Quantum theory, formalized in the first few decades of the 20$^{\mbox{th}}$century, contains elements that are radically different from the classical description of Nature. An important aspect in these fundamental differences is the existence of quantum correlations in the quantum formalism. In the classical description of Nature, if a system is formed by different subsystems, complete knowledge of the whole system implies that the sum of the information of the subsystems makes up the complete information for the whole system. This is no longer true in the quantum formalism. In the quantum world, there exist states of composite systems for which we might have the complete knowledge, while our knowledge about the subsystems might be completely random. In technical terms, one can have pure quantum states of a two-party system, whose local states are completely mixed. One may reach some paradoxical conclusions if one applies a classical description to states which have characteristic quantum signatures.
During the last two decades, it has been realized that these fundamentally nonclassical states, also denoted as “entangled states”, can provide us with something else than paradoxes. They may be *used* to perform tasks that cannot be achieved with classical states. As benchmarks of this turning point in our view of such nonclassical states, one might mention the spectacular discoveries of (entanglement-based) quantum cryptography (1991) [@Ekert], quantum dense coding (1992) [@BW], and quantum teleportation (1993) [@BBCJPW].
In this chapter, we will consider both bipartite and multipartite composite systems. We will define formally what entangled states are, present some important criteria to discriminate entangled states from separable ones, and show how they can be classified according to their capability to perform some precisely defined tasks. Our knowledge in the subject of entanglement is still far from complete, although significant progress has been made in the recent years and very active research is currently underway. We will consider multipartite quantum states (states of more than two parties) in Section \[sec:multi\], until then, we consider only bipartite quantum states.
Bipartite pure states: Schmidt decomposition
============================================
Consider a bipartite system in a shared pure state. The two parties in possession of the system are traditionally denoted as Alice (A) and Bob (B), who can be located in distant regions. Let Alice’s physical system be described by the Hilbert space $\mathcal{H}_A$ and that of Bob by $\mathcal{H}_B$. Then the joint physical system of Alice and Bob is described by the tensor product Hilbert space $\mathcal{H}_A \otimes \mathcal{H}_{B}$.
**Def. 1** *Product and entangled pure states:*\
*A pure state, i.e. a projector $|\psi_{AB}\rangle \langle \psi_{AB}|$ on a vector $|\psi_{AB}\rangle \in {\cal H}_A \otimes {\cal H}_B $, is a product state if the states of local subsystems are also pure states, that is, if $|\psi_{AB}\rangle= |\psi_A\rangle \otimes |\psi_B\rangle$. However, there are states that cannot be written in this form. These states are called entangled states.*
An example of an entangled state is the well-known singlet state ${\ensuremath{|\Psi^-\rangle}}=(\left|01\right\rangle - \left|10\right\rangle)/\sqrt{2}$, where $|0\rangle$ and $|1\rangle$ are two orthonormal states. Operationally, product states correspond to those states, that can be locally prepared by Alice and Bob at two separate locations. Entangled states can, however, be prepared only after the particles of Alice and Bob have interacted either directly or by means of an ancillary system. The second option is necessary due to the existence of the phenomenon of entanglement swapping [@ZZHE]. A very useful representation, only valid for pure bipartite states, is the, so-called, Schmidt representation:\
**Theorem 1** *Schmidt decomposition:*\
*Every $|\psi_{AB}\rangle \in {\cal H}_{A} \otimes {\cal H}_{B}$ can be represented in an appropriately chosen basis as $$|\psi_{AB}\rangle = \sum _{i=1}^{M} a_i |e_i\rangle \otimes |f_i\rangle,
\label{eq:schmidt}$$ where $|e_i\rangle$ ($|f_i\rangle$) form a part of an orthonormal basis in ${\cal H}_{A}$ (${\cal H}_{B}$), $a_i > 0$, and $\sum_{i=1}^{M} a_i^2=1$, where $M\le dim{\cal H}_{A}, dim{\cal H}_{B}$*.\
The positive numbers $a_i$ are known as the Schmidt coefficients and the vectors ${\ensuremath{|e_i\rangle}} \otimes {\ensuremath{|f_i\rangle}}$ as the Schmidt vectors of $|\psi_{AB}\rangle$. Note that product pure states correspond to those states, whose Schmidt decomposition has one and only one Schmidt coefficient. If the decomposition has more than one Schmidt coefficients, the state is entangled. Notice that the squares of the Schmidt coefficients of a pure bipartite state $|\psi_{AB}\rangle$ are the eigenvalues of both the reduced density matrices $\rho_A$ ($= \mbox{tr}_B ({\ensuremath{|\psi_{AB}\rangle}} {\ensuremath{\langle\psi_{AB}|}}) $) and $\rho_B$ ($= \mbox{tr}_A ({\ensuremath{|\psi_{AB}\rangle}} {\ensuremath{\langle\psi_{AB}|}})$) of $|\psi_{AB}\rangle$. The last fact gives us an easy method to find the Schmidt coefficients and the Schmidt vectors.
Bipartite mixed states: Separable and entangled states
======================================================
As discussed in the last section, the question whether a given pure bipartite state is separable or entangled is straightforward. One has just to check if the reduced density matrices are pure. This condition is equivalent to the fact that a bipartite pure state has a single Schmidt coefficient. The determination of separability for mixed states is much harder, and currently lacks a complete answer, even in composite systems of dimension as low as ${\mathbb{C}}^2 \otimes {\mathbb{C}}^4$.
To reach a formal definition of separable and entangled states, consider the following preparation procedure of a bipartite quantum state between Alice and Bob. Suppose that Alice prepares her physical system in the state $|e_{i}\rangle$ and Bob prepares his physical system in the state $|f_{i}\rangle$. Then, the combined state of their joint physical system is given by: $$|e_{i}\rangle \langle e_i |\otimes |f_{i}\rangle \langle f_i|.$$ We now assume that they can communicate over a classical channel (a phone line, for example). Then, whenever Alice prepares the state $|e_{i}\rangle$ ($i = 1,2, \ldots, K$), which she does with probability $p_i$, she communicates that to Bob, and correspondingly Bob prepares his system in the state $|f_{i}\rangle$ ($i = 1,2, \ldots, K$). Of course, $\sum_i p_i =1$ and $p_i \geq 0, \ \forall i$. The state that they prepare is then $$\begin{aligned}
\rho_{AB} = \sum_{i=1}^K p_i |e_i\rangle \langle e_i| \otimes |f_i\rangle \langle f_i|.
\label{eqn:sepmixed}\end{aligned}$$ **Def. 2** *Separable and entangled mixed states:*\
*A quantum state $\rho_{AB}$ is separable if and only if it can be represented as a convex combination of the product of projectors on local states as stated in Eq. (\[eqn:sepmixed\]). Otherwise, the state is said to be entangled.*
The important point to note here is that the state displayed in Eq. (\[eqn:sepmixed\]) is the most general state that Alice and Bob will be able to prepare by local quantum operations and classical communication (LOCC) [@Werner]. In LOCC protocols, two parties Alice and Bob perform local quantum operations separately in their respective Hilbert spaces and they are allowed to communicate classical information about the results of their local operations. Let us make the definition somewhat more formal.\
**Local operations and classical communication (LOCC):** Suppose Alice and Bob share a quantum state $\rho_{AB}$ defined on the Hilbert space $\mathcal{H}_A \otimes \mathcal{H}_B$. Alice performs a quantum operation on her local Hilbert space $\mathcal{H}_A$, using a complete set of complete general quantum operations $\{A_i^{(1)}\}$, satisfying $\sum_i (A_i^{(1)})^{\dagger} A_i^{(1)} = I_A$, and sends her measurement result $i$ to Bob via a classical channel. Depending on the measurement result of Alice, Bob operates a complete set of general quantum operations $\{B_{ij}^{(1)}\}$, satisfying $\sum_j (B_{ij}^{(1)})^{\dagger} B_{ij}^{(1)} = I_B$ on his part belonging to the Hilbert space $\mathcal{H}_B$. This joint operation along with the classical communication is called one-way LOCC. Furthermore, Bob can send his result $j$ to Alice, and she can choose another set of local operations $\{A_{ijk}^{(2)}\}$, satisfying $\sum_k (A_{ijk}^{(2)})^{\dagger} A_{ijk}^{(2)} = I_A$, according to Bob’s outcome. They can continue this process as long as required, and the entire operation is termed as LOCC, or two-way LOCC. The operators $I_A$ and $I_B$ are the identity operators on $\mathcal{H}_A$ and $\mathcal{H}_B$ respectively.
Entangled states cannot be prepared by two parties if only LOCC is allowed between them. To prepare such states, the physical systems must be brought together to interact [^1].
The question whether a given bipartite state is separable or not turns out to be quite complicated. Among the difficulties, we notice that for an arbitrary state $\rho_{AB}$, there is no stringent bound on the value of $K$ in Eq. (\[eqn:sepmixed\]), which is only limited by the Caratheodory theorem to be $K \leq (\dim {\cal H} )^2$ with ${\cal H} = {\cal H}_A \otimes {\cal H}_B$ (see [@Pawelbound; @karnas00]). Although the general answer to the separability problem still eludes us, there has been significant progress in recent years, and we will review some such directions in the following sections.
Operational entanglement criteria
=================================
In this section, we will introduce some operational entanglement criteria for bipartite quantum states. In particular, we will discuss the partial transposition criterion [@PeresPPT; @HorodeckiPPT], the majorization criterion [@NielsenKempe], the cross-norm or realignment criterion [@CCN1; @CCN2; @CCN3], and the covariance matrix criterion [@cmc1; @cmc2]. There exist several other criteria (see e.g. Refs. [@reduction; @Doherty; @other_criteria1; @other_criteria2; @other_criteria3; @other_criteria4]), which will not be discussed here. However note that, up to now, a necessary and sufficient criterion for detecting entanglement of an arbitrary given mixed state is still lacking.
Partial Transposition
---------------------
\[sec:PPT\] **Def. 3** *Let $\rho_{AB}$ be a bipartite density matrix, and let us express it as $$\rho_{AB} = \sum_{\substack{1 \leq i,j \leq d_A \\ 1 \leq \mu, \nu \leq d_B}} a_{ij}^{\mu\nu} (|i\rangle\langle j|)_A \otimes (|\mu\rangle\langle \nu|)_B,
\label{eq:mixed_state}$$ where $\{|i\rangle\}$ ($\{|\mu\rangle\}$) is a set of real orthonormal vectors in ${\cal H}_A$ (${\cal H}_B$), with $d_A = \dim {\cal H}_A$ and $d_B = \dim {\cal H}_B$. The partial transposition, $\rho_{AB}^{T_A}$, of $\rho_{AB}$ with respect to subsystem $A$, is defined as* $$\label{eq_partial_trans}
\rho_{AB}^{T_A} =
\sum_{\substack{1 \leq i,j \leq d_A \\ 1 \leq \mu, \nu \leq d_B}} a_{ij}^{\mu\nu} (|j\rangle\langle i|)_A \otimes (|\mu\rangle\langle \nu|)_B.$$
A similar definition exists for the partial transposition of $\rho_{AB}$ with respect to Bob’s subsystem. Notice that $\rho_{AB}^{T_{B}}\ =(\rho_{AB}^{T_{A}})^{T}$. Although the partial transposition depends upon the choice of the basis in which $\rho_{AB}$ is written, its eigenvalues are basis independent. We say that a state has positive partial transposition (PPT) , whenever $\rho_{AB}^{T_A} \geq 0$, i.e. the eigenvalues of $\rho_{AB}^{T_{A}}$ are non-negative. Otherwise, the state is said to be non-positive under partial transposition (NPT).
**Theorem 2** [@PeresPPT]\
*If a state $\rho_{AB}$ is separable, then $\rho_{AB}^{T_{A}}\ \ge\ 0$ and $\rho_{AB}^{T_{B}}\ =\left(\rho_{AB}^{T_{A}}\right)^{T}\ \ge\ 0$.*
Proof:\
Since $\rho_{AB}$ is separable, it can be written as $$\begin{aligned}
\rho_{AB} &
= \sum_{i=1}^{K}\ p_i |e_i \rangle \langle e_i |\otimes |f_i\rangle \langle f_i | \ge 0.\end{aligned}$$ Now performing the partial transposition w.r.t. A, we have $$\begin{aligned}
\rho_{AB}^{T_{A}} &=& \sum_{i=1}^{K}\ p_i\left(|e_i\rangle \langle e_i |
\right)^{T_{A}}\otimes |f_i \rangle \langle f_i |\nonumber\\
&=& \sum_{i=1}^{K}\ p_i |e_i^*\rangle \langle e_i^*| \otimes | f_i\rangle \langle f_i | \ge 0.\end{aligned}$$ Note that in the second line, we have used the fact that $A^{\dagger}=\left(A^*\right)^{T}$. $\square$
The *partial transposition criterion*, for detecting entanglement is simple: Given a bipartite state $\rho_{AB}$, find the eigenvalues of any of its partial transpositions. A negative eigenvalue immediately implies that the state is entangled. Examples of states for which the partial transposition has negative eigenvalues include the singlet state.
The partial transposition criterion allows to detect in a straightforward manner all entangled states that are NPT states. This is a huge class of states. However, it turns out that there exist PPT states which are not separable, as pointed out in Ref. [@Pawelbound] (see also [@Horodeckibound]). Moreover, the set of PPT entangled states is not a set of measure zero [@koto-volume-re]. It is, therefore, important to have further independent criteria of entanglement detection which permits to detect entangled PPT states. It is worth mentioning here that PPT states which are entangled, form the only known examples of the “bound entangled states” (see Refs. [@Horodeckibound; @biyog] for details). Bound entangled states of bipartite quantum states are the states that cannot be *distilled* i.e., converted to singlet states under LOCC [@bennett_dist; @bennett_eof2], with other entangled states being distillable. We will talk about distillation of quantum states later in this chapter in a bit more detail. Although as yet not found, it is conjectured that there also exist NPT bound entangled states [@biyog]. Note also that both separable as well as PPT states form convex sets. Figure \[fig:phase\] depicts the structure of the state space with respect to the partial transposition criteria and distillability.
![The structure of the state space in light of the partial transposition criteria and distillability. Separable and as well as PPT states form convex sets (while NPT states do not). It also shows the conjectured NPT bound entangled states. It is not clear whether the set of non-distillable states is convex.[]{data-label="fig:phase"}](phase.pdf)
Theorem 2 is a necessary condition of separability in any arbitrary dimension. However, for some special cases, the partial transposition criterion is both a necessary and a sufficient condition for separability:\
**Theorem 3** [@HorodeckiPPT]\
*In ${\mathbb{C}}^2 \otimes {\mathbb{C}}^2$ or ${\mathbb{C}}^2 \otimes {\mathbb{C}}^3$, a state $\rho$ is separable if and only if $\rho^{T_{A}}\ \ge\ 0$*.
As mentioned above, PPT bound entangled states exist. However, as the Theorem 3 shows, they can exist only in dimensions higher than $\mathbb{C}^2 \otimes \mathbb{C}^2$ and $\mathbb{C}^2 \otimes \mathbb{C}^3$.
Majorization
------------
The partial transposition criterion, although powerful, is not able to detect entanglement in a finite volume of states. It is, therefore, interesting to discuss other independent criteria. The majorization criterion, to be discussed in this subsection, has been shown to be *not* more powerful in detecting entanglement. We choose to discuss it here, mainly because it has independent roots. Moreover, it reveals a very interesting thermodynamical property of entanglement.
Before presenting the criterion, we present a definition of majorization [@majorizationBhatia].\
**Def. 4** *Let $x = (x_1, x_2, \ldots, x_d )$, and $y = (y_1, y_2, \ldots, y_d )$ be two probability distributions, arranged in decreasing order, i.e. $x_1 \geq x_2 \geq \ldots \geq x_d$ and $y_1 \geq y_2 \geq \ldots \geq y_d$. Then we define “$x$ majorized by $y$”, denoted as $x\prec y$, as* $$\sum_{i=1}^l x_i \leq \sum_{i=1}^l y_i,$$ *where $ l = 1,2, \ldots d-1$, and equality holds when $l = d$*.
**Theorem 4** [@NielsenKempe]\
*If a state $\rho_{AB}$ is separable, then* $$\label{dibakar}
\lambda (\rho_{AB}) \prec \lambda (\rho_A ), \quad and \quad \lambda (\rho_{AB}) \prec \lambda (\rho_B ),$$ *where $\lambda (\rho_{AB})$ is the set of eigenvalues of $ \rho_{AB} $, and $\lambda ( \rho_A )$ and $ \lambda ( \rho_B ) $ are the sets of eigenvalues of the reduced density matrices of the state $\rho_{AB}$, and where all the sets are arranged in decreasing order.*
The majorization criterion: Given a bipartite state, it is entangled if Eq. (\[dibakar\]) is violated. However, it was shown in Ref. [@Hiroshima], that a state that is not detected by the positive partial transposition criterion, will not be detected by the majorization criterion either. Nevertheless, the criterion has other important implications. We will now discuss one such.
Let us reiterate an interesting fact about the singlet state: The global state is pure, while the local states are completely mixed. In particular, this implies that the von Neumann entropy [^2] of the singlet is lower than the von Neumann entropies of either of the local states. Since the von Neumann entropy can be used to quantify disorder in a given state, there exist global states whose disorder is lower than the any of the local states. This is a nonclassical fact as for two classical random variables, the Shannon entropy [^3] of the joint distribution cannot be smaller than that of either. In Ref. [@horo_ent_sep], it was shown that a similar fact is true for separable states:\
**Theorem 5**\
*If a state $\rho_{AB}$ is separable, $$\label{chandrima}
S (\rho_{AB}) \geq S (\rho_A ), \quad and \quad S (\rho_{AB}) \geq S (\rho_B ).$$* Although the von Neumann entropy is an important notion for quantifying disorder, the theory of majorization is a more stringent quantifier [@majorizationBhatia]: For two probability distributions $x$ and $y$, $x \prec y$ if and only if $x = Dy$, where $D$ is a doubly stochastic matrix [^4]. Moreover, $x \prec y$ implies that $H(\{x_i\}) \geq H(\{y_i\})$. Quantum mechanics therefore allows the existence of states for which global disorder is greater than local disorder even in the sense of majorization.
A density matrix that satisfies Eq. (\[dibakar\]), automatically satisfies Eq. (\[chandrima\]). In this sense, Theorem 4 is a generalization of Theorem 5.
Cross-norm or matrix realignment
--------------------------------
The cross-norm or matrix realignment criterion [@CCN1; @CCN2; @CCN3] provides another way to delineate separable and entangled states, and more importantly, can successfully detect various PPT entangled states. There are various ways to formulate this criterion. Here we present a formulation given in Ref. [@CCN3] as Corollary 18.
A density matrix $\rho_{AB}$ on a Hilbert space $\mathcal{H}_{A} \otimes \mathcal{H}_{B}$, where $d_{A}$ and $d_{B}$ are the dimensions of the Hilbert spaces $\mathcal{H}_A$ and $\mathcal{H}_B$ respectively, can be written as $$\rho_{AB} = \sum_{\substack{1 \leq i,j \leq d_A \\ 1 \leq \mu, \nu \leq d_B}} a_{ij}^{\mu \nu} ({\ensuremath{|i\rangle}}{\ensuremath{\langlej|}})_A \otimes ({\ensuremath{|\mu\rangle}}{\ensuremath{\langle\nu|}})_B
= \sum_{k,l} \xi_{kl} \tilde{G}_{k}^{A} \otimes \tilde{G}_{l}^{B}.
\label{eq:ccn1}$$ We have used the same notations as in Eq. (\[eq:mixed\_state\]), except that we have added zeros to the tensor $a_{ij}^{\mu \nu}$, so that the indices run until the dimensions of the Hilbert spaces. $\{\tilde{G}_{k}^{A}\} = \{{\ensuremath{|i\rangle}}{\ensuremath{\langlej|}}\}$ and $\{\tilde{G}_{l}^{B}\} = \{{\ensuremath{|\mu\rangle}}{\ensuremath{\langle\nu|}}\} $ are complete sets of orthonormal Hermitian operators on the Hilbert spaces $\mathcal{H}_A$ and $\mathcal{H}_B$ respectively, with $ 1 \leq k \leq d_{A}^{2} $ and $ 1 \leq l \leq d_{B}^{2}$. Without loss of generality, we assume that $d_{A} \geq d_{B}$. After singular value decomposition of the matrix $\xi$, we have $$\xi = U \Sigma V^{\dagger},$$ where $U$ and $V$ are $ d_{A}^{2} \times d_{A}^{2} $ and $ d_{B}^{2} \times d_{B}^{2} $ dimensional unitary matrices respectively, and $ \Sigma $ is a $ d_{A}^{2} \times d_{B}^{2} $ dimensional diagonal matrix. Denoting the $k$th column vector of $ U $ and $ V $ by $ {\ensuremath{|u_k\rangle}} $ and $ {\ensuremath{|v_k\rangle}} $, the above expression becomes $$\xi = \sum_{k = 1}^{d_B^2} \lambda_{k} {\ensuremath{|u_{k}\rangle}} {\ensuremath{\langlev_{k}|}},$$ where $ \lambda_{k}$ are the diagonal elements of $\Sigma$. So, we have the matrix elements of $\xi$ as $$\xi_{kl} = \lambda_{k} \delta_{kl}.
\label{eq:ccn2}$$ If $G_k^A$ and $G_l^B$ are the matrix representations of $\tilde{G}_k^A$ and $\tilde{G}_l^B$ in $\{{\ensuremath{|u_k\rangle}}\}$ and $\{{\ensuremath{|v_k\rangle}}\}$ basis respectively, then using Eqs. (\[eq:ccn1\]) and (\[eq:ccn2\]), we obtain $$\rho_{AB} = \sum_{k = 1}^{d_B^2} \lambda_{k} G_{k}^{A} \otimes G_{k}^{B}.
\label{eq:ccn3}$$ Eq. (\[eq:ccn3\]) can be interpreted as the Schmidt decomposition of the density matrix $\rho_{AB}$ in operator space, where the singular values $\lambda_k$ are real and non-negative. The cross-norm or realignment criterion of separability is given by the following theorem:
**Theorem 6** [@CCN3]\
*If a shared quantum state $\rho_{AB}$ is separable, then $$\sum_{k} \lambda_{k} \leq 1,$$ where $\lambda_{k}$ are the singular values of $\rho_{AB}$ as given in Eq. (\[eq:ccn3\]). If the inequality is violated, one can conclude that $\rho_{AB}$ must be an entangled state.*
Covariance matrix
-----------------
There exist several other operational criteria in the literature to detect whether a quantum state is separable or entangled [@reduction; @Doherty; @other_criteria1; @other_criteria2; @other_criteria3; @other_criteria4]. We will conclude this section by briefly illustrating one such separability criteria, known as the covariance matrix criterion [@cmc1; @cmc2], which provides a general framework to link and understand several existing criteria including the cross-norm or realignment criterion. Like the cross-norm or realignment criterion, this method can identify entangled state for which the partial transposition criterion fails. Before delving into the theory of the covariance matrix criterion, let us first discus the definition and properties of the covariant matrices.
### Definition and properties
**Def. 5** *Given a quantum state $\rho$ and a complete set of orthonormal observables $\{M_k\}$ on a $d$-dimensional Hilbert space, the $d^2 \times d^2$ covariant matrix $\gamma = \gamma(\rho,\{M_k\})$ and symmetrized covariant matrix $\gamma^S = \gamma^S(\rho,\{M_k\})$ are defined as $$\begin{aligned}
\gamma_{ij} &=& \langle M_i M_j \rangle - \langle M_i \rangle \langle M_j \rangle, \\
\gamma^S_{ij} &=& \frac{\langle M_i M_j \rangle + \langle M_j M_i \rangle}{2} - \langle M_i \rangle \langle M_j \rangle,\end{aligned}$$ where $\langle M \rangle = \mbox{tr}(\rho M)$ defines the expectation of the operator $M$ with respect to the state $\rho$.*
The complete set of orthonormal observables $\{M_k\}$ has to satisfy the Hilbert-Schmidt orthonormality condition $\mbox{tr}(M_i M_j) = \delta_{ij},$ $\forall i,j = 1,2,...,d$. One example for such a set of observables for the case of single qubit in terms of the Pauli matrices, can be given by $$M_1 = \frac{I}{\sqrt{2}}, \ M_2 = \frac{\sigma_x}{\sqrt{2}}, \ M_3 = \frac{\sigma_y}{\sqrt{2}},
\ M_4 = \frac{\sigma_z}{\sqrt{2}}.$$ In general, for the $d$-dimensional case, one can consider following matrices to form the complete set of orthonormal observables: $$\begin{aligned}
X_i &=& {\ensuremath{|i\rangle}}{\ensuremath{\langlei|}}, \ \ i = 1,2,...,d, \\
Y_{ij} &=& \frac{1}{\sqrt{2}}({\ensuremath{|i\rangle}}{\ensuremath{\langlej|}} + {\ensuremath{|j\rangle}}{\ensuremath{\langlei|}}), \ \ 1 \leq i < j \leq d, \\
Z_{kl} &=& \frac{i}{\sqrt{2}}({\ensuremath{|k\rangle}}{\ensuremath{\langlel|}} - {\ensuremath{|l\rangle}}{\ensuremath{\langlek|}}), \ \ 1 \leq k < l \leq d.\end{aligned}$$
Let us now focus on the situation in which the Hilbert space is a tensor product $\mathcal{H}=\mathcal{H}_A \otimes \mathcal{H}_B$ of Hilbert spaces of two subsystems $A$ and $B$ with dimensions $d_A$ and $d_B$ respectively. We can consider the complete set of orthonormal observables in $A$ as $\{ A_k: k = 1,2...,d_A^2\}$ and in $B$ as $\{ B_k: k = 1,2...,d_B^2 \}$, and construct a set of $d^2_A + d^2_B$ observables as $\{ M_k \} =\{ A_k \otimes I, I \otimes B_k\}$. Although this set is not complete, it can be utilized to define a very useful form of covariant matrices, known as the block covariant matrices. The block covariant matrix for a given bipartite state $\rho_{AB}$ and orthonormal observables $\{M_k\}$ is defined as follows.
**Def. 6** *Let $\rho_{AB}$ be a quantum state of a bipartite system, and let $\{ M_k \} =\{ A_k \otimes I, I \otimes B_k\}$ be a set of orthonormal observables as stated above. Then, the block covariance matrix $\gamma = \gamma(\rho_{AB},\{M_k\})$ is given in terms of its matrix elements as $\gamma_{i,j} = \langle M_i M_j \rangle - \langle M_i \rangle \langle M_j \rangle$, and has the block structure $$\gamma =
\begin{pmatrix}
\mathcal{A} && \mathcal{X} \\
\mathcal{X}^{T} && \mathcal{B}
\end{pmatrix},
\label{eq:bcmc}$$ where $\mathcal{A} = \gamma(\rho_A,\{A_k\})$ and $\mathcal{B} = \gamma(\rho_B,\{B_k\})$ are the covariant matrices of the reduced subsystems, and $\mathcal{X}_{i,j} = \langle A_i \otimes B_j \rangle - \langle A_i \rangle \langle B_j \rangle$.*
Similarly, we can define the symmetric version of the block covariance matrix by replacing $\mathcal{A}$ and $\mathcal{B}$ with their symmetrized counterparts, while keeping $\mathcal{X}$ unchanged. Clearly, if $\rho_{AB} = \rho_A \otimes \rho_B$ is a product state, then its block covariant matrix reduces to the block diagonal form, $\gamma(\rho_{AB}, \{M_k\}) = \mathcal{A} \oplus \mathcal{B}$, as $\mathcal{X}_{i,j}$ become zero $\forall i,j$.
If $\rho$ is a pure state on a $d$-dimensional Hilbert space, then the corresponding covariance matrix $\gamma$ satisfies following properties:
1. The rank of $\gamma$ is equal to $d-1$.
2. The nonzero eigenvalues of $\gamma$ are equal to 1, hence $\mbox{tr}(\gamma)=d-1$.
3. $\gamma^2 = \gamma$.
The corresponding symmetric covariance matrix $\gamma^S$ satisfies the following:
1. The rank of $\gamma^S$ is equal to $2(d-1)$.
2. The nonzero eigenvalues of $\gamma^S$ are equal to 1/2, hence $\mbox{tr}(\gamma^S)=d-1$.
For mixed state $\rho$ on a $d$-dimensional Hilbert space, we have $\mbox{tr}[\gamma(\rho)] = d - \mbox(\rho^2)$, and the same for $\gamma^S(\rho)$. The covariance matrix (symmetric and non-symmetric) also satisfies the concavity property, i.e., if $\rho = \sum_k p_k \rho_k$ is a convex combination of states $\rho_k$, then $$\gamma(\rho) \geq \sum_k p_k \gamma(\rho_k).$$
### Covariance matrix criterion for separability
**Theorem 7**[@cmc1; @cmc2]\
*Let $\rho_{AB}$ be a separable state on $\mathcal{H}_A \otimes \mathcal{H}_B$ and $A_k$ and $B_k$ be the orthonormal observables on the Hilbert spaces $\mathcal{H}_A$ and $\mathcal{H}_B$, with the latter having dimensions $d_A$ and $d_B$ respectively. Define $\{ M_k \} =\{ A_k \otimes I, I \otimes B_k\}$ as discussed previously. Then there exist pure states ${\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}}$ for $A$, ${\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}$ for $B$, and convex weights $p_k$ such that if we define $\kappa_A = \sum_k p_k \gamma({\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}}, \{ A_{k'} \})$ and $\kappa_B = \sum_k p_k \gamma({\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}, \{ B_{k'} \})$, the inequality $$\gamma(\rho_{AB},\{M_k\}) \geq \kappa_A \oplus \kappa_B \Leftrightarrow
\begin{pmatrix}
\mathcal{A} && \mathcal{X} \\
\mathcal{X}^T && \mathcal{B}
\end{pmatrix}
\geq
\begin{pmatrix}
\kappa_A && 0 \\
0 && \kappa_B
\end{pmatrix}$$ holds.*\
Proof:\
Let us assume $\rho_{AB}$ be a separable state with the following pure state decomposition, $$\rho_{AB} = \sum_k p_k ({\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}} \otimes {\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}).$$ Then using the properties of covariance matrices, we get $$\begin{aligned}
\gamma(\rho_{AB},\{M_k\}) &=& \gamma(\sum_k p_k ({\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}} \otimes {\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}), \{M_{k'}\}) \nonumber \\
&\geq& \sum_k p_k \gamma({\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}} \otimes {\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}, \{M_{k'}\}) \nonumber \\
&=& \sum_k p_k \{\gamma({\ensuremath{|\psi_k\rangle}}{\ensuremath{\langle\psi_k|}}, \{A_{k'}\}) \oplus \gamma({\ensuremath{|\phi_k\rangle}}{\ensuremath{\langle\phi_k|}}, \{B_{k''}\})\} \nonumber \\
&=& \kappa_A \oplus \kappa_B,\end{aligned}$$ where we have used concavity property of the covariance matrices in the second line, and in the third line, we have used the fact that the block covariant matrix of a product state takes the block diagonal form. This theorem can also be proven for the symmetric covariant matrices in the same manner. Clearly, if there exist no such $\kappa_A$ and $\kappa_B$, the state $\rho_{AB}$ must be entangled.$\square$
Clearly, it is not evident from Theorem 7 that the covariance matrix criterion leads to an efficient and physically plausible operational indication for separability. The main problem is to identify possible $\kappa_A$ and $\kappa_B$, as this requires an optimization over all pure state decompositions of $\rho_{AB}$. Therefore, we now focus on the cases, where the above criterion can be used efficiently to identify entangled states, by stating several corollaries of the above theorem.
**Corollary 1**\
*Let $
\gamma =
\begin{pmatrix}
\mathcal{A} && \mathcal{X} \\
\mathcal{X}^{T} && \mathcal{B}
\end{pmatrix}
$ be the block covariance matrix of a bipartite state $\rho_{AB}$. Then, if $\rho_{AB}$ is separable, we have* $$||\mathcal{X}||^2 \leq [1 - \mbox{tr}(\rho_A^2)][1 - \mbox{tr}(\rho_B^2)],$$ *where* $||A|| = \mbox{tr} \ \sqrt{A^{\dagger}A}$ *is the matrix trace norm.*
**Corollary 2**\
*Let $\rho_{AB}$ be a quantum state shared between two subsystems $A$ and $B$ with dimensions $d_A$ and $d_B$ respectively (with $d_A \leq d_B$), $
\gamma^S =
\begin{pmatrix}
\mathcal{A} && \mathcal{X} \\
\mathcal{X}^{T} && \mathcal{B}
\end{pmatrix}
$ be the corresponding symmetric covariance matrix, and let $\mathcal{J} = \{j_1,...,j_{d_A^2}\} \left(\subset \{1,...,d_B^2\}\right)$ be a set of $d_A^2$ distinct indices. Then if $\rho_{AB}$ is separable, we have* $$2 \sum_{i=1}^{d_A^2} \sum_{j \in \mathcal{J}} |\mathcal{X}_{i,j}| \leq [1 - \mbox{tr}(\rho_A^2)] + [1 - \mbox{tr}(\rho_B^2)].$$
Now we will give another operational entanglement criterion based on the Schmidt decomposition on operator space, then try to relate covariance matrix criterion with the cross-norm or realignment criterion. A general bipartite quantum state $\rho_{AB}$ on a Hilbert space $\mathcal{H}_{A} \otimes \mathcal{H}_{B}$, where $d_{A}$ and $d_{B}$ are the dimensions of the Hilbert spaces $\mathcal{H}_A$ and $\mathcal{H}_B$ respectively, can be written as $$\rho_{AB} = \sum_{k=1}^{d_A^2}\sum_{l=1}^{d_B^2} \xi_{kl} \tilde{G}^A_k \otimes \tilde{G}^B_l,
\label{eq:sd_svd}$$ where $\xi_{kl}$ are real quantities, and $\{ \tilde{G}^A_i \}$ and $\{ \tilde{G}^B_j \}$ are complete sets of orthonormal Hermitian operators on the Hilbert spaces $\mathcal{H}_A$ and $\mathcal{H}_B$ respectively. As we have seen earlier, $\rho_{AB}$ in Eq. (\[eq:sd\_svd\]), can be written in the Schmidt decomposed-like form (in operator space) via the singular value decomposition as $$\rho_{AB} = \sum_{k=1}^{d_B^2} \lambda_k G^A_k \otimes G^B_k,$$ where singular values $\lambda_k$ are real and non-negative, and we have assumed that $d_A \geq d_B$.
**Corollary 3**\
*Let $\rho_{AB}$ be a separable quantum state shared between two subsystems $A$ and $B$. Then $$2\sum_i |\lambda_i - \lambda_i^2 g_i^A g_i^B| \leq 2 - \sum_i \lambda_i^2[(g_i^A)^2 + (g_i^B)^2],
\label{eq:col3}$$ where $g_i^{A(B)} = \mbox{tr} [ G_i^{A(B)}]$.*\
Proof:\
Let $
\gamma^S =
\begin{pmatrix}
\mathcal{A} && \mathcal{X} \\
\mathcal{X}^{T} && \mathcal{B}
\end{pmatrix}
$ be the symmetric block covariance matrix of the bipartite state $\rho_{AB}$. Using the orthogonality of the matrices $\{G_i^A\}$ and $\{G_i^B\}$, one can deduce $\mathcal{X}_{i,j} = \left(\lambda_i - \lambda_i^2 g_i^A g_i^B\right)$. Further, one can get $\mbox{tr}(\rho_A^2) = \sum_i \lambda_i^2 (g_i^B)^2$, and $\mbox{tr}(\rho_B^2) = \sum_i \lambda_i^2 (g_i^A)^2$. Together with Corollary 2, we can prove the claim. $\square$
Now using the relations $|a-b| \geq |a| - |b|$ and $a^2 + b^2 \geq 2|ab|$ we can have, $$\begin{aligned}
2\sum_i |\lambda_i - \lambda_i^2 g_i^A g_i^B| &\geq & 2\sum_i (\lambda_i - \lambda_i^2 |g_i^A g_i^B|), \\
2(1 - \sum_i \lambda_i^2 |g_i^A g_i^B|) &\geq & 2 - \sum_i \lambda_i^2[(g_i^A)^2 + (g_i^B)^2].
\label{eq:col32}\end{aligned}$$ Using inequalities (\[eq:col3\])-(\[eq:col32\]), we get $$\begin{aligned}
2\sum_i |\lambda_i - \lambda_i^2 g_i^A g_i^B| &\leq & 2 - \sum_i \lambda_i^2[(g_i^A)^2 + (g_i^B)^2] \nonumber \\
&\leq & 2(1 - \sum_i \lambda_i^2 |g_i^A g_i^B|) \nonumber \\
&=& (2 - 2\sum_i \lambda_i) + 2\sum_i (\lambda_i - \lambda_i^2 |g_i^A g_i^B|) \nonumber \\
&\leq & (2 - 2\sum_i \lambda_i) + 2\sum_i |\lambda_i - \lambda_i^2 g_i^A g_i^B| \nonumber \\
\Leftrightarrow \sum_i \lambda_i &\leq & 1.\end{aligned}$$ This is the cross-norm or realignment criterion of separability mentioned earlier, which we get as a corollary of the covariance matrix criterion.
There are several other corollaries of the covariance matrix criterion, which enable one to efficiently detect entangled states in several cases. Moreover, the covariance matrix criterion can be improved by using local filtering operation [@local_filter]. See Ref. [@cmc2] for details.
Non-operational entanglement criteria
=====================================
In this section, we will discuss three further entanglement criteria. We will show how the Hahn-Banach theorem can be used to obtain “entanglement witnesses”. We will also introduce the notion of positive maps, and present the entanglement criterion based on it. And finally we will present the range criterion of separability. All three criteria are “non-operational”, in the sense that they are not state-independent. Nevertheless, they provide important insight into the structure of the set of entangled states. Moreover, the concept of entanglement witnesses can be used to detect entanglement experimentally, by performing only a few *local* measurements, assuming some prior knowledge of the density matrix [@sakkhi12; @sakkhiprl].
### Technical Preface
The following lemma and observation will be useful for later purposes.\
**Lemma 1**\
$\mbox{tr}(\rho_{AB}^{T_{A}}\sigma_{AB}) = \mbox{tr}(\rho_{AB}\sigma_{AB}^{T_{A}})$. **Observation:**\
The space of linear operators acting on ${\cal H}$ (denoted by ${\cal B}({\cal H})$) is itself a Hilbert space, with the (Euclidean) scalar product $$\langle A|B\rangle = \mbox{tr}(A^\dagger B) \qquad A,B \in {\cal B}({\cal H}).$$ This scalar product is equivalent to writing $A$ and $B$ row-wise as vectors, and scalar-multiplying them: $$\mbox{tr}(A^\dagger B) = \sum_{ij}A^\ast_{ij}B_{ij} = \sum_{k=1}^{(\dim {\cal H})^2} a_{k}^\ast b_k.$$
Entanglement Witnesses
----------------------
### Entanglement Witness from the Hahn-Banach theorem
Central to the concept of entanglement witnesses, is the Hahn-Banach theorem, which we will present here limited to our situation and without proof (see e.g. [@alt:1985] for a proof of the more general theorem):\
**Theorem 8**\
*Let $S$ be a convex compact set in a finite dimensional Banach space. Let $\rho$ be a point in the space with $\rho \not\in S$. Then there exists a hyperplane*[^5] *that separates $\rho$ from $S$.*
![Schematic picture of the Hahn-Banach theorem. The (unique) unit vector orthonormal to the hyperplane can be used to define *right* and *left* in respect to the hyperplane by using the signum of the scalar product.[]{data-label="fig:hyper"}](hyperplane.pdf)
The statement of the theorem is illustrated in figure \[fig:hyper\]. The figure motivates the introduction of a new coordinate system located within the hyperplane (supplemented by an orthogonal vector $W$ which is chosen such that it points away from $S$). Using this coordinate system, every state $\rho$ can be characterized by its distance from the plane, by projecting $\rho$ onto the chosen orthonormal vector and using the trace as scalar product, i.e. $\mbox{tr}(W\rho)$. This measure is either positive, zero, or negative. We now suppose that $S$ is the convex compact set of all separable states. According to our choice of basis in figure \[fig:hyper\], every separable state has a positive distance while there are some entangled states with a negative distance. More formally this can be phrased as:\
**Def. 7** *A Hermitian operator (an observable) $W$ is called an entanglement witness (EW) if and only if* $$\exists \ \rho \quad \mbox{such that} \quad \mbox{tr}(W \rho) < 0, \qquad
\mbox{while} \quad \forall \sigma \in S, \quad \mbox{tr}(W\sigma) \geq 0.$$ **Def. 8** *An EW is decomposable if and only if there exists operators $P$, $Q$ with* $$W = P + Q^{T_{A}}, \qquad P,Q \geq 0.$$ **Lemma 2**\
*Decomposable EW cannot detect PPT entangled states.* Proof:\
Let $\delta$ be a PPT entangled state and $W$ be a decomposable EW. Then $$\label{eqn:dEWnd}
\mbox{tr}(W\delta) = \mbox{tr}(P\delta) + \mbox{tr}(Q^{T_{A}} \delta)
= \mbox{tr}(P\delta)+\mbox{tr}(Q\delta^{T_{A}}) \geq 0.$$ Here we used Lemma 1.$\square$
**Def. 9** *An EW is called non-decomposable entanglement witness (nd-EW) if and only if there exists at least one PPT entangled state which is detected by the witness.* Using these definitions, we can restate the consequences of the Hahn-Banach theorem in several ways:\
**Theorem 9** [@Woronowicz; @HorodeckiPPT; @Terhal; @sakkhi-ager]
1. $\rho$ *is entangled if and only if, $\exists$ a witness $W$, such that* $\mbox{tr}(\rho W) < 0$.
2. $\rho$ *is a PPT entangled state if and only if $\exists$ a nd-EW, $W$, such that* $\mbox{tr}(\rho W) < 0$.
3. $\sigma$ *is separable if and only if $\forall$ EW,* $\mbox{tr}(W\sigma) \geq 0$.
From a theoretical point of view, the theorem is quite powerful. However, it does not give any insight of how to construct for a given state $\rho$, the appropriate witness operator.
![Schematic view of the Hilbert-space with two states $\rho_1$ and $\rho_2$ and two witnesses $EW1$ and $EW2$. $EW1$ is a decomposable EW, and it detects only NPT states like $\rho_1$. $EW2$ is an nd-EW, and it detects also some PPT states like $\rho_2$. Note that neither witness detects *all* entangled states.[]{data-label="fig:schhil"}](ew_vgl.pdf)
### Examples {#sec:E}
For a decomposable witness $$W' = P + Q^{T_{A}},$$ $$\label{narayani}
\mbox{tr}(W'\sigma) \geq 0,$$ for all separable states $\sigma$.
Proof:\
If $\sigma$ is separable, then it can be written as a convex sum of product vectors. So if Eq. (\[narayani\]) holds for any product vector $|e,f\rangle$, any separable state will also satisfy the same. $$\begin{aligned}
\mbox{tr}(W' |e,f\rangle \langle e,f|) &=&
\langle e,f |W' |e,f\rangle \nonumber \\
&=&
\underbrace{\langle e,f | P |e,f \rangle}_{\geq 0} + \underbrace{\langle e,f | Q^{T_{A}} |e,f \rangle }_{\geq 0} , \end{aligned}$$ because $$\langle e,f | Q^{T_{A}} |e,f \rangle = \mbox{tr} (Q^{T_{A}} |e,f\rangle \langle e,f |) = \mbox{tr} (Q |e^\ast,f \rangle \langle e^\ast,f | )
\geq 0.$$ Here we used Lemma 1, and $P,Q \geq 0$. $\square$
This argumentation shows that $W=Q^{T_{A}}$ is a suitable witness also. Let us now consider the simplest case of ${\mathbb{C}}^2 \otimes {\mathbb{C}}^2$. We can use $$|\Phi^+ \rangle = \frac{1}{\sqrt{2}}\left(|00\rangle + |11\rangle\right),$$ to write the density matrix $$Q = \left( \begin{array}{cccc}
\frac12 & 0 & 0 & \frac12 \\
0 & 0 & 0 & 0 \\
0 & 0 & 0 & 0 \\
\frac12 & 0 & 0 & \frac12 \end{array}\right). \quad \mbox{Then} \quad
Q^{T_{A}} = \left( \begin{array}{cccc}
\frac12 & 0 & 0 & 0 \\
0 & 0 & \frac12 & 0 \\
0 & \frac12 & 0 & 0 \\
0 & 0 & 0 & \frac12 \end{array}\right).$$ One can quickly verify that indeed $W=Q^{T_{A}}$ fulfills the witness requirements. Using $$|\Psi^- \rangle = \frac{1}{\sqrt{2}} \left(|01 \rangle - |10 \rangle\right),$$ we can rewrite the witness: $$\label{eqn:exwit}
W = Q^{T_{A}} = \frac{1}{2} \left(I_4 - 2 |\Psi^- \rangle \langle \Psi^- |\right),$$ where $I_4$ denotes the identity operator on $\mathbb{C}^2 \otimes \mathbb{C}^2$. This witness now detects $|\Psi^- \rangle$: $$\mbox{tr} (W |\Psi^- \rangle \langle \Psi^- |) = -\frac{1}{2}.$$
Positive maps
-------------
### Introduction and definitions
So far we have only considered states belonging to a Hilbert space ${\cal H}$, and operators acting on the Hilbert space. However, the space of operators ${\cal B}({\cal H})$ has also a Hilbert space structure. We now look at transformations of operators, the so-called maps which can be regarded as *superoperators*. As we will see, this will lead us to an important characterization of entangled and separable states. We start by defining linear maps:\
**Def. 10** *A linear, self-adjoint map $\epsilon$ is a transformation* $$\begin{aligned}
\epsilon: {\cal B} ({\cal H}_{B}) \rightarrow {\cal B} ({\cal H}_{C}),\end{aligned}$$ *which*
- *is linear, i.e.* $$\begin{aligned}
\epsilon(\alpha O_1+\beta O_2) &= \alpha \epsilon(O_1) +
\beta \epsilon(O_2)\quad\forall O_1,\,O_2\in {\cal B}({\cal H}_{B}),
$$ *where $\alpha$, $\beta$ are complex numbers,*
- *and maps Hermitian operators onto Hermitian operators, i.e.* $$\begin{aligned}
\epsilon(O^{\dagger})&=\left(\epsilon(O)\right)^{\dagger}\qquad\forall O \in {\cal B}({\cal H}_{B}).\end{aligned}$$
For brevity, we will only write “linear map”, instead of “linear self adjoint map”. The following definitions help to further characterize linear maps.\
**Def. 11** *A linear map $\epsilon$ is called trace preserving if* $$\begin{aligned}
\mbox{tr}(\epsilon(O))=\mbox{tr}(O)\quad\forall O\in {\cal B}({\cal H}_{B}).\end{aligned}$$
**Def. 12** *Positive map:*\
*A linear, self-adjoint map $\epsilon$ is called positive if* $$\begin{aligned}
\forall \rho \in {\cal B}({\cal H}_{B}), \quad
\rho \geq 0 \quad \Rightarrow \quad
\epsilon(\rho) \geq 0.\end{aligned}$$ Positive maps have, therefore, the property of mapping positive operators onto positive operators. It turns out that by considering maps that are a tensor product of a positive operator acting on subsystem A, and the identity acting on subsystem B, one can learn about the properties of the composite system.
**Def. 13** *Completely positive map:*\
*A positive linear map $\epsilon$ is completely positive if for any tensor extension of the form* $$\epsilon' = {\cal I}_A\otimes\epsilon,$$ where $$\epsilon' : {\cal B}({\cal H}_{A} \otimes {\cal H}_{B})
\rightarrow {\cal B}({\cal H}_{A} \otimes {\cal H}_{C}),$$ *$\epsilon'$ is positive. Here ${\cal I}_A$ is the identity map on ${\cal B}({\cal H}_{A})$.*
**Example: Hamiltonian evolution of a quantum system.** Let $O\in {\cal B}({\cal H}_B)$ and $U$ an unitary matrix and let us define $\epsilon$ by $$\begin{aligned}
\epsilon: {\cal B}( {\cal H}_A) & \rightarrow & {\cal B}( {\cal H}_A) \nonumber \\
\epsilon(O) & = & UOU^{\dagger}.\end{aligned}$$ As an example for this map, consider the time-evolution of a density matrix. It can be written as $\rho(t)=U(t)\rho(0)U^{\dagger}(t)$, i.e. in the form given above. Clearly this map is linear, self-adjoint, positive and trace-preserving. It is also completely positive, because for $0\leq w\in {\cal B}({\cal H}_A \otimes {\cal H}_A)$, $$({\cal I}_A\otimes\epsilon)w = (I_A\otimes U)w(I_A\otimes U^{\dagger})
=\tilde{U}w\tilde{U}^{\dagger},$$ where $\tilde{U}$ is unitary. But then $\langle \psi |\tilde{U}w\tilde{U}^{\dagger} |\psi \rangle \geq 0$, if and only if $\langle \psi | w |\psi \rangle \geq 0$ (since positivity is not changed by unitary evolution).
**Example: Transposition.** An example of a positive but not completely positive map is the transposition $T$ defined as: $$\begin{aligned}
T: {\cal B}({\cal H}_B) & \rightarrow & {\cal B}({\cal H}_B) \nonumber \\
T(\rho) & = & \rho^{T}.\end{aligned}$$ Of course this map is positive, but it is not completely positive, because $$({\cal I}_A\otimes T)w=w^{T_{B}},$$ and we know that there exist states for which $\rho\geq 0$, but $\rho^{T_{B}}\not\geq 0$.
**Def. 14** *A positive map is called decomposable if and only if it can be written as* $$\epsilon=\epsilon_1+\epsilon_2 T$$ *where $\epsilon_1$, $\epsilon_2$ are completely positive maps and $T$ is the operation of transposition.*
### Positive maps and entangled states
Partial transposition can be regarded as a particular case of a map that is positive but not completely positive. We have already seen that this particular positive but not completely positive map gives us a way to discriminate entangled states from separable states. The theory of positive maps provides with stronger conditions for separability, as shown in Ref. [@HorodeckiPPT].\
**Theorem 10**\
*A state $\rho\in {\cal B} ({\cal H}_A \otimes {\cal H}_B)$ is separable if and only if for all positive maps* $$\epsilon: {\cal B} ({\cal H}_B) \rightarrow {\cal B} ({\cal H}_C),$$ *we have* $$({\cal I}_A\otimes\epsilon)\rho\geq 0.$$
Proof:
\[$\Rightarrow$\] As $\rho$ is separable, we can write it as $$\rho
=\sum_{k=1}^P p_k |e_k \rangle \langle e_k | \otimes |f_k \rangle \langle f_k |,
$$ for some $P>0$. On this state, $({\cal I}_A\otimes\epsilon)$ acts as $$({\cal I}_A\otimes\epsilon)\rho
=\sum_{k=1}^P p_k |e_k \rangle \langle e_k | \otimes \epsilon \left(|f_k \rangle \langle f_k |\right)
\geq 0,$$ where the last $\geq$ follows because $|f_k \rangle \langle f_k |\geq0$, and $\epsilon$ is positive.\
$[\Leftarrow]$ The proof in this direction is not as easy as the only if direction. We shall prove it at the end of this section.\
Theorem 10 can also be recast into the following form:\
**Theorem 11** [@HorodeckiPPT]\
*A state $\rho\in {\cal B} ({\cal H}_A \otimes {\cal H}_B)$ is entangled if and only if there exists a positive map $\epsilon: {\cal B} ({\cal H}_B) \rightarrow {\cal B} ({\cal H}_C)$, such that* $$\label{chot-bogoley}
({\cal I}_A\otimes\epsilon)\rho\not\geq 0.$$ Note that Eq. (\[chot-bogoley\]) can never hold for maps, $\epsilon$, that are completely positive, and for non-positive maps, it may hold even for separable states. Hence, any positive but not completely positive map can be used to detect entanglement.
### Choi-Jamio[ł]{}kowski Isomorphism
In order to complete the proof of Theorem 10, we introduce first the Choi-Jamio[ł]{}kowski isomorphism [@jamiolkowski] between operators and maps. Given an operator $E\in {\cal B} ({\cal H}_B \otimes {\cal H}_C)$, and an orthonormal product basis $|k,l \rangle$, we define a map by $$\begin{aligned}
\epsilon: {\cal B} ({\cal H}_B) &\rightarrow & {\cal B} ({\cal H}_C) \nonumber \\
\epsilon(\rho) & = & \sum_{k_1,l_1,k_2,l_2}\,{}_{BC}\langle k_1l_1| E | k_2l_2 \rangle_{BC} \quad
|l_1\rangle_{CB} \langle k_1 | \rho |k_2\rangle_{BC} \langle l_2|,\end{aligned}$$ or in short form, $$\epsilon(\rho)= \mbox{tr}_{B}(E\rho^{T_{B}}).$$ This shows how to construct the map $\epsilon$ from a given operator $E$. To construct an operator from a given map we use the state $$|\Psi^+ \rangle =\frac{1}{\sqrt{M}}\sum_{i=1}^{M} |i\rangle_{B'} |i\rangle_{B}$$ (where $M=\dim {\cal H}_B$) to get $$M\left(I_{B'}\otimes\epsilon\right)\left( |\Psi^+\rangle \langle \Psi^+| \right)=E.$$ This isomorphism between maps and operators results in the following properties:\
**Theorem 12** [@jamiolkowski; @Woronowicz; @HorodeckiPPT; @Terhal; @sakkhi-ager] To indicate further how this equivalence between maps and operators works, we develop here a proof for the “only if" direction of the second statement. Let $E \in {\cal B}({\cal H}_B \otimes {\cal H}_C)$ be an entanglement witness, then $\langle e,f| E |e,f \rangle \geq 0$. By the Jamio[ł]{}kowski isomorphism, the corresponding map is defined as $\epsilon(\rho)=\mbox{tr}_{B}(E\rho^{T_{B}})$ where $\rho \in {\cal B}({\cal H}_B)$. We have to show that $${}_C \langle \phi | \epsilon(\rho) |\phi\rangle_C={}_C \langle \phi| \mbox{tr}(E\rho^{T_{B}}) |\phi \rangle_C \geq 0\qquad
\forall |\phi\rangle_C \in {\cal H}_C.$$ Since $\rho$ acts on Bob’s space, using the spectral decomposition of $\rho$, $\rho=\sum_i\lambda_i |\psi_i \rangle \langle \psi_i| $, leads to $$\rho^{T_{B}}=\sum_i\lambda_i
|\psi_i^\ast \rangle \langle \psi_i^\ast |,$$ where all $\lambda_i\ge 0$. Then $$\begin{aligned}
{}_C\langle\phi | \epsilon(\rho)|\phi\rangle_C &=&
{}_C\langle\phi | \sum_i \mbox{tr}_B (E\lambda_i |\psi_i^\ast\rangle_B {}_B \langle \psi_i^\ast |) |\phi\rangle_C \nonumber\\
&=& \sum_i\lambda_i {}_{BC} \langle \psi_i^\ast,\phi | E |\psi_i^\ast,\phi \rangle_{BC} \geq 0.\end{aligned}$$ $\square$ We can now proof the $\Leftarrow$ direction of Theorem 10 or, equivalently, the $\Rightarrow$ direction of Theorem 11. We thus have to show that if $\rho_{AB}$ is entangled, there exists a positive map $\epsilon: {\cal B}({\cal H}_A) \rightarrow {\cal B}({\cal H}_C)$, such that $\left(\epsilon\otimes {\cal I}_B\right)\rho$ is not positive definite. If $\rho$ is entangled, then there exists an entanglement witness $W_{AB}$ such that $$\begin{aligned}
\mbox{tr}(W_{AB}\rho_{AB}) < 0, \quad \mbox{and} \nonumber \\
\mbox{tr}(W_{AB}\sigma_{AB}) \geq 0,\end{aligned}$$ for all separable $\sigma_{AB}$. $W_{AB}$ is an entanglement witness (which detects $\rho_{AB}$) if and only if $W_{AB}^{T}$ (note the complete transposition!) is also an entanglement witness (which detects $\rho_{AB}^{T}$). We define a map by $$\begin{aligned}
\epsilon: {\cal B}({\cal H}_A) & \rightarrow & {\cal B}({\cal H}_C),\\
\epsilon(\rho) & = & \mbox{tr}_{A} (W^{T}_{AC}\rho^{T_{A}}_{AB}),\end{aligned}$$ where $\dim {\cal H}_C=\dim {\cal H}_B = M$. Then $$(\epsilon\otimes {\cal I}_B)(\rho_{AB})= \mbox{tr}_{A}(W_{AC}^{T}\rho_{AB}^{T_{A}})=\mbox{tr}_{A}(W_{AC}^{T_{C}}\rho_{AB})=
\tilde{\rho}_{CB},$$ where we have used Lemma 1, and that $T=T_{A} \circ T_{C}$. To complete the proof, one has to show that $\tilde{\rho}_{CB}\not\ge 0$, which can be done by showing that ${}_{CB}\langle \psi^+| \tilde{\rho}_{CB} |\psi^+\rangle_{CB} < 0$, where $|\psi^+ \rangle_{CB}=\frac{1}{\sqrt{M}} \sum_i |ii\rangle_{CB}$, with $\{|i\rangle\}$ being an orthonormal basis. $\square$
Range criterion
---------------
The range criterion [@Pawelbound] gives a non-operational condition for separability, which is based on the range of the density matrix and is, in particular, independent of the partial transposition criterion. The range criterion may not detect inseparability in some states for which the partial transposition criterion succeeds but it works efficiently in many cases, especially for the bound entangled states, where the other one fails.
**Def. 15** *Range of a matrix $M$ on a Hilbert space $\mathcal{H}$ is defined as the span (i.e., the set of all possible linear combinations) of its column vectors. Alternatively, it can be defined as $\mathcal{R}(M) \equiv \{{\ensuremath{|\psi\rangle}} \in \mathcal{H} \ | \ M{\ensuremath{|\phi\rangle}}= {\ensuremath{|\psi\rangle}} \mbox{ for some } {\ensuremath{|\phi\rangle}} \in \mathcal{H}\}$.*
It can be easily shown that for a density matrix $\rho$ in a Hilbert space $\mathcal{H}$ having spectral decomposition $$\rho = \sum_i p_i {\ensuremath{|\psi_i\rangle}}{\ensuremath{\langle\psi_i|}},$$where $p_i > 0, \forall i$, the set of vectors $\{{\ensuremath{|\psi_i\rangle}}\}$ spans the range of $\rho$, $\mathcal{R}(\rho)$. The range criteria of separability is given by the following theorem.
**Theorem 13** [@Pawelbound]\
\
Proof:\
If $\rho_{AB}$ is a separable state on the Hilbert space $\mathcal{H} = \mathcal{H}_A \otimes \mathcal{H}_B$, it can be written as a convex combination of $N \leq d_A^2 d_B^2$ products of projectors $P_{\psi_i} \otimes Q_{\phi_k}$ as $$\rho_{AB} = \sum_{i,k} p_{ik} P_{\psi_i} \otimes Q_{\phi_k} = \sum_{i,k} p_{ik} \left({\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi_{k}\rangle}}\right) \left({\ensuremath{\langle\psi_{i}|}} \otimes {\ensuremath{\langle\phi_{k}|}}\right),
\label{eq:range1}$$ where $p_{ik}$ are non-zero probabilities satisfying $\sum_{i,k}p_{ik} = 1$. Now the transposition of a Hermitian operator is simply equivalent to the complex conjugation of its matrix elements, i.e., $Q_{\phi_k}^T = Q_{\phi_k}^* = {\ensuremath{|\phi^*_k\rangle}}{\ensuremath{\langle\phi^*_k|}}$. Thus we obtain the partial transposition of $\rho_{AB}$ as $$\rho_{AB}^{T_B} =\sum_{i,k} p_{ik} P_{\psi_i} \otimes Q^T_{\phi_k} = \sum_{i,k} p_{ik} \left({\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi^*_{k}\rangle}}\right) \left({\ensuremath{\langle\psi_{i}|}} \otimes {\ensuremath{\langle\phi^*_{k}|}}\right).
\label{eq:range2}$$ It is evident from Eq. (\[eq:range1\]) and (\[eq:range2\]), that the ensembles $\{p_{ik},{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi_{k}\rangle}}\}$, and $\{p_{ik},{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi^*_{k}\rangle}}\}$ correspond to the states $ \rho_{AB} $ and $ \rho_{AB}^{T_{B}} $ respectively, and the vectors $\{{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi_{k}\rangle}}\}$ and $\{{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi_{k}^{*}\rangle}}\}$ span the ranges of $\rho_{AB}$ and $ \rho_{AB}^{T_{B}} $ respectively. $\square$
As an example, let us consider a state in $\mathbb{C}^3 \otimes \mathbb{C}^3$, given by $$\begin{aligned}
\rho_{AB}(a)={1 \over 8a + 1}
\left[ \begin{array}{ccccccccc}
a &0&0&0&a&0&0&0& a \\
0&a&0&0&0&0&0&0&0 \\
0&0&a&0&0&0&0&0&0 \\
0&0&0&a&0&0&0&0&0 \\
a &0&0&0&a&0&0&0& a \\
0&0&0&0&0&a&0&0&0 \\
0&0&0&0&0&0&{1+a \over 2}&0&{\sqrt{1-a^2} \over 2}\\
0&0&0&0&0&0&0&a&0 \\
a &0&0&0&a&0&{\sqrt{1-a^2} \over 2}&0&{1+a \over 2}\\
\end{array}
\right ], \ \ \
\label{tran}\end{aligned}$$ with $0 < a < 1$. The partial transposition of this density matrix, $\rho_{AB}^{T_B}(a) $ turns out to be positive. In Ref. [@Pawelbound], it was demonstrated that $\rho_{AB}(a)$ is entangled, which can be successfully detected by the range criterion. For $ a \neq 0,1$, one can find all product vectors $\{{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi^*_{k}\rangle}}\}$ in the range of $ \rho_{AB}^{T_{B}}(a) $. It was shown that the partial complex conjugation with respect to $B$ i.e., $ \{{\ensuremath{|\psi_{i}\rangle}} \otimes {\ensuremath{|\phi_{k}\rangle}}\} $ cannot span the range of $ \rho_{AB}(a) $, thus violating the condition 2 of Theorem 13.
Bell inequalities
=================
The concept of locality with respect to shared quantum states was first brought into light by Einstein, Podolsky, and Rosen (EPR) in their seminal paper in 1935 [@EPRparadox]. They argued that since nonclassical correlations of entangled states of the form $|\psi_{AB}\rangle = \sum _{i} a_i |e_i\rangle \otimes |f_i\rangle$ cannot be explained by any physical theory satisfying the notions of “locality" and “realism", quantum mechanics must be incomplete. In 1964, J. S. Bell [@Bell] provided a formulation of the problem that made the assumptions of locality and realism more precise, and more importantly, showed that the assumptions are actually *testable* in experiments. He derived a mathematical inequality that must be satisfied by any physical theory of nature, which is local as well as realistic.
As we shall see, Bell inequalities are essentially a special type of entanglement witness. An additional property of Bell inequalities is that any entangled state detected by them is nonclassical in a particular way: It violates “local realism”. The inequality is actually a constraint on a linear function of results of certain experiments. Modulo some so-called loopholes (see e.g. [@gerakol]), these inequalities have been shown to be actually violated in experiments (see e.g. [@Paris-ghyama] and references therein). In this section, we will first derive a Bell inequality [^6] and then show how this inequality is violated by the singlet state. Consider a two spin-1/2 particle state where the two particles are far apart. Let the particles be called $A$ and $B$. Let projection valued measurements in the directions $a$ and $b$ be done on $A$ and $B$ respectively. The outcomes of the measurements performed on the particles $A$ and $B$ in the directions $a$ and $b$, are respectively $A_a$ and $B_b$. The measurement result $A_a$ ($B_b$), whose values can be $\pm 1$, may depend on the direction $a$ ($b$) and some other uncontrolled parameter $\lambda$ which may depend on anything, that is, may depend upon system or measuring device or both. Therefore we assume that $A_a$ ($B_b$) has a definite pre-measurement value $A_a (\lambda)$ ($B_b (\lambda)$). Measurement merely uncovers this value. This is the *assumption of reality*. $\lambda$ is usually called a hidden variable and this assumption is also termed as the hidden variable assumption. Moreover, the measurement result at $A$ ($B$) does not depend on what measurements are performed at $B$ ($A$). That is, for example $A_a(\lambda)$ does not depend upon $b$. This is the *assumption of locality*, also called the Einstein’s locality assumption. The parameter $\lambda$ is assumed to have a probability distribution, say $\rho(\lambda)$. Therefore $\rho(\lambda)$ satisfies the following: $$\int \rho(\lambda) d\lambda = 1, \quad \rho(\lambda) \geq 0.$$ The correlation function of the two spin-1/2 particle state for a measurement in a fixed direction $a$ for particle $A$ and $b$ for particle $B$, is then given by (provided the hidden variables exist) $$\label{eq_correlation}
E(a, b) = \int A_a(\lambda) B_b(\lambda) \rho(\lambda) d\lambda.$$ Here $$A_a(\lambda) = \pm 1, \quad \mbox{and} \quad B_b(\lambda) = \pm 1,$$ because the measurement values were assumed to be $\pm 1$.
Let us now suppose that the observers at the two particles $A$ and $B$ can choose their measurements from two observables $a$, $a{'}$ and $b$, $b{'}$ respectively, and the corresponding outcomes are $A_a$, $A_{a{'}}$ and $B_b$, $B_{b{'}}$ respectively. Then $$\begin{aligned}
E(a, b) + E(a, b{'}) + E(a{'}, b) - E(a{'}, b{'}) \nonumber \\ =
\int [A_a (\lambda)(B_b(\lambda)+ B_{b{'}}( \lambda)) +
A_{a{'}} (\lambda)(B_b(\lambda)- B_{b{'}}( \lambda)) ]
\rho(\lambda) d\lambda .\end{aligned}$$ Now $B_b(\lambda)+ B_{b{'}}( \lambda)$ and $B_b(\lambda)- B_{b{'}}( \lambda)$ can only be $\pm 2$ and $0$, or $0$ and $\pm 2$ respectively. Consequently, $$\label{eq_Bell_inequality}
-2 \leq E(a, b) + E(a, b{'}) + E(a{'}, b) - E(a{'}, b{'}) \leq 2.$$ This is the well-known CHSH inequality. Note here that in obtaining the above inequality, we have never used quantum mechanics. We have only assumed Einstein’s locality principle and an underlying hidden variable model. Consequently, a Bell inequality is a constraint that any physical theory that is both, local and realistic, has to satisfy. Below, we will show that this inequality can be violated by a quantum state. Hence quantum mechanics is incompatible with an underlying local realistic model.
Detection of entanglement by Bell inequality
--------------------------------------------
Let us now show how the singlet state can be detected by a Bell inequality. This additionally will indicate that quantum theory is incompatible with local realism. For the singlet state ${\ensuremath{|\Psi^-\rangle}} = \frac{1}{\sqrt{2}} ({\ensuremath{|01\rangle}} - {\ensuremath{|10\rangle}})$, the quantum mechanical prediction of the correlation function $E(a, b)$ is given by $$\label{pukurchuri}
E(a, b) = \left \langle \Psi^{-} | \sigma_a \cdot \sigma_b| \Psi^{-} \right\rangle = - \cos(\theta_{ab}),$$ where $\sigma_a = \vec{\sigma} \cdot \vec{a} $ and similarly for $\sigma_b$. $\vec{\sigma}= (\sigma_x, \sigma_y, \sigma_z)$, where $\sigma_x$, $\sigma_y$ and $\sigma_z$ are the Pauli spin matrices. And $\theta_{ab}$ is the angle between the two measurement directions $a$ and $b$.
So for the singlet state, one has $$\begin{aligned}
\label{khalchuri}
B_{CHSH}&= & E(a, b) + E(a, b{'}) + E(a{'}, b) - E(a{'}, b{'}) \nonumber \\
&= & - \cos \theta_{ab} - \cos \theta_{a b{'}} - \cos \theta_{a{'}b} + \cos \theta_{a{'}b{'}}.\end{aligned}$$ The maximum value of this function is attained for the directions $a$, $b$, $a{'}$, $b{'}$ on a plane, as given in Fig. \[fig\_direction\_highest\], and in that case $$|B_{CHSH}| = 2 \sqrt{2}.$$
=0.5mm
(80,50)(0,-10) (10,15)[(1,0)[60]{}]{} (40,48)[$a$]{} (70,45)[$b$]{}
(70,15)[ $a{'}$]{} (70,-15)[$b{'}$]{}
(40,-15)[(0,1)[60]{}]{} (15,-10)[(1,1)[50]{}]{} (15,40)[(1,-1)[50]{}]{} (45, 30)[$\frac{\pi}{4}$]{}
This clearly violates the inequality in Eq. (\[eq\_Bell\_inequality\]). But Eq. (\[eq\_Bell\_inequality\]) was a constraint for any physical theory which has an underlying local hidden variable model. As the singlet state, a state allowed by the quantum mechanical description of nature, violates the constraint (\[eq\_Bell\_inequality\]), quantum mechanics cannot have an underlying local hidden variable model. In other words, quantum mechanics is not local realistic. This is the statement of the celebrated Bell theorem.
Moreover, it is easy to convince oneself that any separable state does have a local realistic description, so that such a state cannot violate a Bell inequality. Consequently, the violation of Bell inequality by the singlet state indicates that the singlet state is an entangled state. Further, the operator (cf. Eqs. (\[pukurchuri\]) and (\[khalchuri\])) $$\tilde{B}_{CHSH} = \sigma_a \cdot \sigma_b + \sigma_a \cdot \sigma_{b{'}} + \sigma_{a{'}} \cdot \sigma_b
- \sigma_{a{'}} \cdot \sigma_{b{'}}$$ can, by suitable scaling and change of origin, be considered as an entanglement witness for the singlet state, for $a$, $b$, $a{'}$, $b{'}$ chosen as in figure \[fig\_direction\_highest\] (cf. [@POMD]).
Note that violation of Bell inequalities is stronger than entanglement. For example, the Werner state $\rho^W_{AB}(p) = p|\Psi^-\rangle \langle \Psi^- | + \frac{1-p}{4} I \otimes I$ is entangled for $p>1/3$, but it violates CHSH inequality for $p > 1/\sqrt{2}$ [@Werner; @horo_bell].
Quantification of entanglement
==============================
\[sec:ent\_measures\]
The entanglement content of a pure two-party quantum state was initially quantified by the usefulness of the state in communication protocols, e.g., quantum teleportation, quantum dense coding, etc. [@bennett_eof0; @bennett_eof1]. Since entangled quantum states can be used to perform teleportation and dense coding with efficiencies exceeding those situations in which no entanglement is available, entanglement is considered to be a “resource". Moreover, it was found that the singlet state ${\ensuremath{|\Psi^-\rangle}} = \frac{1}{\sqrt{2}} ({\ensuremath{|01\rangle}} - {\ensuremath{|10\rangle}})$ can perform these tasks with the maximal possible efficiency, thus it was assumed that the singlet state or any other state which is connected to the singlet state by local unitary transformations is a maximally entangled state in $\mathbb{C}^2 \otimes \mathbb{C}^2$. It was further assumed that maximally entangled states in $\mathbb{C}^2 \otimes \mathbb{C}^2$ has a unit amount of entanglement, or has 1 *ebit* (“entanglement bit"). What if one has a shared entangled state ${\ensuremath{|\psi_{AB}\rangle}} \neq {\ensuremath{|\Psi^-\rangle}}$? In that case, one can show that given many copies of ${\ensuremath{|\psi_{AB}\rangle}} \in \mathbb{C}^2 \otimes \mathbb{C}^2$, one can extract a fewer number of singlets using LOCC, which can thereafter be used in quantum communication schemes. Conversely, if one has a collection of singlets, then it can be converted into a collection of ${\ensuremath{|\psi_{AB}\rangle}}$ via LOCC. Bennett et al. showed that $n$ copies of an entangled state ${\ensuremath{|\psi_{AB}\rangle}}$, shared between Alice and Bob can be *reversibly* converted, using only LOCC between Alice and Bob, into $m$ copies of singlets, where $m/n$ tends to $S(\rho_A) = S(\rho_B)$ and the fidelity of the conversion approaches unity for large $n$ [@bennett_eof1]. This led to the quantification of the entanglement content of a pure quantum state ${\ensuremath{|\psi_{AB}\rangle}}$ by the von Neumann entropy of its reduced density matrices [@bennett_eof1]: $$\mathcal{E}_E({\ensuremath{|\psi_{AB}\rangle}}) = S(\rho_A) = S(\rho_B).
\label{ent_entropy}$$ This quantification also remains valid in higher dimensions. We refer to the quantity $\mathcal{E}_E({\ensuremath{|\psi_{AB}\rangle}})$ as the the “entropy of entanglement" (or simply entanglement) of ${\ensuremath{|\psi_{AB}\rangle}}$. Clearly, for a disentangled pure state ${\ensuremath{|\psi_{AB}\rangle}} = {\ensuremath{|\psi_A\rangle}} \otimes {\ensuremath{|\psi_B\rangle}}$, $\rho_A$ and $\rho_B$ are also pure states, for which the von Neumann entropies vanish, and $\mathcal{E}_E({\ensuremath{|\psi_{AB}\rangle}}) = 0$. But if ${\ensuremath{|\psi_{AB}\rangle}}$ is an entangled state of the form $|\psi_{AB}\rangle = \sum_i a_i |e_i\rangle \otimes |f_i\rangle$ with more than one nonzero $a_i$, then we have $\rho_A = \sum_i |a_i|^2 {\ensuremath{|e_i\rangle}}{\ensuremath{\langlee_i|}}$ and $\rho_B = \sum_i |a_i|^2 {\ensuremath{|f_i\rangle}}{\ensuremath{\langlef_i|}}$. In this case, we have $\mathcal{E}_E({\ensuremath{|\psi_{AB}\rangle}}) = S(\rho_A) = S(\rho_B) > 0$, and is given by the Shannon entropy of the probability distribution $\{|a_i^2|\}$. Entropy of entanglement ranges from zero for a product state to $\log_2 d$ for a maximally entangled state in a Hilbert space of dimension $d \otimes d$. Clearly for the singlet state ${\ensuremath{|\Psi^-\rangle}}$, the entropy of entanglement $\mathcal{E}_E = 1$.
Before extending the quantitative theory of entanglement to the more general situation in which Alice and Bob share a mixed state $\rho_{AB}$, we present essential conditions that any measure of entanglement $\mathcal{E}(\rho)$ has to satisfy [@vedral_ent; @horo_review; @horo_ent_limit].
1. Non-negativity, i.e., $\mathcal{E}(\rho) \geq 0$ for any bipartite quantum state $\rho$.
2. Entanglement vanishes for separable states, i.e., $\mathcal{E}(\sigma) = 0$ if $\sigma$ is separable.
3. Invariance under local unitary transformations, i.e., $\mathcal{E}(U_A \otimes U_B \rho_{AB} U^{\dagger}_A \otimes U^{\dagger}_B) = \mathcal{E}(\rho_{AB})$.
4. Entanglement cannot increase under local operations and classical communication, i.e., for a given LOCC $\Lambda$, $\mathcal{E}(\Lambda(\rho)) \leq \mathcal{E}(\rho)$.
Condition 1 is there by convention. In any resource theory, the quantification of the resource must be done by a quantity that does not increase under the free operations. Moreover, the quantity must be zero for the states that can be created by these free operations. In the resource theory of entanglement, the free operations are the LOCC, and thus entanglement measures cannot increase under LOCC and separable states must not have any entanglement. This accounts for conditions 2 and 4. Condition 3 arises as local unitary transformations represent only a local change of basis and do not change any correlation. A quantity that satisfies these conditions can be called an entanglement measure, and is eligible for the quantification of the entanglement content of a quantum state. They are also often referred to as *entanglement monotones*. Some authors also impose *convexity* and *additivity* properties for entanglement measures:
- Entanglement is a convex function, i.e., $\mathcal{E}\left(\sum_i p_i \rho_i\right) \leq \sum_i p_i \mathcal{E}(\rho_i)$, for an ensemble $\{p_i, \rho_i\}$.
- Entanglement is additive, i.e., $\mathcal{E}(\rho^{\otimes n}) = n \mathcal{E}(\rho)$.
Below, we briefly discuss a few measures of entanglement.
Entanglement of formation
-------------------------
One way to widen the theory of entanglement measures to the mixed state regime is by the convex roof extension of pure state entanglement measures [@uhlmann_convex]. The first measure introduced by this technique was the *entanglement of formation* [@bennett_eof2].\
**Def. 16** *Entanglement of formation of a quantum state $\rho_{AB}$ shared between Alice and Bob is defined as* $$\mathcal{E}_{EoF}(\rho_{AB}) = \underset{\{p_i,{\ensuremath{|\psi^i_{AB}\rangle}}\}}{\mbox{min}} \sum_i p_i \mathcal{E}_E({\ensuremath{|\psi^i_{AB}\rangle}}),
\label{eq:eof}$$ *where the minimization is taken over all possible pure state decompositions, $\rho_{AB} = \sum_i p_i {\ensuremath{|\psi^i_{AB}\rangle}}{\ensuremath{\langle\psi^i_{AB}|}}$, of $\rho_{AB}$, and $\mathcal{E}_E({\ensuremath{|\psi^i_{AB}\rangle}})$ is the entropy of entanglement of the pure state ${\ensuremath{|\psi^i_{AB}\rangle}}$.*
Clearly, the entanglement of formation for a pure state collapses to the corresponding entropy of entanglement. Using the singlet as the basic unit of entanglement, one can perceive the operational meaning of the entanglement of formation in the following manner:
1. Decompose $\rho_{AB}$ into a pure state ensemble as $\rho_{AB} = \sum_i p_i {\ensuremath{|\psi^i_{AB}\rangle}}{\ensuremath{\langle\psi^i_{AB}|}}$.
2. Choose the state ${\ensuremath{|\psi^i_{AB}\rangle}}$ according to the corresponding probability $p_i$.
3. Prepare the state ${\ensuremath{|\psi^i_{AB}\rangle}}$ from singlets via local operations and classical communication.
4. Finally, forget the identity of the chosen ensemble state.
In this way, one needs on average $\sum_i p_i \mathcal{E}_E({\ensuremath{|\psi^i_{AB}\rangle}})$ singlets, and then one can choose the pure state ensemble for which the average is minimum. This minimum number of singlets required to prepare $\rho_{AB}$ in this procedure gives the entanglement of formation of $\rho_{AB}$.
The convex roof optimization given in Eq. (\[eq:eof\]) is formidable to compute for general mixed states. However, the exact closed form of entanglement of formation is known for two-qubit mixed states in terms of the “*concurrence*".
### Concurrence
Concurrence for pure states was first introduced in Ref. [@hill_conc]. For a two-qubit pure state, ${\ensuremath{|\psi_{AB}\rangle}}$, the concurrence is defined as $$\mathcal{C}(\psi_{AB}) = |\langle\psi_{AB}|\tilde{\psi}_{AB}\rangle|,
\label{eq:pure_conc}$$ where ${\ensuremath{|\tilde{\psi}_{AB}\rangle}} = \sigma_y \otimes \sigma_y {\ensuremath{|\psi^*_{AB}\rangle}}$, with ${\ensuremath{|\psi^*_{AB}\rangle}}$ being the complex conjugate of ${\ensuremath{|\psi_{AB}\rangle}}$ in the standard computational basis $\{ {\ensuremath{|00\rangle}}, {\ensuremath{|01\rangle}}, {\ensuremath{|10\rangle}}, {\ensuremath{|11\rangle}} \}$. For two-qubit mixed states, a closed form expression of the convex roof extension of concurrence can be obtained [@wootters_conc]. For a two-qubit density matrix $\rho_{AB}$, let us first define the spin-flipped density matrix as $\tilde{\rho}_{AB} = (\sigma_y \otimes \sigma_y) \rho^*_{AB}(\sigma_y \otimes \sigma_y)$, and the operator $R = \sqrt{\sqrt{\rho_{AB}}\rho^*_{AB}\sqrt{\rho_{AB}}}$. The convex roof extended concurrence of $\rho_{AB}$ is then given by $$\mathcal{C}(\rho_{AB}) = \mbox{min}\{0,\lambda_1 -\lambda_2 - \lambda_3 -\lambda_4\},
\label{eq:mixed_conc}$$ where the $\lambda_i$’s are the eigenvalues of $R$ in decreasing order. A computable formula for entanglement of formation of two-qubit quantum states can be expressed in terms of the concurrence [@hill_conc; @wootters_conc].\
**Theorem 14** [@wootters_conc]\
*The entanglement of formation of a two-qubit quantum state $\rho_{AB}$ is given by $\mathcal{E}_{EoF}(\rho_{AB}) = \mathcal{F}(\mathcal{C}(\rho_{AB}))$, where the function $\mathcal{F}(\mathcal{C})$ is defined as $$\begin{aligned}
\mathcal{F}(\mathcal{C}) = h \left( \frac{1+ \sqrt{1-\mathcal{C}^2}}{2}\right),\end{aligned}$$ with $h(x) = -x\log_2x - (1-x)\log_2(1-x).$*
Since $\mathcal{F}(\mathcal{C})$ is a monotonically increasing function of $\mathcal{C}$ and goes from 0 to 1 as $\mathcal{C}$ goes from 0 to 1, we can also consider the concurrence as a measure of entanglement in $\mathbb{C}^2 \otimes \mathbb{C}^2$.
### Entanglement cost
As we have seen earlier, the entropy of entanglement of a pure state ${\ensuremath{|\psi_{AB}\rangle}}$ quantifies the average number of singlets needed to asymptotically construct ${\ensuremath{|\psi_{AB}\rangle}}$ via LOCC. We went over to the mixed state scenario by using the concept of entanglement of formation. However, the definition of entanglement of formation consists of a combination of asymptotic and non-asymptotic LOCC transformations. Let us now present a purely asymptotic entanglement measure, known as *entanglement cost*.\
**Def. 17** *Given $m$ copies of singlet state $({\ensuremath{|\Psi^-\rangle}}^{\otimes m})$, consider all LOCC protocols $\mathcal{P}$ that can transform the $m$ singlets into a state $\sigma_n$ such that $\mathcal{D}(\rho_{AB}^{\otimes n}, \sigma_n) \rightarrow 0$ as $n \rightarrow \infty$, with $\mathcal{D}$ being a suitable distance functional. The entanglement cost of $\rho_{AB}$ is defined as* $$\mathcal{E}_C(\rho_{AB}) = \underset{\mathcal{P}} {\mbox{min}} \left(\underset{n \rightarrow \infty}{\lim} \frac{m}{n}\right),
\label{eq:ent_cost}$$ *where the minimization is taken over all such LOCC protocols $\mathcal{P}$.*
Hayden et al. have shown than entanglement cost is equal to the regularized entanglement of formation [@hayden_cost], given by $$\mathcal{E}_C = \underset{n \rightarrow \infty} {\lim} \frac{\mathcal{E}_{EoF}(\rho_{AB}^{\otimes n})}{n}.$$ Clearly, if entanglement of formation is additive, entanglement cost will be equal to the entanglement of formation. For pure states, entanglement cost reduces to the entropy of entanglement.
Distillable entanglement
------------------------
Distillable entanglement [@bennett_eof2; @bennett_dist; @plenio_ent; @rains] is a measure dual to entanglement cost. In case of entanglement cost, we looked at the asymptotic rate at which one can prepare the given state from maximally entangled states via LOCC, whereas in this case, we will be interested in the rate of “distillation" of a given state into singlets, via LOCC. The formal definition of distillable entanglement of a bipartite quantum state $\rho_{AB}$ shared between Alice and Bob is as follows:
**Def. 18** *Given $n$ copies of the shared quantum state $\rho_{AB}$, if one can prepare a state $\sigma_n$ by a LOCC protocol $\mathcal{P}$, such that $\mathcal{D}\left(({\ensuremath{|\Psi^-\rangle}}{\ensuremath{\langle\Psi^-|}})^{\otimes n}, \sigma_n\right) \rightarrow 0$ as $n \rightarrow \infty$, with $\mathcal{D}$ being a suitable distance functional, then $\mathcal{P}$ is referred to as a distillation protocol. The distillable entanglement of $\rho_{AB}$ is defined as* $$\mathcal{E}_D = \underset{\mathcal{P}} {\mbox{max}} \left(\underset{n \rightarrow \infty}{\lim} \frac{m}{n}\right),
\label{eq:dist_ent}$$ *where the maximization is taken over all distillation protocols $\mathcal{P}$.*
For pure states, optimal entanglement transformations are reversible, and thus distillable entanglement and entanglement cost coincide and reduce to the entropy of entanglement. But in general, $\mathcal{E}_D \leq \mathcal{E}_C$. To understand this inequality, we note that if the opposite is allowed, one can get more singlets by distilling a state than the amount of singlets required to create it, leading to a *perpetuum mobile*. Bound entangled states cannot be distilled, and so the distillable entanglements for bound entangled states are always zero, whereas since these states are entangled, their entanglements of formation are non-zero. There are examples of bound entangled states, whose entanglement costs have also been proven to be non-zero [@vidal_bound], leading to irreversibility in asymptotic entanglement transformations.
Relative entropy of entanglement
--------------------------------
\[sec:rel\_ent\]
A qualitatively different way to quantify entanglement is based on the geometry of quantum states. It is defined as the distance between an entangled state and its closest separable state [@GM1]. If $\mathcal{S}$ is the set of all separable states, then a distance-based measure of entanglement for a bipartite shared state $\rho_{AB}$ is given by [@GM1; @vedral_ent; @vedral_rel_ent; @vedral_rmp_re] $$\mathcal{E}_\mathcal{G}(\rho_{AB}) = \underset{\sigma \in \mathcal{S}} {\mbox{min}} \ \mathcal{D}(\rho_{AB}, \sigma),
\label{eq:measure_d}$$ where $\mathcal{D}$ is a suitably chosen distance measure. For $\mathcal{E}_\mathcal{G}(\rho_{AB})$ to be a “good" measure of entanglement, the distance measure $\mathcal{D}$ can be required to satisfy following properties.
1. $\mathcal{D}(\rho, \sigma) \geq 0$ for any two states $\rho$ and $\sigma$; equality holds iff $\rho = \sigma$.
2. Invariance under unitary operations. $\mathcal{D}(U \rho U^{\dagger}, U \sigma U^{\dagger}) = \mathcal{D}(\rho, \sigma)$.
3. $\mathcal{D}(\rho, \sigma)$ is non-increasing under every completely positive and trace preserving map $\Lambda$, i.e., $\mathcal{D}(\Lambda(\rho), \Lambda(\sigma)) \leq \mathcal{D}(\rho, \sigma)$.
The reason for the distance measure $\mathcal{}$ to satisfy these properties is that they imply conditions 1 - 4 for entanglement measures mentioned earlier in this section.
One of the most famous members of this family of distance-based measures is the *relative entropy of entanglement*, where we take the von Neumann relative entropy, which is defined in analogy with the classical Kullback-Leibler distance, as the diatance measure. For two density matrices, $\rho$ and $\sigma$, it is defined as [@NChuang] $$S(\rho || \sigma) = \mbox{tr} \left(\rho \log_2 \frac{\rho}{\sigma}\right) = \mbox{tr} \left\{\rho (\log_2\rho - \log_2\sigma) \right\}.$$ It is to be noted that the relative entropy $S(\rho || \sigma)$ is not symmetric in its arguments, $\rho$ and $\sigma$.
**Def. 19** *The relative entropy of entanglement of a bipartite state $\rho_{AB}$ shared between Alice and Bob is defined as* $$\mathcal{E}_{RE}(\rho_{AB}) = \underset{\sigma \in \mathcal{S}} {\mbox{min}} \ S(\rho_{AB} || \sigma),
\label{eq:measure_d}$$ *where $\mathcal{S}$ denotes the set of all separable states.*\
We now state two important theorems on relative entropy of entanglement.
**Theorem 15** [@vedral_ent; @vedral_rel_ent]\
*For pure bipartite states, the relative entropy of entanglement reduces to the entropy of entanglement.*
**Theorem 16** [@vedral_rel_ent]\
*The relative entropy of entanglement $\mathcal{E}_{RE}(\rho_{AB})$ provides an upper bound to the distillable entanglement $\mathcal{E}_D(\rho_{AB})$, and a lower bound to the entanglement of formation $\mathcal{E}_{EoF}(\rho_{AB})$, i.e., $\mathcal{E}_{D}(\rho_{AB})
\leq \mathcal{E}_{RE}(\rho_{AB}) \leq \mathcal{E}_{EoF}(\rho_{AB})$.*
Although computation of the relative entropy of entanglement for arbitrary mixed states is quite hard, one can characterize the set of entangled states for all of whom a given separable state is the closest separable state, when relative entropy is considered as the distance measure [@rel_ent_closed].
Based on other distance measures, several “geometric” entanglement measures have also been introduced (see Sec. \[sec:ksep\]).
Negativity and logarithmic negativity
-------------------------------------
The partial transposition criterion for entanglement, mentioned in Sec. \[sec:PPT\], provides another quantity to quantify the entanglement content of a given quantum state. This quantity is known as the *negativity* [@eisert_neg; @vidal_neg; @karol_neg; @plenio_neg], given by the absolute sum of negative eigenvalues of the partial transposed density matrix. In other words, the negativity of a shared quantum state $\rho_{AB}$ is defined as $$\mathcal{N}(\rho_{AB}) = \frac{||\rho_{AB} ^{T_B}|| - 1}{2} = \frac{||\rho_{AB} ^{T_A}|| - 1}{2},
\label{eq:neg}$$ where $||A|| = \mbox{tr} \ \sqrt{A^{\dagger}A}$ is the matrix trace norm. Although $\mathcal{N}(\rho)$ satisfies the convexity property, it is not additive. Based on negativity, one can define an additive entanglement measure, known as *logarithmic negativity*, and is given by $$\begin{aligned}
\mathcal{E}_{LN}(\rho_{AB}) &=& \log_2 ||\rho_{AB}^{T_B}|| = \log_2 ||\rho_{AB}^{T_A}|| \nonumber \\
&=& \log_2\left(2 \mathcal{N}(\rho_{AB}) + 1 \right).
\label{eq:log_neg}\end{aligned}$$ $\mathcal{E}_{LN}(\rho_{AB})$ is a monotone under deterministic LOCC operations. However, it fails to be a convex function. It was also shown to be an upper bound of distillable entanglement [@vidal_neg].
A major advantage of negativity and logarithmic negativity is that they are easy to compute for general, possibly mixed, quantum states of arbitrary dimensions. Clearly, for PPT bound entangled states, $\mathcal{N}(\rho_{AB})$ and $\mathcal{E}_{LN}(\rho_{AB})$ are zero and cannot be used to quantify entanglement. But in $\mathbb{C}^2 \otimes \mathbb{C}^2$ and $\mathbb{C}^2 \otimes \mathbb{C}^3$, their non-zero values are necessary and sufficient for detecting entanglement (see Theorem 3).
Classification of bipartite states with respect to quantum dense coding
=======================================================================
Up to now, we have been interested in splitting the set of all bipartite quantum states into separable and entangled states. However, one of the main motivations behind the study of entangled states is that some of them can be used to perform certain tasks, which are not possible if one uses states without entanglement. It is, therefore, important to find out which entangled states are useful for a given task. We discuss here the particular example of quantum dense coding [@BW].
Suppose that Alice wants to send two bits of classical information to Bob. Then a general result known as the Holevo bound (to be discussed below), shows that Alice must send two qubits (i.e. 2 two-dimensional quantum systems) to Bob, if only a noiseless quantum channel is available. However, if additionally Alice and Bob have previously shared entanglement, then Alice may have to send less than two qubits to Bob. It was shown by Bennett and Wiesner [@BW], that by using a previously shared singlet (between Alice and Bob), Alice will be able to send two bits to Bob, by sending just a single qubit.
The protocol of dense coding [@BW] works as follows. Assume that Alice and Bob share a singlet state $$|\Psi^-\rangle = \frac{1}{\sqrt{2}}\left(|01\rangle - |10\rangle\right).$$ The crucial observation is that this entangled two-qubit state can be transformed into four orthogonal states of the two-qubit Hilbert space by performing unitary operations on just a single qubit. For instance, Alice can apply a rotation (the Pauli operations) or do nothing to her part of the singlet, while Bob does nothing, to obtain the three triplets (or the singlet): $$\begin{aligned}
\sigma_x \otimes I |\Psi^-\rangle = - |\Phi^-\rangle, & \quad & \sigma_y \otimes I |\Psi^-\rangle = i |\Phi^+\rangle, \nonumber\\
\sigma_z \otimes I |\Psi^-\rangle = |\Psi^+\rangle, & \quad & I \otimes I |\Psi^-\rangle = |\Psi^-\rangle, \end{aligned}$$ where $$\begin{aligned}
\label{gorur-garir-headlight}
|\Psi^\pm\rangle &=& \frac{1}{\sqrt{2}}\left(|01\rangle \pm |10\rangle\right), \nonumber\\
|\Phi^\pm\rangle &=& \frac{1}{\sqrt{2}}\left(|00\rangle \pm |11\rangle\right),
$$ are the Bell states and $I$ is the qubit identity operator. Suppose that the classical information that Alice wants to send to Bob is $i$, where $i= 0,1,2,3$. Alice and Bob previously agree on the following correspondence between the operations applied at Alice’s end and the information $i$ that she wants to send: $$\begin{aligned}
\sigma_x \Rightarrow i=0, & \quad & \sigma_y \Rightarrow i=1, \nonumber\\
\sigma_z \Rightarrow i=2, & \quad & I \Rightarrow i=3.\end{aligned}$$ Depending on the classical information she wishes to send, Alice applies the appropriate rotation on her part of the shared singlet, according to the above correspondence. Afterwards, Alice sends her part of the shared state to Bob, via the noiseless quantum channel. Bob now has in his possession, the entire two-qubit state, which is in any of the four Bell states $\left\{|\Psi^\pm\rangle, |\Phi^\pm\rangle \right\}$. Since these states are mutually orthogonal, he will be able to distinguish between them and hence find out the classical information sent by Alice.
To consider a more realistic scenario, usually two avenues are taken. One approach is to consider a *noisy* quantum channel, while the additional resource is an arbitrary amount of shared bipartite *pure* state entanglement (see e.g. [@Bennett-ek; @Bennett-dui], see also [@MarieCurie; @Debu]). The other approach is to consider a *noiseless* quantum channel, while the assistance is by a given bipartite *mixed* entangled state (see e.g. [@MarieCurie; @Debu; @ek1; @dui1; @char1; @tin1]).
Here, we consider the second approach, and derive the capacity of dense coding in this scenario, for a given state, where the the capacity is defined as the number of classical bits that can be accessed by the receiver, per usage of the noiseless channel. This will lead to a classification of bipartite states according to their ability to assist in dense coding. In the case where a noisy channel and an arbitrary amount of shared pure entanglement is considered, the capacity refers to the channel (see e.g. [@Bennett-ek; @Bennett-dui]). However, in our case when a noiseless channel and a given shared (possibly mixed) state is considered, the capacity refers to the state. Note that the mixed shared state in our case can be thought of as an output of a noisy channel. A crucial element in finding the capacity of dense coding is the Holevo bound [@ref-halum], which is a universal upper bound on classical information that can be decoded from a quantum ensemble. Below we discuss the bound, and subsequently derive the capacity of dense coding.
The Holevo bound
----------------
\[sec-halum\]
The Holevo bound is an upper bound on the amount of classical information that can be accessed from a quantum ensemble in which the information is encoded. Suppose therefore that Alice ($A$) obtains the classical message $i$ that occurs with probability $p_i$, and she wants to send it to Bob ($B$). Alice encodes this information $i$ in a quantum state $\rho_i$, and sends it to Bob. Bob receives the ensemble $\{p_i, \rho_i\}$, and wants to obtain as much information as possible about $i$. To do so, he performs a measurement, that gives the result $m$, with probability $q_m$. Let the corresponding post-measurement ensemble be $\{p_{i|m}, \rho_{i|m}\}$. The information gathered can be quantified by the mutual information between the message index $i$ and the measurement outcome [@Chennai]: $$I(i:m)= H(\{p_i\}) - \sum_m q_m H(\{p_{i|m}\}).$$ Note that the mutual information can be seen as the difference between the initial disorder and the (average) final disorder. Bob will be interested to obtain the maximal information, which is maximum of $I(i:m)$ for all measurement strategies. This quantity is called the accessible information: $$I_{acc} = \max I(i:m),$$ where the maximization is over all measurement strategies.
The maximization involved in the definition of accessible information is usually hard to compute, and hence the importance of bounds [@ref-halum; @Utpakhi]. In particular, in Ref. [@ref-halum], a universal upper bound, the Holevo bound, on $I_{acc}$ is given: $$I_{acc}(\{p_i, \rho_i\}) \leq \chi(\{p_i, \rho_i\}) \equiv S(\overline{\rho}) - \sum_i p_i S(\rho_i).$$ See also [@Rajabazar1; @Khajuraho; @Rajabazar2]. Here $\overline{\rho} = \sum_ip_i\rho_i$ is the average ensemble state, and $S(\varsigma)= - \mbox{tr}(\varsigma \log_2 \varsigma)$ is the von Neumann entropy of $\varsigma$.
The Holevo bound is asymptotically achievable in the sense that if the sender Alice is able to wait long enough and send long strings of the input quantum states $\rho_i$, then there exists a particular encoding and a decoding scheme that asymptotically attains the bound. Moreover, the encoding consists in collecting certain long and “typical” strings of the input states, and sending them all at once [@babarey; @maarey].
Capacity of quantum dense coding
--------------------------------
\[koto-capa-re?\]
Suppose that Alice and Bob share a quantum state $\rho_{AB}$. Alice performs the unitary operation $U_i$ with probability $p_i$, on her part of the state $\rho_{AB}$. The classical information that she wants to send to Bob is $i$. Subsequent to her unitary rotation, she sends her part of the state $\rho^{AB}$ to Bob. Bob then has the ensemble $\{p_i, \rho_i\}$, where $$\rho_i = U_i \otimes I \rho_{AB} U_i^\dagger \otimes I.$$
The information that Bob is able to gather is $I_{acc}(\{p_i, \rho_i\})$. This quantity is bounded above by $\chi(\{p_i, \rho_i\})$, and is asymptotically achievable. The “one-capacity” $C^{(1)}$ of dense coding for the state $\rho_{AB}$ is the Holevo bound for the best encoding by Alice: $$\label{Kohinoor}
C^{(1)}(\rho) = \max_{p_i,U_i} \chi(\{p_i, \rho_i\}) \equiv \max_{p_i,U_i} \left(S(\overline{\rho}) - \sum_i p_i S(\rho_i)\right).$$ The superscript $(1)$ reflects the fact that Alice is using the shared state once at a time, during the asymptotic process. She is not using entangled unitaries on more than one copy of her parts of the shared states $\rho_{AB}$. As we will see below, encoding with entangled unitaries does not help her to send more information to Bob.
![Classification of bipartite quantum states according to their usefulness in dense coding. The convex innermost region, marked as S, consists of separable states. The shell surrounding it, marked as PPT, is the set of PPT entangled states. The next shell, marked as n-DC, is the set of all states that are NPT, but not useful for dense coding. The outermost shell is that of dense-codeable states.[]{data-label="fig:ghorar-dim"}](dimdense-book.pdf)
In performing the maximization in Eq. (\[Kohinoor\]), first note that the second term in the right hand side (rhs) is $-S(\rho)$, for all choices of the unitaries and probabilities. Secondly, we have $$S(\overline{\rho}) \leq S(\overline{\rho}_A)
+ S(\overline{\rho}_B) \leq \log_2 d_A + S(\overline{\rho}_B),$$ where $d_A$ is the dimension of Alice’s part of the Hilbert space of $\rho_{AB}$, and $ \overline{\rho}_A = \mbox{tr}_B \overline{\rho}$, $ \overline{\rho}_B = \mbox{tr}_A \overline{\rho}$. Moreover, $S(\overline{\rho}_B) = S(\rho_B)$, as nothing was done at Bob’s end during the encoding procedure. (In any case, unitary operations does not change the spectrum, and hence the entropy, of a state.) Therefore, we have $$\max_{p_i,U_i} S(\overline{\rho}) \leq \log_2 d_A + S(\rho_B).$$ But the bound is reached by any complete set of orthogonal unitary operators $\{W_j\}$, to be chosen with equal probabilities, which satisfy the *trace rule* $\frac{1}{d_A^2}\sum_j W_j^\dagger \Xi W_j=\mbox{tr}[\Xi]I$, for any operator $\Xi$. Therefore, we have $$C^{(1)}(\rho) = \log_2 d_A + S( \rho_B ) - S(\rho).$$
The optimization procedure above sketched essentially follows that in Ref. [@tin1]. Several other lines of argument are possible for the maximization. One is given in Ref. [@dui1] (see also [@TajMahal]). Another way to proceed is to guess where the maximum is reached (maybe from examples or by taking the most symmetric option), and then perturb the guessed result. If the first order perturbations vanish, the guessed result is correct, as the von Neumann entropy is a concave function and the maximization is carried out over a continuous parameter space.
Without using the additional resource of entangled states, Alice will be able to reach a capacity of just $\log_2d_A$ bits. Therefore, entanglement in a state $\rho_{AB}$ is useful for dense coding if $S( \rho_B ) - S(\rho)>0$. Such states will be called dense-codeable (DC) states. Such states exist, an example being the singlet state.
Note here that if Alice is able to use entangled unitaries on two copies of the shared state $\rho$, the capacity is not enhanced (see Ref. [@ar-keu-korechhey?]). Therefore, the one-capacity is really the asymptotic capacity, in this case. Note however that this additivity is known only in the case of encoding by unitary operations. A more general encoding may still have additivity problems (see e.g. [@Debu]). Here, we have considered unitary encoding only. This case is both mathematically more accessible, and experimentally more viable.
A bipartite state $\rho_{AB}$ is useful for dense coding if and only if $S( \rho_B ) - S(\rho)>0$. It can be shown that this relation cannot hold for PPT entangled states [@MarieCurie] (see also [@TajMahal]). Therefore a DC state is always NPT. However, the converse is not true: There exist states which are NPT, but not useful for dense coding. Examples of such states can be obtained by the considering the Werner state $\rho^W_{AB}(p) = p|\Psi^-\rangle \langle \Psi^- | + \frac{1-p}{4} I \otimes I$ [@Werner].
The discussions above leads to the following classification of bipartite quantum states:
1. Separable states: These states are of course not useful for dense coding. They can be prepared by LOCC.
2. PPT entangled states: These states, despite being entangled, cannot be used for dense coding. Moreover, their entanglement cannot be detected by the partial transposition criterion.
3. NPT non-DC states: These states are entangled, and their entanglement can be detected by the partial transposition criterion. However, they are not useful for dense coding.
4. DC states: These entangled states can be used for dense coding.
The above classification is illustrated in figure \[fig:ghorar-dim\]. A generalization of this classification has been considered in Refs. [@TajMahal; @ar-keu-korechhey?].
Multipartite states
===================
\[sec:multi\]
The discussion about detection of bipartite entanglement presented above is of course quite far from complete. And yet, in this section, we present a few remarks on multipartite states and multipartite entanglement. The case of detection of entanglement of pure states is again simple, although there are different types of entanglement present in a multipartite system. One quickly realizes that a multipartite pure state is entangled if and only if it is entangled in at least one bipartite splitting. So, for example, the Greenberger-Horne-Zeilinger (GHZ) state $|\mbox{GHZ}\rangle = \frac{1}{\sqrt{2}} ( |000\rangle + |111\rangle ) $ [@GHZ], shared between three parties $A$, $B$ and $C$, is entangled, because it is entangled in the $A$:$BC$ bipartition (as also in all others), whereas the state $|\phi\rangle = \frac{1}{\sqrt{2}} (|000\rangle + |101\rangle) $ is entangled in the $A$:$BC$ and $AB$:$C$ bipartite splits but not in the $B$:$AC$ one.
![ Geometric representation of the hierarchy of multipartite entangled states among $N$-party quantum states. Each red-shaded set contains $k$-separable states (for $2 \leq k \leq N$), while the grey-shaded set contains all genuinely multipartite entangled states. The dotted lines indicate that there can be several $k$-separable state sets between $2$-separable and $N$-separable ones.[]{data-label="fig:p_sep"}](p-sep.pdf)
$k$-separable, fully-separable, and genuine multipartite entangled states
-------------------------------------------------------------------------
\[sec:ksep\] Among multipartite states, there exists a hierarchical structure of states with respect to their entanglement quality. An $N$-party pure quantum state is called $k$-separable ($2 \leq k \leq N$), if it is separable in at least $k-1$ bipartite splitting. Similarly, an $N$-party pure quantum state is $N$-separable or fully-separable if it is separable in all bipartite splittings. A pure quantum state possesses genuine multipartite entanglement if and only if it is entangled in all possible bipartite cuts. For example, the state $|\phi\rangle = \frac{1}{\sqrt{2}} (|000\rangle + |101\rangle) $ is bi-separable or $2$-separable, whereas the GHZ state is genuinely multipartite entangled.
The case of mixed states is more involved. A possibly mixed quantum state $\rho$ of $N$ parties is $k$-separable, if in every pure state decomposition of $\rho = \sum_i p_i {\ensuremath{|\psi^i\rangle}}{\ensuremath{\langle\psi^i|}}$, there exist at least one $k$-separable pure state and no other state with separability lesser than $k$. Similarly, a possibly mixed quantum state is genuinely multipartite entangled, if it has at least one genuine multipartite entangled pure state in every pure state decomposition of it. For example, in the three qubit case, the equal mixture of the W state ${\ensuremath{|W\rangle}} = \frac{1}{\sqrt{3}}({\ensuremath{|001\rangle}} + {\ensuremath{|010\rangle}} + {\ensuremath{|100\rangle}})$ [@dur_vidal_cirac; @W_state] and its “complement" ${\ensuremath{|\bar{W}\rangle}} = \frac{1}{\sqrt{3}}({\ensuremath{|110\rangle}} + {\ensuremath{|101\rangle}} + {\ensuremath{|011\rangle}})$ is genuinely multipartite entangled [@qc_wt_cc], and the equal mixture of ${\ensuremath{|\psi_1\rangle}} = \frac{1}{\sqrt{2}}({\ensuremath{|001\rangle}} + {\ensuremath{|010\rangle}})$ and ${\ensuremath{|\psi_2\rangle}} = \frac{1}{\sqrt{2}}({\ensuremath{|001\rangle}} + {\ensuremath{|100\rangle}})$ is bi-separable. Figure \[fig:p\_sep\] depicts the schematic geometric picture of this hierarchical structure of multipartite entanglement.
One avenue to quantify the degree of such multipartite entanglement relies on the above geometric structure of multipartite entangled states [@GM1; @GM2; @GM3]. Given a distance functional $\mathcal{D}$, that satisfies conditions (1)-(3) given in Sec. \[sec:rel\_ent\], the quantity $$\mathcal{E}^{\mathcal{D}}_k(\rho) = \underset{\sigma \in k\mbox{-sep}}{\mbox{min}} \ \mathcal{D}(\rho, \sigma),
\label{eq:gm}$$ gives a measure of $k$-*inseparable* multipartite entanglement in the state $\rho$. As two special cases, for $k=N$, Eq. (\[eq:gm\]) gives the minimum distance from fully-separable states, and thus quantifies the “total" multipartite entanglement, while for $k=2$, $\mathcal{E}^{\mathcal{D}}_2$ gives a measure of genuine multipartite entanglement. Optimization in Eq. (\[eq:gm\]) is a formidable problem for general multipartite states. But there exist forms of the geometric measure $\mathcal{E}^{\mathcal{D}}_k(\rho)$ for various families of states (pure and mixed) corresponding to certain distance measures [@GGM1; @GGM2; @GGM3]. For example, in case of pure states, if we take the following distance measure $$\mathcal{D}(\psi, \phi) = 1 - |\langle\psi|\phi\rangle|^2,$$ we get the “geometric measures" of multipartite entanglement for pure states [@GM1; @GM3; @GGM1; @GGM2]. In case the minimum distance is from bi-separable states, the corresponding measure has been termed as the generalized geometric measure [@GGM1; @GGM2], $$\mathcal{G}(\psi) = \underset{{\ensuremath{|\phi\rangle}} \ \in \ 2\mbox{-sep}}{\mbox{min}} \left(1-|\langle\psi|\phi\rangle|^2\right),
\label{eq:ggm1}$$ which measures the genuine multiparty entanglement in ${\ensuremath{|\psi\rangle}}$. In this case, we get a computable form of the measure for an arbitrary $N$-party pure state, ${\ensuremath{|\psi\rangle}}$, shared between $A_1, A_2, ..., A_N$, in arbitrary dimensions, given by [@GGM2] $$\mathcal{G}(\psi) = 1 - \mbox{max}\{\lambda^2_{A:B} | A \cup B = A_1, A_2, ..., A_N, A \cap B = \phi\},
\label{eq:ggm2}$$ where $\lambda_{A:B}$ is the maximum Schmidt coefficient in each possible bipartition split of the type $A : B$ of the given state ${\ensuremath{|\psi\rangle}}$.
Until now, in the case of multiparty mixed states, we have considered only the distance-based measures. However, it is also possible to use the convex-roof approach to define entanglement measures for multiparty mixed states, after choosing a certain measure for pure states [@GGM4; @GGM5].
Three qubit case: GHZ class vs. W class
---------------------------------------
For three-qubit pure states, the above classification of multipartite states boils down to three categories, namely
1. fully-separable states of the form ${\ensuremath{|\psi_A\rangle}} \otimes {\ensuremath{|\psi_B\rangle}} \otimes {\ensuremath{|\psi_C\rangle}}$,
2. bi-separable states of three types: $A$:$BC$, $B$:$AC$, and $C$:$AB$, where $A$:$BC$ type states are separable in the $A$:$BC$ splitting but not in others and so on,
3. and finally, genuine tripartite entangled states.
Another classification is possible by considering interconversion of states through stochastic local operations and classical communication (SLOCC) [@slocc], i.e, through LOCC but with a non-unit probability. In this scenario, we call two states ${\ensuremath{|\psi\rangle}}$ and ${\ensuremath{|\phi\rangle}}$ to be equivalent if there is a non-vanishing probability of success when trying to convert ${\ensuremath{|\psi\rangle}}$ into ${\ensuremath{|\phi\rangle}}$ as well as in the opposite direction through SLOCC. For example, in the two-qubit case, every entangled state is equivalent to any other entangled states, and the entropies of entanglement quantify these conversion rates (see Sec. \[sec:ent\_measures\]). This happy situation is absent already in the case of pure three-qubit states [@dur_vidal_cirac]. It turns out that any genuine three-qubit pure entangled state can be converted into either the GHZ state or the W state, but not both, using SLOCC. This divides the set of genuine three-qubit pure entangled states into two sets which are incompatible under SLOCC. In other words, if a state ${\ensuremath{|\psi\rangle}}$ is convertible into ${\ensuremath{|GHZ\rangle}}$ and another state ${\ensuremath{|\phi\rangle}}$ is convertible into ${\ensuremath{|W\rangle}}$ via SLOCC, then one cannot transform ${\ensuremath{|\psi\rangle}}$ to ${\ensuremath{|\phi\rangle}}$ or vice-versa, with any non-zero probability. These two sets of genuine three-qubit pure entangled states are termed as the GHZ-class and the W-class respectively. Figure \[fig:three\_qubit\] shows these different classes of three-qubit pure states and possible SLOCC transformations between the different classes.
![ Different classes of three-qubit pure states. Two states in the same class are SLOCC equivalent, i.e., those can be converted to one another under SLOCC operations. The direction of the arrows shows which non-invertible transformations between classes are possible via SLOCC. Reproduced figure with permission from the Authors of Ref. [@dur_vidal_cirac]. Copyright (2000) of the American Physical Society.[]{data-label="fig:three_qubit"}](three-qubit.pdf)
Dür et al. [@dur_vidal_cirac] presented general forms of the members of each class. A member of the GHZ-class can be expressed as $${\ensuremath{|\psi_{GHZ}\rangle}} = \sqrt{K}(c_\delta|0\rangle|0\rangle|0\rangle+s_\delta e^{i\varphi}|\varphi_A\rangle|\varphi_B\rangle|\varphi_C\rangle),$$ where $$\begin{aligned}
{\ensuremath{|\varphi_A\rangle}}&=&c_{\alpha}{\ensuremath{|0\rangle}}+s_{\alpha}{\ensuremath{|1\rangle}},\nonumber\\
{\ensuremath{|\varphi_B\rangle}}&=&c_{\beta}{\ensuremath{|0\rangle}}+s_{\beta}{\ensuremath{|1\rangle}},\nonumber\\
{\ensuremath{|\varphi_C\rangle}}&=&c_{\gamma}{\ensuremath{|0\rangle}}+s_{\gamma}{\ensuremath{|1\rangle}},\end{aligned}$$ $K=(1+2 c_\delta s_\delta c_\alpha c_\beta c_\gamma c_\varphi)^{-1} \in
(1/2,\infty)$ is a normalization factor, and $\delta \in (0,\pi/4]$, $\alpha,\beta,\gamma \in (0,\pi/2]$ and $\varphi \in
[0,2\pi)$. Here $c_{\alpha} = \cos \alpha$, $s_{\alpha} = \sin \alpha$, etc. Similarly, a member state of the W-class, upto a local unitary transformation, can be written as, $${\ensuremath{|\psi_W\rangle}} = \sqrt{a} {\ensuremath{|001\rangle}} +\sqrt{b} {\ensuremath{|010\rangle}} + \sqrt{c} {\ensuremath{|100\rangle}} +\sqrt{d} {\ensuremath{|000\rangle}},$$ where $a, b, c >0$, and $d = 1- (a + b+ c) \geq 0$.
For multipartite states with $N \geq 4$, there exist infinitely many inequivalent kinds of such entanglement classes under SLOCC [@dur_vidal_cirac]. See Refs. [@frank_9; @miyake] for further results.
Monogamy of quantum entanglement
--------------------------------
The concept of monogamy [@bennett_eof1; @terhal_mono; @ckw; @mono_review] is an inherent feature of multipartite quantum correlations, and in particular, of sharing of two-party entanglements in multiparty quantum states. Unlike classical correlations, quantum entanglement cannot be freely shared among many parties. For example, given three parties $A$, $B$, and $C$, if party $A$ is maximally entangled with party $B$, e.g., if they share a singlet state ${\ensuremath{|\Psi^-\rangle}} = ({\ensuremath{|01\rangle}} - {\ensuremath{|10\rangle}})/\sqrt{2}$, then $A$ cannot be simultaneously entangled with party $C$. In other words, there exist a trade-off between $A$’s entanglement with $B$ and its entanglement with $C$. In principle, and in its simplest form, for a two-party entanglement measure $\mathcal{E}$ and a three party system shared between $A$, $B$, and $C$, any relation providing an upper bound to the sum $\mathcal{E}_{A:B} + \mathcal{E}_{A:C}$ that is stronger than the sum of individual maxima of $\mathcal{E}_{A:B}$ and $\mathcal{E}_{A:C}$, can be termed as a monogamy relation for $\mathcal{E}$. However, in Ref. [@ckw], an intuitive reasoning for the validity of the relation $$\mathcal{E}_{A:B} + \mathcal{E}_{A:C} \leq \mathcal{E}_{A:BC},
\label{eq:monogamy}$$ is given. As in Ref. [@ckw], we will call an entanglement measure $\mathcal{E}$ to be monogamous in a certain three-party system, if the relation (\[eq:monogamy\]) is valid for all quantum states in that system.
**Theorem 17** [@ckw]\
*For an arbitrary three-qubit pure state shared between $A$, $B$, and $C$, the squared concurrence $\mathcal{C}_{A:B}^2$ between $A$ and $B$, plus the squared concurrence $\mathcal{C}_{A:C}^2$ between $A$ and $C$, cannot be greater than the squared concurrence $\mathcal{C}_{A:BC}^2$ between $A$ and the pair $BC$.*
It was shown later that $(\mathcal{C}^2_{A:B} + \mathcal{C}^2_{A:C}) \leq \mathcal{C}^2_{A:BC}$ holds also for arbitrary mixed three-qubit states [@frank_mono], where $\mathcal{C}^2_{A:BC}$ is defined via convex-roof extension. Using Theorem 17, one can define a positive quantity in terms of squared concurrence, named *tangle*, or *three-tangle*, for three-qubit pure states, as [@ckw] $$\tau_{ABC} = \mathcal{C}^2_{A:BC} - (\mathcal{C}^2_{A:B} + \mathcal{C}^2_{A:C}).
\label{eq:tangle}$$ The tangle $\tau_{ABC}$, also called “residual entanglement", is independent of the choice of the “node" or “focus", which is the party A here. It has been argued that the tangle $\tau_{ABC}$ gives a quantification of three-qubit entanglement. The generalization of the tangle to mixed states can e.g. be obtained by the convex roof extension, which is difficult to compute. The tangle is a proper entanglement monotone, as it does not increase on average under LOCC [@dur_vidal_cirac]. It also successfully distinguishes the two inequivalent SLOCC classes in three-qubit pure state scenario, namely the GHZ-class and the W-class. It has been shown that tangle vanishes for states in the W-class, whereas it is always non-zero for states in the GHZ-class [@dur_vidal_cirac]. Therefore, to quantify entanglement content of states from the W-class, one has to look for other multipartite entanglement measures, different from tangle.
In the $N$-party scenario, generalization of inequality (\[eq:monogamy\]) can be written as $$\mathcal{E}_{A_1:A_2} + \mathcal{E}_{A_1:A_3} + ... + \mathcal{E}_{A_1:A_N} \leq \mathcal{E}_{A_1:A_2A_3...A_N}.
\label{eq:monogamy2}$$ In the same spirit as for the definition of the tangle in Eq. (\[eq:tangle\]), we can define the “monogamy score" [@mono_score] corresponding to a bipartite entanglement measure $\mathcal{E}$ as $$\delta^{A_1}_{\mathcal{E}} = \mathcal{E}_{A_1:A_2A_3...A_N} - (\mathcal{E}_{A_1:A_2} + \mathcal{E}_{A_1:A_3} + ... + \mathcal{E}_{A_1:A_N}),
\label{eq:mono_score}$$ with party $A_1$ as “nodal". Like for the tangle, it has been argued that the monogamy score, $\delta^{A_1}_{\mathcal{E}}$, can act as a measure of multiparty entanglement [@ckw; @mono_score], obtained by subtracting the bipartite contributions $(\mathcal{E}_{A_1:A_2} , \mathcal{E}_{A_1:A_3} , ... , \mathcal{E}_{A_1:A_N})$ in the “total" entanglement $\mathcal{E}_{A_1:A_2A_3...A_N}$ in the $A_1:A_2A_3...A_N$ partition. Unlike the tangle, monogamy scores for certain entanglement measures can possess negative values for some $N$-party quantum states. For further information on recent works about the monogamy of quantum entanglement and correlations, see Refs. [@mono_review] and references therein.\
For further results about entanglement criteria, detection, and classification of multipartite states, see e.g. [@panch; @saat; @Mumbai; @satattor; @Horo-realign; @sakkhiprl; @ek; @chhoi; @char; @tin; @dui; @sotero], and references therein.
Problems
========
**Problem 1** Show that the singlet state ${\ensuremath{|\Psi^-\rangle}} = \frac{1}{\sqrt{2}} ({\ensuremath{|01\rangle}} - {\ensuremath{|10\rangle}})$ has non-positive partial transposition.
**Problem 2** Consider the Werner state $\rho^W_{AB}(p) = p |\Psi^- \rangle \langle \Psi^- | + (1 -p) I/4$ in $\mathbb{C}^2 \otimes \mathbb{C}^2$, where $0\leq p \leq 1$ [@Werner]. Find the values of the mixing parameter $p$, for which entanglement in the Werner state can be detected by the partial transposition criterion.
**Problem 3** Show that in ${\mathbb{C}}^2 \otimes {\mathbb{C}}^2$, the partial transposition of a density matrix can have at most one negative eigenvalue.
**Problem 4** Given two random variables $X$ and $Y$, show that the Shannon entropy of the joint distribution cannot be smaller than that of either.
**Problem 5** Prove Theorem 5.
**Problem 6** Consider the following state in $\mathbb{C}^{3}\otimes\mathbb{C}^{3}$: $$\rho_{AB}(\alpha) = \frac{2}{7}|\psi\rangle\langle \psi| + \frac{\alpha}{7}\varrho_+ + \frac{5 -\alpha}{7}\varrho_-,
\label{3x3state2}$$ where $ \varrho_+=(|01\rangle\langle 01|+|12\rangle\langle 12|+|20\rangle\langle 20|)/3$, $\varrho_-=(|10\rangle\langle 10| + |21\rangle\langle 21| + |02\rangle\langle 02|)/3$, $|\psi\rangle = \frac{1}{\sqrt{3}}\sum_{i = 0}^2 |ii\rangle$, and $0\leq\alpha\leq5$ [@be_horo]. Find the ranges of the parameter $\alpha$, for which entanglement in the state $\rho_{AB}(\alpha)$ can be detected by the majorization and the cross-norm criteria.
**Problem 7** Prove Corollaries 1 and 2.
**Problem 8** Prove Lemma 1.
**Problem 9** Prove Theorem 10.
**Problem 10** Consider the the following set of orthogonal product states in $\mathbb{C}^3 \otimes \mathbb{C}^3$ [@UPBPRL]: $$\begin{aligned}
{|\psi _{0}\rangle } &=&{\frac{1}{\sqrt{2}}}{|0\rangle } \otimes ({|0\rangle }-{\
|1\rangle }),\ \ {|\psi _{1}\rangle }={\frac{1}{\sqrt{2}}}({|0\rangle }-{\
|1\rangle })\otimes{|2\rangle }, \nonumber \\
{|\psi _{2}\rangle } &=&{\frac{1}{\sqrt{2}}|2\rangle }\otimes({|1\rangle }-{\
|2\rangle }),\text{ \ }{|\psi _{3}\rangle }={\frac{1}{\sqrt{2}}} \otimes ({|1\rangle }
-{|2\rangle }){|0\rangle }, \nonumber \\
{|\psi _{4}\rangle } &=&{\frac{1}{3}}({|0\rangle }+{|1\rangle }+{|2\rangle )} \otimes
({|0\rangle }+{|1\rangle }+{|2\rangle )}.\end{aligned}$$ Using the range criterion, show that the quantum state $$\rho =\frac{1}{4}(I-\sum_{i=0}^{4}{|\psi _{i}\rangle \langle \psi
_{i}|})$$ is entangled, where $I$ denotes the identity operator on $\mathbb{C}^3 \otimes \mathbb{C}^3$.
**Problem 11** Prove Theorem 15.
**Problem 12** Find the relative entropy of entanglement for Werner state $\rho^W_{AB}(p) = p |\Psi^- \rangle \langle \Psi^- | + (1 -p) I/4$ in $\mathbb{C}^2 \otimes \mathbb{C}^2$, where $0\leq p \leq 1$ [@Werner].
**Problem 13** Show that each of the shells depicted in figure \[fig:ghorar-dim\] are nonempty, and of nonzero measure. Show also that all the boundaries are convex.
**Problem 14** Show that the entanglement of formation is non-monogamous for three-qubit states.
**Problem 15** Consider the geometric measure of genuine multipartite entanglement $\mathcal{G}(\psi)$ given in Eq. (\[eq:ggm1\]), and then show that it assumes the computable closed form given in Eq. (\[eq:ggm2\]).
Acknowledgements {#acknowledgements .unnumbered}
================
ML acknowledge financial support from the John Templeton Foundation, the EU grants OSYRIS (ERC-2013-AdG Grant No. 339106), QUIC (H2020-FETPROACT-2014 No. 641122), and SIQS (FP7-ICT-2011-9 No. 600645), the Spanish MINECO grants FOQUS (FIS2013-46768-P), FÏSICATEAMO (FIS2016-79508-P), and “Severo Ochoa" Programme (SEV-2015-0522), the Generalitat de Catalunya support (2014 SGR 874) and CERCA/Programme, and Fundació Privada Cellex. AS acknowledges financial support from the Spanish MINECO projects FIS2013-40627-P,FIS2016-80681-P the Generalitat de Catalunya CIRIT (2014-SGR-966).
[999]{} A.K. Ekert, Phys. Rev. Lett. **67**, 661 (1991).
C.H. Bennett and S.J. Wiesner, Phys. Rev. Lett. **69**, 2881 (1992).
C.H. Bennett, G. Brassard, C. Crepeau, R. Josza, A Peres, and W.K. Wootters, Phys. Rev. Lett. **70**, 1895 (1993).
M. Żukowski, A. Zeilinger, M.A. Horne and A.K. Ekert, Phys. Rev. Lett. **71**, 4287 (1993); M. Żukowski, A. Zeilinger, and H. Weinfurter, Annals N.Y. Acad. Sci. **755**, 91 (1995); S. Bose, V. Vedral, and P.L. Knight, Phys. Rev. A **57**, 822 (1998); S. Bose, V. Vedral, and P.L. Knight, Phys. Rev. A **60**, 194 (1999).
R.F. Werner, Phys. Rev. A **40**, 4277(1989).
P. Horodecki, Phys. Lett. A **232**, 333 (1997).
S. Karnas and M. Lewenstein, J. Phys. A **34**, 6919 (2001).
A. Peres, Phys. Rev. Lett **77**, 1413 (1996).
M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Lett. A **223**, 1 (1996).
M.A. Nielsen and J. Kempe, Phys. Rev. Lett. **86**, 5184 (2001).
O. Rudolph, J. Phys. A: Math. Gen. [**33**]{} 3951 (2000).
K. Chen and L.-A. Wu, Quant. Inf. Comput., **3**, 193 (2003).
O. Rudolph, Quant. Info. Process., [**4**]{}, 219 (2005).
O. Gühne, P. Hyllus, O. Gittsovich, and J. Eisert, Phys. Rev. Lett. [**99**]{}, 130504 (2007).
O. Gittsovich, O. Gühne, P. Hyllus, and J. Eisert, Phys. Rev. A [**78**]{}, 052319 (2008).
M. Horodecki and P. Horodecki, Phys. Rev. A **59**, 4206 (2000).
A. Doherty, P. Parillo, and F. Spedalieri, Phys. Rev. Lett. **88**, 187904 (2002); Phys. Rev. A **69**, 022308 (2004); F. Hulpke and D. Bru[ß]{}, J. Phys. A: Math. Gen. **38**, 5573 (2005).
H. F. Hofmann and S. Takeuchi, Phys. Rev. A [**68**]{}, 032103 (2003).
J. I. de Vicente, Quantum Inf. Comput. [**7**]{}, 624 (2007).
J. I. de Vicente, J. Phys. A [**41**]{}, 065309 (2008).
C.-J. Zhang, Y.-S. Zhang, S. Zhang, and G.-C. Guo, Phys. Rev. A [**76**]{}, 012334 (2007); C.-J. Zhang, Y.-S. Zhang, S. Zhang, and G.-C. Guo, Phys. Rev. A [**77**]{}, 060301(R) (2008).
M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. **80**, 5239 (1998), and references therein.
K. Życzkowski, P. Horodecki, A. Sanpera, and M. Lewenstein, Phys. Rev. A, **58**, 883 (1998); K. Życzkowski, Phys. Rev. A **60**, 3496 (1999); S. Szarek, Phys. Rev. A [**72**]{}, 032304 (2005), and references therein.
D.P. DiVincenzo, P.W. Shor, J.A. Smolin, B.M. Terhal, and A.V. Thapliyal, Phys. Rev. A **61**, 062312 (2000); W. Dür, J.I. Cirac, M. Lewenstein, and D. Bru[ß]{}, *ibid.* **61**, 062313 (2000). C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, J. A. Smolin, and W. K. Wootters, Phys. Rev. Lett. [**76**]{}, 722 (1996). C. H. Bennett, D. P. DiVincenzo, J. A. Smolin, and W. K. Wootters, Phys. Rev. A [**54**]{}, 3824 (1996).
A.W. Marshall and I. Olkin, *Inequalities: Theory of Majorization and Its Applications* (Academic Press, New York, 1979); P.M. Alberti and A. Uhlmann, *Stochasticity and Partial order: Doubly Stochastic Maps and Unitary Mixing* (Dordrecht, Boston, 1982); R. Bhatia, *Matrix Analysis* (Springer, New York, 1997).
T. Hiroshima, Phys. Rev. Lett. **91**, 057902 (2003).
R. Horodecki and P. Horodecki, Phys. Lett. A [**194**]{}, 147 (1994).
N. Gisin, Phys. Lett. A [**210**]{}, 151 (1996); M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. [**78**]{}, 574 (1997); A. Kent, N. Linden, and S. Massar, Phys. Rev. Lett. [**83**]{}, 2656 (1999); F. Verstraete, J. Dehaene, and B. De Moor, Phys. Rev. A [**65**]{}, 032308 (2002).
O. G[ü]{}hne, P. Hyllus, D. Bru[ß]{}, A. Ekert, M. Lewenstein, C. Macchiavello, and A. Sanpera, Phys. Rev. A **66**, 062305 (2002); O. G[ü]{}hne, P. Hyllus, D. Bru[ß]{}, A. Ekert, M. Lewenstein, C. Macchiavello, and A. Sanpera, J. Mod. Opt. **50**, 1079 (2003).
M. Bourennane, M. Eibl, C. Kurtsiefer, S. Gaertner, H. Weinfurter, O G[ü]{}hne, P. Hyllus, D. Bru[ß]{}, M. Lewenstein, and A. Sanpera, Phys. Rev. Lett. **92**, 087902 (2004).
H.W. Alt, *Lineare Funktionalanalysis*, (Springer-Verlag, 1985).
S.L. Woronowicz, Commun. Math. Phys. **51**, 243 (1976); P. Kryszynski and S.L. Woronowicz, Lett. Math. Phys. **3**, 319 (1979); M.D. Choi, Proc. Sympos. Pure Math. **38**, 583 (1982).
B.M. Terhal, Lin. Alg. Appl. **323**, 61 (2001).
M. Lewenstein, B. Kraus, J.I. Cirac, and P. Horodecki, Phys. Rev. A **62**, 052310 (2000); D. Bru[ß]{}, J.I. Cirac, P. Horodecki, F. Hulpke, B. Kraus, M. Lewenstein, and A. Sanpera, J. Mod. Opt. **49**, 1399 (2002).
E. C. G. Sudarshan, P. M. Mathews, and J. Rau, Phys. Rev. [**121**]{}, 920 (1961); A. Jamio[ł]{}kowski, Rep. Math. Phys., **3**, 275 (1972); M.-D. Choi, Linear Alg. Appl. [**10**]{}, 285 (1975); K. Kraus, *States, Effects, and Operations: Fundamental Notions of Quantum Theory*, *Lecture notes in Physics*, vol. 190 (Spring-Verlag, New York, 1983).
A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. **57**, 777 (1935).
J.S. Bell, Physics **1**, 195 (1964).
E. Santos, Phys. Rev. Lett. **66**, 1388 (1991); E. Santos, Phys. Rev. A **46**, 3646 (1992); P.M. Pearle, Phys. Rev. D **2**, 1418 (1970); J.F. Clauser and M.A. Horne, Phys. Rev. D **10**, 526 (1974); P.G. Kwiat, P.H. Eberhard, A.M. Steinberg, and R.Y. Chiao, Phys. Rev. A **49**, 3209 (1994); N. Gisin and B. Gisin, Phys. Lett. A **260**, 323 (1999); S. Massar, S. Pironio, J. Roland, and B. Gisin, Phys. Rev. A **66**, 052112 (2002); R. Garcia-Patrón, J. Fiurásek, N.J. Cerf, J. Wenger, R. Tualle-Brouri, and Ph. Grangier, Phys. Rev. Lett. **93**, 130409 (2004); N. Brunner, N. Gisin, V. Scarani, and C. Simon, Phys. Rev. Lett. [**98**]{}, 220403 (2007); R. Garcia-Patrón, J. Fiurásek, and N. J. Cerf, Phys. Rev. A [**71**]{}, 022105 (2005); T. Vértesi, S. Pironio, and N. Brunner, Phys. Rev. Lett. [**104**]{}, 060401 (2010); N. Sangouard, J.-D. Bancal, N. Gisin, W. Rosenfeld, P. Sekatski, M. Weber, and H. Weinfurter, Phys. Rev. A [**84**]{}, 052122 (2011); A. Cabello and F. Sciarrino, Phys. Rev. X [**2**]{}, 021010 (2012); J. Larsson, J. Phys. A [**47**]{}, 424003 (2014).
S.J. Freedman and J.S. Clauser, Phys. Rev. Lett. **28**, 938 (1972); A. Aspect, P. Grangier, and G. Roger, Phys. Rev. Lett. **47**, 460 (1981); *ibid.* **49**, 91 (1982); A. Aspect, J. Dalibard, and G. Roger, *ibid.* **49**, 1804 (1982); P.G. Kwiat, K. Mattle, H. Weinfurter, A. Zeilinger, A.V. Sergienko, and Y. Shih, Phys. Rev. Lett. **75**, 4337 (1995); G. Weihs, T. Jennewein, C. Simon, H. Weinfurter, and A. Zeilinger, Phys. Rev. Lett. **81**, 5039 (1998); W. Tittel, J. Brendel, B. Gisin, T. Herzog, H. Zbinden, and N. Gisin, Phys. Rev. A **57**, 3229 (1998); A. Zeilinger Rev. Mod. Phys. **71**, S288 (1999); M.A. Rowe, D. Kielpinski, V. Meyer, C.A. Sackett, W.M. Itano, C. Monroe, and D.J. Wineland, Nature **409**, 791 (2001); M. Giustina *et al.*, Nature [**497**]{}, 227 (2013); B. Hensen *et al.*, Nature [**526**]{}, 682 (2015); M. Giustina *et al.*, Phys. Rev. Lett. [**115**]{}, 250401 (2015); L. K. Shalm *et al.*, Phys. Rev. Lett. [**115**]{}, 250402 (2015); H.-P. Lo, C.-M. Li, A. Yabushita, Y.-N. Chen, C.-W. Luo, T. Kobayashi, Scientific Reports [**6**]{}, 22088 (2016).
J.F. Clauser, M.A. Horne, A. Shimony, and R.A. Holt, Phys. Rev. Lett. **23**, 880 (1969).
J.S. Bell, in *Foundations of Quantum Mechanics*, ed. B. d’Espagnat (Academic, New York, 1971).
P. Hyllus, O. G[ü]{}hne, D. Bru[ß]{}, and M. Lewenstein Phys. Rev. A **72**, 012321 (2005).
R. Horodecki, P. Horodecki, and M. Horodecki, Phys. Lett. A [**200**]{}, 340 (1995).
C. H. Bennett, G. Brassard, S. Popescu, B. Schumacher, J. Smolin, and W. K. Wootters, Phys. Rev. Lett 78, 2031 (1996).
C. H. Bennett, H. J. Bernstein, S. Popescu, and B. Schumacher, Phys. Rev. A [**53**]{}, 2046 (1996).
V. Vedral, M. B. Plenio, M. A. Rippin, and P. L. Knight, Phys. Rev. Lett. [**78**]{}, 2275 (1997).
M. Horodecki, P. Horodecki, and R. Horodecki, Phys. Rev. Lett. [**84**]{}, 2014 (2000).
R. Horodecki, P. Horodecki, M. Horodecki, and K. Horodecki, Rev. Mod. Phys. [**81**]{}, 865 (2009).
A. Uhlmann, Open Sys. Inf. Dyn. [**5**]{}, 209 (1998). S. Hill and W. K. Wootters, Phys. Rev. Lett. [**78**]{}, 5022 (1997).
W. K. Wootters, Phys. Rev. Lett. [**80**]{}, 2245 (1998).
P. M. Hayden, M. Horodecki, and B. M. Terhal, J. Phys. A: Math. Gen. [**34(35)**]{}, 6891 (2001).
E. M. Rains, Phys. Rev. A, [**60**]{}, 173 (1999).
M. B. Plenio and S. Virmani, Quant. Inf. Comput. [**7**]{}, 1 (2007).
G. Vidal and J. I. Cirac, Phys. Rev. Lett. [**86**]{}, 5803 (2001); G. Vidal and J. I. Cirac, Phys. Rev. A [**65**]{}, 012323 (2001).
A. Shimony, Ann. N.Y. Acad. Sci. [**755**]{}, 675 (1995).
V. Vedral and M. B. Plenio, Phys. Rev. A [**57**]{}, 1619 (1998).
V. Vedral, Rev. Mod. Phys. [**74**]{}, 197 (2002).
M. Nielsen and I. Chuang, *Quantum Computation and Quantum Information* (Cambridge University Press, Cambridge, 2000), ISBN 9781139495486.
A. Miranowicz and S. Ishizaka, Phys. Rev. A [**78**]{}, 032310 (2008); S. Friedland and G. Gour, J. Math. Phys. [**52**]{}, 052201 (2011).
K. Życzkowski, P. Horodecki, A. Sanpera, and M. Lewenstein, Phys. Rev. A [**58**]{}, 883 (1998).
J. Eisert and M. B. Plenio, J. Mod. Opt. [**46**]{}, 145 (1999); J. Lee, M. S. Kim, Y. J. Park, and S. Lee, J. Mod. Opt. [**47**]{}, 2151 (2000).
G. Vidal and R.F. Werner, Phys. Rev. A [**65**]{}, 032314 (2002).
M.B. Plenio, Phys. Rev. Lett. [**95**]{}, 090503 (2005).
C.H. Bennett, P.W. Shor, J.A. Smolin, and A.V. Thapliyal, Phys. Rev. Lett. **83**, 3081 (1999).
C.H. Bennett, P.W. Shor, J.A. Smolin, and A.V. Thapliyal, *Entanglement-assisted capacity of a quantum channel and the reverse Shannon theorem*, quant-ph/0106052.
M. Horodecki, P. Horodecki, R. Horodecki, D. Leung, and B. Terhal, Q. Inf. and Comput. **1**, 70 (2001).
A. Winter, J. Math. Phys. **43**, 4341 (2002).
S. Bose, M.B. Plenio, and V. Vedral, J. Mod. Opt. **47**, 291 (2000).
T. Hiroshima, J. Phys. A: Math. Gen. **34**, 6907 (2001).
X.S. Liu, G.L. Long, D.M. Tong, and F. Li, Phys. Rev. A **65**, 022304 (2002).
M. Ziman and V. Bu[ž]{}ek, Phys. Rev. A **67**, 042321 (2003).
J.P. Gordon, in *Proc. Int. School Phys. “Enrico Fermi, Course XXXI”*, ed. P.A. Miles, p. 156 (Academic Press, NY 1964); L.B. Levitin, in *Proc. VI National Conf. Inf. Theory, Tashkent*, p. 111 (1969); A.S. Holevo, Probl. Pereda. Inf. **9**, 3 1973 \[Probl. Inf. Transm. **9**, 110 (1973)\].
T. M. Cover and J. A. Thomas, *Elements of Information Theory* (Wiley, New York, 1991).
R. Josza, D. Robb, and W.K. Wotters, Phys. Rev. A, **49**, 668 (1994).
B. Schumacher, M. Westmoreland, and W. K. Wootters, Phys. Rev. Lett. **76**, 3452 (1996).
P. Badzi[a]{}g, M. Horodecki, A. Sen(De), and U. Sen, Phys. Rev. Lett. **91**, 117901 (2003).
M Horodecki, J. Oppenheim, A. Sen(De), and U. Sen, Phys. Rev. Lett. **93**, 170503 (2004).
B. Schumacher and M.D. Westmoreland, Phys. Rev. A **56**, 131 (1997). A.S. Holevo, IEEE Trans. Inf. Theory **44**, 269 (1998).
D. Bru[ß]{}, G.M. D’Ariano, M. Lewenstein, C. Macchiavello, A. Sen(De), and U. Sen, Phys. Rev. Lett. **93**, 210501 (2004).
D. Bru[ß]{}, G.M. D’Ariano, M. Lewenstein, C. Macchiavello, A. Sen(De), and U. Sen, Int. J. Quant. Inf. [**4**]{}, 415 (2006).
D.M. Greenberger, M.A. Horne, and A. Zeilinger, in *Bell’s Theorem, Quantum Theory, and Conceptions of the Universe*, ed. M. Kafatos (Kluwer Academic, Dordrecht, 1989).
W. Dür, G. Vidal, and J. I. Cirac, Phys. Rev. A [**62**]{}, 062314 (2000).
A. Zeilinger, M. Horne, and D. Greenberger, in *Squeezed States and Quantum Uncertainty*, edited by D. Han, Y. S. Kim, and W. W. Zachary (NASA Conference Publication 3135, NASA, College Park, 1992).
D. Kaszlikowski, A. Sen(De), U. Sen, V. Vedral, and A. Winter, Phys. Rev. Lett. [**101**]{}, 070502 (2008).
M. B. Plenio and V. Vedral, J. Phys. A: Math. Gen. [**34**]{} 6997 (2001).
H. Barnum and N. Linden, J. Phys. A [**34**]{}, 6787 (2001); T.-C. Wei and P.M. Goldbart, Phys. Rev. A [**68**]{}, 042307 (2003).
M. Blasone, F. Dell’Anno, S. De Siena, and F. Illuminati, Phys. Rev. A [**77**]{}, 062304 (2008).
A. Sen(De) and U. Sen, Phys. Rev. A [**81**]{}, 012308 (2010); A. Sen(De) and U. Sen, *Bound Genuine Multisite Entanglement: Detector of Gapless-Gapped Quantum Transitions in Frustrated Systems*, arXiv:1002.1253 \[quant-ph\].
M. Cianciaruso, T. R. Bromley, and G. Adesso, Nature npj Quantum Information [**2**]{}, 16030 (2016).
L. E. Buchholz, T. Moroder, and O. Gühne, Ann. Phys. (Berlin) [**528**]{}, 278 (2016).
T. Das, S. S. Roy, S. Bagchi, A. Misra, A. Sen(De), U. Sen, Phys. Rev. A [**94**]{}, 022336 (2016).
C. H. Bennett, S. Popescu, D. Rohrlich, J. A. Smolin, and A. V. Thapliyal, Phys. Rev. A [**63**]{}, 012307 (2000).
F. Verstraete, J. Dehaene, B. De Moor, and H. Verschelde, Phys. Rev. A [**65**]{}, 052112 (2002); D. Li, X. Li, H. Huang, and X. Li, Phys. Lett. A [**359**]{}, 428 (2006); L. Lamata, J. León, D. Salgado, and E. Solano, Phys. Rev. A [**75**]{}, 022318 (2007).
A. Miyake, Phys. Rev. A [**67**]{}, 012108 (2003).
V. Coffman, J. Kundu, and W. K. Wootters, Phys. Rev. A [**61**]{}, 052306 (2000).
B. M. Terhal, *Is Entanglement Monogamous?*, arxiv:quant-ph/0307120 (2013), and references therein.
J. S. Kim, G. Gour, and B. C. Sanders, Contemp. Phys. [**53**]{}, 417 (2012); H. S. Dhar, A. K. Pal, D. Rakshit, A. Sen(De), and U. Sen, arXiv:1610.01069 (2016).
T. J. Osborne and F. Verstraete, Phys. Rev. Lett. [**96**]{}, 220503 (2006).
M. N. Bera, R. Prabhu, A. Sen(De), and U. Sen, Phys. Rev. A [**86**]{}, 012319 (2012).
W. D[ü]{}r, J.I. Cirac, and R. Tarrach, Phys. Rev. Lett. **83**, 3562 (1999).
R.F. Werner and M.M. Wolf, Phys. Rev. A **64**, 032112 (2001); M. Żukowski and [Č]{}. Brukner, Phys. Rev. Lett. **88**, 210401 (2002)
G. Svetlichny, Phys. Rev. D **35**, 3066 (1987); D. Collins, N. Gisin, S. Popescu, D. Roberts, and V. Scarani Phys. Rev. Lett. **88**, 170405 (2002); M. Seevinck and G. Svetlichny, Phys. Rev. Lett. **89**, 060401 (2002); S.M. Roy, Phys. Rev. Lett. **94**, 010402 (2005).
A. Ac[í]{}n, D. Bru[ß]{}, M. Lewenstein, and A. Sanpera, Phys. Rev. Lett. **87**, 040401 (2001).
M. Horodecki, P. Horodecki, and R. Horodecki, Open Sys. Inf. Dyn. [**13**]{}, 103 (2006).
C. Moura Alves and D. Jaksch Phys. Rev. Lett. **93**, 110501 (2004).
W. Laskowski, T. Paterek, M. Żukowski, and [Č]{}. Brukner, Phys. Rev. Lett. **93**, 200401 (2004).
F.G.S.L. Brandao and R.O. Vianna, Phys. Rev. Lett. **93**, 220503 (2004).
G. Toth and O G[ü]{}hne, Phys. Rev. Lett. **94**, 060501 (2005).
A.C. Doherty, P.A. Parrilo, and F.M. Spedalieri, Phys. Rev. A **71**, 032333 (2005).
O. G[ü]{}hne, G. Toth, and H.J. Briegel, New J. Phys. [**7**]{}, 229 (2005).
C.H. Bennett, D.P. DiVincenzo, T. Mor, P.W. Shor, J.A. Smolin, and B.M. Terhal, Phys. Rev. Lett. **82**, 5385 (1999).
[^1]: \[foot:ent\_swap\] Because of the phenomenon of entanglement swapping [@ZZHE], one must suitably enlarge the notion of preparation of entangled states. So, an entangled state between two particles can be prepared if and only if, either the two particles (call them A and B) themselves came together to interact at a time in the past, or two *other* particles (call them C and D) does the same, with C (D) having interacted beforehand with A (B).
[^2]: The von Neumann entropy of a state $\rho$ is $S(\rho) = -\mbox{tr}\rho \log_2 \rho$.
[^3]: The Shannon entropy of a random variable $X$, taking up values $X_i$, with probabilities $p_i$, is given by $H(X) = H(\{p_i\}) = - \sum_i p_i \log_2 p_i$.
[^4]: A matrix $D= (D_{ij})$ is said to be doubly stochastic, if $D_{ij}\geq 0$, and $\sum_i D_{ij} = \sum_j D_{ij} =1$.
[^5]: A hyperplane is a linear subspace with dimension one less than the dimension of the space itself.
[^6]: We do not derive here the original Bell inequality, which Bell derived in 1964 [@Bell]. Instead, we derive the stronger form of the Bell inequality which Clauser, Horne, Shimony, and Holt (CHSH) derived in 1969 [@CHSHineq]. A similar derivation was also given by Bell himself in 1971 [@Bell71].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For a sequence $a_n$ satisfying a linear recurrence relation over ${\mathbb{Q}}$, we prove some results about the residue classes $a_p\!\!\mod p$ as $p$ ranges over the primes.'
author:
- Julian Rosen
title: Terms in prime position in a recurrent sequence
---
Introduction
============
A *recurrent sequence* (over ${\mathbb{Q}}$) is an infinite sequence $a_0,a_1,\ldots\in{\mathbb{Q}}$ that satisfies a linear recurrence relation $$\label{eqrec}
a_n=c_1a_{n-1}+\ldots+c_k a_{n-k}, {\hspace{5mm}}n\geq k,$$ with $c_i\in {\mathbb{Q}}$. The most famous example is the sequence of Fibonacci numbers $F_n$, satisfying $F_0=0$, $F_1=1$, and $$F_n=F_{n-1}+F_{n-2},{\hspace{5mm}}n\geq 2.$$ It is a classical result (see [@Rib12], p. 60) that for $p\neq 2,5$ a prime, the Fibonacci numbers satisfy $$\label{eqfib}
F_p\equiv\begin{cases}1&:p\equiv \pm 1\mod 5,\\
-1&:p\equiv \pm 2\mod 5.
\end{cases}$$ The purpose of this paper is to generalize to an arbitrary recurrent sequence.
Results
-------
Our first result is a classification of the possible sets of primes $$\{p:a_p\equiv r\mod p\}$$ when $(a_n)$ is a recurrent sequence and $r\in{\mathbb{Q}}$. Adding a constant to a recurrent sequence gives another recurrent sequence, so it suffices to consider the case $r=0$.
A set $P$ of primes is *Frobenian* if there exist a finite Galois extension $L/{\mathbb{Q}}$ and a subset $C\subset{\mathrm{Gal}}(L/{\mathbb{Q}})$ closed under conjugation such that $P$ has finite symmetric difference with the set $$\{p:\phi_p\subset C\},$$ where $\phi_p\subset{\mathrm{Gal}}(L/{\mathbb{Q}})$ is the conjugacy class of the Frobenius at $p$.
The Chebotarev density theorem implies that the natural density of a Frobenian set exists and is a rational number, and that the density is positive if the set is infinite.
We prove the following theorem.
\[thintro1\] Let $P$ be a set of primes. Then there exists a recurrent sequence $(a_n)$ with $P=\{p:a_p\equiv0 \mod p\}$ if and only if $P$ is Frobenian.
The Skolem-Mahler-Lech Theorem describes the possible sets $\{n:a_n=0\}$. The theorem is not effective, and given $(a_n)$, there is no known algorithm to determine whether $\{n:a_n=0\}$ is empty. By constrast, Theorem \[thintro1\] is effective, and there is an algorithm to determine whether $\{p:a_p\equiv 0\mod p\}$ is empty.
If $(a_n)$ is an arbitrary recurrent sequence, there need not exist a finite list $r_1,\ldots,r_k\in{\mathbb{Q}}$ such that for every $p$, $a_p\equiv r_i\mod p$ for some $i$. However, our next theorem says that there is always a polynomial $f(x)$ such that $a_p$ is a root of $f(x)$ modulo $p$ for every $p$. Recall that the characteristic polynomial of the recurrence is the polynomial $$X^k-c_1X^{k-1}-c_2X^{k-2}-\ldots-c_k\in{\mathbb{Q}}[x].$$
\[thintro2\] For every recurrent sequence $(a_n)$, there exists a non-zero polynomial $f(x)\in{\mathbb{Q}}[x]$ such that $$\label{eqfp}
f(a_p)\equiv 0\mod p$$ for all primes $p$. Moreover, every $f$ satisfying for all $p$ has a rational root. If the characteristic polynomial of the recurrence satisfied by $(a_n)$ splits over an abelian extension of ${\mathbb{Q}}$, we can choose $f$ to have all rational roots.
The polynomial $f(x)$ given by Theorem \[thintro2\] is usually not irreducible, and for each of its irreducible factors $f_i(x)$ (at least one of which must be linear), the set $\{p:f_i(a_p)\equiv 0\mod p\}$ is Frobenian, hence has rational density. Our next result concerns the density of this set.
\[thintro3\] Fix a polynomial $f(x)\in{\mathbb{Q}}[x]$ with no rational root. Then for every recurrent sequence $(a_n)$, there is a strict inequality $$\label{eqineq}
\delta\bigg(\big\{p:f(a_p)\equiv 0\!\!\mod p\big\}\bigg) < \delta\bigg(\big\{p:f\text{ has a root mod } p\big\}\bigg),$$ where $\delta$ denotes natural density. The inequality is sharp: given $f$, we can choose the sequence $(a_n)$ so that is arbitrarily close to an equality.
Finite periods
--------------
A *period* is a complex number that, roughly speaking, arises from integrating an algebraic differential form. The set of all periods is a countable subalgebra of ${\mathbb{C}}$ containing ${\overline{\mathbb{Q}}}$. It has been observed (see e.g. [@Jar15b], [@Kan16], [@Ros18a]) that certain elements of the ring $${\mathcal{A}}:=\frac{\prod_p{\mathbb{Z}}/p{\mathbb{Z}}}{\bigoplus_p{\mathbb{Z}}/p{\mathbb{Z}}}$$ act like analogues of periods. Each recurrent sequence $(a_n)$ is $p$-integral for all sufficiently large $p$, so determines an element $$(a_n)_{\mathcal{A}}:=\big(a_p\!\!\mod p\big)_p\in{\mathcal{A}}.$$ In Sec. \[secfrob\], we explain how the elements $(a_n)_{\mathcal{A}}$ should be viewed as the ${\mathcal{A}}$-valued analogues of the algebraic numbers.
Functions on a Galois group {#secfungal}
===========================
In this section we show that a recurrent sequence determines a function on a Galois group. Let $L/{\mathbb{Q}}$ be a finite Galois extension, with ring of integers ${\mathcal{O}}_L$ and Galois group $\Gamma:={\mathrm{Gal}}(L/{\mathbb{Q}})$.
We define $A(L)$ to be the set of functions $g:\Gamma\to L$ satisfying $$\label{eqg}
g(\sigma\tau\sigma^{-1})=\sigma(g(\tau))$$ for all $\sigma$, $\tau\in \Gamma$. Pointwise addition and multiplication make $A(L)$ into a commutative ${\mathbb{Q}}$-algebra of finite dimension.
For $g\in A(L)$, let $p$ be a rational prime unramified in $L$ that is coprime to the denominators of all values of $g$. Let $\mathfrak{P}$ be a prime of $L$ over $p$, with Frobenius element $\phi_\mathfrak{P}\in \Gamma$. It follows from that the residue class $$\label{eqres}
g(\phi_\mathfrak{P})\mod \mathfrak{P}$$ is fixed by $\phi_\mathfrak{P}$, so is an element of ${\mathbb{Z}}/(p)\subset {\mathcal{O}}_L/\mathfrak{P}$. It can be checked that the value of $g(\phi_\mathfrak{P})$ mod $\mathfrak{P}$ is independent of the choice of $\mathfrak{P}|p$ (see [@Ros17], §4), and we write $g(\phi_p)$ mod $p$ for this residue class in ${\mathbb{Z}}/(p)$. We leave $g(\phi_p)$ mod $p$ undefined for the finitely many primes that are ramified in $L$ or not coprime to the denominators of $g$.
The following result is the key ingredient in the proofs that appear in Sec. \[secproofs\].
\[thseqA\] If $(a_n)$ is a recurrent sequence over ${\mathbb{Q}}$, there exist a finite Galois extension $L/{\mathbb{Q}}$ and an element $g\in A(L)$ such that $$\label{eqcong}
a_p\equiv g(\phi_p)\mod p$$ for all sufficiently large $p$. Conversely, given $L$ and $g\in A(L)$, there exists a recurrent sequence $(a_n)$ such that holds for all sufficiently large $p$.
Let $(a_n)$ be a recurrent sequence. There exist column vectors $u$, $v$ and an invertible matrix $M$ (with entries in ${\mathbb{Q}}$) such that $$a_n=u^T M^nv$$ for all $n\in{\mathbb{Z}}$. There is a Jordan-Chevalley decomposition $$M=M_{ss}M_u,$$ where $M_{ss}$ is semi-simple, $M_u$ is unipotent, and $M_{ss}$ commutes with $M_u$. For every prime $p$ larger than the size of $M_u$ that is coprime to all denominators appearing in $M_u$, the $p$-th power $M_u^p$ is congruent to the identity matrix modulo $p$, and if in addition $p$ is coprime to denominators appearing in $u$ and $v$, then $$\label{eqss}
a_p\equiv u^TM_{ss}^pv\mod p.$$
Let $L$ be a finite Galois extension of ${\mathbb{Q}}$ over which $M_{ss}$ diagonalizes, let $\lambda_1,\ldots,\lambda_k\in L$ be the eigenvalues of $M_{ss}$, and write $\Gamma={\mathrm{Gal}}(L/{\mathbb{Q}})$. Using the Jordan normal form of $M_{ss}$, it follows from that there are elements $b_1,\ldots,b_k\in L$ such that $$a_p\equiv \sum_i b_i \lambda_i^p \mod p,$$ and $\Gamma$ permutes the pairs $b_i$, $\lambda_i$, i.e. the element $$\alpha:=\sum_i b_i\otimes\lambda_i\in L\otimes_{\mathbb{Q}}L$$ is invariant under the diagonal action of $\Gamma$. There is a canonical isomorphism $$\begin{aligned}
\varphi:L\otimes_{\mathbb{Q}}L&\to {\mathrm{Hom}}(\Gamma,L),\\
x\otimes y&\mapsto \bigg(\sigma\mapsto x\sigma(y)\bigg)\end{aligned}$$ taking the $\Gamma$-invariant elements of $L\otimes L$ to $A(L)$, and we let $g=\varphi(\alpha)\in A(L)$. If $p$ is a rational prime unramified in $L$ coprime to every denominator of the values of $g$, then for every prime $P$ of $L$ over $P$, $$\begin{aligned}
g(\phi_P)&=\sum_i b_i \phi_P(\lambda_i)\\
&\equiv\sum_i b_i \lambda_i^p\mod P\\
&\equiv a_p\mod P.\end{aligned}$$ This proves the first part of the theorem.
Conversely, suppose $g\in A(L)$ is given, and let $$\varphi^{-1}(g)=\sum_i b_i\otimes \lambda_i\in (L\otimes L)^\Gamma.$$ Then the sequence $$a_n:= \sum_i b_i \lambda_i^n$$ is recurrent, and takes values in ${\mathbb{Q}}$ because $\sum b_i\otimes\lambda_i$ is $\Gamma$-invariant. By the computation above, we see that $$a_p\equiv g(\phi_p)\mod p$$ for all sufficiently large $p$. This completes the proof.
Proofs of the theorems {#secproofs}
======================
In this section, we prove Theorems \[thintro1\], \[thintro2\], and \[thintro3\].
Theorem \[thintro1\] is a direct consequence of Theorem \[thseqA\].
Suppose $(a_n)$ is a recurrent sequence. Let $g\in A(L)$ be as in the statement of Theorem \[thseqA\]. Then for all $p$ unramified in $L$ coprime to the numerators and denominators of all non-zero values of $L$ and all $\mathfrak{P}|p$, we have $$a_p\equiv 0\mod p\Leftrightarrow g(\phi_\mathfrak{P})=0.$$ So we may take $C=\{\sigma\in{\mathrm{Gal}}(L/{\mathbb{Q}}):g(\sigma)=0\}$, which is a union of conjugacy classes by .
Conversely, suppose $L/{\mathbb{Q}}$ and $C\subset{\mathrm{Gal}}(L/{\mathbb{Q}})$ are given. Let $g\in A(L)$ be the characteristic function of $C$, and let $a_n$ be a recurrent sequence such that $$a_p\equiv g(\phi_p)\mod p$$ for all but finitely many $p$ (which exists by Theorem \[thseqA\]). Then $\{p:a_p\equiv 0\mod p\}$ has finite symmetric difference with $\{p:\phi_p\subset C\}$. We can multiply the sequence through by a constant rational number to modify $\{p:a_p\equiv 0\mod p\}$ by any finite set. This completes the proof.
Theorem \[thintro2\] also follows from Theorem \[thseqA\].
Given a recurrent sequence $(a_n)$, let $L/{\mathbb{Q}}$ and $g\in A(L)$ be given by Theorem \[thseqA\]. Since $A(L)$ is finite-dimensional, there is a non-zero $f(x)\in{\mathbb{Q}}[x]$ such that $f(g)=0$, which implies $$\label{eq0}
f(a_p)\equiv f(g(\phi_p))\equiv 0\mod p$$ for all sufficiently large $p$. We can scale $f(x)$ by a rational constant to make hold for all $p$.
Now suppose we are given $f(x)\in{\mathbb{Q}}[x]$ with $f(a_p)\equiv 0$ mod $p$ for all $p$. There are infinitely many primes $p$ that split completely in $L$, and for all but finitely many of these $p$, we have $$\label{zerop}
f(a_p)\equiv f(g(1))\equiv 0\mod p,$$ where $1\in{\mathrm{Gal}}(L/{\mathbb{Q}})$ is the identity element. Since holds for arbitrarily large $p$, it follows that $f(g(1))=0$. Finally, implies that $g(1)\in{\mathbb{Q}}$, so we conclude $f(x)$ has a rational root.
If the characteristic polynomial of $(a_n)$ splits over an abelian extension, then we may choose $L$ to be abelian. In this case every element of $A(L)$ takes only rational values, so we can choose $f(x)$ to have all rational roots.
Before proving Theorem \[thintro3\], we need some preliminary results. Suppose $f(x)\in{\mathbb{Q}}[x]$ is monic, and let $L/{\mathbb{Q}}$ be a finite Galois extension over which $f(x)$ splits into linear factors. Define $\Gamma:={\mathrm{Gal}}(L/{\mathbb{Q}}$). Let $\alpha_1,\ldots,\alpha_n$ be the roots of $f$ in $L$, and for $1\leq i\leq n$ set $\Gamma_i={\mathrm{Gal}}(L/{\mathbb{Q}}(\alpha_i))\subset\Gamma$.
\[lem1\] Let $p$ be a rational prime unramified in $L$ that is coprime to the denominators of coefficients of $f$. Then $f(x)$ has a root modulo $p$ if and only if the Frobenius conjugacy class $\phi_p\subset \Gamma$ is contained in $$S_1:=\bigcup_i \Gamma_i.$$
There is a root of $f(x)$ in ${\mathbb{Z}}/(p)$ if and only if for some (equivalently, every) prime $\mathfrak{P}$ of $L$ over $p$, there is some $i$ for which $\alpha_i\mod \mathfrak{P}$ is in ${\mathbb{Z}}/(p)\subset{\mathcal{O}}_L/\mathfrak{P}$. Now, $\alpha_i\mod \mathfrak{P}$ is in ${\mathbb{Z}}/(p)$ if and only if $\phi_\mathfrak{P}(\alpha_i)=\alpha_i$, which happens if and only if $\phi_\mathfrak{P}\in\Gamma_i$. So $f$ has a root in ${\mathbb{Z}}/(p)$ if and only if there exists $\mathfrak{P}|p$ with $\phi_\mathfrak{P}\in\bigcup\Gamma_i$. Since $\bigcup\Gamma_i$ is closed under conjugation, this is equivalent to the condition that $\phi_p\subset\bigcup\Gamma_i$.
We also need the following fact.
\[lem2\] Define a set $$S_2:=\bigcup_i\big\{\sigma\in \Gamma:C_\Gamma(\sigma)\subset \Gamma_i\big\},$$ where $C_\Gamma(\sigma)$ is the centralizer of $\sigma$ inside $\Gamma$. The for every $g\in A(L)$, we have $$\label{eqsett}
\big\{\sigma\in \Gamma:f(g(\sigma))=0\big\}\subseteq S_2,$$ and there exists $g\in A(L)$ for which is an equality of sets.
If $f(g(\sigma))=0$, then $g(\sigma)=\alpha_i$ for some $i$. By , $g(\sigma)$ is fixed by $C_\Gamma(\sigma)$, so we must have $C_\Gamma(\sigma)\subset {\mathrm{Gal}}(L/{\mathbb{Q}}(\alpha_i))=H_i$. This proves the containment . To show that we can choose $g\in A(L)$ for which is equality, let $\sigma_1,\ldots,\sigma_k\in \Gamma$ be a system of conjugacy class representatives. For $1\leq j\leq k$, if there does not exist $i$ for which $C_\Gamma(\sigma_j)\subset H_i$, then define $g$ to be $0$ on the conjugacy class of $\sigma_j$. If there does exist $i$, the define $g$ on the conjugacy class of $\sigma_j$ by $$g(\tau\sigma_j\tau^{-1})=\tau(\alpha_i).$$ We have $g\in A(L)$ and $\{\sigma:f(g(\sigma))=0\}=S_2$.
We also need a group-theoretic fact about.
\[lemgroup\] Let $\Gamma$ and $A$ be finite groups, with $A$ abelian, and consider the wreath product $$\Gamma':=A^\Gamma\rtimes\Gamma.$$ Let $\pi:\Gamma'\to\Gamma$ be the projection. Then at least $${\left(}1-\frac{|\Gamma|^2}{|A|}{\right)}\big|\Gamma'\big|$$ elements $\xi\in\Gamma'$ satisfy $$\label{eqx}
\pi\big(C_{\Gamma'}(\xi)\big)\subset\langle\pi(\xi)\rangle.$$
We identify elements of $\Gamma'$ with pairs $(\varphi,\sigma)$, where $\varphi:\Gamma\to A$ and $\sigma\in \Gamma$. Under this identification, multiplication in $\Gamma'$ is given by $$(\varphi,\sigma)\circ(\psi,\tau)=\big(\varphi+\psi\circ R_{\sigma},\sigma\tau\big)$$ (here $R_{\sigma}:\Gamma\to \Gamma$ is right multiplication by $\sigma$). A direct computation shows that $(\varphi,\sigma)$, $(\psi,\tau)\in \Gamma'$ commute if and only if $\sigma$ and $\tau$ commute and $$\label{eqcom}
\varphi-\varphi \circ R_\tau=\psi-\psi\circ R_\sigma.$$ For $\eta:\Gamma\to A$, there exists $\psi:\Gamma\to A$ with $\eta=\psi-\psi\circ R_\sigma$ if and only if $$\label{eqsumo}
\sum_{n=0}^{ord(\sigma)-1} \eta \circ R_{\sigma^n}=0.$$ Combining and , we see that, if $\varphi$, $\sigma$, and $\tau$ are fixed, then there exists $\psi$ such that $(\varphi,g)$ and $(\psi,h)$ commute if and only if $$\label{eqEcom}
\sum_{n=0}^{ord(\sigma)-1}\big( \varphi\circ R_{\sigma^n\tau}-\varphi \circ R_{\sigma^n}\big)=0.$$
For each $\sigma$, $\tau\in\Gamma$, define a group homomorphism $$\begin{aligned}
\chi_{\sigma,\tau}:A^\Gamma&\to A,\\
\varphi&\mapsto\sum_{n=0}^{ord(\sigma)-1}\big( \varphi(\sigma^n\tau)-\varphi (\sigma^n)\big).\end{aligned}$$ If $\tau\not\in\langle \sigma\rangle$, then the elements $\sigma^n$ and $\sigma^n\tau$ are all distinct. In this case $\chi_{\sigma,\tau}$ is seen to be surjective, and the kernel of $\chi_{\sigma,\tau}$ has index $|A|$ in $A^\Gamma$. It follows from that, for fixed $\tau$, $\sigma$ with $\tau\not\in\langle\sigma\rangle$, there are at most $|A|^{|\Gamma|-1}$ functions $\varphi :\Gamma\to A$ for which $\tau\in\pi(C_{\Gamma'}((\sigma,\varphi))$. Taking the union over all $\sigma$, $\tau\in\Gamma$ with $\tau\not\in\langle\sigma\rangle$, we find that the number of elements $\xi\in \Gamma'$ for which does *not* hold is at most $$|\Gamma|^2|A|^{|\Gamma|-1}.$$ This completes the proof.
We are now ready to prove Theorem \[thintro3\].
We are given a polynomial $f(x)\in{\mathbb{Q}}[x]$ having no rational root. Let $\Gamma\supset S_1\supset S_2$ be as in the statements of Lemmas \[lem1\] and \[lem2\]. By the Chebotarev density theorem, $$\label{eqd1}
\delta{\left(}\big\{p:f\text{ has a root modulo } p\big\}{\right)}=\frac{\#S_1}{\#\Gamma},$$ $$\label{eqd2}
\max_{g\in A(L)}\delta{\left(}\big\{p:g(\phi_p)\equiv 0\mod p\big\}{\right)}=\frac{\#S_2}{\#\Gamma}.$$ We get the strict inequality of Theorem \[thintro3\] because the identity element of $\Gamma$ is in $S_1$ but not in $S_2$. To show that the inequality is sharp, we pass from $L$ to an extension $L'/L$ with the property that in ${\mathrm{Gal}}(L'/{\mathbb{Q}})$, most elements have small centralizers (in a sense to be made precise).
For $L'/{\mathbb{Q}}$ a finite Galois extension containing $L$, write $\Gamma'={\mathrm{Gal}}(L'/{\mathbb{Q}})$ and $\pi:\Gamma'{\twoheadrightarrow}\Gamma$ for the restriction map. Let $$\Gamma_i'=\pi^{-1}(\Gamma_i)={\mathrm{Gal}}(L'/{\mathbb{Q}}(\alpha_i))\subset \Gamma',$$ for $i=1,\ldots,n$. If an element $\sigma\in \Gamma'_i$ satisfies $$\label{eqxx}
\pi\big(C_{\Gamma'}(\sigma)\big)\subset\langle\pi(\sigma)\rangle,$$ then $C_{\Gamma'}(\sigma)\subset\Gamma_i'$. We will show that for every $\epsilon>0$, we can choose the $L'/L$ so that holds for at least $(1-\epsilon)|\Gamma'|$ elements $\sigma$ of $\Gamma'$. This will prove the theorem.
Let $\epsilon>0$ be given, and choose a positive integer $r$ such that $$\frac{|\Gamma|^2}{2^{r}}<\epsilon.$$ Let $p_1,\ldots,p_r$ be distinct rational primes that split completely in the Hilbert class field of $L$. For each $i$, let $\beta_i\in {\mathcal{O}}_L$ be a generator for a (necessarily degree $1$ and principal) prime of $L$ over $p_i$. Let $L'$ be the extension of $L$ obtained by adjoining a square root of $\sigma(\beta_i)$ for all $\sigma\in \Gamma$ and $1\leq i\leq r$. Then $\Gamma'$ is isomorphic to a wreath product $$\Gamma'\cong A^\Gamma\rtimes \Gamma,$$ with $A= ({\mathbb{Z}}/2)^r$. The result now follows from Lemma \[lemgroup\].
Given a recurrent sequence $(a_n)$, let $f(x)\in{\mathbb{Q}}[x]$ be a non-zero polynomial of minimal degree such that $f(a_p)\equiv 0$ mod $p$ for all $p$ (which exists by Theorem \[thintro2\]). Then $f(x)=f_1(x)\ldots f_r(x)$ for distinct irreducible polynomials $f_i\in{\mathbb{Q}}[x]$. For $1\leq i\leq r$, the set $$V_i:=\big\{p:f_i(a_p)\equiv 0\mod p\big\}$$ has some positive density $\delta_i$, which is rational. The densities $\delta_i$ satisfy $\sum\delta_i=1$, and for each $i$ such that $\deg(f_i)\geq 2$, there is strict inequality $$\delta_i<\delta{\left(}\big\{p:f_i\text{ has a root modulo } p\big\}{\right)}.$$ Given $(a_n)$, the irreducible polynomials $f_1,\ldots,f_r$ and densities $\delta_1,\ldots,\delta_r$ can be computed explicitly. For example, the (shifted) tribonacci numbers are defined by $T_0=T_1=0$, $T_1=1$, and $$T_n=T_{n-1}+T_{n-2}+T_{n-3},{\hspace{5mm}}n\geq 3,$$ and we find (with the help of a computer):
$f_i(x)$ $\delta_i$
-------------------------------------- ---------------
$x$ $\frac{1}{6}$
$x^2-x+\frac{3}{11}$ $\frac{1}{3}$
$x^3-x^2+\frac{3}{11}x+\frac{7}{22}$ $\frac{1}{2}$
In [@Evi18], it is shown (Theorem 1.1) that for $p\geq 23$, $T_p\equiv 0\mod p$ if and only if $p$ can be written in the form $p=x^2+11y^2$, with $x$, $y\in{\mathbb{Z}}$.
Periods {#secfrob}
=======
In this section we use the ring $${\mathcal{A}}:=\frac{\prod_p{\mathbb{Z}}/p{\mathbb{Z}}}{\bigoplus_p{\mathbb{Z}}/p{\mathbb{Z}}}.$$ An element of ${\mathcal{A}}$ is a prime-indexed sequence $(a_p)_p$, with $a_p\in{\mathbb{Z}}/p{\mathbb{Z}}$, and two sequences are equal in ${\mathcal{A}}$ if they agree for all sufficiently large $p$. If $(a_n)$ is a recurrent sequence over ${\mathbb{Q}}$, then only finitely many primes divide denominators of terms in the sequence, so we get an element $$(a_n)_{\mathcal{A}}:=\big(a_p\!\!\mod p\big)_p\in{\mathcal{A}}.$$ Separately, for each finite Galois extension $L/{\mathbb{Q}}$, there is a ${\mathbb{Q}}$-algebra homomorphism $$\begin{aligned}
{per}_{\mathcal{A}}:A(L)&\to{\mathcal{A}},\\
g&\mapsto\big(g(\phi_p)\!\!\mod p\big)_p.\end{aligned}$$ Theorem \[thseqA\] is the statement that the set of elements of ${\mathcal{A}}$ of the form $(a_n)_{\mathcal{A}}$ equals the set of elements of the form ${per}_{\mathcal{A}}(g)$.
We define ${\mathcal{P}}^0_{\mathcal{A}}\subset{\mathcal{A}}$ to be the set of elements of the form $(a_n)_{\mathcal{A}}$, or equivalently, the set of elements of the form ${per}_{\mathcal{A}}(g)$.
It can be seen that ${\mathcal{P}}^0_{\mathcal{A}}$ is a countable ${\mathbb{Q}}$-algebra.
Here we give another description of ${\mathcal{P}}^0_{\mathcal{A}}$. Let $L/{\mathbb{Q}}$ be a finite Galois extension, with ring of integers ${\mathcal{O}}_L$. For each rational prime $p$, the $p$-th power map is a ${\mathbb{Z}}/p{\mathbb{Z}}$-algebra endomorphism $F_{p,L}$ of ${\mathcal{O}}_L/p{\mathcal{O}}_L$, which is an automorphism if $p$ is unramified in $L$. There is an isomorphism of ${\mathcal{A}}$-algebras $$L\otimes_{\mathbb{Q}}{\mathcal{A}}\cong\frac{\prod_p{\mathcal{O}}_L/p}{\bigoplus_p{\mathcal{O}}_L/p}.$$
The *${\mathcal{A}}$-valued Frobenius automorphism* is the ${\mathcal{A}}$-algebra automorphism $F_{{\mathcal{A}},L}$ of $L\otimes_{\mathbb{Q}}{\mathcal{A}}$ induced by $F_{p,L}$ in the $p$-th factor.
If we choose a basis for $L$ as a ${\mathbb{Q}}$-vector space, we can represent $F_{{\mathcal{A}},L}$ by a square matrix with entries in ${\mathcal{A}}$, and the ${\mathbb{Q}}$-span of the matrix entries does not depend on the choice of basis.
The subspace ${\mathcal{P}}^0_{\mathcal{A}}\subset{\mathcal{A}}$ is equal to the ${\mathbb{Q}}$-span of matrix entries for $F_{{\mathcal{A}},L}$ as $L$ ranges over all finite Galois extensions of ${\mathbb{Q}}$.
We prove the stronger statement that for each fixed $L$, the ${\mathbb{Q}}$-span of matrix coefficients for $F_{{\mathcal{A}},L}$ is equal to the image of $A(L)$ under ${per}_{\mathcal{A}}$.
The ${\mathbb{Q}}$-span of matrix coefficients for $F_{{\mathcal{A}},L}$ is the image of the map $$\begin{aligned}
\label{map}L^{\vee}\otimes L &\to {\mathcal{A}},\\
\varphi\otimes y&\mapsto \big(\varphi(y^p)\mod p\big)_p.\nonumber\end{aligned}$$ Here $L^{\vee}$ is the ${\mathbb{Q}}$-linear dual of $L$, and the tensor product is over ${\mathbb{Q}}$. The trace form induces an isomorphism of $L$ with $L^{\vee}$, so the image of is equal to the image of $$\begin{aligned}
L\otimes L&\to {\mathcal{A}},\\
x\otimes y&\mapsto {\left(}\Bigg(\sum_{\sigma\in \Gamma}\sigma(xy^p)\Bigg)\mod p{\right)}_p,\end{aligned}$$ where $\Gamma={\mathrm{Gal}}(L/{\mathbb{Q}})$.
It follows from the proof of Theorem \[thseqA\] that ${per}_{\mathcal{A}}(A(L))$ is equal to the image of the map $$\begin{aligned}
(L\otimes L)^\Gamma&\to {\mathcal{A}},\\
\sum_i x_i\otimes y_i&\mapsto {\left(}\Bigg(\sum_i x_i y_i^p\Bigg)\mod P{\right)}_p,\end{aligned}$$ where for each $p$ we have chosen a prime $P$ of $L$ over $p$. The result now follows from the fact that $$\begin{aligned}
L\otimes L&\to (L\otimes L)^\Gamma,\\
x\otimes y&\mapsto\sum_\sigma \sigma(x)\otimes\sigma(y)\end{aligned}$$ is surjective.
We may view $L$ as the algebraic de Rham cohomology of the $0$-dimensional variety ${\mathrm{Spec}}(L)$. Thus ${\mathcal{P}}^0_{\mathcal{A}}$ is the ${\mathbb{Q}}$-span of the matrix coefficients for the isomorphism $$H^0_{dR}({\mathrm{Spec}}(L))\otimes{\mathcal{A}}{\xrightarrow{\sim}}H^0_{dR}({\mathrm{Spec}}(L))\otimes{\mathcal{A}},$$ for $L$ ranging over the finite Galois extensions of ${\mathbb{Q}}$. If instead we look at de Rham-Betti comparison isomorphisms $$H^0_{dR}({\mathrm{Spec}}(L))\otimes{\mathbb{C}}{\xrightarrow{\sim}}H^0_{B}({\mathrm{Spec}}(L))\otimes{\mathbb{C}}$$ for varying $L$, the ${\mathbb{Q}}$-span of the matrix coefficients is ${\overline{\mathbb{Q}}}$. For this reason, we view ${\mathcal{P}}^0_{\mathcal{A}}\subset{\mathcal{A}}$ as analogous to ${\overline{\mathbb{Q}}}\subset{\mathbb{C}}$. By contrast, the integral closure of ${\mathbb{Q}}$ inside ${\mathcal{A}}$ is uncountable.
Generalization to positive-dimensional varieties
------------------------------------------------
The preceeding construction of ${\mathcal{P}}^0_{\mathcal{A}}$ can be generalized to give elements of ${\mathcal{A}}$ associated with an arbitrary algebraic variety. If $X$ is a variety defined over ${\mathbb{Q}}$, then for all sufficiently large $p$, there is a distinguished automorphism $$\label{eqdefFp}
F_{p,X}:H_{dR}^*(X)\otimes{\mathbb{Q}}_p{\xrightarrow{\sim}}H_{dR}^*(X)\otimes{\mathbb{Q}}_p$$ coming from crystalline cohomology (see [@Ked09]). Each matrix coefficient for is $p$-integral for all sufficiently large $p$, so reduction modulo $p$ (for all large $p$ at once) gives an element of ${\mathcal{A}}$. These elements are called ${\mathcal{A}}$-valued periods in [@Ros18a]. It is convenient to allow Tate twists, which means we divide out by a suitable power of $p$ before reducing modulo $p$. Details can be found in §6 of [@Ros18a].
If we instead use the de Rham-Betti comparison isomorphism $$H^*_{dR}(X)\otimes{\mathbb{C}}{\xrightarrow{\sim}}H^*_{B}(X)\otimes{\mathbb{C}},$$ matrix coefficients are the ordinary (complex) periods of $X$.
In the language of motives, the ring $A(L)$ defined in Sec. \[secfungal\] is the ring of de Rham motivic periods of ${\mathrm{Spec}}(L)$ (see [@Bro14], §1.2, and [@Ros17], §5).
[1]{}
Francis Brown. Single-valued motivic periods and multiple zeta values. In [*Forum of Mathematics, Sigma*]{}, volume 2. Cambridge University Press, 2014.
Tim Evink and Paul Alexander Helminck. Tribonacci numbers and primes of the form $p=x^2+11y^2$. 2018, arXiv:1801.04605.
David Jarossay. Indirect computation of $p$-adic cyclotomic multiple zeta values. 2015, arXiv:1501.04893.
Masanobu Kaneko. Finite multiple zeta values (in [J]{}apanese). .
Kiran S Kedlaya. p-adic cohomology. , 2:667, 2009.
Paulo Ribenboim. . Springer Science & Business Media, 2012.
Julian Rosen. A choice-free absolute [G]{}alois group and [A]{}rtin motives. , 2017.
Julian Rosen. Sequential periods of the crystalline [F]{}robenius. 2018, Preprint.
Jean-Pierre Serre. . AK Peters/CRC Press, 2016.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We define the notion of a $(P,{\widetilde}{P})$-structure on a universe $p$ in a locally cartesian closed category category with a binary product structure and construct a $(\Pi,\lambda)$-structure on the C-systems $CC({\cal C},p)$ from a $(P,{\widetilde}{P})$-structure on $p$.
We then define homomorphisms of C-systems with $(\Pi,\lambda)$-structures and functors of universe categories with $(P,{\widetilde}{P})$-structures and show that our construction is functorial relative to these definitions.
---
= 2mm
[**The [$(\Pi,\lambda)$]{}-structures on the [C-systems]{} defined by universe categories**]{}[^1]$^,$[^2]
[**Vladimir Voevodsky**]{}[^3]
Introduction {#Sec.1}
------------
The concept of a C-system in its present form was introduced in [@Csubsystems]. The type of the C-systems is constructively equivalent to the type of contextual categories defined by Cartmell in [@Cartmell0] and [@Cartmell1] but the definition of a C-system is slightly different from Cartmell’s foundational definition.
In this paper, which extends the series started with [@Cfromauniverse], [@fromunivwithPiI] and [@presheavesOb], we continue to consider what might be the most important structure on C-systems - the structure that corresponds, for the syntactic C-systems, to the operations of dependent product, $\lambda$-abstraction and application. The first C-system formulation of this structure was introduced by John Cartmell in [@Cartmell0 pp. 3.37 and 3.41] as a part of what he called a strong M.L. structure. It was later studied by Thomas Streicher in [@Streicher p.71] who called a C-system (contextual category) together with such a structure a “contextual category with products of families of types”.
In [@fromunivwithPiI] we introduced an alternative formulation of this structure that we called a $(\Pi,\lambda)$-structure and constructed a bijection between the sets of Cartmell-Streicher structures and $(\Pi,\lambda)$-structures on any C-system $CC$.
In this paper we consider the case of C-systems of the form $CC({\cal C},p)$ introduced in [@Cfromauniverse]. They are defined, in a functorial way, by a category $\cal C$ with a final object and a morphism $p:{\widetilde}{U}{\rightarrow}U$ together with the choice of pullbacks of $p$ along all morphisms in $\cal C$. A morphism with such choices is called a universe in $\cal C$. As a corollary of general functoriality we obtain a construction of isomorphisms that connect the C-systems $CC({\cal C},p)$ corresponding to different choices of pullbacks and different choices of final objects. It allows us to use the notation $CC({\cal C},p)$ that only mentions $\cal C$ and $p$.
In [@presheavesOb] a number of results about presheaves on universe categories and on the C-systems $CC({\cal C},p)$ has been established. These results are of general nature and do not refer to the $(\Pi,\lambda)$-structures. However, they are highly useful for the constructions such as the one presented in this paper.
The main result of the paper - Construction \[2015.03.17.constr3\], produces a $(\Pi,\lambda)$-structure on $CC({\cal C},p)$ from what we call a $(P,{\widetilde}{P})$-structure on $p$ and what is, in essence, two morphisms in $\cal C$ completing two other morphisms to a pullback. Its combination with the construction of [@fromunivwithPiI], without the part that concerns the bijection, was originally stated in [@CMUtalk Proposition 2] with a sketch of a proof given in the 2009 version of [@NTS]. It and the ideas that it is based on are among the most important ingredients of the construction of the univalent model of the Martin-Lof type theory in Kan simplicial sets.
In view of Lemma \[2016.09.09.l1\], Construction \[2015.03.17.constr3\] can be used not only to construct the $(\Pi,\lambda)$-structures on C-systems, but also to prove that such structures do not exist. It is possible, that a similar technique may be used with other systems of inference rules of type theory, for example, to show that for a given universe $p$ no model of a given kind of higher inductive types exists on $CC({\cal C},p)$.
In the following section we define homomorphisms of C-systems with $(\Pi,\lambda)$-structures and functors of universe categories with $(P,{\widetilde}{P})$-structures and show, in Theorem \[2015.03.21.th1\], that our construction is functorial relative to these definitions.
Theorem \[2015.03.21.th1\] is interesting also in that that its proof indirectly uses almost all results of [@presheavesOb]. On the other hand, modulo these results, the proof is very short and straightforward.
The $(\Pi,\lambda)$-structures correspond to the $(\Pi,\lambda,app,\beta,\eta)$-system of inference rules. In [@fromunivwithPiI Remark 4.4] we outline the definitions of classes of structures that correspond to the similar systems but without the $\beta$- or $\eta$-rules. Such structures appear as natural variations of the $(\Pi,\lambda)$-structures. The results of the present paper admit straightforward modifications needed to construct and sometimes classify such partial $(\Pi,\lambda)$-structures on C-systems of the form $CC({\cal C},p)$.
One may wonder how the construction of this paper relates to the earlier ideas of Seely [@Seely1984] and their refinement by Clairambault and Dybjer [@ClairambaultDybjer2014]. This question requires further study.
The methods of this paper are fully constructive and, in fact, almost entirely essentially algebraic.
The paper is written in the same formalization-ready style as the previous ones. The main intended base for its formalization is Zermelo-Fraenkel theory. However, it can also be formalized in the existing formal systems for the univalent foundations such as the UniMath.
Because of the importance of constructions for this paper we continue to use a special pair of names Problem-Construction for the specification of the goal of a construction and the description of the particular solution.
We also continue to use the diagrammatic order of writing compositions of morphisms, i.e., for $f:X{\rightarrow}Y$ and $g:Y{\rightarrow}Z$ the composition of $f$ and $g$ is denoted by $f\circ g$. This rule applies to functions between sets, morphisms in categories, functors etc.
For a functor $\Phi:{\cal C}{\rightarrow}{\cal C}'$, we let $\Phi^{\circ}$ denote the functor $PreShv(C'){\rightarrow}PreShv(C)$ given by the pre-composition with a functor $\Phi^{op}:{\cal C}^{op}{\rightarrow}({\cal C}')^{op}$, that is, $$\Phi^{\circ}(F)(X)=F(\Phi(X))$$ In the literature this functor is denoted both by $\Phi^*$ and $\Phi_*$ and we decided to use a new unambiguous notation instead.
Acknowledgements are at the end of the paper.
From $(P,{\widetilde}{P})$- to $(\Pi,\lambda)$-structures – the construction {#Sec.2}
----------------------------------------------------------------------------
In this section we describe a method of constructing $(\Pi,\lambda)$-structures on C-systems of the form $CC({\cal C},p)$ where $\cal C$ is a locally cartesian closed universe category $({\cal C},p)$ with a binary product structure.
Let us recall the following definition from [@fromunivwithPiI]:
[\[2015.03.09.def1\]]{} Let $CC$ be a C-system. A pre-$(\Pi,\lambda)$-structure on $CC$ is a pair of morphisms of presheaves $$\Pi:{{\mathcal Ob}}_2{\rightarrow}{{\mathcal Ob}}_1$$ $$\lambda:{{{\widetilde}{\mathcal Ob}}}_2{\rightarrow}{{{\widetilde}{\mathcal Ob}}}_1$$ such that the square
[\[2015.03.09.eq1\]]{}
[[[Ob]{}]{}]{}\_2@>>> [[[Ob]{}]{}]{}\_1\
@VVV @VV V\
[[Ob]{}]{}\_2 @>>> [[Ob]{}]{}\_1
commutes.
A pre-$(\Pi,\lambda)$-structure is called a $(\Pi,\lambda)$-structure if the square (\[2015.03.09.eq1\]) is a pullback.
The functors $I_p$ used in the following definition are defined in [@presheavesOb Sec. 2.6].
[\[2015.03.29.def1\]]{} Let $\cal C$ be a locally cartesian closed category with a binary product structure and $p:{\widetilde}{U}{\rightarrow}U$ a universe in $\cal C$. A pre-$(P,{\widetilde}{P})$-structure on $p$ is a pair of morphisms $${\widetilde}{P}:I_p({\widetilde}{U}) {\rightarrow}{\widetilde}{U}$$ $$P:I_p(U) {\rightarrow}U$$ such that the square
[\[2009.prod.square\]]{}
I\_p([U]{}) @>[P]{}>> [U]{}\
@V I\_p(p) VV @VV p V\
I\_p(U) @>P>> U
commutes.
A pre-$(P,{\widetilde}{P})$-structure is called a $(P,{\widetilde}{P})$-structure if the square (\[2009.prod.square\]) is a pullback.
We will often say pre-$P$-structure (resp. $P$-structure) instead of pre-$(P,{\widetilde}{P})$-structure (resp. $(P,{\widetilde}{P})$-structure).
[\[2015.03.17.prob0\]]{} Let $({\cal C},p)$ be a locally cartesian closed universe category with a binary product structure. Let $(P,{\widetilde}{P})$ be a pre-$(P,{\widetilde}{P})$-structure on $p$. To construct a pre-$(\Pi,\lambda)$-structure on $CC({\cal C},p)$.
[\[2015.03.17.constr3\]]{}Consider the diagram:
[\[2016.12.09.eq1\]]{}
[[[Ob]{}]{}]{}\_2 @>\_2>> int\^(Yo(I\_p([U]{}))) @>int\^(Yo([P]{}))>> int\^(Yo([U]{})) @>\_1\^[-1]{}>> [[[[Ob]{}]{}]{}]{}\_1\
@VVV @VV int\^(Yo(I\_p(p))) V @VV int\^(Yo(p)) V @VVV\
[[Ob]{}]{}\_2 @>\_2>> int\^(Yo(I\_p(U))) @>int\^(Yo(P))>> int\^(Yo(U)) @>\_1\^[-1]{}>> [[Ob]{}]{}\_1
where $\mu_n$ and ${\widetilde}{\mu}_n$ are isomorphisms defined in [@presheavesOb Sec. 2.6]. The left hand side and the right hand side squares of this diagram commute because the squares in [@presheavesOb Problem 2.6.8] commute. The middle square commutes because the square (\[2009.prod.square\]) commutes and both $Yo$ and $int^{\circ}$ are functors. Therefore, the outside rectangle commutes and we conclude that the pair of morphisms
[\[2016.12.09.eq3\]]{}
=\_2int\^(Yo([P]{}))\_1\^[-1]{}\
=\_2int\^(Yo(P))\_1\^[-1]{}
is a pre-$(\Pi,\lambda)$-structure on $CC({\cal C},p)$.
[\[2017.01.07.l4\]]{} In the context of Construction \[2015.03.17.constr3\], if $(P,{\widetilde}{P})$ is a $(P,{\widetilde}{P})$-structure then the pre-$(\Pi,\lambda)$-structure constructed there is a $(\Pi,\lambda)$-structure.
We need to show that the external square of the diagram (\[2016.12.09.eq1\]) is a pullback.
Horizontal composition of pullbacks is a pullback. The left hand side square is a pullback because it is a commutative square with two parallel sides being isomorphisms. The right hand side square is a pullback for the same reason.
It remains to show that the middle square is pullback. This square is obtained by applying first the functor $Yo$ and then the functor $int^{\circ}$ to the pullback square (\[2009.prod.square\]).
Our claim follows now from two facts:
1. the Yoneda functor $Yo:{\cal C}{\rightarrow}PreShv({\cal C})$ takes pullbacks to pullbacks,
2. for any functor $F:{\cal C}'{\rightarrow}{\cal C}$, the functor $$F^{\circ}:PreShv({\cal C}){\rightarrow}PreShv({\cal C}')$$ of pre-composition with $F^{op}$, takes pullbacks to pullbacks.
We assume that these two facts are known.
There is an important class of cases when the function from $(P,{\widetilde}{P})$-structures on $p$ to $(\Pi,\lambda)$-structures on $CC({\cal C},p)$ defined by Construction \[2015.03.17.constr3\] is a bijection.
[\[2016.09.09.l1\]]{} Let $({\cal C},p)$ be a universe category such that the functor $$Yo\circ int^{\circ}:{\cal C}{\rightarrow}PreShv(CC({\cal C},p))$$ is fully faithful. Then the function from the pre-$(P,{\widetilde}{P})$-structures on $p$ to the pre-$(\Pi,\lambda)$-structures on $CC({\cal C},p)$ defined by Construction \[2015.03.17.constr3\] is a bijection.
Moreover, the restriction of this function to the function from $(P,{\widetilde}{P})$-structures to $(\Pi,\lambda)$-structures, which is defined in view of Lemma \[2017.01.07.l4\], is a bijection as well.
Let $${\widetilde}{\alpha}:Mor_{PreShv(CC({\cal C},p))}(int^{\circ}(Yo(I_p({\widetilde}{U}))),int^{\circ}(Yo({\widetilde}{U}))){\rightarrow}Mor_{\cal C}(I_p({\widetilde}{U}),{\widetilde}{U})$$ $${\alpha}:Mor_{PreShv(CC({\cal C},p))}(int^{\circ}(Yo(I_p({U}))),int^{\circ}(Yo({U}))){\rightarrow}Mor_{\cal C}(I_p({U}),{U})$$ be the inverses to $(Yo\circ int^{\circ})_{I_p({\widetilde}{U}),{\widetilde}{U}}$ and $(Yo\circ int^{\circ})_{I_p(U),U}$ respectively.
Given a pre-$(\Pi,\lambda)$-structure $(\Pi,\lambda)$ let
[\[2016.09.09.eq1\]]{}
[P]{}=(\_2\^[-1]{}\_1)\
P=(\_2\^[-1]{}\_1)
Then ${\widetilde}{P}:I_p({\widetilde}{U}){\rightarrow}{\widetilde}{U}$ and $P:I_p(U){\rightarrow}U$. Let $S$ be the square that ${\widetilde}{P}$ and $P$ form with $I_p(p)$ and $p$. Then the square $(Yo\circ int^{\circ})(S)$ is of the form
[\[2017.01.07.eq7\]]{}
int\^(Yo(I\_p([U]{}))) @>\_2\^[-1]{}\_1>> int\^(Yo([U]{}))\
@Vint\^(Yo(I\_p(p))) VV @VV int\^(Yo(p)) V\
int\^(Yo(I\_p(U))) @>\_2\^[-1]{}\_1>> int\^(Yo(U))
Since the left and right squares of (\[2016.12.09.eq1\]) commute and their horizontal arrows are isomorphisms, the square $(Yo\circ int^{\circ})(S)$ is isomorphic to the original square formed by $\Pi$ and $\lambda$ and as a square isomorphic to a commutative square is commutative. Since $Yo\circ int^{\circ}$ is faithful, that is, injective on morphisms between a given pair of objects we conclude that $S$ is commutative, that is, $(P,{\widetilde}{P})$ defined in (\[2016.09.09.eq1\]) is a pre-$(P,{\widetilde}{P})$-structure.
One verifies immediately that the function from pre-$(\Pi,\lambda)$-structures to pre-$(P,{\widetilde}{P})$-structures that this construction defines is both left and right inverse to the function defined by Construction \[2015.03.17.constr3\].
Assume now that we started with a $(\Pi,\lambda)$-structure. Then the square $(Yo\circ int^{\circ})(S)$ is isomorphic to a pullback and therefore is a pullback. By our assumption, the functor $Yo\circ int^{\circ}$ is fully-faithful. Fully-faithful functors reflect pullbacks, that is, if the image of a square under a fully-faithful functor is a pullback than the original square is a pullback. We conclude that both the direct and the inverse bijections map the subsets of $(P,{\widetilde}{P})$-structures and $(\Pi,\lambda)$-structures to each other. Therefore, e.g. by [@fromunivwithPiI Lemma 5.1], the restrictions of the total bijections to these subsets are bijections as well.
The lemma is proved.
[\[2016.12.09.prob2\]]{} Let $({\cal C},p)$ be a universe category.
To construct a function from the set of $(P,{\widetilde}{P})$-structures on $p$ to the set of structures of products of families of types on $CC({\cal C},p)$.
To show that if the functor $Yo\circ int^{\circ}$ is fully faithful than this function is a bijection.
[\[2016.12.09.constr2\]]{} The required function is the composition of the function of Construction \[2015.03.17.constr3\] with the construction for [@fromunivwithPiI Problem 4.5] described in that paper.
[\[2017.01.07.rem1\]]{} One can define a mixed $(P,{\widetilde}{P})$-structure (or pre-$(P,{\widetilde}{P})$-structure) as follows:
[\[2009.10.27.def1\]]{} Let $\cal C$ be an lcc category and let $p_i:{\widetilde}{U}_i{\rightarrow}U_i$, $i=1,2,3$ be three morphisms in $\cal C$. A $(P,{\widetilde}{P})$-structure on $(p_1,p_2,p_3)$ is a pullback of the form
[\[Pisq1\]]{}
I\_[p\_1]{}([U]{}\_2) @>[P]{}>> [U]{}\_3\
@VI\_[p\_1]{}(p\_2)VV @VVp\_3V\
I\_[p\_1]{}(U\_2) @>P>> U\_3
Then a $(P,{\widetilde}{P})$-structure on $p$ is a $(P,{\widetilde}{P})$-structure on $(p,p,p)$. This concept can be used to construct universes in C-systems that participate in impredicative $(\Pi,\lambda)$-structures.
From $(P,{\widetilde}{P})$- to $(\Pi,\lambda)$-structures – the functoriality {#Sec.3}
-----------------------------------------------------------------------------
Recall that in [@fromunivwithPiI pp. 1067-68] we have constructed, for any homomorphism $H:CC{\rightarrow}CC'$ of C-systems, and any $n\ge 0$, natural transformations $$H{{\mathcal Ob}}_n:{{\mathcal Ob}}_i{\rightarrow}H^{\circ}({{\mathcal Ob}}_i)$$ where for $\Gamma\in CC$ and $T\in {{\mathcal Ob}}_i(\Gamma)$ one has $$H{{\mathcal Ob}}_n(T)=H_{Ob}(T)$$ and $$H{{\widetilde}{{{\mathcal Ob}}}}_n:{{\widetilde}{{{\mathcal Ob}}}}_i{\rightarrow}H^{\circ}({{\widetilde}{{{\mathcal Ob}}}}_i)$$ where for $\Gamma\in CC$ and $o\in {{\widetilde}{{{\mathcal Ob}}}}_n(\Gamma)$ one has $$H{{\widetilde}{{{\mathcal Ob}}}}_n(o)=H_{Mor}(o)$$
[\[2016.09.13.def1\]]{} Let $H:CC{\rightarrow}CC'$ be a homomorphism of C-systems. Let $(\Pi,\lambda)$ and $(\Pi',\lambda')$ be pre-$(\Pi,\lambda)$-structures on $CC$ and $CC'$ respectively.
Then $H$ is called a $(\Pi,\lambda)$-homomorphism if the following two squares commute $$\begin{CD}
{{\mathcal Ob}}_2 @>\Pi>> {{\mathcal Ob}}_1\\
@VH{{\mathcal Ob}}_2 VV @VVH{{\mathcal Ob}}_1 V\\
H^{\circ}({{\mathcal Ob}}_2) @>H^{\circ}(\Pi')>> H^{\circ}({{\mathcal Ob}}_1)
\end{CD}
{{\,\,\,\,\,\,\,}}{{\,\,\,\,\,\,\,}}{{\,\,\,\,\,\,\,}}{{\,\,\,\,\,\,\,}}\begin{CD}
{{\widetilde}{{{\mathcal Ob}}}}_2 @>\lambda>> {{\widetilde}{{{\mathcal Ob}}}}_1\\
@VH{{\widetilde}{{{\mathcal Ob}}}}_2 VV @VVH{{\widetilde}{{{\mathcal Ob}}}}_1 V\\
H^{\circ}({{\widetilde}{{{\mathcal Ob}}}}_2) @>H^{\circ}(\lambda')>> H^{\circ}({{\widetilde}{{{\mathcal Ob}}}}_1)
\end{CD}$$ If $(\Pi,\lambda)$ and $(\Pi',\lambda')$ are $(\Pi,\lambda)$-structures then $H$ is called a $(\Pi,\lambda)$-homomorphism if it is a $(\Pi,\lambda)$-homomorphism with respect to the corresponding pre-$(\Pi,\lambda)$-structures.
Unfolding the definition of $H{{\mathcal Ob}}_i$ and $H{{\widetilde}{{{\mathcal Ob}}}}_i$ we see that $H$ is a $(\Pi,\lambda)$-homomorphism if and only if for all $\Gamma\in CC$ one has
1. for all $T\in {{\mathcal Ob}}_2(\Gamma)$ one has
[\[2016.09.13.eq1\]]{} H(\_(T))=’\_[H()]{}(H(T))
2. for all $o\in {{\widetilde}{{{\mathcal Ob}}}}_2(\Gamma)$ one has
[\[2016.09.13.eq2\]]{} H(\_(o))=’\_[H()]{}(H(o))
The morphisms $\xi$ and ${\widetilde}{\xi}$ used in the following definition are defined in [@presheavesOb Sec. 3.4].
[\[2017.01.13.def1\]]{} Let $({\cal C},p)$ and $({\cal C}',p')$ be universe categories with locally cartesian closed and binary product structures and let $(P,{\widetilde}{P})$, $(P',{\widetilde}{P}')$ be pre-$(P,{\widetilde}{P})$-structures on $p$ and $p'$ respectively.
A universe category functor ${\bf\Phi}=(\Phi,\phi,{\widetilde}{\phi})$ is said to be a pre-$(P,{\widetilde}{P})$-functor relative to the structures $(P,{\widetilde}{P})$ and $(P',{\widetilde}{P}')$ if the squares
[\[2015.03.23.sq1\]]{}
(I\_p(U)) @>(P) >> (U)\
@V\_[[****]{},1]{}VV @VV V\
I\_[p’]{}(U’) @>P’>> U’
(I\_p([U]{})) @>([P]{}) >> ([U]{})\
@V\_[[****]{},1]{}VV @VV V\
I\_[p’]{}([U]{}’)@>[P]{}’ >> [U]{}’
commute.
If $(P,{\widetilde}{P})$ and $(P',{\widetilde}{P}')$ are $(P,{\widetilde}{P})$-structures then $\bf\Phi$ satisfying the above condition is called a $(P,{\widetilde}{P})$-functor.
[\[2015.03.21.th1\]]{} Let $({\cal C},p)$ and $({\cal C}',p')$ be universe categories with locally cartesian closed and binary product structures. Let let $(P,{\widetilde}{P})$ and $(P',{\widetilde}{P}')$ be pre-$(P,{\widetilde}{P})$-structures on $p$ and $p'$ respectively.
If ${\bf\Phi}=(\Phi,\phi,{\widetilde}{\phi})$ is a pre-$(P,{\widetilde}{P})$-universe category functor relative to $(P,{\widetilde}{P})$ and $(P',{\widetilde}{P}')$, then the homomorphism $$H(\Phi,\phi,{\widetilde}{\phi}):CC({\cal C},p){\rightarrow}CC ({\cal C}',p')$$ is a homomorphism of C-systems with pre-$(\Pi,\lambda)$-structures relative to the structures obtained from $(P,{\widetilde}{P})$ and $(P',{\widetilde}{P}')$ by Construction \[2015.03.17.constr3\].
We have to show that for all $\Gamma\in Ob(CC({\cal C},p))$, $T\in Ob_2(\Gamma)$ and $o\in {{\widetilde}{{{\mathcal Ob}}}}_2(\Gamma)$ the equalities (\[2016.09.13.eq1\]) and (\[2016.09.13.eq2\]) hold. We will prove the first equality. The proof of the second one is strictly parallel to the proof of the first. We have $$H(\Pi(T))=$$$$H(\mu_1^{-1}(\mu_2(T)\circ P))=
\mu_1^{-1}(\psi(\Gamma)\circ \Phi(\mu_2(T)\circ P)\circ \xi_0)=$$$$\mu_1^{-1}(\psi(\Gamma)\circ \Phi(\mu_2(T))\circ \Phi(P)\circ \phi)=
\mu_1^{-1}(\psi(\Gamma)\circ \Phi(\mu_2(T))\circ \xi_{1}\circ P')=
\mu_1^{-1}(\mu_2(H(T))\circ P')=$$$$\Pi'(H(T))$$ where the first equality holds by the definition of $\Pi$, the second by [@presheavesOb Eq. 3.45], the third by the composition axiom of functor $\Phi$ and [@presheavesOb Eq. 3.41], the fourth by the commutativity of (\[2015.03.23.sq1\]), the fifth by [@presheavesOb Eq. 3.43], and the sixth one by the definition of $\Pi'$.
Acknowledgements {#Sec.4}
----------------
I am grateful to the Department of Computer Science and Engineering of the University of Gothenburg and Chalmers University of Technology for its the hospitality during my work on the first version of the paper.
Work on this paper was supported by NSF grant 1100938.
This material is based on research sponsored by The United States Air Force Research Laboratory under agreement number FA9550-15-1-0053. The US Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright notation thereon.
The views and conclusions contained herein are those of the author and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of the United States Air Force Research Laboratory, the U.S. Government or Carnegie Mellon University.
[10]{}
John Cartmell. Generalised algebraic theories and contextual categories. , 1978. <http://www.cs.ru.nl/~spitters/Cartmell.pdf>.
John Cartmell. Generalised algebraic theories and contextual categories. , 32(3):209–243, 1986.
Pierre Clairambault and Peter Dybjer. The biequivalence of locally cartesian closed categories and [M]{}artin-[L]{}öf type theories. , 24(6):e240606, 54, 2014.
R. A. G. Seely. Locally [C]{}artesian closed categories and type theory. , 95(1):33–48, 1984.
Thomas Streicher. . Progress in Theoretical Computer Science. Birkhäuser Boston Inc., Boston, MA, 1991. Correctness, completeness and independence results, With a foreword by Martin Wirsing.
Vladimir Voevodsky. Notes on type systems. 2009–2012. <https://github.com/vladimirias/old_notes_on_type_systems>.
Vladimir Voevodsky. The equivalence axiom and univalent models of type theory. , pages 1–11, 2010. <http://arxiv.org/abs/1402.5556>.
Vladimir Voevodsky. A [C]{}-system defined by a universe category. , 30(37):1181–1215, 2015. <http://www.tac.mta.ca/tac/volumes/30/37/30-37.pdf>.
Vladimir Voevodsky. Products of families of types and [$(\Pi,\lambda)$]{}-structures on [C]{}-systems. , 31(36):1044–1094, 2016. <http://www.tac.mta.ca/tac/volumes/31/36/31-36.pdf>.
Vladimir Voevodsky. Subsystems and regular quotients of [C-systems]{}. In [*Conference on Mathematics and its Applications, (Kuwait City, 2014)*]{}, number 658 in Contemporary Mathematics, pages 127–137, 2016.
Vladimir Voevodsky. -systems defined by universe categories: presheaves. , 32(3):53 – 112, 2017. <http://www.tac.mta.ca/tac/volumes/32/3/32-03.pdf>.
[^1]: *2000 Mathematical Subject Classification: 03F50, 18C50 03B15, 18D15,*
[^2]: For the published version of the paper see <http://www.tac.mta.ca/tac/volumes/32/4/32-04abs.html>
[^3]: School of Mathematics, Institute for Advanced Study, Princeton NJ, USA. e-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'For positive $p$-harmonic functions on Riemannian manifolds, we derive a gradient estimate and Harnack inequality with constants depending only on the lower bound of the Ricci curvature, the dimension $n$, $p$ and the radius of the ball on which the function is defined. Our approach is based on a careful application of the Moser iteration technique and is different from Cheng-Yau’s method [@CY] employed by Kostchwar and Ni [@kn], in which a gradient estimate for positive $p$-harmonic functions is derived under the assumption that the sectional curvature is bounded from below.'
address:
- |
Department of Mathematics\
Michigan State University\
East Lansing, MI 48824
- |
Department of Mathematics\
University of Florida\
358 Little Hall, P. O. Box 118105\
Gainesville, FL 32611-8105
author:
- Xiaodong Wang
- Lei Zhang
title: 'Local Gradient Estimate for $p$-harmonic functions on Riemannian Manifolds'
---
[^1]
Introduction
============
The study of harmonic functions on Riemannian manifolds has been one of the central subjects in geometric analysis. In their classical work Cheng-Yau [@SY] derived the following gradient estimate for positive harmonic functions on Riemannian manifolds:
*Theorem A (Cheng-Yau) Let* $M$ *be an* $n$*-dimensional complete Riemannian manifold with* $Ric\geq -\left( n-1\right) \kappa $*, where* $\kappa \geq 0$ *is a constant. Suppose that* $u$ *is a positive harmonic function on a geodesic ball* $B\left( o,R\right) $*. Then*$$\sup_{B\left( o,R/2\right) }\frac{\left\vert \nabla u\right\vert }{u}\leq
C_{n}\frac{1+R\sqrt{\kappa }}{R}, \label{yau2}$$ *where* $C_{n}$* is a constant depending only on* $n$*.*
An important feature of Cheng-Yau’s estimate is that the RHS (which stands for the right hand side) of (\[yau2\]) depends only on $n$, $k$ and $R$, it does not depend on the lower bound of the injectivity radius or a global coordinate system. From PDE viewpoints, deriving a Harnack inequality requires some bounds on the coefficients in some fixed coordinate system, thus not suitable for many problems defined on manifolds. We also observe that the RHS of (\[yau2\]) is optimal in the sense that even for $\kappa>0$, the bound stays bounded when $R\rightarrow \infty$.
There are two major ingredients in the proof of Theorem A. First a Bochner formula is used to derive a lower bound of the Laplacian of $|\nabla u|^{2}$ for a harmonic function $u$ in terms of the lower bound of the Ricci tensor. The second major ingredient is a clever application of the maximum principle. The trick is to multiply $|\nabla u|^{2}$ by a cut-off function, derive a new differential inequality for the product and then apply the maximum principle. The cut-off function is constructed using the distance function. As a result, the new differential inequality involves the Laplacian of the distance function. As is well known, the Riemannian distance function is uniformly Lipschitz and its Laplacian has an upper bound depending on the lower bound of the Ricci tensor.
Cheng-Yau’s approach turned out to be very useful and some important results for other problems are deeply influenced by Theorem A. For example, P. Li [@pli] obtained the sharp lower bound for the first eigenvalue of a manifold, which was later generalized by Li-Yau [@liyau1]. Similar results were also obtained by Li-Yau [@liyau2] for heat equations. S. Y. Cheng [@cheng] and H. I. Choi [@choi] obtained gradient estimates for harmonic mappings, etc. We refer to [@SY; @schoen] and the more recent survey [@Li] for an overview of the subject.
$p$-harmonic functions are natural extensions of harmonic functions from a variational point of view. It has been extensively studied because of its various interesting features and applications. Compared with the theory for harmonic functions the study of $p$-harmonic functions is generally harder because the equation, even though elliptic, is degenerate and the regularity results are far weaker (see, for example [@tolk]). Recently, there has been renewed interest in $p$-harmonic functions. In particular R. Moser [M]{} established a nice connection between $p$-harmonic functions and the inverse mean curvature flow. In a recent paper [@kn] Kotschwar and Ni derived, among other things, a local gradient estimate for $p$-harmonic functions under the assumption that the sectional curvature is bounded from below. It is remarkable that the constant in their estimate does not blowup when $p\rightarrow 1$, which leads to interesting results on the inverse mean curvature flow problems. Their proof follows the same strategy introduced by Cheng-Yau [@CY] for harmoinc functions (i.e. $p=2$). However for general $p$-harmonic functions, the computation involves the full Hessian of the distance function when the cut-off function is introduced. As a result, a lower bound on the sectional curvature has to be assumed in [@kn].
Kotschwar and Ni speculated that their estimate may hold if only a lower bound on the Ricci tensor is assumed. The main result of this paper is to establish the following theorem:
\[p-harm\] Let $\left( M^{n},g\right) $ be a complete Riemannian manifold with $\mathrm{Ric}\geq -\left( n-1\right) \kappa $. Assume that $v$ is a positive $p$-harmonic function on the ball $B(o,R)\subset M$. Then there exists a constant $C_{p,n}$ such that$$\frac{|\nabla v|}{v}\leq C_{p,n}\left( 1+\sqrt{\kappa }R\right) /R\quad %
\mbox{on}\quad B(o,R/2).$$
The proof of Theorem \[p-harm\] will be presented in section two. As far as the second major ingredient of Cheng-Yau’s proof is concerned, our approach follows a different strategy by carefully using the Moser iteration technique. This approach only involves differentiating the distance function once and hence bypasses the difficulty of handling the full Hessian of the distance function. In the special case $p=2$, when $p$-harmonic function are just harmonic functions, Theorem \[p-harm\] is exactly Cheng-Yau’s theorem. An immediate consequence of Theorem \[p-harm\] is the following Harnack inequality.
Let $\left( M^{n},g\right) $ be a complete Riemannian manifold with $\mathrm{Ric}\geq -\left( n-1\right) \kappa $. Assume that $v$ is a positive $p$-harmonic function on the ball $B(o,R)\subset M$. Then there exists a constant $C_{p,n}$ such that for any $x,y\in B(o,R/2)$,$$v\left( x\right) /v\left( y\right) \leq e^{C_{p,n}\left( 1+\sqrt{\kappa }%
R\right) }.$$
It follows that if $\mathrm{Ric}\geq 0$, then we have a uniform constant $%
c_{p,n}$ (independent of $R$) s.t. $$\sup_{B(o,R/2)}v\leq c_{p,n}\inf_{B(o,R/2)}v. \label{harn}$$This was already proved by Rigoli, Salvatori, and Vignati [@RSV]. In fact, they proved the stronger result that (\[harn\]) holds provided that the volume is doubling and a weak Poincare inequality holds. See also [Hol]{}.
Another standard application of Theorem \[p-harm\] is the following Liouville theorem, which was also deduced from the Harnack inequality in [@RSV].
* Let* $u$* be a* $p$*-harmonic function bounded from above or below on a complete Riemannian manifold with non-negative Ricci tensor, then* $u$* is constant.*
Finally we point out that our constant $C_{p,n}$ in Theorem \[p-harm\] becomes unbounded as $p\rightarrow 1$, while in Kotschwar-Ni’s result, all the constants stay bounded when $p\rightarrow 1$. We do not know if the method can be tweaked to remove this defect.
The gradient estimate for the $p$-harmonic functions
====================================================
$p$-harmonic functions arise naturally as critical points of the $%
L^{p}\,\,(p>1)$ norm of the gradient. Let $\left( M^{n},g\right) $ be a complete Riemannian manifold and $\Omega \subset M$ an open set. A function $%
v\in W_{loc}^{1,p}\left( \Omega \right) $ is $p$-harmonic if$$\mathrm{div}\left( \left\vert \nabla v\right\vert ^{p-2}\nabla v\right) =0$$in the weak sense, i.e.$$\int_{U}\left\vert \nabla v\right\vert ^{p-2}\left\langle \nabla v,\nabla
\xi \right\rangle =0$$for all $\xi \in W_{0}^{1,p}\left( \Omega \right) $. By [@tolk] for example, $v$ must be $C^{1,\alpha }$. Moreover $v\in W_{loc}^{2,2}$ if $%
p\geq 2$; $v\in W_{loc}^{2,p}$ if $1<p<2$. Away from $\left\{ \nabla
v=0\right\} $, $v$ is in fact smooth.
Suppose that $v$ is positive. Set $u=-\left( p-1\right) \log v$. Then $u$ satisfies$$\mathrm{div}\left( \left\vert \nabla u\right\vert ^{p-2}\nabla u\right)
=\left\vert \nabla u\right\vert ^{p}. \label{equ}$$Let $f=\left\vert \nabla u\right\vert ^{2}$. We define$$\mathcal{L}\left( \psi \right) =\mathrm{div}\left( f^{p/2-1}A\left( \nabla
\psi \right) \right) -pf^{p/2-1}\left\langle \nabla u,\nabla \psi
\right\rangle ,$$where $$A=id+\left( p-2\right) \frac{\nabla u\otimes \nabla u}{\left\vert \nabla
u\right\vert ^{2}}.$$
We need the following lemma from [@kn] and the proof is by direct calculation.
\[nilem\] $$\mathcal{L}\left( f\right) =2f^{p/2-1}\left( \left\vert D^{2}u\right\vert
^{2}+\mathrm{Ric}\left( \nabla u,\nabla u\right) \right) +\left( \frac {p}{2}%
-1\right) \left\vert \nabla f\right\vert ^{2}f^{p/2-2}.$$
Lemma \[nilem\] holds point-wisely in $\{x:\,\,f(x)>0\}$. From the gradient estimate of [@tolk] we know that $f=|\nabla u|^{2}\in C^{\alpha
}$ for some $\alpha >0$ and $f\in W_{loc}^{1,\beta }$ for some $\beta >1$.
We choose a local orthonormal frame $\left\{ e_{i}\right\} $ with $%
e_{1}=\nabla u/\left\vert \nabla u\right\vert $. Then (\[equ\]) takes the following form$$\left( p-1\right) u_{11}+\sum_{i=2}^{n}u_{ii}=f.$$Therefore$$\begin{aligned}
\left\vert D^{2}u\right\vert ^{2}& \geq
u_{11}^{2}+2\sum_{i=2}^{n}u_{1i}^{2}+\sum_{i=2}^{n}u_{ii}^{2} \\
& \geq u_{11}^{2}+2\sum_{i=2}^{n}u_{1i}^{2}+\frac{1}{n-1}\left(
\sum_{i=2}^{n}u_{ii}\right) ^{2} \\
& =u_{11}^{2}+2\sum_{i=2}^{n}u_{1i}^{2}+\frac{1}{n-1}\left( f-\left(
p-1\right) u_{11}\right) ^{2} \\
& =\frac{1}{n-1}f^{2}-\frac{2\left( p-1\right) }{n-1}fu_{11}+\left( 1+\frac{%
\left( p-1\right) ^{2}}{n-1}\right) u_{11}^{2}+2\sum_{i=2}^{n}u_{1i}^{2} \\
& \geq \frac{1}{n-1}f^{2}-\frac{2\left( p-1\right) }{n-1}fu_{11}+a_{0}%
\sum_{i=1}^{n}u_{1i}^{2},\end{aligned}$$where $a_{0}=1+\min \left( \frac{\left( p-1\right) ^{2}}{n-1},1\right) >1$. Using the identities $$\begin{aligned}
2fu_{11}& =\left\langle \nabla u,\nabla f\right\rangle , \\
\sum_{i=1}^{n}u_{1i}^{2}& =\frac{1}{4}\frac{\left\vert \nabla f\right\vert
^{2}}{f},\end{aligned}$$we end up with $$\left\vert D^{2}u\right\vert ^{2}\geq \frac{1}{n-1}f^{2}-\frac{\left(
p-1\right) }{n-1}\left\langle \nabla u,\nabla f\right\rangle +\frac{a_{0}}{4}%
\frac{\left\vert \nabla f\right\vert ^{2}}{f}.$$Assume that $\mathrm{Ric}\geq -\left( n-1\right) \kappa $. Therefore $$\begin{aligned}
\mathcal{L}\left( f\right) & \geq -2\left( n-1\right) \kappa f^{p/2}+\left(
\frac{p+a_{0}}{2}-1\right) \left\vert \nabla f\right\vert ^{2}f^{p/2-2}
\label{930e1} \\
& +\frac{2}{n-1}f^{p/2+1}-\frac{2\left( p-1\right) }{n-1}f^{p/2-1}\left%
\langle \nabla u,\nabla f\right\rangle \notag \\
& \geq -2\left( n-1\right) kf^{p/2}+\frac{2}{n-1}f^{p/2+1}-\frac{2\left(
p-1\right) }{n-1}f^{p/2-1}\left\langle \nabla u,\nabla f\right\rangle \notag\end{aligned}$$
Equation (\[930e1\]) holds wherever $f$ is strictly positive. Let $%
K=\{x\in \Omega :f(x)=0\}$. Then for any nonnegative function $\psi $ with compact support in $\Omega \setminus K$, we have $$\begin{aligned}
&&\int_{\Omega }\left\langle f^{p/2-1}\nabla f+(p-2)f^{p/2-2}\left\langle
\nabla u,\nabla f\right\rangle \nabla u,\nabla \psi \right\rangle
\label{lfweak} \\
&&+p\int_{\Omega }f^{p/2-1}\left\langle \nabla u,\nabla f\right\rangle \psi +%
\frac{2}{n-1}\int_{\Omega }f^{p/2+1}\psi \notag \\
&\leq &2(n-1)k\int_{\Omega }f^{p/2}\psi +\frac{2(p-1)}{n-1}\int_{\Omega
}f^{p/2-1}\left\langle \nabla u,\nabla f\right\rangle \psi . \notag\end{aligned}$$In particular, let $\epsilon >0$ and $\psi =f_{\epsilon }^{b}\eta ^{2}$ where $f_{\epsilon }=(f-\epsilon )^{+}$, $\eta \in C_{0}^{\infty }(B_{R})$ is non-negative, $b>1$ is to be determined later. Then direct computation yields $$\nabla \psi =bf_{\epsilon }^{b-1}\nabla f\eta ^{2}+2f_{\epsilon }^{b}\eta
\nabla \eta .$$Using the above in (\[lfweak\]) we obtain $$\begin{aligned}
&&b\int_{B_{R}}\bigg (f^{p/2-1}f_{\epsilon }^{b-1}|\nabla
f|^{2}+(p-2)f^{p/2-2}\left\langle \nabla u,\nabla f\right\rangle
^{2}f_{\epsilon }^{b-1}\bigg )\eta ^{2} \label{930e3} \\
&+&2(p-2)\int_{B_{R}}f^{p/2-2}\left\langle \nabla u,\nabla f\right\rangle
f_{\epsilon }^{b}\eta (\nabla u,\nabla \eta )+2\int_{\Omega
}f^{p/2-1}f_{\epsilon }^{b}\eta \left\langle \nabla f,\nabla \eta
\right\rangle \notag \\
&+&p\int_{\Omega }f^{p/2-1}\left\langle \nabla u,\nabla f\right\rangle
f_{\epsilon }^{b}\eta ^{2}+\frac{2}{n-1}\int_{\Omega }f^{p/2+1}f_{\epsilon
}^{b}\eta ^{2} \notag \\
&\leq &2(n-1)\kappa \int_{\Omega }f^{p/2-1}f_{\epsilon }^{b}\eta ^{2}+\frac{%
2(p-1)}{n-1}\int_{\Omega }f^{p/2-1}\left\langle \nabla u,\nabla
f\right\rangle f_{\epsilon }^{b}\eta ^{2}. \notag\end{aligned}$$Since $u\in C^{1,\alpha }$, $f\in C^{\alpha }$ and $\nabla f\in L^{\beta
}(\Omega )$ for some $\alpha >0$ and $\beta >1$, we see that except for the first term, all the other terms converge to the corresponding form without $%
\epsilon $. For the first term, observe that $$f^{p/2-1}f_{\epsilon }^{b-1}|\nabla f|^{2}+(p-2)f^{p/2-2}\left\langle \nabla
u,\nabla f\right\rangle ^{2}f_{\epsilon }^{b-1}\geq
a_{1}f^{p/2-1}f_{\epsilon }^{b-1}|\nabla f|^{2}$$where $a_{1}=1$ if $p\geq 2$ and $a_{1}=(p-1)$ if $p\in (1,2)$. Thus by passing $\epsilon $ to $0$ we have $$\begin{aligned}
&&ba_{1}\int_{\Omega }f^{p/2+b-2}|\nabla f|^{2}\eta ^{2} \label{930e2} \\
&+&2(p-2)\int_{\Omega }f^{\frac{p-4}{2}+b}\left\langle \nabla u,\nabla
f\right\rangle \eta \left\langle \nabla u,\nabla \eta \right\rangle
+2\int_{\Omega }f^{\frac{p-2}{2}+b}\eta \left\langle \nabla f,\nabla \eta
\right\rangle \notag \\
&+&p\int_{\Omega }f^{\frac{p-2}{2}+b}\left\langle \nabla u,\nabla
f\right\rangle \eta ^{2}+\frac{2}{n-1}\int_{\Omega }f^{\frac{p+2}{2}+b}\eta
^{2} \notag \\
&\leq &2(n-1)k\int_{\Omega }f^{\frac{p-2}{2}+b}\eta ^{2}+\frac{2(p-1)}{n-1}%
\int_{\Omega }f^{\frac{p-2}{2}+b}\left\langle \nabla u,\nabla f\right\rangle
\eta ^{2}. \notag\end{aligned}$$
From now on we use $a_{1},a_{2},\cdots $ etc. to denote constants depending only on $p$ and $n$. Combining terms in (\[930e2\]) using the definition of $f$ we have $$\begin{aligned}
&&a_{1}b\int_{\Omega }f^{p/2+b-2}|\nabla f|^{2}\eta ^{2}+\frac{2}{n-1}%
\int_{\Omega }f^{p/2+1+b}\eta ^{2} \label{921e1} \\
&\leq &2(n-1)\kappa \int_{\Omega }f^{p/2+b}\eta ^{2}+a_{2}\int_{\Omega }f^{%
\frac{p-1}{2}+b}|\nabla f|\eta ^{2} \notag \\
&&+a_{3}\int_{\Omega }f^{p/2+b-1}|\nabla f||\nabla \eta |\eta . \notag\end{aligned}$$For $R_{3}$ (the third term on the RHS, $L_{1},L_{2},R_{1}$ etc. are understood similarly) in (\[921e1\]) we have $$|R_{3}|\leq \frac{a_{1}b}{4}\int_{\Omega }f^{b+p/2-2}|\nabla f|^{2}\eta ^{2}+%
\frac{a_{4}}{b}\int_{\Omega }|\nabla \eta |^{2}f^{b+p/2}.$$Also by Cauchy’s inequality $R_{2}$ of (\[921e1\]) can be estimated as $$|R_{2}|\leq \frac{a_{1}b}{4}\int_{\Omega }f^{p/2+b-2}|\nabla f|^{2}\eta ^{2}+%
\frac{a_{5}}{b}\int_{\Omega }f^{b+p/2+1}\eta ^{2}.$$With these two inequalities we obtain $$\begin{aligned}
&&\frac{a_{1}b}{2}\int_{\Omega }f^{p/2+b-2}|\nabla f|^{2}\eta ^{2}+\frac{2}{%
n-1}\int_{\Omega }f^{p/2+1+b}\eta ^{2} \label{923e1} \\
&\leq &\int_{\Omega }\left( 2(n-1)\kappa \eta ^{2}+\frac{a_{4}}{b}|\nabla
\eta |^{2}\right) f^{p/2+b}+\frac{a_{5}}{b}\int_{\Omega }f^{b+p/2+1}\eta
^{2}. \notag\end{aligned}$$By requiring $$\frac{a_{5}}{b}<\frac{1}{n-1} \label{assumb}$$we see that the last term in on the RHS of (\[923e1\]) is majorized by the last term on the LHS. Therefore we have $$\begin{aligned}
&&\frac{a_{1}b}{2}\int_{\Omega }f^{p/2+b-2}|\nabla f|^{2}\eta ^{2}+\frac{1}{%
n-1}\int_{\Omega }f^{p/2+1+b}\eta ^{2} \\
&\leq &2(n-1)\kappa \int_{\Omega }f^{p/2+b}\eta ^{2}+\frac{a_{4}}{b}%
\int_{\Omega }|\nabla \eta |^{2}f^{p/2+b}.\end{aligned}$$For the first term on the LHS we use $$|\nabla (f^{p/4+b/2}\eta )|^{2}\leq \frac{1}{2}(\frac{p}{2}%
+b)^{2}f^{p/2+b-2}|\nabla f|^{2}\eta ^{2}+2f^{p/2+b}|\nabla \eta |^{2}.$$From the above we obtain $$\int_{\Omega }|\nabla (f^{p/4+b/2}\eta )|^{2}+d_{1}\int_{\Omega
}f^{p/2+1+b}\eta ^{2}\leq \kappa d_{2}\int_{\Omega }f^{p/2+b}\eta
^{2}+a_{7}\int_{\Omega }|\nabla \eta |^{2}f^{b+p/2}. \label{929e1}$$where $d_{1}\sim b,d_{2}\sim b$ (recall $b>1$, $d_{1}\sim b$ means $d_{1}$ is comparable to $b$, $d_{2}\sim b$ is understood the same way).
The following Sobolev embedding theorem of Saloff-Coste plays an important role in our approach:
*Theorem B (Theorem 3.1 of [@saloff]) Let* $\left(
M^{n},g\right) $* be a complete Riemannian manifold with* $Ric\geq
-\left( n-1\right) \kappa $*. For* $n>2$*, there exists* $C$*, depending only on* $n$*, such that for all* $B\subset M$* of radius* $R$* and volume* $V$* we have for* $f\in C_{0}^{\infty
}\left( B\right) $$$\left( \int \left\vert f\right\vert ^{2q}\right) ^{1/q}\leq e^{C\left( 1+%
\sqrt{\kappa }R\right) }V^{-2/n}R^{2}\left( \int \left\vert \nabla
f\right\vert ^{2}+R^{-2}f^{2}\right) ,$$* where* $q=n/(n-2)$*. For* $n=2$*, the above inequality holds with* $n$* replaced by any fixed* $n^{\prime }>2$*.*
From now on, we assume $\Omega =B\left( o,R\right) $. Theorem B gives $$\begin{aligned}
&&\left( \int_{\Omega }f^{\frac{n(p/2+b)}{n-2}}\eta ^{\frac{2n}{n-2}}\right)
^{\left( n-2\right) /n} \label{929e2} \\
&\leq &e^{c_{0}(1+\sqrt{\kappa }R)}V^{-\frac{2}{n}}\bigg (R^{2}\int_{\Omega
}|\nabla (f^{p/4+b/2}\eta )|^{2}+\int_{\Omega }f^{p/2+b}\eta ^{2}\bigg ).
\notag\end{aligned}$$where $c_{0}(n,p)>0$ depends only on $n,p$. Let $b_{0}=c_{1}(n,p)(1+\sqrt{%
\kappa }R)$ with $c_{1}(n,p)$ large enough to make $b_{0}$ satisfy ([assumb]{}), then (\[929e1\]) and (\[929e2\]) combined gives $$\begin{aligned}
&&\left( \int_{\Omega }f^{\frac{n(p/2+b)}{n-2}}\eta ^{\frac{2n}{n-2}}\right)
^{\left( n-2\right) /n}+a_{8}bR^{2}e^{c_{2}b_{0}}V^{-2/n}\int_{\Omega
}f^{p/2+1+b}\eta ^{2} \label{921moser} \\
&\leq &a_{9}b_{0}^{2}be^{c_{2}b_{0}}V^{-2/n}\int_{\Omega }f^{p/2+b}\eta
^{2}+a_{10}e^{c_{2}b_{0}}V^{-2/n}R^{2}\int_{\Omega }|\nabla \eta
|^{2}f^{p/2+b} \notag\end{aligned}$$
\[923lem1\] Let $b_{1}=(b_{0}+\frac{p}{2})\frac{n}{n-2}$. Then there exists $c_{3}(n,p)>0$ such that $$\Vert f\Vert _{L^{b_{1}}(B_{3R/4})}\leq c_{3}\frac{b_{0}^{2}}{R^{2}}%
V^{1/b_{1}}. \label{923e2}$$
**Proof of Lemma \[923lem1\]:** Let $b=b_{0}$ in ([921moser]{}), then by comparing $L_{2}$ and $R_{1}$ of (\[921moser\]) we observe that $$a_{9}b_{0}^{3}f^{p/2+b_{0}}<\frac{1}{2}a_{8}b_{0}R^{2}f^{p/2+1+b_{0}}$$if $f>a_{11}b_{0}^{2}R^{-2}$. Thus in the evaluation of $R_{1}$ we decompose $\Omega $ into two subregions, one over the places where $f\leq
a_{11}b_{0}^{2}R^{-2}$ and the second region is the complement of the first region. With this decomposition we have $$R_{1}\leq a_{12}^{b_{0}}b_{0}^{3}\left( \frac{b_{0}}{R}\right)
^{p+2b_{0}}e^{c_{2}b_{0}}V^{1-2/n}+\frac{L_{2}}{2}.$$Now (\[921moser\]) with $b=b_{0}$ can be written as $$\begin{aligned}
&&\left( \int_{\Omega }f^{\frac{n(p/2+b_{0})}{n-2}}\eta ^{\frac{2n}{n-2}%
}\right) ^{\left( n-2\right) /n}+\frac{a_{8}}{2}%
b_{0}R^{2}e^{c_{2}b_{0}}V^{-2/n}\int_{\Omega }f^{p/2+1+b_{0}}\eta ^{2}
\label{923e3} \\
&\leq &a_{12}^{b_{0}}b_{0}^{3}\left( \frac{b_{0}}{R}\right)
^{p+2b_{0}}e^{c_{2}b_{0}}V^{1-2/n}+a_{10}e^{c_{2}b_{0}}V^{-2/n}R^{2}\int_{%
\Omega }|\nabla \eta |^{2}f^{p/2+b_{0}} \notag\end{aligned}$$Now we choose $\eta $ to make $R_{2}$ in (\[923e3\]) dominated by the LHS. Let $\eta _{1}\in C_{0}^{\infty }(B_{R})$ satisfy $$0\leq \eta _{1}\leq 1,\quad \eta _{1}\equiv 1\quad \mbox{in}\quad
B_{3R/4},\quad |\nabla \eta _{1}|\leq C(n)/R.$$Let $\eta =\eta _{1}^{m}$ where $m=b_{0}+\frac{p}{2}+1$. Direct computation shows $$R^{2}|\nabla \eta |^{2}\leq a_{13}b_{0}^{2}\,\eta ^{\frac{2b_{0}+p}{%
b_{0}+p/2+1}}. \label{923e4}$$By (\[923e4\]) and Young’s inequality, the $R_{2}$ of (\[923e3\]) can be written as $$\begin{aligned}
&&a_{10}\,R\,^{2}\int |\nabla \eta |^{2}f^{b_{0}+p/2} \\
&\leq &a_{14}\,b_{0}^{2}\,\int_{\Omega }f^{b_{0}+p/2}\eta ^{\frac{2b_{0}+p}{%
b_{0}+p/2+1}} \\
&\leq &a_{14}\,b_{0}^{2}\left( \int_{\Omega }f^{b_{0}+p/2+1}\eta ^{2}\right)
^{\frac{b_{0}+p/2}{b_{0}+p/2+1}}V^{\frac{1}{b_{0}+p/2+1}} \\
&\leq &\frac{a_{8}b_{0}}{2}\,R^{2}\,\int_{\Omega }f^{b_{0}+p/2+1}\eta
^{2}+a_{15}^{b_{0}+p/2}\frac{b_{0}^{b_{0}+p/2+2}}{R^{2b_{0}+p}}V.\end{aligned}$$
With the estimates on $R_{1},R_{2}$ we arrive at $$\bigg (\int_{B_{3R/4}}f^{(b_{0}+p/2)n/\left( n-2\right) }\bigg )^{\left(
n-2\right) /n}\leq a_{16}^{b_{0}}e^{c_{2}b_{0}}V^{1-2/n}b_{0}^{3}\left(
\frac{b_{0}}{R}\right) ^{p+2b_{0}}. \label{923e7}$$
Recall $b_{1}=(b_{0}+\frac{p}{2})\frac{n}{n-2}$. Taking the $1/(b_{0}+p)$ root on both sides of (\[923e7\]) we have $$\Vert f\Vert _{L^{b_{1}}(B_{3R/4})}\leq a_{17}V^{\frac{1}{b_{1}}%
}b_{0}^{2}/R^{2}.$$Lemma \[923lem1\] is established. $\Box $
Now we go back to (\[921moser\]), by ignoring $L_{2}$ we have
$$\begin{aligned}
&&\left( \int_{\Omega }f^{\frac{n(p/2+b)}{n-2}}\eta ^{\frac{2n}{n-2}}\right)
^{\left( n-2\right) /n} \label{923e9} \\
&\leq &a_{17}\frac{e^{c_{2}b_{0}}}{V^{2/n}}\int_{\Omega }\bigg (%
(b_{0}^{2}b\eta ^{2}+R^{2}|\nabla \eta |^{2}\bigg )f^{b+p/2}. \notag\end{aligned}$$
To apply the Moser iteration we set $$b_{l+1}=b_{l}\frac{n}{n-2},\quad \Omega _{l}=B(o,\frac{R}{2}+\frac{R}{4^{l}}%
),\quad l=1,2..$$and choose $\eta _{l}\in C_{0}^{\infty }\left( \Omega \right) $ s.t. $$\eta _{l}\equiv 1\,\,\mbox{in}\,\,\Omega _{l+1},\quad \eta _{l}\equiv 0\,\,%
\mbox{in}\,\,\Omega \setminus \Omega _{l},\quad |\nabla \eta _{l}|\leq \frac{%
C4^{l}}{R},\quad 0\leq \eta _{l}\leq 1.$$Then in (\[923e9\]), by letting $b+\frac{p}{2}=b_{l}$, $\eta =\eta _{l}$ we have $$\left( \int_{\Omega _{l+1}}f^{b_{l+1}}\right) ^{\frac{1}{b_{l+1}}}\leq
\left( a_{17}\frac{e^{c_{2}b_{0}}}{V^{\frac{2}{n}}}\right) ^{\frac{1}{b_{l}}%
}\left( \int_{\Omega _{l}}\left( b_{0}^{2}b_{l}+R^{2}|\nabla \eta
_{l}|^{2}\right) f^{b_{l}}\right) ^{\frac{1}{b_{l}}}.$$By the estimate of $|\nabla \eta _{l}|$ $$\Vert f\Vert _{L^{b_{l+1}}(\Omega _{l+1})}\leq \left( a_{17}\frac{%
e^{c_{2}b_{0}}}{V^{2/n}}\right) ^{1/b_{l}}\left(
b_{0}^{2}b_{l}+16^{l}\right) ^{1/b_{l}}\Vert f\Vert _{L^{b_{l}}(\Omega
_{l})}. \label{923e10}$$Notice that $\sum_{l=1}^{\infty }\frac{1}{b_{l}}=\frac{n}{2b_{1}}$, then (\[923e10\]) leads to $$\begin{aligned}
&&\Vert f\Vert _{L^{\infty }(B_{R/2})} \label{923e11} \\
&\leq &\left( a_{18}\frac{e^{c_{2}b_{0}}}{V^{2/n}}\right)
^{\sum_{l=1}^{\infty }1/b_{l}}\prod\limits_{l=1}^{\infty }\left(
b_{0}^{3}\left( \frac{n}{n-2}\right) ^{l}+16^{l}\right) ^{1/b_{l}}\Vert
f\Vert _{L^{b_{1}}(B_{3R/4})} \notag \\
&\leq &a_{19}\frac{e^{\frac{nc_{2}b_{0}}{2b_{1}}}}{V^{1/b_{1}}}b_{0}^{\frac{%
3n}{2b_{1}}}\Vert f\Vert _{L^{b_{1}}(B_{3R/4})}. \notag\end{aligned}$$Using Lemma \[923lem1\] in (\[923e11\]) we obtain $$\Vert f\Vert _{L^{\infty }(B_{R/2})}\leq a_{20}b_{0}^{2}/R^{2}.
\label{923e12}$$Thus Theorem \[p-harm\] is established. $\Box $
[99]{} S. Y. Cheng, Liouville theorem for harmonic maps. Geometry of the Laplace operator (Proc. Sympos. Pure Math., Univ. Hawaii, Honolulu, Hawaii, 1979), pp. 147–151, Proc. Sympos. Pure Math., XXXVI, Amer. Math. Soc., Providence, R.I., 1980.
S.Y. Cheng, S. T. Yau, Differential equations on Riemannian manifolds and their geometric applications. Comm. Pure Appl. Math. 28 (1975), 333-354.
H. I. Choi, On the Liouville theorem for harmonic maps. Proc. Amer. Math. Soc. 85 (1982), no. 1, 91–94.
I. Holopainen, Volume growth, Green’s function, and parobolicity of ends. Duke Math. Journal 97 (1999), 319-346.
B. Kotschwar, L. Ni, Local gradient estimates of $p$-harmonic functions, $1/H$-flow, and an entropy formula. Ann. Sci. Éc. Norm. Supér. (4) 42 (2009), no. 1, 1–36.
O. A. Ladyzhenskaya, N. Ural’tseva, Linear and quasilinear elliptic equations. Translated from the Russian by Scripta Technica, Inc. Translation editor: Leon Ehrenpreis Academic Press, New York-London 1968 xviii+495 pp.
P. Li, Harmonic functions and applications to complete manifolds, preprint (available on the author’s homepage).
P. Li, A lower bound for the first eigenvalue of the Laplacian on a compact manifold. Indiana Univ. Math. J. 28 (1979), no. 6, 1013–1019.
P. Li, S. T. Yau, Estimates of eigenvalues of a compact Riemannian manifold. Geometry of the Laplace operator (Proc. Sympos. Pure Math., Univ. Hawaii, Honolulu, Hawaii, 1979), pp. 205–239, Proc. Sympos. Pure Math., XXXVI, Amer. Math. Soc., Providence, R.I., 1980.
P. Li, S. T. Yau, On the parabolic kernel of the Schrödinger operator. Acta Math. 156 (1986), no. 3-4, 153–201.
R. Moser, The inverse mean curvature flow and $p$-harmonic functions. J. Eur. Math. Soc. 9 (2007), 77-83.
M. Rigoli, M. Salvatori, and M. Vignati, A note on $p$-subharmonic functions on complete manifolds. Manuscripta Math. 92 (1997), 339–359.
L. Saloff-Coste, Uniformly elliptic operators on Riemannian manifolds. J. Differential Geom. 36 (1992), no. 2,417–450.
R. Schoen, The effect of curvature on the behavior of harmonic functions and mappings. Nonlinear partial differential equations in differential geometry (Park City, UT, 1992), 127–184, IAS/Park City Math. Ser., 2, Amer. Math. Soc., Providence, RI, 1996.
R. Schoen, S. T. Yau, Lectures on Differential Geometry. International Press, Boston, 1994.
P. Tolksdorf, Regularity for a more general class of quasilinear elliptic equations. J. Differential Equations 51 (1984), no. 1, 126–150.
S. T. Yau, Harmonic functions on complete Riemannian manifolds. Comm. Pure Appl. Math. 28 (1975), 201–228.
[^1]: Wang acknowledges support from NSF grant DMS-0905904. Zhang is supported in part by NSF Grant 0900864(1027628)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The nature of monetary arrangements is often discussed without any reference to its detailed construction. We present a graph representation which allows for a clear understanding of modern monetary systems. First, we show that systems based on commodity money are incompatible with credit. We then study the current chartalist systems based on pure fiat money, and we discuss the consolidation of the central bank with the Treasury. We obtain a visual explanation about how commercial banks are responsible for endogenous money creation whereas the Treasury and the central bank are in charge of the total amount of net money. Finally we draw an analogy between systems based on gold convertibility and currency pegs to show that fixed exchange rates can never be maintained.\
\
address: 'Université Pierre & Marie Curie - Paris VI, Sorbonne Universités, Institut d’Astrophysique de Paris, CNRS-UMR 7095, 98 bis Bd Arago, 75014 Paris, France'
author:
- Cyril Pitrou
title: 'Graph representation of balance sheets: from exogenous to endogenous money'
---
Monetary theory ,graph theory,endogenous money ,central bank ,chartalism
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We consider the problem of threshold estimation for autoregressive time series with a “space switching” in the situation, when the regression is nonlinear and the innovations have a smooth, possibly non Gaussian, probability density. Assuming that the unknown threshold parameter is sampled from a continuous positive prior density, we find the asymptotic distribution of the Bayes estimator. As usually in the singular estimation problems, the sequence of Bayes estimators is asymptotically efficient, attaining the minimax risk lower bound.'
author:
- |
P. <span style="font-variant:small-caps;">Chigansky</span>, [The Hebrew University, Jerusalem, Israel]{}\
Yu. A. <span style="font-variant:small-caps;">Kutoyants</span>, [Université du Maine, Le Mans, France]{}
date:
title: On nonlinear TAR processes and threshold estimation
---
\[section\] \[section\] \[section\] \[section\] \[section\] \[section\] \[section\]
ł
[**Key words and phrases:**]{} Bayes estimator, compound Poisson process, likelihood inference, limit distribution, nonlinear threshold models, singular estimation.\
[*AMS 1991 subject classifications: Primary 62G30; secondary 62M10.*]{}
Introduction
============
The simplest threshold autoregressive (TAR) process is the time series, generated by the recursion $$X_{j+1}=\rho _1X_j\,\1_{\left\{X_j<\theta \right\}}+
\rho_2X_j\,\1_{\left\{X_j\geq \theta \right\}}+\varepsilon _{j+1} ,\quad
\quad j=0,\ldots, n-1,$$ where $\varepsilon _j\sim{\cal N}\left(0,\sigma ^2\right)$ are i.i.d. random variables and $ \rho _1\not=\rho _2$ and $\sigma ^2$ are known constants. The unknown [ *threshold*]{} parameter $\theta \in \Theta =\left(\alpha ,\beta \right)$ is to be estimated from the data $X^n=\left(X_0,X_1,\ldots,X_n\right)$. This model and some of its generalizations has been extensively studied during the last decades (see e.g. [@Ch93]-[@DN11],[@Tong11] and the references therein). Particularly, much attention focused on the properties of the least squares (LS) estimator $$\theta _n^*={\rm argmin}_{\theta \in \Theta }
\sum_{j=0}^{n-1}\left[X_{j+1}-\rho _1X_j\,\1_{\left\{X_j<\theta \right\}}
-\rho_2X_j\,\1_{\left\{X_j\geq \theta \right\}}\right]^2.$$ Assuming that $|\rho_1|\vee |\rho_2|<1$ and thus that $(X_j)$ is geometric mixing with the unique invariant density $\varphi \left(x,\theta \right)$, Chan [@Ch93] proved consistency of $\theta _n^*$ with the rate $n$ (rather than $\sqrt{n}$ as in regular problems) and showed that the limit distribution is related to certain compound Poisson process (see below). Note that if $\varepsilon_1\sim {\cal
N}\left(0,\sigma ^2\right)$, the LS estimator coincides with the maximum likelihood (ML) estimator.
This work continues the study of the Bayes estimator for the TAR models, initiated in [@ChK11] and developed further in [@ChK10] and [@CKL1] (see also [@Kut10] for the continuous time counterpart). We consider the following more general nonlinear TAR(1) model $$\label{nG}
X_{j+1}=h \left(X_j\right)\,\1_{\left\{X_j< \theta
\right\}}+g\left(X_j\right)\,\1_{\left\{X_j\geq \theta
\right\}} +\varepsilon _{j+1} , \qquad j=0,\ldots, n-1,$$ where $h(x)$ and $g(x)$ are known functions, $(\varepsilon _j)$ are i.i.d. random variables with a known density function $f\left(x\right)>0,x\in
\RR$ and the initial condition $X_0$ is independent of $(\varepsilon_j)$ and has a probability density $f_0(x)$.
Throughout we shall assume that the following conditions are in force
1. \[c1\] The parameter $\theta \in \left(\alpha ,\beta
\right)\equiv \Theta , -\infty<\alpha <\beta <\infty $ is sampled from the continuous positive prior density $p\left(\theta \right),\theta \in\Theta $.
2. \[c2\] The functions $h$ and $g$ are continuous and satisfy $$\inf_{v\in\Theta }\left|\delta \left(v\right)\right|>0,\qquad \delta
\left(v\right):=g\left(v\right)-h\left(v\right).$$
3. \[c3\] The random variables $(\varepsilon _j)_{j\ge 1}$ are i.i.d. with a known continuous bounded density function $f\left(x\right)>0,x\in \RR$
4. \[c4\] The functions $h\left(x\right),g\left(x\right)$ and $f\left(x\right)$ are such that the time series, generated by , is geometric mixing with the unique positive bounded invariant density $\varphi
\left(x,\theta\right)$, i.e. for any measurable function $|\psi(x)|\le 1$ $$\Ex\bigg|\Ex \big(\psi(X_j)|\F_i\big)-\int_\RR \psi(x)\varphi(x,\theta)dx\bigg|\le C r^{|j-i|},\quad j>i$$ with positive constants $C$ and $r<1$.
5. \[c5\] The function $$J(z):=\int_{-\infty }^{\infty
}\left|\ln\frac{f\left(y+z\right)}{f\left(y\right)}\right|
f\left(y\right)\,{\rm d}y, \quad \min_{\theta\in \Theta} \delta(\theta)\le z\le \max_{\theta\in \Theta} \delta(\theta)$$ is bounded.
The likelihood function of the sample $X^n$ is given by $$L\left(\theta ,X^n\right)=f_0\left(X_0\right)\prod_{j=0}^{n-1}
f\Big(X_{j+1}-h \left(X_j\right)\,\1_{\left\{X_j< \theta
\right\}}-g\left(X_j\right)\,\1_{\left\{X_j\geq \theta \right\}} \Big),$$ and the Bayes estimator $\tilde\theta _n$ with respect to the mean square risk is the conditional expectation $$\tilde\theta _n=\Ex\left(\theta | X^n\right)=\frac{\int_{\Theta }^{}\theta \,p\left(\theta \right)\,
L\left(\theta ,X^n\right)\;{\rm d}\theta}{\int_{\Theta }^{}p\left(\theta
\right)\, L\left(\theta ,X^n\right)\;{\rm d}\theta}.$$ Since the likelihood $L(\theta,X^n)$ is piecewise constant in $\theta$, the estimate can be computed efficiently (see [@ChK11]).
The asymptotic properties of $(\tilde\theta_n)$ are formulated in terms of the following compound Poisson process $$\label{Z}
{Z}\left(u\right)= \left\{\begin{array}{ll}
&\exp\left( \,\sum_{l=1}^{N_+\left(\;u\,\right)}\,\, \ln
\frac{f\left(\varepsilon_l^++\delta \left(\theta _0\right)
\right)}{f\left(\varepsilon_l^+ \right)} \right),\qquad u\geq 0 ,
\\ &\exp\left( \,\sum_{l=1}^{N_-\left(-u\right)} \ln
\frac{f\left(\varepsilon_l^--\delta \left(\theta _0\right)
\right)}{f\left(\varepsilon_l^- \right)} \right),\qquad u< 0.
\end{array}
\right.$$ Here $\theta_0 $ is the true value of the parameter, $\epsilon_l^\pm $ are independent random variables with the density function $f\left(x\right)$, $N_+\left(\cdot \right)$, $N_-\left(\cdot \right)$ are independent Poisson processes with the same intensity $\lambda =\varphi \left(\theta
_0,\theta _0\right)$ ($Z(u):=1$ on the sets $\{N_\pm(u)=0\}$).
Define the random variable $$\tilde{u}=\frac{\int_{R}^{}u\,Z\left(u\right)\,{\rm d}u}{\int_{R}^{}Z\left(u\right)\,{\rm d}u}.$$
As shown in [@ChK10] (see [@IH] for the general theory), we have the following lower bound on the mean square risk of an arbitrary sequence of estimates $(\bar\theta _n)$: $$\Liminf_{\delta \rightarrow 0}\Liminf_{n\rightarrow \infty
}\sup_{\left|\theta -\theta _0\right|<\delta } n^2\Ex_\theta
\left(\bar\theta _n-\theta \right)^2\geq \Ex_{\theta _0}\tilde u^2,$$ and the Bayes estimates $(\widetilde \theta_n)$ are [*efficient*]{}, attaining this lower bound asymptotically. Our main result is the following
\[T1\] Under the conditions \[c1\]-\[c5\], the sequence of estimates $(\tilde \theta _n)$ is consistent, the convergence in distribution $$n\left(\tilde\theta _n-\theta _0\right)\Longrightarrow \tilde u$$ holds and the moments converge: $$\lim_{n\rightarrow \infty }n^{p}\Ex_{\theta _0}\left|\tilde\theta
_n-\theta _0\right|^p=\Ex_{\theta _0}\left|\tilde u\right|^p, \quad p>0.$$
The assumption \[c4\] is often easy to check, using the standard ergodic theory as e.g. in [@MT09]. The assumption \[c5\] is satisfied for many common densities. For example, for the Gaussian innovations $\varepsilon _j\sim N\left(0,\sigma ^2\right)$, $$J\left(z\right)\le \frac{z^2}{2\sigma ^2}+\frac{|z|}{\sigma }.$$ In this case, the limit compound Poisson process $Z\left(u\right)$ has Gaussian jumps: $$\ln
\frac{f\left(\varepsilon_1^\pm\pm \delta \left(\theta _0\right)
\right)}{f\left(\varepsilon_1^\pm \right)}
=-\frac{\delta \left(\theta_0 \right)^2}{2\sigma ^2} \mp
\frac{\delta \left(\theta_0 \right)}{\sigma ^2}\;\varepsilon^\pm_1\sim{\cal
N}\left( -\frac{\delta ^2\left(\theta _0\right)}{2\sigma ^2},\frac{\delta
^2\left(\theta _0\right)}{\sigma ^2} \right).$$
Similarly the assumption \[c5\] is checked for the Laplace density $f\left(y\right)=\left(2\sigma
\right)^{-1}e^{-\frac{\left|y\right|}{\sigma }}$ and the limit process has jumps of the form $$\ln
\frac{f\left(\varepsilon_1^\pm\pm\delta \left(\theta _0\right)
\right)}{f\left(\varepsilon_1^\pm \right)}
=\frac 1 {\sigma^2} \Big(\big|\eps^{\pm}_1\big|-\big|\eps^{\pm}_1\pm\delta(\theta_0)\big|\Big).$$
The Proof
=========
We shall verify the conditions of the Theorem 1.10.2 in [@IH], where the properties of the Bayes estimators, announced in Theorem \[T1\], are derived from the convergence of the normalized likelihood ratios $$Z_n\left(u\right)=\frac{ L\left(\theta_0+u/n,X^n \right) }{
L\left(\theta_0,X^n\right)},\qquad u\in \UU_n=\left[n\left(\alpha
-\theta _0\right),n\left(\beta -\theta _0\right)\right]$$ to the limit process $Z\left(u\right),u\in \RR$ and the two inequalities and , presented below. The change of variables $\theta =\theta _0+u/n$ gives $$\tilde\theta _n=\frac{\int_{\UU_n}^{}\left(\theta
_0+\frac{u}{n}\right)\, p\left(\theta _0+\frac{u}{n}\right)
\frac{L\left(\theta _0+\frac{u}{n},X^n \right)}{L\left(\theta
_0,X^n\right)} {\rm d}u} {\int_{\UU_n}^{} p\left(\theta _0+\frac{u}{n}\right)
\frac{L\left(\theta _0+\frac{u}{n},X^n \right)}{L\left(\theta
_0,X^n\right)} {\rm d}u}
=\theta
_0+\frac{1}{n}\frac{\int_{\UU_n}u\left[p\left(\theta
_0\right)+o\left(1\right)\right] Z_n\left(u\right){\rm d}u}{\int_{\UU_n}\left[p\left(\theta
_0\right)+o\left(1\right)\right] Z_n\left(u\right){\rm d}u}.$$ Then, informally, we have $$\begin{aligned}
\tilde u_n=n\left(\tilde\theta _n-\theta
_0\right)=\frac{\int_{\UU_n}u\;Z_n\left(u\right){\rm
d}u}{\int_{\UU_n}Z_n\left(u\right){\rm d}u}+o\left(1\right)\Longrightarrow
\frac{\int_{\RR}u\;Z\left(u\right){\rm
d}u}{\int_{\RR}Z\left(u\right){\rm d}u}\equiv \tilde u.\end{aligned}$$ Theorem 1.10.2 in [@IH] validates this convergence along with the convergence of moments. Similar program has been realized in the preceding works [@ChK11], [@ChK10] and [@CKL1].
To avoid inessential technicalities, we shall assume that $(X_j)$ is stationary, i.e. $X_0\sim \varphi(\cdot,\theta_0)$. Due to the mixing property \[c4\], all the results below can be derived without stationarity assumption, along the same lines with minor adjustments (see [@CKL1] for details).
Below, $C$, $C'$, $c$, $C_p$, etc. denote constants, whose values are not important and may change from line to line. We shall denote by $\Pb_{\theta}$ and $\Ex_{\theta}$ the probability and the expectation, corresponding to the particular value of the unknown parameter $\theta\in \Theta$ and set $\F_j:=\sigma\{\eps_i, i\le j\}$. The standard $O(\cdot)$ and $o(\cdot)$ notations will be used and we set $\sum_{i=k}^m(...)=0$ and $\prod_{i=k}^m
(...)=1$ for $k>m$.
Convergence of f.d.f.
---------------------
We shall prove the convergence of the finite dimensional distributions: $$\label{wc1}
\big(\ln Z_n(u_1),...,\ln Z_n(u_d)\big)
\Longrightarrow
\big(\ln Z(u_1),...,\ln Z(u_d)\big), \quad u\in \RR^d,$$ following [@CKL1]. We shall restrict the consideration to $0=u_0<u_1<...<u_d$, leaving out the similar complementary case. To this end, note that the declared limit process $\ln Z(u)$ has independent increments and $$\begin{gathered}
\Ex_{\theta_0} \exp\left(\sum_{j=1}^d \iunit\lambda_j \Big(\ln Z(u_j)-\ln Z(u_{j-1})\Big) \right) =\\
\exp \left(
\sum_{j=1}^d (u_j-u_{j-1})\varphi(\theta_0,\theta_0) \Big(\Psi(\lambda_j)-1\Big)
\right)=: e^{H(\lambda)}, \quad \lambda \in \RR^d,\end{gathered}$$ where (recall that $\delta:=g-h$) $$\Psi(\lambda_j) := \Ex_{\theta_0} \exp\left(\iunit \lambda_j \ln
\frac{f\big(\varepsilon_1+\delta \left(\theta _0\right)
\big)}{f\big(\varepsilon_1 \big)}\right).$$ Since $\ln Z(0)=0$ a.s., follows from the convergence of characteristic functions of the increments $$\lim_n \Ex_{\theta_0} \exp\left(\sum_{j=1}^d \iunit\lambda_j \Big(\ln Z_n(u_j)-\ln Z_n(u_{j-1})\Big) \right)
=
e^{H(\lambda)}, \quad \lambda \in \RR^d.$$ Let $m(x,\theta):= h(x)\1_{\{x<\theta\}}+g(x)\1_{\{x\ge \theta\}}$ and note that $$m(x,\theta_0+u_{j-1}/n)-m(x,\theta_0+u_j/n) = \delta(x)\1_{\{x\in \DD^n_j\}},$$ where $\DD^n_j:= [\theta_0+u_{j-1}/n, \theta_0+u_j/n)$. Let $\BB^n_{j-1}:= [\theta_0,\theta_0+u_{j-1}/n)$, then $$\label{eln}
\begin{aligned}
\ln Z_n(u_j)-&\ln Z_n(u_{j-1}) = \sum_{k=0}^{n-1}
\ln \frac
{
f\Big(X_{k+1}-m(X_k,\theta_0+u_j/n)\Big)
}
{
f\Big(X_{k+1}-m(X_k,\theta_0+u_{j-1}/n)\Big)
}= \\
&
\sum_{k=0}^{n-1}
\ln \frac
{
f\Big(\eps_{k+1}+m(X_k, \theta_0)-m(X_k,\theta_0+u_j/n)\Big)
}
{
f\Big(\eps_{k+1}+m(X_k, \theta_0)-m(X_k,\theta_0+u_{j-1}/n)\Big)
}=\\
&
\sum_{k=0}^{n-1}
\ln \frac
{
f\Big(\eps_{k+1}+\delta(X_k)\1_{\{X_k\in \BB^n_{j-1}\}}+\delta(X_k)\1_{\{X_k\in \DD^n_j\}})\Big)
}
{
f\Big(\eps_{k+1}+\delta(X_k)\1_{\{X_k\in \BB^n_{j-1}\}}\Big)
}=\\
&
\sum_{k=0}^{n-1}
\1_{\{X_k\in \DD^n_j\}}\ln \frac
{
f\Big(\eps_{k+1}+\delta(X_k)\1_{\{X_k\in \BB^n_{j-1}\}}+\delta(X_k)\Big)
}
{
f\Big(\eps_{k+1}+\delta(X_k)\1_{\{X_k\in \BB^n_{j-1}\}})\Big)
}\stackrel{\dagger}{=}\\
&
\sum_{k=0}^{n-1}
\1_{\{X_k\in \DD^n_j\}}\ln \frac
{
f\big(\eps_{k+1}+\delta(X_k)\big)
}
{
f\big(\eps_{k+1}\big)
}
=: \sum_{k=0}^{n-1} s^j_{k}
\end{aligned}$$ where the equality $\dagger$ holds $\Pb_{\theta_0}$-a.s., since $\Pb_{\theta_0}\Big(X_{k-1}\in \BB^n_{j-1}\cap \DD^n_j\Big)=0$. Further, define $$S_n:= \sum_{j=1}^d \lambda_j \Big(\ln Z_n(u_j)-\ln Z_n(u_{j-1})\Big)=\sum_{j=1}^d \lambda_j \sum_{k=0}^{n-1} s^j_{k}.$$ We shall partition $n$ terms of this sum into $n^{1/2}$ consecutive blocks of size $n^{1/2}$ and discard from each block its $n^{1/4}$ first entries. As we shall see, this does not alter the asymptotic distribution of $S_n$, but makes the blocks almost independent. Since in each block, the single event $\{X_k\in D^n_j\}$ occurs with probability of order $n^{1/2}$, the Poisson behavior emerges. To implement these heuristics, define $$S_{m,n} := \sum_{j=1}^d \lambda_j \sum_{k = (m-1) n^{1/2}+n^{1/4}}^{m n^{1/2}} s_k^j, \quad m= 1,...,n^{1/2},$$ and set $
\widetilde S_n:= \sum_{m=1}^{n^{1/2}}S_{m,n}
$ (this is the sum, in which the $n^{1/4}$ entries of each block have been discarded). By the triangle inequality $$\begin{gathered}
\label{tri}
\Big|\Ex_{\theta_0} e^{\iunit S_n}- e^{H(\lambda)}\Big|\le
\Big|\Ex_{\theta_0} e^{\iunit S_n}- \Ex_{\theta_0} e^{\iunit \widetilde S_n}\Big| + \\
\bigg|\Ex_{\theta_0} e^{\iunit \widetilde S_n} -\Big(\Ex_{\theta_0} e^{\iunit S_{1,n}}\Big)^{n^{1/2}}\bigg|
+
\bigg|\Big(\Ex_{\theta_0} e^{\iunit S_{1,n}}\Big)^{n^{1/2}}- e^{H(\lambda)}\bigg|.\end{gathered}$$ We shall show that all the terms on the right hand side vanish as $n\to\infty$. By stationarity and the assumption \[c5\], $$\begin{aligned}
&\Big|\Ex_{\theta_0} e^{\iunit S_n}-\Ex_{\theta_0} e^{\iunit\widetilde
S_n}\Big|\le \Ex_{\theta_0} \Big| e^{\iunit S_n}-e^{\iunit\widetilde
S_n}\Big| \le
\Ex_{\theta_0} \Big| S_n-\widetilde S_n\Big|\le\\
&n^{3/4} \max_j |\lambda_j|\Ex_{\theta_0} \1_{\{X_0\in \DD^n_j\}}
\left|
\ln \frac
{
f\big(\eps_{1}+\delta(X_0)\big)
}
{
f\big(\eps_{1}\big)
}
\right| = \\
&
n^{3/4} \max_j |\lambda_j|\Ex_{\theta_0} \1_{\{X_0\in \DD^n_j\}} J\big(\delta(X_0)\big)\le \\
&n^{3/4} \max_j |\lambda_j| \frac{u_j-u_{j-1}}{n}\sup_{x\in \RR}\varphi(x,\theta_0)\sup_{\theta\in \Theta}J(\delta(\theta))
\xrightarrow{n\to\infty}0,\end{aligned}$$ i.e. the first term in converges to zero.
Further, note that by the Markov property of $(X_j)$ and \[c4\] $$\bigg|\Ex_{\theta_0} \Big(e^{\iunit S_{\ell,n}}\big|\F_{(\ell-1)n^{1/2}}\Big)-\Ex_{\theta_0} e^{\iunit S_{1,n}}\bigg| \le C r^{n^{1/4}}, \quad
\ell=1,...,n^{1/2}$$ and hence $$\begin{aligned}
&\bigg|\Ex_{\theta_0} e^{\iunit \widetilde S_n} -\Big(\Ex_{\theta_0} e^{\iunit S_{1,n}}\Big)^{n^{1/2}}\bigg| =
\bigg|\Ex_{\theta_0}\prod_{m=1}^{n^{1/2}} e^{\iunit S_{m,n}} -\Big(\Ex_{\theta_0} e^{\iunit S_{1,n}}\Big)^{n^{1/2}}\bigg|=\\
&\bigg|\sum_{\ell=1}^{n^{1/2}}
\bigg(
\Ex_{\theta_0}\prod_{m=1}^{\ell} e^{\iunit S_{m,n}}
\Big( \Ex_{\theta_0} e^{\iunit S_{1,n}} \Big)^{n^{1/2}-\ell}
-
\Ex_{\theta_0}\prod_{m=1}^{\ell-1} e^{\iunit S_{m,n}}
\Big( \Ex_{\theta_0} e^{\iunit S_{1,n}} \Big)^{n^{1/2}-\ell+1}
\bigg)
\bigg|=\\
&
\bigg|\sum_{\ell=1}^{n^{1/2}}
\bigg(
\Ex_{\theta_0}\prod_{m=1}^{\ell-1} e^{\iunit S_{m,n}}
\Big(
e^{\iunit S_{\ell,n}}-\Ex_{\theta_0} e^{\iunit S_{1,n}}
\Big)
\Big( \Ex_{\theta_0} e^{\iunit S_{1,n}} \Big)^{n^{1/2}-\ell}
\bigg)
\bigg|=\\
&
\bigg|\sum_{\ell=1}^{n^{1/2}}
\bigg(
\Ex_{\theta_0}\prod_{m=1}^{\ell-1} e^{\iunit S_{m,n}}
\Big(
\Ex_{\theta_0} \Big(e^{\iunit S_{\ell,n}}\big|\F_{(\ell-1)n^{1/2}}\Big)-\Ex_{\theta_0} e^{\iunit S_{1,n}}
\Big)
\Big( \Ex_{\theta_0} e^{\iunit S_{1,n}} \Big)^{n^{1/2}-\ell}
\bigg)
\bigg|\le \\
&
\sum_{\ell=1}^{n^{1/2}}\Ex_{\theta_0}
\bigg|
\Ex_{\theta_0} \Big(e^{\iunit S_{\ell,n}}\big|\F_{(\ell-1)n^{1/2}}\Big)-\Ex_{\theta_0} e^{\iunit S_{1,n}}
\bigg|\le C n^{1/2} r^{n^{1/4}}\xrightarrow{n\to\infty}0.\end{aligned}$$ It is left to show that the last term in converges to zero. Let $\DD^n = \bigcup_{j=1}^d \DD^n_j$ and introduce the following events $$\begin{aligned}
& A_0 := \bigcap_{\ell\le n^{1/2}}\{X_\ell\not\in \DD^n\},\quad A_1 :=
\bigcup_{j=1}^d\bigcup_{\ell=0}^{n^{1/2}} A_{\ell,j},\qquad A_{2+} :=
\Big(A_0\cup A_1\Big)^c \\
& A_{k,j} := \{X_k\in \DD^n_j\}\cap
\bigcap_{\ell\le n^{1/2}, \ell\ne k} \{X_\ell\not \in \DD^n\} .\end{aligned}$$ In words, $A_0$ is the event, on which none of the first $n^{1/2}$ samples falls in any of $\DD^n_j$’s, $A_1$ is the event of having exactly single sample visiting one of $\DD^n_j$’s, etc. On the event $A_{k,j}$, $$S_{1,n} = \sum_{i=1}^d \lambda_i \sum_{\ell =n^{1/4}}^{ n^{1/2}} \1_{\{X_\ell\in \DD^n_i\}}\ln \frac
{
f\big(\eps_{\ell+1}+\delta(X_\ell)\big)
}
{
f\big(\eps_{\ell+1}\big)
}=
\lambda_j \ln \frac
{
f\big(\eps_{k+1}+\delta(X_k)\big)
}
{
f\big(\eps_{k+1}\big)
}$$ and, since $\{X_k\in \DD^n_j\}=A_{k,j}\biguplus\Big(\{X_k\in \DD^n_j\}\cap \bigcup_{\ell\ne k} \{X_\ell \in \DD^n\}\Big)$, $$\begin{aligned}
\label{plugme}
&\Ex_{\theta_0} e^{\iunit S_{1,n}}\1_{A_1} = \sum_{j=1}^d \sum_{k=0}^{n^{1/2}} \Ex_{\theta_0} e^{\iunit S_{1,n}}\1_{A_{k,j}}=
\sum_{j=1}^d \sum_{k=0}^{n^{1/4}-1} \Pb_{\theta_0}(A_{k,j})+\\
\nonumber
&\sum_{j=1}^d \sum_{k=n^{1/4}}^{n^{1/2}} \Ex_{\theta_0} \exp\bigg(\iunit \lambda_j \ln \frac
{
f\big(\eps_{k+1}+\delta(X_k)\big)
}
{
f\big(\eps_{k+1}\big)
}\bigg)\Big(\1_{\{X_k\in \DD^n_j\}}-\1_{\{X_k\in \DD^n_j\}\cap \bigcup_{\ell\ne k} \{X_\ell \in \DD^n\}}\Big).\end{aligned}$$ By continuity of $\varphi(x,\theta_0)$ and $\delta(x)$, $$\Pb_{\theta_0}(A_{k,j})\le \Pb_{\theta_0}(X_k\in \DD^n_j)=\frac{u_j-u_{j-1}}{n} \varphi(\theta_0,\theta_0) + o(n^{-1}),$$ and $$\begin{aligned}
&\Ex_{\theta_0} \exp\bigg(\iunit \lambda_j \ln \frac
{
f\big(\eps_{k+1}+\delta(X_k)\big)
}
{
f\big(\eps_{k+1}\big)
}
\bigg)\1_{\{X_k\in \DD^n_j\}} =\\
&\Ex_{\theta_0} \exp\bigg(\iunit \lambda_j \ln \frac
{
f\big(\eps_{k+1}+\delta(\theta_0)\big)
}
{
f\big(\eps_{k+1}\big)
}
\bigg)\1_{\{X_k\in \DD^n_j\}} + o(n^{-1})=\\
&\Psi(\lambda_j)\frac{u_j-u_{j-1}}{n}\varphi(\theta_0,\theta_0) + o(n^{-1}).\end{aligned}$$ Further, by the Markov property, for $k<\ell$ $$\begin{aligned}
& \Pb_{\theta_0}\big(X_k\in \DD^n_j, X_\ell \in \DD^n\big) = \Ex_{\theta_0}\1_{X_k\in \DD^n_j}\Pb_{\theta_0}\big(X_\ell \in \DD^n|\F_{\ell-1}\big) =\\
&
\Ex_{\theta_0}\1_{X_k\in \DD^n_j}\int_{\DD^n}
f\Big(x-h(X_{\ell-1})\1_{\{X_{\ell-1}<\theta_0\}}-g(X_{\ell-1})\1_{\{X_{\ell-1}\ge\theta_0\}}\Big)dx \le \\
& C_1 n^{-1} \Pb_{\theta_0}\big(X_k\in \DD^n_j\big) \le C_2n^{-2},\end{aligned}$$ where the inequalities hold, since the density $f(x)$ and therefore the invariant density $\varphi(x,\theta_0)$, $x\in \RR$ are bounded. Similar bound holds for $k>\ell$ and it follows that $$\Pb_{\theta_0}\bigg(\{X_k\in \DD^n_j\}\cap \bigcup_{\ell\ne k, \ell\le n^{1/2}} \{X_\ell \in \DD^n\}\bigg) \le
\sum_{\ell\le n^{1/2}, \ell\ne k}\Pb_{\theta_0}\big(X_k\in \DD^n_j, X_\ell \in \DD^n\big) \le C_3 n^{-3/2}.$$ Plugging these estimates into , we get $$\Ex_{\theta_0} e^{\iunit S_{1,n}}\1_{A_1}=n^{-1/2}\sum_{j=1}^ d\Psi(\lambda_j)\big(u_j-u_{j-1}\big)\varphi(\theta_0,\theta_0)+o(n^{-1/2}).$$ If we set all $\lambda_j$’s to zeros, we also obtain $$\label{A1}
\Pb_{\theta_0}(A_1)=n^{-1/2}\sum_{j=1}^ d\big(u_j-u_{j-1}\big)\varphi(\theta_0,\theta_0)+o(n^{-1/2}).$$ Further, $$\begin{gathered}
\Pb_{\theta_0}(A_0)=1- \Pb_{\theta_0}\left(\bigcup_{\ell\le n^{1/2}}\{X_\ell\in \DD^n\}\right) \ge 1-
\sum_{\ell\le n^{1/2}}\Pb_{\theta_0}\big(X_\ell\in \DD^n\big)= \\
1-
\sum_{\ell\le n^{1/2}}\sum_{j=1}^d\Pb_{\theta_0}\big(X_\ell\in \DD^n_j\big)=1-
n^{-1/2}\sum_{j=1}^ d\big(u_j-u_{j-1}\big)\varphi(\theta_0,\theta_0)+o(n^{-1/2}).\end{gathered}$$ On the other hand, $\Pb_{\theta_0}(A_0)\le 1-\Pb_{\theta_0}(A_1)$ and in view of , it follows that $$\label{A0}
\Pb_{\theta_0}(A_0)=1-
n^{-1/2}\sum_{j=1}^ d\big(u_j-u_{j-1}\big)\varphi(\theta_0,\theta_0)+o(n^{-1/2}).$$ Finally, using and , we also have $$\Pb_{\theta_0}(A_{2+}) = 1-\Pb_{\theta_0}(A_0)-\Pb_{\theta_0}(A_1) = o(n^{-1/2}).$$ Assembling all parts together, we obtain the asymptotic $$\begin{gathered}
\Ex_{\theta_0} e^{\iunit S_{1,n}} = \Pb_{\theta_0}(A_0)+\Ex_{\theta_0} e^{\iunit S_{1,n}}\1_{A_1} + \Ex_{\theta_0} e^{\iunit S_{1,n}}\1_{A_{2+}} =\\
1+n^{-1/2}\bigg(\sum_{j=1}^ d \Big(\Psi(\lambda_j)-1\Big)\big(u_j-u_{j-1}\big)\varphi(\theta_0,\theta_0)
\bigg) +o(n^{-1/2}),\end{gathered}$$ and, in turn, $$\lim_n \bigg|\Big(\Ex_{\theta_0} e^{\iunit S_{1,n}}\Big)^{n^{1/2}}- e^{H(\lambda)}\bigg|=0.$$ The claim now follows from .
Equicontinuity
--------------
The next step is to show that for some $C>0$ $$\label{e1}
\Ex_{\theta _0}\left(Z_n^{1/2}\left(u_2\right)-Z_n^{1/2}\left(u_1\right)\right)^2\leq C\,\left|u_2-u_1\right|.$$ As in Lemma 2.4 in [@CKL1], for e.g. $u_2>u_1>0$, gives $$\begin{aligned}
&\Ex_{\theta_0}\left(Z_n^{1/2}\left(u_2\right)-Z_n^{1/2}\left(u_1\right)\right)^2\leq
\Ex_{\theta _0+{u_1}/n} \ln
\frac{Z_n\left(u_1\right)}{Z_n\left(u_2\right)}\\
& \le \Ex_{\theta _0+{u_1}/n}
\sum_{j=0}^{n-1}\left|\ln\frac{f \left(\varepsilon _{j+1} \right)
}{f\left(\delta \left(X_j\right)+\varepsilon _{j+1} \right)}\right| \,\1_{\left\{\theta
_0 +\frac{u_1}{n}\leq X_j < \theta_0+\frac{u_2}{n} \right\}}\\
& =\Ex_{\theta _0+{u_1}/n}
\sum_{j=0}^{n-1}\Ex_{\theta _0+{u_1}/n} \left(\left|\ln\frac{f
\left(\varepsilon _{j+1} \right)
}{f\left(\delta \left(X_j\right)+\varepsilon _{j+1} \right)}\right|\Big|{\cal F}_j \right) \,\1_{\left\{\theta
_0 +\frac{u_1}{n}\leq X_j < \theta_0+\frac{u_2}{n} \right\}}\\
& =\Ex_{\theta _0+{u_1}/n}
\sum_{j=0}^{n-1}J\left(\delta \left(X_j\right)\right) \,\1_{\left\{\theta
_0 +\frac{u_1}{n}\leq X_j < \theta_0+\frac{u_2}{n} \right\}}\\
& =n\int_{\theta
_0 +\frac{u_1}{n}}^{\theta_0+\frac{u_2}{n} }J\left(\delta
\left(x\right)\right) \,\varphi \left(x,\theta_0+u_1/n\right)\,{\rm d}x \leq C\,\left|u_2-u_1\right|,\end{aligned}$$ as required.
Large deviations estimate
-------------------------
Finally we shall prove that for any $p>0$ there exists a constant $C_p>0$ such that $$\label{e2}
\Ex_{\theta _0}Z_n^{1/2}\left(u\right)\leq \frac{C_p}{\left|u\right|^p}.$$ We shall only sketch the proof, as most of the arguments can be directly adopted from the proof of Lemma 2.2 in [@ChK11] or Lemma 2.5, [@CKL1]. Note that for any $c>0$, $$\begin{gathered}
\label{une}
\Ex_{\theta _0}Z_n^{1/2}(u) = \Ex_{\theta _0}Z_n^{1/2}(u)\1_{\{Z_n^{1/2}(u)>e^{-c|u|}\}}
+
\Ex_{\theta _0}Z_n^{1/2}(u)\1_{\{Z_n^{1/2}(u)\le e^{-c|u|}\}}\le \\
\big(\Ex_{\theta _0}Z_n(u)\big)^{1/2}\Pb_{\theta_0}^{1/2}\big(Z_n^{1/2}(u)>e^{-c|u|}\big)
+
e^{-c|u|} = \Pb_{\theta _0}^{1/2}\big(\ln
Z_n^{1/2}\left(u\right)>-c\left|u\right|\big) +e^{-c\left|u\right|}\end{gathered}$$ and hence it suffices to show that for some $c>0$, $$\Pb_{\theta _0}\big(\ln
Z_n^{1/2}\left(u\right)>-c\left|u\right|\big)\le \frac{C_p}{\left|u\right|^p}, \quad p>0.$$ For $u>0$ (and similarly for $u<0$), $$\begin{gathered}
\label{doux}
\Pb_{\theta _0}\left\{\ln
Z_n^{1/2}\left(u\right)>-c\left|u\right|\right\}\\
=\Pb_{\theta
_0}\left\{\sum_{j=0}^{n-1}\ln\left[\frac{f \left(\delta
\left(X_j\right)+\varepsilon _{j+1} \right) }{f\left(\varepsilon _{j+1}
\right)}\right] \,\1_{\left\{\theta _0 \leq X_j <
\theta_0+u/n \right\}}>-2cu\right\}.\end{gathered}$$ Let $
\ell\left(x,y\right):=\ln\left[\frac{f \left(\delta
\left(x\right)+y \right) }{f\left(y
\right)}\right]
$ and introduce the notations $$\begin{aligned}
G\left(\delta \right)&=-\ln H\left(\delta\right),\\
S_n^{\left(1\right)}&=\sum_{j=0}^{n-1}\ell\left(X_j,\varepsilon
_{j+1}\right) \,\1_{\left\{X_j\in\BB^n \right\}},\quad
S_n^{\left(2\right)}=\sum_{j=0}^{n-1}G\left(\delta\left(X_j\right)
\right) \,\1_{\left\{X_j\in\BB^n \right\}},\end{aligned}$$ where $\BB^n=[\theta_0, \theta_0+u/n]$ and $$H\left(\delta\right):=\int_{-\infty }^{\infty }
\left(\frac{f\left(\delta
+y\right)}{f\left(y\right)}\right)^{1/2}f\left(y\right)\,{\rm
d}y$$ is the Hellinger integral of order $1/2$. By the Jensen inequality for all $\delta\ne 0$, $H(\delta)<1$ and hence $G\left(\delta \right)>0$.
Further, we have the following identity $$\begin{aligned}
\label{eg}
\Ex_{\theta _0}e^{\frac 1 2 S_n^{\left(1\right)}+S_n^{\left(2\right)}}=1.\end{aligned}$$ Indeed $$\begin{aligned}
\Ex_{\theta
_0}e^{\frac 1 2 S_n^{\left(1\right)}+S_n^{\left(2\right)}}=\Ex_{\theta
_0}e^{\frac 1 2 S_{n-1}^{\left(1\right)}+S_{n-1}^{\left(2\right)}}\Ex_{\theta
_0}\left(\left. e^{\left[\frac 1 2 \ell\left(X_{n-1},\varepsilon
_{n}\right)+G\left(\delta
\left(X_{n-1}\right)\right)\right]\1_{\left\{X_{n-1}\in\BB^n
\right\}}}\right|{\cal F}_{n-1}\right)\end{aligned}$$ and $$\begin{aligned}
&\Ex_{\theta _0}\left(\left. e^{ \frac 1 2 \ell\left(X_{n-1},\varepsilon
_{n}\right)\1_{\left\{X_{n-1}\in\BB^n \right\}}}\right|{\cal
F}_{n-1}\right)=\\
&\qquad \Ex_{\theta _0}\left(\left. e^{\frac 1 2
\ell\left(X_{n-1},\varepsilon _{n}\right)}\right|{\cal F}_{n-1}\right)
\1_{\left\{X_{n-1}\in\BB^n
\right\}}+\1_{\left\{X_{n-1}\not\in\BB^n\right\}} =\\
&\qquad
\int_{-\infty }^{\infty }\left(\frac{f\left(\delta
\left(X_{n-1}\right)+y\right)}{f\left(y\right)}\right)^{1/2}f\left(y\right){\rm
d}y \;\1_{\left\{X_{n-1}\in\BB^n\right\}}+\1_{\left\{X_{n-1}\not\in\BB^n \right\}}=\\
&\qquad
\exp\Big(- G \left(\delta
\left(X_{n-1}\right)\right)\;\1_{\left\{X_{n-1}\in\BB^n\right\}}\Big).\end{aligned}$$ Hence $$\Ex_{\theta
_0}\left(\left. e^{\left[\frac 1 2 \ell\left(X_{n-1},\varepsilon
_{n}\right)+G\left(\delta
\left(X_{n-1}\right)\right)\right]\1_{\left\{X_{n-1}\in\BB^n
\right\}}}\right|{\cal F}_{n-1}\right)=1$$ and follows. Now we have $$\begin{gathered}
\Pb_{\theta _0}\left\{\sum_{j=0}^{n-1}\ell\left(X_j,\varepsilon _{j+1}\right)
\,\1_{\left\{X_j\in \BB^n \right\}}>-2cu\right\}=\Pb_{\theta _0}\left\{
\frac 1 2 S_n^{\left(1\right)}+S_n^{\left(2\right)} - S_n^{\left(2\right)}
>-cu\right\} \\
\leq \Pb_{\theta _0}\left\{
\frac 1 2 S_n^{\left(1\right)}+S_n^{\left(2\right)}
>\frac 1 2 cu\right\} +\Pb_{\theta _0}\left\{
- S_n^{\left(2\right)}
>-\frac 3 2cu\right\} \leq e^{-\frac 1 2 cu} +\Pb_{\theta _0}\left\{
S_n^{\left(2\right)}
<\frac 3 2 cu\right\}\end{gathered}$$ where we used . In view of and , it is left to show that for all $p>1$, $$\Pb_{\theta _0}\left\{
\sum_{j=0}^{n-1}G \left(\delta
\left(X_{j}\right)\right)\1_{\left\{X_j\in \BB^n
\right\}}
<\frac 3 2 cu\right\}\leq \frac{C_p}{\left|u\right|^p}.$$
Following [@ChK11], we shall split the consideration into the cases $u<n^s$ and $n^s\leq u<n\left(\beta -\alpha\right)$, for some $s\in\left(0,1\right)$.
To this end, note that the Hellinger integral $H(\delta)$ is a continuous function of $\delta$: $$\begin{aligned}
&\big(H(\delta)-H(\delta+\eta)\big)^2 = \left(\int_{-\infty}^\infty \left[\left(\frac{f(\delta+y)}{f(y)}\right)^{1/2}-
\left(\frac{f(\delta+\eta+y)}{f(y)}\right)^{1/2} \right]f(y) dy\right)^2 \le \\
& \int_{-\infty}^\infty \left( \left(\frac{f(\delta+y)}{f(y)}\right)^{1/2}-
\left(\frac{f(\delta+\eta+y)}{f(y)}\right)^{1/2}\right)^2 f(y) dy =\\
& 2- 2
\int_{-\infty}^\infty \sqrt{ f(\delta+y) f(\delta+\eta+y)} dy =
\int_{-\infty}^\infty \Big( \sqrt{f( y)} - \sqrt{f( \eta+y)}\Big)^{2} dy \le \\
& \int_{-\infty}^\infty \big|f( y)-f( \eta+y)\big|dy\xrightarrow{\eta\to 0}0\end{aligned}$$ where we used LeCam’s inequality for the Hellinger and the total variation distances and the convergence holds by Scheffe’s lemma.
By continuity of $G(\delta)=-\ln H(\delta)$ and since $G(\delta)>0$ for all $\delta\ne 0$, the assumption \[c2\] implies that for $u<n^s$ $$\begin{aligned}
G\left(\delta
\left(X_{j}\right)\right)\1_{\left\{X_j\in \BB^n
\right\}}\geq \inf_{\theta _0\leq v\leq \theta _0+n^{s-1}}G\left(\delta
\left(v\right)\right)\geq c_0\end{aligned}$$ with some constant $c_0>0$ and $$\begin{aligned}
\Pb_{\theta _0}\left\{ S_n^{\left(2\right)} <\frac 3 2cu\right\}\leq
\Pb_{\theta _0}\left\{ \sum_{j=0}^{n-1}\1_{\left\{X_j\in \BB^n \right\}}
<\frac 3 2\frac{ c}{c_0}\,u\right\}.\end{aligned}$$ Now let $\eta _j\left(u\right)=\Ex_{\theta
_0}\1_{\left\{X_j\in \BB^n \right\}}
-\1_{\left\{X_j\in \BB^n \right\}} $. Since the density $f(x)$ is continuous and positive, so is the invariant density $\varphi \left(x,\theta _0\right)$ and $$\begin{aligned}
S_n^{\left(3\right)}&=\sum_{j=0}^{n-1}\Ex_{\theta
_0}\1_{\left\{X_j\in \BB^n\right\}} =n\int_{\theta
_0}^{\theta _0+u/n}\varphi \left(x,\theta _0\right)\,{\rm d}x\geq C'\,u,\end{aligned}$$ with a positive constant $C'$. Then $$\begin{gathered}
\Pb_{\theta _0}\left\{ \sum_{j=0}^{n-1}\1_{\left\{X_j\in \BB^n
\right\}} <\frac 3 2\frac{c}{c_0}\,u\right\} =\Pb_{\theta _0}\left\{
-\sum_{j=0}^{n-1} \eta _j\left(u\right)
<-\left(S_n^{\left(3\right)}-\frac 3 2\frac{c}{c_0}u \right) \right\}\\
\qquad
\qquad \leq \Pb_{\theta _0}\left\{ \sum_{j=0}^{n-1} \eta _j\left(u\right)
>\kappa u \right\} \leq \frac{\Ex_{\theta _0} \left|\sum_{j=0}^{n-1} \eta
_j\left(u\right) \right|^{2p}}{\left|\kappa u\right|^{2p}},\end{gathered}$$ where we chose $c$ small enough, so that $C'-\frac 3 2c/c_0=\kappa >0$. Using the geometric mixing property \[c4\] and an appropriate version of Rosenthal’s inequality as in Lemma 2.2 [@ChK11], we get $$\Ex_{\theta _0} \left|\sum_{j=0}^{n-1} \eta
_j\left(u\right) \right|^{2p}\leq C\left(p\right)\:\left|u\right|^p$$ which yields for $|u|<n^s$. The complementary case, $n^s\le |u|\le (\beta-\alpha)n$ is treated exactly as in Lemma 2.2, [@ChK11] or Lemma 2.5, [@CKL1].
Discussion
==========
Theorem \[T1\] can be directly generalized to the multi-threshold autoregression $$X_{j+1}=\sum_{k=0}^{K}h_k\left(X_j\right)\1_{\left\{\theta _{k}<X_j\leq
\theta _{k+1}\right\}}+\varepsilon _{j+1},\quad j=0,1,\ldots, n,$$ where ${\bm \theta} =\left(\theta _1,\ldots,\theta _K\right)$ is the unknown parameter (and $\theta_0=-\infty$ and $\theta_{K+1}=\infty$ are set). As in , $(\varepsilon _j)$ are independent random variables with known density $f\left(x\right)>0,x\in \RR$ and the functions $h_k\left(\cdot \right) $ and $f\left(\cdot \right)$ are continuous and such that $(X_j)$ is geometrically mixing. Assume that $\theta _k\in\left(\alpha _k,\beta _k\right)$, where $\beta_k<\alpha _{k+1}$.
For all sufficiently large $n$ and $u_k\ge 0$, the normalized likelihood ratio is given by $$\begin{aligned}
Z_n\left({\bf
u}\right)&=\prod_{j=0}^{n-1}\frac{f\left(X_{j+1}-\sum_{k=0}^{K}h_k\left(X_j\right)\1_{\left\{\theta
_{k}+\frac{u_k}{n}<X_j\leq
\theta_{k+1}+\frac{u_{k+1}}{n}\right\}}\right)
}{f\left(X_{j+1}-\sum_{k=0}^{K}h_k\left(X_j\right)\1_{\left\{\theta
_{k}<X_j\leq \theta_{k+1}\right\}}\right)}\\
&=\prod_{j=0}^{n-1}\frac{f\left(\sum_{k=0}^{K}h_k\left(X_j\right)\left[\1_{\left\{\theta
_{k}<X_j\leq \theta_{k+1}\right\}}
-
\1_{\left\{\theta
_{k}+\frac{u_k}{n}<X_j\leq
\theta_{k+1}+\frac{u_{k+1}}{n}\right\}}\right]+\varepsilon _{j+1}\right)
}{f\left(\varepsilon _{j+1}\right)}\\
&=\prod_{j=0}^{n-1}\frac{f\left(\sum_{k=1}^{K}\left[h_{k-1}\left(X_j\right)-h_{k}\left(X_j\right)\right]
\1_{\left\{\theta
_{k}<X_j\leq
\theta_{k}+\frac{u_{k}}{n}\right\}}+\varepsilon _{j+1}\right)
}{f\left(\varepsilon _{j+1}\right)},\end{aligned}$$ and thus $$\begin{aligned}
\ln Z_n\left({\bf u}\right)=&
\sum_{j=0}^{n-1}\ln \frac{f\left(\sum_{k=1}^{K}\left[h_{k-1}\left(X_j\right)-h_{k}\left(X_j\right)\right]
\1_{\left\{\theta
_{k}<X_j\leq
\theta_{k}+\frac{u_{k}}{n}\right\}}+\varepsilon _{j+1}\right)
}{f\left(\varepsilon _{j+1}\right)} =\\
&
\sum_{k=1}^{K}
\sum_{j=0}^{n-1}\ln \frac{f\big(\delta_k(X_j)
+\varepsilon _{j+1}\big)
}{f\big(\varepsilon _{j+1}\big)} \1_{\left\{\theta
_{k}<X_j\leq
\theta_{k}+\frac{u_{k}}{n}\right\}},\end{aligned}$$ where $\delta_k(x):= h_{k-1}(x)-h_{k}(x)$. Using the same approach as in the proof of Theorem \[T1\], it can be seen that $$\begin{aligned}
\ln Z_n\left({\bf u}\right)\quad \Longrightarrow\quad
\sum_{k=1}^{K}\sum_{l=1}^{N_k^+\left(u_k\right)} \ln \frac{f\big(
\varepsilon^+_{k,l}+\delta_k(\theta_k)\big)
}{f\big(\varepsilon^+_{k,l}\big)},\end{aligned}$$ where $N_k^+\left(u_k\right), u_k\geq 0$ are independent Poisson processes with intensities $\varphi \left(\theta_k,\theta _k\right)$ and $\eps^+_{k,l}$ are i.i.d. random variables with the density $f$. Similar asymptotic is obtained for $u_k<0$. Consequently the limit likelihood ratio is a product on $K$ independent one-dimensional copies of the process defined (with $\theta_0$ replaced by $\theta_k$’s) and the corresponding Bayes estimates $\widetilde \theta_{k,n}$, $k=1,...,K$ are asymptotically independent with the asymptotic distribution as in Theorem \[T1\].
Acknowledgement {#acknowledgement .unnumbered}
===============
The authors are grateful to the referee for the careful proofreading of the manuscript and the suggested improvements.
[1]{}
K. S. Chan. Consistency and limiting distribution of the least squares estimator of a threshold autoregressive model. , 21(1):520–533, 1993.
N.H. Chan and Yu. A. Kutoyants. Recent developments of threshold estimation for nonlinear time series. , 40(2):277–308, 2010.
N.H. Chan and Yu. A. Kutoyants. On parameter estimations of threshold autoregressive models. , 1:81–104, 2012.
P. Chigansky, Yu. A. Kutoyants, and R. Liptser. Threshold estimation in autoregressive models driven by colored noise. arXiv preprint 1010.5105.
S. Dachian and I. Negri. On compound Poisson processes arising in change-point type statistical models as limiting likelihood ratio. , 14(3):255–271, 2011.
I. A. Ibragimov and R. Z. Has’minskii. . New York, 1981.
Yu. A. Kutoyants. On identification of the threshold diffusion processes. , 64(2):383-413, 2012.
S. Meyn and R. L. Tweedie. . Cambridge University Press, Cambridge, second edition, 2009.
H. Tong. Threshold models in time series analysis - 30 years on. , 4(2):107–118, 2011.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Quasinormal mode (QNM) gravitational radiation from black holes is expected to be observed in a few years. A perturbative formula is derived for the shifts in both the real and the imaginary part of the QNM frequencies away from those of an idealized isolated black hole. The formulation provides a tool for understanding how the astrophysical environment surrounding a black hole, e.g., a massive accretion disk, affects the QNM spectrum of gravitational waves. We show, in a simple model, that the perturbed QNM spectrum can have interesting features.'
address:
- '${}^{(1)}$Department of Physics, The Chinese University of Hong Kong, Hong Kong'
- '${}^{(2)}$McDonnell Center for the Space Sciences, Department of Physics, Washington University, St Louis, MO 63130, U S A'
author:
- 'P.T. Leung${}^{(1)}$, Y. T. Liu${}^{(1)}$, W.-M. Suen${}^{(1, 2)}$, C. Y. Tam${}^{(1)}$ and K. Young${}^{(1)}$'
title: Quasinormal Modes of Dirty Black Holes
---
[2]{}
[*1. Introduction.*]{} The new generation of gravitational wave observatories (LIGO, VIRGO) will soon be able to probe black holes in their dynamical interactions with the astrophysical environment (e.g., matter or another black hole falling into it). Numerical simulations [@num] show that the gravitational waves emitted in this process will carry a signature associated with the well-defined quasinormal mode (QNM) frequencies of the black hole, and thereby confirm its existence.
A stationary neutral black hole in an otherwise empty and asymptotically flat spacetime is a Kerr hole (Schwarzschild hole in the case of zero angular momentum) [@nohair]. Linearized gravitational waves propagating on the Kerr or Schwarzschild background can be described by the Klein-Gordon equation [@chand]:
$$\left[ {\partial_{t}^{2}} - {\partial_{x}^{2}} +V(x) \right] \phi
(x,t) = 0 ,
\label{eq:kg}$$
where $x$ is a radial (tortoise) coordinate, $\phi$ is the radial part of a combination of the metric functions representing the gravitational wave. The potential $V(x)$ describes the scattering of the gravitational waves by the background geometry. The outgoing wave boundary condition is appropriate for waves escaping to infinity, and a monochromatic solution \[$\phi \propto \exp(-i \omega t)$\] is a QNM, with Im $\omega < 0$. The QNM spectra of Kerr and Schwarzschild black holes have been extensively studied [@chand], and provide a template against which one can try to determine the nature of the source; for an isolated black hole, the no-hair theorem [@nohair] implies that the QNM spectrum depends only on the mass $M$ and the angular momentum $J$.
However, the black holes that are observed will not be isolated, but will be situated at the centers of galaxies, or will be surrounded by accretion disks. Therefore the observed spectra should not be matched against those of a pure Kerr or Schwarzschild hole, but to a black hole perturbed by interactions with its surrounding — a dirty black hole. So far, the perturbation of black hole QNMs has attracted little attention, partly because a perturbative formalism for the QNMs of open systems, as opposed to the normal modes (NMs) of conservative systems, has not hitherto been available. In this paper we develop such a formalism, which then opens the way to inferring the astrophysical environment of the black holes from the observed signal, beyond $M$ and $J$.
Two kinds of perturbations are involved here. In the standard black hole perturbation theory [@chand], (\[eq:kg\]) is obtained by linearizing the metric about the Kerr or Schwarzschild background, and the time-independent eigenvalue problem (with the outgoing wave boundary condition) determines the QNM spectrum. The second type of perturbations are [*the perturbations that change the background*]{} on which the wave propagates, e.g., by the presence of an accretion disk; these are often quasi-static, and hence separable from that of the gravitational wave perturbation by the time scales involved (in a suitable gauge choice). In this paper we focus on time-independent perturbation of the background, described by (\[eq:kg\]) with a potential $V(x) = V_0(x) + \mu V_1(x)$, $|\mu| \ll 1$. Therefore we are led to study the following eigenvalue problem in powers of $\mu$:
$$- \phi''(x) + \left[ V_0(x) + \mu V_1(x) \right] \phi = \omega^2 \phi
\label{eq:kg1}$$
While reminiscent of standard textbook problems, e.g., the usual Rayleigh-Schrödinger perturbation theory (RSPT), the problem here is fundamentally different: the outgoing wave condition renders the system physically nonconservative (energy escapes to infinity) and the associated operator $ - d^2/dx^2 +V(x)$ non-hermitian; hermiticity underpins the usual RSPT.
The difficulty can be seen in several guises if one tries naively to transcribe the usual formulas. The first-order shift cannot be given by the usual formula $\langle \phi_0 | \mu V_1 | \phi_0 \rangle / \langle
\phi_0 | \phi_0 \rangle$, in obvious notation — the usual inner product leads to $\langle \phi_0 | \phi_0 \rangle = \mbox{\Large
$\int$}_{-\infty}^{\infty} dx \phi_0^* \phi_0 = \infty$ since a QNM $
\phi_0$ extends over all space (and indeed grows exponentially at infinity). Higher-order shifts are even more problematic, since the usual RSPT formula involves a sum over intermediate eigenstates, but now the unperturbed eigenstates do not in general form a complete basis [@kg-comp], at least not in the case of black holes.
[*2. Formulation.*]{} Our formulation generalizes the logarithmic perturbation theory (LPT) [@3d] to QNM systems; LPT has the property that it does not require a complete set of eigenstates. Attention is focussed on the logarithmic derivative $f(x)=\phi'(x) / \phi(x)$. From (\[eq:kg1\])
$$f'(x) + f^2(x) - \left[ V_0(x) + \mu V_1(x) \right] + \omega^2 = 0
\label{eq:ricc}$$
For any $\omega$, we define two solutions $f_{\pm}(\omega,x)$ by the boundary conditions $f_{\pm}(\omega,x) \rightarrow \pm i\omega$ as $x \rightarrow \pm \infty$. At an eigenvalue $\omega$, $f_+(\omega,x) = f_-(\omega,x)$.
For many cases of interest, $V(x) = V_0(x) + \mu V_1(x)$ is nontrivial only in a finite domain $(L_-,L_+)$, and is relatively simple in the asymptotic regions $(-\infty, L_-)$ and $(L_+,\infty)$. In particular, we assume that the asymptotic regions can be solved with the outgoing wave conditions to give the logarithmic derivatives $D_{\pm}(\omega) = f_{\pm}(\omega,L_{\pm})$. We then expand all quantities in powers of $\mu$: $ f \equiv f_0 + g = f_0 + \mu
g_1 + \mu^2 g_2 + \cdots$; $\omega = \omega_0 + \mu
\omega_1 + \cdots$; $D_{\pm} = D_{\pm 0} + \mu D_{\pm 1}+ \cdots$.
While the details of the derivation will be given elsewhere, the central result for the $n$th order shift is
$$\omega_n = \frac{ \langle \phi_0 | V_n | \phi_0 \rangle}
{ 2 \omega_0 \langle \phi_0 | \phi_0 \rangle }
\label{eq:result}$$
in which we have introduced the suggestive notation
$$\begin{aligned}
\langle \phi_0 | V_n | \phi_0 \rangle &= &
\int_{L_-}^{L_+} \mbox{\Large{}}V_n(x) \phi_0^2(x) dx \nonumber \\
&-& \Delta_{+n} \phi_0^2(L_+) + \Delta_{-n} \phi_0^2(L_-)
\label{eq:matelm}\end{aligned}$$
$$\begin{aligned}
\langle \phi_0 | \phi_0 \rangle &=&
\int_{L_-}^{L_+} \phi_0^2(x) dx \nonumber \\
&+& \frac{1}{2\omega_0} \left\{ D_{+0}' \phi_0^2(L_+)
- D_{-0}' \phi_0^2(L_-) \right\}
\label{eq:norm}\end{aligned}$$
Here $V_1$ is the perturbing potential in (\[eq:kg1\]), and for $n>1$, $V_n(x)=-\sum_{i=1}^{n-1} \left[ g_i(x) g_{n-i}(x) + \omega_i \omega_{n-i}
\right]$, with
$$\begin{aligned}
\phi_0^2(x) g_n(x) = & & \left[
\omega_n D_{-0}'(\omega_0) + \Delta_{-n} \right] \phi_0^2(L_-)
\nonumber \\
&+&
\int_{L_-}^x dy \left[ V_n(y) - 2\omega_0 \omega_n
\right] \phi_0^2(y)
\label{eq:gn}\end{aligned}$$
Here $\Delta_{\pm n}$ is the $n$th-order part of $D_{\pm}(\omega)-D_{\pm 0}(\omega_0)$; explicitly $\Delta_{\pm 1} = D_{\pm 1}$, $\Delta_{\pm 2} = D_{\pm 2}+
\omega_1 D_{\pm 1}' + \frac{1}{2} \omega_1^2 D_{\pm 0}'' $ etc., where all $D_{\pm n}$ and their derivatives are understood to be evaluated at the unperturbed frequency $\omega_0$. These results express the $n$th order correction to the eigenvalue in quadrature in terms of lower-order quantities. One can hence in principle obtain the corrections to any order. Similar to LPT for conservative systems, a sum over intermediate states is not needed.
[*3. Properties of the Perturbed Spectrum For Open Systems in General.*]{} The result in (\[eq:result\]) has been written in a way formally similar to the hermitian case. The factor $2\omega_0$ occurs because the eigenvalue is $\omega^2$ rather than $\omega$. The numerator and the denominator in (\[eq:result\]) are separately independent of $L_{\pm}$, so that they can be given physical interpretations as a generalized matrix element and a generalized norm respectively.
The generalized norm has some unusual properties [@waveeq; @two-comp]. (a) It involves $\phi_0^2$ rather than $|\phi_0|^2$, and is in general complex. (b) It involves surface terms at $x = L_{\pm}$, though the value of the entire expression is independent of the choice of $L_{\pm}$. Thus, it is not a norm in the strict sense, but rather a useful bilinear map. Nevertheless, in cases where the system parameters can be tuned so that the leakage of the wavefunction approaches zero (e.g., $V_0(x)$ contains tall barriers on both sides), the generalized norm does reduce to the usual (real and positive-definite) norm for a NM.
It is useful to define a function $H(x)$ for each QNM which depends only on the original unperturbed system
$$\frac{\delta \omega}{\delta ( \mu V_1(x))} \equiv H(x)
= \frac{\phi_0(x)^2}{ 2 \omega_0 \langle \phi_0 | \phi_0 \rangle }$$
Both the magnitude and the phase of $H(x)$ are well defined and physically significant. The magnitude implies that we can now give a precise meaning to the normalization of a QNM, even though the wavefunction diverges at infinity. The phase of $H$ determines the phase of the first-order shift $\omega_1$ for a real and positive localized perturbation $V_1(x)$. The phase is intriguing because it has no counterpart for a hermitian system — in that case, $H(x)$ must be real and non-negative. The functions $H(x)$ are then convenient objects for discussing the effect of any perturbation on the QNMs of a given system. We next present some properties of $H(x)$ for the Schwarzschild black hole.
[*4. General Properties of the Perturbed Spectrum of a Schwarzschild Hole.*]{} Waves propagating on the exact Schwarzschild background geometry is described by (1) with the potential [@chand]
$$V_{Sc} (M,x) = \left( 1 - \frac{2M}{r} \right)
\left[ \frac{l(l+1)}{r^2} + (1 - s^2) \frac{2M}{r^3} \right]
\label{schw}$$
with $x = r + 2M \ln \left( r/2M -1 \right)$, where $s$ is the spin of the field ($s=2$ for gravitational waves).
In Fig. 1 we plot for the $s=0, l=1$ case the functions $H(x)$, which [*depends only on the unperturbed potential*]{}. The diagrams refer to the lowest QNMs (labeled as $j = 0, 1, \cdots , 5$). We note that for a localized perturbation $\mu V_1(x) = \mu \delta(x-x_1)$, the frequency shift $\omega_1$ is given by $H(x_1)$, and therefore can be read out directly from the figure. Both $\mbox{Re $H(x)$}$ and $\mbox{Im $H(x)$}$ alternate in sign as $(-1)^j$ near the event horizon. The patterns are different for different values of $x_1$ and not simple, demonstrating that a localized perturbation will push the QNMs along different directions in the complex frequency plane, generating a rich pattern of frequency shifts (in contrast to shifts all of the same phase in the case of the NMs of a conservative system). This implies much better prospects for extracting information about the perturbing potential from the observed shifts.
-0.5cm [Fig. 1 Graph of Re $\left[ H(x)e^{-2\gamma \sqrt{1+(x/2M)^2}}
\right]$ (solid line) and Im $\left[ H(x)e^{-2\gamma \sqrt{1+(x/2M)^2}} \right]$ (dashed line) vs $r/2M$, where $\gamma$=Im($-2M\omega$)]{}
0.5cm The richness of the pattern could be diluted if the perturbation has a spatial extent $\Delta x$ large compared to the typical wavelength of oscillation of $H(x)$, $\lambda \approx 2\pi / |\mbox{Re $\omega_0$}|
\approx $ a few $M$. Next we discuss a model problem with an effective potential which extends over an infinite range of $x$.
[*5. Perturbed Spectrum of a Schwarzschild Black Hole in a Model Problem.*]{} Consider a Schwarzschild hole surrounded by a static shell of matter. Denote the total mass of the system as measured at infinity (ADM mass) by $M_o$, and the mass of the black hole as measured by its horizon surface area by $M_a$. The perturbation is characterized by $\mu \equiv (M_o-M_a)/M_a$ and the circumferential radius $r=r_s$ where the shell is placed. For scalar wave ($s=0$), both the unperturbed potential $V_0$ and the perturbation $V_1$ can be given in terms of $V_{Sc}$ in (\[schw\]): $V_0 (x)=V_{Sc}(M_o, x)$; $ \mu V_1 (x) = \kappa \delta (x - x_s) + (\beta / \alpha) V_{Sc}(M_a, x) -
V_{Sc}(M_o, x)$, for $x < x_s$, and $V_1=0$ for $x > x_s$, where $x_s$ is the tortoise coordinate at $r_s$. The constants $\kappa , \alpha$ and $\beta$ are given by $M_o, \mu $ and $r_s$ in some complicated expressions. This perturbation consists of a $\delta$-function at the shell, plus a contribution inside the shell extending all the way to the horizon ($x \rightarrow -\infty$, $r \rightarrow 2M_a$). There is no perturbation outside the shell; in terms of the ADM mass, the outside metric is exactly that of a Schwarzschild hole with $M_o$.
For $x<0$, the full potential $V=V_0 + \mu V_1$ can be expressed as a sum of exponentials, for which (\[eq:kg1\]) with the outgoing wave boundary condition can be integrated analytically, thus giving the log derivatives $D_-$, whereas the log derivative $D_+$ is trivial because the perturbation vanishes outside the shell. The details of the treatment of exponential potentials will be given elsewhere [@expon].
We first demonstrate the convergence of the perturbation results. Fig. 2 shows the magnitude of the error in the frequencies in the 0th, 1st and 2nd order results versus $\mu$, for $l=1$ scalar waves, compared to the exact numerical results (which can be obtained by brute force in this simple case [@others].) The error of the $n$th order result goes as $\mu^{n+1}$, as it should. Detailed estimation of the error and the large $n$ behavior of the perturbation expansion will be given elsewhere.
-0.5cm [ Fig. 2 The magnitude of the error $e$ in the frequencies of the 0th (circles), 1st (squares) and 2nd order (triangles) perturbation for $l=1$, $s=0$, $j=1$, and $r_s=2.52 M_a$, vs the size of the perturbation $\mu$.]{}
0.5cm We next study the dependence on the parameters of the shell. Fig. 3a shows the trajectories of the lowest damping QNMs ($j=0,1,\cdots,6$) for different $r_s$ for the case of $l=1$ scalar wave with $\mu=0.01$ based on exact numerical calculation. We note the rich features of the perturbed spectra. As $r_s$ changes, the QNMs execute complicated trajectories on the complex $\omega$ plane, with the higher-order modes moving more rapidly as $r_s$ varies. This behavior is readily understood from the perturbation formula (\[eq:result\]), which gives
$$\omega_1 \sim e^{2i\omega_0 x_s}/x_s^2~~~~~\mbox{for~~~~}x_s/2M_a \gg 1.$$
With $\mbox{Im $\omega_0 < 0$}$, the QNMs move away from the unperturbed positions in an exponential fashion as $x_s$ increases, and the higher-order modes ($- \mbox{Im }2M_a \omega_0 \gg 1$) move with higher speed.
More intriguingly there are complicated fine structures in these trajectories. Fig. 3b shows the fine details of the trajectory of the $j=0$ mode. The results obtained by direct numerical integration and by the 1st order perturbation formula are shown. For larger $x_s$, the trajectory shows a spiral structure, which can be explained from the first-order perturbation formula:
$$\omega_1 = \int_{-\infty}^{x_s} H(x)V_1(x) dx +\kappa(x_s)H(x_s).$$
The asymptotic behavior of $H(x)$ is $H(x) \sim e^{2 i \omega_0 x}$, so for large $x_s$, ($'=d/dx_s$)
$$\begin{aligned}
\omega_1 ' \sim \left[
V_1(x_s)+\kappa'(x_s)+2i\omega_0 \kappa(x_s) \right] e^{2i\omega_0 x_s}.\end{aligned}$$
The exponential factor $e^{2i\omega_0 x_s}$ gives the spirial structure.
-1.5cm
-0.5cm [Fig. 3a (the upper graph) The trajectory of the lowest QNMs of $l=1$ scalar waves for $\mu=0.01$ and $r_s/M_a$ varying from $2.22$ to $60$. The circles are QNMs of a bare Schwarzschild hole with mass $M_o$; the squares are the QNMs for $r_s=2.22M_a$ (The dominant energy condition is violated when $r_s<2.22M_a$); the triangles show the positions of QNMs at $r_s/M_a=$ $6$ to $60$ in intervals of $6$. Fig. 3b (the lower graph) shows the detail of the trajectory of the $j=0$ mode based on exact (solid line) and 1st order (dashed line) calculation.]{}
0.5cm [*6. Conclusion.*]{} We have developed a formulation for the perturbation of QNMs, in close parallel to the familiar perturbation theory, which is directly applicable to black holes. With QNM gravitational wave signals from black holes to be detected soon, and many black holes expected to be perturbed by their astrophysical environments, e.g., accretion disks, this formulation will be of interest to gravitational wave astronomy.
Although the QNMs of any system can in principle be obtained through brute force numerical integration, perturbation formulas are often more revealing. We note the usefulness of perturbation theory in conventional conservative systems,e.g., in quantum mechanics. Moreover, the numerical integration of QNM spectrum is much more difficult than for NM system.
In summary, we raise the importance of studying the QNMs of dirty black holes, and have developed a perturbation formulation for this purpose. The formulation opens the way to extracting rich information from gravitational wave signals from black hole events, and leads the way to study of the inverse problem. We show in a simple example that the perturbed spectrum shows interesting features, which can be understood with the perturbation formula.
This work is supported in part by the Hong Kong Research Grants Council grant 452/95P, and the US NSF grant PHY 96-00507. WMS also wants to thank the support of the Institute of Mathematical Science of The Chinese University of Hong Kong. We thank C. K. Au for discussions about LPT.
See e.g., P. Anninos [*et. al.*]{}, Phys. Rev. Lett. [**71**]{}, 2851 (1993) and references therein.
See e.g., S. W. Hawking and G. F. R. Ellis, [*The Large Scale Structure of Spacetime*]{} (Cambridge U. Press, 1973).
See e.g., S. Chandrasekhar, [*The Mathematical Theory of Black Holes*]{} (Oxford U. Press, 1983); W. H. Press, Ann. New York Acad. Sc. [**224**]{}, 272, (1973), and references therein.
E.S.C. Ching, P.T. Leung, W.M. Suen, and K. Young, Phys. Rev. Lett. [**74**]{}, 4588 (1995); Phys. Rev. D [**54**]{}, 3778 (1996).
Y. Aharonov and C.K. Au, Phys. Rev. Lett. [**42**]{}, 1582 (1979); C.K. Au, Phys Rev A [**29**]{}, 1034 (1984).
P.T. Leung, S.Y. Liu and K. Young, Phys. Rev. A [**49**]{}, 3057 (1994); P.T. Leung, S.Y. Liu, S.S. Tong and K. Young, Phys. Rev. A [**49**]{}, 3068 (1994); P.T. Leung, S.Y. Liu and K. Young, Phys. Rev. A [**49**]{}, 3982 (1994).
P. T. Leung, et. al., “Two-Component Eigenfunction Expansion for open Systems Described by the Wave Equation I & II”, to appear in J. Phys. A.
P.T. Leung et. al. in preparation.
R. H. Price, Phys. Rev. D [**5**]{}, 2439, (1972); E.W. Leaver, Proc. R. Soc. London A [**402**]{}, 285 (1985); J.W. Guinn, C.M. Will, Y. Kojima and B.F. Schutz, Class. Quantum Grav. [**7**]{}, L47 (1990); E. Leaver, Class. Quantum Grav. [**9**]{}, 1643 (1992); F. Andersson and S. Linnaeus, Phys. Rev. D [**46**]{}, 4179 (1992); H.-P. Nollert, Phys. Rev. D [**47**]{}, 5253 (1993).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'If Gamma-Ray Bursts (GRBs) are cosmologically distributed standard candles and are associated with the luminous galaxies, then the observed angular distribution of all GRBs is altered due to weak gravitational lensing of bursts by density inhomogeneities. The amplitude of the effect is generally small. For example, if the current catalogs extend to $z_{max}\sim 1$ and we live in a flat $\Omega=1$ Universe, the angular auto-correlation function of GRBs will be enhanced by $\sim 8\%$ due to lensing, on all angular scales. For an extreme case of $z_{max}= 1.5$ and ($\Omega$, $\Lambda$)=(0.2, 0.8), an enhancement of $\sim 33\%$ is predicted. If the observed distribution of GRBs is used in the future to derive power spectra of mass density fluctuations on large angular scales, the effect of weak lensing should probably be taken into account.'
author:
- 'L. L. R. Williams'
title: 'The Effect of Weak Gravitational Lensing on the Angular Distribution of Gamma-Ray Bursts'
---
Introduction
============
After more than two decades of GRB observations their physical nature and distances remain largely a mystery. The distribution of bursts is isotropic in the sky (Quashnock 1996, Tegmark 1996). This observation rules out Galactic disk/bulge models, but leaves Galactic halo (Lamb 1995) and cosmological models as possible scenarios. The Spectroscopy Detectors on board the Compton Gamma Ray Observatory detected no convincing spectral features to help determine GRB redshifts (Palmer 1993). Searching for counterparts of GRBs at other wavelengths (optical, IR, X-rays) could help determine their distances, however none were found within the error boxes of GRBs (Shaefer 1993, but see Irwin & Żytkow 1994). In fact, the lack of any bright optical or IR galaxies in the error boxes of 10 brightest GRBs indicates that the bursts must originate at distances greater than 1 Gpc (Shaefer 1990).
If the bursts are extragalactic, they may be expected to trace luminous matter in the Universe. For example, if they are associated with coalescing black holes or neutron stars, their rate would be roughly proportional to the density of normal stellar populations. In particular, the bursts should have clustering properties similar to those of the galaxies. All attempts to find angular correlation of GRBs with themselves (Blumenthal 1993, Quashnock 1996, Tegmark 1996), and with nearby cosmological structures (Hartmann 1996, Nemiroff 1993) have so far failed. These non-detections can be used to place lower limits on the “effective" distance to the GRB population. Using GRBs from the 1B Catalog, Blumenthal (1993) derive a lower limit of $z\sim 0.05$. If typical faint GRBs were any closer one would detect a positive angular correlation signal, since a smaller effective distance means less smearing of the clustering signal due to projection effects. Based on the data from the 3B Catalog, Quashnock (1996) derives a minimum sampling distance of $z=0.25$, at 95% confidence level. Based on the lack of any Supergalactic anisotropies, Hartmann (1996) derive a minimum sampling distance of $z\sim 0.065$.
All auto- and cross- correlation analyses with the BATSE Catalog are restricted to angles greater than $\sim 5^\circ$. Smaller separations cannot be probed reliably because of the low surface density of the bursts (roughly 1 GRB per 40 square degrees in the 3B Catalog), and more importantly, because of the large positional errors in the burst locations. Positions of individual GRBs in the BATSE Catalog suffer from random and systematic errors. The random errors range from 0.5 to 15 degrees depending on the flux of the GRB, while the systematic errors are about 4 degrees (Graziani & Lamb 1996).
In this paper we investigate the effect of weak gravitational lensing on the distribution of GRBs. If the GRBs are standard candles and trace the luminous matter, should we expect to see weak lensing effects in the angular distribution of bursts of all fluxes?
To show how weak lensing could affect the observed distribution of GRBs one can explore the following toy model. Imagine a positive density inhomogeneity (a clump of galaxies) located somewhere between us and the typical faint GRBs. The clump itself will produce an increase in the number of GRBs seen in that direction. It will also act as a lens for the background matter. The effects of lensing can be calculated in a straightforward fashion. Due to flux magnification, the maximum redshift of observed GRBs will be extended, and the surface number density of background GRBs will be diluted; therefore, that direction in the sky will have an excess of bursts due to the galaxy clump and lensing flux magnification, as well as a deficit due to lensing area distortion. How do the strengths of these effects compare? In Section 2 we will see that lensing can produce changes in the distribution of GRBs at the level of several percent. Due to projection effects, however, the signal will be smeared out and is not expected to be seen in the current data sets.
Before proceeding, we would like to note that the influence of strong lensing on GRBs has been considered in the literature, see Paczyński (1986, 1987). In general, if GRBs extend to $z\sim 2$, the fraction of strongly lensed bursts should be comparable to that of QSOs, $\sim4$ in every 500 GRBs should be multiply imaged (see Maoz & Rix 1993). Therefore there could be several multiply imaged GRBs in the current BATSE Catalog.
The Model
=========
Assumptions about the GRB population
------------------------------------
Throughout this paper we assume that GRBs are cosmological in origin, all the bursts have identical spectral properties, and identical luminosities. We assume that the luminosity and number density of GRBs do not evolve with redshift. There is no convincing physical reason for such simplifications, except that a model based on these very simple assumptions correctly predicts the log$N$-log$f$ distribution of sources over 3 decades in flux (Mao & Paczyński 1992).
In particular, bright bursts have a cumulative number counts distribution with slope of $-{3\over 2}$, expected from a uniform spatial distribution of nonevolving sources in flat space. In addition, the well documented roll-off in the observed number counts of GRBs at faint fluxes is competely consistent with the geometry of an expanding Universe. The minimum distance to the faintest GRBs, derived using best fits to the log$N$-log$f$ curves, is $z\simgt 1$ (Wickramasinghe 1993, Mao & Paczyński 1992). These distances are in agreement with the lower limits derived using auto- and cross- correlation analyses (see Introduction). One must remember, however, that the assumptions of no-evolution and single intrinsic burst luminosity are not necessary, and may not be sufficient to describe the current GRB data.
We assume, following Mao and Paczyński (1992), that the spectral slope of the GRBs is $a=1$, such that $L_{\nu}d\nu \propto \nu^{a-1} d\nu$, and the flux from a burst at $z$ is given by $$f\propto (1+z)^a/[4\pi D_L(z)^2],\eqno(1)$$ where $D_L$ is the luminosity distance. Spectral slope is important for K-corrections and determination of observed fluxes, since the frequency range of the BATSE detector corresponds to different energy bands depending on the source redshift. Since the bursts are assumed to be standard candles, the maximum redshift of observed GRBs, $z_{max}$, is directly related to the faintest detectable flux, $f_{min}$, through equation (1).
Cosmological time dilation affects GRBs in two ways: it changes the durations of individual bursts, as well as the rate of events seen by the observer. It is interesting to point out that the time dilation of event durations may have already been detected (Norris 1994, Norris 1996, but see Brainerd 1994). In the present work time dilation of individual events will not be important. Only the reduction of GRB rate with redshift is relevant for us here.
We assume that the rate of production of GRBs is proportional to the density of luminous matter. Also, the mass distribution on large scales is traced by the light distribution. This assertion is consistent with the observations of the large scale flows; the mass density, reconstructed from the peculiar velocity field of the galaxies, agrees reasonably well with the galaxy distribution (Dekel 1994).
We consider three sets of cosmological parameters, ($\Omega$, $\Lambda$)=(1, 0), (0.2, 0), and (0.2, 0.8), and a Hubble constant of 75 km s$^{-1}$ Mpc$^{-1}$ where appropriate.
Effects of weak lensing
-----------------------
It is instructive to study lensing effects due to a single isolated density inhomogeneity. Imagine looking at a patch of the sky of angular diameter $\theta$. GRBs in that patch will originate from a range of redshifts, or from a conical ‘tube’ with its apex at the observer and extending up to $z_{max}$. Let the mass distribution within the tube be completely smooth. The rate of GRBs (of all fluxes) from that tube, as seen by an observer, is proportional to $F_u$, where subscript $u$ stands for ‘unlensed’. Let us now place a mass concentration (a ‘clump’) at a redshift $z_l$ within this tube, centered on the tube’s axis. Let the clump contribute an additional rate of GRBs proportional to $F_c$. The clump will weakly lense the background sources, thereby altering the rate of GRBs originating in the tube. The altered rate of GRBs from the tube alone is proportional to $F_l$, where subscript $l$ stands for ‘lensed’.
The average, background subtracted, surface mass density of the clump, $\Sigma_c(r_c)$, is assumed to be proportional to $r_c^{-\beta}$, within radius $r_c$. This relation can be normalized such that $\Sigma_c r_c^\beta=C$, a constant. A reasonable value for $\beta$ is probably around 1, corresponding to a singular isothermal sphere, or 0.8, which is the slope of the observed angular correlation function of galaxies. The exact values of $\beta$ and $C$ are not important, as will be shown later (see equation \[7\]).
Since the angular scales that can be probed with the GRB population are of the order of a few degrees or more, the cosmological structures defined by such scales are correspondingly large in extent and represent small deviations from the average mass density of the Universe. To illustrate this, consider an angular scale of 10 degrees, about the smallest angular scale that can be reliably sampled with the BATSE data, given both systematic and random positional errors. At a nearby distance of $z\sim 0.025$, $10^\circ$ corresponds to $\sim 13h^{-1}$Mpc. On this scale, the rms scatter in $\delta \rho / \rho$ is less than one. Therefore, all the mass clumps considered here have low surface mass densities, and are all sub-critical in the lensing sense, $\Sigma_c \ll \Sigma_{crit}$. Here, $\Sigma_{crit}={{c^2}\over{4\pi G}}{{D_{os}}\over{D_{ol}D_{ls}}}$ is the critical surface mass density for lensing, and $D$’s are the angular diameter distances between the source, lens, and observer. Strong lensing is unimportant on such scales. Moreover, weak lensing is quite weak indeed; therefore it is sufficient to approximate magnification $A$ of a lens with $\kappa=\Sigma_c/\Sigma_{crit}$, by $A^{-1}\approx(1-\kappa)^2\approx(1-2\kappa)$.
We will now evaluate the expected rates of GRBs, $F_c$, $F_u$, and $F_l$. Assuming the production of GRBs traces luminous matter, the clump will generate GRBs at a rate proportional to $M_c/(M/L)_c$, where $M_c$ and $(M/L)_c$ are its mass and mass to light ratio. The rate of bursts is then given by, $$F_c(z_l)\propto {1\over{(M/L)_c}}\Sigma_c r_c^2 (1+z_l)^2 (1+z_l)^{-1}=
{1\over{(M/L)_c}}C r_c^{2-\beta} (1+z_l) .\eqno(2)$$ The time dilation term, $(1+z_l)^{-1}$, is due to the reduced rate of GRBs originating at cosmological redshifts. The physical surface density of a density inhomogeneity increases as $(1+z_l)^2$. Later, we will need to express $r_c$ in terms of its angular size, and distance, $r_c\approx \theta~D(z_l)$. The small angle approximation, assumed here, is valid for angles up to a few tens of degrees. The angular correlation function of GRBs on scales larger than that is probably not relevant. Thus, small angle approximation used here, and also in equations (3)-(5) below, is valid for our purposes.
Similarly, in the tube, the rate of bursts is also proportional to the density of luminous matter, ${(M/L)_U}^{-1} \int \rho_U(z) dV_{tube}(z)$. Here, $\rho_U(z)$ and $(M/L)_U$ are the average density and mass to light ratio of the Universe, and $V_{tube}$ is the volume of the tube. The rate of GRBs from the tube is given by, $$F_u(z_{max})\propto{{\rho}\over{(M/L)_U}}\int_0^{z_{max}}
[\theta D(z)]^2{{cdt}\over{dz}}(1+z)^3 (1+z)^{-1} dz,\eqno(3)$$ where $D(z)$ is the angular diameter distance, and $\rho=\rho_U$($z$=0) is the present day average mass density of the Universe.
Weak lensing due to a density inhomogeneity will have two effects on the observed sources: (1) it will extend the redshift of the visible GRBs beyond $z_{max}$, to $z_{max}^\prime$; and (2) it will dilute the surface density of the bursts located beyond $z_l$. Thus, including lensing effects, the rate of GRBs from the tube is given by, $$F_l(z_l,z_{max}^\prime)\propto{{\rho}\over{(M/L)_U}}\Biggl[\int_0^{z_l}
[\theta D(z)]^2{{cdt}\over{dz}}(1+z)^2 dz
+\int_{z_l}^{z_{max}^\prime}
[\theta D(z)]^2{{cdt}\over{dz}}(1+z)^2 {1\over{A(z_l,z)}} dz\Biggr].
\eqno(4)$$ Average magnification due to the clump at $z_l$ of sources at $z>z_l$ is $$A^{-1}(z_l,z)\approx1-2{{\Sigma_c}\over{\Sigma_{crit}}}=
1-{{2C(1+z_l)^2[\theta D(z)]^{-\beta}}\over{\Sigma_{crit}}}.\eqno(5)$$
Let us estimate $z_{max}^\prime$, the redshift of the faintest observable GRBs if lensing magnification is included. The minimum observable (unlensed) flux is reduced to $f_{min}^\prime=f_{min} A^{-1}(z_l,z_{max})\approx f (1-2\kappa)$. The fractional change in flux is then $df/f\approx
-2\kappa$, and since ${{df}\over f}={{df}\over{dz}}{{1}\over f}dz$, the redshift change is $$dz\approx 2\kappa {{(w^{0.5}-1)}\over{(a-2)w^{-0.5}-(a-1)w^{-1}}}
=x(w)\kappa,\eqno(6)$$ where $w=1+z_{max}$, and ($\Omega$, $\Lambda$)=(1, 0) cosmology is assumed. With weak lensing, maximum redshift of observed bursts is $z_{max}^\prime=z_{max}+dz$. For $z_{max}\sim 1$, and $a$ between 0.5 and 1.5, $x(w)$ is of the order of 1.
Now let us examine the effects of weak lensing due a single clump at $z_l$, on the observed rate of GRBs seen in a given direction. An interesting quantity to consider is the ratio of the change in the number of bursts due to lensing to the number of bursts due to the clump itself, $R(z_l)={{(F_l-F_u)}\over{F_c}}$. Making use of equations (2)-(6) we get, $$R(z_l)\approx\rho{{(M/L)_c}\over{(M/L)_U}}(1+z_l){1\over{D(z_l)^2}}
\Biggl[x(w){{D(z_{max})^2(1+z_{max})^2}\over{\Sigma_{crit}(z_l,z_{max})}}
{{cdt}\over{dz}}{\Bigg\vert}_{z_{max}}{\hskip 0.5in}$$ $${\hskip 3.0in}
-{\int_{z_l}^{z_{max}}{{D(z)^2(1+z)^2}\over{\Sigma_{crit}(z_l,z)}}
{{cdt}\over{dz}}dz}\Biggr].\eqno(7)$$ The two terms in equation (7) represent additional sources due to fainter observable fluxes, and the deficit in the source number density due to the dilution effects of lensing, respectively. Ratio $R(z_l)$ can be either positive or negative depending on which of the two terms in equation (7) dominates. In general, area dilution term dominates if the column of matter between the lens and the edge of the source population is large.
$R(z_l)$ is independent of $\beta$, $\theta$, and $C$, as long as lensing is weak. In fact, it only depends on cosmology through $D$’s, $\rho$, and $(M/L)$’s, and weakly, on $a$, the spectral index of GRBs.
Figure 1 shows $R(z_l)$ as a function of log $z_l$, for ($\Omega$, $\Lambda$)=(1, 0). There are five solid curves; from the top they are for $z_{max}=$0.5, 0.2, 0.7, 1.0, and 1.5. These curves assume ${(M/L)_c}={(M/L)_U}$, the density inhomogeneities on relevant scales, $>10h^{-1}$Mpc, have a biasing factor $b={{(M/L)_U}\over{(M/L)_c}}=1$. The long-dash and short-dash curves are for $z_{max}=$1.0, and $b=2$ and $b=0.5$, respectively. Since $R(z_l)$ is simply proportional to ${{(M/L)_c}\over{(M/L)_U}}=b^{-1}$, the rest of the $z_{max}$ cases with $b\ne1 $ are not shown on the plot to avoid crowding. The actual value of $b$ is probably somewhere between 0.5 and 2, and depends on the type of galaxies that are used to delineate the density inhomogeneities. For example, the biasing factor for IRAS galaxies is probably around 0.7, while optical galaxies have a larger $b$ (see Dekel 1993, Strauss 1992).
Notice that $R(z_l)$ changes very rapidly as redshift $z_l$ decreases (Figure 1 is a log-linear plot). In fact, the analytical expression in equation (7) diverges as $1\over {z_l}$, as $z_l$ approaches 0. Physically, the quantity $R(z_l)$ loses its meaning as $z_l$ gets very small—since the expected number of bursts is well below 1.
Figure 1 implies that weak lensing by the nearby cosmological structures has a non-negligible effect ($\vert R\vert $ of the order of 1) on the distribution of GRBs of all fluxes. For example, if $z_{max}=1$, the excess of bright bursts due to the cosmological structures at $z_l\sim 0.015$ (log $z_l\sim -1.82$) will be almost exactly offset by the deficit of fainter bursts. In other words, if GRBs of all fluxes are considered, there will be no net change of the burst rate due to structures at $z_l\sim 0.015$. Of course, with the present numbers of bursts in the BATSE Catalog ($\sim 1000$), and the fraction of bursts originating at $z_l\simlt 0.015$ ($\sim 1.6\cdot 10^{-5}$), one would not expect to see either these structures, or their lensing effects.
To gauge the overall importance of weak lensing, one needs to consider lenses at all redshifts up to $z_{max}$. Since $R(z_l)$ is independent of the particulars of cosmological density inhomogeneities, a rough estimate of the average value of $R(z_l)$ is its volume weighted average, $$\langle R\rangle=\int_0^{z_{max}} R(z_l){{dV}\over{dz}}dz {\Bigg/}
\int_0^{z_{max}} {{dV}\over{dz}}dz,\eqno(8)$$ where $dV$ is the comoving volume element.
Figure 2 shows the dependence of $\langle R\rangle$, expressed as a percentage, on the effective depth of the observed GRB population, $z_{max}$, for three different sets of cosmological parameters: ($\Omega$, $\Lambda$)=(0.2, 0.8), (0.2, 0), and (1, 0). As expected, a $\Lambda$ dominated Universe shows the largest change due to lensing.
Volume weighted $\langle R\rangle$ is strongly dominated by the positive $R$’s at $z_l\simgt 0.1$, for all values of $z_{max}$ and all sets of ($\Omega$, $\Lambda$). This means that weak lensing will enhance the contrast of individual density inhomogeneities by $\langle R\rangle$. Therefore, the observed angular correlation function of [*all*]{} bursts will be increased by $\langle R\rangle$ on [*all*]{} angular scales. Given that the present day limits on $\omega(\theta)$ of GRBs are rather weak, an increase of $\simlt 35\%$ would not be detected, especially if the observed GRBs extend to large redshifts, $z_{max}\simgt 0.5$.
Conclusions and Discussion
==========================
In this paper we examined the effect of weak lensing on the angular distribution of GRBs. Since the nature of and distance to GRBs is unknown, a set of assumptions about the GRB population had to be adopted (see Section 2.1). These assumptions are common to most models of GRBs currently discussed in the literature. In particular, all the bursts are considered to be at cosmological distances, and associated with luminous matter. Another assumption crucial to the current discussion is that GRBs are standard candles, or at least have a narrow intrinsic luminosity function. The main conclusions are as follows.
The dominant effect of weak lensing is to extend the faint flux cutoff of the observed bursts. This wins over the area dilution effects of lensing, and hence leads to an enhancement of the density contrast of cosmological structures, as seen in the distribution of bursts of all fluxes. The amplitude of the enhancement depends on the depth of the present GRB Catalogs, the cosmological model adopted, and the biasing factor, $b$ on scales $>~10h^{-1}$ Mpc. The amplitude is insensitive to the mass profile of density inhomogeneities, $\beta$, their absolute mass density normalization, $C$, and the angular scale of observations, $\theta$. It is slightly sensitive to the assumed spectral index of GRBs, $a$.
The overall importance of weak lensing is summarized in Figure 2. $\langle R\rangle$ can be roughly interpreted as the change in the amplitude of the angular correlation function of GRBs. For a canonical set of assumptions, $z_{max}=1$ and $(\Omega$, $\Lambda)$=(1, 0), $\omega(\theta)$ is expected to be increased by $\sim 8\%$ on all angular scales. Such a change will not be detected given the present numbers of cataloged bursts.
Recently, it was suggested by Lamb and Quashnock (1993) that if GRBs trace luminous matter, their distribution on $\simgt~1^\circ$ scales can be used to probe the large scale structure in the Universe. The authors estimate that if $\simgt 3000$ GRBs are observed, these would provide the power spectrum of density fluctuations on scales currently probed by pencil beam surveys and superclusters. Density fluctuations on such scales are probably primordial in nature and may be compared to those implied by the COBE observations of microwave background. If, after a few more years of BATSE observations, the accumulated number of bursts makes such an investigation possible, then weak lensing effects should be taken into account.
Since the weak lensing scenario described in this paper seems to be rather general, one may wonder if it applies to populations of other extragalactic objects seen in projection, for example, galaxies. The answer is no. Here, GRBs are treated as standard candles. Galaxies, on the other hand, have a broad distribution of intrinsic luminosities. In fact, a typical $d$log $N(m)$/$dm$ slope for galaxies, 0.4 (Jarvis & Tyson 1981), leads to no net change in the surface number density of a lensed population, hence the angular correlation function of galaxies will not be affected by weak lensing in this case. (For an extended discussion of the effect of weak lensing on the observed angular clustering of galaxies see Villumsen 1995.)
I would like to acknowledge the support of the PPARC fellowship at the Institute of Astronomy, Cambridge, UK. I am also grateful to the anonymous referee for useful comments on the paper.
Blumenthal, G. R., Hartmann, D. H., & Linder, E. V. 1993, In GRB Second Workshop, AIP Conf. Proc. 307 , eds. G. J. Fishman, J. J. Brainerd, & K. Hurley, p. 117
Brainerd, J. J. 1994, ApJ, 428, L1
Dekel, A. 1994, ARA&A, 32, 371
Dekel, A., Bertschinger, E., Yahil, A., Strauss, M. A., Davis, M., & Huchra, J. P. 1993, ApJ, 412, 1
Graziani, C., & Lamb, D. Q. 1996, astro-ph/9601143
Hartmann, D. H., Briggs, M. S., & Mannheim, K. 1996, astro-ph/9601057
Irwin, M., & Żytkow, A. N. 1994, ApJ, 433, L81
Jarvis, J. F., & Tyson, A. T. 1981, AJ, 86, 476
Lamb, D., Q. 1995, PASP, 107, 1152
Lamb, D. Q., & Quashnock, J. M. 1993, ApJ 415, L1
Mao, S., & Paczyński, B. 1992, ApJ, 388, L45
Maoz, D., & Rix, H.-W. 1993, ApJ, 416, 425
Nemiroff, R. J. et al. 1993, In GRB Second Workshop, AIP Conf. Proc. 307, eds. G. J. Fishman, J. J. Brainerd, & K. Hurley, p. 137
Norris, J. P., Nemiroff, R. J., Bonnell, J. T., & Scargle, J. D. 1996, astro-ph/9601178
Norris, J. P. et al. 1994. ApJ, 424, 540
Paczyński, B. 1987, ApJ, 317, L55
Paczyński, B. 1986, ApJ, 308, L43
Palmer, D. M., et al. 1993, In GRB Second Workshop, AIP Conf. Proc. 307, eds. G. J. Fishman, J. J. Brainerd, & K. Hurley, p. 247
Quashnock, J. M. 1996, astro-ph/9512119
Shaefer, B. E. 1993, In GRB Second Workshop, AIP Conf. Proc. 307, eds. G. J. Fishman, J. J. Brainerd, & K. Hurley, p. 382
Shaefer, B. E. 1990, In GRB—Observations, Analyses, and Theories, Los Alamos Workshop on GRBs, eds. C. Ho, R. I. Epstein, & E. E. Fenimore, p. 107
Strauss, M. A., Davis, M., Yahil, A., & Huchra, J. P. 1992, ApJ, 385, 421
Tegmark, M., Hartmann, D., Briggs, M., & Meegan, C. 1996, astro-ph/9602039
Villumsen, J. V. 1995, astro-ph/9512001, submitted to MNRAS
Wickramasinghe, W. A. D. T. et al. 1993, ApJ, 411, L55
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Andrew Phillip Davies
bibliography:
- 'allrefs.bib'
title: Cocycle twists of algebras
---
Let $A$ be a $k$-algebra where $k$ is an algebraically closed field and $G$ be a finite abelian group whose order is not divisible by the characteristic of $k$. If $G$ acts on $A$ by $k$-algebra automorphisms then the action induces a $G$-grading on $A$ which, in conjunction with a normalised 2-cocycle of the group, can be used to twist the multiplication of the algebra. Such twists can be formulated as Zhang twists [@zhang1998twisted] as well as in the language of Hopf algebras, as done in [@montgomery1993hopf §7.5].
We investigate such cocycle twists with an emphasis on the situation where $A$ also possesses a connected graded structure and the action of $G$ respects this grading. The twisting operation uses the induced $G$-grading of $A$ rather than its [$\mathbb{N}$]{}-grading, as is often the case with Zhang twists. As a result, geometric data encoded by special modules in the category of graded right modules, $\text{grmod}(A)$, may not be preserved. Nevertheless, we show using an alternative construction of the twist and faithful flatness arguments that many properties are preserved; the strongly noetherian property, finite global dimension and Artin-Shelter regularity for example.
The above concepts are then illustrated by applying cocycle twists to the 4-dimensional Sklyanin algebras, $A:=A(\alpha,\beta,\gamma)$, first studied ring-theoretically in [@smith1992regularity]. Such algebras are examples of noncommutative projective surfaces and display many interesting geometric properties. Much of this geometry is controlled by a factor ring $B:=B(E,{\ensuremath{\mathcal{L}}},\sigma)$, which is a twisted homogeneous coordinate ring.
We define an action of the Klein four-group $G=(C_2)^2$ on $A$ such that the action restricts to the factor ring $B$. The cocycle twists under the induced $G$-grading, namely $A^{G,\mu}$ and $B^{G,\mu}$ respectively, have very different geometric properties to their untwisted counterparts. While $A$ has point modules parameterised by an elliptic curve $E$ and four extra points, $A^{G,\mu}$ has only 20 point modules when the automorphism $\sigma$ has infinite order. The point modules over $A$ can be used to construct fat point modules of multiplicity 2 over $A^{G,\mu}$, and there are isomorphisms among such objects corresponding to orbits of a natural action of $G$ on $E$. This is just one example of the interplay between the 1-critical modules over $A$ and $A^{G,\mu}$.
The ring $B^{G,\mu}$ falls under the purview of Artin and Stafford’s classification of noncommutative curves [@artin2000semiprime], and we describe it in geometric terms. It can be expressed as a more general twisted homogeneous coordinate ring, where the key object is a sheaf of orders over ${\ensuremath{\mathcal{O}}}_{E^{G}}$. In fact, the fat point modules over $A^{G,\mu}$ — which are also $B^{G,\mu}$-modules — can be explained by an equivalence of categories consequent to this work.
Other examples of twists are also studied, one of which relates to Rogalski and Zhang’s classification of AS-regular algebras of dimension 4 with three degree 1 generators and an additional ${\ensuremath{\mathbb{Z}}}^{2}$-grading [@rogalski2012regular]. Their work classified such algebras into 8 families up to isomorphism; we demonstrate that several of these families are related by cocycle twists.
I would first like to thank EPSRC for their generous grant enabling me to study for a PhD, since undoubtedly I could not have afforded to do so otherwise.
My supervisor, Professor Toby Stafford, has been a superb guide and mentor throughout the last few years; I have certainly come to understand why his former students hold him in such high esteem. Whenever I may have been feeling frustrated with maths, a meeting with Toby would invariably result in me leaving full of his contagious enthusiasm.
I would also like to acknowledge the fellow PhD students I have known at Manchester, particularly those with whom I have shared an office. Their willingness to discuss maths, other ‘enlightening’ conversations and numerous distractions have all made my time as a PhD student thoroughly enjoyable.
Finally, I would like to thank my family and my girlfriend Kirsti – jag älskar dig. I’m very lucky to have you, and I preemptively forgive you if you give up reading after a couple of pages/paragraphs/words…
Introduction {#chap: introduction}
============
Overview {#sec: introoverview}
--------
In this thesis we study a twisting operation on algebras. This operation is closely related to the classical notion of using a 2-cocycle to deform the multiplication in a group-graded algebra. Our primary concern is to establish whether certain properties are preserved under the twisting operation, the first results in this direction being proved in [@zhang1998twisted] and [@montgomery2005algebra]. In the former the twists we study are phrased as Zhang twists, while Montgomery’s results in the latter hold for generalised twists in the setting of Hopf algebras.
We use another construction of these *cocycle twists* that was independently described in [@bazlov2012cocycle]. This construction gives one much greater traction when trying to prove that properties are preserved under twisting, since it enables the use of faithful flatness arguments. Our main results are stated later in this introduction, namely in Proposition \[prop: twoconstrsaresame\] and Theorems \[thm: thebigone\], \[thm: sklyanin\] and \[thm: geometricthcrdesc\].
The motivation for our work was an example of Odesskii, appearing in the survey paper [@odesskii2002elliptic] on *elliptic algebras*. Although the example as we state it does not use a specific type of algebra, the original example in [@odesskii2002elliptic] used a *4-dimensional Sklyanin algebra*. In Chapter \[chap: sklyanin\] we will study cocycle twists of such Sklyanin algebras, although they will also be discussed briefly in §\[subsec: examples\] (see Definition \[def: sklyanin relations\] for a presentation by generators and relations).
In order to state Odesskii’s example, we need to make the following assumptions. Let $k= {\ensuremath{\mathbb{C}}}$ and consider an associative $k$-algebra $A$ that is generated as a $k$-algebra by $x_0,x_1,x_2$ and $x_3$. We will assume that the Klein four-group $G=(C_2)^2= \langle g_1, g_2 \rangle$ acts on $A$ by algebra automorphisms, i.e. $G \rightarrow
\text{Aut}_{\text{alg}}(A)$. Denoting the action of $g \in G$ by the superscript $-^g$, the action is defined on generators by $$\label{eq: odesskiiaction}
x_i^{g_{1}} = x_{i+2},\;\;\; x_i^{g_{2}} = (-1)^i x_{i},$$ where $i \in \{0,1,2,3\}$ and indices are taken modulo 4.
There is an action of $G$ by $k$-algebra automorphisms on $M_2(k)$, the ring of $2 \times 2$ matrices over $k$. This action is defined by $$\label{eq: matrixaction}
M^{g_{1}} =\begin{pmatrix}
-1 & 0 \\
0 & 1
\end{pmatrix}M\begin{pmatrix}
-1 & 0 \\
0 & 1
\end{pmatrix},\; M^{g_{2}}=\begin{pmatrix}
0 & 1 \\
1 & 0
\end{pmatrix}M\begin{pmatrix}
0 & 1 \\
1 & 0
\end{pmatrix},$$ for a matrix $M \in M_2(k)$ and the group generators $g_1$ and $g_2$. A 2-cocycle arises here via an action of $G$ on $k^2$ – such objects are important for the constructions we will soon describe.
Now consider the tensor product of $k$-algebras $A \otimes_{k} M_2(k)$. Given the action of $G$ in and the embedding $G \rightarrow \text{Aut}_{\text{alg}}(A)$, there is a natural diagonal action of $G$ on the tensor product, where $$\label{eq: diagaction}
(a \otimes_k M)^g = a^g \otimes_k M^g,$$ for all $a \in A$ and $M \in M_2(k)$.
Odesskii’s construction concludes by taking the invariant ring of $A \otimes_{k} M_2(k)$ under this diagonal action. We highlight this algebra for future reference.
\[ex: odesskii\] Odesskii’s example is given by the invariant ring $(A \otimes_{k} M_2(k))^G$.
It is natural to ask if the properties of $A$ are shared by $(A \otimes_{k} M_2(k))^G$. Moreover, can the construction can be generalised to any algebra or group? Our attempts to answer these questions form the basis of the work in this thesis.
### Twisting theory {#subsec: theoryoftwist}
Before stating some of our main results we must define the constructions involved in our work. Let us make the following assumptions that will remain in place until the end of §\[subsec: theoryoftwist\]. Our base field $k$ will be assumed to be algebraically closed and $A$ will denote an associative $k$-algebra with identity. The finite abelian group $G$ will satisfy $\text{char}(k) \nmid |G|$ and will act on $A$ by $k$-algebra automorphisms. We will also fix a duality isomorphism between $G$ and its group of characters $G^{\vee}$, where the character to which an element $g \in G$ is mapped is denoted by $\chi_g$.
#### Construction 1 {#subsubsec: constr1}
Let us define the first of two twisting constructions that we will use. There are two key components of Construction 1; first, the inducement of a $G$-grading on $A$ via the action of $G$, and second, the notion of using a normalised 2-cocycle to twist the $G$-graded multiplication in $A$.
The definition of a *$G$-graded algebra* is given in Definition \[defn: ggradedalgebra\]. For now it suffices to state that for all $g\in G$ the corresponding component of the induced $G$-grading on $A$ is defined by $$\label{eq: inducedGgrading}
A_g := \{a \in A:\; a^h = \chi_{g^{-1}}(h)a,\; \forall \; h \in G \}.$$
Given such a grading, one can use a normalised 2-cocycle to deform the multiplicative structure of $A$.
\[def: normalised2cocycle\] Consider a function $\mu: G \times G \rightarrow k^{\times}$ satisfying the following relations for all $g,h,l \in G$ and the group identity element $e$: $$\label{eq: cocycleidinto}
\mu(g,h)\mu(gh,l)=\mu(g,hl)\mu(h,l), \;\;\; \mu(e,g)=\mu(g,e)=1.$$ We say that $\mu$ is a *normalised 2-cocycle* of $G$ with values in $k^{\times}$.
Although normalised 2-cocycles can be interpreted in terms of cohomology — as we do in §\[subsec: cocycletwists\] — for now we will use them solely to define a new multiplication on $A$. We will write $(A,\cdot)$ to denote the $G$-graded algebra $A$ and the standard multiplication $\cdot : A \times A \rightarrow A$, usually written as juxtaposition of elements.
One can define a new multiplication on the underlying $G$-graded vector space structure of $A$ as follows: for homogeneous elements $a \in A_g$ and $b \in A_h$ define $a \ast_{\mu} b:=\mu(g,h)ab$. This gives rise to the algebra $(A, \ast_{\mu})$. By virtue of $\mu$ satisfying this new algebra is associative and has the same algebra identity element as $(A,\cdot)$ (proved in Proposition \[prop: preserveassociative\]). We will use the notation $A^{G,\mu}:=(A, \ast_{\mu})$ and call such an object a cocycle twist of the induced $G$-grading on $A$ using the normalised 2-cocycle $\mu$. For brevity we may also use the term *cocycle twist* of $A$, since the notation makes clear both the group acting and the normalised 2-cocycle.
Perhaps the simplest nontrivial example of a cocycle twist is that of a twisted group algebra.
\[ex: twistedgroupalgebra\] Consider the group algebra $AG := A \otimes_k kG$, with $A$ and $G$ subject to the assumptions at the beginning of §\[subsec: theoryoftwist\]. The multiplication in this algebra is given by $$\label{eq: groupalgebramult}
(a \otimes_k g)(b \otimes_k h) = ab \otimes_k gh,$$ for all $a,b \in A$ and $g,h \in G$.
The group algebra has an obvious $G$-graded structure given by $A_g=\{a \otimes_k g : \; a \in A\}$. Twisting this grading using a normalised 2-cocycle $\mu$ produces a *twisted group algebra* $AG_{\mu}$. The multiplication in this algebra is defined by $$(a \otimes_k g) \ast_{\mu} (b \otimes_k h) =\mu(g,h)(ab \otimes_k gh),$$ for all $a, b \in A$ and $g, h \in G$.
Cocycle twists have been studied in many other places and generalised vastly – they can be described as Zhang twists for example (see Theorem \[thm: cocycleaszhang\]), which were studied in detail by Zhang in [@zhang1998twisted]. One can also recover them from the *twisted $H$-comodule algebra* construction of [@montgomery1993hopf §7.5] by using the Hopf algebra $H=kG$, the group algebra of a finite abelian group.
Let us exhibit some of the results already in the literature in relation to the preservation of properties under cocycle twisting. Although some of the results hold more generally, we state them in the setting of Construction 1.
\[prop: propspreservedalready\] Assume that $A$ and $G$ are subject to the assumptions at the beginning of §\[subsec: theoryoftwist\] and let $\mu$ be a normalised 2-cocycle. Then if $A$ has one of the following properties, so does $A^{G,\mu}$:
- it is finitely generated as a $k$-algebra [@montgomery2005algebra cf. Proposition 3.1(1)];
- it satisfies the polynomial identity (PI) property (cf. [@zhang1998twisted Proposition 5.6], [@montgomery2005algebra Proposition 3.1(4)]);
- it is noetherian (cf. [@zhang1998twisted Proposition 5.1], [@montgomery2005algebra Proposition 3.1(3)]).
One can replicate this construction without using an identification of $G$ and $G^{\vee}$ by instead using a 2-cocycle over $G^{\vee}$.
#### Construction 2 {#subsubsec: constr2}
We now move on to the second construction. One first defines an action of $G$ on the twisted group algebra $kG_{\mu}$ by $k$-algebra automorphisms. The algebra $kG_{\mu}$ can be obtained by taking $A=k$ in Example \[ex: twistedgroupalgebra\].
The action of $G$ on $kG_{\mu}$ is defined as follows: for all $g,h \in G$ and $\alpha \in k$, let $(\alpha \otimes_k
g)^h:=\chi_g(h) \alpha\otimes_k g$. By assumption one has an action of $G$ on $A$ by $k$-algebra automorphisms, thus one can define a diagonal action of $G$ on the tensor product $A \otimes_k kG_{\mu}$ by $$\label{eq: diagactionconstr2}
(a \otimes_k g)^h := \chi_g(h)(a^h \otimes_k g),$$ for all $a \in A$ and $g, h \in G$.
The construction is then completed by taking the invariant ring under this action, $(A \otimes_k kG_{\mu})^G$, which is isomorphic to $(AG_{\mu})^G$ as a $k$-algebra.
Note that for the Klein-four group $G$ and a particular normalised 2-cocycle $\mu$ one has $kG_{\mu} \cong M_2(k)$ as $k$-algebras by Lemma \[lem: kgmuiso\]. This provides some indication that Example \[ex: odesskii\] can be described using Construction 2. The demonstration of this is delayed until §\[subsec: odesskiiegtwist\].
#### Consequences {#subsubsec: consequences}
We now gather together a collection of our main results regarding Constructions 1 and 2, the first of which shows that they are in fact the same.
\[prop: twoconstrsaresame\] Assume that $A$ and $G$ satisfy the base assumptions of §\[subsec: theoryoftwist\]. Then for any normalised 2-cocycle $\mu$ one has an isomorphism of $k$-algebras $A^{G,\mu} \cong (AG_{\mu})^G$.
The equivalence of the two twisting constructions allows us to adopt the notation $A^{G,\mu}$ for a cocycle twist formed using *either* construction. Our main tool for proving the preservation of properties under twisting is provided by the following lemma; the twisted bimodule structures involved are defined prior to Proposition \[prop: fflat\].
\[lem: bimoddecomp\] Under the base assumptions of §\[subsec: theoryoftwist\], the twisted group algebra $AG_{\mu}$ has the following decomposition as an $(A^{G,\mu},A^{G,\mu})$-bimodule: $$AG_{\mu} \cong \bigoplus_{g \in G} {^{\text{id}}(A^{G,\mu})^{\phi_{g}}}.$$ Here $\phi_{g}$ is some $k$-algebra automorphism of $A^{G,\mu}$, with $\phi_e=\text{id}$. Each summand is free as a left and right $A^{G,\mu}$-module. Consequently, $AG_{\mu}$ is a faithfully flat extension of $A^{G,\mu}$ on both the left and the right. Furthermore, $AG_{\mu}$ is a faithfully flat extension of $A$ on both the left and the right too.
Figure \[fig: twists\] illustrates how we can use Proposition \[lem: bimoddecomp\] to adopt a new strategy when trying to show that properties are preserved under cocycle twists. Instead of the direct route from $A$ to $A^{G,\mu}$, we can attempt to push and pull properties along the faithfully flat extensions via the twisted group algebra $AG_{\mu}$.
(1.7,1.7) – (4,2.9); (7.2,1.8) – (5,2.9); (1.7,1.3) .. controls (3.45,0.8) and (5.45,0.8) .. (7.2,1.3);
at (4.5,3) [$AG_{\mu}$]{}; at (1.5,1.5) [$A$]{}; at (7.5,1.5) [$A^{G,\mu}\cong(AG_{\mu})^G$]{};
at (2.5,2.5) ; at (6.5,2.5) ; at (4.5,0.7) ;
Using the new technique highlighted by Figure \[fig: twists\] allows us to prove the following main theorem concerning the preservation of properties under cocycle twists. It brings together Lemma \[lem: hilbseries\], Corollaries \[cor: uninoeth\] and \[cor: asreg\], and Propositions \[prop: koszul\] and \[prop: cohenmac\]. Connected graded algebras and Hilbert series are defined in Definition \[def: conngraded\] and Definition \[def: hilbseries\] respectively. Definitions of the remaining properties involved are deferred until their appearance in §\[sec: preservation\].
\[thm: thebigone\] Further to the base assumptions of §\[subsec: theoryoftwist\], assume that $A$ is connected graded and the action of $G$ on $A$ by $k$-algebra automorphisms preserves this grading. Then for any normalised 2-cocycle $\mu$, $A$ has one of the following properties if and only if $A^{G,\mu}$ does:
- it is strongly noetherian;
- it is AS-regular of global dimension $n$ for some $n \in {\ensuremath{\mathbb{N}}}$;
- it is Koszul;
- it is Auslander regular;
- it is Cohen-Macaulay.
Furthermore, $A$ and $A^{G,\mu}$ have the same Hilbert series.
### Examples of twists {#subsec: examples}
Having described some of our results regarding properties preserved under cocycle twists, let us now turn to examples. We will be primarily concerned with twisting AS-regular algebras of global dimension 4, whose classification is an ongoing and central project in noncommutative algebraic geometry. This term will refer throughout to the school that was created in the wake of the seminal papers of Artin, Tate and Van den Bergh in the early 1990’s [@artin1990some; @artin1991modules].
The algebras that will form the main focus of this thesis are the *4-dimensional Sklyanin algebras*, whose twists are studied in Chapter \[chap: sklyanin\]. Such algebras were studied in [@smith1992regularity; @smith1993irreducible; @levasseur1993modules], not to mention [@tate1996homological], where they are viewed as a special case of a more general construction.
In the following definition we use the notation $[x,y] := xy-yx$ and $[x,y]_{+} :=
xy+yx$ to denote certain types of commutator in an algebra.
\[def: sklyanin relations\] Let $k$ be an algebraically closed field with $\text{char}(k)\neq 2$ and $\alpha, \beta, \gamma \in k$ be scalars satisfying $$\label{eq: 4sklyanincoeffcondintro}
\alpha + \beta + \gamma + \alpha \beta \gamma = 0 \;\text{ and }\; \{\alpha,\beta, \gamma\} \cap \{0,\pm 1\}=\emptyset.$$ We define the *4-dimensional Sklyanin algebra* $A(\alpha,\beta,\gamma)$ associated to such parameters to be the quotient of the free $k$-algebra $k\{x_0,x_1,x_2,x_3\}$ by the ideal whose generators are the six quadratic relations in [@smith1992regularity Equation 0.2.2]: $$\label{eq: 4sklyaninrelnsintro}
\begin{array}{llll}
[x_0,x_1]-\alpha[x_2,x_3]_+, &&& [x_0,x_1]_+ -[x_2,x_3], \\ \relax
[x_0,x_2]-\beta[x_3,x_1]_+, &&& [x_0,x_2]_+ -[x_3,x_1], \\ \relax
[x_0,x_3]-\gamma[x_1,x_2]_+, &&& [x_0,x_3]_+ -[x_1,x_2].
\end{array}$$
The latter condition of allows us to avoid certain degenerate cases (see [@smith1992regularity §1]). We will omit the parameters of $A(\alpha,\beta,\gamma)$ for the remainder of §\[subsec: examples\].
The construction of $A$ can also be phrased in terms of a smooth elliptic curve $E$ and the translation automorphism $\sigma$ associated to a point on the curve (as discussed in detail in [@smith1992regularity §2.9-2.14]). By generalising this construction one can define $n$-dimensional Sklyanin algebras for all integers $n \geq 3$. Indeed, it is this family which was studied in [@tate1996homological]. The 4-dimensional Sklyanin algebra has many good properties, including being AS-regular of global dimension 4 and having Hilbert series $1/(1-t)^4$ by [@smith1992regularity Theorem 0.3].
There are many actions of the Klein-four group — which we denote by $G$ for the remainder of §\[subsec: examples\] — on $A$ by [$\mathbb{N}$]{}-graded algebra automorphisms. In Chapter \[chap: sklyanin\] we study a cocycle twist $A^{G,\mu}$ for a particular action of $G$ and a normalised 2-cocycle $\mu$. Some of the more surprising properties of $A^{G,\mu}$ are grouped together in parts (ii) and (iii) of the following result; it summarises Proposition \[claim: fatpoints\] and Theorems \[thm: 4sklytwistprops\] and \[prop: finitepointscheme\].
\[thm: sklyanin\] Let $E$ and $\sigma$ be the elliptic curve and automorphism associated to a 4-dimensional Sklyanin algebra $A$. There is an action of the Klein-four group $G$ on $A$ by [$\mathbb{N}$]{}-graded algebra automorphisms which, together with a normalised 2-cocycle $\mu$, produces a cocycle twist $A^{G,\mu}$ with the following properties:
- it is strongly noetherian AS-regular domain of global dimension 4, which has Hilbert series $1/(1-t)^4$;
- when $|\sigma| = \infty$ its point scheme consists of only 20 points (see Definition \[def: pointscheme\]);
- it has a family of fat point modules of multiplicity 2 parameterised by $E^G$, an elliptic curve whose underlying structure comes from the orbit space of $E$ under a natural action of $G$.
The behaviour exhibited in Theorem \[thm: sklyanin\] (ii) is particularly surprising given that the point scheme of $A$ consists of the associated elliptic curve $E$ and four extra points [@smith1992regularity Propositions 2.4 and 2.5].
The presence of the modules occurring in Theorem \[thm: sklyanin\] (iii) can be related to a factor ring of $A$. There exist two central elements $\Omega_1,\Omega_2 \in A_2$ by [@smith1992regularity Corollary 3.9]. The factor ring $A/(\Omega_1,\Omega_2)$ is isomorphic to the twisted homogeneous coordinate ring $B:=B(E,{\ensuremath{\mathcal{L}}},\sigma)$, with such rings being defined in Definition \[def: thcr\]. One can use the Noncommutative Serre’s theorem (Theorem \[thm: noncomserrethm\]) to see that the point modules over $A$ that are parameterised by $E$ correspond geometrically to skyscraper sheaves over the elliptic curve.
Since $\Omega_1$ and $\Omega_2$ are fixed by the action of $G$ on $A$ used in Theorem \[thm: sklyanin\], the action of $G$ is compatible with factoring out the ideal $(\Omega_1,\Omega_2)$. We prove the following result, which summarises Theorem \[thm: geomdescthcrtwist\] and Proposition \[prop: fatpointsincohE\]. Any undefined terms are defined in §\[subsec: geomdata\] or §\[subsec: geomdesc\].
\[thm: geometricthcrdesc\] Let $A^{G,\mu}$ be as in Theorem \[thm: sklyanin\]. There is a factor ring of $A^{G,\mu}$, labelled $B^{G,\mu}$, which is a cocycle twist of a factor ring of $A$. The ring $B^{G,\mu}$ satisfies the following properties:
- it can be described as a twisted ring over $\mathcal{E}$, where $\mathcal{E}$ is a sheaf of ${\ensuremath{\mathcal{O}}}_{E^G}$-orders;
- it has a family of fat point modules of multiplicity 2 parameterised by $E^G$, arising in a natural way from certain coherent sheaves of $\mathcal{E}$-modules.
The importance of this result will be stressed in §\[sec: noncommgeom\] in relation to canonical maps to twisted rings. Briefly, it illustrates that twisted rings can capture geometric properties of noncommutative algebras, in this case $A^{G,\mu}$.
Other interesting examples can be found by studying the algebras classified by Rogalski and Zhang in [@rogalski2012regular]. Such algebras are AS-regular of dimension 4 with three degree 1 generators and an additional ${\ensuremath{\mathbb{Z}}}^{\times 2}$ grading. We disregard algebras with these properties that are normal extensions of AS-regular algebras of dimension 3. The classification up to isomorphism of the remaining algebras consists of 8 families $\mathcal{A}-\mathcal{H}$. Theorem \[thm: rogzhang\] shows that one can relate several of these families using cocycle twists.
\[thm: rogzhang\] Consider Rogalski and Zhang’s classification in [@rogalski2012regular]. For the Klein-four group $G$ and a certain normalised 2-cocycle $\mu$, one has the following isomorphisms of $k$-algebras: $$\mathcal{A}(1,-1)^{G,\mu}\cong \mathcal{D}(1,1),\;\; \mathcal{B}(1)^{G,\mu} \cong \mathcal{C}(1), \;\;
\mathcal{E}(1,\gamma)^{G,\mu}\cong \mathcal{E}(1,-\gamma), \;\; \mathcal{G}(1,\gamma)^{G,\mu} \cong
\mathcal{G}(1,\overline{\gamma}).$$
### Contents {#subsec: contents}
We now give a brief description of the contents of this thesis. The remainder of the current chapter consists of §\[sec: notation\], where our basic conventions and notation are given.
In Chapter \[chap: background\] we describe some of the background material that is needed. We begin with §\[sec: noncommgeom\], which consists of an introduction to noncommutative algebraic geometry emphasising the aspects that are most pertinent to cocycle twists. This is followed by §\[sec: twistsofalgebras\], in which we study several related concepts involving the twisting of algebras; cocycle twists, Zhang twists and crossed products are all defined. The chapter ends with the short section §\[sec: goldietheory\], which describes the Goldie theory that we will use.
The next chapter is the first containing original work. In §\[sec: construction\] two twisting constructions are defined and then subsequently shown to be the same in Proposition \[prop: twoconstrequal\]. The preservation of properties under such twists is then treated in §\[sec: preservation\]. A result of particular note is Corollary \[cor: asreg\], where the property of being AS-regular is shown to be preserved under cocycle twists. In §\[sec: modules\] we study the relationship between point modules over an algebra and fat point modules over a cocycle twist of it. The machinery we construct is used in several of the examples later in the thesis.
Chapter \[chap: sklyanin\] deals with the main examples of cocycle twists studied in the thesis, namely twists of 4-dimensional Sklyanin algebras, which we denote by $A^{G,\mu}$. In §\[subsec: 4sklyanintwistandptscheme\] such examples are shown to satisfy many good properties (in particular Theorem \[thm: 4sklytwistprops\]), while in Theorem \[prop: finitepointscheme\] their point schemes are determined. We also show that the full version of Odesskii’s example from [@odesskii2002elliptic Introduction] (which was simplified in Example \[ex: odesskii\]), can be described as a cocycle twist of a Sklyanin algebra (Proposition \[prop: odesskiiegdone\]). The existence of fat point modules over $A^{G,\mu}$ is proved in §\[subsec: fatpoints\]; such modules arise as direct sums of point modules over Sklyanin algebras by Proposition \[claim: fatpoints\]. This is just one indication of the interplay between 1-critical modules over $A$ and $A^{G,\mu}$, and we prove several more results in this direction.
In Chapter \[chap: thcrtwist\] we study a twist of the factor ring $B=A/(\Omega_1,\Omega_2)$. Under the induced $G$-grading on $A$, $B$ is a $G$-graded factor ring. Thus one can regard the twist $B^{G,\mu}$ as a factor of $A^{G,\mu}$, which was studied in the Chapter \[chap: sklyanin\]. The algebra $B^{G,\mu}$ is shown — under some conditions — to have trivial centre in Proposition \[prop: centrethcrtwist\]. From this proposition we derive Corollary \[cor: centretwistskly\], which describes the centre of $A^{G,\mu}$. In §\[subsubsec: geomdescrthcrtwist\] we describe $B^{G,\mu}$ in terms of the geometric classification in [@artin2000semiprime], concerning semiprime, noetherian rings of quadratic growth. This description is given in Theorem \[thm: geomdescthcrtwist\], after which we show that the twist is a prime ring (Corollary \[cor: actuallyprime\]). The fat point modules over $A^{G,\mu}$ and the fact that $B^{G,\mu}$ possesses no point modules can both be related to this geometric realisation (see Propositions \[prop: onlyfatpoints\] and \[prop: fatpointsincohE\]).
Chapter \[chap: othertwists\] considers cocycle twists of other algebras, beginning in §\[subsec: staffordalgs\] with cocycle twists of algebras discovered by Stafford in [@stafford1994regularity]. The next examples are the algebras classified in [@rogalski2012regular] by Rogalski and Zhang. These are shown in Theorem \[thm: rogzhangmymain\] to have some hidden relations via cocycle twists. Of particular note amongst the remaining examples in the chapter is a twist of an algebra from [@vancliff1994quadratic] in §\[sec: vancliffql\]. We contrast the geometry relating to a factor ring of such a twist and compare it with the twist studied in Chapter \[chap: thcrtwist\]. In §\[subsec: gradedskewclifford\] we twist a graded skew Clifford algebra, while the main portion of the thesis ends by briefly addressing the ungraded situation – we study twists of a universal enveloping algebra and its homogenisation in §\[subsec: homenvelopalg\].
Appendix \[app: calc\] contains any calculations that have been omitted from the main thesis, while Appendix \[app: comp\] contains any computer code used to prove earlier results.
Basic conventions and notation {#sec: notation}
------------------------------
In this short section we state our basic conventions with regard to cocycle twists and also some more general notation.
### Basic conventions {#subsec: conventions}
The conventions that follow will hold throughout this thesis apart from certain occurrences where deviation from them will be stated explicitly.
We assume that $0 \in {\ensuremath{\mathbb{N}}}$ for grading purposes and work over a base field $k$ which is assumed to be algebraically closed. Any algebra $A$ will be an associative $k$-algebra with identity that is finitely generated (f.g.) as an algebra. By $G$ we will denote a finite *abelian* group with identity element $e$. Furthermore, we will assume that the characteristic of the base field does not divide the order of $G$, that is $\text{char}(k) \nmid |G|$. A normalised 2-cocycle will be denoted by $\mu$. Unadorned tensor products $\otimes$ will mean tensor products over $k$, $\otimes_k$.
### Notation {#subsec: notation}
In this section we will describe our basic notation that will be in place throughout the rest of the thesis, unless otherwise stated. More specialised notation may subsequently be defined as and when it is needed.
#### Group actions
The group of *algebra automorphisms* of $A$ is denoted by $\text{Aut}_{\text{alg}}(A)$, while the group of *group automorphisms* of $G$ is denoted by $\text{Aut}_{\text{grp}}(G)$. One of our base assumptions in §\[subsec: theoryoftwist\] was that $G$ acts on $A$ by $k$-algebra automorphisms, i.e. $G
\rightarrow \text{Aut}_{\text{alg}}(A)$. As $G$ is abelian we can assume that the group acts on the right without loss of generality. We will denote the action of an element $g \in G$ by $-^g$. The embedding of $G$ in the automorphism group means that for all $a,b \in A$, $g \in G$ and $\alpha \in k$ one has $(\alpha a)^g=\alpha a^g$ and $(ab)^g=a^gb^g$.
We will also define actions of $G$ on objects other than algebras later in the thesis. The same superscript notation will be used in such circumstances, although the precise nature of the action will be given at the time.
Since $G$ is a finite abelian group, we can write it as a product of cyclic groups. We will denote the cyclic group of order $n$ by $C_n$. The Klein-four group will be pivotal in all of our examples from Chapter \[chap: sklyanin\] onwards, and using this notation it can be denoted by $(C_2)^2$.
#### Gradings
One of the fundamental definitions required for the work in this thesis is the following.
\[defn: ggradedalgebra\] The algebra $A$ is said to be *$G$-graded* if it possesses a $k$-vector space decomposition $A=\bigoplus_{g \in G}A_g$ for which $1 \in A_e$ and $A_g \cdot A_h \subset A_{gh}$ for all $g, h \in G$.
By replacing $G$ with ${\ensuremath{\mathbb{N}}}$ in this definition one obtains the definition of an *[$\mathbb{N}$]{}-graded algebra*.
We will make the additional assumption that $A$ is connected graded for certain results in §\[sec: preservation\]. Moreover, all of our examples of cocycle twists (apart from §\[subsec: homenvelopalg\]) involve such algebras.
\[def: conngraded\] A $k$-algebra $A$ is said to be *connected graded* or *c.g.* if it has an ${\ensuremath{\mathbb{N}}}$-grading $A = \bigoplus_{n \in {\ensuremath{\mathbb{N}}}}A_n$ for which $A_0=k$ and $\text{dim}_k A_n < \infty$ for all $n \in {\ensuremath{\mathbb{N}}}$.
If $A$ is connected graded then one can define the notion of a Hilbert series.
\[def: hilbseries\] If $A$ is a connected graded algebra then its *Hilbert series* with respect to its [$\mathbb{N}$]{}-grading is $H_A(t):=\sum_{n \in {\ensuremath{\mathbb{N}}}}(\text{dim}_kA_n)t^n$.
From now on we will refer to connected graded algebras as c.g. algebras for brevity. When $A$ is [$\mathbb{N}$]{}-graded or a c.g.algebra we will assume that $G$ acts on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms. This means that in addition to the requirement of acting by algebra automorphisms, if $a \in A_n$ then $a^g \in A_n$. The group of ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms is denoted by $\text{Aut}_{{\ensuremath{\mathbb{N}}}-\text{alg}}(A)$.
If $A$ is ${\ensuremath{\mathbb{N}}}$-graded then we assume that it is generated by finitely many elements that are homogeneous with respect to its ${\ensuremath{\mathbb{N}}}$-grading. We do not, however, necessarily assume that the generators lie in degree 1. Our algebras will often be written as quotients of a free algebra $k\{x_0,\ldots,x_n\}$, which can also be described as the tensor algebra $T(V)$ over the vector space $V=kx_0 + \ldots + kx_n$. If we work with an algebra of the form $T(V)/I$ for some ideal $I$, then we will refer to the generators of $I$ as the *defining relations* of the algebra.
At times we will focus our attention on quadratic algebras, particularly when studying the Koszul property in §\[subsec: koszul\].
\[def: quadalg\] Let $V$ be a finite-dimensional vector space over $k$ and $T(V)$ be the tensor algebra over $V$. For an ideal $I$ whose generators lie in $V \otimes V$ we say that the quotient $T(V)/I$ is a *quadratic algebra* over $k$.
Note that our definition implies that quadratic algebras are c.g. $k$-algebras that are generated in degree 1 and whose defining relations lie in degree 2.
#### Modules
We will work with right modules unless otherwise stated. A module $M$ over the algebra $A$ may sometimes be written $M_A$ (or $_AM$ if it is a left module). If a module is *finitely generated* then we will abbreviate this to *f.g.*. Although this abbreviation is also used for finitely generated algebras, it will be clear from the context to which we refer.
Let us now assume that $A$ is ${\ensuremath{\mathbb{N}}}$-graded.
\[def: gradedmodule\] A module $M$ over $A$ is an *${\ensuremath{\mathbb{N}}}$-graded right $A$-module* if it has $k$-vector space decomposition $M=\bigoplus_{n \in {\ensuremath{\mathbb{N}}}}M_n$ such that $M_n \cdot A_m \subset M_{n+m}$ for all $n, m \in {\ensuremath{\mathbb{N}}}$.
Using this definition allows us to define the category of modules in which we will sometimes work.
\[def: gradedmodulecategory\] By $\text{GrMod}(A)$ we denote the *category of [$\mathbb{N}$]{}-graded right $A$-modules*. The objects of this category are the [$\mathbb{N}$]{}-graded right $A$ modules as defined in Definition \[def: gradedmodule\]. A morphism $\phi: M \rightarrow N$ in this category is a homomorphism of right $A$-modules such that $\phi(M_n) \subset
N_n$ for all $n \in {\ensuremath{\mathbb{N}}}$.
Let $M \in \text{GrMod}(A)$. For any $d \in {\ensuremath{\mathbb{N}}}$ we define $M[d]:=\bigoplus_{n \in N} M_n'$ with $M_n'=M_{n+d}$ to be a *degree shift of $M$ by degree $d$*. We also define $M_{\geq d}:=\bigoplus_{n \geq d} M_n$ to be a *tail* of $M$.
For $M, N \in \text{GrMod}(A)$ one can also consider the ${\ensuremath{\mathbb{Z}}}$-graded group $$\text{HOM}_A(M,N) = \bigoplus_{j \in {\ensuremath{\mathbb{Z}}}} \text{Hom}_{\text{GrMod}(A)}(M,N[j]).$$ If $M$ and $N$ are f.g. modules then $\text{HOM}_A(M,N)$ coincides with $\text{Hom}_{A}(M,N)$, the abelian group of ungraded left $A$-module homomorphisms. To prove this, consider a map $\phi \in \text{Hom}_A(M,N)$ and let $m_1,\ldots,m_p$ denote the homogeneous generators of $M$ of degrees $d_1,\ldots, d_p$ respectively. Then $\phi(m_i)$ can be written as a sum of homogeneous elements in $N$, that is $\phi(m_i) \in \bigoplus_{j} N_{r_{i,j}}$ for some integers $r_{i,j} \in {\ensuremath{\mathbb{Z}}}$. Hence $\phi \in \bigoplus_{i,j} \text{Hom}_{\text{GrMod}(A)}(M,N[r_{i,j}-d_j])$. Additionally, the abelian group $\text{Hom}_{A}(M,N)$ inherits a ${\ensuremath{\mathbb{Z}}}$-graded structure from the modules.
For $i \in {\ensuremath{\mathbb{N}}}$ the derived functors $$\label{eq: EXT}
\text{EXT}^i_A(M,N) = \bigoplus_{j \in {\ensuremath{\mathbb{Z}}}} \text{Ext}^i_{\text{GrMod}(A)}(M,N[j]).$$ have a ${\ensuremath{\mathbb{Z}}}$-graded structure. Again, when $M$ and $N$ are f.g. modules $\text{EXT}^i_A(M,N)$ coincides with the ungraded $\text{Ext}^i_{A}(M,N)$. On several occasions we will use this graded structure on cohomology groups. In fact, when either $M$ or $N$ has an $(A,B)$-bimodule structure for some ${\ensuremath{\mathbb{N}}}$-graded $k$-algebra $B$, the cohomology groups $\text{Ext}^i_{A}(M,N)$ become ${\ensuremath{\mathbb{Z}}}$-graded $B$-modules (see [@rotman2008introduction Theorem 1.15] for a description of the relevant module structures).
For a noetherian c.g. algebra $A$ and any f.g. module $M \in \text{GrMod}(A)$, one can define the Hilbert series of $M$ in the same manner as in Definition \[def: hilbseries\]. Such an object is denoted by $H_M(t)$.
One can define $G$-graded modules in an analogous manner to Definition \[def: gradedmodule\] for [$\mathbb{N}$]{}-graded modules. This allows the following category to be defined.
\[def: Ggradedmodcat\] Suppose that $A$ is $G$-graded. By $\text{GrMod}_{G}(A)$ we will denote the *category of $G$-graded right $A$-modules*. The objects in this category are the $G$-graded right $A$-modules. A morphism $\phi: M \rightarrow N$ in this category is a homomorphism of right $A$-modules such that $\phi(M_g) \subset
N_g$ for all $g \in G$.
#### Dimension functions
Gelfand-Kirillov dimension, shortened to *GK dimension*, is a dimension function for algebras and their modules that is defined as follows.
\[def: gkdim\] Let $k$ be a field and $A$ an associative $k$-algebra with identity. Let $V \subset A$ be a $k$-vector space for which $1 \in V$ and define $V^n$ be the vector space spanned by all products of $n$ elements from $V$. The *Gelfand-Kirillov dimension* of $A$ is defined by $$\text{GKdim }A\index{notation}{GKdim $A$}= \underset{V}{\text{sup}} \underset{n \rightarrow \infty}{\varlimsup}
\left(\frac{\text{log dim}_kV^n}{\text{log }n}\right).$$ Here the supremum is taken over all such vector subspaces $V$.
For the basic properties of this dimension function, consult [@krause2000growth]. We will also use several homological dimension functions, defined as follows.
\[def: homdimfunc\] Let $A$ be a $k$-algebra and $M$ be a right $A$-module. Then
- $\text{pdim }M$ denotes the *projective dimension* of $M$. This is the minimal length of a projective resolution of $M$.
- $\text{lgldim }A$ and $\text{rgldim }A$ denote the *left* and *right global dimension* of $A$ respectively. One has $$\text{rgldim }A = \text{sup}\{\text{pdim }M:\; M\text{ is a right }A\text{-module}\},$$ with $\text{lgldim }A$ being defined analogously for left $A$-modules.
- $\text{idim }M$ denotes the *injective dimension* of $M$. This is the minimal length of an injective resolution of $M$.
A good reference for such dimension functions is [@mcconnell2001noncommutative §7.1].
Background material {#chap: background}
===================
Noncommutative algebraic geometry {#sec: noncommgeom}
---------------------------------
In this section we give an introduction to the field of noncommutative algebraic geometry in the sense of [@artin1990some] and [@artin1991modules]. Our perspective is informed by the cocycle twists that we study in this thesis.
Broadly speaking, the aim of noncommutative algebraic geometry is to study noncommutative algebras using the techniques and intuition used to study commutative algebras in algebraic geometry. Unfortunately, there are some fundamental obstacles preventing the translation of the theory verbatim; prime ideals are crucial to the commutative theory, but are relatively scarce in noncommutative algebras.
Throughout §\[sec: noncommgeom\] we will assume that $k$ is an algebraically closed field of characteristic 0 unless otherwise stated. We will work with noetherian c.g. algebras (see Definition \[def: conngraded\]) that are assumed to be f.g. as algebras by degree 1 elements, unless otherwise stated. While the latter assumption is not always necessary, often results which do not require it cannot be stated as succinctly.
### Noncommutative spaces {#subsec: noncommspaces}
Instead of working with prime ideals, one can approach the problem from a categorical perspective. Let $A$ denote a noetherian c.g. $k$-algebra and recall the definition of $\text{GrMod}(A)$ given in Definition \[def: gradedmodulecategory\]. Consider the following subcategories of $\text{GrMod}(A)$:
- $\text{grmod}(A)$, the full subcategory of all f.g. ${\ensuremath{\mathbb{N}}}$-graded $A$-modules;
- $\text{fdmod}(A)$, the full subcategory of finite-dimensional ${\ensuremath{\mathbb{N}}}$-graded $A$-modules.
One can use these subcategories to define the following category, often referred to as the *category of tails*.
\[defn: cattails\] The *noncommutative space* associated to $A$ is the quotient category $$\label{eq: qgrade}
\text{qgr}(A)\index{notation}{q@qgr$(A)$}:=\text{grmod}(A)/\text{fdmod}(A).$$ While $\text{qgr}(A)$ has the same objects as $\text{grmod}(A)$, its morphisms are different. Each morphism in $\text{Hom}_{\text{qgr}(A)}(M,N)$ arises from a morphism in $\text{Hom}_{\text{gr}(A)}(M',N')$, where $M'$ is any graded submodule of $M$ for which $M/M' \in \text{fdmod}(A)$ and $N'\cong N/N''$ in $\text{grmod}(A)$ for some $N'' \in
\text{fdmod}(A)$.
\[rem: qgrdefn\] In [@artin1990twisted §1] the authors define a category $\text{QGr}(A)$ by taking the quotient of $\text{GrMod}(A)$ by the subcategory consisting of direct limits of right bounded modules. If $A$ is right noetherian then [@artin1994noncommutative Proposition 2.3] shows that $\text{QGr}(A)$ is determined up to equivalence by $\text{qgr}(A)$. In this thesis we will work will noetherian algebras, hence Definition \[defn: cattails\] is sufficient for our needs.
Definition \[defn: cattails\] implies that two modules $M, N \in \text{grmod}(A)$ become isomorphic in $\text{qgr}(A)$ if they are isomorphic (as graded modules) in sufficiently high degree, that is $M_{\geq n}\cong N_{\geq n}$ for some $n
\in {\ensuremath{\mathbb{N}}}$. Using the terminology introduced in §\[subsec: notation\] we can say that their tails are isomorphic in high degree. Finite-dimensional ${\ensuremath{\mathbb{N}}}$-graded modules become trivial in $\text{qgr}(A)$, thus are sometimes referred to as *torsion*.
There is a canonical functor related to the noncommutative space $\text{qgr}(A)$, namely $\pi: \text{grmod}(A)
\rightarrow \text{qgr}(A)$, which sends a module to the corresponding object in $\text{qgr}(A)$. For $M \in
\text{grmod}(A)$ we will refer to $\pi(M)$ as the *tail* of $M$. This fits in with our previous definition of a tail of a module since for $M \in \text{grmod}(A)$ one has $\pi(M) \cong \pi(M_{\geq n})$ in $\text{qgr}(A)$ for all $n
\in {\ensuremath{\mathbb{N}}}$. It will be clear from the context whether the term tail refers to a module of the form $M_{\geq n}$ or to $\pi(M)$.
Let us now begin to motivate our use of the term *noncommutative space* in reference to the category defined in Definition \[defn: cattails\]. We need to define a class of rings which is fundamental to noncommutative algebraic geometry.
\[def: thcr\] Let $(X,\sigma,\mathcal{L})$ be a triple consisting of a projective scheme $X$, an automorphism $\sigma$ and an invertible sheaf of $\mathcal{O}_X$-modules $\mathcal{L}$. The shorthand notation $\mathcal{L}^{\sigma}=\sigma^{\ast}\mathcal{L}$ will be used for pullbacks. The *twisted homogeneous coordinate ring* associated to this triple is $$B(X,\mathcal{L},\sigma)\index{notation}{b@$B(X,\mathcal{L},\sigma)$}=\bigoplus_{n \in \mathbb{N}}
\Gamma(X,\mathcal{L}_n),$$ where $\mathcal{L}_0={\ensuremath{\mathcal{O}}}_X$ and the sheaf $\mathcal{L}_n$ for $n \geq 1$ is defined by $$\mathcal{L}_n=\mathcal{L} \otimes_{{\mathcal{O}_X}} \mathcal{L}^{\sigma} \otimes_{{\mathcal{O}_X}} \ldots
\otimes_{{\mathcal{O}_X}} \mathcal{L}^{{\sigma}^{n-1}}.$$ There are natural morphisms of invertible sheaves $$\mathcal{L}_n \otimes_{{\mathcal{O}_X}} \mathcal{L}_m \rightarrow \mathcal{L}_n \otimes_{{\mathcal{O}_X}}
\mathcal{L}_m^{\sigma^{n}} \rightarrow \mathcal{L}_{m+n},$$ which induce the multiplication in $B(X,\mathcal{L},\sigma)$ via taking global sections: $$\label{eq: globalsecs}
H^0(X,\mathcal{L}_n) \otimes H^0(X,\mathcal{L}_m) \rightarrow H^0(X,\mathcal{L}_n) \otimes
H^0(X,\mathcal{L}_m^{\sigma^{n}}) \rightarrow H^0(X,\mathcal{L}_{m+n}).$$
If the invertible sheaf associated to a twisted homogeneous coordinate ring is $\sigma$-ample — whose definition follows — then the ring is noetherian.
\[defn: sigmaample\] Let $(X,\mathcal{L},\sigma)$ be as in Definition \[def: thcr\]. We say that $\mathcal{L}$ is *$\sigma$-ample* if for any coherent sheaf $\mathcal{F}$ of $\mathcal{O}_X$-modules $$H^p(X,\mathcal{F}\otimes \mathcal{L} \otimes \mathcal{L}^{\sigma}\otimes \ldots \otimes \mathcal{L}^{\sigma^{n-1}})=0,$$ for all $p>0$ and $n \gg 0$.
This was originally the definition of a *left $\sigma$-ample* invertible sheaf. Keeler proved in [@keeler2000criteria Theorem 1.2] that this definition is equivalent to a notion of *right $\sigma$-ampleness* for projective schemes over algebraically closed fields. We will mainly work with twisted homogeneous coordinate rings for which the associated projective scheme is an elliptic curve defined over such a field. In that case — as noted in the proof of [@artin1995noncommutative Theorem 4.7] — Corollary 1.6 from [@artin1990twisted] implies that ample sheaves are always $\sigma$-ample and so we do not need to worry about this property.
\[rem: finesse\] We now remark upon another consequence of $\sigma$-ampleness, as demonstrated by [@smith1994center Proposition 2.1]: if $(X,\mathcal{L},\sigma)$ is a triple as in Definition \[def: thcr\] for which $X$ is an integral scheme and $\mathcal{L}$ is $\sigma$-ample, then there is an embedding of algebras $B:=B(X,\mathcal{L},\sigma)
\hookrightarrow k(X)[z,z^{-1};\sigma]$, where $k(X)$ is the function field of $X$. In this setting one may define $\overline{B}_n:=\text{H}^0(X,{\ensuremath{\mathcal{L}}}_n) \subset k(X)$ for all $n \in {\ensuremath{\mathbb{N}}}$, in which case one has $B_n=\overline{B}_n z^n$ upon considering $B$ as a subring of the Ore extension. This notational finesse allows one to concretely see the twisted multiplication in $B$ arising from the structure of the Ore extension. We will use such ideas in §\[subsec: geomdata\].
Let us now try to justify our usage of the term noncommutative space in Definition \[defn: cattails\] by first considering what happens in the commutative situation. We therefore assume until further notice that $A$ is commutative in addition to our base assumptions of being a noetherian c.g algebra that is generated in degree 1.
Associated to $A$ is a projective scheme $X= \text{Proj(A)}$ and an embedding $j: X \hookrightarrow {\ensuremath{\mathbb{P}_{k}^{n}}}$ for some $n \in {\ensuremath{\mathbb{N}}}$. The following sheaf is crucial to the proof of our next result, Serre’s theorem (Theorem \[thm: serresthm\]).
\[def: twistingsheaf\] Let $A$, $X$ and $j$ be as in the previous paragraph. The *twisting sheaf* ${\ensuremath{\mathcal{O}}}_X(1)$ is defined to be the pullback $j^{\ast}{\ensuremath{\mathcal{O}}}_{{\ensuremath{\mathbb{P}_{k}^{n}}}}(1)$.
\[thm: serresthm\] Let $A$ be a commutative, noetherian c.g. algebra generated in degree 1. Then there is an equivalence of categories between $\text{qgr}(A)$ and the category of coherent $\mathcal{O}_X$-modules, $\text{coh}({\ensuremath{\mathcal{O}}}_X)$, where $X= \text{Proj(A)}$ is the projective scheme determined by $A$.
Serre’s theorem can be proved by focusing on the distinguished element $\pi(A) \in \text{qgr}(A)$. The module $A_A$ can be reconstructed in high degrees by taking global sections of the invertible sheaves $\mathcal{O}_{X}(n)={\ensuremath{\mathcal{O}}}_X(1)^{\otimes n}$ for $n \gg 0$, along with a natural multiplication obtained from by taking $\sigma=\text{id}$ and ${\ensuremath{\mathcal{L}}}={\ensuremath{\mathcal{O}}}_{X}(1)$. In light of this observation it is not surprising that twisted homogeneous coordinate rings form the gateway to a theory of noncommutative algebraic geometry.
Let us now return to the noncommutative setting and our base assumptions. Thus $A$ is a noetherian c.g. algebra that is generated in degree 1. The following theorem, known as the *Noncommutative Serre’s theorem*, shows that for twisted homogeneous coordinate rings the noncommutative space $\text{qgr}(B(X,\mathcal{L},\sigma))$ is actually a commutative geometric object.
\[thm: noncomserrethm\] Let $X$ be a projective scheme over a field $k$, $\sigma$ an automorphism of $X$, and $\mathcal{L}$ a $\sigma$-ample invertible sheaf on $X$. Then the twisted homogeneous coordinate ring $B(X,\mathcal{L},\sigma)$ is a f.g. noetherian algebra and there is an equivalence of categories $$\text{qgr}(B(X,\mathcal{L},\sigma)) \simeq \text{coh}(\mathcal{O}_X).$$
Some of the central objects of study in noncommutative algebraic geometry are the *Artin-Schelter-regular algebras* (AS-regular). These noncommutative algebras are defined in Definition \[defn: asregular\] and display many of the properties of commutative polynomial rings. For this reason, an AS-regular algebra of global dimension $n$ is often regarded as the coordinate ring of a noncommutative [$\mathbb{P}_{k}^{n-1}$]{}. The noncommutative projective planes (where $n = 3$) were classified in [@artin1990some; @artin1991modules] for algebras generated in degree 1 and in [@stephenson1996artin] for the general case. Their classification relied heavily on geometric techniques which we will soon discuss.
The results in this thesis, particularly those in Chapter \[chap: sklyanin\], relate to the classification of noncommutative [$\mathbb{P}_{k}^{3}$]{}’s, or, equivalently, AS-regular algebras of dimension 4. While no classification of such algebras is on the horizon, many families of such algebras have been found. Examples include 4-dimensional Sklyanin algebras and their close relations [@smith1992regularity; @stafford1994regularity], homogenisations of universal enveloping algebras [@lebruyn1993homogenized], algebras associated to quadrics [@vancliff1994quadratic; @vancliff1997embedding], double Ore extensions [@zhang2008double], algebras with an additional ${\ensuremath{\mathbb{Z}}}^2$-grading [@lu2007regular; @rogalski2012regular] and the algebras in [@vancliff1998some; @cassidy2006generalized; @cassidy2010generlizations].
We now consider an example of how a particular AS-regular algebra of low dimension fits in with Theorem \[thm: noncomserrethm\],
\[eg: 3dimsklyanin\] Consider a 3-dimensional Sklyanin algebra $S(a,b,c)$, associated to the parameters $(a,b,c) \in {\ensuremath{\mathbb{P}_{k}^{2}}}$ and generated by three degree 1 elements. Such algebras are those of type $A$ with $r= 3$ in Artin and Schelter’s classification in [@artin1987graded] (see (10.14) op. cit. for the defining relations).
Apart from 12 choices of parameters in ${\ensuremath{\mathbb{P}_{k}^{2}}}$, $S(a,b,c)$ is AS-regular of global dimension 3. Thus $\text{qgr}(S(a,b,c))$ is a noncommutative ${\ensuremath{\mathbb{P}_{k}^{2}}}$. Furthermore, there is a graded surjection from $S(a,b,c)$ to the twisted homogeneous coordinate ring $B(E,\mathcal{L},\sigma)$, where $E \subset {\ensuremath{\mathbb{P}_{k}^{2}}}$ is an elliptic curve, $\mathcal{L}={\ensuremath{\mathcal{O}}}_E(1)$ a very ample invertible sheaf with 3 global sections and $\sigma$ a translation automorphism.
At the level of noncommutative spaces we see that the noncommutative [$\mathbb{P}_{k}^{2}$]{} associated to $S(a,b,c)$ contains $\text{qgr}(B(E,\mathcal{L},\sigma))$. By the Noncommutative Serre’s Theorem the latter category is equivalent to $\text{coh}({\ensuremath{\mathcal{O}}}_X)$, thus $\text{qgr}(S(a,b,c))$ contains a commutative geometric category.
In general, one can ask if there exists a graded surjection from a given c.g. algebra to a twisted homogeneous coordinate ring that gives geometric information about the algebra. The following property is relevant.
\[defn: strongnoeth\] Let $A$ be a noetherian c.g. algebra. $A$ is *strongly noetherian* if for all commutative noetherian $k$-algebras $R$, $A \otimes R$ is noetherian.
Under the hypothesis that an algebra is strongly noetherian and generated in degree 1, one always has a map from it to some noetherian twisted homogeneous coordinate ring [@rogalski2008canonical Theorem 1.1] (in fact, the scheme involved is the *point scheme* of the algebra, see Definition \[def: pointscheme\]). In particular, many AS-regular algebras of low dimension are strongly noetherian. Thus a noncommutative [$\mathbb{P}_{k}^{n-1}$]{} will often contain a bona fide projective scheme in the sense that $\text{coh}({\ensuremath{\mathcal{O}}}_X)\subset \text{qgr}(A)$ for some projective scheme $X$ (as in Example \[eg: 3dimsklyanin\]). Unfortunately, not every noetherian algebra is strongly noetherian as Rogalski’s examples in [@rogalski2004generic] first illustrated.
Theorems \[thm: sklyanin\] and \[thm: geometricthcrdesc\] suggest that it may be possible to prove a result in the vein of [@rogalski2008canonical Theorem 1.1] for *twisted rings* rather than twisted homogeneous coordinate rings. Twisted rings are defined in Definition \[defn: twistedhomring\] and are one of the fundamental components of Artin and Stafford’s classification of noncommutative curves, begun in [@artin1995noncommutative] and completed in [@artin2000semiprime]. Artin and Stafford’s work in these papers allows all semiprime noetherian algebras of GK dimension 2 to be described in terms of twisted rings (see Theorem \[thm: artinstaffordmain\] for a more precise formulation). We will return to the theme of canonical maps to twisted rings in §\[subsec: irrinqgr\].
Having associated a noncommutative space to a c.g. algebra, one can ask the following two questions to motivate the development of the subject:
\[que: motivquenag\]
- What are the irreducible objects in $\text{qgr}(A)$?
- What are the noncommutative analogues of points, lines, etc. in $\text{qgr}(A)$?
In the next section we will give an overview of attempts to answer these questions.
### Irreducible objects in noncommutative spaces {#subsec: irrinqgr}
In order to answer Questions \[que: motivquenag\] we will once again consider what happens in the commutative case. Thus we assume until further notice that $A$ is a commutative noetherian c.g. algebra generated in degree 1. Associated to $A$ is a projective scheme $X$, whose embedding in projective space gives rise to a twisting sheaf as in Definition \[def: twistingsheaf\].
Before considering irreducible objects in $\text{qgr}(A)$ we need the following definition.
\[def: idealsheaf\] Consider a projective scheme $X$ and a closed point $p \in X$, along with the inclusion morphism $i: p \hookrightarrow
X$. The *ideal sheaf* of $p$, denoted by $\mathcal{I}_p$, is the kernel of the morphism $i^{\sharp}: {\ensuremath{\mathcal{O}}}_X \rightarrow
i_{\ast}{\ensuremath{\mathcal{O}}}_{X,p}$. Here ${\ensuremath{\mathcal{O}}}_{X,p}$ denotes the local ring of $X$ at $p$.
The equivalence of categories in Theorem \[thm: serresthm\] tells us that the irreducible objects in $\text{qgr}(A)$ correspond to the irreducible objects in $\text{coh}({\ensuremath{\mathcal{O}}}_X)$. In the latter these are the skyscraper sheaves $k_p:={\ensuremath{\mathcal{O}}}_X/\mathcal{I}_p$, supported only at a closed point $p \in X$.
The functor $\text{coh}({\ensuremath{\mathcal{O}}}_X) \rightarrow \text{qgr}(A)$ is defined on a coherent sheaf of ${\ensuremath{\mathcal{O}}}_X$-modules ${\ensuremath{\mathcal{F}}}$ by $$\label{eq: serrecatequivdefn}
\mathcal{F} \mapsto \pi\left(\bigoplus_{n \in {\ensuremath{\mathbb{N}}}} H^0(X,\mathcal{F}\otimes_{{\ensuremath{\mathcal{O}}}_X} {\ensuremath{\mathcal{O}}}_X(n))\right),$$ where $\pi: \text{grmod}(A) \rightarrow \text{qgr}(A)$ is the canonical functor to the associated noncommutative space.
The sheaf $k_p$ is sent under the equivalence described in to the tail of the module $M_p:=A/I_p$ for an appropriate ideal $I_p$. For any closed point $p
\in X$ the module $M_p$ has Hilbert series $1/(1-t)$. If $N$ denotes a nontrivial ${\ensuremath{\mathbb{N}}}$-graded submodule of $M_p$ then, since the factor module $M_p/N$ is a finite-dimensional $k$-vector space, one has $\pi(N) \cong \pi(M_p)$ in $\text{qgr}(A)$. Equivalently, $M_p$ is irreducible in $\text{qgr}(A)$.
Let us now return to our base assumptions, thus $A$ is a noetherian c.g. $k$-algebra generated in degree 1. Combining our observations regarding the commutative case with the Noncommutative Serre’s Theorem suggests that the following is a good definition.
\[defn: pointmodule\] Let $A$ be a noetherian c.g. algebra. An ${\ensuremath{\mathbb{N}}}$-graded right $A$-module $M$ is a *point module* if it is cyclic and has Hilbert series $1/(1-t)$.
Note that these conditions guarantee that any ${\ensuremath{\mathbb{N}}}$-graded submodule of a point module has finite codimension when $A$ is generated in degree 1. Thus if $M_p$ is a point module then its tail $\pi(M_p)$ will be irreducible in $\text{qgr}(A)$. However, there can exist irreducible objects in $\text{qgr}(A)$ other than the tails of point modules. As remarked at the beginning of [@smith1994four §7], this is comparable to the fact that noncommutative algebras can have simple modules of dimension greater than 1.
\[defn: fatpointmodule\] Let $A$ be a noetherian c.g. algebra. An ${\ensuremath{\mathbb{N}}}$-graded right $A$-module $M$ is a *fat point module of multiplicity $e$* if it has Hilbert series $e/(1-t)$, is generated in degree 0 and any non-zero ${\ensuremath{\mathbb{N}}}$-graded submodule has finite codimension.
\[rem: multiplicitymodule\] By [@artin1991modules Proposition 2.21(iii)], the multiplicity of a f.g. ${\ensuremath{\mathbb{N}}}$-graded module $M$ can be defined more generally: it is the leading coefficient of the expansion of the Hilbert series $H_M(t)$ in terms of powers of $(1-t)$. It is clear that this more general definition of multiplicity coincides with the use of the term in Definition \[defn: fatpointmodule\].
One recovers the definition of a point module from that of a fat point module of multiplicity 1; as remarked after Definition \[defn: pointmodule\], the finite codimension condition on ${\ensuremath{\mathbb{N}}}$-graded submodules automatically holds for a cyclic module with Hilbert series $1/(1-t)$. For general fat point modules this condition is part of the definition to ensure that the tails of such modules are irreducible in $\text{qgr}(A)$. We will refer to the tail (in $\text{qgr}(A)$) of a fat point module as a *fat point*, and similarly a *point* will refer to the tail of a point module.
One could approach the question of irreducibility from another angle. The condition in Definition \[defn: fatpointmodule\] regarding every ${\ensuremath{\mathbb{N}}}$-graded submodule having finite codimension can be restated in terms of critical modules.
\[def: criticalmodule\] A module $M \in \text{grmod}(A)$ is said to be *$n$-critical* if $\text{GKdim}(M)=n$ and $\text{GKdim}(M/N)<n$ for any non-zero ${\ensuremath{\mathbb{N}}}$-graded submodule $N$.
A basic property of GK dimension is that the finite-dimensional modules over any algebra have GK dimension 0. This has two consequences for us; firstly, it means that 1-critical modules have irreducible tails in $\text{qgr}(A)$, and secondly it implies that fat point modules of any multiplicity are 1-critical.
In relation to Question \[que: motivquenag\](i), irreducible objects in $\text{qgr}(A)$ are, unfortunately, not always tails of 1-critical modules (nor fat point modules). An example of Smith [@smith1992the pg. 5] demonstrates this: if $A=k[x,y]$ is given the grading $x \in A_1$, $y \in A_2$, then the tail of the module $A/xA$ is irreducible in $\text{qgr}(A)$. This module has multiplicity 1/2 and so is not a fat point module. Smith’s manuscript also gives an example of a subring of a differential operator ring over which there is a module with Hilbert series $1/(1-t)^2$, whose tail is irreducible in the associated noncommutative space.
Nonetheless, fat point modules do exhaust all irreducible objects in $\text{qgr}(A(\alpha,\beta,\gamma))$ for a large class of 4-dimensional Sklyanin algebras by [@smith1994four Proposition 7.1] (recall Definition \[def: sklyanin relations\] for the construction of $A(\alpha,\beta,\gamma)$). We will elaborate on this point shortly.
Finding the point modules of a c.g. algebra $A$ is often the starting point to determining the geometry of $\text{qgr}(A)$. A crucial object in relation to point modules is the *point scheme* of an algebra, which was first introduced in [@artin1990some §3]. We must define some ingredients to be able to describe the point scheme of an algebra, namely truncated point modules and multilinearisations.
\[def: truncated\] Let $A$ be a noetherian c.g. algebra generated by $n$ elements of degree 1. For $d \in {\ensuremath{\mathbb{N}}}$ a *truncated point module of length $d+1$* is a module $M \in \text{grmod}(A)$ that is cyclic and has Hilbert series $H_M(t)=\sum_{i=0}^{d}t^i$.
In the following definition we use the shorthand notations $\overline{\{n\}}:=
\{0,1,\ldots,n\}$ and $\underline{i} :=
(i_0,\ldots,i_{d-1})$ for a vector in $I:=\overline{\{n-1\}}^d$, where the length of the vector is clear from the context.
\[def: multilin\] Let $A=T(V)/I$ be as in Definition \[def: truncated\], with $V = kx_0 + \ldots + k x_{n-1}$. Let $d \in {\ensuremath{\mathbb{N}}}$ and $f \in
I_d$ be a relation, where $f = \sum_{I}\alpha_{\underline{i}}x_{i_{0}}\ldots x_{i_{d-1}}$ for some $\alpha_{\underline{i}} \in k$. Define $\tilde{f}$ to be the element of the homogeneous coordinate ring of $({\ensuremath{\mathbb{P}_{k}^{n-1}}})^{\times d}$ given by $$\sum_{I}\alpha_{\underline{i}}x_{i_{0},0}x_{i_{1},1}\ldots x_{i_{d-1},d-1},$$ where the coordinate ring of the $(j+1)$’th copy of ${\ensuremath{\mathbb{P}_{k}^{n-1}}}$ is $k[x_{0,j},\ldots, x_{n-1,j}]$. We say that $\tilde{f}$ is the *multilinearisation* of $f$.
Multilinearisations allow truncated point modules of length $d+1$ to be parameterised by a projective scheme $\Gamma_d$, as proved in [@artin1990some Proposition 3.9]. Furthermore, this parameterisation is functorial in the sense that there are natural maps in each situation that correspond to each other (formally, $\Gamma_d$ represents the functor of flat families of truncated point modules of length $d+1$). To see this correspondence, consider the map that sends a truncated point module of length $d+1$, $M = \bigoplus_{i=0}^{d}M_i$, to $\bigoplus_{i=0}^{d-1}M_i$ by factoring out the highest degree component. This map sends $M$ to a truncated point module of length $d$. A morphism of schemes $\Gamma_{d} \rightarrow \Gamma_{d-1}$ is induced and, as [@artin1990some Proposition 3.7(i)] shows, the induced morphism is nothing more than the projection $({\ensuremath{\mathbb{P}_{k}^{n-1}}})^{\times d} \rightarrow ({\ensuremath{\mathbb{P}_{k}^{n-1}}})^{\times d-1}$ which ‘forgets’ the final component of the product.
\[def: pointscheme\] Let $A$ be as in Definition \[def: truncated\]. The *point scheme* of $A$ is the inverse limit $\Gamma$ of the following diagram: $$\begin{tikzpicture}
\draw [<-] (0.3,2) -- (1.2,2);
\draw [<-] (1.8,2) -- (2.7,2);
\draw [<-] (3.3,2) -- (4.1,2);
\draw [<-] (4.9,2) -- (5.65,2);
\draw [<-] (6.4,2) -- (7.2,2);
\node at (0,2) {$\Gamma_0$};
\node at (1.5,2) {$\Gamma_1$};
\node at (3,2) {$\ldots$};
\node at (4.5,2) {$\Gamma_{d-1}$};
\node at (6,2) {$\Gamma_{d}$};
\node at (7.5,2) {$\ldots$};
\node at (3.75,0) {\large{$\Gamma$}};
\draw [->] (3.58,0.18) -- (0,1.7);
\draw [->] (3.69,0.3) -- (1.5,1.7);
\draw [->] (3.79,0.3) -- (4.5,1.7);
\draw [->] (3.92,0.18) -- (6,1.7);
\end{tikzpicture}$$ Here $\Gamma_d$ denotes the projective scheme parameterising truncated point modules of length $d+1$, with the maps $\Gamma_{d}\rightarrow \Gamma_{d-1}$ as described in the previous paragraph.
By [@artin1990some Corollary 3.13] the points of $\Gamma$ are in 1-1 correspondence with point modules of $A$. This means that $\Gamma$ ‘parameterises’ point modules in the following sense: $\Gamma$ represents the functor mapping a f.g. commutative $k$-algebra $C$ to the set of isomorphism classes of factors of $A \otimes_k C$ with Hilbert series $1/(1-t)$. The use of the word parameterises in Theorem \[thm: artzhanglimit\] below reflects this.
In particularly nice cases one of the morphisms in the inverse limit structure of the point scheme is an isomorphism. Proposition 3.7(ii) from [@artin1990some] then implies that all higher degree morphisms are also isomorphisms. Intrepreted in terms of modules, this means that given a truncated point module $M$ of sufficiently great length $d \gg
0$, there exists a unique point module $M'$ such that $M \cong M'/M'_{\geq d}$.
For strongly noetherian c.g. algebras the following theorem shows that the point scheme has a particularly nice form.
\[thm: artzhanglimit\] Assume that $A$ is a strongly noetherian c.g. algebra over a field $k$. Then the point scheme of $A$ is a projective scheme that parameterises the point modules over $A$.
\[rem: justifyscheme\] While the term *point scheme* is misleading in the sense that such an object is not necessarily a scheme, Theorem \[thm: artzhanglimit\] shows that the term is justified under certain hypotheses.
All of the algebras we will study in this thesis are strongly noetherian, thus their point schemes are projective schemes.
Let us now consider Rogalski and Zhang’s result [@rogalski2008canonical Theorem 1.1], describing a canonical map from a strongly noetherian c.g. algebra generated in degree 1 to a twisted homogeneous coordinate ring over the point scheme of the algebra. Theorem \[thm: artzhanglimit\] has implications for the relevance of Theorems \[thm: sklyanin\] and \[thm: geometricthcrdesc\] to a generalisation of Rogalski and Zhang’s result; before even considering a canonical map to a twisted ring as defined in Definition \[defn: twistedhomring\], one should try to discover the structure of the geometric object — if such an object exists — parameterising fat point modules of higher multiplicity. In fact, [@artin2001abstract Theorem E5.1] provides an answer to this question for algebras satisfying a strong homological condition, with the geometric objects shown to be countable unions of projective schemes.
Let us assume now that our algebra $A$ is strongly noetherian in addition to our other hypotheses. More information about point modules can be uncovered by using the following technique, described in the introduction of [@artin1991modules].
\[defn: ptschemeautomorphism\] Let $A$ be a strongly noetherian c.g. algebra with point scheme $\Gamma$, and let $M_p$ be a point module over $A$ corresponding to a point $p \in \Gamma$. The shift $M_p[1]_{\geq 0}$ is also a point module over $A$, therefore it corresponds to some point $p^{\sigma}\in \Gamma$. This shifting operation induces a scheme automorphism $\sigma: \Gamma
\rightarrow \Gamma$ which we call the *associated automorphism of the point scheme*.
We will now state a result of Shelton and Vancliff concerning the point schemes of certain algebras. It will be useful to us on several occassions, notably in §\[subsec: 4sklypointscheme\]. Any undefined properties in the statement of the result will be defined in §\[subsec: cohenmac\].
\[thm: pointschemenice\] Assume that $k$ is an algebraically closed field for which $\text{char}(k) \neq 2$. Let $A$ be a c.g. $k$-algebra with point scheme $\Gamma$, and suppose that the following are true for $A$:
- it is generated in degree 1 and has Hilbert series $1/(1-t)^4$;
- it is noetherian and Auslander regular of global dimension 4;
- it is Cohen-Macaulay.
Then the projective scheme $\Gamma_2$ corresponding to the multilinearisations of the defining relations of $A$ is isomorphic to $$\label{eq: graphptscheme}
\Gamma_2 = \{(p,p^{\sigma}):\; p \in \Gamma\} \subset {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}},$$ the graph of $\Gamma$ under the associated automorphism $\sigma$.
\[not: projmap\] Theorem \[thm: pointschemenice\] will be applicable to many of the algebras that we study. We introduce the notation $\pi_i: {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}} \rightarrow {\ensuremath{\mathbb{P}_{k}^{3}}}$ for $i =1,2$ to denote the projection morphism to the $i$’th coordinate. In the setting of the theorem one may therefore write $\pi_1(\Gamma_2)
= \Gamma$.
In the language of Definition \[def: pointscheme\], Theorem \[thm: pointschemenice\] tells us that the projection morphism $\Gamma_2 \rightarrow \Gamma_1$ defines an isomorphism from $\Gamma_2$ to the point scheme $\Gamma$. Translated into a statement about modules, it implies that each truncated point module of length 3 over such an algebra uniquely determines a point module; to study point modules it suffices to study the multilinearisations of the quadratic relations in the algebra.
The point scheme of a c.g. algebra is an important invariant that is preserved by ${\ensuremath{\mathbb{N}}}$-graded isomorphisms. Moreover, it is preserved by certain instances of a twisting operation on algebras called *Zhang twists*. While we will discuss Zhang twists fully in §\[subsec: zhangtwist\], we define a special class of them here.
\[def: algebraictwist\] Suppose that $A$ is a c.g. algebra and $\phi$ is an ${\ensuremath{\mathbb{N}}}$-graded algebra automorphism of $A$. One can define a new multiplicative structure on $A$ by defining $a \ast_{\phi} b := a \phi^n(b)$ for all $a \in A_n$ and $b \in A_m$. The algebra $(A,\ast_{\phi})$ is called a *Zhang twist* of $A$ by $\phi$.
If a c.g. algebra $B$ is a Zhang twist of the $A$ as in Definition \[def: algebraictwist\], then by [@zhang1998twisted Theorem 3.1] one has $\text{GrMod}(A) \simeq \text{GrMod}(B)$, hence $\text{qgr}(A) \simeq \text{qgr}(B)$. Thus the geometry encoded by point modules and fat point modules is preserved under such twists.
In general, Zhang twists can involve $G$-gradings for any semigroup $G$, and as Theorem \[thm: cocycleaszhang\] demonstrates, the cocycle twists in Proposition \[prop: twoconstrsaresame\] can be phrased as Zhang twists. In that situation, [@zhang1998twisted Theorem 3.1] shows that there is an equivalence of categories $\text{GrMod}_{G}(A)
\simeq \text{GrMod}_{G}(A^{G,\mu})$, where $G$ is the relevant semigroup. When $G \neq {\ensuremath{\mathbb{N}}}$ this equivalence gives relatively little information about the relationship between $\text{qgr}(A)$ and $\text{qgr}(A^{G,\mu})$.
Our results show that $\text{qgr}(A)$ and $\text{qgr}(A^{G,\mu})$ can be very different. Let us focus on the 4-dimensional Sklyanin algebra $A(\alpha, \beta, \gamma)$, under the additional assumption that the associated automorphism $\sigma$ has infinite order. All fat point modules over $A(\alpha, \beta, \gamma)$ are known, and by [@smith1994four Proposition 7.1] fat point modules constitute all of the irreducible objects in $\text{qgr}(A(\alpha, \beta, \gamma))$. In particular:
- its points are parameterised by an elliptic curve $E$ and four extra points in [$\mathbb{P}_{k}^{3}$]{} [@smith1992regularity Propositions 2.4 and 2.5];
- fat points of all multiplicities were classified by Smith and Staniszkis in [@smith1993irreducible].
By contrast, Theorem \[thm: sklyanin\] shows that $A(\alpha, \beta, \gamma)^{G,\mu}$ has infinitely many fat point modules of multiplicity 2 and only 20 point modules, the former valid without the assumption that $|\sigma|=\infty$.
Before moving on we briefly address our method of constructing the fat point modules of multiplicity 2 over $A(\alpha,
\beta, \gamma)^{G,\mu}$. Fat point modules over $A(\alpha, \beta, \gamma)$ were realised in [@smith1993irreducible] as factor modules of line modules (see Definition \[defn: linearmodules\] with $n=1$). They also appear as a byproduct of some categorical equivalences described in [@van1996translation Proposition 7.5.2].
Our method uses the fact that $kG_{\mu} \cong M_2(k)$ for the Klein-four group $G$ and some normalised 2-cocycle $\mu$ (see Lemma \[lem: kgmuiso\]). This means that $A(\alpha, \beta, \gamma)^{G,\mu}$ is a subring of $M_2(A(\alpha, \beta,
\gamma))$ via Construction 2 of the twist. It is then natural to take direct sums of modules over $A(\alpha, \beta,
\gamma)$ and consider them as $M_2(A(\alpha, \beta, \gamma))$-modules, before restricting them down to the cocycle twist. For a point module over $A(\alpha, \beta, \gamma)$ corresponding to a point on the elliptic curve $E$, this method produces a fat point module of multiplicity 2 over $A(\alpha, \beta, \gamma)^{G,\mu}$.
Let us now address Question \[que: motivquenag\](ii) – what are the noncommutative analogues of lines, planes, etc.? Once again we can look to the commutative case for inspiration, which suggests the following definition.
\[defn: linearmodules\] Let $A$ be a noetherian c.g. algebra. A module $M \in \text{grmod}(A)$ is a *linear module of dimension $n$* if it is cyclic and has Hilbert series $1/(1-t)^{n+1}$.
For $n=0$ one recovers point modules, while modules in the $n=1,2$ cases are referred to as *line modules* and *plane modules* respectively. Thus lines and planes in $\text{qgr}(A)$ can be defined as the tails of line and plane modules in $\text{grmod}(A)$ similarly. Another indication that $\text{qgr}(A(\alpha,\beta,\gamma))$ is well-understood is given by Staniszkis’s classification of linear modules of all dimensions in [@staniszkis1996linear].
We saw in Theorem \[thm: artzhanglimit\] that the point scheme of an algebra parameterises its point modules under the strongly noetherian hypothesis. Is there an object which plays a similar role for larger linear modules? This is the question addressed by the research in [@shelton2002schemes]. One of the results in that paper, namely Corollary 1.5 op. cit., shows that there is a scheme $\Omega_{\infty}(A,n)$ whose closed points parameterise the $n$-linear modules over $A$. When $n=1$ this scheme is called the *line scheme* of the algebra, while for $n=0$ one recovers the point scheme.
Now consider the following conditions on an algebra $A$.
\[cond: linescheme\] Suppose that $A$ satisfies the following conditions:
- $A$ has Hilbert series $H_A(t)=1/(1-t)^4$;
- $A$ is a domain;
- any plane module $M$ over $A$ is 3-critical, and any nontrivial ${\ensuremath{\mathbb{N}}}$-graded submodule of $M$ also has GK dimension 3.
If $A$ satisfies these conditions then its line scheme will not only be a projective scheme, but can be characterised in several different ways [@shelton2002schemes cf. Lemma 2.4]. Furthermore, by Corollary 2.6 op. cit. the irreducible components of the line scheme of an algebra satisfying Conditions \[cond: linescheme\] are at least 1-dimensional.
There are some examples in the literature of AS-regular algebras of dimension 4 with a 1-dimensional line scheme; see [@vancliff1998some Proposition 3.5] or the algebras defined in [@cassidy2010generlizations Example 5.1] for example. Some of the algebras that we study later in this thesis will be shown to have a 1-dimensional line scheme (see Propositions \[prop: linescheme1dim\] and \[prop: staffordlinescheme\]).
### Algebras with a 0-dimensional point scheme {#sec: finitedimptscheme}
One of the major themes of the papers [@vancliff1998some; @shelton1999some; @cassidy2006generalized; @stephenson2007constructing; @cassidy2010generlizations] is the study and construction of AS-regular algebras of dimension 4 with a 0-dimensional point scheme. As Theorem \[thm: sklyanin\] shows, some examples of cocycle twists that we study can be considered alongside these examples.
The original example of a noncommutative algebra with a 0-dimensional point scheme is given in the unpublished manuscript [@van1988example]. Within it, Van den Bergh shows that graded Clifford algebras with ‘generic’ relations have 20 point modules, related in pairs by the shifting automorphism of Definition \[defn: ptschemeautomorphism\]. These algebras were generalised in [@cassidy2010generlizations], with a cocycle twist of such a generalisation being studied in §\[subsec: gradedskewclifford\].
As remarked in several papers, for example in [@shelton1999embedding §1.2], an AS-regular algebra of dimension 4 that is generated in degree 1 with Hilbert series $1/(1-t)^4$ and ‘generic’ relations has a 0-dimensional point scheme. Furthermore, if such an algebra has a 0-dimensional point scheme then it has at most 20 point modules; the precise number depends on the multiplicities of the closed points in the point scheme (see the remarks prior to [@goetz2003noncommutative Definition II.1.7]).
Since we have been unable to find a proof of the latter fact in the literature, we give one below. In the proof we consider certain irreducible projective schemes as projective varieties in order that we may use the notion of degree for such objects, as defined in [@hartshorne1977algebraic Chapter I, Definition pg. 52].
\[prop: genericpointscheme\] Let $k$ be any algebraically closed field and assume that the $k$-algebra $A=T(V^{\ast})/I$ satisfies conditions (i)-(iii) in Theorem \[thm: pointschemenice\]. If the point scheme of $A$ is 0-dimensional then $A$ has 20 point modules counting multiplicity.
By assumption $A$ satisfies the hypotheses of Theorem \[thm: pointschemenice\]. That result implies that the point scheme of $A$ is $\Gamma :=\pi_1(\Gamma_2)$, where $$\Gamma_2 \subset \mathbb{P}(V) \times \mathbb{P}(V) = {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$$ is the scheme determined by the multilinearisations of the quadratic relations of $A$. Furthermore, $\Gamma_2$ is the graph of $\Gamma$ under an automorphism. Combined with the assumption that $\Gamma$ is 0-dimensional, this implies that $\Gamma_2$ must also be 0-dimensional. The closed points of $\Gamma$ parameterise point modules over $A$, thus it suffices to prove that $\Gamma_2$ contains 20 points counting multiplicity.
We may use the Segre embedding to consider ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ (and thus $\Gamma_2$) as a closed subscheme of $\mathbb{P}(V \otimes V) = {\ensuremath{\mathbb{P}_{k}^{15}}}$. Note that $\Gamma_2$ consists of precisely those points in ${\ensuremath{\mathbb{P}_{k}^{3}}} \times
{\ensuremath{\mathbb{P}_{k}^{3}}}$ at which all elements in $I_2$ vanish upon evaluation. One naturally has $I_2^{\perp} \subset V \otimes V$, in which case $\Gamma_2$ can be described inside ${\ensuremath{\mathbb{P}_{k}^{15}}}$ as the intersection of ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ with the projectivisation of $I_2^{\perp}$, namely $\mathbb{P}(I_2^{\perp})$. Thus $\Gamma_2=({\ensuremath{\mathbb{P}_{k}^{3}}} \times
{\ensuremath{\mathbb{P}_{k}^{3}}}) \cap
\mathbb{P}(I_2^{\perp}) \subset {\ensuremath{\mathbb{P}_{k}^{15}}}$.
To see that $\Gamma_2$ consists of 20 points counting multiplicity we can apply Bezout’s Theorem [@hartshorne1977algebraic Chapter I, Theorem 7.7]. In the notation of that result we take $H=\mathbb{P}(I_2^{\perp})$ and $Y =
{\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$, which can both be considered as projective varieties. By assumption, $I_2$ is a 6-dimensional $k$-vector space, hence $I_2^{\perp} \subset V \otimes V$ has dimension 10. It follows that $\mathbb{P}(I_2^{\perp})= {\ensuremath{\mathbb{P}_{k}^{9}}}$, whence it has degree 1. Using [@hartshorne1977algebraic Chapter I, Exercise 7.1(b)] one can see that ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ has degree $\binom{3+3}{3}$, which equals 20. Bezout’s theorem tells us that the sum of the multiplicities of the points in $\Gamma_2$ is equal to $\text{deg}({\ensuremath{\mathbb{P}_{k}^{3}}} \times
{\ensuremath{\mathbb{P}_{k}^{3}}})\cdot\text{deg}(\mathbb{P}(I_2^{\perp}))$, which is 20.
In light of Proposition \[prop: genericpointscheme\], Theorem \[thm: sklyanin\](ii) tells us that the Sklyanin twist $A(\alpha,\beta,\gamma)^{G,\mu}$ has the maximal number of point modules given that its point scheme is 0-dimensional.
The following result shows that certain algebras are determined by the scheme $\Gamma_2$.
\[thm: pointschemereconstruct\] Fix a base field $k$ that is algebraically closed and for which $\text{char}(k) \neq 2$. Let $T(V)/I$ be a quadratic algebra on four generators with six defining relations, and let $\Gamma_2 \subset {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ denote the zero locus of $I_2$ in the sense of Definitions \[def: multilin\] and \[def: pointscheme\]. If $\Gamma_2$ is 0-dimensional then $$ I_2 = \{ f \in V \otimes V : f|_{\Gamma_{2}} = 0 \}.$$
Remarkably, this result does not require any hypotheses other than a specified number of generators and relations. Applied to our results it implies that the Sklyanin twist $A(\alpha,\beta,\gamma)^{G,\mu}$ is determined up to isomorphism by the 20 points in its point scheme and the associated automorphism, which together describe $\Gamma_2$ for that algebra. We show in Theorem \[thm: new\] that $A(\alpha,\beta,\gamma)^{G,\mu}$ is not isomorphic to any of the existing examples in the literature of AS-regular algebras of dimension 4 with 20 point modules.
A similar result to Theorem \[thm: pointschemereconstruct\] was proved for the line scheme of an AS-regular algebra of dimension 4 in [@shelton2002schemes Theorem 4.3]. When the line scheme of such an algebra has dimension 1 its relations can be reconstructed from those functions vanishing on the line scheme.
Such results highlight the importance of finding examples of algebras that are AS-regular of dimension 4, have a 0-dimensional point scheme and a 1-dimensional line scheme, since information in $\text{qgr}(A)$ essentially determines them. In addition to having a 0-dimensional point scheme, the Sklyanin twist $A(\alpha,\beta,\gamma)^{G,\mu}$ has a 1-dimensional line scheme by Proposition \[prop: linescheme1dim\].
Twists of algebras {#sec: twistsofalgebras}
------------------
In this section we will describe three constructions relating to twisting algebras. The first two, namely cocycle twists and Zhang twists, involve twisting the multiplicative structure in an algebra. These objects are the subject of sections §\[subsec: cocycletwists\] and §\[subsec: zhangtwist\] respectively. Cocycle twists can be formulated as Zhang twists, as shown in Theorem \[thm: cocycleaszhang\]. A partial converse to this result is given by Proposition \[prop: recoverztwist\] in the next chapter, which gives a class of Zhang twists that can be phrased as cocycle twists. The section ends with crossed products being defined in §\[sec: crossedproduct\]. Once again there is some intersection between such objects and the twisting constructions previously introduced.
### Cocycle twists {#subsec: cocycletwists}
Cocycle twists are a classical construction and the subsequent material can be found in many places, such as [@karpilovsky1987schur §2.3]. The idea of twisting algebraic structures has been vastly extended; the recent paper [@bazlov2012cocycle] studies cocycle twists in a monoidal category for example. In this section we expand on the brief introduction to classical cocycle twists given in Construction 1 of §\[subsec: theoryoftwist\].
We will assume until further notice that $k$ is an algebraically closed field and $A$ is an associative $k$-algebra with identity. By $G$ we will denote a finite group, not necessarily abelian, for which $\text{char}(k) \nmid |G|$ and $A$ admits a $G$-graded structure. Recall from Definition \[defn: ggradedalgebra\] that this means that there is a direct sum decomposition of $k$-vector spaces $A=\bigoplus_{g \in G} A_g$, with $A_g A_h \subset A_{gh}$ for all $g, h \in G$ and $1 \in A_e$.
In Definition \[def: normalised2cocycle\] we defined a *normalised 2-cocycle* of $G$. All 2-cocycles that we use will be normalised, therefore we make the following definition for the remainder of the thesis.
\[def: 2cocycle\] A *2-cocycle* of $G$ with values in $k^{\times}$ is a function $\mu: G \times G
\rightarrow k^{\times}$ satisfying the following relations for all $g,h,l \in G$: $$\begin{aligned}
\mu(g,h)\mu(gh,l) &=\mu(g,hl)\mu(h,l),\label{eq: cocycleidinto1}\\
\mu(e,g) =&\mu(g,e)=1.\label{eq: cocycleidinto2}\end{aligned}$$
Let $Z^2(G,k^{\times})$ denote the set of 2-cocycles of $G$ with values in $k^{\times}$. Under pointwise multiplication $Z^2(G,k^{\times})$ forms an abelian group whose identity element is $\mu_e$, where $\mu_e(g,h)=1$ for all $g, h \in G$.
One can deform the $G$-graded multiplication in $A$ using a 2-cocycle as we saw in Construction 1 of §\[subsec: theoryoftwist\]. However, we will now take a more general approach. Let $\mu: G \times G \rightarrow k^{\times}$ be any function satisfying for all $g \in G$. We define a map $\ast_{\mu}: A \times A \rightarrow A$ as follows. First, for homogeneous elements $a \in A_g$ and $b \in A_h$ define $a \ast_{\mu} b := \mu(g,h) ab$, where juxtaposition denotes the original multiplication in $A$. This can then be extended by $k$-linearity to a function on the whole of $A \times A$. One obtains a new algebra structure on $A$ with the same identity element, and we denote this algebra by $(A,\ast_{\mu})$.
The next proposition shows that condition in Definition \[def: 2cocycle\] governs the preservation of associativity under twisting.
\[prop: preserveassociative\] Let $G$ be a finite group and $A$ be a $G$-graded associative $k$-algebra with identity. Consider a function $\mu: G
\times G \rightarrow k^{\times}$ satisfying . Then $(A,\ast_{\mu})$ is a $k$-algebra with the same algebra identity element as $A$. Moreover, $(A,\ast_{\mu})$ is associative if and only if $\mu$ satisfies .
To see that $(A,\ast_{\mu})$ is a $k$-algebra is suffices to check that $k$ is central under $\ast_{\mu}$. Let $\lambda
\in k^{\times}$. The $G$-grading of $A$ gives a decomposition into $k$-vector spaces, hence $k \subset A_e$ must hold. Thus $$\label{eq: algident}
\lambda \ast_{\mu} a = \mu(e,g) \lambda a \;\text{ and }\; a \ast_{\mu} \lambda = \mu(g,e) \lambda a.$$ When holds $\lambda$ must remain central under the new multiplication. Moreover, by taking $\lambda = 1_A$ in it is clear that the algebra identity in $(A,\ast_{\mu})$ is the same as that in $A$.
Now let $a \in A_g$, $b \in A_h$ and $c \in A_l$ for some $g, h, l \in G$. One has $$\begin{aligned}
a \ast_{\mu} (b \ast_{\mu} c) &= a \ast_{\mu} (\mu(h,l) bc) = \mu(h,l)\mu(g,hl) a(bc),\\
(a \ast_{\mu} b) \ast_{\mu} c &= (\mu(g,h) ab) \ast_{\mu} c = \mu(g,h)\mu(gh,l) (ab)c.\end{aligned}$$ Using the associativity of $A$ it is clear that $\ast_{\mu}$ is associative if and only if is satisfied.
We gave the prototypical examples of cocycle twists in Example \[ex: twistedgroupalgebra\], namely *twisted group algebras*. Let us recall their definition.
\[eg: groupalg\] Let $G$ be a finite group and $k$ a field. The *group algebra* $kG$ is a vector space whose basis elements are indexed in a natural way by elements of $G$, that is, $kG= \bigoplus_{g \in G} kg$. The multiplication in $kG$ is given by $$(\alpha g)(\beta h) = \alpha \beta (gh),$$ for all $\alpha,\beta \in k$ and $g,h \in G$.
Now let $A$ be an associative $k$-algebra. One can form a group algebra over $A$ by taking the tensor product of algebras $AG:=A \otimes kG$. The group algebra $AG$ has a $G$-grading given by $AG= \bigoplus_{g \in G} A \otimes g$, and one can twist this grading using a 2-cocycle. For a 2-cocycle $\mu$, the algebra $AG_{\mu}:= (AG,\ast_{\mu})$ is a *twisted group algebra*. The multiplication in $AG_{\mu}$ is given by $$(a \otimes g) \ast_{\mu} (b \otimes h) = \mu(g,h)ab \otimes gh,$$ for all $a, b \in A$ and $g, h \in G$. It is clear that $AG_{\mu} = A \otimes kG_{\mu}$.
A natural question to ask is the following: for which 2-cocycles is $AG_{\mu}$ isomorphic to $AG$? In order to answer this question let us assume that $A=k$. Consider a 2-cocycle $\mu$ for which there exists a function $\rho: G
\rightarrow k^{\times}$ such that for all $g, h \in G$, $$\label{eq: trivialcocycle}
\mu(g,h)=\rho(g)\rho(h)\rho(gh)^{-1}.$$
Let $B^2(G,k^{\times})$ denote the subgroup of $Z^2(G,k^{\times})$ consisting of all 2-cocycles satisfying . This is the set of *2-coboundaries* of $G$ with coefficients in $k^{\times}$. If two 2-cocycles lie in the same coset modulo the coboundaries then we say they are *cohomologous*.
The next result shows that 2-coboundaries control isomorphisms between twisted group algebras. Although it is stated in [@karpilovsky1987schur] for $k = {\ensuremath{\mathbb{C}}}$, the proof can be adapted to work when $k$ is only assumed to be algebraically closed.
\[thm: coboundary\] The twisted group algebras $kG_{\mu}$ and $kG_{\phi}$ are isomorphic as $G$-graded algebras if and only if $\mu$ and $\phi$ are cohomologous. In particular, if $\mu$ is a coboundary then $kG_{\mu} \cong kG$.
By Theorem \[thm: coboundary\] the abelian group $Z^2(G,k^{\times})/B^2(G,k^{\times})$, which we denote by $H^2(G,k^{\times})$, can be viewed as an object that parameterises isomorphism classes of $G$-graded deformations of $kG$. In fact, when $k = {\ensuremath{\mathbb{C}}}$ this quotient group has a special name.
\[defn: schurmult\] The *Schur multiplier* $M(G)$ of a finite group $G$ is the quotient group $H^2(G,{\ensuremath{\mathbb{C}}}^{\times})$.
\[rem: otherschurmult\] The Schur multiplier of a group $G$ is also related to central extensions and projective representations of $G$, see [@karpilovsky1987schur Theorem 1.4.1] for example.
The Schur multiplier of a group $G$ — and more generally $H^2(G,k^{\times})$ — can be interpreted in terms of group cohomology as in [@karpilovsky1987schur §1.3]. One can use this approach to show that cyclic groups have no nontrivial 2-cocycles in the sense that $H^2(G,k^{\times})$ is isomorphic to the trivial group; for $k = {\ensuremath{\mathbb{C}}}$ this is Proposition 2.1.1 op. cit.. It then follows from the next proposition that there are no nontrivial twists for cyclic groups. The result is standard and is proved in a similar manner to one direction of Theorem \[thm: coboundary\], however we have been unable to find a reference for the form as we state it.
\[prop: trivialtwist\] Let $G$ be a finite group and $A$ a $G$-graded $k$-algebra. For a 2-cocycle $\mu$ that is cohomologous to the trivial 2-cocycle one has an isomorphism of $k$-algebras $(A,\cdot) \cong (A,\ast_{\mu})$, where $\cdot$ denotes the original multiplicative structure of $A$.
By assumption there exists some $\rho: G \rightarrow k^{\times}$ such that $$\mu(g,h)= \rho(g)\rho(h)\rho(gh)^{-1},$$ for all $g, h \in G$, as in . Define a map $\varphi:(A,\cdot) \rightarrow (A,\ast_{\mu})$ by $\varphi(a)=\rho(g)^{-1}a$ for all $a \in A_g$ and extending by $k$-linearity. For homogeneous elements $a \in A_g$ and $b \in A_h$ one has $$\varphi(a) \ast_{\mu} \varphi(b)= \mu(g,h)\rho(g)^{-1}\rho(h)^{-1} ab = \rho(gh)^{-1} ab = \varphi(ab).$$ Thus $\varphi$ is a $k$-algebra homomorphism, and since the two algebras have the same underlying vector space structure it must in fact be an isomorphism.
Another nice consequence of knowing that 2-cocycles over cyclic groups are trivial is given by the following result of Yamazaki, given as an exercise in [@karpilovsky1987schur].
\[prop: yamazaki\] Let $P$, $G_1$ and $G_2$ be abelian groups. Then $$H^2(G_1 \times G_2, P) \cong H^2(G_1, P) \times H^2(G_2, P) \times \text{Hom}_{\text{grp}}(G_1 \otimes_{{\ensuremath{\mathbb{Z}}}} G_2, P)$$
One can use Proposition \[prop: yamazaki\] to calculate $H^2(G,k^{\times})$ for any finite abelian group. In particular, the proposition implies that the Klein-four group $G=(C_2)^2$ is the smallest group for which $H^2(G,k^{\times})$ is nontrivial. Our examples in Chapter \[chap: sklyanin\] show that many interesting phenomena occur for twists involving 2-cocycles over this group.
### Zhang twists {#subsec: zhangtwist}
We now move on to considering Zhang twists, a thorough treatment of which is given in [@zhang1998twisted]. Throughout §\[subsec: zhangtwist\] we will retain the assumptions of §\[subsec: cocycletwists\], except that $G$ can now be any semigroup, primarily to allow for the case $G={\ensuremath{\mathbb{N}}}$.
We will work with $G$-graded $k$-linear automorphisms of $A$. These are $k$-linear maps from $A$ to itself such that the restriction to any graded piece $A_g$ is a vector space isomorphism. As Zhang notes prior to [@zhang1998twisted Definition 2.1], $G$-graded $k$-linear automorphisms are not necessarily algebra automorphisms, although in practice twists arising from algebra automorphisms are more commonly studied (see Example \[eg: autzhangtwist\]).
The main ingredient of a Zhang twist is the following.
\[defn: twistingsystem\] A set $\tau=\{\tau_g : g \in G \}$ of $G$-graded $k$-linear automorphisms of $A$ is called a *twisting system* of $A$ if $$\label{eq: twistingsystem}
\tau_g(y\tau_h(z))=\tau_g(y)\tau_{gh}(z),$$ for all $g,h,l \in G$ and all $y \in A_h$, $z \in A_l$.
We proceed by giving an example of perhaps the simplest manner in which a nontrivial twisting system can arise.
\[eg: autzhangtwist\] Suppose that $A$ is a c.g. algebra and $f$ is an ${\ensuremath{\mathbb{N}}}$-graded algebra automorphism of $A$. Then $\{f^n : n \in {\ensuremath{\mathbb{N}}}\}$ is a twisting system of $A$. This is an example of an *algebraic* twisting system since it arises from a semigroup homomorphism $({\ensuremath{\mathbb{N}}},+) \rightarrow \text{Aut}_{{\ensuremath{\mathbb{N}}}-\text{alg}}(A)$ where $n \mapsto f^n$.
As for 2-cocycles, a twisting system can be used to define a new multiplication on the underlying $G$-graded vector space structure of $A$. As the next proposition demonstrates, the condition in means that associativity is preserved by the new multiplication.
\[prop: zhangtwist\] Let $A$ be a $G$-graded algebra and $\tau$ a twisting system. Then there is a new $G$-graded, associative multiplication $\ast_{\tau}$ on the underlying $G$-graded $k$-vector space $A=\bigoplus_{g \in G} A_g$, defined by $$x \ast_{\tau} y := x \tau_{g}(y),$$ for all $x \in A_g$ and $y \in A_h$. The element $1_{\tau}=\tau_{e}^{-1}(1)$ is the identity element with respect to $\ast_{\tau}$.
The new algebra we have defined is called the *Zhang twist of $A$ by $\tau$* and is denoted by $A^{G,\tau}$. The apparent conflict with this notation also being using in the previous section will be explained shortly. By [@zhang1998twisted Proposition 2.4] we can assume without loss of generality that the identity element in an algebra is preserved under a Zhang twist.
Given a $G$-graded right $A$-module there is a natural way to construct a $G$-graded right $A^{G,\tau}$-module from it.
\[defn: ztwistmodule\] Let $A$ be a $G$-graded algebra and $\tau$ a twisting system with $A^{G,\tau}$ the associated Zhang twist. If $M=
\bigoplus_{g \in G} M_g$ is a $G$-graded right $A$-module then there is a graded right $A^{G,\tau}$-module structure on the underlying $G$-graded $k$-vector space structure of $M$, defined by $$m \ast_{\tau} z:= m \tau_h(z),$$ for all $m \in M_h$ and $z \in A_l$. The graded right $A^{G,\tau}$-module $(M,\ast_{\tau})$ is called a *twist of the module $M$ by $\tau$*, and is denoted by $M^{\tau}$.
Before stating some of the main results of Zhang’s paper, we make the following remark. In the context of noncommutative algebraic geometry the $G$-grading used when twisting usually comes from the semigroup $G={\ensuremath{\mathbb{N}}}$, arising from the underlying connected graded structure of the algebra. Although Zhang twists preserve several properties for gradings by general semigroups, when $G={\ensuremath{\mathbb{N}}}$ stronger results can be proved. One such result is the following Morita-type theorem.
\[thm: ztwistgmodequiv\] Let $A$ and $B$ be two $G$-graded algebras where $G$ is a semigroup. If $B$ is a Zhang twist of the $G$-grading on $A$ then the categories $\text{GrMod}_{G}(A)$ and $\text{GrMod}_{G}(B)$ are equivalent.
When such algebras are connected graded and $A_1 \neq 0$, this result becomes an if and only if statement by [@zhang1998twisted Theorem 3.5].
We now illustrate how cocycle twists can be formulated as Zhang twists.
\[thm: cocycleaszhang\] Cocycle twists as described in §\[subsec: cocycletwists\] can be formulated as Zhang twists.
Let $A$ be a $G$-graded algebra and $\mu$ be a 2-cocycle. Define a twisting system $\tau=\{\tau_g : g \in G \}$ as follows: for all $g,h \in G$ and $y \in A_h$, let $\tau_{g}(y):=\mu(g,h)y$. Such maps are clearly $k$-linear and one can check that the cocycle condition implies that is satisfied. We claim that there is an algebra isomorphism between the Zhang twist $A^{G,\tau}$ and the cocycle twist $A^{G,\mu}$.
To prove this, consider the map which is the identity on the underlying vector space of $A$. For homogeneous elements $x \in A_g$ and $y \in A_h$ we have $$x \ast_{\tau} y= x\tau_g(y)=\mu(g,h)xy=x \ast_{\mu} y.$$ The multiplication in the two twists is the same and thus the map is an isomorphism. For cocycle twists of $G$-graded algebras one therefore has an equivalence of categories of graded modules by Theorem \[thm: ztwistgmodequiv\].
### Crossed products {#sec: crossedproduct}
In Theorem \[thm: cocycleaszhang\] it was shown that cocycle twists of a $G$-graded algebra can be expressed as Zhang twists. In this short section we will define the notion of a crossed product and see that cocycle twists are related to a special case of their construction. Viewing cocycle twists in this manner will be useful to us in certain circumstances, allowing us to use results in the literature on crossed products.
The definition of a crossed product is as follows.
\[def: crossedproduct\] Let $R$ be a ring and $G$ a group. Let $S \supset R$ be a ring containing a set of units $\overline{G}=\{\overline{g} :
g \in G\}$, isomorphic as a set to $G$. $S$ is said to be a *crossed product of $R$ and $G$* if the following conditions hold:
- $S$ is a free right $R$-module with basis $\overline{G}$ and $\overline{1}_G=1_S$;
- for all $g_1,g_2 \in G$, $\overline{g}_{1} R=R\overline{g}_{1}$ and $\overline{g}_{1} \overline{g}_{2}
R=\overline{g_{1}g_{2}}R$.
Such a ring is often written $S=R \ast G$.
We will see in Chapter \[chap: cocycletwists\] that the group ring $AG$ and the skew group ring $AG_{\mu}$ play a key role in many proofs concerning the cocycle twist $A^{G,\mu}$. Such objects are a special case of crossed products; the free basis indexed by $G$ is central in $AG_{\mu}$ rather than just normal as required by condition (ii) in Definition \[def: crossedproduct\].
Goldie theory {#sec: goldietheory}
-------------
In this section we will describe the results from Goldie theory that will be needed in this thesis.
Artin-Wedderburn theory shows that any semisimple artinian ring is isomorphic to a direct sum of matrix rings over division rings [@mcconnell2001noncommutative Theorems 0.1.10 and 0.1.11]. Goldie theory answers the question of when a ring has a semisimple artinian classical quotient ring. The classical quotient ring of a ring $R$ is the ring obtained by inverting the set of non-zero regular elements in $R$.
The rings to which Goldie’s theory of noncommutative localisation apply are defined as follows.
A ring $R$ is *Goldie* if the following conditions hold for the modules $R_R$ and $_RR$:
- any direct sum of submodules is finite;
- any ascending chain of annihilators is also finite.
A noetherian ring is easily seen to be Goldie, therefore the next result is applicable to the rings that we study in this thesis.
\[thm: goldie\] A ring $R$ has a semisimple artinian classical quotient ring if and only if it is semiprime Goldie. Moreover, $R$ has a simple artinian classical quotient ring if and only if it is prime Goldie.
The rings we study will often be graded, and there is a version of Goldie’s theorem for such rings [@nastasescu1982graded Theorem C.I.1.6]. We will only need the ‘prime’ version of that result, stated as Theorem \[thm: grgoldie\] below. In the theorem the ring of fractions refers to inverting the non-zero homogeneous regular elements. Any property that is prefixed by gr- has the same definition as in the ungraded case but only applied to graded modules.
\[thm: grgoldie\] Let $G$ be an abelian group and $R$ a $G$-graded, gr-prime, gr-Goldie ring. Then $R$ has a gr-simple, gr-artinian ring of fractions, denoted by $Q_{\text{gr}}(R)$. We call $Q_{\text{gr}}(R)$ the *graded quotient ring* of $R$.
We now describe such graded quotient rings explicitly when $G = {\ensuremath{\mathbb{Z}}}$, which also encompasses ${\ensuremath{\mathbb{N}}}$-graded algebras. By [@nastasescu1982graded Corollary A.I.4.3 and Theorem A.I.5.8], such a ring is isomorphic to a ring of the form $M_n(D)[z,z^{-1};\sigma]$, where $D$ is a division ring, $\sigma$ is some automorphism of $D$ and $n \geq 1$. Thus the graded quotient ring of a graded prime ring is a skew polynomial ring over a simple artinian ring.
When $R$ is a domain, $Q_{\text{gr}}(R)=D[z,z^{-1};\sigma]$ for some division ring $D$. This division ring is called the *graded division ring* of $R$. One could choose another element in $Dz$ to be the skew parameter, in which case the corresponding automorphism is obtained from the previous one by a conjugation map. This is demonstrated by the following calculation. For all $d, d' \in D$ one has $$dzd'=d\sigma(d')z=(d\sigma(d')d^{-1})dz.$$
Thus, if one takes $dz$ to be the new skew parameter then $\sigma$ is replaced by $c_{d} \circ \sigma$ in the skew structure, where $c_d: D \rightarrow D$ is defined by $c_d(x)=dxd^{-1}$ for all $x \in D$. In our case $D$ will often be a field, in which case the conjugation is trivial and one can change the skew parameter without changing the associated automorphism.
Cocycle twists of automorphism-induced $G$-gradings {#chap: cocycletwists}
===================================================
In this chapter we will elaborate on Constructions 1 and 2 from §\[subsec: theoryoftwist\] and study them in more detail. These two twisting constructions are shown to be the same in Proposition \[prop: twoconstrequal\]. In §\[subsec: odesskiiegtwist\] we then formulate Odesskii’s example (Example \[ex: odesskii\]) in terms of a cocycle twist. This is followed by §\[subsec: zhangtwistascocycle\], in which we show that some Zhang twists of ${\ensuremath{\mathbb{N}}}$-gradings can be described as cocycle twists. To end §\[sec: construction\] we discuss the effect of group automorphisms and choices of duality on twisting in §\[subsec: twistggrading\].
In §\[sec: preservation\] we show that many properties are preserved under cocycle twists, for example AS-regularity in Corollary \[cor: asreg\]. Our main tool is Proposition \[prop: fflat\], which allows the use of faithful flatness arguments.
The chapter concludes with §\[sec: modules\], in which we study the interplay between 1-critical modules over an algebra $A$ and a cocycle twist $A^{G,\mu}$. Under some hypotheses, our work demonstrates that point modules over $A$ can be used to construct fat point modules of multiplicity 2 over $A^{G,\mu}$ (see Proposition \[prop: fatpoints\]).
Constructions of the twists {#sec: construction}
---------------------------
We begin this section by fixing the base assumptions under which we will work.
\[hyp: generalcase\] Let $A$ be a $k$-algebra where $k$ is an algebraically closed field. Assume that a finite abelian group $G$ acts on $A$ by algebra automorphisms, where $\text{char}(k) \nmid |G|$. Fix an isomorphism between $G$ and its group of characters $G^{\vee}$, mapping $g \mapsto \chi_g$.
Our primary interest in cocycle twists is to apply them to ${\ensuremath{\mathbb{N}}}$-graded algebras. As such, we record the following additional assumptions that will be used when dealing with properties related to ${\ensuremath{\mathbb{N}}}$-graded algebras.
\[hyp: gradedcase\] Further to Hypotheses \[hyp: generalcase\], assume that $A$ is ${\ensuremath{\mathbb{N}}}$-graded and $G$ acts on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorhisms, i.e. $G \rightarrow \text{Aut}_{{\ensuremath{\mathbb{N}}}\text{-alg}}(A)$.
### Two constructions {#subsec: twoconstruct}
The first construction is essentially Odesskii’s twist from [@odesskii2002elliptic] in greater generality. Given a 2-cocycle $\mu \in Z^2(G,k^{\times})$ we can form the twisted group algebra $AG_{\mu}=A \otimes kG_{\mu}$, as defined in Example \[eg: groupalg\]. As noted in §\[sec: crossedproduct\], this algebra has a crossed product structure.
We will define an action of $G$ on $kG_{\mu}$ by $g^{h}:=\chi_g(h)g$ for all $g, h \in G$ and extending $k$-linearly. To see that this is indeed an action by algebra automorphisms, note that $$\label{eq: Gactiongroupalgebra}
g^{hk}=\chi_g(hk)g=\chi_g(h)(\chi_g(k)g)=(g^k)^h,$$ and $$(g \ast_{\mu} h)^k = (\mu(g,h) gh)^k = \mu(g,h)\chi_{gh}(k)gh= \mu(g,h)\chi_{g}(k)\chi_{h}(k)gh = g^k \ast_{\mu} h^k.$$ In we have used the fact that $G$ is abelian.
While this is not the obvious action of $G$ on the twisted group algebra, choosing it will simplify our work. Observe that under this action $kG_{\mu}$ affords the regular representation of $G$, with isotypic components of the form $\left(kG_{\mu}\right)^{\chi_{g}}=kg$. We then define a diagonal action of $G$ on the tensor product $A \otimes
kG_{\mu}$, where $$\label{eq: diagonalaction}
\left(\sum_i a_i \otimes g_i \right)^h=\sum_i a_i^h \otimes g_i^h,$$ for all $a_i \in A$, $g_i, h \in G$. The algebra in which we are interested is the invariant ring under this action, $(A
\otimes kG_{\mu})^G=(AG_{\mu})^G$.
We now describe the second construction, which is a special case of [@montgomery1993hopf §7.5.1] obtained by taking $H=kG$. We will avoid Hopf algebra terminology and give the construction in some detail.
Let $G$ and $A$ be as in Hypotheses \[hyp: generalcase\]. We must first address how a $G$-grading is induced on $A$ by the action of $G$. Maschke’s Theorem [@isaacs1976character Theorem 1.9] tells us that since $A$ is a $kG$-module it is completely reducible and therefore splits into a possibly infinite direct sum of irreducible submodules, $A =
\bigoplus_{i \in I}A_i$. Since $G$ is abelian, each submodule $A_i$ is 1-dimensional, thus $A_i=k a_i$ for some $a_i \in
A$.
Having discussed these preliminaries we may now define the $G$-grading on $A$ that is induced by the action of $G$.
\[lem: ggrading\] Define $A_g:=A^{\chi_{g^{-1}}}$ for all $g\in G$, where $A^{\chi_{g^{-1}}}$ is the isotypic component of $A$ corresponding to the character $\chi_{g^{-1}}$. Then $A= \bigoplus_{g \in G}A_g$ defines a $G$-grading on $A$.
Since $G$ acts by $k$-algebra automorphisms one clearly has $1 \in A_e$. To complete the proof, consider homogeneous elements $a \in A_{g_{1}}, b \in A_{g_{2}}$ and $h \in G$, and calculate $$(ab)^h=a^hb^h=\chi_{g_{1}^{-1}}(h)a \chi_{g_{2}^{-1}}(h)b=\chi_{(g_1g_2)^{-1}}(h)ab,$$ which implies that $ab \in A_{g_{1}g_{2}}$. Thus $A_{g_{1}} \cdot A_{g_{2}} \subset A_{g_{1} g_{2}}$ for all $g_1, g_2
\in G$.
We denote the cocycle twist of $A$ by $\mu$ under the induced grading by $A^{G,\mu}$, as in §\[subsec: cocycletwists\].
The two constructions we have given are in fact the same, as the next proposition shows. We remark that this fact was also noted more generally in [@bazlov2012cocycle §3.4] via the coaction of a group algebra on another algebra.
\[prop: twoconstrequal\] If the same isomorphism $G \cong G^{\vee}$ is used in both constructions, then $A^{G,\mu} \cong (AG_{\mu})^G$ as $k$-algebras.
Consider the $G$-grading on $A$ defined in Lemma \[lem: ggrading\]. Since $A^{G,\mu}$ is a deformation of the $G$-graded multiplication on $A$, it is clear that it has the same $G$-grading on the underlying vector space structure it shares with $A$. We define a $k$-algebra homomorphism on $G$-homogeneous elements by $$\label{eq: isobetweentwoconstrs}
\phi: A^{G,\mu} \rightarrow (AG_{\mu})^G,\;\;\; a \in A_g \mapsto a \otimes g,$$ and extend by $k$-linearity. To see that this is well-defined, observe that for all $g,h \in G$ one has $$A^{\chi_{g}} \otimes \left(kG_{\mu}\right)^{\chi_{h}}=\left(A \otimes kG_{\mu} \right)^{\chi_{gh}}$$ under the diagonal action of $G$. Since in one has $a \in A_g=A^{\chi_{g^{-1}}}$ and $g
\in \left(kG_{\mu}\right)^{\chi_{g}}$, the element $a \otimes g $ is indeed invariant in $AG_{\mu}$.
We now check that this map is also a $k$-algebra homomorphism. It is enough to check for homogeneous elements with respect to the $G$-grading, therefore suppose that $a \in A_g$ and $b \in A_h$. Then $$\phi(a) \phi(b) = (a \otimes g)(b \otimes h) =ab \otimes (g \ast_{\mu} h) =
\mu(g,h) ab \otimes gh =\phi(\mu(g,h)ab) = \phi(a \ast_{\mu} b).$$ Define a map $\psi: (AG_{\mu})^G \rightarrow A^{G,\mu}$ by $\psi(a \otimes g)=a$ on pure tensors (where $a \in A_{g}$ by definition of the grading). This shows that $\phi$ must be an isomorphism.
We end this section with a remark.
\[rem: diagrelns\] The decomposition of $A$ into 1-dimensional $kG$-modules makes it clear that one can always choose a vector space basis of $A$ such that $G$ acts diagonally on it. We will assume from now on that the algebra structure is defined in terms of such a basis, and in particular that the relations of the algebra are written using it. The action of $G$ must preserve the relations of $A$, therefore in a similar manner one can choose a basis for the relations on which $G$ acts diagonally – we will assume that this holds as well.
### Odesskii’s example as a cocycle twist {#subsec: odesskiiegtwist}
Let us recall Example \[ex: odesskii\]. For a specific choice of algebra, this example encompasses Odesskii’s example from [@odesskii2002elliptic], and it is this situation on which we wish to focus.
Let $G$ be the Klein-four group $(C_2)^2= \langle g_1, g_2 \rangle$ and let $A$ be a 4-dimensional Sklyanin algebra over $k = {\ensuremath{\mathbb{C}}}$. Consider the action of $G$ on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms as described on a set of algebra generators for $A$ in . We note that the algebra generators appearing there are not those used in Definition \[def: sklyanin relations\].
To satisfy Hypotheses \[hyp: gradedcase\] we must fix an isomorphism between $G$ and $G^{\vee}$. The character table of $G$ in below is labelled to reflect such a choice. $$\label{eq: chartable}
\begin{array}{c|cccc}
& e & g_1 & g_2 & g_1 g_2 \\
\hline
\chi_e & 1 & 1 & 1 & 1 \\
\chi_{g_{1}} & 1 & 1 & -1 & -1 \\
\chi_{g_{2}} & 1 & -1 & 1 & -1 \\
\chi_{g_{1}g_{2}} & 1 & -1 & -1 & 1 \end{array}$$
The next lemma is a key result, not only in relation to Odesskii’s example, but also for our work in §\[sec: modules\].
\[lem: kgmuiso\] Let $k$ be an algebraically closed field with $\text{char}(k)\neq 2$. Consider the 2-cocycle $\mu$ defined by $$\label{eq: mucocycledefn}
\mu(g_1^p g_2^q, g_1^r g_2^s) := (-1)^{ps}\; \text{ for all }p, q, r, s \in \{ 0 , 1 \},$$ the action of $G$ on $kG_{\mu}$ defined in §\[subsec: twoconstruct\] and that on $M_2(k)$ described in . Then there exists an isomorphism of $k$-algebras $\phi: kG_{\mu} \rightarrow M_2(k)$ such that $$\label{eq: preserveisocomps}
\phi((kG_{\mu})^{\chi_{g}}) = M_2(k)^{\chi_{g}},$$ for all $g \in G$.
Consider the map $\phi: kG_{\mu} \rightarrow M_2(k)$ which sends $$\label{eq: diagmat}
e \mapsto {\begin{pmatrix} 1 & 0 \\ 0 & 1 \end{pmatrix}},\; g_{1} \mapsto {\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}},\; g_{2} \mapsto {\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}},\; g_{1}g_{2}
\mapsto {\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}}.$$
The multiplication table for the standard vector space basis of $kG_{\mu}$ is $$\label{eq: multtable}
\begin{array}{c|cccc}
kG_{\mu} & e & g_{1} & g_{2} & g_{1}g_{2} \\
\hline
e & e & g_{1} & g_{2} & g_{1}g_{2} \\
g_{1} & g_{1} & e & -g_{1}g_{2} & -g_{2} \\
g_{2} & g_{2} & g_{1}g_{2} & e & g_{1} \\
g_{1}g_{2} & g_{1}g_{2} & g_{2} & -g_{1} & -e \end{array}$$ One can use to show that $\phi$ is a $k$-algebra homomorphism. We give an example of the calculations needed: $$\phi(g_1 \ast_{\mu} g_2)= \phi(-g_1g_2)= {\begin{pmatrix} 0 & 1 \\ -1 & 0 \end{pmatrix}} ={\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}}{\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}} =
\phi(g_1)\phi(g_2).$$ The remaining verifications are similar and we omit them. Since the two algebras are 4-dimensional over $k$ and the matrices in are linearly independent, $\phi$ must be an isomorphism.
Recall the action of $G$ on $kG_{\mu}$, which was defined by $g^{h}:=\chi_g(h)g$ for all $g, h \in G$. Under this action the isotypic component of $kG_{\mu}$ corresponding to $\chi_g$ is spanned by $g$. Thus, to complete the proof we must check that $\phi(g)$ spans $M_2(k)^{\chi_{g}}$ for all $g \in G$. Once again, we omit some of the necessary calculations but give an example to show how to fill in the details.
To see that $\phi(g_1)$ belongs to $M_2(k)^{\chi_{g_{1}}}$, note that $$\begin{aligned}
\begin{pmatrix}
1 & 0 \\
0 & -1
\end{pmatrix}^{g_{1}} &=\begin{pmatrix}
-1 & 0 \\
0 & 1
\end{pmatrix}
\begin{pmatrix}
1 & 0 \\
0 & -1
\end{pmatrix} \begin{pmatrix}
-1 & 0 \\
0 & 1
\end{pmatrix} = \begin{pmatrix}
1 & 0 \\
0 & -1
\end{pmatrix}= \phi(g_1), \\
\begin{pmatrix}
1 & 0 \\
0 & -1
\end{pmatrix}^{g_{2}} &= \begin{pmatrix}
0 & 1 \\
1 & 0
\end{pmatrix} \begin{pmatrix}
1 & 0 \\
0 & -1
\end{pmatrix} \begin{pmatrix}
0 & 1 \\
1 & 0
\end{pmatrix} = \begin{pmatrix}
-1 & 0 \\
0 & 1
\end{pmatrix}= -\phi(g_1).\end{aligned}$$ Thus $g_1$ fixes $\phi(g_1)$ while the other two non-identity elements of $G$ act by the scalar -1. But this implies precisely what we wanted to show: $\phi(g_1)$ belongs to $M_2(k)^{\chi_{g_{1}}}$. Repeating such calculations for each element of the group implies that $\phi$ respects the isotypic decompositions of $kG_{\mu}$ and $M_2(k)$.
\[rem: cocycle\] The 2-cocycle $\mu$ used in Lemma \[lem: kgmuiso\] is cohomologous to the one which will be used in Lemma \[subsec: hbergcentext\]. Although we did not prove that $\mu$ is a 2-cocycle, one can adapt the proof of Lemma \[subsec: hbergcentext\] easily.
We can now prove the main result in this section.
\[prop: odesskiieg\] Consider Odesskii’s example from [@odesskii2002elliptic Introduction]: $A$ is a 4-dimensional Sklyanin algebra for which the Klein four-group $G$ acts on $M_2(A)$ by algebra automorphisms. The invariant ring under this action, $M_2(A)^G$, can be expressed as a cocycle twist $A^{G,\mu}$ for some 2-cocycle $\mu$.
As discussed at the beginning of §\[subsec: odesskiiegtwist\], the example from [@odesskii2002elliptic Introduction] is described in Example \[ex: odesskii\] when $A$ is chosen to be a 4-dimensional Sklyanin algebra. Under this assumption, we wish to study the algebra $(A \otimes_{k} M_2(k))^G$, where $G$ acts diagonally on the tensor product.
By Lemma \[lem: kgmuiso\], $kG_{\mu} \cong M_2(k)$ as a $k$-algebra, and moreover this isomorphism respects the action of $G$ on these algebras; the action on $kG_{\mu}$ is that defined in §\[subsec: twoconstruct\], while the action on $M_2(k)$ is defined in . The result follows from these observations upon considering the invariant construction of cocycle twists from §\[subsec: twoconstruct\].
### Zhang twists as cocycle twists {#subsec: zhangtwistascocycle}
We recall the definition of Zhang twists given in Proposition \[prop: zhangtwist\]. It will now be shown that — as a partial converse to Theorem \[thm: cocycleaszhang\] — some Zhang twists can be formulated as cocycle twists.
To do this we require a 2-cocycle of $(C_n)^2$ for any $n\geq 2$, with such 2-cocycles being described in the following lemma. When $n=2$ the 2-cocycle we construct is cohomologous to that appearing in Odesskii’s example in §\[subsec: odesskiiegtwist\].
\[subsec: hbergcentext\] Fix an integer $n \geq 2$ and an algebracially closed field $k$ of characteristic coprime to $n$. For $G=(C_n)^2=\langle
g_1,g_2 \rangle$ and a primitive $n$’th root of unity $\lambda \in k$, the function $$\label{eq: heisenbergcocycle}
\mu(g_1^p g_2^q, g_1^r g_2^s)=\lambda^{qr} \; \text{ for all }p,q,r,s \in \{0,1,\ldots, n-1 \},$$ defines a 2-cocycle over $G$.
We must check the conditions in Definition \[def: 2cocycle\]. First, note that $\mu$ certainly satisfies , thus we only give the necessary calculations to check that is satisfied. Consider elements $g,h,l \in G$ of the form $g=g_1^p g_2^q$, $h=g_1^r g_2^s$ and $l=g_1^t
g_2^u$ for some $p,q,r,s,t,u \in \{0,1,\ldots, n-1\}$. We verify that $$\begin{aligned}
\mu(g,h)\mu(gh,l) &=\lambda^{qr}\mu(g_1^{p+r} g_2^{q+s},g_1^t g_2^u) = \lambda^{qr+qt+st},\\
\mu(h,l)\mu(g,hl) &=\lambda^{st}\mu(g_1^p g_2^q,g_1^{r+t} g_2^{s+u}) = \lambda^{qr+qt+st}.\end{aligned}$$ This means that $\mu$ satisfies and therefore is a 2-cocycle.
We also need the following lemma, which shows that if the action of $G$ on $A$ preserves a grading on $A$ unrelated to the induced $G$-grading, then the twist $A^{G,\mu}$ will also possess the additional grading. This result will be useful to us primarily when the extra grading is an [$\mathbb{N}$]{}-grading, although it will also be used in §\[subsec: rogzhangalgebras\] in relation to ${\ensuremath{\mathbb{Z}}}^{2}$-gradings.
\[lem: autpresgrad\] Suppose that $A$ has a $H$-grading for some group $H$ and that a finite abelian group $G$ acts on $A$ by $H$-graded algebra automorphisms. Then any cocycle twist $A^{G,\mu}$ will admit the same $H$-grading as $A$ on their shared underlying vector space structure.
We must show that for all $h_1,h_2 \in H$ and homogeneous elements $x \in A_{h_{1}}$ and $y \in A_{h_{2}}$ one has $x
\ast_{\mu} y \in A_{h_{1}h_{2}}$, since then $A^{G,\mu}_{h_{1}} \cdot A^{G,\mu}_{h_{2}} \subset A^{G,\mu}_{h_{1}h_{2}}$. As $G$ acts on $A$ by $H$-graded algebra automorphisms, one can apply Maschke’s theorem to the $H$-graded components of $A$ under the action of $G$. This allows us to further assume that $x$ and $y$ are homogeneous with respect to the $G$-grading, thus $x \in A_{g_{1}}$ and $y \in A_{g_{2}}$ for some $g_1,g_2 \in G$. Then $$x \ast_{\mu} y = \mu(g_1,g_2)xy \in A_{h_{1}h_{2}},$$ which completes the proof.
\[rem: asdfsdaf\] One interpretation of this lemma is that the action of $G$ induces a $(G,H)$-bigrading on $A$; the induced $G$-grading is compatible with the pre-existing $H$-grading. To apply Lemma \[lem: autpresgrad\] under Hypotheses \[hyp: gradedcase\], one can consider an ${\ensuremath{\mathbb{N}}}$-grading as a ${\ensuremath{\mathbb{Z}}}$-grading whose negative components are all zero.
We now assume until the end of §\[subsec: zhangtwistascocycle\] that $A$ is a c.g. and f.g. algebra, although not necessarily in degree 1.
In the following proposition — which is the main result of §\[subsec: zhangtwistascocycle\] — we consider an algebraic twisting system as defined in Example \[eg: autzhangtwist\].
\[prop: recoverztwist\] Let $\phi$ be an ${\ensuremath{\mathbb{N}}}$-graded algebra automorphism of $A$ which has finite order. Denote the associated twisting system by $\tau=\{\tau_n := \phi^n :\;n \in {\ensuremath{\mathbb{N}}}\}$. Then there exists a finite abelian group $G$ and a 2-cocycle $\mu$ for which:
- the group $G$ acts on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms;
- the Zhang twist $A^{{\ensuremath{\mathbb{N}}},\tau}$ is isomorphic to $A^{G,\mu}$ as an $({\ensuremath{\mathbb{N}}},G)$-bigraded $k$-algebra.
Let $G=(C_n)^2=\langle g_1,g_2\rangle$ and $\mu$ be the 2-cocycle of $G$ from Lemma \[subsec: hbergcentext\]. Our aim is to write the Zhang twist as a cocycle twist involving this data.
Fix a primitive $n$’th root of unity $\lambda \in k$. Let $g_1$ act on $A$ by $\phi$ and $g_2$ act on homogeneous elements by scalar multiplication by $\lambda$, thus $a^{g_2}=\lambda^m a$ for all $a \in A_m$. This extends naturally to an action on the whole of $A$. These two actions commute and combine to give a graded action of $G$ on $A$. Once we have chosen an isomorphism $G \cong G^{\vee}$, this action will induce an $({\ensuremath{\mathbb{N}}},G)$-bigrading on $A$ by Lemma \[lem: autpresgrad\] and Remark \[rem: asdfsdaf\]. For $0 \leq i,j \leq n-1$ we will let $\chi_{g_{1}^{i}g_{2}^{j}}$ denote the character for which $\chi_{g_{1}^{i}g_{2}^{j}}(g_1^rg_2^s)=\lambda^{ir+js}$ for all $0 \leq r,s \leq n-1$. It is an easy check to confirm that this defines an isomorphism $G \cong G^{\vee}$.
To prove the result it suffices to show that the multiplications in the two twists are the same on homogeneous elements with respect to the $({\ensuremath{\mathbb{N}}},G)$-bigrading on $A$. Assume therefore that $a \in A_{m_1,g_{1}^{p}g_{2}^{q}}$ and $b \in
A_{m_2,g_{1}^{r}g_{2}^{s}}$ for some $0 \leq p,q,r,s \leq n-1$ and $m_1,m_2 \in {\ensuremath{\mathbb{N}}}$. One can assume that $q=n-m_1$ and $s=n-m_2$ by definition of the action of $g_2$ and the chosen isomorphism $G \cong G^{\vee}$. Under the Zhang twist multiplication one has $$a \ast_{\tau} b=a \tau_{m_{1}}(b)=a\phi^{n-q}(b)=a
b^{g_1^{n-q}}=\chi_{g_{1}^{n-r}g_{2}^{n-s}}(g_1^{n-q})ab=\lambda^{qr}ab.$$
On the other hand, in the cocycle twist we have the multiplication $$a \ast_{\mu} b = \mu(g_1^{p}g_2^{q},g_1^{r}g_2^{s}) ab=\lambda^{qr}ab.$$ Thus $A^{{\ensuremath{\mathbb{N}}},\tau} \cong A^{G,\mu}$ as $({\ensuremath{\mathbb{N}}},G)$-bigraded $k$-algebras.
Proposition \[prop: recoverztwist\] says that $A^{G,\mu}$ is a Zhang twist of both the ${\ensuremath{\mathbb{N}}}$- and $G$-graded structures on $A$. Thus by Theorem \[thm: ztwistgmodequiv\] there are equivalences of categories $$\text{GrMod}(A^{G,\mu}) \simeq \text{GrMod}(A) \text{ and }\text{GrMod}_{G}(A^{G,\mu}) \simeq
\text{GrMod}_{G}(A).$$
Henceforth we will primarily use the notation $A^{G,\mu}$ for a cocycle twist. On occasions we will write $(AG_{\mu})^G$ for a cocycle twist if we wish to emphasise the invariant construction. Having two different constructions of the twist will be advantageous, since in some places it will be preferable to use one instead of the other.
### Twisting the $G$-grading by an automorphism {#subsec: twistggrading}
In this section we illustrate how seemingly different actions can give rise to isomorphic twists.
We begin with a lemma which relates the action of a group automorphism on two different objects; on the graded structure of an algebra and on a 2-cocycle over the group. Such actions are presumably well-known, but we have not been able to find a reference.
\[lem: autoncocycle\] Let $G$ be a finite abelian group and $A = \bigoplus_{g \in G} A_g$ a $G$-graded $k$-algebra. For a group automorphism $\sigma$ one can define a new $G$-grading on $A$ by $A_{\sigma} := \bigoplus_{g \in G} B_g$, where $B_g=A_{\sigma(g)}$. Moreover, for a 2-cocycle $\mu$ the cocycle twist $(A_{\sigma},\ast_{\mu})$ is isomorphic as a $k$-algebra to $(A,\ast_{\mu^{\left(\sigma^{-1}\right)}})$, where $\mu^{(\sigma^{-1})}$ is the 2-cocycle defined by $$\label{eq: autactoncocycle}
\mu^{(\sigma^{-1})}(g,h) := \mu(\sigma^{-1}(g),\sigma^{-1}(h))\index{notation}{m@$\mu^{(\sigma^{-1})}$},$$ for all $g, h \in G$.
Since $\sigma$ is a group automorphism it follows that the decomposition $A_{\sigma}= \bigoplus_g B_g$ is a $G$-grading and that $\mu^{(\sigma^{-1})}$ as defined in is a 2-cocycle of $G$.
In the twist $(A_{\sigma},\ast_{\mu})$ consider homogeneous elements $a \in B_g$ and $b \in
B_h$. Under the graded structure in $(A,\ast_{\mu^{\left(\sigma^{-1}\right)}})$ one has $a \in A_{\sigma(g)}$ and $b \in
A_{\sigma(h)}$. Writing the multiplication of $a$ and $b$ in $(A_{\sigma},\ast_{\mu})$ gives $$\label{eq: autactoncocyclemult}
a \ast_{\mu} b = \mu(g,h)ab = \mu^{(\sigma^{-1})}(\sigma(g),\sigma(h))ab.$$ Notice that the right-hand side of is precisely the multiplication $a
\ast_{\mu^{\left(\sigma^{-1}\right)}} b$ in $(A,\ast_{\mu^{\left(\sigma^{-1}\right)}})$. This is sufficient to complete the proof.
We will use Lemma \[lem: autoncocycle\] in combination with Proposition \[prop: trivialtwist\] in §\[subsec: permuteaction\], where cocycle twists related to different gradings are shown to be isomorphic.
We now show that the choice of isomorphism $G \cong G^{\vee}$ is not totally benign, but that its consequences can be explained using Lemma \[lem: autoncocycle\]. Let us use the notation $(A,\phi,\mu)$ for a triple consisting of an algebra, an isomorphism $G \rightarrow G^{\vee}$, and a 2-cocycle respectively. When $G$ acts on $A$ by algebra automorphisms each such triple can be naturally associated to a cocycle twist.
\[prop: benign\] Let $G$ act on $A$ by algebra automorphisms. Let $\phi$ and $\rho$ be isomorphisms $G \cong G^{\vee}$ and $\mu$ be a 2-cocycle. Then there exists an automorphism of $G$, $\tau$ say, such that the cocycle twists corresponding to the triples $(A,\phi,\mu)$ and $(A,\rho,\mu^{(\tau^{-1})})$ are isomorphic as $k$-algebras.
Given $\phi$, we will identify $\rho$ with an automorphism of $G$ as follows. Firstly, there exists an automorphism $\psi: G^{\vee} \rightarrow G^{\vee}$ such that $\phi = \psi \circ \rho$. Suppose we have an element $x \in A_g$, where the grading is determined under the duality given by $\phi$. This means that for all $h \in G$, $$x^h = \phi(g)^{-1}(h)x=\psi(\rho(g))^{-1}(h)x.$$
Since all maps involved are isomorphisms, for all $g \in G$ there exists $k_g \in G$ such that $\psi(\rho(g))=\rho(k_g)$. We claim that the map $\tau:\; g \mapsto k_g$ defines an isomorphism of $G$. To see this, note that for all $g,h \in G$ one has $$\rho(k_{gh})=\psi(\rho(gh))=\psi(\rho(g)) \psi(\rho(h)) =\rho(k_{g})\rho(k_{h}).$$
As $\rho$ and $\psi$ and are isomorphisms, it follows that $\tau$ is a homomorphism and bijective as claimed. Under the duality isomorphism $\rho$ one has $x \in A_{k_{g}}$, since $$\label{eq: dualgrading1}
x^h =\psi(\rho(g))^{-1}(h)x=\rho(k_g)^{-1}(h)x,$$ for all $h \in G$.
We are now in a position to show that for homogeneous elements the two twists give the same multiplication, from which the result follows. Suppose that $x \in A_g$ and $y \in A_h$ for some $g,h \in G$ under the duality given by $\phi$. Thus $x
\ast_{\mu} y=\mu(g,h)xy$ in $(A,\phi,\mu)$. Under the duality given by $\rho$ one has $x \in A_{k_{g}}$ by , and in a similar manner $y \in A_{k_{h}}$. Thus in $(A,\rho,\mu^{(\tau^{-1})})$ the multiplication is $$x \ast_{\mu^{\left(\tau^{-1}\right)}} y = \mu^{(\tau^{-1})}(k_g,k_h)xy=\mu(\tau^{-1}(k_g),\tau^{-1}(k_h))xy=\mu(g,h)xy,$$ which completes the proof.
Preservation of properties {#sec: preservation}
--------------------------
In this section we prove that many properties are preserved by the twists defined in §\[sec: construction\]. A table giving a summary of our results can be found in Appendix \[app: preserve\]. We begin with some basic results and then prove Proposition \[prop: fflat\], which allows us to use faithful flatness arguments; these will be our main tools when dealing with more advanced properties such as AS-regularity in §\[subsec: asregular\].
Most results will use either Hypotheses \[hyp: generalcase\] or Hypotheses \[hyp: gradedcase\], with this being made clear in the statement of each result. Under the former, more general hypotheses, the fact that a cocycle twist can be formulated as a Zhang twist means that certain properties are preserved under twisting, as we saw in Proposition \[prop: propspreservedalready\]. However, the stronger results from [@zhang1998twisted] in this vein are only valid when the grading being twisted is an ${\ensuremath{\mathbb{N}}}$-grading, and this only occurs as a degenerate case for us (see Proposition \[prop: recoverztwist\]).
Our first result is a useful lemma regarding the behaviour of regular and normal elements under a twist. Although this result is not stated explicitly in [@zhang1998twisted], its proof is essentially contained in the proof of Proposition 2.2(1) op. cit.. Nevertheless, since it will be used several times later in the thesis we state it in a manner that will be most useful for our purposes and give a proof.
Before stating it we make the following definition: an element $a \in A$ is *right regular* if whenever $ab=0$ for some $b \in A$, then $b=0$. There is an analogous definition for being left regular, with a *regular element* satisfying both conditions.
\[prop: stillregular\] Assume Hypotheses \[hyp: generalcase\]. Any element $a \in A$ that is homogeneous with respect to the $G$-grading is regular (normal) in $A$ if and only if it is regular (normal) in $A^{G,\mu}$.
Suppose that $a \in A_g$ is regular in $A$, but not in $A^{G,\mu}$. There must exist some $b \in A^{G,\mu}$ such that $a
\ast_{\mu} b=0$. We can further assume that $b$ is homogeneous with respect to the $G$-grading on $A$. However, this implies that $\mu(g,h)ab = 0$ for some $h \in G$ by definition of the new multiplication, which contradicts the regularity of $a$ in $A$. The proof of the other direction is identical.
An element $a \in A_g$ is normal if and only if for all $G$-homogeneous elements $b \in A_h$ we have $ba=ab'$ for some $b' \in A_h$. The latter statement is equivalent to $b \ast_{\mu} a = \frac{\mu(h,g)}{\mu(g,h)} a \ast_{\mu} b'$ in the twist, i.e. $a$ is normal with respect to the new multiplication also.
We now prove a lemma concerning the behaviour of $G$-graded ideals under a cocycle twist.
\[lem: defrelns\] Assume Hypotheses \[hyp: generalcase\]. Let $I$ be a $G$-graded ideal of $A$, with a homogeneous generating set $\{f_i\}_{i \in I}$. Then $I$ remains an ideal in $A^{G,\mu}$, and moreover it is still generated by $\{f_i\}_{i \in I}$ with respect to the new multiplication.
That $I$ is still an ideal in the twist is proved in [@montgomery2005algebra Proposition 3.1(2)]. To complete the proof it suffices to deal with the case that $I=(f)$ for some homogeneous element $f \in A_g$. Suppose that $a \in A$ with homogeneous decomposition $a = \sum_{h \in G} a_h$. One has $$fa = f \ast_{\mu} \left(\sum_{h \in G} \frac{a_h}{\mu(g,h)} \right)\; \text{ and }\; af = \left(\sum_{h \in G}
\frac{a_h}{\mu(h,g)} \right) \ast_{\mu} f,$$ which proves the result.
\[rem: defrelns\] It is clear from the proof of Lemma \[lem: defrelns\] that the statement is also true for one-sided ideals.
Our next result is related to a remark in [@odesskii2002elliptic] in which Odesskii asserts that the algebra in his example — which we studied in §\[subsec: odesskiiegtwist\] — has the same Hilbert series as the 4-dimensional Sklyanin algebra of which it is a twist. We state our result for general cocycle twists.
\[lem: hilbseries\] Assume Hypotheses \[hyp: gradedcase\]. Then the Hilbert series of $A$ is preserved under twisting, that is $H_A(t)=H_{A^{G,\mu}}(t)$.
By Lemma \[lem: autpresgrad\], $A^{G,\mu}$ possesses the same ${\ensuremath{\mathbb{N}}}$-graded structure as $A$, thus the dimensions of the graded components are the same for both.
The following result of Montgomery is particularly useful in the graded context of our examples, although it holds under more general hypotheses.
\[lemma: finitelygenerated\] Assume Hypotheses \[hyp: generalcase\]. Then $A$ is f.g. as a $k$-algebra if and only if $A^{G,\mu}$ is too.
\[rem: montfingenremark\] By consulting the proof of this result in [@montgomery2005algebra], one can see that a generating set for $A^{G,\mu}$ can be obtained as follows: take a generating set of $A$ and find a vector space $V$ which contains this generating set and is preserved by the action of $G$. Then $A^{G,\mu}$ will be generated by $V$ under the new multiplication on the shared underlying vector space.
In most of our examples we will assume that, in addition to Hypotheses \[hyp: gradedcase\], $A$ is a c.g.algebra that is generated in degree 1. In that case we can conclude using Lemma \[lemma: finitelygenerated\] that $A^{G,\mu}$ is also generated in degree 1. A basis of $V$ which generates both $A$ and $A^{G,\mu}$ as algebras (under their respective multiplications) and on which $G$ acts diagonally will be called a *diagonal basis*.
In the next proposition — which will be crucial for our subsequent results — it will be useful to consider the invariant construction of the twist. While its consequences regarding faithful flatness are used in many results, its information on bimodule structures is particularly useful in relation to global dimension (see Proposition \[prop: gldim\]).
Before stating the result, we recall the concept of twisting a module by an automorphism. Let $A$ be an algebra and $\phi$ be an algebra automorphism. For a right $A$-module $M$, one can define a new right $A$-module $M^{\phi}$ through the multiplication $m \ast_{\phi} a=m\phi(a)$ for all $a \in A$, $m \in M$. This can be recovered from Definition \[defn: ztwistmodule\] by using a trivial grading.
One can twist both sides of an $(A,A)$-bimodule in this manner simultaneously. Suppose that $A$ is graded and $_{A}M_{A}$ is a graded bimodule that is free of rank 1 on both sides. By considering the action of homogeneous elements on the left free generator $m$, we can see that $a\cdot m=m \cdot \phi(a)$ for all $a \in A_n$, where $\phi$ is some ${\ensuremath{\mathbb{N}}}$-graded algebra automorphism. Such a bimodule therefore has the form ${^{\text{id}}}A^{\phi}$, that is, an $(A,A)$-bimodule that is twisted on one side.
\[prop: fflat\] Assume Hypotheses \[hyp: generalcase\]. As an $(A^{G,\mu},A^{G,\mu})$-bimodule there is a decomposition $$\label{eq: decompbimod}
AG_{\mu} \cong \bigoplus_{g \in G} {^{\text{id}}(A^{G,\mu})^{\phi_{g}}},$$ for some automorphisms $\phi_{g}$ of $A^{G,\mu}$, with $\phi_e=\text{id}$. Each summand is free of rank 1 as a left and right $A^{G,\mu}$-module. Consequently, $AG_{\mu}$ is a faithfully flat extension of $A^{G,\mu}$ on both the left and the right. Similarly, $_A(AG_{\mu})$ and $(AG_{\mu})_A$ are free modules of finite rank, thus $AG_{\mu}$ is a faithfully flat extension of $A$ on both the left and the right.
We will proceed as in the proof of the main theorem of [@smith1989can]. Let $AG_{\mu}=\bigoplus_{g \in G}
M^{\chi_{g}}$ be the isotypic decomposition of $AG_{\mu}$ under the action of $G$. Observe that $A^{G,\mu}=M^{\chi_{e}}$ and $M^{\chi_{g}}M^{\chi_{h}}=M^{\chi_{gh}}$ for all $g,h \in G$, since $G$ acts by algebra automorphisms. This means that each isotypic component $M^{\chi_{g}}$ has an $(A^{G,\mu},A^{G,\mu})$-bimodule structure.
The isotypic component $M^{\chi_{g}}$ contains the element $1 \otimes g$. An arbitrary element in this component is a sum of pure tensors of the form $a \otimes h$ for some $a \in A_{g^{-1}h}=A^{\chi_{gh^{-1}}}$. Thus $a \otimes g^{-1}h
\in A^{G,\mu}$ and therefore $$a \otimes h= \left(\frac{a \otimes g^{-1}h}{\mu(g^{-1}h,g)}\right) \cdot (1 \otimes g)=(1 \otimes g) \cdot \left(\frac{a
\otimes g^{-1}h}{\mu(g,g^{-1}h)}\right).$$ Thus $M^{\chi_{g}}$ is cyclic as a left or a right $A^{G,\mu}$-module. Note that $1 \otimes g$ is regular in $AG$, therefore by Lemma \[prop: stillregular\] it is also regular in $AG_{\mu}$. This proves that $M^{\chi_{g}}$ is a free $A^{G,\mu}$-module of rank 1 on both the left and the right.
By the discussion prior to the statement of the proposition, we know that the bimodule generated by $1 \otimes g$ is isomorphic to ${^{\text{id}}}(A^{G,\mu})^{\phi_{g}}$ for some algebra automorphism $\phi_{g}$. To describe $\phi_g$ it suffices to look at the left action of a homogeneous element in $A^{G,\mu}$ on $1 \otimes g$, which can be taken to be a free generator for the left $A^{G,\mu}$-module structure. Consider a homogeneous element $a \otimes h \in A^{G,\mu}_h$. One has $$(a \otimes h) \cdot (1 \otimes g) = \mu(h,g) a \otimes hg = (1 \otimes g) \cdot \frac{\mu(h,g)}{\mu(g,h)}(a \otimes h).$$
Define a map $\phi_g: A^{G,\mu} \rightarrow A^{G,\mu}$ by $a \otimes h \mapsto \frac{\mu(h,g)}{\mu(g,h)}(a \otimes h)$ on homogeneous elements and extending $k$-linearly. To see that this is a $G$-graded automorphism, consider homogeneous elements $a \otimes h \in A^{G,\mu}_h$ and $b \otimes l \in A^{G,\mu}_l$. Then $$\label{eq: phighom}
\phi_g(a \otimes h)\phi_g(b \otimes l) = \frac{\mu(h,g)\mu(l,g)\mu(h,l)}{\mu(g,h)\mu(g,l)}(ab\otimes hl).$$
On the other hand one can use to see that $$\label{eq: phighom1}
\phi_g(\mu(h,l)(ab\otimes hl)) = \frac{\mu(hl,g)\mu(h,l)}{\mu(g,hl)} (ab\otimes hl) =
\frac{\mu(h,lg)\mu(l,g)\mu(h,l)}{\mu(g,h)\mu(gh,l)} (ab\otimes hl).$$ Observe that $\frac{\mu(h,lg)}{\mu(gh,l)} = \frac{\mu(h,g)}{\mu(g,l)}$, which follows from $G$ being abelian together with another use of . Substituting this expression into produces the expression in . It is clear that $\phi_g$ is injective, therefore it must be a $G$-graded automorphism of $A^{G,\mu}$ as claimed.
The result is trivial for $A$ by the definition of $AG_{\mu}$.
We now state the definition of a property which generalises that of being strongly noetherian, which was defined in Definition \[defn: strongnoeth\].
Let $A$ be a noetherian $R$-algebra. $A$ is *universally noetherian* if for any $R$-algebras $R'$ the following conditions are satisfied:
- if $R'$ is right noetherian then so is $A \otimes_R R'$;
- if $R'$ is left noetherian then so is $A \otimes_R R'$.
One can replace the word universal in the following corollary with strongly — or indeed omit it completely — without changing the veracity of the statement. It is a generalisation (in some sense) of [@montgomery2005algebra Proposition 3.1(3)], where the preservation of being noetherian is proved in the Hopf algebra setting (see [@zhang1998twisted Proposition 5.1] also).
\[cor: uninoeth\] Assume Hypotheses \[hyp: generalcase\]. Then $A$ is universally noetherian if and only if $A^{G,\mu}$ is.
We make no distinction between left and right since the proof is identical. Assume that $A$ is universally noetherian. Then $AG_{\mu}$ is a f.g. $A$-module by the proof of Proposition \[prop: fflat\], hence by [@artin1999generic Proposition 4.1(1a)] $AG_{\mu}$ is also universally noetherian. Using Proposition \[prop: fflat\] again, the extension $A^{G,\mu} \hookrightarrow AG_{\mu}$ is faithfully flat on both sides, therefore we can apply [@artin1999generic Proposition 4.1(2a)] to reach the desired conclusion.
In the other direction we can use the same argument but with $AG_{\mu}$ replaced with $A^{G,\mu}G_{\mu^{-1}}$; Proposition \[prop: fflat\] tells us that $A^{G,\mu}G_{\mu^{-1}}$ is a faithfully flat extension of both $A^{G,\mu}$ and of $A$.
### AS-regularity {#subsec: asregular}
The aim of this section is to prove that the AS-regular property is preserved under cocycle twists. As was discussed in §\[sec: noncommgeom\], this property is very important in relation to a geometric theory of noncommutative algebras. Before giving its definition we must first define the AS-Gorenstein property.
\[defn: asgor\] An algebra $A$ is said to be *AS-Gorenstein* (of global dimension $d$) if it is connected graded and satisfies the condition $$\text{Ext}^i_A(k,A)= \left\{ \begin{array}{cl} k & \text{if }i=d, \\ 0 & \text{if }i \neq d, \end{array}\right.$$ when $k$ and $A$ are considered as left ${\ensuremath{\mathbb{N}}}$-graded $A$-modules.
One can calculate this particular Ext group in either $\text{Mod}(A)$ or in $\text{GrMod}(A)$; since $A$ and $k$ are f.g. modules, the discussion in [@levasseur1992some §1.4] shows that the two Ext groups in question are the same.
We can now give the definition of an AS-regular algebra.
\[defn: asregular\] Let $d \in \mathbb{N}$. A connected graded $k$-algebra $A$ is said to be *AS-regular of dimension $d$* if the following conditions are satisfied:
- $A$ has finite GK dimension;
- $\text{gldim }A=d$;
- $A$ is AS-Gorenstein.
\[rem: fgldim\] Note that condition (ii) means that the left and right global dimensions of $A$ agree and equal $d$ (see Definition \[def: homdimfunc\]). The phrase *‘$A$ has finite global dimension’* will therefore be used to encapsulate this.
Under Hypotheses \[hyp: gradedcase\], Proposition \[lem: hilbseries\] implies that $A$ and $A^{G,\mu}$ have the same Hilbert series. In particular, this means that their GK dimensions are equal. Whilst we will mostly be concerned with ${\ensuremath{\mathbb{N}}}$-graded algebras in this section, we can prove that condition (i) of Definition \[defn: asregular\] is preserved in the ungraded case, where one does not have recourse to Hilbert series arguments.
\[prop: gkdim\] Under Hypotheses \[hyp: generalcase\], $\text{GKdim }A= \text{GKdim }A^{G,\mu}$.
By Proposition \[prop: fflat\], $AG_{\mu}$ is a f.g. module over $A$ and $A^{G,\mu}$ on both sides. Applying [@krause2000growth Proposition 5.5] twice, first to $A \subset AG_{\mu}$, then $A^{G,\mu} \subset AG_{\mu}$, proves the result.
Let us move on to global dimension. For the purposes of this section, we only need to show that finite global dimension is preserved under cocycle twists. However, we will prove the more general result that left and right global dimension are preserved, regardless of whether they are equal or not. The algebras we will consider are all noetherian, in which case left and right global dimensions coincide [@mcconnell2001noncommutative §7.1.11].
We will need the following technical result to compare the global dimension of $A^{G,\mu}$ with that of the twisted group algebra $AG_{\mu}$. There is an analogous version for left modules.
\[thm: mcrobtechnical\] Let $R,S$ be rings with $R \subset S$ such that $R$ is an $(R,R)$-bimodule direct summand of $S$. Then $$\text{rgldim }R \leq \text{rgldim }S +\text{pdim }S_R.$$
Without further ado, we show that left and right global dimension are preserved under cocycle twists.
\[prop: gldim\] Assume Hypotheses \[hyp: generalcase\]. One has $\text{rgldim }A=\text{rgldim }A^{G,\mu}$ and $\text{lgldim
}A=\text{lgldim }A^{G,\mu}$.
We will give the proof for right global dimension, from which a left-sided proof can easily be derived. Recall from §\[sec: crossedproduct\] that $AG_{\mu}$ has the structure of a crossed product. We can therefore apply [@mcconnell2001noncommutative Theorem 7.5.6(iii)] to conclude that $\text{rgldim }A=\text{rgldim }AG_{\mu}$. By Proposition \[prop: fflat\] we know that $A^{G,\mu}$ is an $(A^{G,\mu},A^{G,\mu})$-bimodule direct summand of $AG_{\mu}$. We may therefore apply Theorem \[thm: mcrobtechnical\], which tells us that $$\label{eq: gldiminequality}
\text{rgldim }A^{G,\mu} \leq \text{rgldim }AG_{\mu} + \text{pdim }(AG_{\mu})_{A^{G,\mu}}= \text{rgldim }AG_{\mu}.$$
Here $\text{pdim }(AG_{\mu})_{A^{G,\mu}}=0$ since the module is free by Proposition \[prop: fflat\]. We have proved that $\text{rgldim }A^{G,\mu} \leq \text{rgldim }A$.
Now we repeat the argument with the roles of $A$ and $A^{G,\mu}$ reversed, considering them as subrings of $A^{G,\mu}G_{\mu^{-1}}$. One obtains the opposite inequality, which proves the result.
We have addressed two of the conditions necessary to be AS-regular without any assumption regarding an [$\mathbb{N}$]{}-grading. However, we must now assume that $A$ is c.g. and that $G$ acts by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms.
Our main tool for criterion (iii) in Definition \[defn: asregular\] will be a result of Brown and Levasseur for left modules. We also state a version of their result for right modules which can be proved in an analogous manner.
\[prop: brownlevass\] Let $R$ and $S$ be rings and $R \rightarrow S$ a ring homomorphism such that $S$ is flat as a left and right $R$-module.
- Let $X$ be an $(R,R)$-bimodule such that the $(R,S)$-bimodule $X \otimes_R S$ is an $(S,S)$-bimodule. Then for every f.g. left $R$-module $M$ and all $i \geq 0$, there are isomorphisms of right $S$-modules, $$\text{Ext}_R^i(M,X)\otimes_R S \cong \text{Ext}_S^i(S \otimes_R M,X \otimes_R S).$$
- Similarly, suppose that $X$ is an $(R,R)$-bimodule such that the $(S,R)$-bimodule $S \otimes_R X$ is an $(S,S)$-bimodule. Then for every f.g. right $R$-module $M$ and all $i \geq 0$, there are isomorphisms of left $S$-modules, $$S \otimes_R \text{Ext}_R^i(M,X) \cong \text{Ext}_S^i(M \otimes_R S,S \otimes_R X).$$
\[rem: x=s\] For some of our applications of Proposition \[prop: brownlevass\] we will take $R=X$. In that case $X \otimes_R S
\cong S$ as an $(R,S)$-bimodule, from which $X \otimes_R S$ inherits a natural $(S,S)$-bimodule structure.
Before stating the result regarding the AS-Gorenstein property, we need a preliminary lemma.
\[lem: tensor\] In addition to Hypotheses \[hyp: gradedcase\], assume that $A$ is a c.g. algebra. As right $AG_{\mu}$-modules we have isomorphisms $$k \otimes_A AG_{\mu} \cong k \otimes_{A^{G,\mu}} AG_{\mu} \cong kG_{\mu}.$$ As left $AG_{\mu}$-modules we have isomorphisms $$AG_{\mu} \otimes_A k \cong AG_{\mu} \otimes_{A^{G,\mu}} k \cong kG_{\mu}.$$
Under the ${\ensuremath{\mathbb{N}}}$-grading on $AG_{\mu}$ one has $A_{\geq 1} \cdot AG_{\mu} = (AG_{\mu})_{\geq 1}$. Thus $$\begin{gathered}
\begin{align*}k \otimes_A AG_{\mu} \cong \frac{A}{A_{\geq 1}} \otimes_A AG_{\mu} \cong \frac{AG_{\mu}}{A_{\geq 1} \cdot AG_{\mu}} =
\frac{AG_{\mu}}{(AG_{\mu})_{\geq 1}} \cong kG_{\mu}.
\end{align*}\end{gathered}$$
To complete the proof we must first show that the following equality holds: $$\label{eq: genindegreeonecocycletwist}
(A^{G,\mu})_{\geq 1} \cdot AG_{\mu} = (AG_{\mu})_{\geq 1}.$$ Clearly $(A^{G,\mu})_{\geq 1} \cdot AG_{\mu} \subseteq (AG_{\mu})_{\geq 1}$. To prove the opposite inclusion, observe that the action of $G$ on $AG_{\mu}$ in respects its ${\ensuremath{\mathbb{N}}}$-graded structure under our hypotheses. Remark \[rem: asdfsdaf\] then tells us that $AG_{\mu}$ is endowed with an $({\ensuremath{\mathbb{N}}},G)$-bigrading. By linearity it therefore suffices to consider a bihomogeneous element $a \otimes g \in (AG_{\mu})_{(n,h)}$ for some $g,h \in G$ and integer $n \geq 1$. Then $a \otimes gh^{-1} \in (AG_{\mu})_{(n,e)} \subset A^{G,\mu}_{\geq 1}$, and thus $$\frac{1}{\mu(gh^{-1},h)}(a \otimes gh^{-1})(1 \otimes h) = a \otimes g,$$ which proves that $(A^{G,\mu})_{\geq 1} \cdot AG_{\mu} \supseteq (AG_{\mu})_{\geq 1}$. One can then use to see that $$\begin{gathered}
\begin{align*}k \otimes_{A^{G,\mu}} AG_{\mu} \cong \frac{A^{G,\mu}}{(A^{G,\mu})_{\geq 1}} \otimes_{A^{G,\mu}} AG_{\mu} \cong
\frac{AG_{\mu}}{(A^{G,\mu})_{\geq 1}\cdot AG_{\mu}} = \frac{AG_{\mu}}{(AG_{\mu})_{\geq 1}} \cong kG_{\mu}.
\end{align*}\end{gathered}$$
We omit the proof of the statement for left $AG_{\mu}$-modules.
This lemma can be interpreted as saying that under the $({\ensuremath{\mathbb{N}}},G)$-bigrading on $AG_{\mu}$, the subalgebra consisting of elements that have degree zero is the twisted group algebra $kG_{\mu}$.
\[prop: asgor\] Assume in addition to Hypotheses \[hyp: gradedcase\] that $A$ is a c.g. algebra. Then $A$ is AS-Gorenstein of global dimension $d$ if and only if $A^{G,\mu}$ shares this property.
We will give the proof in the only if direction when $k$ and $A$ are considered as left $A$-modules. The proof in the opposite direction is identical by untwisting, while the proof for right modules is almost identical to that below; the only difference is that it requires the use of Proposition \[prop: brownlevass\](ii) rather than part (i) of that result.
[Proposition \[prop: asgor\]]{} We wish to apply Proposition \[prop: brownlevass\](i) with $R=X=A$, $S=AG_{\mu}$ and $M=k$. To see that the hypotheses of that result are satisfied, observe that $A \subset AG_{\mu}$ is flat by Proposition \[prop: fflat\] and recall Remark \[rem: x=s\]. Applying Proposition \[prop: brownlevass\](i) gives $$\label{eq: asgorfirststep}
\text{Ext}^i_A(k,A) \otimes_A AG_{\mu} \cong \text{Ext}^i_{AG_{\mu}}(AG_{\mu} \otimes_A k,A \otimes_A AG_{\mu}) \cong
\text{Ext}^i_{AG_{\mu}}(kG_{\mu},AG_{\mu}),$$ by using Lemma \[lem: tensor\]. Since $A$ is AS-Gorenstein of global dimension $d$ we know that the left hand side is non-zero only for $i=d$. For this value of $i$ it is equal to $k \otimes_A AG_{\mu} \cong kG_{\mu}$ using Lemma \[lem: tensor\] once again.
We would now like to apply Proposition \[prop: brownlevass\](i) a second time, using $R=X=A^{G,\mu}$, $S=AG_{\mu}$ and $M=k$. To see that these data satisfy the hypotheses of that proposition we may apply the same argument as used earlier in the proof, mutatis mutandis. Applying Proposition \[prop: brownlevass\](i) we obtain $$\begin{gathered}
\begin{aligned}\label{eq: asgorsecondstep}
\text{Ext}^i_{A^{G,\mu}}(k,A^{G,\mu}) \otimes_{A^{G,\mu}} AG_{\mu} & \cong \text{Ext}^i_{AG_{\mu}}(AG_{\mu}
\otimes_{A^{G,\mu}} k, A^{G,\mu} \otimes_{A^{G,\mu}} AG_{\mu}) \\
& \cong \text{Ext}^i_{AG_{\mu}}(kG_{\mu}, AG_{\mu}).
\end{aligned}\end{gathered}$$ Combining the information from and gives $$\text{Ext}^i_{A^{G,\mu}}(k,A^{G,\mu}) \otimes_{A^{G,\mu}} AG_{\mu} \cong \left\{ \begin{array}{cl} kG_{\mu} & \text{if
}i=d, \\ 0 & \text{if }i \neq
d. \end{array}\right.$$
Since $A^{G,\mu} \subset AG_{\mu}$ is a faithfully flat extension on the left, $\text{Ext}^i_{A^{G,\mu}}(k,A^{G,\mu})$ must vanish in all degrees for which $i \neq d$. When $i=d$ we have $$\label{eq: asgorfinalstep}
\text{Ext}^d_{A^{G,\mu}}(k,A^{G,\mu}) \otimes_{A^{G,\mu}} AG_{\mu} \cong kG_{\mu},$$ as right $AG_{\mu}$-modules.
Recall from §\[subsec: notation\] that $\text{Ext}^i_{A^{G,\mu}}(k,A^{G,\mu})$ is a ${\ensuremath{\mathbb{Z}}}$-graded group since $k$ and $A^{G,\mu}$ are f.g. ${\ensuremath{\mathbb{N}}}$-graded left $A^{G,\mu}$-modules. This ${\ensuremath{\mathbb{Z}}}$-grading is compatible with the right $A^{G,\mu}$-module structure, in which case the graded module structure on $\text{Ext}^i_{A^{G,\mu}}(k,A^{G,\mu})$ allows us to complete the proof as follows. One may use the $(A^{G,\mu},A^{G,\mu})$-bimodule structure of $AG_{\mu}$ described in Proposition \[prop: fflat\] to see that upon restricting the isomorphism in to $A^{G,\mu}$, one obtains $$\bigoplus_{g\in G} \text{Ext}^d_{A^{G,\mu}}(k,A^{G,\mu})^{\phi_{g}} \cong (kG_{\mu})_{A^{G,\mu}}.$$ By considering the $G$-graded components of this isomorphism and noting that $\phi_e$ is the identity, one obtains the isomorphism of right $A^{G,\mu}$-modules $\text{Ext}^d_{A^{G,\mu}}(k,A^{G,\mu}) \cong k$, which proves the result.
\[cor: asreg\] Assume in addition to Hypotheses \[hyp: gradedcase\] that $A$ is a c.g. algebra. Then $A$ is AS-regular if and only if $A^{G,\mu}$ is. Moreover, if $A$ has global and GK dimension less than or equal to 4, then $A$ is a domain if and only if $A^{G,\mu}$ is a domain.
The statement about AS-regularity follows from Lemma \[lem: hilbseries\] and Propositions \[prop: gldim\] and \[prop: asgor\]. The second part of the corollary follows from [@artin1991modules Theorem 3.9].
\[rem: domain\] Remark \[rem: bgmunilp\] will illustrate that being a domain is *not* preserved in general by cocycle twists. This should be expected since cocycle twists are Zhang twists by Theorem \[thm: cocycleaszhang\]; such twists preserve being a domain when $G$ is an ordered semigroup but not in general (see [@zhang1998twisted §5] and Proposition 5.2 op. cit. in particular). One should contrast this with Lemma \[prop: stillregular\], which shows that regular homogeneous elements remain regular under twisting.
### The Koszul property {#subsec: koszul}
Our next result will be to show that the *Koszul* property is preserved under cocycle twists. This property is often studied in relation to quadratic algebras as defined in Definition \[def: quadalg\], although it can be defined for more general algebras too. Whilst we will cite [@koszul1996beilinson] for any definitions, we suggest [@krahmernotes] as a good introduction to the subject.
Recall from §\[subsec: notation\] that for $M, N \in \text{grmod}(A)$ the cohomology group $\text{Ext}_{A}^i(M,N)$ is ${\ensuremath{\mathbb{Z}}}$-graded. We use this idea to define a Koszul algebra.
\[defn: koszulcomplex\] A c.g. $k$-algebra is *Koszul* if and only if for all $i \geq 0$ the ${\ensuremath{\mathbb{Z}}}$-graded components of $\text{Ext}_A^i(k,k)$ vanish in all degrees other than degree $i$.
\[prop: koszul\] In addition to Hypotheses \[hyp: gradedcase\] assume that $A$ is a quadratic algebra over $k$. Then $A$ is Koszul if and only $A^{G,\mu}$ is.
We wish to apply Proposition \[prop: brownlevass\](i) with $R=A$, $S=AG_{\mu}$, $X={_A}k_A$ and $M={_A}k$. To do so we must check that the hypotheses of that result are satisfied. Observe that $A \subset AG_{\mu}$ is flat by Proposition \[prop: fflat\], while $X \otimes_R S = kG_\mu$ by Lemma \[lem: tensor\], whence it has a natural $(AG_{\mu},AG_{\mu})$-bimodule structure. We may therefore apply Proposition \[prop: brownlevass\](i), in which case one has $$\begin{gathered}
\begin{aligned}\label{eq: koszulbrown}
\text{Ext}_A^i(k,k)\otimes_A AG_{\mu} &\cong \text{Ext}_{AG_{\mu}}^i(AG_{\mu} \otimes_A k,k \otimes_A AG_{\mu}) \\
&\cong \text{Ext}_{AG_{\mu}}^i(kG_{\mu},kG_{\mu}).
\end{aligned}\end{gathered}$$ Note that Lemma \[lem: tensor\] has been used to pass from the first line of to the second.
Now set $R= A^{G,\mu}$, $S=AG_{\mu}$, $X={_{A^{G,\mu}}}k_{A^{G,\mu}}$ and $M={_{A^{G,\mu}}}k$. One can use the same argument as earlier in the proof, mutatis mutandis, to see that these data also satisfy the hypotheses of Proposition \[prop: brownlevass\](i). Applying that result we obtain $$\begin{gathered}
\begin{aligned}\label{eq: koszulbrown1}
\text{Ext}_{A^{G,\mu}}^i(k,k)\otimes_{A^{G,\mu}} AG_{\mu} &\cong \text{Ext}_{AG_{\mu}}^i(AG_{\mu} \otimes_{A^{G,\mu}}
k,k \otimes_{A^{G,\mu}} AG_{\mu}) \\
&\cong \text{Ext}_{AG_{\mu}}^i(kG_{\mu},kG_{\mu}),
\end{aligned}\end{gathered}$$ where we have used Lemma \[lem: tensor\] once again.
The ${\ensuremath{\mathbb{Z}}}$-grading on $\text{Ext}_A^i(k,k)$ and $\text{Ext}_{A^{G,\mu}}^i(k,k)$ is compatible with their right $A$- and $A^{G,\mu}$-module structures respectively. Thus the tensor products $\text{Ext}_A^i(k,k)\otimes_A AG_{\mu}$ and $\text{Ext}_{A^{G,\mu}}^i(k,k)\otimes_{A^{G,\mu}} AG_{\mu}$ are naturally ${\ensuremath{\mathbb{Z}}}$-graded right $AG_{\mu}$-modules. The ${\ensuremath{\mathbb{Z}}}$-grading on the cohomology group $\text{Ext}_{AG_{\mu}}^i(kG_{\mu},kG_{\mu})$ is also compatible with its right $AG_{\mu}$-module structure. Moreover, one can see from the proof of Proposition \[prop: brownlevass\](i) (from [@brown1985cohomology Proposition 1.6]) that the isomorphisms in and respect these ${\ensuremath{\mathbb{Z}}}$-graded structures. We may therefore conclude that there is an isomorphism $$\label{eq: comparedims}
\text{Ext}_A^i(k,k)\otimes_A AG_{\mu} \cong \text{Ext}_{A^{G,\mu}}^i(k,k)\otimes_{A^{G,\mu}} AG_{\mu}$$ of ${\ensuremath{\mathbb{Z}}}$-graded right $AG_{\mu}$-modules.
Using the free module structures of $_A(AG_{\mu})$ and $_{A^{G,\mu}}(AG_{\mu})$ described in Proposition \[prop: fflat\], we may express the isomorphism in as $$\label{eq: comparedims1}
\bigoplus_{|G|}\text{Ext}_A^i(k,k) \cong \bigoplus_{|G|}\text{Ext}_{A^{G,\mu}}^i(k,k),$$ at the level of vector spaces. Furthermore, since $AG_{\mu}$ is an ${\ensuremath{\mathbb{N}}}$-graded left module over $A$ and over $A^{G,\mu}$, the isomorphism in respects the ${\ensuremath{\mathbb{Z}}}$-graded structure.
Since $A$ is Koszul, we know that the ${\ensuremath{\mathbb{Z}}}$-graded components of the left hand side of vanish in all degrees other than degree $i$. It follows that $\text{Ext}_{A^{G,\mu}}^i(k,k)$ must also vanish in all degrees other than degree $i$, hence $A^{G,\mu}$ must be Koszul.
Let us now define the Koszul dual of a c.g. algebra with quadratic relations.
\[def: kdual\] Let $A=T(V)/(R)$ be a quadratic algebra over $k$. The *Koszul dual* of $A$ is the $k$-algebra $A^!:=T(V^{\ast})/(R^{\perp})$. Here $R^{\perp}
\subset V^{\ast} \otimes V^{\ast}$ is the space of functions which vanish on the quadratic relations of $A$.
If $x \in V$ then we will denote by $\overline{x}$ the element in $V^{\ast}$ that vanishes away from the 1-dimensional space spanned by $x$ and for which $\overline{x}(x)=1$.
Our next aim is to show that taking the Koszul dual of an algebra and applying a cocycle twist almost commute with each other. To achieve this aim we first need to define a graded action of $G$ on the Koszul dual given a graded action on $A$. Such an action can be defined on generators by $\overline{x}^g = \overline{x^g}$ for all $g \in G$ and $\overline{x} \in
V^{\ast}$. This action induces a $G$-grading on $A^{!}$ in the manner of §\[subsec: twoconstruct\]; we claim that if $x \in A_g$ then $\overline{x} \in A^{!}_g$. Indeed, observe that for all $h \in G$ we have $\overline{x}^h=
\overline{x^{h}}=\chi_{g^{-1}}(h)\overline{x}$.
We are now in a position to prove the result.
\[prop: koszuldualtwistcommute\] In addition to Hypotheses \[hyp: gradedcase\], assume that $A$ is a quadratic algebra over $k$ that is Koszul. Then there is an isomorphism of $k$-algebras $$\label{eq: kdualcocyclecommute}
(A^{!})^{G,\mu^{-1}} \cong (A^{G,\mu})^{!}.$$
We first fix our notation. Let $V=\text{span}_k(x_0,\ldots,x_{n})$ be a diagonal basis of generators of $A$, where $x_i
\in A_{g_{i}}$ for some $g_i \in G$. We will denote the generators of the other algebras involved by $$(A^!,\overline{x_{i}}),\;\; ((A^!)^{G,\mu^{-1}},y_i),\;\; (A^{G,\mu},v_i),\;\text{ and }\;
((A^{G,\mu})^!,\overline{v_{i}}).$$
Although twists have the same underlying vector space structure, the new generators allow us to write the twisted multiplication by juxtaposition. Thus $$\label{eq: twistedmultx2}
v_iv_j =x_i \ast_{\mu} x_j=\mu(g_i,g_j)x_i x_j\;\text{ and }\; y_iy_j =\overline{x_i} \ast_{\mu^{-1}}
\overline{x_j}=\frac{1}{\mu(g_i,g_j)} \overline{x_i}\overline{x_j}.$$
Consider the map $\phi: (A^{!})^{G,\mu^{-1}} \rightarrow (A^{G,\mu})^{!}$ sending $y_i \mapsto \overline{v_i}$. Note that $$\overline{v_i}\overline{v_j} (v_kv_l) = \left\{
\begin{array}{l l}
1 & \quad \text{if $(k,l)=(i,j)$},\\
0 & \quad \text{otherwise}.
\end{array} \right.$$ One can regard elements of both $(A^{!})^{G,\mu^{-1}}$ and $(A^{G,\mu})^{!}$ as functions on the underlying vector space of $A$ by untangling the twists involved. Using we can interpret $\overline{v_i}\overline{v_j}$ as follows: $$\overline{v_i}\overline{v_j} (x_k x_l) = \left\{
\begin{array}{l l}
\frac{1}{\mu(g_{i},g_{j})} & \quad \text{if $(k,l)=(i,j)$},\\
0 & \quad \text{otherwise}.
\end{array} \right.$$
By one can also conclude that $\overline{v_i}\overline{v_j}=\frac{1}{\mu(g_{i},g_{j})}\overline{x_i}\overline{x_j}=y_i y_j$. This expression implies that the relations in both $(A^{!})^{G,\mu^{-1}}$ and $(A^{G,\mu})^{!}$ are controlled by those in $A^{!}$, therefore $\phi$ is both well-defined and a surjection.
Since $A$ is Koszul, the Hilbert series of $A^!$ depends only on the Hilbert series of $A$; the precise relationship between their Hilbert series was first given in [@froberg1975determination §4, Theorem]. By Proposition \[prop: koszul\] the same is true of the Hilbert series of $(A^{G,\mu})^{!}$ in relation to that of $A^{G,\mu}$. In combination with Lemma \[lem: hilbseries\], this shows that the two algebras in must have the same Hilbert series. Thus the map we have defined is an isomorphism, which completes the proof.
For an application of this proposition, see §\[par: kdual\].
### The Cohen-Macaulay property and Auslander regularity {#subsec: cohenmac}
In this section we will prove that several more homological properties of rings are preserved under cocycle twists. Their definitions can be stated in the graded situation, however we state them — and prove their preservation — in full generality.
The definitions that follow can all be found in [@levasseur1993modules §1.2]. The first concept that we need is the notion of the grade of a module. The *grade* of a f.g. left or right $A$-module $M$ is defined to be the value $$j_A(M)=\text{inf}\{i: \text{Ext}_A^i(M,A)\neq 0\} \in {\ensuremath{\mathbb{N}}}\cup \{+\infty\}.$$
\[def: cm\] A ring $A$ is said to satisfy the *Cohen-Macaulay property* or be *Cohen-Macaulay* if for all non-zero f.g. $A$-modules $M$, one has $$\text{GKdim }M+j_A(M)=\text{GKdim }A.$$
The *Auslander condition* also uses the notion of the grade of a module; it is satisfied by ring $A$ if for every f.g. left or right module $M$, all $i \geq 0$ and every $A$-submodule $N$ of $\text{Ext}^i_A(M,A)$, one has $j_A(N)\geq i$.
\[def: auslanderprops\] A ring $A$ is said to be *Auslander-Gorenstein* if in addition to satisfying the Auslander condition it has finite left and right injective dimension. It is said to be *Auslander regular* if in addition to satisfying the Auslander condition it has finite global dimension.
Before proceeding, recall that Remark \[rem: fgldim\] applies to our usage of the phrase finite global dimension.
The following result shows that the properties defined in Definitions \[def: cm\] and \[def: auslanderprops\] are preserved under a cocycle twist.
\[prop: cohenmac\] In addition to Hypotheses \[hyp: generalcase\], assume that $A$ is noetherian. Then $A$ has one of the following properties if and only if $A^{G,\mu}$ does as well:
- it is Cohen-Macaulay;
- it is Auslander-Gorenstein;
- it is Auslander regular.
\(i) Assume that $A$ is Cohen-Macaulay. We first show that $AG_{\mu}$ shares this property. As we saw in Proposition \[prop: gkdim\], $\text{GKdim }A=\text{GKdim }AG_{\mu}=\text{GKdim }A^{G,\mu}$. Let $M$ be a f.g right $AG_{\mu}$-module. It must also be f.g. as an $A$-module since the extension $A \subset AG_{\mu}$ is finite by Proposition \[prop: fflat\]. By [@ardakov2007primeness Lemma 5.4] it is clear that the grades of $M_{AG_{\mu}}$ and $M_A$ are equal. One can then apply [@lorenz1988on Lemma 1.6] to conclude that $\text{GKdim }M_{AG_{\mu}}=\text{GKdim }M_{A}$. Piecing this together, we find that $$\begin{gathered}
\begin{aligned}\label{eq: cohenaagmu}
\text{GKdim }M_{AG_{\mu}}+ j_{AG_{\mu}}(M)= \text{GKdim }M_A+j_A(M) &=\text{GKdim }A \\ &=\text{GKdim }AG_{\mu},
\end{aligned}\end{gathered}$$ and therefore $AG_{\mu}$ is Cohen-Macaulay.
Now let $M$ be a f.g. right $A^{G,\mu}$-module. By applying Proposition \[prop: brownlevass\](ii) with $R=X=A^{G,\mu}$ and $S=AG_{\mu}$ we obtain $$\label{eq: brownlevasscohenmacappl}
AG_{\mu} \otimes_{A^{G,\mu}} \text{Ext}_{A^{G,\mu}}^i \left(M,A^{G,\mu}\right) \cong \text{Ext}_{AG_{\mu}}^i\left(M
\otimes_{A^{G,\mu}} AG_{\mu},AG_{\mu}\right).$$
When combined with faithful flatness of the extension $A^{G,\mu} \subset AG_{\mu}$ (by Proposition \[prop: fflat\]), this implies that $$j_{A^{G,\mu}}(M)=j_{AG_{\mu}}(M \otimes_{A^{G,\mu}} AG_{\mu}).$$
If we can show that $\text{GKdim }M_{A^{G,\mu}}=\text{GKdim }(M \otimes_{A^{G,\mu}} AG_{\mu})_{AG_{\mu}}$ then the result follows from an equality like that in . By faithful flatness of the extension $A^{G,\mu}
\subset AG_{\mu}$, $M$ is contained in $M \otimes_{A^{G,\mu}} AG_{\mu}$. Therefore by the definition of GK dimension one has $$\label{eq: gkineq1}
\text{GKdim }M_{A^{G,\mu}} \leq \text{GKdim }(M \otimes_{A^{G,\mu}} AG_{\mu})_{A^{G,\mu}}.$$ By [@krause2000growth Proposition 5.6] one has the inequality $$\label{eq: gkineq2}
\text{GKdim }M_{A^{G,\mu}} \geq \text{GKdim }(M \otimes_{A^{G,\mu}} AG_{\mu})_{AG_{\mu}}.$$
Applying [@lorenz1988on Lemma 1.6] to $M \otimes_{A^{G,\mu}} AG_{\mu}$ allows one to combine the inequalities in and , from which the preservation of the Cohen-Macaulay property under cocycle twists follows.
\(ii) Using [@yi1995injective Proposition 3.9(i)] one can see that if $A$ satisfies the Auslander condition then so must $AG_{\mu}$. The twist $A^{G,\mu}$ then satisfies the Auslander condition by [@teo1996homological Theorem 2.2(iv)], since the only hypothesis needed is that the extension be flat – this is true by Proposition \[prop: fflat\].
It remains to show that finite left and right injective dimension are preserved. Consider the $G$-grading on $AG_{\mu}$ for which $(AG_{\mu})_g = A \otimes g$ for all $g \in G$. Under this grading $AG_{\mu}$ is a strongly $G$-graded ring, thus one can apply [@nastasescu1983strongly Corollary 2.7] with $R = N = AG_{\mu}$ and $\sigma = e$. That result implies that $$\text{idim }_{AG_{\mu}}AG_{\mu} = \text{idim }_{(AG_{\mu})_{e}}(AG_{\mu})_{e}= \text{idim }_{A}A.$$ A right-sided analogue of Năstăsescu’s result shows that the same is also true for right injective dimension.
Now consider the $G$-grading on $AG_{\mu}$ under which $(AG_{\mu})_g = A^{G,\mu}(1 \otimes g)$ for all $g \in G$. This $G$-grading is induced by the diagonal action of $G$ on $AG_{\mu}$. It is clear that $AG_{\mu}$ is a strongly $G$-graded ring under this grading as well. One can therefore apply [@nastasescu1983strongly Corollary 2.7] with this new $G$-grading on $R = N = AG_{\mu}$, together with $\sigma = e$, to see that $$\text{idim }_{AG_{\mu}}AG_{\mu} = \text{idim }_{(AG_{\mu})_{e}}(AG_{\mu})_{e}= \text{idim }_{A^{G,\mu}}A^{G,\mu}.$$ An analogous argument, used in conjunction with a right-sided analogue of the corollary of Năstăsescu, proves that $AG_{\mu}$ and $A^{G,\mu}$ also have equal right injective dimensions. This completes the proof.
\(iii) We saw in the proof of (ii) that the Auslander condition is preserved. One can then see from Proposition \[prop: gldim\] that the global dimensions of $A$ and $A^{G,\mu}$ are equal, which completes the proof.
Modules under twisting {#sec: modules}
----------------------
In this section we explore the interplay between modules over $A$ and those over a cocycle twist $A^{G,\mu}$. Our ultimate aim is to apply the results we obtain to the algebras studied in Chapter \[chap: sklyanin\]. Let us define the hypotheses that we will work under for the duration of this section.
\[hyp: genhypforfatpts\] Let $k$ be an algebraically closed with $\text{char}(k)\neq 2$. Assume that $A$ is a $k$-algebra that satisfies the hypotheses of Theorem \[thm: pointschemenice\], with degree 1 generators $x_0, x_1 ,x_2$ and $x_3$. Let $G=\langle g_1,g_2 \rangle$ be the Klein four-group, which acts on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms. Furthermore, assume that the action of $G$ on $A_1$ affords the regular representation, inducing the $G$-grading on generators: $$\label{eq: gengrading}
x_0 \in A_{e},\; x_1 \in A_{g_{1}},\; x_2 \in A_{g_{2}},\; x_3 \in A_{g_{1}g_{2}}.$$ Finally, assume that $\mu$ is the 2-cocycle of $G$ defined by $\mu(g_1^p g_2^q, g_1^r
g_2^s) = (-1)^{ps}$ for all $p, q, r, s \in \{ 0 , 1\}$.
Recall that $kG_{\mu}\cong M_2(k)$ for the group $G$ and 2-cocycle $\mu$ by Lemma \[lem: kgmuiso\]. Under Hypotheses \[hyp: genhypforfatpts\] one therefore has a cocycle twist $A^{G,\mu}$ that embeds inside $M_2(A)\cong A \otimes
M_2(k)$. The following result describes the generators of the twist under this matrix embedding.
\[lem: matrixgens\] The degree 1 generators of $A^{G,\mu}$, denoted by $v_0, v_1, v_2$ and $v_3$, are given by the following matrices in $M_2(A)$: $$\begin{aligned}
\label{eq: matrixembedding}
v_0 = \begin{pmatrix} x_0 & 0 \\ 0 & x_0 \end{pmatrix},\; v_1 = \begin{pmatrix} x_1 & 0 \\ 0 &
-x_1 \end{pmatrix},\; v_2 = \begin{pmatrix} 0 & x_2 \\ x_2 & 0 \end{pmatrix}, \;\;
v_3 = \begin{pmatrix} 0 & -x_3 \\ x_3 & 0 \end{pmatrix}.\end{aligned}$$
Consider the embedding $A^{G,\mu} \hookrightarrow A\otimes kG_{\mu}$ from Proposition \[prop: twoconstrequal\]. A homogeneous element $x \in A_{g}$ is sent to $x \otimes g$. An isomorphism $kG_{\mu}\cong M_2(k)$ was described in . By Lemma \[lem: kgmuiso\] this isomorphism respects the isotypic components of each algebra under their respective $G$-actions. One can pair up the generators $x_i$ with matrices in using the grading in . The result follows upon using the isomorphism $M_2(A)\cong A \otimes M_2(k)$.
Let us now introduce some notation for point modules over $A$. Under Hypotheses \[hyp: genhypforfatpts\] one may apply Theorem \[thm: pointschemenice\] to conclude that point modules over $A$ are parameterised by the point scheme $\Gamma
\subset {\ensuremath{\mathbb{P}_{k}^{3}}}$. Furthermore, the shifting operation on point modules is controlled by a scheme automorphism $\sigma$.
We will denote the point module corresponding to a point $p=(p_0,p_1,p_2,p_3) \in \Gamma$ by $M_p= \bigoplus_{j \in {\ensuremath{\mathbb{N}}}}
km_j^p$. For the action of the generators of $A$ on $M_p$ we will use the notation $m_j^p \cdot x_i= \alpha_{j,i}^p
m_{j+1}^p$, where $\alpha_{j,i}^p \in k$. By standard point module theory one has $M_p[j]_{\geq 0} \cong
M_{\sigma^j(p)}$ for all $j \in {\ensuremath{\mathbb{N}}}$. The scalar $\alpha_{j,i}^p$ is therefore determined by the $(i+1)$’th coordinate of $\sigma^j(p)$, thus $\alpha_{0,i}^p=p_i$ for $i=0,1,2,3$ in particular.
For a point $p \in \Gamma$ it is clear that $M_p^2$ is an ${\ensuremath{\mathbb{N}}}$-graded right $M_2(A)$-module with Hilbert series $2/(1-t)$, whose action is given by matrix multiplication on the right. It is natural to restrict such a module to the subring $A^{G,\mu}$, whose generators act by the matrices in . We can prove the following proposition about such modules.
\[prop: fatpoints\] Suppose that at least three coordinates of $p \in \Gamma$ are non-zero. Then $M_p^2$ is a fat point module over $A^{G,\mu}$ of multiplicity 2.
We must show that $M_p^2$ is 1-critical and generated in degree 0; let us proceed by proving the latter statement. Let $N$ denote the submodule generated by $(M_p^2)_0$. We prove by induction that $(M_p^2)_j \subset N$ for all $j \geq 0$, where the base case $j=0$ is clear. Suppose that $(M_p^2)_j \subset N$ for some $j \geq 0$. Since $M_p[j]_{\geq 0} \cong
M_{\sigma^j(p)}$, the action of the generators of $A$ on $m_j^p$ is given by the coordinates of $\sigma^j(p)$. At least one generator does not annihilate $m_j^p$, in which case letting the corresponding generator of $A^{G,\mu}$ act on $(m_j^p,0)$ and $(0,m_j^p)$ shows that $(M_p^2)_{j+1} \subset N$. By induction, $N=M_p^2$ and so $M_p^2$ is generated in degree 0.
To prove that $M_p^2$ is 1-critical it is sufficient to show that any cyclic ${\ensuremath{\mathbb{N}}}$-graded submodule has finite codimension. Consider the submodule generated by an element $(m_j^p,\lambda m_j^p) \in M_p^2$, where $\lambda \in
k^{\times}$. We will show that $(m_{j+1}^p,0)$ and $(0,m_{j+1}^p)$ belong to the submodule. Since $M_p^2[j+1]_{\geq 0}
\cong M_{\sigma^{j+1}(p)}^2$, the argument of the previous paragraph implies that these two elements generate $M_p^2$ in high degree, in which case the submodule generated by $(m_j^p,\lambda m_j^p)$ must have finite codimension.
By assumption $\alpha_{j,i}^p \neq 0$ for at least three of the generators. This means that either $\alpha_{j,0}^p,\alpha_{j,1}^p \neq 0$ or $\alpha_{j,2}^p,\alpha_{j,3}^p \neq 0$. In the former case $x_0$ and $x_1$ do not annihilate $m_j^p$ and thus $$\begin{aligned}
(m_j^p,\lambda m_j^p) \cdot \left(v_0+ \frac{\alpha_{j,0}^p}{\alpha_{j,1}^p} v_1\right) &=
(2\alpha_{j,0}^p m_{j+1}^p,0), \\
(m_j^p, \lambda m_j^p) \cdot \left(v_0 - \frac{\alpha_{j,0}^p}{\alpha_{j,1}^p} v_1 \right) &=
(0,2 \lambda \alpha_{j,0}^p m_{j+1}^p).\end{aligned}$$ On the other hand, if $\alpha_{j,2},\alpha_{j,3} \neq 0$ then $x_2$ and $x_3$ do not annihilate $m_j^p$ and so $$\begin{aligned}
(m_j^p,\lambda m_j^p) \cdot \left(v_2 + \frac{\alpha_{j,2}^p}{\alpha_{j,3}^p} v_3 \right) &=
(2 \lambda \alpha_{j,3}^p m_{j+1}^p,0), \\
(m_j^p, \lambda m_j^p) \cdot \left(v_2 - \frac{\alpha_{j,2}^p}{\alpha_{j,3}^p} v_3 \right) &=
(0,2 \alpha_{j,2}^p m_{j+1}^p). \end{aligned}$$ These equations prove that the submodule generated by $(m_j^p, \lambda m_j^p)$ has finite codimension in $M_p^2$.
It remains to show that the submodules generated by either $(m_j^p,0)$ or $(0,m_j^p)$ have finite codimension. We give the argument for $(m_j^p,0)$, the argument for $(0,m_j^p)$ being similar. By assumption either $\alpha_{j,0}^p,\alpha_{j,2}^p \neq 0$ or $\alpha_{j,1}^p,\alpha_{j,3}^p \neq 0$. If $\alpha_{j,0}^p,\alpha_{j,2}^p \neq
0$ then one has $$(m_j^p,0)\cdot v_0 = (\alpha_{j,0}^p m_{j+1}^p,0)\; \text{ and }\; (m_j^p,0)\cdot v_2 = (0,\alpha_{j,2}^p m_{j+1}^p),$$ while if $\alpha_{j,1}^p,\alpha_{j,3}^p \neq 0$ one has $$(m_j^p,0)\cdot v_1 = (\alpha_{j,1}^p m_{j+1}^p,0)\; \text{ and }\;(m_j^p,0)\cdot v_3 = (0,-\alpha_{j,3}^p m_{j+1}^p).$$ Once again, this is sufficient to show the submodule generated by $(m_j^p,0)$ has finite codimension.
There is a natural action of $G$ on point modules of $A$, which we now describe. Since $G$ acts on $A$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms, one can use Definition \[defn: ztwistmodule\] to twist such a module by the action of a particular element of the group. Explicitly, for a point module $M$ and $g \in G$, define a new $A$-module $M^g$ by the multiplication $m \ast_g a=ma^g$ for all $m \in M$ and $a \in A$. The underlying ${\ensuremath{\mathbb{N}}}$-graded vector space structure of $M$ remains unchanged and therefore $M^g$ still has the same Hilbert series as a point module. Moreover, $g$ acts by an automorphism and so the twisted module remains cyclic, therefore $M^g$ must also be a point module.
This action on point modules induces an action of $G$ on $\Gamma$: for all $g \in G$, $p \in \Gamma$, define $p^g$ to be the point $q \in \Gamma$ for which $M_q \cong (M_p)^g$. Before stating a result concerning this action, let us introduce some notation.
\[not: idealaut\] Let $I$ be a right ideal in an algebra $A$, on which a finite group $G$ acts by algebra automorphisms. Then for all $g
\in G$ we define $g(I) := \{a^{g} : \; a \in I\}$.
\[lem: actiononpoints\] Assume Hypotheses \[hyp: genhypforfatpts\]. Then the group $G$ acts on a point $p=(p_0,p_1,p_2,p_3) \in \Gamma$ in the following manner: $$\begin{gathered}
\begin{aligned}\label{eq: Gactonpoints}
&p^e =p,\;\; p^{g_{1}}=(p_0,p_1,-p_2,-p_3),\;\; p^{g_{2}}=(p_0,-p_1,p_2,-p_3),\\ &p^{g_{1}g_{2}} =(p_0,-p_1,-p_2,p_3).
\end{aligned}\end{gathered}$$ In particular, this action preserves the condition on a point having at least three non-zero coordinates.
Consider the point module $M_p=A/I_p$, where $I_p$ is a right ideal. For an element $g \in G$ we claim that $(M_p)^g
\cong A/g^{-1}(I_p)$. To see this, recall that $(M_p)^g$ has the same underlying ${\ensuremath{\mathbb{N}}}$-graded vector space structure as $M_p$. Thus we may consider the map $\varphi: (M_p)^g \rightarrow A/g^{-1}(I_p)$ defined by $a + I_p \mapsto a^{g^{-1}}
+ g^{-1}(I_p)$ for $a \in A$. This map is well-defined on the coset structures involved and preserves the ${\ensuremath{\mathbb{N}}}$-graded vector space structures, thus it remains to check that it is an isomorphism of right $A$-modules. For $a, b \in A$ one has $$\begin{aligned}
\varphi((a+I_p) \ast_g b) &= \varphi(ab^{g} + I_p) = (ab^g)^{g^{-1}} + g^{-1}(I_p) = a^{g^{-1}}b + g^{-1}(I_p),\\
\varphi(a+I_p)b &= (a^{g^{-1}} + g^{-1}(I_p))b = a^{g^{-1}}b + g^{-1}(I_p).\end{aligned}$$ Thus $\varphi$ is a homomorphism of ${\ensuremath{\mathbb{N}}}$-graded right $A$-modules. As $G$ acts by automorphisms, the modules $(M_p)^g$ and $A/g^{-1}(I_p)$ have the same Hilbert series. Consequently, $\varphi$ is an isomorphism.
The right ideal $I_p$ is generated by the degree 1 elements $$\label{eq: ptmoddeg1gens}
p_0x_1 - p_1 x_0,\;\;\; p_0x_2 - p_2 x_0,\;\;\; p_0x_3 - p_3 x_0.$$ The isomorphism $(M_p)^g \cong A/g^{-1}(I_p)$ and Lemma \[lem: defrelns\] indicate that the behaviour of the three generators in under the action of $g^{-1}$ govern $p^g$. The result is clear when $g=e$ since the identity acts trivially. We give a proof for $g=g_1$, with the remaining two cases being similar. Noting that $g_1$ has order 2, one has $$\begin{gathered}
\begin{aligned}\label{eq: ptmoddeg1gensact}
(p_0x_1 - p_1 x_0)^{g_{1}} &= p_0x_1^{g_{1}} - p_1 x_0^{g_{1}} = p_0x_1 - p_1 x_0,\\
(p_0x_2 - p_2 x_0)^{g_{1}} &= p_0x_2^{g_{1}} - p_2 x_0^{g_{1}} = -p_0x_2 - p_2 x_0,\\
(p_0x_3 - p_3 x_0)^{g_{1}} &= p_0x_3^{g_{1}} - p_1 x_3^{g_{1}} = -p_0x_3 - p_3 x_0.
\end{aligned} \end{gathered}$$ The right ideal generated by the three elements on the right-hand side of corresponds to the point $q = (p_0,p_1,-p_2,-p_3)$. Thus $p^{g_{1}}=q$ as in the statement of the lemma.
In Proposition \[prop: fatpoints\] we constructed fat point modules over the twist $A^{G,\mu}$. One can repeat the same trick, using the embedding of $A=(A^{G,\mu})^{G,\mu}$ inside $M_2(A^{G,\mu})$ to prove the following result.
\[prop: fatpointsotherwaygen\] Consider $M_p^2$, a fat point module over $A^{G,\mu}$ constructed in Proposition \[prop: fatpoints\]. The direct sum $(M_p^2)^2$ is an ${\ensuremath{\mathbb{N}}}$-graded right $M_2(A^{G,\mu})$-module. On restriction to a module over the subalgebra $A$, one has a decomposition of ${\ensuremath{\mathbb{N}}}$-graded right $A$-modules $$\label{eq: orbitdecomp}
(M_p^2)^2 \cong \bigoplus_{g \in G} M_{p^{g}}.$$
The fat point module $M_p^2$ was obtained by restricting the $M_2(A)$-module $M_p^2$ to $A^{G,\mu}$. Thus $(M_p^2)^2$ can be considered as the $M_4(A)$-module $M_p^4$, which becomes an $M_2(A^{G,\mu})$-module upon restriction. One can then regard $(M_p^2)^2$ as an $A$-module by restricting a second time, with the action on homogeneous pieces given by $4 \times 4$ matrices.
Indeed, one can write the action of the generators of $A$ on $(M_p^2)^2$ explicitly by unravelling the composite embedding of $A$ into $M_4(A)$. Let us use the notation $m_j^p \cdot x_i= \alpha_{j,i}^p m_{j+1}^p$ — which was introduced before Proposition \[prop: fatpoints\] — for the action of the generators of $A$ on the point module $M_p$. For all $j \in {\ensuremath{\mathbb{N}}}$ and $c_l \in k$, the action of the generators of $A$ on $(M_p^2)^2$ is given by $$(c_0 m_j^p, c_1 m_j^p,c_2 m_j^p,c_3 m_j^p) \cdot x_i := (c_0 m_{j+1}^p, c_1 m_{j+1}^p,c_2 m_{j+1}^p,c_3 m_{j+1}^p) \cdot
Q_i,$$ where $$\begin{gathered}
\begin{aligned}\label{eq: actonfatptdoubles}
Q_0 &= \begin{pmatrix} \alpha_{j,0}^p &0&0&0 \\0& \alpha_{j,0}^p &0&0 \\ 0&0& \alpha_{j,0}^p &0 \\ 0&0&0& \alpha_{j,0}^p
\end{pmatrix},\;\;
Q_1 = \begin{pmatrix} \alpha_{j,1}^p &0&0&0 \\0& -\alpha_{j,1}^p &0&0 \\ 0&0& -\alpha_{j,1}^p &0 \\ 0&0&0&
\alpha_{j,1}^p \end{pmatrix},\\
Q_2 &= \begin{pmatrix} 0&0&0& \alpha_{j,2}^p \\0&0& \alpha_{j,2}^p &0 \\ 0& \alpha_{j,2}^p &0&0 \\ \alpha_{j,2}^p &0&0&0
\end{pmatrix},\;\;
Q_3 = \begin{pmatrix} 0&0&0& \alpha_{j,3}^p \\0&0& -\alpha_{j,3}^p &0 \\ 0& -\alpha_{j,3}^p &0&0 \\ \alpha_{j,3}^p
&0&0&0 \end{pmatrix}.
\end{aligned}\end{gathered}$$
One can see from the matrices in that there is a decomposition of $A$-modules $$\label{eq: forbitdecomp}
(M_p^2)^2=(M_p,0,0,M_p) \oplus (0,M_p,M_p,0).$$
We claim that the decomposition in the statement of the proposition is then given by the isomorphisms of right $A$-modules $$\begin{gathered}
\begin{aligned}\label{eq: decompasptmods}
(M_p,0,0,M_p) &\cong (m_0,0,0,m_0)A \oplus (m_0,0,0,-m_0)A \cong M_p \oplus M_{p^{g_{1}}}, \\
(0,M_p,M_p,0) &\cong (0,m_0,m_0,0)A \oplus (0,m_0,-m_0,0)A \cong M_{p^{g_{2}}} \oplus M_{p^{g_{1}g_{2}}}.
\end{aligned}\end{gathered}$$ To see this, note that the submodules $$(m_0,0,0,m_0)A,\; (m_0,0,0,-m_0)A, \; (0,m_0,m_0,0)A \; \text{ and }\;(0,m_0,-m_0,0)A$$ certainly have the correct Hilbert series to be point modules and are cyclic. By calculating which degree 1 elements of $A$ annihilate them, one can see that these cyclic submodules are indeed isomorphic to the point modules indicated in .
Proposition \[prop: fatpointsotherwaygen\] allows us to determine the isomorphisms in $\text{grmod}(A^{G,\mu})$ between the fat point modules constructed in Proposition \[prop: fatpoints\].
\[cor: fatpointisoclasses\] The only isomorphisms in $\text{grmod}(A^{G,\mu})$ between the fat point modules over $A^{G,\mu}$ described in Proposition \[prop: fatpoints\] are of the form $M_p^2 \cong M_{p^g}^2$ for all $g \in G$.
To see that $M_p^2 \cong M_{p^g}^2$ holds for all $g \in G$, we first use the final part of Lemma \[lem: actiononpoints\]: if $p \in \Gamma$ has three non-zero coordinates then so does $p^g$ for all $g \in G$. Thus one can construct $M_{p^g}^2$ as in Proposition \[prop: fatpoints\] for any $g \in G$. The isomorphisms are governed by the matrices in :
- $M_p^2 \cong M_{p^{g_{1}}}^2$ via right multiplication by ${\begin{pmatrix} 1 & 0 \\ 0 & -1 \end{pmatrix}}$;
- $M_p^2 \cong M_{p^{g_{2}}}^2$ via right multiplication by ${\begin{pmatrix} 0 & 1 \\ 1 & 0 \end{pmatrix}}$;
- $M_p^2 \cong M_{p^{g_{1}g_{2}}}^2$ via right multiplication by ${\begin{pmatrix} 0 & -1 \\ 1 & 0 \end{pmatrix}}$.
Suppose now that $p, q \in \Gamma$ both have at least three non-zero coordinates and their associated fat point modules are isomorphic, thus $M_p^2 \cong M_{q}^2$. By Proposition \[prop: fatpointsotherwaygen\], we can take direct sums of these fat point modules and consider them as right $M_2(A^{G,\mu})$-modules. On restricting them down to the subalgebra $A$ we obtain via the following isomorphism: $$\label{eq: orbitdecomp2}
\bigoplus_{g \in G} M_{p^{g}} \cong \bigoplus_{g \in G} M_{q^{g}}.$$
Both modules in are f.g. ${\ensuremath{\mathbb{N}}}$-graded modules of GK dimension 1. By [@smith1992the Proposition 1.5] the factors in a critical composition series of such a module are, when considered in high degree, unique up to permutation and isomorphism. In our case this implies that we must have $\pi(M_q) \cong \pi(M_{p^g})$ for some $g \in G$. Thus there exists some $n \in {\ensuremath{\mathbb{N}}}$ such that $M_q[n]_{\geq 0}\cong M_{p^g}[n]_{\geq 0}$ in $\text{grmod}(A)$. But $M_q[n]_{\geq 0} \cong M_{q^{\sigma^{n}}}$ for any point in $\Gamma$, in which case one has an isomorphism in $\text{grmod}(A)$ of the form $M_{q^{\sigma^{n}}} \cong M_{(p^g)^{\sigma^{n}}}$. Since point modules over $A$ are parameterised up to isomorphism by the closed points of $\Gamma$, we may conclude that $q^{\sigma^{n}}=(p^g)^{\sigma^{n}}$. As $\sigma$ is an automorphism it follows that $q=p^g$.
To end this section we recap the ideas we have used. For $G$ and $\mu$ as in Hypotheses \[hyp: genhypforfatpts\] we can play the following game: take a 1-critical $A$-module $M$ and consider $M^2$ as a module over $A^{G,\mu} \cong
M_2(A)^G$. This module has GK dimension 1 and, by replacing it with $M^2_{\geq n}$ for some $n \in {\ensuremath{\mathbb{N}}}$ if necessary, has a critical composition series in which each composition factor is also 1-critical. This technique provides an effective way of discovering 1-critical modules over a cocycle twist $A^{G,\mu}$ given knowledge of those over $A$.
Twists of Sklyanin algebras {#chap: sklyanin}
===========================
In this chapter we will consider twists of 4-dimensional Sklyanin algebras. Such algebras can be presented as the quotient of the free $k$-algebra $k\{x_0,x_1,x_2,x_3\}$ by the ideal generated by the relations: $$\label{eq: 4sklyaninrelns}
\begin{array}{ll}
f_1:=[x_0,x_1]-\alpha[x_2,x_3]_+, & f_2:=[x_0,x_1]_+ -[x_2,x_3], \\ \relax
f_3:=[x_0,x_2]-\beta[x_3,x_1]_+, & f_4:=[x_0,x_2]_+ -[x_3,x_1], \\ \relax
f_5:=[x_0,x_3]-\gamma[x_1,x_2]_+, & f_6:=[x_0,x_3]_+ -[x_1,x_2],
\end{array}$$ where $\alpha, \beta, \gamma \in k$ satisfy $$\label{eq: 4sklyanincoeffcond}
\alpha + \beta + \gamma + \alpha \beta \gamma = 0 \;\text{ and }\; \{\alpha,\beta, \gamma\} \cap \{0,\pm 1\}=\emptyset.$$ As noted in [@smith1992regularity Equation 1.1.1], the first condition in can be rewritten as $$\label{eq: 4sklyanincoeffcond1}
(\alpha+1)(\beta+1)(\gamma+1)=(1-\alpha)(1-\beta)(1-\gamma),$$ and this form will be useful in some of our computations. The relations in were first given in [@smith1992regularity Equation 0.2.2].
Many properties of the algebra $A(\alpha, \beta, \gamma)$ are controlled by an elliptic curve $E \subset {\ensuremath{\mathbb{P}_{k}^{3}}}$ defined by the following elements in $k[y_0, y_1, y_2, y_3]$, the homogeneous coordinate ring of ${\ensuremath{\mathbb{P}_{k}^{3}}}$ [@smith1992regularity Proposition 2.4]: $$y_0^2+y_1^2+y_2^2+y_3^2, \;\;\; y_3^2 - \left(\frac{1-\gamma}{1+\alpha}\right)y_1^2
+\left(\frac{1+\gamma}{1-\beta}\right)y_1^2$$
One can consider the algebra defined by the relations for any $\alpha, \beta, \gamma \in k$ satisfying , where $k$ is an algebraically closed field with $\text{char}(k)\neq 2$. Such algebras will be referred to as *4-dimensional Sklyanin algebras* and be denoted by $A(\alpha,\beta,\gamma)$ (or simply $A$ if we can omit the parameters without ambiguity). The characteristic assumption on $k$ will be made throughout this chapter and the next. Since the group that will be used in these chapters has order 4, the hypothesis that $\text{char}(k)\nmid |G|$ which was used in Chapter \[chap: cocycletwists\] will always hold.
Before beginning our study of twists of such algebras we make some general remarks pertaining to the notation we will use for elements in cocycle twists. Our notation will be like that used in the proof of Proposition \[prop: koszuldualtwistcommute\]: for the remaining chapters (apart from §\[subsec: homenvelopalg\]) we will denote the generators of the algebra that we wish to twist by $x_i$. If such generators are not acted on diagonally by the group then we will use the notation $w_i$ for a diagonal basis of the generating space. Despite the fact that $A$ and $A^{G,\mu}$ share the same underlying $k$-vector space structure, we will denote the corresponding generators of the twist by $v_i$ to avoid confusion and allow us to write the twisted multiplication as juxtaposition. Thus if $x_i \in
A_g$ and $x_j \in A_h$ for some $g, h \in G$, then $v_iv_j =x_i \ast_{\mu} x_j=\mu(g,h)x_i x_j$.
Properties of the twist {#subsec: 4sklyanintwistandptscheme}
-----------------------
Although there are other twists of the 4-dimensional Sklyanin algebra — as discussed in §\[subsec: permuteaction\] and Remark \[rem: otheractions\] — the one that we will focus our attention upon is the following. Consider the isomorphism between $G=(C_2)^2 = \langle g_1,g_2 \rangle$ and its dual $G^{\vee}$ given by the character table . Define an action of the generators of $G$ on those of $A(\alpha,\beta,\gamma)$ by $$\label{eq: sklyaninactionIuse}
x_0^{g_{1}}=x_0,\;\;\; x_1^{g_{1}}=x_1,\;\;\; x_2^{g_{1}}=-x_2,\;\;\; x_3^{g_{1}}=-x_3\;\; \text{ and }\;\; x_i^{g_{2}}
= (-1)^i x_i,$$ for $i = 0,1,2,3$. The generators of $A(\alpha,\beta,\gamma)$ are acted on diagonally and therefore are homogeneous with respect to the induced $G$-grading, lying in the following components: $$\label{eq: 4sklyaningrading}
x_0 \in A(\alpha,\beta,\gamma)_{e},\; x_1 \in A(\alpha,\beta,\gamma)_{g_{1}},\; x_2 \in
A(\alpha,\beta,\gamma)_{g_{2}},\; x_3 \in A(\alpha,\beta,\gamma)_{g_{1}g_{2}}.$$
The action in is not the same as that used in the example studied in §\[subsec: odesskiiegtwist\]. Nevertheless, we will show in Proposition \[prop: odesskiiegdone\] that Odesskii’s original example is isomorphic to an algebra in the same family as that which we study.
The next lemma describes the relations of the algebra that we will primarily focus our attention on. We will use the 2-cocycle that is used here throughout the rest of the thesis.
\[lem: relationsoftwist\] Let $\mu$ be the 2-cocycle of $G$ defined in . Then the algebra $A(\alpha,\beta,\gamma)^{G,\mu}$ is the quotient of the free $k$-algebra $k\{v_0,v_1,v_2,v_3\}$ by the ideal generated by the following six quadratic relations: $$\label{eq: twistrelns}
\begin{array}{ll}
f_1^{\mu}:=[v_0,v_1]-\alpha[v_2,v_3], & f_2^{\mu}:=[v_0,v_1]_{+} -[v_2,v_3]_{+}, \\ \relax
f_3^{\mu}:=[v_0,v_2]-\beta[v_3,v_1], & f_4^{\mu}:=[v_0,v_2]_{+} -[v_3,v_1]_{+}, \\ \relax
f_5^{\mu}:=[v_0,v_3]+\gamma[v_1,v_2], & f_6^{\mu}:=[v_0,v_3]_{+} +[v_1,v_2]_{+}.\end{array}$$
We begin by computing how the defining relations of $A(\alpha,\beta,\gamma)$ from behave under the twist: $$\begin{aligned}
0 &= x_0x_1-x_1x_0-\alpha x_2 x_3-\alpha x_3 x_2 \\
&= \frac{x_0 \ast_{\mu} x_1}{\mu(e,g_1)}-\frac{x_1\ast_{\mu} x_0}{\mu(g_1,e)}-\alpha \frac{x_2 \ast_{\mu}
x_3}{\mu(g_2,g_1 g_2)}-\alpha \frac{x_3 \ast_{\mu} x_2}{\mu(g_1 g_2,g_2)} \\
&= x_0 \ast_{\mu} x_1-x_1 \ast_{\mu}x_0-\alpha x_2 \ast_{\mu}x_3+\alpha x_3 \ast_{\mu}x_2,\\
0 &= x_0x_1+x_1x_0- x_2 x_3+ x_3 x_2 \\
&= \frac{x_0 \ast_{\mu} x_1}{\mu(e,g_1)}+\frac{x_1\ast_{\mu} x_0}{\mu(g_1,e)}-\frac{x_2 \ast_{\mu} x_3}{\mu(g_2,g_1
g_2)}+\frac{x_3 \ast_{\mu} x_2}{\mu(g_1 g_2,g_2)} \\
&= x_0 \ast_{\mu}x_1+x_1 \ast_{\mu}x_0- x_2 \ast_{\mu}x_3- x_3 \ast_{\mu}x_2,\\
0 &= x_0x_2-x_2x_0-\beta x_3 x_1-\beta x_1 x_3 \\
&= \frac{x_0 \ast_{\mu} x_2}{\mu(e,g_2)}-\frac{x_2\ast_{\mu} x_0}{\mu(g_2,e)}-\beta \frac{x_3 \ast_{\mu} x_1}{\mu(g_1
g_2,g_1)}-\beta \frac{x_1 \ast_{\mu} x_3}{\mu(g_1,g_1 g_2)} \\
&= x_0 \ast_{\mu}x_2-x_2 \ast_{\mu}x_0-\beta x_3 \ast_{\mu}x_1+\beta x_1 \ast_{\mu}x_3,\\
0 &= x_0 x_2+x_2 x_0- x_3 x_1 + x_1 x_3 \\
&= \frac{x_0 \ast_{\mu} x_2}{\mu(e,g_2)}+\frac{x_2 \ast_{\mu} x_0}{\mu(g_2,e)}-\frac{x_3 \ast_{\mu} x_1}{\mu(g_1
g_2,g_1)}+ \frac{x_1 \ast_{\mu} x_3}{\mu(g_1, g_1 g_2)} \\
&= x_0 \ast_{\mu}x_2+x_2 \ast_{\mu}x_0-x_3 \ast_{\mu}x_1 - x_1 \ast_{\mu}x_3,\\
0 &= x_0x_3-x_3x_0-\gamma x_1 x_2-\gamma x_2 x_1 \\
&= \frac{x_0 \ast_{\mu} x_3}{\mu(e,g_1 g_2)}-\frac{x_3\ast_{\mu} x_0}{\mu(g_1 g_2,e)}-\gamma \frac{x_1 \ast_{\mu}
x_2}{\mu(g_1, g_2)}-\gamma \frac{x_2 \ast_{\mu} x_1}{\mu(g_2,g_1)} \\
&= x_0 \ast_{\mu}x_3-x_3 \ast_{\mu}x_0+\gamma x_1 \ast_{\mu}x_2-\gamma x_2 \ast_{\mu}x_1,\\
0 &= x_0x_3+x_3x_0-x_1 x_2+ x_2 x_1 \\
&= \frac{x_0 \ast_{\mu} x_3}{\mu(e,g_1 g_2)}+\frac{x_3\ast_{\mu} x_0}{\mu(g_1 g_2,e)}-\frac{x_1 \ast_{\mu}
x_2}{\mu(g_1,g_2)}+ \frac{x_2 \ast_{\mu} x_1}{\mu(g_2, g_1)} \\
&= x_0 \ast_{\mu}x_3+x_3 \ast_{\mu}x_0+ x_1 \ast_{\mu}x_2+ x_2 \ast_{\mu}x_1.\end{aligned}$$ Recall our remarks regarding notation for cocycle twists at the beginning of Chapter \[chap: sklyanin\]; rewriting the relations above in terms of the new generators $v_i$ produces the relations in . By regarding $A(\alpha,\beta,\gamma)^{G,\mu}$ as a twist of the presentation of $A(\alpha,\beta,\gamma)$ by the relations in , one can use Lemma \[lem: defrelns\] to see that the ideal of relations in the twist is generated by the relations in .
Notice that $A(\alpha,\beta,\gamma)$ has a generating set which is homogeneous with respect to the $G$-grading by . Thus Remark \[lemma: finitelygenerated\] implies that $A(\alpha,\beta,\gamma)^{G,\mu}$ is generated as an algebra by the same generating set, hence by the $v_i$.
\[rem: reducestudy\] In §\[subsec: permuteaction\] we will show that the study of a whole family of cocycle twists can be reduced to studying the algebra defined by the relations in .
By virtue of being a cocycle twist, we immediately get the following result.
\[thm: 4sklytwistprops\] Assume that the parameter triple $(\alpha, \beta, \gamma)$ satisfies . Then the algebra $A(\alpha,\beta,\gamma)^{G,\mu}$ is f.g. in degree 1 and has the following properties:
- it is a universally noetherian domain;
- it has Hilbert series $1/(1-t)^4$;
- it is AS-regular of global dimension 4;
- it is Auslander regular;
- it satisfies the Cohen-Macaulay property;
- it is Koszul.
Note that $A(\alpha,\beta,\gamma)^{G,\mu}$ is f.g. in degree 1 by Lemma \[lem: relationsoftwist\]. We proceed by showing that part of (i) holds. By [@artin1999generic Corollary 4.12], $A(\alpha,\beta,\gamma)$ is universally noetherian. Using Corollary \[cor: uninoeth\] shows that the same is true for $A(\alpha,\beta,\gamma)^{G,\mu}$.
By [@smith1992regularity Thm 5.5] and [@levasseur1993modules Corollary 1.9], $A(\alpha,\beta,\gamma)$ has the properties stated in (ii)-(vi). We can then use Lemma \[lem: hilbseries\] (for (ii)), Corollary \[cor: asreg\] and Proposition \[prop: gldim\] (for (iii)), Proposition \[prop: cohenmac\] (for (iv) and (v)) and Proposition \[prop: koszul\] (for (vi)) to show that $A(\alpha,\beta,\gamma)^{G,\mu}$ also has the respective properties. To complete the proof one can use [@artin1991modules Theorem 3.9] to conclude that $A(\alpha,\beta,\gamma)^{G,\mu}$ is a domain.
In fact Theorem \[thm: 4sklytwistprops\] is valid for any parameter triple $(\alpha, \beta, \gamma)$ that is not of the form $(-1,1,\gamma)$, $(\alpha,-1,1)$ or $(1,\beta,-1)$. For those three triples the associated 4-dimensional Sklyanin algebra is not a domain, while for the remaining triples not covered by the theorem one obtains an iterated Ore extension over $k$ (see [@smith1992regularity §1]). We leave the proof of the outstanding cases to the reader.
### Permuting the $G$-action {#subsec: permuteaction}
Our aim in this section is to determine the behaviour of some other twists of $A(\alpha,\beta,\gamma)$. We will show that such twists are isomorphic up to a change of parameters to the algebra whose relations are given in .
\[lem: 24to1\] Let $G$ be the Klein-four group and $(\alpha,\beta,\gamma)$ a parameter triple satisfying . There are 24 actions of $G$ on $A(\alpha,\beta,\gamma)$ by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms for which the following hold:
- $G$ acts diagonally on the generators $x_0,x_1,x_2$ and $x_3$;
- the action of $G$ on $A(\alpha,\beta,\gamma)_1$ affords the regular representation.
For any such action of $G$, consider the twist of the induced $G$-grading on $A(\alpha,\beta,\gamma)$ by the 2-cocycle $\mu$ from Lemma \[lem: relationsoftwist\]. The algebra obtained is isomorphic to the twist studied in that lemma up to a change of parameters which still satisfy .
In the previous section we studied the action on $A(\alpha,\beta,\gamma)$ corresponding to the grading $$\label{eq: 4sklyaninusualgrading}
x_0 \in A(\alpha,\beta,\gamma)_{e},\; x_1 \in A(\alpha,\beta,\gamma)_{g_{1}},\; x_2 \in
A(\alpha,\beta,\gamma)_{g_{2}},\; x_3 \in A(\alpha,\beta,\gamma)_{g_{1}g_{2}}.$$ We identify each $G$-graded component of $A$ with the corresponding index of the generator it contains in . In this manner, any other grading corresponding to an action of $G$ affording the regular representation gives rise to a permutation in the symmetric group $S_4$. For example, the action which induces the grading $$x_1 \in A(\alpha,\beta,\gamma)_{e},\; x_0 \in A(\alpha,\beta,\gamma)_{g_{1}},\; x_3 \in
A(\alpha,\beta,\gamma)_{g_{2}},\; x_2 \in A(\alpha,\beta,\gamma)_{g_{1}g_{2}}.$$ corresponds to the permutation $(01)(23)$. It is trivial to check using the relations in that each permutation in $S_4$ corresponds to a genuine $G$-grading.
The action which induces the grading in corresponds to the identity permutation, hence *for this proof only the associated twist will be denoted $A(\alpha,\beta,\gamma)^{G,\mu;(id)}$*. Our aim can therefore be reformulated as trying to understand $A(\alpha,\beta,\gamma)^{G,\mu;\sigma}$ for other permutations $\sigma \in S_4$.
Let us now assume that any $G$-grading is one of the 24 arising from an action of $G$ on the generators by the regular representation. The automorphism group of $G$ is isomorphic to $S_3$, which acts by permuting the order 2 elements. We will use the convention that products of permutations are applied from right to left.
Notice that two $G$-gradings are twists of each other by an automorphism of $G$ if and only if the components of their $G$-gradings corresponding to $e$ are equal. This is a consequence of the automorphism group of $G$ being isomorphic to $S_3$. Recast in terms of permutations, we obtain a partition of $S_4$ by the subsets $$H_j=\{\sigma \in S_4 : \; \sigma^{-1}(0)=j\} \;\text{ for }\;j=0,1,2,3.$$ Let us choose the identity map and the transpositions $(0j)$ for $j = 1,2,3$ as representatives of these subsets.
We now show that the action of $\text{Aut}_{\text{grp}}(G)$ on 2-cocycles of $G$ is trivial. To do this we first identify the subgroup of $S_4$ given by $$\label{eq: 0stabperms}
H_0=\{(id),(12),(23),(13),(123),(132)\},$$ with $\text{Aut}_{\text{grp}}(G)$. The identification we use arises naturally from our prior identification of elements of $G$ with the set $\{0,1,2,3\}$ using .
Observe that the following table describes for each permutation $\sigma \in H_0$ how the 2-cocycles $\mu$ and $\mu^{(\sigma^{-1})}$ are cohomologous via the function $\rho: G \rightarrow k^{\times}$, where $i$ denotes a primitive $4^{\text{th}}$ root of unity in $k$: $$\label{eq: sigmaactcoboundary}
\begin{array}{c|cccc}
\sigma & \rho(e) & \rho(g_1) & \rho(g_2) & \rho(g_1g_2) \\
\hline
(12) & 1 & -1 & 1 & 1 \\
(13) & 1 & i & 1 & i \\
(23) & 1 & 1 & i & i \\
(123) & 1 & i & -1 & i \\
(132) & 1 & 1 & i & -i
\end{array}$$
Now consider $\sigma \in H_j$. We already know that the $G$-grading associated to $\sigma$ is obtained from that associated to $(0j)$ by twisting the grading by an automorphism of $G$. But shows that the action of $\text{Aut}_{\text{grp}}(G)$ on 2-cocycles is trivial. We may conclude by Lemma \[lem: autoncocycle\] that the twists $A(\alpha,\beta,\gamma)^{G,\mu;\sigma}$ and $A(\alpha,\beta,\gamma)^{G,\mu;(0j)}$ are twists of the same $G$-grading by cohomologous 2-cocycles. By Proposition \[prop: trivialtwist\] it follows that these algebras are isomorphic.
To complete the proof we describe the changes of parameters needed to show that the three algebras of the form $A(\alpha,\beta,\gamma)^{G,\mu;(0j)}$ belong to the family $A(\alpha',\beta',\gamma')^{G,\mu;(id)}$ for some $(\alpha',\beta',\gamma')$ satisfying .
For the permutation $(01)$, one can use the rescaling $$(v_0,v_1,v_2,v_3) \mapsto \left(v_0,\frac{i}{\sqrt{\beta
\gamma}}v_1,-\frac{1}{\sqrt{\gamma}}v_2,-\frac{i}{\sqrt{\beta}}v_3\right),$$ to show that there is an isomorphism of ${\ensuremath{\mathbb{N}}}$-graded $k$-algebras $$\label{eq: 01iso}
A(\alpha,\beta,\gamma)^{G,\mu;(01)} \cong A\left(\alpha, \frac{1}{\beta}, \frac{1}{\gamma}\right)^{G,\mu;(id)}.$$
Similarly, for $(02)$ there is an isomorphism of ${\ensuremath{\mathbb{N}}}$-graded $k$-algebras $$A(\alpha,\beta,\gamma)^{G,\mu;(02)} \cong A\left(\frac{1}{\alpha}, \beta, \frac{1}{\gamma}\right)^{G,\mu;(id)},$$ given by the rescaling $$(v_0,v_1,v_2,v_3) \mapsto
\left(v_0,\frac{i}{\sqrt{\gamma}}v_1,\frac{i}{\sqrt{\alpha\gamma}}v_2,\frac{1}{\sqrt{\alpha}}v_3\right).$$
Likewise, for $(03)$ there is an isomorphism of ${\ensuremath{\mathbb{N}}}$-graded $k$-algebras $$A(\alpha,\beta,\gamma)^{G,\mu;(03)} \cong A\left(\frac{1}{\alpha}, \frac{1}{\beta}, \gamma\right)^{G,\mu;(id)},$$ given by the rescaling $$(v_0,v_1,v_2,v_3) \mapsto
\left(v_0,\frac{i}{\sqrt{\beta}}v_1,\frac{1}{\sqrt{\alpha}}v_2,\frac{i}{\sqrt{\alpha\beta}}v_3\right).$$
If the parameters $(\alpha, \beta, \gamma)$ satisfy then so does the triple obtained by negation, permutation, or taking the reciprocal of two of the three parameters. This completes the proof.
\[rem: otheractions\] There are of course other actions of $G$ on $A(\alpha,\beta,\gamma)$. For example, we could consider the action inducing the grading $$x_0, x_1 \in A(\alpha,\beta,\gamma)_{g_{1}},\; x_2,x_3 \in A(\alpha,\beta,\gamma)_{g_{1}g_{2}}.$$ The cocycle twist of this grading by $\mu$ is a Zhang twist of $A(\alpha,\beta,\gamma)$ by the automorphism by which $g_1$ acts. This is an artefact of $g_2$ acting by scalar multiplication, which brings to mind the proof of Proposition \[prop: recoverztwist\]; we defined a group action where one generator acted by scalar multiplication in order to recover a Zhang twist as a cocycle twist.
Proposition \[lem: 24to1\] implies that our study of $A(\alpha,\beta,\gamma)^{G,\mu;(id)}$ encompasses 24 twists up to an allowable change of parameters. Thus, from the beginning of §\[subsec: 4sklypointscheme\] onwards, the notation $A^{G,\mu}$ will refer to that twist.
We now address the question of whether Odesskii’s example from §\[subsec: odesskiiegtwist\] is isomorphic to one of the twists we have just analysed. We therefore assume that $k={\ensuremath{\mathbb{C}}}$ for the rest of §\[subsec: permuteaction\]. Recall that in Proposition \[prop: odesskiieg\] it was shown that Odesskii’s example can be formulated as a cocycle twist.
We will use results of Staniszkis and Smith from [@smith1993irreducible]. Although the primary focus of their paper is to classify the finite-dimensional simple $A$-modules, they also describe the ${\ensuremath{\mathbb{N}}}$-graded automorphism group of $A$, denoted $\text{Aut}_{{\ensuremath{\mathbb{N}}}\text{-alg}}(A)$, which we describe below. For this purpose it is useful to consider the associated elliptic curve $E$ as the quotient group ${\ensuremath{\mathbb{C}}}/\Lambda$, where $\Lambda$ is an integer lattice in ${\ensuremath{\mathbb{C}}}$ generated by a complex number $\nu$ such that $\text{im}(\nu)\neq 0$ (see the section on elliptic functions in [@hartshorne1977algebraic Chapter IV, pgs. 326-329]).
Following [@smith1992regularity §2.10], let us introduce the following holomorphic functions on ${\ensuremath{\mathbb{C}}}$ associated to the lattice $\Lambda$. That such *theta functions* are related to the Sklyanin algebra associated to $E$ is clear from below.
\[def: thetafunctions\] For $a,b \in \{0,1\}$ define $\Theta_{ab}: {\ensuremath{\mathbb{C}}}\rightarrow {\ensuremath{\mathbb{C}}}$ to be a holomorphic function satisfying the relations $$\Theta_{ab}(z+1) = (-1)^a \Theta_{ab}(z), \;\; \Theta_{ab}(z+\nu)=\text{exp}(-\pi i \nu - 2\pi i z - \pi i b)
\Theta_{ab}(z),$$ for all $z \in {\ensuremath{\mathbb{C}}}$.
Let the automorphism associated to the point scheme of $A$ be translation by the point $\tau \in E$. It is shown in [@smith1993irreducible Theorem 2.2] that apart from one exceptional case there is a short exact sequence of groups $$\label{eq: sesAutA}
{\ensuremath{1 \rightarrow {\ensuremath{\mathbb{C}}}^{\times} \rightarrow \text{Aut}_{{\ensuremath{\mathbb{N}}}\text{-alg}}(A) \rightarrow E_4 \rightarrow 0}},$$ where $E_4$ denotes the 4-torsion on $E$. When $E$ is regarded as a lattice, such torsion can easily be described as the 16 cosets of the form $\frac{n}{4}+\frac{m}{4}\nu + \Lambda$ for $n,m \in \{0,1,2,3\}$. By [@hartshorne1977algebraic Chapter IV, Theorem 4.16], $E_4$ is isomorphic as a group to $({\ensuremath{\mathbb{Z}}}/4{\ensuremath{\mathbb{Z}}})^2$. For any $n \in {\ensuremath{\mathbb{N}}}$ the automorphism associated to $\lambda \in {\ensuremath{\mathbb{C}}}^{\times}$ acts on $A_n$ by scalar multiplication by $\lambda^n$.
The exceptional case mentioned above occurs when $|\tau|=3$ and $E$ has a special form. In that case there is still an exact sequence like in , but with $E_4$ replaced with $E_4 \times ({\ensuremath{\mathbb{Z}}}/3 {\ensuremath{\mathbb{Z}}})$ (see [@smith1993irreducible Theorem 2.2(c)]). As we are mainly interested in the situation when $\tau$ has infinite order, we will not consider this case.
On pg. 64 op. cit. some automorphisms of $A$ are given which are labelled by the generators of $E_4$, and one can in fact associate all points in $E_4$ with automorphisms. The exact sequence in is non-split because the composition of automorphisms parameterised by $E_4$ only respects the addition on $E$ coming from their corresponding points up to scalar automorphism.
By taking compositions of the automorphisms corresponding to the two generators of $E_4$ (which are exhibited on [@smith1993irreducible pg. 64]), we can recover all of the automorphisms parameterised by $E_4$ up to scalar. This is enough for our purposes: to prove Proposition \[prop: odesskiiegdone\] we will only need to know whether certain automorphisms act diagonally on the generators, thus working up to scalar is sufficient.
Some of the automorphisms exhibited in Smith and Staniszkis’s paper can be written in terms of the theta functions introduced above in Definition \[def: thetafunctions\]. The parameters in the relations can be written in terms of those functions as follows: $$\label{eq: thetafunctionsoldnew}
\alpha =\left(\frac{\Theta_{11}(\tau)\Theta_{00}(\tau)}{\Theta_{01}(\tau)\Theta_{10}(\tau)} \right)^2 ,\;\; \beta=-
\left(\frac{\Theta_{11}(\tau)\Theta_{01}(\tau)}{\Theta_{00}(\tau)\Theta_{10}(\tau)} \right)^2,\;\; \gamma=-
\left(\frac{\Theta_{11}(\tau)\Theta_{10}(\tau)}{\Theta_{00}(\tau)\Theta_{01}(\tau)} \right)^2.$$
Using these expressions, one can rewrite the automorphisms in terms of the new parameters, as we now illustrate. Consider the automorphism corresponding to $\frac{1}{4} + \frac{1}{4}\nu \in E_4$, which sends $$(X_{11},X_{00},X_{01},X_{10}) \mapsto \left(\frac{\Theta_{11}(\tau)}{\Theta_{00}(\tau)} X_{00},i
\frac{\Theta_{00}(\tau)}{\Theta_{11}(\tau)} X_{11},\frac{\Theta_{01}(\tau)}{\Theta_{10}(\tau)} X_{10}, -i
\frac{\Theta_{10}(\tau)}{\Theta_{01}(\tau)} X_{01}\right).$$
Using the identification $(X_{11},X_{00},X_{01},X_{10})=(x_0,x_1,x_2,x_3)$ and , we can rewrite this as $$(x_0,x_1,x_2,x_3) \mapsto \left( (\beta \gamma)^{\frac{1}{4}} x_1,i (\beta \gamma)^{-\frac{1}{4}}
x_0,\left(\frac{\beta}{\gamma}\right)^{\frac{1}{4}} x_3, -i \left(\frac{\beta}{\gamma}\right)^{-\frac{1}{4}} x_2
\right).$$
We now prove that Odesskii’s twist must be one that we have previously encountered.
\[prop: odesskiiegdone\] Suppose that $|\tau|=\infty$. The example of Odesskii in §\[subsec: odesskiiegtwist\] is isomorphic to an algebra with relations for some choice of parameters $(\alpha',\beta', \gamma')$ satisfying .
Note that the algebra being twisted in Odesskii’s example, $A(\alpha,\beta,\gamma)$ say, must correspond to parameters satisfying due to its construction using a smooth elliptic curve. Smith and Stafford’s work in [@smith1992regularity] shows that the Sklyanin algebra is only ‘elliptic’ in this sense under the restriction on parameters.
The action in Odesskii’s example — which was described in — affords the regular representation when restricted to the degree 1 generators. We will show that the action of $G$ on $A$ in this example is by automorphisms that are diagonal in the new basis (for which the relations are given by ). In that case, the induced $G$-grading on $A$ will be one of the 24 considered Proposition \[lem: 24to1\]. That result implies that the twist must be isomorphic to $A(\alpha', \beta', \gamma')^{G,\mu}$ for some choice of parameters satisfying the required condition. It is therefore sufficient to show that all automorphisms of order 2 act diagonally on our generators.
We are not in the exceptional case, hence there is an exact sequence as given in . Let $\rho \in
\text{Aut}_{{\ensuremath{\mathbb{N}}}\text{-alg}}(A)$ have order 2. As discussed prior to the proposition, there exists $\lambda \in
{\ensuremath{\mathbb{C}}}^{\times}$ and one of the automorphisms parameterised by $E_4$, $\phi$ say, such that $\lambda \rho= \phi$. Since $\lambda$ commutes with all automorphisms, one has $\phi^2= \lambda^2 \in {\ensuremath{\mathbb{C}}}^{\times}$. Knowing the automorphisms parameterised by $E_4$ up to scalar allows us to conclude that the only such automorphisms whose square is a scalar automorphism are those corresponding to the 2-torsion points. It can be seen from [@smith1993irreducible pg. 64]) that such automorphisms act diagonally on the generators, from which we can conclude that $\rho$ must also act diagonally on them. As was shown in the previous paragraph, this is sufficient to prove the result.
### The point scheme {#subsec: 4sklypointscheme}
We now begin to study the point scheme of $A(\alpha,\beta,\gamma)^{G,\mu}$, which we will denote by $\Gamma$. Our ultimate aim is to prove the following result.
Suppose that $A(\alpha,\beta,\gamma)$ is associated to the elliptic curve $E$ and automorphism $\sigma$, which has infinite order. Then the point scheme $\Gamma$ of $A(\alpha,\beta,\gamma)^{G,\mu}$ consists of 20 points.
The proof of this result will appear later as Theorem \[prop: finitepointscheme\]. The delay is needed because the proof uses the interplay between 1-critical modules over $A(\alpha,\beta,\gamma)$ and $A(\alpha,\beta,\gamma)^{G,\mu}$ – this is studied in §\[sec: modulesoversklytwist\].
Let us begin by recalling to the reader’s attention Theorem \[thm: pointschemenice\], regarding the multilinearisations of the quadratic relations in certain algebras.
\[lem: multilins\] Consider the multilinearisations of the relations of $A(\alpha,\beta,\gamma)^{G,\mu}$ given in : $$\begin{gathered}
\begin{aligned}\label{eqn: multilins}
&m_1:=v_{01}v_{12}-v_{11}v_{02}-\alpha v_{21}v_{32}+\alpha v_{31}v_{22},\;\;\;\;
m_2:=v_{01}v_{12}+v_{11}v_{02}-v_{21}v_{32}-v_{31}v_{22}, \\
&m_3:=v_{01}v_{22}-v_{21}v_{02}+\beta v_{11}v_{32}-\beta v_{31}v_{12},\;\;\;\;
m_4:=v_{01}v_{22}+v_{21}v_{02}-v_{31}v_{12}-v_{11}v_{32}, \\
&m_5:=v_{01}v_{32}-v_{31}v_{02}+\gamma v_{11}v_{22}-\gamma v_{21}v_{12},\;\;\;\;
m_6:=v_{01}v_{32}+v_{31}v_{02}+v_{11}v_{22}+v_{21}v_{12}.
\end{aligned}\end{gathered}$$ The closed subscheme $\Gamma_2 \subset {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ determined by the equations in is isomorphic to the graph of the point scheme $\Gamma$ under a scheme automorphism $\phi$. Furthermore, the closed points of $\Gamma$ parameterise point modules over $A(\alpha,\beta,\gamma)^{G,\mu}$.
One can use Theorem \[thm: 4sklytwistprops\] to see that $A(\alpha,\beta,\gamma)^{G,\mu}$ satisfies the hypotheses of Theorem \[thm: pointschemenice\], from which the result follows.
We will often study $\Gamma_2$, since it incorporates information about both the point scheme and the associated automorphism. Lemma \[lem: multilins\] allows us to write $$\label{eq: 4sklyaninpointlocus}
\Gamma_2=\{q = (p,p^{\phi}) \in {\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}: m_i(q)=0\;\text{ for $i=1,\ldots,6$} \}.$$
Our next result exhibits 20 closed points of the scheme $\Gamma_2$. Under some additional hypotheses, such points will be shown in Theorem \[prop: finitepointscheme\] to compose all of the closed points of $\Gamma_2$. Let us introduce the following notation for certain points in ${\ensuremath{\mathbb{P}_{k}^{3}}}$ prior to stating the result: $$\label{eq: 4sklyanineipts}
e_0:=(1,0,0,0),\;\; e_1:=(0,1,0,0),\;\; e_2:=(0,0,1,0),\;\; e_3:=(0,0,0,1),$$
\[lem: ptschemecontains\] The scheme $\Gamma_2$ contains the closed points $(e_j,e_j)$ for $j = 0,1,2,3$, as well as the following 16 points, where $i^2= -1$: $$\begin{gathered}
\begin{aligned}\label{eq: 4sklyanintwistpts}
&((1,\pm i, \pm i ,1),(1, \pm i, \pm i , 1)),\;\;\; ((1,\pm i, \mp i ,-1),(1, \pm i, \mp i , -1)),\\
&\left(\left(1,-(\beta \gamma)^{-\frac{1}{2}},\mp \gamma^{-\frac{1}{2}},\mp\beta^{-\frac{1}{2}} \right),\left(1,-(\beta
\gamma)^{-\frac{1}{2}},\pm \gamma^{-\frac{1}{2}},\pm\beta^{-\frac{1}{2}} \right)\right), \\
&\left(\left(1,(\beta \gamma)^{-\frac{1}{2}},\mp \gamma^{-\frac{1}{2}},\pm\beta^{-\frac{1}{2}} \right),\left(1,(\beta
\gamma)^{-\frac{1}{2}},\pm \gamma^{-\frac{1}{2}},\mp\beta^{-\frac{1}{2}} \right)\right), \\
&\left(\left(1,\pm i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}} \right),\left(1,
\mp i\gamma^{-\frac{1}{2}},(\alpha \gamma)^{-\frac{1}{2}},\mp i \alpha^{-\frac{1}{2}} \right)\right), \\
&\left(\left(1, \mp i\gamma^{-\frac{1}{2}},- (\alpha \gamma)^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}} \right),\left(1,
\pm i\gamma^{-\frac{1}{2}},- (\alpha \gamma)^{-\frac{1}{2}},\mp i \alpha^{-\frac{1}{2}} \right)\right), \\
&\left(\left(1,\pm \beta^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right),\left(1,\mp
\beta^{-\frac{1}{2}},\mp i \alpha^{-\frac{1}{2}}, i (\alpha \beta)^{-\frac{1}{2}} \right)\right), \\
&\left(\left(1,\pm \beta^{-\frac{1}{2}},\mp i \alpha^{-\frac{1}{2}},- i (\alpha \beta)^{-\frac{1}{2}} \right),\left(1,
\mp \beta^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}},- i (\alpha \beta)^{-\frac{1}{2}} \right)\right).
\end{aligned}\end{gathered}$$ In particular, $\Gamma = \pi_1(\Gamma_2)$ contains at least 20 points.
It is easy to see that $(e_j,e_j) \in \Gamma_2$ for $j = 0,1,2,3$, since these points satisfy the multilinearisations in . A routine verification — using where necessary — confirms that the points in also satisfy the equations in , and therefore belong to $\Gamma_2$.
To see that the 20 points are distinct, note that by it suffices to show that the points in $\Gamma$ are distinct. Our assumptions on $\alpha$, $\beta$, $\gamma$ from imply that $e_j \notin \Gamma'$ for $j = 0,1,2,3$. It is easy to partition $\Gamma'$ into four sets of order 4 based on which pair from the parameters $\alpha$, $\beta$, $\gamma$ (if any) is needed to describe the coordinates of the point: $$\begin{gathered}
\begin{aligned}\label{eq: blah}
&\{(1,\pm i, \pm i ,1), (1, \pm i, \mp i , 1)\},\\
&\left\lbrace\left(1,-(\beta \gamma)^{-\frac{1}{2}},\mp \gamma^{-\frac{1}{2}},\mp\beta^{-\frac{1}{2}} \right),
\left(1,(\beta \gamma)^{-\frac{1}{2}},\mp \gamma^{-\frac{1}{2}},\pm\beta^{-\frac{1}{2}} \right)\right\rbrace ,\\
&\left\lbrace\left(1,\pm i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}} \right),
\left(1, \mp i\gamma^{-\frac{1}{2}},- (\alpha \gamma)^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}}
\right)\right\rbrace,\\
&\left\lbrace\left(1,\pm \beta^{-\frac{1}{2}},\pm i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right),
\left(1,\pm \beta^{-\frac{1}{2}},\mp i \alpha^{-\frac{1}{2}},- i (\alpha \beta)^{-\frac{1}{2}} \right)\right\rbrace.
\end{aligned}
\end{gathered}$$
The points in each set are clearly distinct and differ from each other only by multiplying two of their coordinates by -1. Now consider the following function ${\ensuremath{\mathbb{P}_{k}^{3}}} \rightarrow k$, which sends $$(a,b,c,d) \mapsto \left(\frac{ab}{cd}, \frac{ac}{bd},\frac{ad}{bc}\right) \text{ where }a,b,c,d \in k,\; abcd \neq 0.$$ This function takes a different value on each of the four sets in , as the following table shows (again we use the assumptions from on the parameters):
[| c | c | c | c |]{} Representative & $\left(\frac{ab}{cd},\frac{ac}{bd},\frac{ad}{bc}\right)$\
$(1,i,i,1)$ & $(1,1,-1)$\
$\left(1,-(\beta \gamma)^{-\frac{1}{2}}, -\gamma^{-\frac{1}{2}},\beta^{-\frac{1}{2}} \right)$ & $(-1,-\beta,-\gamma)$\
$\left(1, i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}} \right)$ & $(\alpha,-1,\gamma)$\
$\left(1,\beta^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right)$ & $(-\alpha,\beta,1)$\
This shows that the four sets in are disjoint, from which the result follows.
\[rem: computercalcs\] The points in were discovered through the use of computer calculations for specific parameter choices. The solutions obtained indicated the general form of the points stated above.
We now introduce some new notation.
\[not: gammaprime\] Recall the definition of $\pi_1$ given in Notation \[not: projmap\]. Define $\Gamma'$ to be the set of points $\pi_1(q)$ for $q$ in . Thus $|\Gamma'|=16$, with Lemma \[lem: ptschemecontains\] implying that $\Gamma' \cup \{e_0,e_1,e_2,e_3\} \subset \Gamma$.
Whilst we have not yet proved that the 20 points in Lemma \[lem: ptschemecontains\] constitute the whole of $\Gamma_2$, we can still find the order of $\phi$.
\[lem: phiaut\] The scheme automorphism $\phi$ has order 2. In particular, it fixes 8 of the points given in Lemma \[lem: ptschemecontains\].
We will exploit some of the observations made after [@vancliff1998some Definition 1.4] with regard to graded skew Clifford algebras, which remain valid in our situation. Notice that the multilinearisations in are invariant under the map $v_{i1} \leftrightarrow v_{i2}$. Thus $\Gamma_2$ is invariant under the automorphism which switches components of the ambient space ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$.
Combining this observation with the description of $\Gamma_2$ in , we conclude that $(p,p^{\phi}) \in \Gamma_2$ if and only if $(p^{\phi},p) \in \Gamma_2$. But $\Gamma_2$ is the graph of $\Gamma$ under the automorphism $\phi$, hence we must have $p=p^{\phi^{2}}$. Consequently, $\phi$ has order at most 2.
It is clear by observation that 8 of the points in $\Gamma$ exhibited in Lemma \[lem: ptschemecontains\] are fixed by $\phi$, including those of the form $e_j$. The remaining 12 points of $\Gamma$ that we have given each have order 2 under this automorphism. Consequently, $\phi$ has order 2.
We end this section by remarking that the conclusion of Lemma \[lem: phiaut\] is surprising, since generically the associated automorphism of the point scheme of $A(\alpha,\beta,\gamma)$ will have infinite order.
Modules over the twist {#sec: modulesoversklytwist}
----------------------
In this section we study modules over $A(\alpha,\beta,\gamma)^{G,\mu}$, which will be denoted by $A^{G,\mu}$ when it is possible to omit the parameters. Our first result does not fit neatly into one of the sections that follows, thus we state it here. It concerns the dimension of the line scheme of $A^{G,\mu}$ – we will use the formulation of the line scheme that is used in [@shelton2002schemes Lemma 2.5].
\[prop: linescheme1dim\] Assume that $\text{char}(k) = 0$. Then the line scheme of $A(\alpha,\beta,\gamma)^{G,\mu}$ is 1-dimensional for generic parameters.
A relation in $A^{G,\mu}$ can be written in the form $\sum_{i=1}^{6} t_i f_i^{\mu}$ for some scalars $t_i \in k$. The line scheme of $A^{G\,\mu}$ is the locus $(t_1,\ldots,t_6) \in {\ensuremath{\mathbb{P}_{k}^{5}}}$ for which the matrix $$\begin{pmatrix}
0 & t_1+t_2 & t_3+t_4 & t_5+t_6 \\
t_2-t_1 & 0 & \gamma t_5 + t_6 & \beta t_3 -t_4 \\
t_4 - t_3 & t_6 - \gamma t_5 & 0 & -\alpha t_1 - t_2 \\
t_6 - t_5 & -\beta t_3 - t_4 & \alpha t_1 - t_2 & 0
\end{pmatrix}$$ has rank less than 3 by [@shelton2002schemes Lemma 2.5]. The Macaulay2 code given by Code \[code: lineschemeagmu\] in Appendix \[sec: sklyanincalcs\] shows that this scheme is indeed 1-dimensional. Of course, this code is only valid in characteristic 0 and for generic parameters, hence our assumptions.
### Point modules and their annihilators {#subsec: pointmodbehav}
We already know the existence of 20 isomorphism classes of point modules through their parameterisation by $\Gamma$ (see Notation \[not: gammaprime\]). We study the point modules corresponding to these 20 points, although some of our results hold for a general point module. Since we show that there are no further point modules up to isomorphism in Theorem \[prop: finitepointscheme\], such considerations will eventually prove superfluous.
Let us first describe the behaviour of point modules of $A^{G,\mu}$ under the twisting operation. To avoid confusion we introduce the following notation.
\[not: tilde\] Point modules over $A$ will be denoted by $M_p$, while those over $A^{G,\mu}$ will be denoted by $\widetilde{M}_p$.
\[prop: ptmodbehav\] Let $\widetilde{M}_p=A^{G,\mu}/I_p$ denote the point module corresponding to the point $p \in \Gamma$. If $p$ is fixed by $\phi$ then $I_p$ is a two-sided ideal and $\widetilde{M}_{p} \cong
\widetilde{M}_p[1]_{\geq 0}$. If $p^{\phi}\neq p$ then we have isomorphisms $$\widetilde{M}_{p^{\phi}} \cong \widetilde{M}_p[1]_{\geq 0} \; \text{ and } \; \widetilde{M}_p \cong
\widetilde{M}_{p^{\phi}}[1]_{\geq 0}.$$
Since $\phi$ has order 2, this is a consequence of Theorem \[thm: pointschemenice\] and Definition \[defn: ptschemeautomorphism\].
We will now consider the annihilators of these modules. Our first observation is that all such ideals are prime; point modules are 1-critical and therefore by [@artin1991modules Proposition 2.30(vi)] their annihilators are prime ideals.
For a point module $\widetilde{M}_p$ with $p \in \Gamma$, one has $$\text{Ann}_{A^{G,\mu}}(\widetilde{M}_p)=\bigcap_{n \in {\ensuremath{\mathbb{N}}}} \text{Ann}_{A^{G,\mu}}(\widetilde{M}_p)_{n}.$$
Suppose that $p$ is fixed by $\phi$. The annihilator of each graded piece of $\widetilde{M}_p$ is the same, since it is isomorphic to its shifts by Proposition \[prop: ptmodbehav\]. Thus the annihilator of such a point module is precisely the defining ideal of the point module, which is two-sided by the same result.
On the other hand, the annihilator of the point modules which have order 2 under the shift is $\text{Ann}_{A^{G,\mu}}(\widetilde{M}_p)_0 \cap \text{Ann}_{A^{G,\mu}}(\widetilde{M}_p)_1$. By Proposition \[prop: ptmodbehav\] the annihilator of such a module is $I_p \cap I_{p^{\phi}}$. In particular, this means that this intersection of right ideals is a two-sided ideal.
From now until the beginning of §\[subsec: fatpoints\] let us assume that $p$ has order 2 under $\phi$. We would like to determine the answers to the following two questions:
\[que: annihilator\]
- What is the Hilbert series of $I_p \cap I_{p^{\phi}}$?
- Can generators for this ideal be found?
In order to answer Question \[que: annihilator\](i), we will use the isomorphism of vector spaces $$\label{eq: 2ndisovs}
\frac{I_p}{I_p \cap I_{p^{\phi}}} \cong \frac{I_p + I_{p^{\phi}}}{I_{p^{\phi}}}.$$ This allows us to prove the following lemma.
\[lem: hilbseriesannih\] Let $p \in \Gamma$ be a point such that $p^{\phi}\neq p$. Then $I_p + I_{p^{\phi}}$ contains two elements of degree 1 and has codimension 2 inside $A^{G,\mu}$ in all higher degrees.
Since $(I_{p})_{1}$ and $(I_{p^{\phi}})_{1}$ both have codimension 1 inside $A^{G,\mu}_{1}$, if $A^{G,\mu}_{1} \not\subset I_p + I_{p^{\phi}}$ then we would have $I_p=I_{p^{\phi}}$, which is a contradiction. Thus $A^{G,\mu}_{1} \subset I_p + I_{p^{\phi}}$. By Theorem \[thm: 4sklytwistprops\], $A^{G,\mu}$ is generated in degree 1, thus $A^{G,\mu}_{\geq 1}=I_p + I_{p^{\phi}}$.
Using this information together with allows us to obtain the following relation of Hilbert series: $$\begin{gathered}
\begin{align*} H_{I_{p} \cap I_{p^{\phi}}}(t) = H_{I_{p}}(t)+H_{I_{p^{\phi}}}(t)-H_{A^{G,\mu}_{\geq 1}}(t) &=
2\left(\frac{1}{(1-t)^4}- \frac{1}{(1-t)} \right) - \left( \frac{1}{(1-t)^4}-1 \right) \\
&=\frac{1-(1-t)^3(1+t)}{(1-t)^4}.
\end{align*}\end{gathered}$$ Thus $H_{A^{G,\mu}/(I_{p} \cap I_{p^{\phi}})} = \frac{1+t}{1-t}$. Consequently, the ideal $I_{p} \cap I_{p^{\phi}}$ has codimension 2 in all degrees greater than 0, which proves the result.
We can now describe the generators of the intersection $I_{p} \cap I_{p^{\phi}}$, thus providing an answer to Question \[que: annihilator\](ii).
\[prop: ptmodintersectgens\] Let $p \in \Gamma'$ be a point for which $p^{\phi}\neq p$. Then $I_{p} \cap I_{p^{\phi}}$ is generated as a two-sided ideal by two degree 1 elements. The generators of this ideal for each of the 6 orbits of order 2 are given respectively by:
- $(rs v_0 + v_1, r v_3 - s v_2)$ for $r=\gamma^{-\frac{1}{2}},\; s=\pm \beta^{-\frac{1}{2}}$;
- $(rs v_0 + v_2,s v_1 - r v_3)$ for $r=i \gamma^{-\frac{1}{2}},\; s= \pm i \alpha^{-\frac{1}{2}}$;
- $(rs v_0 - v_3, r v_2 + s v_1)$ for $r=-\beta^{-\frac{1}{2}},\; s= \pm i \alpha^{-\frac{1}{2}}$.
Moreover, the factor ring $A^{G,\mu}/I_{p} \cap I_{p^{\phi}}$ is isomorphic to the quotient of a skew polynomial ring by a central regular element of degree 2.
One can see that $(I_{p} \cap I_{p^{\phi}})_1$ certainly contains the two elements given in the statement of the proposition in each case. We will factor out the two-sided ideal generated by these two elements and show that the factor ring obtained has the Hilbert series $\frac{1+t}{1-t}$. By Lemma \[lem: hilbseriesannih\], this is sufficient to prove the result.
Let us first consider case (i). It is clear that the factor ring is generated as an algebra by $v_0$ and $v_2$. The relations of $A^{G,\mu}$ can be rewritten in terms of these two elements; the relations $f_i^{\mu}$ for $i=1,3,5$ lie in the ideal $(rs v_0 + v_1, r v_3 - s v_2)$, while the remaining relations can be rewritten in the following manner: $$f_2^{\mu} \rightsquigarrow rs v_0^2 + \frac{s}{r}v_2^2,\;\; f_4^{\mu} \rightsquigarrow (1+s^2)[v_0,v_2]_+,\;\;
f_6^{\mu} \rightsquigarrow \left(\frac{1}{r}-r \right)[v_0,v_2]_+.$$ Note that any cancellation involving $r$ and $s$ needed to determine these ‘new’ relations does not depend on the sign of the scalar $s$, hence is valid for both choices.
Since $\beta \neq 0,-1$ and $\gamma^2 \neq 1$, we must have $$\frac{A^{G,\mu}}{(rs v_0 + v_1, r v_3 - s v_2)} \cong \frac{k\{v_0,v_2\}}{(v_0 v_2+v_2 v_0, v_0^2 + \gamma v_2^2)},$$ as $k$-algebras. This is a factor ring of the quantum plane $k_{-1}[v_0,v_2]$ by the ideal generated by a central regular element of degree 2. Such a ring has Hilbert series $(1+t)/(1-t)$, therefore the annihilator must be generated by the two elements as claimed.
The remainder of the proof comprises the same argument repeated in the other cases. In case (ii) the factor ring obtained by factoring out the two-sided ideal generated by $rs v_0 + v_2$ and $s v_1 - r v_3$ is generated as an algebra by $v_0$ and $v_1$. The relations of $A^{G,\mu}$ can be rewritten in terms of the two algebra generators; as in case (i), the relations $f_i^{\mu}$ for $i=1,3,5$ lie in the ideal $(rs v_0 + v_2,s v_1 - r v_3)$, while the others can be rewritten as follows: $$f_2^{\mu} \rightsquigarrow (1+s^2)[v_0,v_1]_+,\;\; f_4^{\mu} \rightsquigarrow rs v_0^2 + \frac{s}{r} v_1^2,\;\;
f_6^{\mu} \rightsquigarrow \left(\frac{1}{r}+r\right)[v_0,v_1]_+.$$
Since $\alpha \neq 0,1$ and $\gamma^2 \neq 1$, we must have $$\frac{A^{G,\mu}}{(rs v_0 + v_2,s v_1 - r v_3)} \cong \frac{k\{v_0,v_1\}}{(v_0 v_1+v_1 v_0, v_0^2 - \gamma v_1^2)},$$ as $k$-algebras. Once again, the factor ring is a factor of a skew polynomial ring by a central regular element of degree 2 and has the correct Hilbert series.
Moving on to case (iii), the factor ring obtained by factoring out the two-sided ideal generated by $rs v_0 - v_3$ and $r v_2 + s v_1$ is generated as an algebra by $v_0$ and $v_1$. Rewriting the relations of $A^{G,\mu}$ in terms of the two algebra generators we find that the relations $f_i^{\mu}$ for $i=1,3,5$ lie in the ideal $(rs v_0 - v_3,r v_2 + s
v_1)$, while the remaining relations can be transformed to the forms given below: $$f_2^{\mu} \rightsquigarrow (1-s^2) [v_0,v_1]_+,\;\; f_4^{\mu} \rightsquigarrow
\left(r-\frac{1}{r}\right)[v_0,v_1]_+,\;\; f_6^{\mu} \rightsquigarrow rs v_0^2+\frac{s}{r} v_2^2.$$
Since $\alpha \neq 0,-1$ and $\beta^2 \neq 1$, we have an isomorphism of $k$-algebras $$\frac{A^{G,\mu}}{(rs v_0 - v_3, r v_2 + s v_1)} \cong \frac{k\{v_0,v_1\}}{(v_0 v_1+v_1 v_0, v_0^2 + \beta v_1^2)}.$$ As in the previous cases, this factor ring is a quotient of a skew polynomial ring by a central regular element of degree 2. The factor ring therefore has the correct Hilbert series, which completes the proof.
The Hilbert series of these factor rings — after removing the degree 0 piece — is that of a fat point module of multiplicity 2. Despite this, one can see that they are not 1-critical as modules over $A^{G,\mu}$ because they have GK dimension 1 factor modules, namely the point modules $\widetilde{M}_p$ and $\widetilde{M}_{p^{\phi}}$.
### Fat point modules of multiplicity 2 {#subsec: fatpoints}
Our aim in this section is to apply the results from §\[sec: modules\] to $A$ and $A^{G,\mu}$. To see that both of these algebras satisfy Hypotheses \[hyp: genhypforfatpts\] one can use Theorem \[thm: 4sklytwistprops\], as well as noting that the action of $G$ on $A^{G,\mu}$ is chosen to be that which induces the $G$-grading inherited from $A$.
Before giving our first result we recap some of the geometry associated to $A$. We recall that the parameters associated to $A$ satisfy , with the following information depending upon this fact. As proved in [@smith1992regularity Propositions 2.4 and 2.5], the point modules over $A$ are parameterised by points on a smooth elliptic curve $E \subset {\ensuremath{\mathbb{P}_{k}^{3}}}$ and four extra points $e_j$ as in . In [@smith1992regularity Corollary 2.8] it is shown that the automorphism associated to the point scheme fixes the four exceptional points and is given by a translation $\sigma$ on $E$.
In order to apply the machinery of §\[sec: modules\] we need to following result.
\[lem: threegensannihilate\] For all $p \in E$ at least three coordinates of $p$ are non-zero. Furthermore, the action of $G$ on the point scheme of $A$ restricts to $E$.
Assume that $p= (p_0,p_1,p_2,p_3) \in E$. We will use [@smith1992regularity Proposition 2.5], which describes the homogeneous coordinate ring of $E$: $$\label{eq: thcroreextn}
\frac{k[y_0,y_1,y_2,y_3]}{\left(y_0^2+y_1^2+y_2^2+y_3^2,y_3^2+\left(\frac{1-\gamma}{1+\alpha}
\right)y_1^2+\left(\frac{1+\gamma}{1-\beta}\right) y_2^2\right)}.$$
The action of $G$ on the point scheme of $A$ is described prior to Lemma \[lem: actiononpoints\], see in particular. It is clear from that equation and above that if $p \in E$ then $p^g \in E$ for all $g \in G$.
Let us now prove the other part of the result. Since $e_j \notin E$ for $j=0,1,2,3$ we can assume that at least two coordinates of $p$ are non-zero. If there were two non-zero entries, $p_l$ and $p_m$ say, then the equations defining $E$ would reduce to the form $$p_l^2+p_m^2=p_l^2+ \lambda p_m^2=0,$$ for some $\lambda \in k$. The only solution when $\lambda \neq 1$ is $p_l=p_m=0$, which results in a contradiction. If $\lambda=1$ then either $\frac{1-\gamma}{1+\alpha}=1$ or $\frac{1+\gamma}{1-\beta}=1$ or $\frac{1-\gamma}{1+\alpha}=\frac{1+\gamma}{1-\beta}$.
If $\frac{1-\gamma}{1+\alpha}=1$ then one has $\alpha=-\gamma$, in which case $\beta=0$ or $\gamma=\pm 1$, contradicting . Similarly, if $\frac{1+\gamma}{1-\beta}=1$ then $\beta=-\gamma$, whereupon $\alpha=0$ or $\gamma=\pm 1$. Once again, such parameters are not permitted by . Finally, assume that $\frac{1-\gamma}{1+\alpha}=\frac{1+\gamma}{1-\beta}$. In this case one has $(1-\beta)(1-\gamma)=(1+\alpha)(1+\gamma)$, which can be rearranged to $\alpha=-\beta$ by using . This forces $\gamma=0$ or $\beta=\pm 1$, contradicting again.
The parameterisation described in the following result also has a geometric interpretation, as can be seen from Proposition \[prop: fatpointsincohE\].
\[claim: fatpoints\] $A^{G,\mu}$ has a family of fat point modules of multiplicity 2 parameterised up to isomorphism by the $G$-orbits of $E$.
Let $p \in E$. By Lemma \[lem: threegensannihilate\] at least three coordinates of $p$ are non-zero and the action of $G$ on the point scheme preserves $E$. Thus one may apply Proposition \[prop: fatpoints\] to obtain a fat point module $M_p^2$ over $A^{G,\mu}$. By Corollary \[cor: fatpointisoclasses\] the only isomorphisms between such modules in $\text{grmod}(A^{G,\mu})$ are of the form $M_p^2 \cong M_{p^g}^2$ for $g \in G$.
\[rem: qgrisoslater\]
- In fact, there are no further isomorphisms between the corresponding fat points in $\text{qgr}(A^{G,\mu})$, although we will only prove this in Corollary \[cor: qgrisos\].
- If one tries to apply the construction of Proposition \[prop: fatpoints\] to the point modules $M_{e_{j}}$, one does not obtain any fat point modules. In fact, the right $M_2(A)$-module $M^2_{e_{j}}$ becomes isomorphic to the direct sum $\widetilde{M}_{e_{j}}^2$ upon restriction to $A^{G,\mu}$.
We will use $[p]$ to denote the $G$-orbit of a point $p \in {\ensuremath{\mathbb{P}_{k}^{3}}}$; when necessary it will be made clear whether this point lies on $E$ or in $\Gamma$, where $\Gamma$ is the point scheme of $A^{G,\mu}$. This notation allows us to define for all $p \in E$ the right $A^{G,\mu}$-module $\widetilde{F}_{[p]}:=
M_{p}^2$; Proposition \[claim: fatpoints\] implies that this is well-defined and furthermore this notation is in harmony with that introduced in Notation \[not: tilde\].
Our next aim is to reverse the process of Proposition \[claim: fatpoints\] by taking direct sums of point modules over $A^{G,\mu}$. Before doing so we need to introduce some preliminary material, including the following result regarding the action of $G$ on the point scheme $\Gamma$.
\[lem: gammaorbits\] Consider $\Gamma' \cup \{e_0,e_1,e_2,e_3\}$, which is contained in the point scheme $\Gamma$. Under the action of $G$ on $\Gamma$ described prior to Lemma \[lem: actiononpoints\], $\Gamma' \cup \{e_0,e_1,e_2,e_3\}$ decomposes as the union of 8 $G$-orbits: $$\begin{gathered}
\begin{aligned}\label{eq: orbitsunderG}
\textit{Singleton orbits: }\; &[e_0],\;\;[e_1],\;\;[e_2],\;\;[e_3].\\
\textit{Order 4 orbits: }\; &\left[\left(1,(\beta \gamma)^{-\frac{1}{2}}, -\gamma^{-\frac{1}{2}},\beta^{-\frac{1}{2}}
\right)\right], \;\; \left[\left(1, i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}}
\right)\right], \\
&\left[\left(1, \beta^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right)\right],\;\;
\left[(1,i,i,1)\right].
\end{aligned}\end{gathered}$$ Furthermore, if $[p]$ is an order 4 orbit in then there exists $h \in G$ such that $(p^g)^{\phi}=(p^{g})^h$ for all $g \in G$. That is, the restriction of $\phi$ to each orbit of order 4 coincides with the action of a particular element of $G$.
One can use to verify that the orbits under the action of $G$ are as stated in .
Let us now address the second part of the statement of the lemma. Since $[(1,i,i,1)]$ is fixed pointwise by $\phi$, it is clear that for $p \in[(1,i,i,1)]$ one has $p^{\phi}=p^e$. Thus the identity element is associated to this orbit. For representatives of the remaining three orbits of order 4, we exhibit in the group element associated to their orbit: $$\begin{gathered}
\begin{aligned}\label{eq: orbitcorrespondence}
\left(1,(\beta \gamma)^{-\frac{1}{2}}, -\gamma^{-\frac{1}{2}},\beta^{-\frac{1}{2}} \right)^{\phi} &=
\left(1,(\beta \gamma)^{-\frac{1}{2}},- \gamma^{-\frac{1}{2}},\beta^{-\frac{1}{2}} \right)^{g_{1}}, \\
\left(1, i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}} \right)^{\phi} &=
\left(1, i\gamma^{-\frac{1}{2}}, (\alpha \gamma)^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}} \right)^{g_{2}}, \\
\left(1, \beta^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right)^{\phi} &=
\left(1, \beta^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}},i (\alpha \beta)^{-\frac{1}{2}} \right)^{g_{1}g_{2}}.
\end{aligned} \end{gathered}$$ One can see that there is a 1-1 correspondence between elements of $G$ and order 4 orbits of $\Gamma$ that we have discovered so far.
Assume now that $\sigma$, the associated automorphism of the point scheme of $A$, has infinite order. In [@smith1993irreducible], Smith and Staniszkis classify fat point modules of all multiplicities over $A$ under this assumption on $\sigma$. Their classification is a by-product of their work classifying the finite-dimensional simple $A$-modules. As the final remark in §4 op. cit. states, their work shows that there are four non-isomorphic fat point modules of each multiplicity $m \geq 2$. These modules can be denoted by $F(\omega^{\sigma^{m-1}})$ for some 2-torsion point $\omega \in E_2$. This notation is natural since $F(\omega^{\sigma^{m-1}})$ arises as the quotient of any line module $M_{p,q}$ associated to points $p, q \in E$ such that $p+q = \omega^{\sigma^{m-1}}$.
Let us now state a result relating point modules over $A^{G,\mu}$ with fat point modules of multiplicity 2 over $A$. Beforehand, recall from Notation \[not: tilde\] that we use tildes to denote point modules over $A^{G,\mu}$.
\[prop: sklyaninfatpts\] Consider $A$ as the invariant subring $M_2(A^{G,\mu})^G$. Four isomorphism classes of fat point modules of multiplicity 2 over $A$ arise as the restriction of modules of the form $\widetilde{M}_p^2$, where $\widetilde{M}_p$ is a point module over $A^{G,\mu}$. When $|\sigma|=\infty$ one recovers in this manner all four fat point modules of multiplicity 2 over $A$.
We remark that $A^{G,\mu}$ satisfies Hypotheses \[hyp: genhypforfatpts\], in which case we have the tools of §\[sec: modules\] at our disposal. Consider the 16 points in $\Gamma'$, which arise from the projection to the first coordinate of those in . By our assumption on scalars from , each of these points has at least three non-zero coordinates. Thus by Proposition \[prop: fatpoints\] one can construct 16 fat point modules of multiplicity 2 over $A$, of the form $\widetilde{M}_p^2$.
The 16 points we are considering are partitioned into the four $G$-orbits described in . Thus by Corollary \[cor: fatpointisoclasses\] there are precisely four isomorphism classes of such fat point modules. When $|\sigma|=\infty$, the work in [@smith1993irreducible] shows that there are precisely four isomorphism classes of fat point modules of multiplicity 2. Thus, under that hypothesis we recover each of these classes using the construction of Proposition \[prop: fatpoints\].
\[rem: degenfatptno\] Applying the construction of Proposition \[prop: fatpoints\] to the point modules $\widetilde{M}_{e_{j}}$ produces behaviour like that explained in Remarks \[rem: qgrisoslater\](ii). The right $M_2(A^{G,\mu})$-module $\widetilde{M}^2_{e_{j}}$ becomes isomorphic to the direct sum $M_{e_{j}}^2$ upon restriction to $A$.
Let $p \in \Gamma'$. Proposition \[prop: sklyaninfatpts\] allows us to associate a 2-torsion point $\omega_{[p]}\index{notation}{o@$\omega_{[p]}$} \in E$ to $p$. Thus $F(\omega_{[p]}^{\sigma})$ denotes the fat point module over $A$ which is isomorphic to the restriction of the $M_2(A^{G,\mu})$-module $\widetilde{M}_{p^{g}}^2$ for all $g \in G$.
Let us also introduce the notation $N_p:=A^{G,\mu}/I_p \cap I_{p^{\phi}}$ for points $p \in
\Gamma'$ such that $p^{\phi}\neq p$. These right $A^{G,\mu}$-modules were studied in §\[subsec: pointmodbehav\]. Although such modules are not 1-critical — as we observed after the proof of Proposition \[prop: ptmodintersectgens\] — one can use them to recover fat point modules over $A$.
\[cor: anncompseries\] Assume that $|\sigma|= \infty$ and let $p \in \Gamma'$ have order 2 under $\phi$. When regarded as a right $A$-module by restriction, the $M_2(A^{G,\mu})$-module $(N_{p}[1]_{\geq 0})^2$ has a critical composition series of length 2. Both of the composition factors are isomorphic to the fat point module $F(\omega_{[p]}^{\sigma})$.
Consider the following chain of $M_2(A^{G,\mu})$-submodules of $(N_{p}[1]_{\geq 0})^2$: $$\label{eq: dasfds}
(N_{p}[1]_{\geq 0})^2=\left( \left(\frac{A^{G,\mu}}{I_p \cap I_{p^{\phi}}}\right)[1]_{\geq 0} \right)^2 \supsetneq
\left( \left(\frac{I_p}{I_p \cap I_{p^{\phi}}}\right)[1]_{\geq 0}\right)^2 \supsetneq 0.$$
Recall from the proof of Lemma \[lem: hilbseriesannih\] that $A^{G,\mu}_{\geq 1}=I_p + I_{p^{\phi}}$. Using this fact and the second isomorphism theorem for modules, the middle term in can be rewritten as follows: $$\left( \left(\frac{I_p}{I_p \cap I_{p^{\phi}}}\right)[1]_{\geq 0}\right)^2 \cong \left( \left(\frac{I_p+
I_{p^{\phi}}}{I_{p^{\phi}}}\right)[1]_{\geq 0}\right)^2 = \left( \left(\frac{A^{G,\mu}_{\geq
1}}{I_{p^{\phi}}}\right)[1]_{\geq 0}\right)^2.$$ By Proposition \[prop: ptmodbehav\] this module is isomorphic to $\widetilde{M}_p^2$ as a right $M_2(A^{G,\mu})$-module. The other factor of the chain of submodules in can be seen to be isomorphic to $\widetilde{M}_{p^{\phi}}^2$ by using the third isomorphism theorem and Proposition \[prop: ptmodbehav\] once again. Lemma \[lem: gammaorbits\] implies that $p^{\phi} \in [p]$, since $p^{\phi}=p^g$ for the group element $g$ associated to the orbit $[p]$.
One can therefore use Proposition \[prop: sklyaninfatpts\] to conclude that, on restriction to modules over $A$, both of the factors of the chain in are isomorphic to a fat point module of the form $F(\omega_{[p]}^{\sigma})$. Thus the chain in is a critical composition series for $(N_{p}[1]_{\geq
0})^2$ of length 2.
We have seen in Propositions \[claim: fatpoints\] and \[prop: sklyaninfatpts\] that by taking direct sums of point modules over $A$ and $A^{G,\mu}$ we can recover fat point modules of multiplicity 2 over the other algebra. The next result shows that this relationship is reciprocal – direct sums of fat point modules over one algebra decompose as direct sums of point modules on restriction to the other algebra.
\[prop: fatpointsotherway\] Assume that $|\sigma| = \infty$.
- Let $p \in E$ and $\omega \in E_2$. On restriction to a module over $A=M_2(A^{G,\mu})^G$, one has $(\widetilde{F}_{[p]})^2 \cong \bigoplus_{g \in G}M_{p^{g}}$.
- Now let $p \in \Gamma'$. The $M_2(A)$-module $F(\omega_{[p]}^{\sigma})^2$ is isomorphic upon restriction to $A^{G,\mu}$ to the direct sum $\bigoplus_{g \in G}\widetilde{M}_{p^{g}}$. In this manner one recovers all 16 point modules of $A^{G,\mu}$ associated to points in $\Gamma'$.
One can prove both (i) and (ii) by a direct application of Proposition \[prop: fatpointsotherwaygen\]. We elaborate the argument with respect to the final statement of (ii). This holds because each $G$-orbit $[p] \subset \Gamma'$ corresponds uniquely to a fat point module $F(\omega_{[p]}^{\sigma})^2$ over $A$ by Proposition \[prop: sklyaninfatpts\]. Proposition \[prop: fatpointsotherwaygen\] then implies that taking a direct sum of this fat point module and restricting returns the $A^{G,\mu}$-module $\bigoplus_{g \in G}\widetilde{M}_{p^{g}}$.
We can now prove that the point scheme of $A^{G,\mu}$ consists of precisely 20 points.
\[prop: finitepointscheme\] Assume that $|\sigma|= \infty$. Then $\Gamma$, the point scheme of $A^{G,\mu}$, has the form $$\Gamma = \Gamma' \cup \{e_0,e_1,e_2,e_3\}.$$ Thus $\Gamma$ contains 20 points, each of which has multiplicity 1.
Suppose that $\widetilde{M}_p$ is a point module over $A^{G,\mu}$, where $p \in {\ensuremath{\mathbb{P}_{k}^{3}}}$. We claim that if $p \neq
e_j$ then at least three coordinates of $p$ are non-zero. To see this, consider the following manner of expressing the multilinearisations in : $$\label{eq: 4sklyaninmatrixform}
\begin{pmatrix}
-v_{11} & v_{01} & \alpha v_{31} & -\alpha v_{21} \\
v_{11} & v_{01} & -v_{31} & -v_{21} \\
-v_{21} & -\beta v_{31} & v_{01} & \beta v_{11} \\
v_{21} & -v_{31} & v_{01} & -v_{11} \\
-v_{31} & -\gamma v_{21} & \gamma v_{11} & v_{01} \\
v_{31} & v_{21} & v_{11} & v_{01} \end{pmatrix} \cdot \begin{pmatrix} v_{02} \\ v_{12} \\ v_{22} \\ v_{32}
\end{pmatrix} =0.$$ We know that $(p,p^{\phi})$ is a solution to these equations, i.e. $(p,p^{\phi}) \in \Gamma_2$. Furthermore, as $\phi$ is an automorphism this is the unique point $q \in \Gamma_2$ for which $p=\pi_1(q)$. In particular, this implies that when the coordinates of $p$ are substituted into the left-hand matrix in , that matrix must have rank less than or equal to 3.
Suppose, therefore, that two coordinates of $p$ are zero. We may assume that the remaining two coordinates are non-zero, otherwise $p$ would be one of the four points of the form $e_j$. Recall that we have assumed that $\{\alpha,\beta,
\gamma\} \cap
\{0,\pm 1\}=\emptyset$. This assumption implies that the matrix in has rank at least 4, regardless of which two coordinates of $p$ are zero. Thus we may assume that at least three coordinates of $p$ are non-zero.
The construction of Proposition \[prop: fatpoints\] shows that $\widetilde{M}_p^2$ is a fat point module of multiplicity 2 over $A$. But when $|\sigma|=\infty$ there are only four such modules up to isomorphism, hence we know that $\widetilde{M}_p^2$ must be isomorphic to one of them. Proposition \[prop: fatpointsotherway\] implies that there exists some point $z \in \Gamma'$ for which $\widetilde{M}_p^2 \cong F(\omega_{[z]}^{\sigma})$ as right $A$-modules.
Applying Proposition \[prop: fatpointsotherwaygen\] to $\widetilde{M}_p^2$ produces the right $A^{G,\mu}$-module $\bigoplus_{g \in G}\widetilde{M}_{p^{g}}$. Using the decomposition of the right $A^{G,\mu}$-module $(F(\omega_{[z]}^{\sigma}))^2$ from Proposition \[prop: fatpointsotherway\], we must have an isomorphism $$\label{eq: dualdecomp}
\bigoplus_{g \in G}\widetilde{M}_{p^{g}} \cong \bigoplus_{g \in G}\widetilde{M}_{z^{g}},$$ of right $A^{G,\mu}$-modules. But by [@smith1992the Proposition 1.5] the factors in a critical composition series of a f.g. ${\ensuremath{\mathbb{N}}}$-graded module of GK dimension 1 are unique up to permutation and isomorphism in high degree.
That result applies to the modules in , hence we may conclude that $\pi(\widetilde{M}_{p}) \cong
\pi(\widetilde{M}_{z^{g}})$ in $\text{qgr}(A^{G,\mu})$ for some $g \in G$. This implies the existence of $n \in {\ensuremath{\mathbb{N}}}$ such that $\widetilde{M}_{p^{\phi^{n}}} \cong \widetilde{M}_{\left(z^{g}\right)^{\phi^{n}}}$ in $\text{gr}(A^{G,\mu})$. Since $\Gamma$ parameterises isomorphism classes of point modules over $A^{G,\mu}$ and $\phi$ is an automorphism, we may conclude that $p= z^g$.
To finish the proof, observe that Proposition \[prop: genericpointscheme\] implies that each of the 20 points in $\Gamma$ has multiplicity 1.
Recall the discussion in §\[sec: finitedimptscheme\] regarding algebras with a 0-dimensional point scheme and a 1-dimensional line scheme being determined by this data. Propositions \[prop: linescheme1dim\] and \[prop: finitepointscheme\] imply that $A^{G,\mu}$ is a desirable AS-regular algebra to study in light of its associated geometry.
Figure \[fig: 1critical\] illustrates the behaviour of 1-critical modules of small multiplicity over $A$ and $A^{G,\mu}$ that we have uncovered in §\[subsec: fatpoints\]. We write $E^G:=E/G$ to denote the orbit space of $E$ under the group action – this will be shown in Lemma \[lem: ellcurve\] to be an elliptic curve. Bold points and lines represent modules of multiplicity 2, while those which are in regular font represent modules of multiplicity 1. The isomorphisms above right-pointing arrows arise by restricting $M_2(A)$-modules to $A^{G,\mu}$, whereas those above arrows pointing to the left arise by restricting $M_2(A^{G,\mu})$-modules to $A$. For example, the top right-facing arrow illustrates the construction of fat point modules over $A^{G,\mu}$ given in Proposition \[claim: fatpoints\], while the left-facing arrow directly below it refers to the ‘reverse process’ of Proposition \[prop: fatpointsotherway\].
at (3,14.5) ; at (9,14.5) ; at (1.5,5.5) ; at (3.5,7.5) ; at (1.5,7.5) ; at (3.5,5.5) ; at (2.5,6.5) ;
at (9,5) ; at (9,6) ; at (9,7) ; at (9,8) ; at (9.5,6.5) ; at (8,5) ; at (8,6) ; at (8,7) ; at (8,8) ; at (10,5) ; at (10,6) ; at (10,7) ; at (10,8) ; at (11,5) ; at (11,6) ; at (11,7) ; at (11,8) ;
at (2,1) ; at (4,3) ; at (2,3) ; at (4,1) ; at (3,2) ; at (8,1) ; at (10,3) ; at (8,3) ; at (10,1) ; at (9,2) ;
at (6,3) ; (4.5,2.25) .. controls (5.5,2.75) and (6.5,2.75) .. (7.5,2.25);
at (6,1) ; (4.5,1.75) .. controls (5.5,1.25) and (6.5,1.25) .. (7.5,1.75);
at (6,7.6) ; (4.5,6.75) .. controls (5.5,7.25) and (6.5,7.25) .. (7.5,6.75);
at (6,5.4) ; (4.5,6.25) .. controls (5.5,5.75) and (6.5,5.75) .. (7.5,6.25);
at (6,12.6) ; (4.5,11.75) .. controls (5.5,12.25) and (6.5,12.25) .. (7.5,11.75);
at (6,10.4) ; (4.5,11.25) .. controls (5.5,10.75) and (6.5,10.75) .. (7.5,11.25);
at (2.7,12.3) ; at (9.25,12.3) ;
at (3.5,12.6) ; at (3.48,12.5) ; at (10.14,12) ; at (10.12,12) ;
plot \[smooth\] coordinates [(1,10) (3.6,12) (2.75,13) (1.9,12) (4.2,10)]{};
plot \[smooth\] coordinates [(7.8,10) (10.1,12) (9.25,13) (8.4,12) (11,10)]{};
Twisting a factor ring of the Sklyanin algebra {#chap: thcrtwist}
==============================================
The 4-dimensional Sklyanin algebra $A(\alpha,\beta,\gamma)$ contains two central elements of degree 2; this is proved in [@smith1992regularity Corollary 3.9], with such elements having the form $$\label{eq: 4sklyanincentre}
\Omega_1:=-x_0^2+x_1^2+x_2^2+x_3^2,\;\; \Omega_2:=x_1^2+\left(\frac{1+\alpha}{1-\beta}
\right)x_2^2+\left(\frac{1-\alpha}{1+\gamma} \right)x_3^2.$$
The same result shows that there is a isomorphism between the ring obtained by factoring out the ideal generated by these two central elements and the twisted homogeneous coordinate ring $B(E,{\ensuremath{\mathcal{L}}},\sigma)$. Recall that $E$ is a smooth elliptic curve, ${\ensuremath{\mathcal{L}}}:={\ensuremath{\mathcal{O}}}_E(1)$ a very ample invertible sheaf with 4 global sections, and $\sigma$ an automorphism of $E$ given by translation by a point.
In this chapter we will assume the following hypothesis, which we highlight to avoid repetition. We remark that there are no known conditions on the parameter triple $(\alpha,\beta,\gamma)$ which imply that it is true.
\[hypsing: siginforder\] The automorphism $\sigma$ has infinite order.
Since $\sigma$ is given by a translation by a point, under Hypothesis \[hypsing: siginforder\] there are no points fixed by $\sigma$. The ring $B(E,{\ensuremath{\mathcal{L}}},\sigma)$ will be denoted by $B$ whenever there is no ambiguity in doing so.
The aim of the chapter is to investigate a twist of $B$ related to the twist that we studied in Chapter \[chap: sklyanin\]. The twist — which we denote by $B^{G,\mu}$ — is shown in Theorem \[thm: thcrtwistprops\] to be Auslander-Gorenstein and Cohen-Macaulay. We then establish the centre of the twist in Proposition \[prop: centrethcrtwist\], which allows us to determine the centre of $A^{G,\mu}$ in Corollary \[cor: centretwistskly\]. The main result of this chapter appears in §\[subsubsec: geomdescrthcrtwist\], in which we study $B^{G,\mu}$ in terms of Artin and Stafford’s classification of noncommutative curves; Theorem \[thm: geomdescthcrtwist\] describes the twist in terms of geometry, and this description allows us to determine the irreducible objects in $\text{qgr}(B^{G,\mu})$ in Proposition \[prop: onlyfatpoints\].
Properties of the twist {#subsec: propthcrtwist}
-----------------------
In this section we give some elementary properties of $B^{G,\mu}$. Let us consider the grading defined on $A$ by . The two central elements exhibited in are homogeneous with respect to both this grading and the [$\mathbb{N}$]{}-grading. This means that the cocycle twist performed in §\[subsec: 4sklyanintwistandptscheme\] can also be applied to $B$. In addition, $\Omega_1$ and $\Omega_2$ lie in the identity component of the $G$-grading. Using Lemma \[prop: stillregular\] we can see that these elements must remain central and regular in the twist. In fact, we can say more: by [@smith1992regularity Theorem 5.4] $\Omega_1$ and $\Omega_2$ form a regular sequence of central elements in $A$. Applying Lemma \[prop: stillregular\] to successive factor rings shows that this remains true for $\Theta_1$ and $\Theta_2$ in $A^{G,\mu}$.
From the discussion of the previous paragraph, the following lemma is straightforward.
\[lem: twistthcr\] The degree 2 elements $$\label{eq: 4sklyanintwistcentre}
\Theta_1:=-v_0^2+v_1^2+v_2^2-v_3^2,\;\; \Theta_2:=v_1^2+\left(\frac{1+\alpha}{1-\beta}
\right)v_2^2-\left(\frac{1-\alpha}{1+\gamma} \right)v_3^2,\index{notation}{t@$\Theta_1, \Theta_2$}$$ in $A^{G,\mu}$ are central and form a regular sequence. Moreover, there is an isomorphism of $k$-algebras $$\label{eq: isofactors}
B^{G,\mu}:=B(E,{\ensuremath{\mathcal{L}}},\sigma)^{G,\mu}\index{notation}{b@$B(E,{\ensuremath{\mathcal{L}}},\sigma)^{G,\mu}$}=\left(\frac{A(\alpha,
\beta,\gamma)}{(\Omega_1,\Omega_2)} \right)^{G,\mu} \cong \frac{A(\alpha, \beta,\gamma)^{G,\mu}}{(\Theta_1,\Theta_2)}.$$
To complete the proof we only need to apply Lemma \[lem: defrelns\] to $A(\alpha, \beta,\gamma)$ and the ideal $(\Omega_1,\Omega_2)$. This confirms that the isomorphism in holds.
\[rem: genindeg1Bgmu\] Since $A^{G,\mu}$ is generated in degree 1, the isomorphism in implies that $B^{G,\mu}$ must share this property. This fact will be crucial in the proof of Theorem \[thm: geomdescthcrtwist\] later on.
We will now prove that the twist has several further properties. Note that the result holds regardless of Hypothesis \[hypsing: siginforder\], which is assumed throughout this chapter.
\[thm: thcrtwistprops\] $B^{G,\mu}$ has the following properties:
- it is universally noetherian;
- it has GK dimension 2;
- it is Auslander-Gorenstein of dimension 2;
- it satisfies the Cohen-Macaulay property;
- it is Koszul.
For (ii), observe that Lemma \[lem: hilbseries\] implies that $\text{GKdim}(B^{G,\mu})=\text{GKdim}(B)$, with the latter being equal to 2 by [@artin1990twisted Proposition 1.5]. To prove (iv) we note that $A$ is Cohen-Macaulay and therefore so is $B$ by [@levasseur1992some Corollary 6.7]. Applying Proposition \[prop: cohenmac\](i) shows that $B^{G,\mu}$ is also Cohen-Macaulay.
By Theorem \[thm: 4sklytwistprops\], $A^{G,\mu}$ has properties (i) and (iii), therefore it suffices to show that these two properties are preserved by going to the factor ring. Lemma \[lem: twistthcr\] shows that $\Theta_1$ and $\Theta_2$ form a regular sequence of normal elements in $B^{G,\mu}$, in which case using [@artin1999generic Proposition 4.9(1)] shows that (i) is true. Applying [@levasseur1992some Theorem 3.6(2)] twice proves (iii). Finally, $B$ is Koszul by [@stafford1994regularity Theorem 3.9]. Using Proposition \[prop: koszul\] shows that this is also true for $B^{G,\mu}$.
\[rem: bgmunilp\] Whereas $B$ is a domain — which follows from the irreducibility of $E$ — $B^{G,\mu}$ contains nilpotent elements in degree 1. Let $v=v_0-i v_1 - i v_2 -v_3$. One can compute by hand or using the computer program Affine that $$\label{eq: nilpotentelement}
v^2 = -\Theta_1-i f_{2}^{\mu}-i f_4^{\mu} - f_6^{\mu} = 0.$$
Since $v^2$ belongs to the relations of $B^{G,\mu}$ (which are $G$-invariant), so must $(v^g)^2=(v^2)^g$ for all $g \in
G$. Thus the other elements in the same $G$-orbit as $v$ are also nilpotent. Curiously, these elements are linearly independent and therefore they generate $A^{G,\mu}$ as an algebra. In that ring the square of each generator is central, since only terms of the form $f_i^{\mu}$ vanish in .
The fact that $B^{G,\mu}$ is not a domain illustrates that Zhang twists do not necessarily preserve this property (see Remark \[rem: domain\] for more discussion on this point).
While it may not be a domain, $B^{G,\mu}$ is prime – this will be proved in Corollary \[cor: actuallyprime\]. We are not in a position to prove this yet, however we prove that $B^{G,\mu}$ is semiprime in the next result. The following definition is needed beforehand: given a ring $R$ on which a finite group $G$ acts by ring automorphisms, we say that $R$ has no *additive $|G|$-torsion* if the $|G|r \neq 0$ for all $r \in R$.
\[prop: semiprime\] $B^{G,\mu}$ is semiprime.
As it is a matrix ring over a domain, $M_2(B)$ is certainly semiprime. Note that $\text{char}(k)\nmid |G|$, hence $|G|
\in k$ acts faithfully on $M_2(B)$ and thus there is no additive $|G|$-torsion. Since $B^{G,\mu}=M_2(B)^G$, one can apply [@montgomery1980fixd Corollary 1.5(1)] to obtain the result.
The next result of this section will show that the centre of $B^{G,\mu}$ is trivial. As a consequence we can determine the centre of $A^{G,\mu}$ as well. To do so we will need to use the following property.
\[defn: projsimple\] A $k$-algebra $A$ is *projectively simple* if for any two-sided ideal of $A$, $I$ say, one has $\text{dim}_k(A/I) < \infty$.
The ring $B$ is projectively simple when $\sigma$ has infinite order by [@rogalski2006proj Proposition 0.1]. Let us now determine the centre of $B^{G,\mu}$.
\[prop: centrethcrtwist\] The centre of $B^{G,\mu}$ is trivial, that is $Z(B^{G,\mu})=k$.
Suppose that $f \in Z(B^{G,\mu})$. We can assume without loss of generality that $f$ is homogeneous with respect to both the ${\ensuremath{\mathbb{N}}}$-grading and the $G$-grading: the homogeneous components (with respect to either grading) of a central element must also be central. Assume further that it has strictly positive [$\mathbb{N}$]{}-degree. We can untwist the factor ring $B^{G,\mu}/(f)$ to obtain $B/(f')$, where $f'$ is normal by Proposition \[prop: stillregular\]. Moreover, $f'$ is regular since $B$ is a domain, hence $f$ is regular in $B^{G,\mu}$ by the same result. One may therefore apply [@levasseur1992some Lemma 5.7], which implies that $\text{GKdim }B/(f')=\text{GKdim }B -1$. By [@artin1990twisted Proposition 1.5], $B$ has GK dimension 2, hence $\text{GKdim }B/(f')=1$. However, $B$ is projectively simple as noted prior to the proposition, therefore by definition any factor ring must be finite-dimensional. This contradiction implies that $Z(B^{G,\mu})=k$.
\[cor: centretwistskly\] The centre of $A^{G,\mu}$ is $Z(A^{G,\mu})=k[\Theta_1,\Theta_2]$.
We will follow the method of [@levasseur1993modules Proposition 6.12]. Lemma \[lem: twistthcr\] shows that $B^{G,\mu}$ is obtained from $A^{G,\mu}$ by factoring out the ideal $(\Theta_1,\Theta_2)$, while Proposition \[prop: centrethcrtwist\] implies that $Z(B^{G,\mu})=k$. Combining these facts together enables one to conclude that $Z(A^{G,\mu}) \subset k+(\Theta_1,\Theta_2)$.
Our next step is to show that $Z(A^{G,\mu}/(\Theta_1))=k[\Theta_2]$ where, once again, knowledge of the centre of $B^{G,\mu}$ tells us that $Z(A^{G,\mu}/(\Theta_1)) \subset k + (\Theta_2)$. Suppose that the centre strictly contains $k[\Theta_2]$ and choose a central element $w \notin k[\Theta_2]$ of minimal degree. We can write $w=\Theta_2 w'$ for some $w'$ of strictly smaller degree. Appealing to Proposition \[prop: stillregular\] we know that $\Theta_2$ is regular and therefore $w'$ must also be central. Since it is of lower degree than $w$ we must have $w' \in k
[\Theta_2]$, which gives a contradiction.
Now let $z \in Z(A^{G,\mu})$ be homogeneous of minimal degree such that $z \notin k[\Theta_1,\Theta_2]$. By the previous paragraph, $z=\Theta_1 y + f$ for some $y \in A^{G,\mu}$ and $f \in k[\Theta_2]$. Note that $\Theta_1 y$ and $f$ have the same degree as $z$ if they are non-zero. Since $z$ and $f$ are central, $\Theta_1 y$ must be too. As argued in the previous paragraph, $\Theta_1$ is regular which implies that $y$ is central as well. But $y$ has strictly lower degree than $z$, contradicting our original assumption.
Corollary \[cor: centretwistskly\] allows us to show that $A^{G,\mu}$ is not isomorphic to one of the previously discovered examples of AS-regular algebras with 20 point modules. Since we have used the computer program Affine in the proof, the result holds only for generic parameters values.
\[thm: new\] Assume that a 4-dimensional Sklyanin algebra $A(\alpha,\beta,\gamma)$ is associated to the elliptic curve $E$ and automorphism $\sigma$, with $|\sigma|=\infty$. Then for generic parameters $\alpha,\beta$ and $\gamma$, $A(\alpha,\beta,\gamma)^{G,\mu}$ is not isomorphic to any of the previously studied examples in the literature of AS-regular algebras of dimension 4 with 20 point modules.
As remarked in [@lebruyn1995central §4], graded Clifford algebras are finite over their centre. By Corollary \[cor: centretwistskly\] the centre of $A^{G,\mu}$ has GK dimension 2, and one can then see from [@krause2000growth Proposition 5.5] that $A^{G,\mu}$ cannot be finite over its centre. Thus $A^{G,\mu}$ is not a graded Clifford algebra. It therefore suffices to consider the algebras studied in Examples 5.1 and 5.2 from [@cassidy2010generlizations], since they encompass the remaining examples in the literature that are not finite over their centre (see [@shelton2001koszul]).
Let us begin with the algebras in [@cassidy2010generlizations Example 5.1], letting $k$ be a field of characteristic 0. Over $k$, the algebras in question are generated by four degree 1 elements $x_1,x_2,x_3$ and $x_4$, subject to the defining relations $$\begin{array}{lll}
x_4x_1 - i x_1 x_4, & x_3^2-x_1^2, & x_3x_1 - x_1x_3 + x_2^2, \\ \relax
x_3x_2 - i x_2 x_3, & x_4^2-x_2^2, & x_4x_2 - x_2x_4 + \gamma x_1^2,
\end{array}$$ where $i^2 =-1$ and $\gamma \in k^\times$. We follow Cassidy and Vancliff’s notation and denote such an algebra by $A(\gamma)$. If $\gamma = \pm 2$ then $A(\gamma)$ does not have 20 point modules, therefore let us assume that $\gamma
\neq \pm 2$. Computer calculations show that the elements $x_1^4$ and $x_2^4$ are central in $A(\gamma)$. Moreover, there are no nontrivial central elements of smaller degree. By Corollary \[cor: centretwistskly\] there are three central elements of degree 4 in $A^{G,\mu}$, namely $\Theta_1^2$, $\Theta_2^2$ and $\Theta_1\Theta_2$. If there was an ${\ensuremath{\mathbb{N}}}$-graded isomorphism $A(\gamma) \cong A^{G,\mu}$ then either $x_1^4$ or $x_2^4$ would map to some $\Theta_i^2$. But there are no central elements of degree 2 in $A(\gamma)$, which is a contradiction.
It remains to deal with the algebras studied in [@cassidy2010generlizations Example 5.2]. Once again, such algebras are generated over the base field $k$ by four degree 1 elements $x_1,x_2,x_3$ and $x_4$, subject to the relations $$\begin{array}{lll}
x_3x_1 + x_1 x_3 - \beta_2 x_2^2, & x_4x_2 + x_2x_4 - x_3^2, & x_4x_1 + x_1x_4 - \alpha_2x_3^2, \\ \relax
\alpha_1 x_3^2 + \beta_1 x_2^2 - x_1^2, & x_2x_3-x_3x_2, & x_2^2 - x_4^2.
\end{array}$$ Here $\alpha_1,\alpha_2,\beta_1, \beta_2 \in k$, with these parameters satisfying the conditions $$\alpha_2 (\alpha_2 -1) = 0 \;\text{ and } \; (\alpha_1^2 + \alpha_2^2 \beta_1)(\beta_1^2 + \beta_2^2 \alpha_1) \neq 0.$$
One can use the computer program Affine to show that for any choice of parameters the following degree 4 elements are central: $$x_1^4, \;\; x_3^4, \;\; x_1^2x_3^2 , \;\; (x_2x_1)^2+(x_1x_2)^2, \;\; (x_4x_3)^2+(x_3x_4)^2.$$ As for the previous class of examples, this contradicts the fact that $\text{dim}_kZ(A^{G,\mu})_4=3$.
Modules over the twist {#subsec: modulesthcrtwist}
----------------------
We will continue our study of $B^{G,\mu}$ by showing that it does not possess any point modules. Further information about 1-critical modules over the twist will be given in §\[subsubsec: geomdescrthcrtwist\] once we have described a geometric construction of it.
Although $B$ has point modules parameterised by the elliptic curve $E$, when the associated automorphism $\sigma$ has infinite order $B^{G,\mu}$ can have at most 20 point modules by Lemma \[lem: twistthcr\] and Theorem \[prop: finitepointscheme\]. We will show in Proposition \[prop: bgmunoptmodules\] that the point scheme of $B^{G,\mu}$ is empty, although we need a useful, more general preliminary result. When applied to point modules it can be used to understand the action given in . The lemma uses the notation we introduced in Notation \[not: idealaut\].
\[lem: annGinvariant\] Let $M$ be a module over an algebra $A$, on which a finite group $G$ acts by algebra automorphisms. Then $\text{Ann}_A(M^g)=g^{-1}(\text{Ann}_A(M))$, and the ideal $\bigcap_{g \in G} \text{Ann}_A(M^g)$ is $G$-invariant. In particular, if $G$ acts by graded automorphisms then the ideal consisting of elements which annihilate all point modules over $A$ is $G$-invariant.
Let $a \in \text{Ann}_A(M^g)$. Then for all $m \in M$, $ma^g=0$, hence $a^g \in \text{Ann}_A(M)$. This implies that $a
\in g^{-1}(\text{Ann}_A(M))$. The other inclusion is proved by the reverse argument. By the first part of the result $$\label{eq: annptmod}
\bigcap_{g \in G} \text{Ann}_A(M^g) = \bigcap_{g \in G} g^{-1}(\text{Ann}_A(M)).$$ Writing the ideal in this manner makes it clear that it is preserved under the action of $G$.
As we saw prior to Notation \[not: idealaut\], if $G$ acts by ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms and $M$ is a point module over $A$, then $M^g$ is also a point module for all $g \in G$. The result is now clear.
Having proved this general lemma we can now return to our standard notational assumptions, thus $A$ denotes a 4-dimensional Sklyanin algebra associated to the elliptic curve $E$ and automorphism of infinite order $\sigma$, while $G=(C_2)^2$.
There are two ways to show that the point scheme of $B^{G,\mu}$ is empty. The first is by abstract arguments and the second is more computational. We relegate the latter to Appendix \[app: calcptmodnonproof\].
\[prop: bgmunoptmodules\] $B^{G,\mu}$ has no point modules.
Let $\Gamma''$ denote the point scheme of $B^{G,\mu}$ and suppose that it is non-empty. By Theorem \[thm: thcrtwistprops\](i), $B^{G,\mu}$ is strongly noetherian. We can therefore use [@rogalski2008canonical Theorem 1.1] to conclude that there exists a graded homomorphism $$\label{eq: canmaptothcr}
B^{G,\mu} \rightarrow B(\Gamma'',\mathcal{M},\phi),$$ that is surjective in high degree, where $\phi$ is an automorphism of $\Gamma''$ and $\mathcal{M}$ is a $\phi$-ample invertible sheaf. We know that the point scheme of $B^{G,\mu}$ is 0-dimensional, therefore by [@artin1990twisted Proposition 1.5] the GK dimension of $B(\Gamma'',\mathcal{M},\phi)$ must be 1.
Let $I$ denote the kernel of the map in . Since it is surjective in high degree there must exist $n \in {\ensuremath{\mathbb{N}}}$ such that $$\left(\frac{B^{G,\mu}}{I}\right)_{\geq n} \cong B(\Gamma'',\mathcal{M},\phi)_{\geq n},$$ as $k$-vector spaces. Thus $B^{G,\mu}/I_{\geq n}$ also has GK dimension 1.
In addition to the existence of the homomorphism in , [@rogalski2008canonical Theorem 1.1] states that $I_{\geq n}$ consists of the elements of $B^{G,\mu}$ that annihilate all of its point modules. By Lemma \[lem: annGinvariant\] this ideal is $G$-invariant, therefore one can twist the $G$-grading on $B^{G,\mu}/I_{\geq n}$ by the 2-cocycle $\mu$. As $\mu$ has order 2, upon twisting one obtains a factor ring $B/I'$ for some ideal $I'$. The factor ring has GK dimension 1 by Proposition \[prop: gkdim\].
By [@artin1995noncommutative Theorem 4.4], ideals of $B$ correspond to $\sigma$-invariant closed subschemes of $E$. Since $\sigma$ is given by translation by a point of infinite order on $E$, there are no nontrivial such subschemes. Thus $I'$ must equal $0$ or have finite codimension in $B$. In either case we get a contradiction: in the former because $B$ has GK dimension 2, and in the latter because $B/I'$ would be finite-dimensional.
We saw in Remark \[rem: bgmunilp\] that $B^{G,\mu}$ contains zero divisors. The existence of such elements could also be proved using the previous proposition and [@artin1995noncommutative Thm 0.2] as follows. The latter result implies that if $B^{G,\mu}$ were a domain then there would be an equivalence of categories $\text{qgr}(B^{G,\mu}) \simeq
\text{coh}(Y)$ for some projective curve $Y$. But this would mean that $B^{G,\mu}$ had a family of pairwise non-isomorphic point modules parameterised by $Y$, contradicting the conclusion of Proposition \[prop: bgmunoptmodules\].
While $B^{G,\mu}$ does not possess any point modules when $\sigma$ has infinite order, it does have a family of fat point modules of multiplicity 2 parameterised by $G$-orbits on $E$. As we will see in Proposition \[prop: onlyfatpoints\], such modules arise as the restriction to $B^{G,\mu}$ of $M_2(B)$-modules of the form $M_p^2$, where $M_p$ is a point module over $B$.
Structure in relation to Artin-Stafford theory {#subsubsec: geomdescrthcrtwist}
----------------------------------------------
In this section we discuss $B^{G,\mu}$ from the viewpoint of Artin and Stafford’s classification of noncommutative projective curves in [@artin1995noncommutative] and [@artin2000semiprime]. In the first of these papers the authors classify connected graded $k$-algebras which are noetherian domains of quadratic growth using geometric techniques. This classification is extended in the latter paper to semiprime algebras.
By Theorem \[thm: thcrtwistprops\] and Proposition \[prop: semiprime\], $B^{G,\mu}$ is a c.g. semiprime noetherian algebra of GK dimension 2. Thus we can study it from the viewpoint of the classification in [@artin2000semiprime]. Although a priori we only know that this ring is semiprime, it will be shown Corollary \[cor: actuallyprime\] that it is in fact a prime ring. Thus we will only describe the geometry needed in the prime case, since that will be sufficient for our needs.
### Geometric data and twisted rings {#subsec: geomdata}
Our first task is to define the geometric rings that feature in the classification. To this end, consider a projective curve $Y$ over $k$, whose function field $K$ has transcendence degree 1 over the base field. As we will describe at the end of §\[subsec: geomdata\], it is perhaps more natural in our situation to obtain such a field $K$ as the graded division ring of a prime noetherian algebra (along with some extra conditions) and then use [@hartshorne1977algebraic Chapter I, Corollary 6.12] to recover the curve $Y$.
Now let $T$ be a central simple $K$-algebra that is finite-dimensional over $K$. The structure sheaf ${\ensuremath{\mathcal{O}}}_Y$ is a subsheaf of the constant sheaf of rational functions on $Y$, namely $K$, therefore the sections of ${\ensuremath{\mathcal{O}}}_Y$ lie inside $K$. Since $K \subset T$ such sections are also contained in $T$ .
\[defn: lattice\] We say that $\mathcal{{\ensuremath{\mathcal{L}}}}$ is an *${\ensuremath{\mathcal{O}}}_Y$-lattice* in $T$ if it is a sheaf of finitely generated ${\ensuremath{\mathcal{O}}}_Y$-submodules of $T$ for which ${\ensuremath{\mathcal{L}}}K = T$.
When the term lattice is used in future we will mean the term in the sense of this definition. Given such a lattice, one can define the *left* and *right orders* of ${\ensuremath{\mathcal{L}}}$ by $$E({\ensuremath{\mathcal{L}}})=\{\alpha \in T:\alpha {\ensuremath{\mathcal{L}}}\subseteq {\ensuremath{\mathcal{L}}}\}\;\text{ and }\;E'({\ensuremath{\mathcal{L}}})=\{\alpha \in T: {\ensuremath{\mathcal{L}}}\alpha\subseteq
{\ensuremath{\mathcal{L}}}\},$$ respectively.
A lattice is said to be *invertible* if it is a locally free left $E({\ensuremath{\mathcal{L}}})$-module of rank 1. Lemma 1.10 from [@artin2000semiprime] implies that an invertible lattice is also a locally free right $E'({\ensuremath{\mathcal{L}}})$-module of rank 1.
A crucial point for us is that for invertible lattices ${\ensuremath{\mathcal{L}}}$ and $\mathcal{M}$ the product lattice ${\ensuremath{\mathcal{L}}}\mathcal{M}$ — where the product takes places inside $T$ — is isomorphic to the tensor product ${\ensuremath{\mathcal{L}}}\otimes_{E(\mathcal{M})}
\mathcal{M}$ if $E'({\ensuremath{\mathcal{L}}})=E(\mathcal{M})$. Following the notation in [@artin2000semiprime], we will denote such a tensor product by ${\ensuremath{\mathcal{L}}}\cdot \mathcal{M}$.
We can now introduce some more geometric objects that will be needed. Let $\mathcal{E}$ be a coherent sheaf of ${\ensuremath{\mathcal{O}}}_Y$-orders inside $T$, $\tau$ an automorphism of $T$, and $\mathcal{B}_1$ an invertible lattice in $T$ such that $E(\mathcal{B}_1)=\mathcal{E}$ and $E'(\mathcal{B}_1)=\mathcal{E}^{\tau}$. One can then define a sequence of sheaves $\{\mathcal{B}_n\}$ by $\mathcal{B}_n=\mathcal{B}_1 \otimes_{\mathcal{E}} \mathcal{B}_1^{\tau}
\otimes_{\mathcal{E}^{\tau}} \ldots \otimes_{\mathcal{E}^{\tau^{n-1}}} \mathcal{B}_1^{\tau^{n-1}}$. The conditions on $E(\mathcal{B}_1)$ and $E'(\mathcal{B}_1)$ imply that $\mathcal{B}_n=\mathcal{B}_1 \cdot \mathcal{B}_1^{\tau} \cdot
\ldots \cdot \mathcal{B}_1^{\tau^{n-1}}$. Using this data one can define a *sheaf of bimodule algebras* $\mathbb{B}(\mathcal{E},\mathcal{B}_1,\tau)=\bigoplus_{n \in {\ensuremath{\mathbb{N}}}}
\mathcal{B}_n$ with $\mathcal{B}_0=\mathcal{E}$ and multiplication given by the formula $\mathcal{B}_i \cdot \mathcal{B}_j^{\tau^{i}} = \mathcal{B}_{i+j}$ for all $i, j
\geq 1$ (see [@artin2000semiprime pg. 102]).
There is a notion of ampleness for sequences of sheaves such as $\{\mathcal{B}_n\}$, and this condition plays the same role for the twisted rings we will construct as $\sigma$-ampleness does for twisted homogeneous coordinate rings. While one can make such a definition in the context of coherent left $\mathcal{E}$-modules, by [@artin2000semiprime Lemma 6.1] one can reduce to the more familiar definition stated below.
\[def: amplebimodule\] Let $\{\mathcal{L}_n\}$ be a sequence of coherent $\mathcal{O}_Y$-modules. The sequence $\{\mathcal{L}_n\}$ is *ample* if for all coherent sheaves $\mathcal{G}$ of $\mathcal{O}_Y$-modules and all $n \gg 0$, the sheaf $\mathcal{G} \otimes_{\mathcal{O}_Y} \mathcal{L}_n$ is generated by global sections, and $H^1(Y,\mathcal{G} \otimes_{\mathcal{O}_Y} \mathcal{L}_n)=0$.
A priori we will not know that the sequence of sheaves we work with is ample, but this is not a problem as we discuss in Remark \[rem: ample\].
Suppose now that $\mathcal{B}_1$ is an invertible lattice such that $E(\mathcal{B}_1)=\mathcal{E}$, $E'(\mathcal{B}_1)=\mathcal{E}^{\tau}$, and the sequence of sheaves $\{\mathcal{B}_n\}$ defined above is ample in the sense of Definition \[def: amplebimodule\]. We say that $\mathcal{B}_1$ is an *ample lattice*. One does not need an ample lattice to construct the rings in the next definition, however this condition does imply that the objects constructed are noetherian. The rings in question are formed by essentially taking global sections of sheaves of bimodule algebras. In the definition we use the ideas discussed in Remark \[rem: finesse\], which allow us to view the twisted multiplicative structure in a more concrete setting.
\[defn: twistedhomring\] Let $Y$, $\tau$, $\mathcal{E}$ and $\{\mathcal{B}_n\}$ be as above. The *generalised twisted homogeneous coordinate ring* associated to this data is the ring $$B(\mathcal{E},\mathcal{B}_1,\tau)=\bigoplus_{n \in {\ensuremath{\mathbb{N}}}} H^0(Y,\mathcal{B}_n)z^n,$$ whose the multiplication is induced by that in the corresponding sheaf of bimodule algebras, i.e. $z\beta=\beta^{\tau}z$.
Such rings were originally studied in [@van1996translation], although Van den Bergh’s definitions differ slightly. For brevity we will refer to such rings as *twisted rings* in future.
Artin and Stafford’s main results need several technical hypotheses, however we can summarise them in a simplified case.
\[thm: artinstaffordmain\] Suppose that $R$ is a semiprime noetherian c.g. algebra of GK dimension 2. Then there is a Veronese subring of $R$ which in high degree is a left ideal in a twisted ring for which $\mathcal{B}_1$ is an ample lattice. Moreover, this twisted ring is a finite left module over the Veronese ring of $R$.
The geometric data needed to construct such a twisted ring from $R$ is obtained in part from certain divisors associated to the graded components $R_n$, generalising the methods of [@artin1995noncommutative]. One piece of data that is easy to recover is the curve $Y$, and we now indicate in the prime case how one can do this. Let $R$ be as in Theorem \[thm: artinstaffordmain\], with the additional hypotheses of being prime and generated in degree 1. As we saw in §\[sec: goldietheory\], such a ring has a graded quotient ring $Q_{\text{gr}}(R)=M_n(D)[z,z^{-1};\tau]$. Define $K:=Z(M_n(D))\cong Z(D)$. By [@artin1995noncommutative Theorem 1.1], $K$ has transcendence degree 1 over $k$ when the base field is algebraically closed. Using [@hartshorne1977algebraic Chapter I, Corollary 6.12] enables one to associate $K$ to a unique smooth projective curve over $k$ by considering the set of all discrete valuation rings inside $K$.
We will show that $B^{G,\mu}$ fits into the classification in a particularly nice way, in the respect that it is *isomorphic* to a twisted ring. Another nice feature in our situation is that $\mathcal{E}=\mathcal{E}^{\tau}$, which is not true in general.
### A geometric description of the twist {#subsec: geomdesc}
To obtain a geometric description of $B^{G,\mu}$, we first go back to $B$. By [@smith1994center Proposition 2.1], this ring embeds in the Ore extension $k(E)[z,z^{-1};\sigma]$, which is its graded quotient ring. The function field $k(E)$ is the graded division ring of the homogeneous coordinate ring of the elliptic curve, defined in .
The action of $G$ on $B$ extends in a natural way to an action on $k(E)[z,z^{-1};\sigma]$ by [$\mathbb{Z}$]{}-graded algebra automorphisms, where $(ab^{-1})^g=a^g (b^g)^{-1}$ for all $a, b \in B$ with $b$ homogeneous and $g \in G$. This action induces a $G$-grading on the localisation under which $B$ is a $G$-graded subring.
As shown in §\[sec: goldietheory\], one can replace the skew generator $z$ with another element in $k(E)z$ up to changing $\sigma$ by a conjugation. In our situation $k(E)$ is a field and so any conjugation is trivial. Moreover, $B_1
\subset k(E)z$ and $x_0 \in B_1 \cap B_e$, hence we can assume that $z$ is fixed by the action of $G$ without changing $\sigma$.
Note that the action of $G$ preserves the skew structure of $k(E)[z,z^{-1};\sigma]$. Indeed, for all $f \in k(E)$ and $g
\in G$ we have $$0=(zf-f^{\sigma}z)^g=z^gf^g-(f^{\sigma})^gz^g=zf^g-(f^{\sigma})^gz=((f^g)^{\sigma}-(f^{\sigma})^g)z.$$ We highlight for future reference that for all $f \in k(E)$ one has $$\label{eq: sigmagcommute}
(f^g)^{\sigma}=(f^{\sigma})^g.$$
As [@hartshorne1977algebraic Chapter I, Theorem 4.4] shows, an algebra automorphism on $k(E)$ induces a unique automorphism on the curve itself. By the induced actions of $G$ and $\sigma$ must also commute on $E$.
\[lem: actionscoincide\] This action of $G$ on $E$ coincides with that induced by the action on point modules of $A$, given in .
Both actions are a consequence of the action of $G$ on $B$; one by localisation, the other through the action on ideals defining points in $\mathbb{P}^3_k$.
This lemma, in conjunction with Proposition \[claim: fatpoints\], allows us to extend the conclusions of that proposition to $\text{qgr}(A^{G,\mu})$.
\[cor: qgrisos\] Let $p, q \in E$. There is an isomorphism $\pi(M_p^2)\cong \pi(M_q^2)$ in $\text{qgr}(A^{G,\mu})$ if and only if $q \in
[p]$.
By Corollary \[cor: fatpointisoclasses\] one has $M_p^2 \cong M_{p^{g}}^2$ in $\text{grmod}(A^{G,\mu})$ for all $g \in
G$. Thus suppose that $\pi(M_p^2)\cong \pi(M_q^2)$ for some $p,q \in E$. As in the proof of Proposition \[claim: fatpoints\], this implies the existence of $n \in {\ensuremath{\mathbb{N}}}$ such that $M_{p^{\sigma^{n}}}^2\cong M_{q^{\sigma^{n}}}^2$. Corollary \[cor: fatpointisoclasses\] then tells us that $\sigma^n(p)$ and $\sigma^n(q)$ lie in the same $G$-orbit. By and Lemma \[lem: actionscoincide\] the actions of $G$ and $\sigma$ on $E$ commute, therefore $p$ and $q$ lie in the same $G$-orbit, which completes the proof.
As $B$ is a $G$-graded subring of its graded quotient ring, we can twist both rings simultaneously and use the invariant ring construction to see that $$\label{eq: twistthcrinvariant}
B^{G,\mu}=M_2(B)^G \hookrightarrow M_2(k(E)[z,z^{-1};\sigma])^G = M_2(k(E))^G[z,z^{-1};\tilde{\tau}],$$ where $\tilde{\tau}$ denotes the induced action of $\sigma$. Here we have used the fact that $G$ acts trivially on $z$.
This embedding suggests that $M_2(k(E))^G$ plays a role in governing the underlying geometry of the algebra. Indeed, this is the case as our main result shows.
\[thm: geomdescthcrtwist\] There is an isomorphism of $k$-algebras $$B^{G,\mu} \cong B(\mathcal{E},\mathcal{B}_1,\tilde{\tau})=\bigoplus_{n \in {\ensuremath{\mathbb{N}}}} H^0(E^G,\mathcal{B}_n)z^n \subset
M_2(k(E))^G[z,z^{-1};\tilde{\tau}],$$ for the following geometric data:
- an elliptic curve $E^G := E/G$, with $\tau$ the action induced by $\sigma$, along with the morphism of curves $r:E \rightarrow E^G$ sending $p \mapsto [p]$;
- $\mathcal{E}=M_2(r_{\ast}{\ensuremath{\mathcal{O}}}_E)^G$, a sheaf of $\mathcal{O}_{E^{G}}$-orders inside $M_2(k(E))^G$ on which the automorphism $\tilde{\tau}$ acts. This automorphism restricts to the induced action of $\tau$ on $k(E^G)$;
- $\mathcal{B}_1=M_2(r_{\ast}{\ensuremath{\mathcal{L}}})^G$, an invertible lattice in $M_2(k(E))^G$, where ${\ensuremath{\mathcal{L}}}$ is as defined after .
Since the proof of Theorem \[thm: geomdescthcrtwist\] is quite technical, we will first prove several preliminary lemmas. Note that the sheaf $\mathcal{E}$ is defined on an open set $U \subseteq E^G$, with $V=r^{-1}(U)$, by $$\mathcal{E}(U):=\left[M_2(r_{\ast}{\ensuremath{\mathcal{O}}}_E)^G\right](U)=M_2((r_{\ast}{\ensuremath{\mathcal{O}}}_E)(U))^G = M_2({\ensuremath{\mathcal{O}}}_E(V))^G.$$ Sheaves of invariant objects are discussed in [@liu2002algebraic Exercises 2.2.14, 2.3.20 and 2.3.21] for example.
Before stating the first lemma, recall that $\sigma$ is given by translation by a point of infinite order on $E$.
\[lem: ellcurve\] Define $E^G:= E/G$, the orbit space of $E$ under the action of $G$. Then $E^G$ is a smooth elliptic curve, with an associated automorphism $\tau$ that is induced by $\sigma$. Furthermore, $\tau$ has infinite order and does not fix any points.
The orbit space $E^G$ is the curve associated to the fixed field $k(E)^G$, which has transcendence degree 1 over $k$. In particular we have $k(E^G)=k(E)^G$. As remarked after , the actions of $G$ and $\sigma$ commute on $E$. One may therefore conclude that there is an induced action of $\sigma$ on $E^G$. Denote this map by $\tau$, which is defined by $[p]^{\tau}:=[p^{\sigma}]$ for all $p \in E$, and suppose that $\tau$ has a fixed point $[p]$. Since $[p]$ contains 4 or fewer points and $\sigma$ is given by translation, this implies that $\sigma$ has finite order which is a contradiction. A similar argument proves that $\tau$ has infinite order.
Let us now show that $E^G$ is smooth. The singular locus of $E^G$ must be finite and preserved by $\tau$. If it were non-empty then this would imply that $\sigma$ has finite order, which is a contradiction. One can now apply Hurwitz’s Theorem [@hartshorne1977algebraic Chapter IV, Corollary 2.4] to see that $E^G$ has genus 0 or 1; if it had genus 0 then it would be birationally equivalent to [$\mathbb{P}_{k}^{1}$]{}, whose automorphisms always fix at least one point. Thus $E^G$ must have genus 1 and hence be an elliptic curve.
Before our next lemma we need to define outer and $X$-outer actions of a group. Our statement of the latter property is given for prime Goldie rings, in which case [@montgomery1980fixd Examples 3.6 and 3.7] allow us to give the formulation below.
\[def: Xouter\] Let $G$ be a finite group acting by ring automorphisms on a ring $R$. We say that $G$ is *outer* if no nontrivial element of $G$ acts by conjugation by an element of $R$. If, in addition, we assume that $R$ is prime Goldie, we say that $G$ is *X-outer* on $R$ if it is outer when extended to the Goldie quotient ring $Q(R)$.
\[lem: Xouter\] The group $G$ is outer on $M_2(k(E))$ and therefore X-outer on the ring $M_2({\ensuremath{\mathcal{O}}}_E(V))$ for each $G$-invariant affine open set $V \subset E$. Consequently, $M_2({\ensuremath{\mathcal{O}}}_E(V))^G$ is prime.
Recall that the action of $G$ on $M_2(k)$ is given in . The action of $G$ on $A$ induces the $G$-grading in . This means that the action of $G$ on the degree 1 component of the factor ring $B=A/(\Omega_1,\Omega_2)$ affords the regular representation. Note that $k(E)$ is graded division ring of $B$ (by [@smith1994center Proposition 2.1] for example). Combining these facts, we observe that each graded component of $k(E)$ under the induced $G$-grading is non-empty. Hence we can find a non-zero element $y \in k(E)_{g_{2}}$, in which case: $$\begin{pmatrix} y & 0 \\ 0 & 0 \end{pmatrix}^{g_{1}}=\begin{pmatrix} -y & 0 \\ 0 & 0 \end{pmatrix}.$$
Suppose that this action were given by conjugation by $\left( \begin{smallmatrix} a&b\\ c&d \end{smallmatrix} \right)
\in M_2(k(E))$: $$\begin{gathered}
\begin{aligned}\label{eq: conjg1calc}
\begin{pmatrix} -y & 0 \\ 0 & 0 \end{pmatrix}
&= \frac{1}{ad-bc} \begin{pmatrix} a & b \\ c & d \end{pmatrix} \begin{pmatrix} y & 0 \\ 0 & 0 \end{pmatrix}
\begin{pmatrix} d & -b \\ -c & a \end{pmatrix} \\
&= \frac{1}{ad-bc} \begin{pmatrix} ad y & -ab y \\ c d y & -bc y \end{pmatrix}.
\end{aligned}\end{gathered}$$ Certainly we would need $b=c=0$ for the correct entries of the matrix to vanish. But then would reduce to $$\label{eq: matricesneq}
\begin{pmatrix} -y & 0 \\ 0 & 0 \end{pmatrix} = \begin{pmatrix} y & 0 \\ 0 & 0 \end{pmatrix},$$ which is a contradiction.
We must repeat these calculations for the other two non-identity elements in the group. For $g=g_2$ or $g_1g_2$, choose a non-zero element $y' \in k(E)_{g_{1}}$. In both cases we have $$\begin{pmatrix} y' & 0 \\ 0 & 0 \end{pmatrix}^{g}= \begin{pmatrix} 0 & 0 \\ 0 & -y' \end{pmatrix}.$$
Using the conjugation in , one can see that $a=d=0$ must hold in order that the correct entries of the matrix vanish. However, regardless of the values of $b$ and $c$ chosen, the conjugation does not coincide with the group action. Thus the actions of $g_2$ and $g_1g_2$ do not arise from conjugation in $M_2(k(E))$ either.
Now consider a $G$-invariant affine open set $V \subset E$ and $M_2({\ensuremath{\mathcal{O}}}_E(V))$. As ${\ensuremath{\mathcal{O}}}_E(V)$ is a domain it is clear that $M_2({\ensuremath{\mathcal{O}}}_E(V))$ is prime. Furthermore, it has Goldie quotient ring $M_2(k(E))$, on which the action of $G$ is outer by the argument above. By Definition \[def: Xouter\] the action of $G$ must be X-outer on $M_2({\ensuremath{\mathcal{O}}}_E(V))$, whereupon we may apply [@montgomery1980fixd Theorem 3.17(2)] to show that $M_2({\ensuremath{\mathcal{O}}}_E(V))^G$ is a prime ring.
We may now define the sheaf of orders which will be needed for the geometric description of $B^{G,\mu}$.
\[lem: maxorder\] Define $\mathcal{E}:=M_2(r_{\ast}{\ensuremath{\mathcal{O}}}_E)^G$, which is considered as a subsheaf of the constant sheaf $M_2(k(E))^G$. Then
- the natural action of $\sigma$ on $M_2(k(E))$ restricts to an automorphism $\tilde{\tau}$ on $\mathcal{E}$, with $\tilde{\tau}$ restricting to the induced action of $\tau$ on $k(E^G)$;
- $\mathcal{E}$ is a sheaf of Dedekind prime rings and therefore a sheaf of maximal orders.
The automorphism $\sigma$ extends naturally from $k(E)$ to $M_2(k(E))$ under the trivial action on the matrix units. Since this automorphism commutes with $G$, it restricts to an automorphism $\tilde{\tau}$ on $\mathcal{E}$. It follows from the construction that the action of $\tilde{\tau}$ on $k(E)^G$ coincides with that of $\tau$.
To prove (ii), let $U \subset E^G$ be an affine open set with $V=r^{-1}(U)$. The open set $V \subset E$ is $G$-invariant, hence $\mathcal{E}(U)= M_2({\ensuremath{\mathcal{O}}}_E(V))^G$ is prime by Lemma \[lem: Xouter\]. As $G$ acts on ${\ensuremath{\mathcal{O}}}_E(V)$ by $k$-algebra automorphisms, one can view $\mathcal{E}(U)$ as a cocycle twist of the form ${\ensuremath{\mathcal{O}}}_E(V)^{G,\mu}$, where $G=(C_2)^2$ and $\mu$ is the same 2-cocycle as used to twist $B$. Consequently, we have the results of Chapter \[chap: cocycletwists\] at our disposal. In particular, ${\ensuremath{\mathcal{O}}}_E(V)$ is noetherian and has global dimension 1 since $E$ is a smooth curve, therefore we may apply Corollary \[cor: uninoeth\] and Proposition \[prop: gldim\] to see that $\mathcal{E}(U)$ is a hereditary noetherian prime (HNP) ring.
We will now show that $\mathcal{E}(U)$ contains no idempotent maximal ideals, which by [@mcconnell2001noncommutative Proposition 5.6.10] will imply that it contains no nontrivial idempotent ideals. The existence of idempotent maximal ideals is a local condition: given a sheaf $\mathcal{M}$ of maximal ideals of $\mathcal{E}$, the sheaf $\mathcal{E}/\mathcal{M}$ is supported at a single point $p \in E^G$, and the stalk $\mathcal{M}_p$ is an idempotent maximal ideal of $\mathcal{E}_p$ if and only if $\mathcal{M}(V)$ is an idempotent maximal ideal of $\mathcal{E}(V)$ for an affine open set $V \ni p$.
Let $\bigcup_{i=1}^n U_i$ be an affine open cover of $E^G$ for some $n \in {\ensuremath{\mathbb{N}}}$, which gives rise to a corresponding cover $\bigcup_{i=1}^n V_i$ of $E$ by $G$-invariant open sets with $V_i = r^{-1}(U_i)$. As the rings $\mathcal{E}(U_i)
\subset M_2(k(E))$ are PI and noetherian, they can have at most finitely many idempotent maximal ideals by [@mcconnell2001noncommutative Theorem 13.7.15]. By the remarks of the preceding paragraph, there can be at most finitely many idempotent maximal ideals in the totality of the stalks of $\mathcal{E}$. We know that $\tau$ has no finite orbits on $E^G$. This means that if a stalk $\mathcal{E}_p$ contains an idempotent maximal ideal $\mathcal{M}_p$, then there are infinitely many other stalks with such an ideal, of the form $\tilde{\tau}^n(\mathcal{M}_p)$ for $n \in {\ensuremath{\mathbb{N}}}$. This is a contradiction.
Now we may apply [@mcconnell2001noncommutative Theorem 5.6.3] to $\mathcal{E}(U)$: it contains no nontrivial idempotent ideals and therefore it is a Dedekind prime ring. In particular, condition (ii) of Theorem 5.2.10 op. cit. implies that $\mathcal{E}(U)$ is a maximal order.
\[lem: twistedring\] Define $\mathcal{B}_1:=M_2(r_{\ast}{\ensuremath{\mathcal{L}}})^G$ which, like $\mathcal{E}$, is considered as a subsheaf of the constant sheaf $M_2(k(E))^G$. Then
- $\mathcal{B}_1$ is an invertible ${\ensuremath{\mathcal{O}}}_{E^G}$-lattice;
- $\mathcal{E}$ is coherent sheaf of maximal ${\ensuremath{\mathcal{O}}}_{E^G}$-orders for which $\mathcal{E}=\mathcal{E}^{\tilde{\tau}}$;
- $\mathcal{E}= E(\mathcal{B}_1) = E'(\mathcal{B}_1)$.
We first show that $\mathcal{B}_1$ is an ${\ensuremath{\mathcal{O}}}_{E^G}$-lattice. To do this we need to prove that $M_2(k(E))^G =
\mathcal{B}_1 \cdot k(E^G)$, where it suffices to work at the level of the global sections of $\mathcal{B}_1$. Observe that the embedding in sends $B^{G,\mu}_1$ to $M_2(\text{H}^0(E,{\ensuremath{\mathcal{L}}}))^Gz$, which is precisely $\text{H}^0(E^G,\mathcal{B}_1)z$ since $$\text{H}^0(E^G,\mathcal{B}_1) = \left[M_2(r_{\ast}{\ensuremath{\mathcal{L}}})^G\right](E^G) = M_2({\ensuremath{\mathcal{L}}}(E))^G=M_2(\text{H}^0(E,{\ensuremath{\mathcal{L}}}))^G.$$ Thus $\text{H}^0(E^G,\mathcal{B}_1)$ contains the elements $$\label{eq: sorted}
\begin{pmatrix} x_0 & 0 \\ 0 & x_0 \end{pmatrix}z^{-1},\begin{pmatrix} x_{1} & 0 \\ 0 &
-x_{1} \end{pmatrix}z^{-1},\begin{pmatrix} 0 & x_{2} \\ x_{2} & 0 \end{pmatrix}z^{-1},\begin{pmatrix} 0 & -x_3 \\ x_{3}
& 0 \end{pmatrix}z^{-1},$$ where $x_0,x_1,x_2$ and $x_3$ are the degree 1 generators of $B$. The elements in are linearly independent over $k(E^G)$. But $M_2(k(E))^G$ is a 4-dimensional vector space over $k(E^G)$, which implies that $M_2(k(E))^G = \mathcal{B}_1 \cdot k(E^G)$ must hold and therefore $\mathcal{B}_1$ is an ${\ensuremath{\mathcal{O}}}_{E^G}$-lattice.
Notice that $\mathcal{B}_1$ is an $\mathcal{E}$-module on the right since $$M_2(r_*\mathcal{L})^G\cdot M_2(r_*\mathcal{O}_E)^G \subseteq
M_2(r_*\mathcal{L}\cdot r_*\mathcal{O}_E)^G \subseteq M_2(r_*\mathcal{L})^G.$$ In particular, one has $\mathcal{E} \subseteq E'(\mathcal{B}_1)$. One may argue in a similar manner for left modules to see that $\mathcal{E} \subseteq E(\mathcal{B}_1)$. Applying [@mcconnell2001noncommutative Lemma 3.1.12(i)] shows that the orders $E(\mathcal{B}_1)$ and $E'(\mathcal{B}_1)$ are equivalent to $\mathcal{E}$. However, we showed in Lemma \[lem: maxorder\] that $\mathcal{E}$ is a sheaf of maximal orders, therefore we must have $\mathcal{E}=
E(\mathcal{B}_1) = E'(\mathcal{B}_1)$. By [@artin2000semiprime Lemma 1.10(i)], $\mathcal{E}$ must be a coherent sheaf of ${\ensuremath{\mathcal{O}}}_{E^G}$-modules. The equality $\mathcal{E}=\mathcal{E}^{\tilde{\tau}}$ follows from Lemma \[lem: maxorder\](i).
Finally, we show that $\mathcal{B}_1$ is invertible over $\mathcal{E}$. Note that $\mathcal{B}_1$ is generated by its global sections (since ${\ensuremath{\mathcal{L}}}$ is), therefore $\mathcal{B}_1$ is a sheaf of f.g. modules over $\mathcal{E}$. Since $\mathcal{E}$ is a sheaf of Dedekind prime rings by Lemma \[lem: maxorder\], [@mcconnell2001noncommutative Lemma 5.7.4] implies that $\mathcal{B}_1 \subset M_2(k(E))^G$ must be a sheaf of torsionfree, projective modules contained in the Goldie quotient ring of $\mathcal{E}$. These facts imply that it must be an invertible lattice.
We are now in a position to prove Theorem \[thm: geomdescthcrtwist\].
[Theorem \[thm: geomdescthcrtwist\]]{} To begin, note that in Lemma \[lem: twistedring\] we proved that $\mathcal{E}= E(\mathcal{B}_1) = E'(\mathcal{B}_1)$, whereupon $$\label{eq: productsheaf}
\mathcal{B}_n = \mathcal{B}_1 \cdot \mathcal{B}_1^{\tilde{\tau}} \cdot \ldots \cdot \mathcal{B}_1^{\tilde{\tau}^{n-1}},$$ for all $n \in {\ensuremath{\mathbb{N}}}$. As all sheaves involved lie inside $M_2(k(E))^G$, we may construct the twisted ring $B(\mathcal{E},\mathcal{B}_1,\tilde{\tau})$ as a subring of the Ore extension $M_2(k(E))^G[z,z^{-1};\tilde{\tau}]$, as described in the statement of the theorem.
The degree 1 piece of $B(\mathcal{E},\mathcal{B}_1,\tilde{\tau})$ is $\text{H}^0(E^G,\mathcal{B}_1)z =
M_2(\text{H}^0(E,{\ensuremath{\mathcal{L}}}))^Gz$. Recall that in the proof of Lemma \[lem: twistedring\] we observed that this is precisely the image of $B_1^{G,\mu}$ under its embedding in the Ore extension. Remark \[rem: genindeg1Bgmu\] tells us that $B^{G,\mu}$ is generated in degree 1, therefore we obtain an embedding of $k$-algebras $B^{G,\mu} \hookrightarrow
B(\mathcal{E},\mathcal{B}_1,\tau)$. In particular, there is an injection of vector spaces $B_n^{G,\mu} \hookrightarrow
H_0(E^G,\mathcal{B}_n)z^n$ for all $n \in {\ensuremath{\mathbb{N}}}$.
Let us now consider an open set $U \subset E^G$, with $V=r^{-1}(U)$. Since $\tau$ is induced by $\sigma$, we have $r^{-1}(\tau^n(U)) = \sigma^n(V)$ for all $n \in {\ensuremath{\mathbb{N}}}$. Note that $\tilde{\tau}$ commutes with the action of $G$ on $M_2(k(E))^G$, which is a consequence of $\sigma$ commuting with the group action. Observe finally that as $\mathcal{E}$ is a sheaf of maximal orders and $\mathcal{B}_1$ is an invertible lattice contained in $M_2(k(E))^G$, the tensor products appearing in the definition of $\mathcal{B}_n$ are actual products inside $M_2(k(E))^G$. Thus $$\begin{aligned}
\mathcal{B}_n(U) &= \mathcal{B}_1(U)\cdot \mathcal{B}_1^{\tilde{\tau}}(U) \cdot \ldots \cdot
\mathcal{B}_1^{\tilde{\tau}^{n-1}}(U)\\
&= \left[M_2(r_{\ast}{\ensuremath{\mathcal{L}}})^G\right](U) \cdot \left[M_2((r_{\ast}{\ensuremath{\mathcal{L}}})^{\tau})^G\right](U) \cdot \ldots \cdot
\left[M_2((r_{\ast}{\ensuremath{\mathcal{L}}})^{\tau^{n-1}})^G\right](U) \\
&= M_2({\ensuremath{\mathcal{L}}}(V))^G \cdot M_2({\ensuremath{\mathcal{L}}}(\sigma(V)))^G \cdot \ldots \cdot M_2({\ensuremath{\mathcal{L}}}(\sigma^{n-1}(V)))^G\\
&\subseteq M_2({\ensuremath{\mathcal{L}}}(V){\ensuremath{\mathcal{L}}}(\sigma(V)) \ldots {\ensuremath{\mathcal{L}}}(\sigma^{n-1}(V)))^G\\
&= M_2({\ensuremath{\mathcal{L}}}{\ensuremath{\mathcal{L}}}^{\sigma} \ldots {\ensuremath{\mathcal{L}}}^{\sigma^{n-1}}(V))^G\\
&= \left[M_2(r_{\ast}{\ensuremath{\mathcal{L}}}_n)^G\right](U).\end{aligned}$$
We therefore have an inclusion $\mathcal{B}_n \hookrightarrow M_2(r_{\ast}{\ensuremath{\mathcal{L}}}_n)^G$, and by taking global sections one has $\text{H}^0(E^G,\mathcal{B}_n) \subset M_2(H^0(E,{\ensuremath{\mathcal{L}}}_n))^G$. But since the latter vector space is precisely $B^{G,\mu}_n$, this implies that the injection $B_n^{G,\mu} \hookrightarrow H_0(E^G,\mathcal{B}_n)z^n$ must be a bijection, which completes the proof.
\[rem: ample\] Although we did not show during the proof of Theorem \[thm: geomdescthcrtwist\] that the sequence $\{\mathcal{B}_n\}$ is ample, it follows from that result as we now indicate. Since $B^{G,\mu}$ is semiprime noetherian of GK dimension 2, the sheaves associated to its graded components form an ample sequence [@artin2000semiprime Proposition 6.4]; Theorem \[thm: geomdescthcrtwist\] shows that these sheaves coincide with the sequence $\{\mathcal{B}_n\}$.
Our first corollary is due to a result which generalises the categorical part of the Noncommutative Serre’s theorem (Theorem \[thm: noncomserrethm\]).
\[cor: astfequivcat\] There is an equivalence of categories $\text{qgr}(B^{G,\mu}) \simeq \text{coh}(\mathcal{E})$, where $\text{coh}(\mathcal{E})$ denotes the category of coherent sheaves over $\mathcal{E}$.
Combine Theorem \[thm: geomdescthcrtwist\] with [@artin2000semiprime Corollary 6.11].
We also obtain another corollary, which shows that the twist is in fact a prime ring.
\[cor: actuallyprime\] $B^{G,\mu}$ is prime.
By Theorem \[thm: geomdescthcrtwist\] we know that the degree 0 component of the graded quotient ring of $B^{G,\mu}$ is $M_2(k(E))^G$. If we can show that this ring is simple artinian then $B^{G,\mu}$ must be prime. By Lemma \[lem: Xouter\] the action of $G$ on $M_2(k(E))$ is outer. As $M_2(k(E))$ is simple artinian, we may apply [@montgomery1980fixd Theorem 2.7(1)] to conclude that $M_2(k(E))^G$ is also simple artinian, which completes the proof.
One can explicitly describe the structure of $M_2(k(E))^G$, as the following result shows. In the proof we use the notion of *PI degree* defined in [@mcconnell2001noncommutative §13.6.7].
\[cor: shapeofsimpleart\] There is an isomorphism of $k$-algebras $M_2(k(E))^G \cong M_2(k(E^G))$, where $k(E^G)=k(E)^G$.
By Corollary \[cor: actuallyprime\], $M_2(k(E))^G$ is simple artinian and therefore isomorphic to a ring of the form $M_n(D)$, where $D$ is some division ring. Considering the PI degree of $M_2(k(E))^G$ enables one to conclude that either $n=2$ and $D$ is a field or $M_2(k(E))^G$ is a division ring that is 4-dimensional over its centre. We can rule out the latter case since $M_2(k(E))^G$ is the degree 0 component of the graded quotient ring of $B^{G,\mu}$, which contains non-regular homogeneous elements by Remark \[rem: bgmunilp\]. Thus $D$ must be a field, and therefore the centre of $M_2(D)$ is isomorphic to $D$ itself. Since $k(E^G)$ is central in $M_2(k(E))^G$, it must embed in $D$ under the isomorphism $M_2(k(E))^G \cong M_2(D)$.
To complete the proof, observe that $M_2(k(E))$ is a 4-dimensional module over $M_2(k(E))^G$ (on either the left or the right), which itself is 4-dimensional over its centre $Z(M_2(k(E))^G) \cong D$. However, it is clear that $M_2(k(E))$ is 16-dimensional over $k(E^G)$, in which case one must have $D=k(E^G)$.
Our work in the remainder of this section concerns the irreducible objects in $\text{qgr}(B^{G,\mu})$. We begin with a preliminary lemma.
\[lem: irrtail1crit\] The irreducible objects in $\text{qgr}(B^{G,\mu})$ are precisely the tails of 1-critical $B^{G,\mu}$-modules.
Let us consider an irreducible object in $\text{qgr}(B^{G,\mu})$. Such an object will have the form $\pi(M)$ for some $M
\in \text{grmod}(B^{G,\mu})$. Theorem \[thm: thcrtwistprops\](ii) implies that $B^{G,\mu}$ has GK dimension 2, in which case [@krause2000growth Proposition 5.1(d)] tells us that $\text{GKdim }M=0,1$ or 2. Note that $M$ cannot have GK dimension 0 since $\pi(M)$ is nontrivial. Let us assume therefore that $\text{GKdim }M=2$. Since $\pi(M)$ is irreducible this means that $M$ must be 2-critical. However, the only 2-critical $B^{G,\mu}$-module is $B^{G,\mu}$ itself, and $\pi(B^{G,\mu})$ is reducible in $\text{qgr}(B^{G,\mu})$. We reach a contradiction, upon which the result is proved.
We can now give a more concrete description of the irreducible objects in $\text{qgr}(B^{G,\mu})$.
\[prop: onlyfatpoints\] Any irreducible object in $\text{qgr}(B^{G,\mu})$ is isomorphic to $\pi(M_p^2)$ for some $p \in E$, where $M_p^2$ is a direct sum of point modules over $B$. In particular, such modules are 1-critical and are therefore fat point modules of multiplicity 2.
By Lemma \[lem: irrtail1crit\], any irreducible object in $\text{qgr}(B^{G,\mu})$ has the form $\pi(N)$ for some 1-critical $B^{G,\mu}$-module $N$. Consider the extension $N \otimes_{B^{G,\mu}} M_{2}(B)$ as a right $M_2(B)$-module, which is ${\ensuremath{\mathbb{N}}}$-graded via the standard grading on a tensor product of graded modules. One may argue in the same manner as in the proof of Proposition \[prop: cohenmac\](i) to show that $N \otimes_{B^{G,\mu}} M_{2}(B)$ has GK dimension 1. This $M_{2}(B)$-module is noetherian, hence there is a 1-critical $M_{2}(B)$-module, $I$ say, that embeds in $\left(N \otimes_{B^{G,\mu}} M_{2}(B)\right)_{\geq n}$ for some sufficiently large $n \in {\ensuremath{\mathbb{N}}}$. One therefore has $\pi(I) \hookrightarrow \pi(N \otimes_{B^{G,\mu}} M_{2}(B))$ in $\text{qgr}(M_2(B))$. The 1-critical modules over $B$ are isomorphic in high degree to point modules by the equivalence of categories $\text{qgr}(B) \simeq
\text{coh}({\ensuremath{\mathcal{O}}}_E)$ from [@artin1990twisted Theorem 1.3]. Thus we may replace $\pi(I)$ with $\pi(M_p^2)$ for some $p \in E$.
Let us now restrict from $\text{qgr}(M_2(B))$ to $\text{qgr}(B^{G,\mu})$. By Proposition \[prop: fflat\] one has $$\label{eq: restinject}
\pi(M_p^2) \hookrightarrow \pi(N \otimes_{B^{G,\mu}} M_2(B)) \cong \pi\left(N \otimes_{B^{G,\mu}} \left(\bigoplus_{g \in
G} {^{\text{id}}}(B^{G,\mu})^{\phi_g} \right)\right) \cong \pi\left(\bigoplus_{g \in G} N^{\phi_g}\right).$$
The summands $N^{\phi_g}$ are 1-critical $B^{G,\mu}$-modules, and moreover they are the factors in a critical composition series for $(N \otimes_{B^{G,\mu}} M_2(B))_{B^{G,\mu}}$. By [@smith1992the Proposition 1.5], these factors are unique up to permutation and isomorphism in high degree. However, we claim that the right $M_2(B)$-module $M_p^2$ remains 1-critical upon restriction to a module over $B^{G,\mu}$, in which case $\pi(M_p^2) \cong
\pi(N^{\phi_{g}})$ for some $g \in G$. To prove this claim, observe that the module $M_p$ is also a point module over $A$, thus $M_p^2$ can be considered as an $M_2(A)$-module that has been restricted first to $A^{G,\mu}$ and then to the factor ring $B^{G,\mu}$. Thus Proposition \[prop: fatpoints\] is applicable and implies that $M_p^2$ is a 1-critical $B^{G,\mu}$-module. Since such a module has Hilbert series $2/(1-t)$, it must be a fat point module of multiplicity 2.
By untwisting one has $\pi((M_p^2)^{\phi_g^{-1}}) \cong \pi(N)$ in $\text{qgr}(B^{G,\mu})$. Note that the $G$-graded automorphism $\phi_g$ extends naturally from $B^{G,\mu}$ to $M_2(B)$ via the $G$-grading on $B$ (i.e. the grading induced by the action of $G$ on $B$ on which the action on $M_2(k)$ has no bearing). Thus we may consider $(M_p^2)^{\phi_g^{-1}}$ as a right $M_2(B)$-module, in which case there exists $L \in \text{grmod}(B)$ such that $(M_p^2)^{\phi_g^{-1}} \cong L^2$ in $\text{grmod}(B)$. By considering Hilbert series it is clear that $L$ must be a point module. On restriction to $B^{G,\mu}$, one obtains an isomorphism $\pi(M_q^2) \cong \pi(N)$ in $\text{qgr}(B^{G,\mu})$ for some $q \in E$. This completes the proof.
\[rem: restrictfatpoints\] As we saw in the proof of Proposition \[prop: onlyfatpoints\], one can consider the fat point modules of multiplicity 2 over $B^{G,\mu}$ as the restriction of the modules over $A^{G,\mu}$ constructed in Proposition \[claim: fatpoints\].
Proposition \[prop: onlyfatpoints\] enables us to describe the geometric origins of the fat point modules of multiplicity 2 that we have studied, as the next result demonstrates.
\[prop: fatpointsincohE\] For $p \in E$ consider the coherent sheaf of right $M_2({\ensuremath{\mathcal{O}}}_E)$-modules $k_p^2$, where $k_p:={\ensuremath{\mathcal{O}}}_E/\mathcal{I}_p$ is a skyscraper sheaf supported at $p$. Under a natural functor $\phi: \text{coh}(M_2(\mathcal{O}_E)) \rightarrow \text{coh}(\mathcal{E})$, the irreducible objects in $\text{coh}(\mathcal{E})$ are of the form $\phi(k_p^2)$. Furthermore, $\phi(k_p^2) \cong \phi(k_q^2)$ if and only if $q \in [p]$.
By [@artin2000semiprime Corollary 6.11] there is an equivalence of categories $\varphi_1:
\text{coh}(M_2(\mathcal{O}_E)) \rightarrow \text{qgr}(M_2(B))$. We label the equivalence from Corollary \[cor: astfequivcat\] by $\varphi_3: \text{qgr}(B^{G,\mu}) \rightarrow \text{coh}(\mathcal{E})$. Finally, let $\varphi_2$ denote the functor from $\text{qgr}(M_2(B))$ to $\text{qgr}(B^{G,\mu})$ obtained by the restriction of modules. Consider the diagram below: $$\label{eq: cohqgrdiag}
\centering
\begin{tikzpicture}
\draw [->] (1.25,0) -- (1.85,0);
\node[below] at (1.55,0) {\tiny{$\sim$}};
\node[above] at (1.55,0) {\footnotesize{$\varphi_1$}};
\draw [dashed,->] (0,-0.25) -- (0,-1.25);
\node[left] at (0,-0.75) {\footnotesize{$\phi$}};
\draw [->] (3,-0.25) -- (3,-1.25);
\node[right] at (3,-0.75) {\footnotesize{$\varphi_2$}};
\draw [<-] (0.65,-1.5) -- (2,-1.5);
\node[above] at (1.325,-1.5) {\tiny{$\sim$}};
\node[below] at (1.325,-1.5) {\footnotesize{$\varphi_3$}};
\node at (0,0) {$\text{coh}(M_2(\mathcal{O}_E))$};
\node at (0,-1.5) {$\text{coh}(\mathcal{E})$};
\node at (3,0) {$\text{qgr}(M_2(B))$};
\node at (3,-1.5) {$\text{qgr}(B^{G,\mu})$};
\end{tikzpicture}$$ We denote the clockwise composition of functors beginning at $\text{coh}(M_2(\mathcal{O}_E))$ by $\phi$.
Let us now consider an irreducible object in ${\ensuremath{\mathcal{F}}}\in \text{coh}(\mathcal{E})$. By Proposition \[prop: onlyfatpoints\] we know that $\varphi_3(\varphi_2(\pi(M_p^2)))={\ensuremath{\mathcal{F}}}$ for some $p \in E$. One also has $\varphi_1(k^2_p)=\pi(M_p^2)$, therefore $\phi(k_p^2)={\ensuremath{\mathcal{F}}}$. Since $\varphi_1$ and $\varphi_3$ are equivalences, the isomorphisms among the sheaves $\phi(k_p^2)$ are governed by the restriction functor $\varphi_2$. Corollary \[cor: qgrisos\] describes the only such isomorphisms that can occur, and this completes the proof.
As was mentioned in the Chapter \[chap: sklyanin\], if $|\sigma|= \infty$ then $A$ possesses four non-isomorphic fat point modules of each multiplicity greater than or equal to 2 by [@smith1993irreducible]. The categorical equivalence given by Theorem \[thm: noncomserrethm\] shows that these are not $B$-modules. Thus the non-existence of fat point modules over $B^{G,\mu}$ of multiplicity other than 2 by Proposition \[prop: onlyfatpoints\] does not preclude the existence of such modules over $A^{G,\mu}$.
To conclude this chapter we include one final lemma, whose significance will become apparent in §\[sec: vancliffql\].
\[lem: orbit4points\] For $p\in E$, $[p]$ contains precisely 4 points.
Suppose that $[p]$ contains fewer than 4 points. Then we must have $p^g=p^h$ for some distinct $g,h \in G$. Looking at any pair of the four points in , one can see that they have the same entry in precisely two coordinates, while the other two coordinates differ by a minus sign. If such a pair defined the same point then at least two of the entries would have to be zero. However, by Lemma \[lem: threegensannihilate\] any point on $E$ must have at least 3 non-zero coordinates.
The Koszul dual {#par: kdual}
---------------
In this short section we study the Koszul dual of $B^{G,\mu}$. We begin by noting that $B$ is Koszul by [@stafford1994regularity Theorem 3.9], which allows us to apply Proposition \[prop: koszuldualtwistcommute\] to $B$ and $B^{G,\mu}$. That proposition tells us that there is an isomorphism of $k$-algebras $$\label{eq: thcrdualtwist}
(B^{G,\mu})^! \cong (B^!)^{G,\mu},$$ since $\mu^2$ is the trivial 2-cocycle over $G$.
The relations of $B^!$ are given in [@stafford1994regularity Lemma 2.2]. Using we can easily write down the relations in the twisted dual: $$\begin{gathered}
\begin{align*}&[\overline{v_{0}},\overline{v_{1}}]+\frac{1}{\alpha}[\overline{v_{2}},\overline{v_{3}}],\;\;
[\overline{v_{0}},\overline{v_{1}}]_+ + [\overline{v_{2}},\overline{v_{3}}]_+,\;\;
[\overline{v_{0}},\overline{v_{2}}]+\frac{1}{\beta}[\overline{v_{3}},\overline{v_{1}}],\;\;
[\overline{v_{0}},\overline{v_{2}}]_+ + [\overline{v_{3}},\overline{v_{1}}]_+, \\
&[\overline{v_{0}},\overline{v_{3}}]-\frac{1}{\gamma}[\overline{v_{1}},\overline{v_{2}}],\;\;
[\overline{v_{0}},\overline{v_{3}}]_+ - [\overline{v_{1}},\overline{v_{2}}]_+, \;\; \left( \frac{1-\alpha}{1+\gamma}
-1\right)\overline{v_{0}}^2 +\left( \frac{1-\alpha}{1+\gamma} \right)\overline{v_{1}}^2+\overline{v_{3}}^2, \\
&\left( \frac{1+\alpha}{1-\beta} -1\right)\overline{v_{0}}^2 +\left( \frac{1+\alpha}{1-\beta}
\right)\overline{v_{1}}^2-\overline{v_{2}}^2.
\end{align*}\end{gathered}$$
Since $(B^!)^{G,\mu}$ has the same underlying $k$-vector space structure as $B^!$, $(B^{G,\mu})^!$ must have Hilbert series $\left(\frac{1+t}{1-t}\right)^2$. Moreover, this implies that the 8 degree 2 elements corresponding to those exhibited in [@stafford1994regularity Proposition 2.3] form a basis of $(B^{G,\mu})^!_2$. These elements are $$\begin{gathered}
\begin{aligned}\label{eq: kdualdeg2basis}
&t_1=\overline{v_{0}}^2,\;\; t_2=\overline{v_{1}}^2, \;\; t_3=[\overline{v_{0}}, \overline{v_{1}}], \;\;
t_4=[\overline{v_{0}}, \overline{v_{1}}]_+, \;\; t_5=[\overline{v_{0}}, \overline{v_{2}}], \\ &t_6=[\overline{v_{0}},
\overline{v_{2}}]_+, \;\; t_7=[\overline{v_{0}} ,\overline{v_{3}}], \;\; t_8=[\overline{v_{0}}, \overline{v_{3}}]_+.
\end{aligned}\end{gathered}$$
Note that in the twist none of the scalar coefficients have changed from their untwisted counterparts.
It is shown in [@stafford1994regularity Proposition 2.3(ii)] that each degree 2 element is normal or central in $B^!$, and a description is given of how they commute with the generators $\overline{x_{i}}$. We will show that the elements in are also either normal or central, but with different commutation rules.
Let us adopt Stafford’s notation. For a normal element $y\in (B^{G,\mu})^!$ and $i=0,1,2,3$ there exist $r_i \in
(B^{G,\mu})^!$ such that $\overline{v_{i}}y=yr_i$. This is encoded by $(\overline{v_{0}},\overline{v_{1}},\overline{v_{2}},\overline{v_{3}})^y=(r_0,r_1,r_2,r_3)$.
\[prop: normalsinkdual\] The elements $t_1,t_2,t_4,t_6$ and $t_8$ are central in $(B^{G,\mu})^!$, while $t_3, t_5$ and $t_7$ are normal. For $j=3,5,7$ we have $(\overline{v_{0}},\overline{v_{1}},\overline{v_{2}},\overline{v_{3}})^{t_{j}}=(-\overline{v_{0}},-\overline{v_{1}},
-\overline{v_{2}},-\overline{v_{3}})$.
The elements described in [@stafford1994regularity Proposition 2.3(ii)] are all normal and homogeneous with respect to the $G$-grading on $B^!$. Lemma \[prop: stillregular\] then implies that their twisted counterparts, exhibited in , will be normal in $(B^{G,\mu})^!$.
To see that the commutation rules are as described in the statement of the result, one simply computes using the information from $B^!$. For example, the commutation rule for $t_3$ in $B^!$ is $$\label{eq: t3commutation}
(\overline{x_{0}},\overline{x_{1}},\overline{x_{2}},\overline{x_{3}})^{t_3}=(-\overline{x_{0}},-\overline{x_{1}},
\overline{x_{2}},\overline{x_{3}}).$$
Under the $G$-grading, $t_3$ lies in the component $B^!_{g_1}$. By regarding $g_1$ as an element of $kG_{\mu}$ one can see that it satisfies the commutation rule $$\label{eq: normalkernel1}
(e,g_1,g_2,g_1g_2)^{g_{1}}= (e,g_1,-g_2,-g_1g_2).$$ in $kG_{\mu}$. It is clear by combining the commutation rules in and that $t_3$ behaves as indicated in the statement of the result. Similar calculations for the remaining elements completes the proof.
\[cor: twistkdualPI\] $(B^{G,\mu})^!$ is PI.
The Veronese ring $((B^{G,\mu})^!)^{(2)}$ is commutative (hence PI), which can be seen from the commutation rules in Proposition \[prop: normalsinkdual\]. Since $(B^{G,\mu})^!$ is a finite module over this Veronese subring, one can then apply [@mcconnell2001noncommutative Corollary 13.4.9(i)] to reach the desired conclusion.
As noted in [@stafford1994regularity §2.4] for $B$ and $B^!$, the algebras $B^{G,\mu}$ and $(B^{G,\mu})^!$ are very different. The latter has a large centre, while the former has a trivial centre by Proposition \[prop: centrethcrtwist\].
Other twists {#chap: othertwists}
============
The examples of cocycle twists that we have studied thus far have all been either twists of the 4-dimensional Sklyanin algebra or a factor ring of it. The goal of this section is to illustrate that there are other algebras that admit interesting cocycle twists.
The examples in §\[subsec: staffordalgs\] will concern the algebras defined by Stafford in [@stafford1994regularity]. Such algebras are related to the ring $B$, a twist of which was studied in Chapter \[chap: thcrtwist\]. The algebras that we will study in §\[subsec: rogzhangalgebras\] are those that were first introduced by Rogalski and Zhang in [@rogalski2012regular]. They are examples of AS-regular algebras of dimension 4 with three degree 1 generators. In §\[sec: vancliffql\] we will study cocycle twists of the algebras defined in [@vancliff1994quadratic]. Such algebras depend on a nonsingular quadric $Q$ and a line $L$ in [$\mathbb{P}_{k}^{3}$]{}, together with an automorphism $\sigma$ of $Q \cup L$. The algebras twisted in §\[subsec: gradedskewclifford\] generalise graded Clifford algebras, while our final examples in §\[subsec: homenvelopalg\] are twists of the universal enveloping algebra of $\mathfrak{sl}_2({\ensuremath{\mathbb{C}}})$ and its homogenisation.
Many of the examples in this thesis are twists of AS-regular algebras of dimension 4, for which no classification exists but many examples are known. Several families of such algebras have appeared in this thesis, for example 4-dimensional Sklyanin algebras and their cocycle twists, while there are further examples given by double Ore extensions [@zhang2008double], algebras with an additional ${\ensuremath{\mathbb{Z}}}^2$-grading [@lu2007regular] (like Rogalski and Zhang’s algebras in [@rogalski2012regular] but with two degree 1 generators) and the generalised Laurent polynomial rings studied in [@cassidy2006generalized].
One may wonder if there are any interesting twists of AS-regular algebras of smaller dimension. In dimensions 2 and 3 there exist classifications of such algebras (see [@artin1987graded] and [@artin1990some; @artin1991modules; @stephenson1996artin] respectively), which are preserved under cocycle twists by Corollary \[cor: asreg\]. For such algebras we have been unable to find any cocycle twists that are not Zhang twists of their [$\mathbb{N}$]{}-graded structure (in the manner of Proposition \[prop: recoverztwist\] or the example given in Remark \[rem: otheractions\]). Certainly for AS-regular algebras of dimension 2 there are no other cocycle twists – there are only two families of such algebras up to isomorphism, with neither possessing enough ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms to accommodate a twist that is not a Zhang twist of their ${\ensuremath{\mathbb{N}}}$-grading.
It would be interesting to know if in the dimension 3 case any of the families in the classification are related by cocycle twists. If that were the case then this behaviour would be similar to that which we will uncover in §\[subsec: rogzhangalgebras\] for Rogalski and Zhang’s algebras.
Twists of Stafford’s algebras {#subsec: staffordalgs}
-----------------------------
In this section we will study twists of the algebras introduced by Stafford in [@stafford1994regularity]. In that paper it is assumed that $\text{char}(k)\neq 2$, thus we make the same assumption for the duration of §\[subsec: staffordalgs\].
We will first explain how Stafford’s algebras were discovered, before showing that certain twists of them possess many good properties. We then briefly discuss how the fat point modules of multiplicity 2 over $A^{G,\mu}$ described in Proposition \[claim: fatpoints\] are fat point modules over the twists of Stafford’s algebras as well. Following this, we describe the point scheme of one family of twists and conjecture what the point scheme is for the remaining algebras.
Let us begin by stating the question which motivated the work in [@stafford1994regularity]. Recall that $B$ is an ${\ensuremath{\mathbb{N}}}$-graded factor ring of some 4-dimensional Sklyanin algebra $A$, thus there is a natural surjection $A
\twoheadrightarrow B$.
\[que: staffordasregexist\] Are there any other AS-regular algebras of dimension 4 which possess an ${\ensuremath{\mathbb{N}}}$-graded surjection onto $B$, or does this property characterise the 4-dimensional Sklyanin algebra $A$?
The defining relations of $B$ are given by the 8-dimensional space of quadratic elements $$\label{eq: 8defeqns}
R_B=\text{span}_k(f_1,f_2,f_3,f_4,f_5,f_6,\Omega_1,\Omega_2),$$ where the $f_i$ and $\Omega_i$ are given explicitly in and respectively. Any algebra answering Question \[que: staffordasregexist\] would, for Hilbert series reasons, have relations given by a 6-dimensional subspace of $R_B$. Stafford proved the existence of subspaces other than that spanned by the $f_i$ with the required property. Thus the surjection $A \twoheadrightarrow B$ is not a uniquely defining characteristic of the 4-dimensional Sklyanin algebra $A$ among AS-regular algebras of dimension 4.
As we did in §\[subsec: 4sklyanintwistandptscheme\], we will assume that the scalars $\alpha,\beta,\gamma \in k$ satisfy . Let $V=\text{span}_k(x_0,x_1,x_2,x_3)$. The algebras defined in [@stafford1994regularity] are presented by $$\label{eq: staffordinftyrelns}
S_{\infty}\index{notation}{s@$S_{\infty}$}:=T(V)/(R_{\infty})\text{, where }
R_{\infty}=\text{span}_k(f_1,f_2,f_3,f_4,\Omega_1,\Omega_2),$$ and $$\label{eq: staffordlincombrelns}
S_{d,i}\index{notation}{s@$S_{d,i}$}:=T(V)/(R_{d,i})\text{, where } R_{d,i}=\text{span}_k(d_1\Omega_1+d_2\Omega_2,f_j:\;
1 \leq j \leq 6,\; j \neq i),$$ for some $1 \leq i \leq 6$ and $d=(d_1,d_2)\in {\ensuremath{\mathbb{P}_{k}^{1}}}$.
There are also two further variants of $S_{\infty}$, obtained by replacing the relations $f_i$ and $f_{i+1}$ in $A$ with $\Omega_1$ and $\Omega_2$ for either $i=1$ or $i=3$; we denote these algebras by $S_{\infty,1,2}$ and $S_{\infty,3,4}$ respectively. In this notation one could therefore write $S_{\infty}=S_{\infty,5,6}$. Lemma \[lem: cyclicisostafford\] will allow us to dispense with this cumbersome notation.
With some restrictions on $d$, the algebras defined in and are shown by Stafford in [@stafford1994regularity Theorem 0.4] to be noetherian domains which are AS-regular of dimension 4. The method used to show this is to prove that their Koszul duals have certain properties which have good consequences for the algebras themselves.
We will twist the algebras in and in a manner that is consistent with the twists of $A$ and $B$ that we have already studied. We will let $S=T(V)/(R_S)$ denote one of Stafford’s algebras if we do not wish to specify which family it belongs to.
Consider the familiar action of $G=(C_2)^2 = \langle g_1, g_2 \rangle$ on $V$ from . Under the isomorphism $G \cong G^{\vee}$ defined by , this corresponds to the $G$-grading on generators $$\label{eq: sklyaninfullgrading}
x_0 \in S_{e},\;\; x_1 \in S_{g_{1}},\;\; x_2 \in S_{g_{2}},\;\; x_3 \in S_{g_{1}g_{2}}.$$
One can then use the 2-cocycle $\mu$ defined in to twist Stafford’s algebras. Since we already know the behaviour of the relations of $B$ under this twist, it is a simple matter to write down the new relations. Recall that the elements $f_i^{\mu}$ and $\Theta_i$ are described explicitly in and respectively.
\[lem: stafftwists\] Twisting the algebras defined in and using the data described above produces the following algebras. Firstly, $S_{\infty}^{G,\mu}:=T(V)/(R_{\infty}^{\mu})$, whose relations are given by the ideal generated by $$\label{eq: sinftymu}
R_{\infty}^{\mu} = \text{span}_k(f_1^{\mu},f_2^{\mu},f_3^{\mu},f_4^{\mu},\Theta_1,\Theta_2).$$ The other family of algebras is given by $S_{d,i}^{G,\mu}:=T(V)/(R_{d,i}^{\mu})$, whose relations are given by $$\label{eq: sdimu}
R_{d,i}^{\mu} = \text{span}_k(d_1\Theta_1+d_2\Theta_2,f_j^{\mu}:\; 1 \leq j \leq 6,\; j \neq i),$$ for some $1 \leq i \leq 6$ and $d=(d_1,d_2)\in {\ensuremath{\mathbb{P}_{k}^{1}}}$.
Since we are using the same twisting data as in Chapters \[chap: sklyanin\] and \[chap: thcrtwist\], we know the behaviour under twisting of each of the eight relations in . Explicitly, see equations and . An application of Lemma \[lem: defrelns\] is needed to see that the ideal of relations in each twist is generated by the elements we have indicated.
\[rem: otherstaffalgs\] As mentioned previously, there are two further variants of $S_{\infty}$ which can be twisted in the same manner to form the algebras $S_{\infty,1,2}^{G,\mu}$ and $S_{\infty,3,4}^{G,\mu}$.
Let us show immediately that, under some conditions on the parameters, such algebras have many good properties.
\[thm: stafftwistprops\] Assume that $\alpha,\beta$ and $\gamma$ satisfy . Let $S=S_{d,i}$ for some $d=(d_1,d_2)\in
{\ensuremath{\mathbb{P}_{k}^{1}}}$ and $1 \leq i \leq 2$ or let $S=S_{\infty}$. If $S=S_{d,1}$, assume that $d \neq (1,0),(1,-1-\beta \gamma)$ and if $S=S_{d,2}$ assume that $d \neq (1,\beta-1),(1,-1-\gamma)$. Then $S^{G,\mu}$ has the following properties:
- it is universally noetherian domain;
- it is an AS-regular algebra of dimension 4;
- it has Hilbert series $1/(1-t)^4$;
- it is Koszul;
- it is Auslander regular and satisfies the Cohen-Macaulay property.
Firstly, note that $S$ surjects onto $B$, with the kernel being generated by a regular sequence of normal elements. This is a consequence of Proposition 2.10, Lemma 2.14 and then Theorem 1.3(v) from [@stafford1994regularity] (see the remarks at the beginning of the proof of the latter result). By Theorem \[thm: thcrtwistprops\](i), $B$ is universally noetherian. Applying [@artin1999generic Proposition 4.9(1)] twice shows that $S$ is also universally noetherian, and using Corollary \[cor: uninoeth\] then gives a proof of part of (i).
Let us now move on to properties (ii)-(iv). The algebra $S$ has these properties by [@stafford1994regularity Theorem 0.4]. That (ii)-(iv) are true for $S^{G,\mu}$ then follows from Lemma \[lem: hilbseries\], Corollary \[cor: asreg\] and Proposition \[prop: koszul\]. One can then apply [@artin1991modules Theorem 3.9] to conclude that $S^{G,\mu}$ is also a domain, completing the proof of (i). To show that $S^{G,\mu}$ has the properties in (v) we can apply [@levasseur1992some Corollary 6.7] to $S$, followed by Proposition \[prop: cohenmac\].
Although Theorem \[thm: stafftwistprops\] is only stated for $S_{\infty}^{G,\mu}$ and $S_{d,i}^{G,\mu}$ for $1 \leq i
\leq 2$, it remains valid for $S_{\infty,1,2}^{G,\mu}$, $S_{\infty,3,4}^{G,\mu}$ and $S_{d,i}^{G,\mu}$ for $3 \leq i
\leq 6$. To see this, note that cyclically permuting the generators $x_1,x_2$ and $x_3$ of $S_{\infty,i,i+1}$ or $S_{d,i}$ defines an isomorphism with an algebra in the family $S_{\infty,i+2,i+3}$ or $S_{d,i+2}$ respectively (see the remarks after [@stafford1994regularity Theorem 0.4]). The indices here are considered mod 6. We can use the same ideas to find similar relations among the cocycle twists, as the following lemma shows.
\[lem: cyclicisostafford\] Cyclically permuting the generators $v_1,v_2$ and $v_3$ gives the following isomorphisms:
- $S_{\infty,5,6}^{G,\mu}(\alpha, \beta, \gamma) \cong S_{\infty,1,2}^{G,\mu}(\gamma, \alpha, \beta) \cong
S_{\infty,3,4}^{G,\mu}(\beta, \gamma, \alpha)$;
- $S_{d,1}^{G,\mu}(\alpha, \beta, \gamma) \cong S_{d',3}^{G,\mu}(\gamma, \alpha, \beta) \cong
S_{d'',5}^{G,\mu}(\beta, \gamma, \alpha)$ where $d'=(\left(\frac{1+\gamma}{1-\alpha}\right)d_1,d_2)$ and $d''=(\left(\frac{1-\beta}{1+\alpha}\right)d_1,d_2)$;
- $S_{d,2}^{G,\mu}(\alpha, \beta, \gamma) \cong S_{d',4}^{G,\mu}(\gamma, \alpha, \beta) \cong
S_{d'',6}^{G,\mu}(\beta, \gamma, \alpha)$ where $d'$ and $d''$ are as in (ii).
We give a proof for (i) and remark that the remaining cases are similar. Since we will need to distinguish between different parameters, we introduce the notation $f_{i,(\alpha,\beta,\gamma)}^{\mu}$ to indicate the parameters associated to a particular relation.
The ideal of relations in $S_{\infty,5,6}^{G,\mu}(\alpha, \beta, \gamma)$ is given in . Now permute the generators using the permutation $(123)$ on indices, then apply the automorphism $\varphi_{1}$ which sends $v_1
\mapsto -i v_1$ and $v_3 \mapsto i v_3$, where $i^2 =-1$. The following equations indicate the relations obtained: $$\begin{gathered}
\begin{aligned}\label{eq: permute1relns1}
\begin{array}{ccccccl}
f_{1,(\alpha,\beta,\gamma)}^{\mu} &\overset{(123)}{\rightsquigarrow} &[v_0,v_2]-\alpha[v_3,v_1]
&\overset{\varphi_{1}}{\rightsquigarrow} &[v_0,v_2]-\alpha[v_3,v_1] &= &f_{3,(\gamma,\alpha,\beta)}^{\mu},\\
f_{3,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_3]-\beta[v_1,v_2] &\rightsquigarrow
&i[v_0,v_3]+i\beta[v_1,v_2] &= & i f_{5,(\gamma,\alpha,\beta)}^{\mu},\\
f_{2,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_2]_{+} -[v_3,v_1]_{+} &\rightsquigarrow
&[v_0,v_2]_{+} -[v_3,v_1]_{+} &= & f_{4,(\gamma,\alpha,\beta)}^{\mu},\\
f_{4,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_3]_{+} -[v_1,v_2]_{+} &\rightsquigarrow
&i[v_0,v_3]_{+}+i[v_1,v_2]_{+} &= & i f_{6,(\gamma,\alpha,\beta)}^{\mu},\\
\Theta_{1,(\alpha,\beta,\gamma)} &\rightsquigarrow & -v_0^2+v_2^2+v_3^2-v_1^2 &\rightsquigarrow
&-v_0^2+v_2^2-v_3^2+v_1^2 &= &\Theta_{1,(\gamma,\alpha,\beta)}.
\end{array}
\end{aligned}\end{gathered}$$ Under the same composition of maps, the relation $\Theta_{2,(\alpha,\beta,\gamma)}$ is sent to $$\label{eq: permute1relns2}
v_2^2-\left(\frac{1+\alpha}{1-\beta} \right)v_3^2+\left(\frac{1-\alpha}{1+\gamma} \right)v_1^2 =
\left(\frac{1-\alpha}{1+\gamma} \right) \Theta_{2,(\gamma,\alpha,\beta)},$$ where we have used to obtain the equality. The relations obtained in and are those in $S_{\infty,1,2}^{G,\mu}(\gamma, \alpha, \beta)$, which proves the first part of (i).
We proceed in a similar manner to prove the second part of (i). We first apply the permutation $(132)$ to the generators, then apply the automorphism $\varphi_{2}$ which sends $v_2 \mapsto - i v_2$ and $v_3 \mapsto i v_3$, where $i^2 =-1$ once again. The relations in are transformed under this composition as indicated below: $$\begin{gathered}
\begin{aligned}\label{eq: permute2relns1}
\begin{array}{ccccccl}
f_{1,(\alpha,\beta,\gamma)}^{\mu} &\overset{(132)}{\rightsquigarrow} &[v_0,v_3]-\alpha[v_1,v_2]
&\overset{\varphi_{2}}{\rightsquigarrow} & i[v_0,v_3]+ i\alpha[v_1,v_2] &= &if_{5,(\beta,\gamma,\alpha)}^{\mu},\\
f_{3,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_1]-\beta[v_2,v_3] &\rightsquigarrow
&[v_0,v_1]-\beta[v_2,v_3] &= & f_{1,(\beta,\gamma,\alpha)}^{\mu},\\
f_{2,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_3]_{+} -[v_1,v_2]_{+} &\rightsquigarrow
&i[v_0,v_3]_{+} +i[v_1,v_2]_{+} &= & if_{6,(\beta,\gamma,\alpha)}^{\mu},\\
f_{4,(\alpha,\beta,\gamma)}^{\mu} &\rightsquigarrow &[v_0,v_1]_{+} -[v_2,v_3]_{+} &\rightsquigarrow
&[v_0,v_1]_{+}-[v_2,v_3]_{+} &= & f_{2,(\gamma,\alpha,\beta)}^{\mu},\\
\Theta_{1,(\alpha,\beta,\gamma)} &\rightsquigarrow & -v_0^2+v_3^2+v_1^2-v_2^2 &\rightsquigarrow
&-v_0^2-v_3^2+v_1^2+v_2^2 &= &\Theta_{1,(\beta,\gamma,\alpha)}.
\end{array}
\end{aligned}\end{gathered}$$ Under the same maps the relation $\Theta_{2,(\alpha,\beta,\gamma)}$ is sent to $$\label{eq: permute2relns2}
-v_3^2+\left(\frac{1+\alpha}{1-\beta} \right)v_1^2+\left(\frac{1-\alpha}{1+\gamma} \right)v_2^2 =
\left(\frac{1+\alpha}{1-\beta} \right) \Theta_{2,(\beta,\gamma,\alpha)},$$ where we have used to obtain the equality. The relations obtained in and are those in $S_{\infty,3,4}^{G,\mu}(\beta,\gamma, \alpha)$, which completes the proof.
By Lemma \[lem: cyclicisostafford\] one obtains the following corollary.
\[cor: staff3456props\] Under conditions on $d$ determined by the isomorphisms in Lemma \[lem: cyclicisostafford\], the properties listed in Theorem \[thm: stafftwistprops\] also hold for the algebras $S_{d,i}^{G,\mu}$ for $3 \leq i \leq 6$. The same is true for $S_{\infty,1,2}^{G,\mu}$ and $S_{\infty,3,4}^{G,\mu}$ also.
In light of the isomorphisms given in Lemma \[lem: cyclicisostafford\], we will assume now that $S$ is subject to the hypotheses of Theorem \[thm: stafftwistprops\].
We chose the $G$-grading on $S$ given in in order that it be compatible with the $G$-grading on $B$. A consequence of this is that there is a graded surjection $S^{G,\mu} \twoheadrightarrow B^{G,\mu}$. As for the Sklyanin algebra, we have three families of algebras which surject onto the same geometric ring; instead of this being a twisted homogeneous coordinate ring it is now $B^{G,\mu}$, which was shown in Theorem \[thm: geomdescthcrtwist\] to be a twisted ring. The situation is illustrated by Figure \[fig: sklyanintwists\].
at (-1,0) [$A$]{}; at (1,0) [$A^{G,\mu}$]{}; at (-1,2) [$S_{\infty}$]{}; at (1,2) [$S_{\infty}^{G,\mu}$]{}; at (-1,-2) [$S_{d,i}$]{}; at (1,-2) [$S_{d,i}^{G,\mu}$]{}; (-0.65,2) – (0.55,2); (-0.7,0) – (0.6,0); (-0.6,-2) – (0.55,-2); at (0,0.32) ; at (-4,0) [$B$]{}; at (4,0) [$B^{G,\mu}$]{}; (1.3,1.8) – (3.7,0.3); (1.5,0) – (3.2,0); (1.4,-1.7) – (3.7,-0.3); (-1.3,1.8) – (-3.7,0.3); (-1.2,0) – (-3.2,0); (-1.3,-1.8) – (-3.7,-0.3); (-3.4,0.3) .. controls (-1,1) and (1,1) .. (3.4,0.3);
\[rem: staffordpermisos\] In §\[subsec: permuteaction\] we studied cocycle twists of $A$ obtained by using the 24 possible actions of $G$ by the regular representation on the standard generators. All such actions also respect the relations in $S_{\infty}$ and $S_{d,i}$, thus one could study the associated cocycle twists. We remark that it was shown in the proof of Proposition \[lem: 24to1\] that $\text{Aut}_{\text{grp}}(G)$ stabilises $\mu$, hence in each case there are four isomorphism classes of cocycle twists. As in §\[subsec: permuteaction\], these are governed by the generator that belongs to the identity component of the induced $G$-grading.
We will now consider fat point modules over the algebras we have just constructed. The situation encapsulated by Figure \[fig: sklyanintwists\] implies that the point schemes of the algebras $A$, $S_{\infty}$ and $S_{d,i}$ all contain the elliptic curve $E$. Note that the full point schemes of the latter two algebras are given by [@stafford1994regularity Lemma 2.16]. In contrast, the surjections from $S_{\infty}^{G,\mu}$ and $S_{d,i}^{G,\mu}$ onto $B^{G,\mu}$ do not give us any information regarding point modules when $|\sigma|=\infty$. This is because the point scheme of $B^{G,\mu}$ was proved to be empty under that hypothesis in Proposition \[prop: bgmunoptmodules\].
However, the surjections onto $B^{G,\mu}$ do have the following consequence. The fat point modules of multiplicity 2 over $B^{G,\mu}$ described in Proposition \[prop: onlyfatpoints\] are fat point modules over the algebras $S_{\infty}^{G,\mu}$ and $S_{d,i}^{G,\mu}$. Recall that these modules can be obtained by restricting $M_2(S)$-modules of the form $M_p^2$ for $p \in E$ to $S^{G,\mu}$.
Let us now restrict our attention to point modules and describe the point scheme of algebras in the family $S_{\infty}^{G,\mu}$. As we did for $A^{G,\mu}$, we will study the graph of the point scheme under the associated automorphism. Unlike in that case, however, we cannot give an explicit description of all of the points, merely their existence. Computer calculations suggest that there are 20 distinct points of multiplicity 1, as was the case for $A^{G,\mu}$.
We will prove the following proposition through a series of lemmas.
\[prop: staffordptschemes\] Assume that $\text{char}(k) =0$. For generic parameters $\alpha, \beta$ and $\gamma$ the point scheme $\Gamma''$ of $S_{\infty}^{G,\mu}$ consists of 20 points up to multiplicity.
The hypotheses of Proposition \[prop: staffordptschemes\] are needed due to the use of computer calculations in the proof.
The first lemma allows us to focus on the multilinearisations of the defining relations of $S_{\infty}^{G,\mu}$. One should recall the definition of the points $e_j \in {\ensuremath{\mathbb{P}_{k}^{3}}}$ from .
\[lem: inftyptscheme1\] Let $\Gamma_2$ denote the vanishing locus of the multilinearisations of the quadratic relations of $S_{\infty}^{G,\mu}$. Then $\Gamma_2$ is the graph of $\Gamma''$ under the associated automorphism $\tau$. Moreover, $\tau$ has order 2.
By Theorem \[thm: stafftwistprops\], $S_{\infty}^{G,\mu}$ satisfies the hypotheses of Theorem \[thm: pointschemenice\]. Applying that theorem implies the first part of the lemma. Consider the multilinearisations of the quadratic relations of $S_{\infty}^{G,\mu}$: $$\begin{gathered}
\begin{aligned}\label{eqn: multilinsquasiprojsinfty1}
&m_1:= v_{01}v_{12}- v_{11}v_{02} -\alpha v_{21}v_{32}+\alpha v_{31}v_{22},\\
&m_2:=v_{01}v_{12}+v_{11}v_{02}-v_{21}v_{32}-v_{31}v_{22}, \\
&m_3:=v_{01}v_{22}-v_{21}v_{02}+\beta v_{11}v_{32}-\beta v_{31}v_{12},\\
&m_4:=v_{01}v_{22}+v_{21}v_{02} -v_{31}v_{12}-v_{11}v_{32}, \\
&m_5:=-v_{01}v_{02} + v_{11}v_{12}+v_{21}v_{22}- v_{31}v_{32}, \\
&m_6:=v_{11}v_{12}+\left(\frac{1+ \alpha}{1-\beta}
\right)v_{21}v_{22}-\left(\frac{1-\alpha}{1+\gamma}\right)v_{31}v_{32}.
\end{aligned}\end{gathered}$$ It is clear that the multilinearisations in are invariant under the map $v_{i1}
\leftrightarrow v_{i2}$, and moreover that $(e_0,e_3)$ is a solution to them. As in the proof of Lemma \[lem: phiaut\], one can conclude that $\tau$ has order 2.
\[lem: inftyptscheme2\] Consider the quasiprojective subscheme of $\Gamma_2$ in which $v_{01}\neq 0$ and $v_{02}=0$. This subscheme contains only one point, namely $(e_0,e_3)$.
In this quasiprojective subscheme we may assume that $v_{01}=1$. The multilinearisations in then become $$\begin{gathered}
\begin{align*}&m_1':=v_{12}-\alpha v_{21}v_{32}+\alpha v_{31}v_{22},\;\; m_2':=v_{12}-v_{21}v_{32}-v_{31}v_{22}, \\
&m_3':=v_{22}+\beta v_{11}v_{32}-\beta v_{31}v_{12},\;\; m_4':=v_{22}-v_{31}v_{12}-v_{11}v_{32}, \\
&m_5':=v_{11}v_{12}+v_{21}v_{22}- v_{31}v_{32}, \;\; m_6':=v_{11}v_{12}+\left(\frac{1+ \alpha}{1-\beta}
\right)v_{21}v_{22}-\left(\frac{1-\alpha}{1+\gamma}\right)v_{31}v_{32}.
\end{align*}\end{gathered}$$ By equating $m_1'$ and $m_2'$ one obtains the equation $$\label{eq: sinftyptscheme1}
(\alpha-1)v_{21}v_{32}=(\alpha +1)v_{31}v_{22},$$ while doing the same for $m_3'$ and $m_4'$ gives $$\label{eq: sinftyptscheme2}
(\beta-1)v_{31}v_{12}=(\beta +1)v_{11}v_{32}.$$ In light of we split the analysis into three cases; either $v_{21}=0$, $v_{32}=0$ or $v_{21},v_{32}\neq 0$.
*Case 1:* If $v_{21}=0$, then implies that either $v_{31}=0$ or $v_{22}=0$.
*Case 1(a):* If $v_{31}=0$ then implies that either $v_{11}=0$ or $v_{32}=0$. In either situation one can use $m_1',m_2',m_3'$ and $m_4'$ to see that $v_{12}=v_{22}=0$, which results in a contradication when $v_{32}=0$. If $v_{11}=0$ then the only solution that we obtain is $(e_0,e_3)$, which we have already seen lies in $\Gamma_2$.
*Case 1(b):* If $v_{22}=0$ then one must have $v_{12}=0$ by $m_1'$. Using this implies that either $v_{11}=0$ or $v_{32}=0$, as in the previous case. One can proceed in a similar manner to conclude that the only solution in this case is $(e_0,e_3)$ once again.
*Case 2:* Now assume that $v_{32}=0$. Then implies that either $v_{31}=0$ or $v_{12}=0$. If the former is true then one can use $m_1'$ and $m_3'$ to show that $v_{12}=v_{22}=0$, which is absurd. In the latter situation one can use $m_3'$ to conclude that $v_{22}=0$, then use $m_1'$ to show that $v_{12}=0$. This gives a contradiction once again.
*Case 3:* Finally, suppose that $v_{21},v_{32}\neq 0$. In this case one can see from that $v_{31},v_{22}\neq 0$ must hold. By one must therefore have either $v_{11}=v_{12}=0$ or $v_{11},v_{12} \neq 0$. If the former is true then $m_3'$ implies that $v_{22}=0$, which is a contradiction. Suppose therefore that $v_{11},v_{12} \neq 0$. The Macaulay2 code given by Code \[code: lineschemeainfty2\] in Appendix \[subsec: staffordcalcs1\] shows that there are no solutions to the multilinearisations in this case. Note that we can always scale the second copy of ${\ensuremath{\mathbb{P}_{k}^{3}}}$ so that $v_{11}v_{21}v_{31}v_{12}v_{22}v_{32}=1$.
\[lem: inftyptscheme3\] Consider the closed subscheme of $\Gamma_2$ in which $v_{01}=v_{02}=0$. This subscheme contains only 2 points, namely $(e_2,e_1)$ and $(e_1,e_2)$.
In this subscheme the multilinearisations from become $$\begin{aligned}
&m_1'':=-\alpha v_{21}v_{32}+\alpha v_{31}v_{22}, \quad\quad\> m_2'':=-v_{21}v_{32}-v_{31}v_{22}, \\
&m_3'':=\beta v_{11}v_{32}-\beta v_{31}v_{12}, \quad\quad\quad m_4'':=-v_{31}v_{12}-v_{11}v_{32}, \\
&m_5'':=v_{11}v_{12}+v_{21}v_{22}- v_{31}v_{32}, \;\; m_6'':=v_{11}v_{12}+\left(\frac{1+ \alpha}{1-\beta}
\right)v_{21}v_{22}-\left(\frac{1-\alpha}{1+\gamma}\right)v_{31}v_{32}. \end{aligned}$$ Using the equations $m_1'',\ldots,m_4''$ and our assumptions on $(\alpha,\beta,\gamma)$, we must have $$v_{21}v_{32}=v_{31}v_{22}=v_{31}v_{12}=v_{11} v_{32}=0.$$
If $v_{32}\neq 0$ then $v_{11}=v_{21}=0$ which is absurd. Similarly, if $v_{31}\neq 0$ we obtain a contradiction. Thus we can assume that $v_{31}=v_{32}=0$. Since $\frac{1+ \alpha}{1-\beta} \neq 1$ (this would imply $\gamma=0$ or $\beta=-1$), we must have $v_{11}v_{12}=v_{21}v_{22}=0$. The only admissible solutions to these equations are the points $(e_2,e_1)$ and $(e_1,e_2)$, the second of which we could have deduced by symmetry from the first, or vice versa.
We are now in a position to prove that the point scheme of $S_{\infty}^{G,\mu}$ consists of 20 points up to multiplicity.
[Proposition \[prop: staffordptschemes\]]{} By Lemma \[lem: inftyptscheme1\] we know that $\Gamma_2$ is the graph of $\Gamma''$ under the automorphism $\tau$. We will first study $\Gamma_2$, which is covered by the following four subschemes of ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$: $$U_1:\; v_{01},v_{02} \neq 0,\;\; U_2:\; v_{01} \neq 0, v_{02}=0,\;\; U_3:\; v_{01}=0,v_{02} \neq 0,\;\; U_4:\;
v_{01}=v_{02}=0.$$
Note that the Macaulay2 code given by Code \[code: lineschemeainfty1\] in Appendix \[subsec: staffordcalcs1\] shows that $U_1 \cap \Gamma_2$ contains 16 points up to multiplicity. By Lemma \[lem: inftyptscheme2\] we know that $U_2\cap
\Gamma_2$ contains only a single point of the form $(e_0,e_3)$. But by Lemma \[lem: inftyptscheme1\] the multilinearisations defining $\Gamma_2$ are symmetric. Thus Lemma \[lem: inftyptscheme2\] implies that $U_3\cap
\Gamma_2$ also contains only a single point, namely $(e_3,e_0)$. Finally, by Lemma \[lem: inftyptscheme3\] we know that $U_4\cap \Gamma_2$ contains only the two points $(e_2,e_1)$ and $(e_1,e_2)$.
We have shown that $\Gamma_2$ is a 0-dimensional scheme, therefore $\Gamma''= \pi_1(\Gamma_2)$ must also be 0-dimensional. To complete the proof we note that Proposition \[prop: genericpointscheme\] implies that the multiplicities of the points in $\Gamma''$ must add up to 20. In particular, this fact in conjunction with Code \[code: lineschemeainfty1\] implies that the four points we have exhibited explicitly must each have multiplicity 1.
Let $k$ be an algebraically closed field with $\text{char}(k) \neq 2$. Assume that $\alpha,\beta$ and $\gamma$ satisfy and $d \neq (1,\beta-1),(1,-1-\gamma)$. Then the point schemes of $S_{\infty}^{G,\mu}$ and $S_{d,i}(\alpha,\beta,\gamma)^{G,\mu}$ consist of 20 points counting multiplicity.
As evidence for this conjecture in relation to $S_{d,i}(\alpha,\beta,\gamma)^{G,\mu}$, we note that it can be verified by computer for some specific values of the parameters.
To end this section we will compute the dimension of the line scheme of $S_{\infty}^{G,\mu}$.
\[prop: staffordlinescheme\] Assume that $\text{char}(k)=0$. For generic parameters $\alpha,\beta$ and $\gamma$ the line scheme of $S_{\infty}^{G,\mu}$ is a 1-dimensional projective scheme.
As in Proposition \[prop: linescheme1dim\], our hypotheses reflect the fact that the proof relies on computer calculations. Let a general relation in $S_{\infty}^{G,\mu}$ be written in the form $$t_1f_1^{\mu}+t_2 f_2^{\mu}+t_3 f_3^{\mu}+t_4 f_4^{\mu}+ t_5 \Theta_1 + t_6 \Theta_2,$$ for some $t_i \in k$. Mimicking the method used to prove Proposition \[prop: linescheme1dim\], we form the following matrix: $$\begin{pmatrix}
-t_5 & t_1+t_2 & t_3+t_4 & 0 \\
t_2-t_1 & t_5+t_6 & 0 & \beta t_3-t_4 \\
t_4-t_3 & 0 & t_5+t_6 \left(\frac{1+\alpha}{1-\beta}\right) & -\alpha t_1-t_2 \\
0 & -\beta t_3-t_4 & \alpha t_1-t_2 & -t_5-t_6 \left(\frac{1-\alpha}{1+\gamma}\right)
\end{pmatrix}.$$
The line scheme of $S_{\infty}^{G,\mu}$ is the closed subscheme of points $(t_1,\ldots,t_6) \in {\ensuremath{\mathbb{P}_{k}^{5}}}$ such that this matrix has rank less than or equal to 2. The Macaulay2 code given by Code \[code: lineschemeainfty\] in Appendix \[subsec: staffordcalcs1\] shows that, when regarded as defining the affine cone of the line scheme in $\mathbb{A}_k^6$, the ideal generated by the $3 \times 3$ minors of this matrix has codimension 4. Equivalently, the affine scheme it defines has dimension 2. Since its affine cone has dimension 2, the line scheme must have dimension 1 when considered as a projective scheme in ${\ensuremath{\mathbb{P}_{k}^{5}}}$.
As we discussed in §\[subsec: irrinqgr\], AS-regular algebras of dimension 4 with a 0-dimensional point scheme and a 1-dimensional line are considered important examples of regular algebras. When $\text{char}(k)= 0$ and for generic parameters $\alpha, \beta$ and $\gamma$, Propositions \[prop: staffordptschemes\] and \[prop: staffordlinescheme\] imply that $S_{\infty}^{G,\mu}$ is such an algebra.
Twists in Rogalski and Zhang’s classification {#subsec: rogzhangalgebras}
---------------------------------------------
The algebras that we will now study were classified by Rogalski and Zhang in their paper [@rogalski2012regular]. We will need to work over an algebraically closed field of characteristic 0 for the duration of this section.
Rogalski and Zhang’s algebras are AS-regular domains of dimension 4 satisfying two extra conditions; they are generated by three degree 1 elements and admit a *proper ${\ensuremath{\mathbb{Z}}}^{2}$-grading*. Properness of such a grading, $A= \bigoplus_{n,m \in {\ensuremath{\mathbb{Z}}}}
A_{m,n}$ say, means that $A_{0,1}\neq 0$ and $A_{1,0}\neq 0$.
As the authors of [@rogalski2012regular] note in their introduction, AS-regular algebras of dimension 4 can have either two, three or four generators. Their aim was to develop examples in the least studied of these cases; the 4-dimensional Sklyanin algebras and Stafford’s algebras from [@stafford1994regularity] are of course examples of the four generator case (not to mention their cocycle twists), whilst examples with two generators were studied in [@lu2007regular].
Rogalski and Zhang’s main results are summarised below.
\[thm: rogzhangmain\] Let $A$ be an AS-regular domain of dimension 4 which is generated by three degree 1 elements and properly ${\ensuremath{\mathbb{Z}}}^{2}$-graded. Then either $A$ is a normal extension of an AS-regular algebra of dimension 3, or up to isomorphism it falls into one of eight 1 or 2 parameter families, $\mathcal{A}-\mathcal{H}$. Moreover, any such algebra is strongly noetherian, Auslander regular and Cohen-Macaulay.
The final three properties in Theorem \[thm: rogzhangmain\] hold under weaker conditions as [@rogalski2012regular Corollary 0.3] shows: one no longer needs to assume that the algebra is a domain or make an assumption on the number of generators.
There is another family of such algebras, namely $\underline{\mathcal{F}}$. However, by the isomorphism in [@rogalski2012regular Example 3.11] we can ignore such algebras and work with the family $\mathcal{F}$ instead. We will not concern ourselves with the algebras that are normal extensions.
In order to apply our cocycle twist construction we require graded algebra automorphisms, where in this case graded refers to the connected graded structure rather than the additional ${\ensuremath{\mathbb{Z}}}^{2}$-grading. Section 5 of Rogalski and Zhang’s paper is concerned with precisely this topic. The key result is the following, where generic means avoiding some finite set of parameters given in the statement of [@rogalski2012regular Lemma 5.1]:
\[thm: rogzhangauts\] Consider a generic AS-regular algebra $A$ in one of the families $\mathcal{A} - \mathcal{H}$. The graded automorphism group of $A$ is isomorphic either to $k^{\times}\times k^{\times}$ or to $k^{\times}\times
k^{\times} \times C_2$. The first case occurs for the families $\mathcal{A}(b,q)$ with $q \neq -1$, $\mathcal{D}(h,b)$ with $h
\neq b^4$, $\mathcal{F}$, $\underline{\mathcal{F}}$ and $\mathcal{H}$. The second case occurs if $A$ belongs to one of the families $\mathcal{A}(b,-1)$, $\mathcal{B}$, $\mathcal{C}$, $\mathcal{D}(h,b)$ with $h=b^4$, $\mathcal{E}$ or $\mathcal{G}$.
We now show that any cocycle twist of an algebra in one of these families must be isomorphic to another algebra in the classification. Before stating the result, let us fix some notation for the remainder of the section. The algebra $A$ will be generated by the three degree 1 elements $x_1,x_2$ and $x_3$, where $x_1, x_2 \in A_{1,0}$ and $x_3 \in
A_{0,1}$. We will follow Rogalski and Zhang in referring to the extra automorphism of order 2 as the *quasi-trivial* automorphism. This automorphism interchanges $x_1$ and $x_2$ whilst fixing $x_3$.
\[lem: preserveclassrogzhang\] Let $A$ be a generic algebra in one of the families $\mathcal{A} - \mathcal{H}$. Any cocycle twist $A^{G,\mu}$ must also belong to one of these families. Moreover, if $A$ possesses the quasi-trivial automorphism then so must $A^{G,\mu}$.
First note that all of the automorphisms described in Theorem \[thm: rogzhangauts\] preserve the ${\ensuremath{\mathbb{Z}}}^{2}$-grading on the algebra. Therefore by Lemma \[lem: autpresgrad\] any cocycle twist must also possess a proper ${\ensuremath{\mathbb{Z}}}^{2}$-grading. By Corollary \[cor: asreg\] and Lemma \[lem: hilbseries\] respectively, such a twist will be AS-regular of dimension 4 and have the same Hilbert series as $A$. Moreover, Lemma \[lemma: finitelygenerated\] and Remark \[rem: montfingenremark\] imply that $A^{G,\mu}$ must have three degree 1 generators. The proof of the first part of the lemma is completed by applying [@artin1991modules Theorem 3.9], which implies that such a twist is also a domain.
Suppose now that $A$ admits the quasi-trivial automorphism $\phi$. It suffices to show that this preserves any induced $G$-grading, since in that case for all homogeneous elements $x \in A_g$, $y \in A_h$ one has $$\phi(x \ast_{\mu} y)=\mu(g,h)\phi(xy)=\mu(g,h)\phi(x)\phi(y)=\phi(x) \ast_{\mu} \phi(y),$$ as $\phi(x) \in A_g$ and $\phi(y) \in A_h$. Observe that the following elements must be homogeneous with respect to any induced $G$-grading, since any automorphism acts on them diagonally: $$\label{eq: rogzhangcob}
w_1=x_1+x_2,\;\; w_2=x_1-x_2,\;\; w_3=x_3.$$
In particular, $\phi$ acts on them diagonally. These elements generate $A$ and therefore $\phi$ must preserve any induced $G$-grading.
The automorphisms corresponding to $k^{\times}\times k^{\times}$ come from scaling components of the ${\ensuremath{\mathbb{Z}}}^{2}$-grading. The additional presence of the quasi-trivial automorphism implies the existence of cocycle twists relating algebras in different families, as we now show. Recall that Lemma \[lem: defrelns\] concerns generators of ideals under cocycle twists; we use this lemma implicitly in the proof.
\[thm: rogzhangmymain\] Let $G=(C_2)^2 = \langle g_1,g_2 \rangle$ and let $\mu$ denote the 2-cocycle of $G$ defined in . Then there are $k$-algebra isomorphisms $$\mathcal{A}(1,-1)^{G,\mu}\cong \mathcal{D}(1,1),\;\; \mathcal{B}(1)^{G,\mu} \cong \mathcal{C}(1), \;\;
\mathcal{E}(1,\gamma)^{G,\mu}\cong \mathcal{E}(1,-\gamma), \;\; \mathcal{G}(1,\gamma)^{G,\mu} \cong
\mathcal{G}(1,\overline{\gamma}).$$
Once again, we will use the isomorphism $G \cong G^{\vee}$ given by . Let us begin by defining the action of $G$ which we will use for each of the cocycle twists we perform. Note that all of the algebras in the statement of the result admit the quasi-trivial automorphism. Therefore we can let $g_1$ act via the quasi-trivial automorphism and $g_2$ act by multiplying $x_3$ by -1 and fixing the other two generators.
Since the standard generators are not diagonal with respect to this action, we will instead use the generators $w_1,w_2,w_3$ described above in . Denoting the algebra we wish to twist by $A$, the induced $G$-grading on the new generators is given by $$w_1 \in A_e,\;\; w_2 \in A_{g_{2}},\;\; w_3 \in A_{g_{1}}.$$
The defining relations of any algebra in one of the eight families belong to different components of the ${\ensuremath{\mathbb{Z}}}^2$-grading. Observe that the algebras $\mathcal{A}(1,-1)$, $\mathcal{B}(1)$, $\mathcal{C}(1)$ and $\mathcal{D}(1,1)$ share three relations, only being distinguished from each other by their relations in the $(2,1)$-component. Writing the shared relations in terms of the diagonal basis we show that they are left invariant under the twist – the first two are quadratic relations: $$\begin{aligned}
0 &= w_1^2 - w_2^2 = \frac{w_1 \ast_{\mu} w_1}{\mu(e,e)} - \frac{w_2 \ast_{\mu} w_2}{\mu(g_2,g_2)} = w_1 \ast_{\mu} w_1
-
w_2 \ast_{\mu} w_2 = v_1^2 - v_2^2,\\
0 &= w_3 w_1 - w_1 w_3 = \frac{w_3 \ast_{\mu} w_1}{\mu(g_1,e)} - \frac{w_1 \ast_{\mu} w_3}{\mu(e,g_1)} = w_3 \ast_{\mu}
w_1 - w_1 \ast_{\mu} w_3 = v_3 v_1 - v_1 v_3,\end{aligned}$$ while the third relation is cubic: $$\begin{aligned}
0 = w_3^2w_2 - w_2 w_3^2 &= \frac{w_3 \ast_{\mu} w_3 \ast_{\mu} w_2}{\mu(g_1,g_1)\mu(e,g_2)} - \frac{w_2 \ast_{\mu} w_3
\ast_{\mu} w_3}{\mu(g_2,g_1)\mu(g_1g_2,g_1)} \\
&= w_3 \ast_{\mu} w_3 \ast_{\mu} w_2 - w_2 \ast_{\mu} w_3 \ast_{\mu} w_3 \\ &= v_3^2 v_2 - v_2 v_3^2.\end{aligned}$$
Thus, to verify the first two isomorphisms in the statement of the result it suffices to consider the behaviour under the twist of the only relation they do not share. We first twist this relation in the algebra $\mathcal{A}(1,-1)$, having once again written it in terms of the new generators beforehand: $$\begin{aligned}
0 &= [w_3,[w_1,w_2]_+] \\
&= w_3 w_1 w_2 + w_3 w_2 w_1 - w_1 w_2 w_3 - w_2 w_1 w_3 \\
&= \frac{w_3 \ast_{\mu} w_1 \ast_{\mu} w_2}{\mu(g_1,e)\mu(g_1,g_2)} + \frac{w_3 \ast_{\mu} w_2 \ast_{\mu}
w_1}{\mu(g_1,g_2)\mu(g_1g_2,e)} - \frac{w_1 \ast_{\mu} w_2 \ast_{\mu} w_3}{\mu(e,g_2)\mu(g_2,g_1)} - \frac{w_2
\ast_{\mu} w_1 \ast_{\mu} w_3}{\mu(g_2,e)\mu(g_2,g_1)} \\
&= -w_3 \ast_{\mu} w_1 \ast_{\mu} w_2 - w_3 \ast_{\mu} w_2 \ast_{\mu} w_1 - w_1 \ast_{\mu} w_2 \ast_{\mu} w_3 - w_2
\ast_{\mu} w_1 \ast_{\mu} w_3 \\
&= -[v_3,[v_1,v_2]_+]_+.\end{aligned}$$ This relation is the same as that in $\mathcal{D}(1,1)$ under the new generators, which proves the first isomorphism.
Let us now move on to $\mathcal{B}(1)$. Twisting the non-shared relation we see that $$\begin{aligned}
0 &= [w_3,[w_2,w_1]]_+ \\
&= w_3 w_2 w_1 - w_3 w_1 w_2 + w_2 w_1 w_3 - w_1 w_2 w_3 \\
&= \frac{w_3 \ast_{\mu} w_2 \ast_{\mu} w_1}{\mu(g_1,g_2)\mu(g_1g_2,e)} - \frac{w_3 \ast_{\mu} w_1 \ast_{\mu}
w_2}{\mu(g_1,e)\mu(g_1,g_2)} + \frac{w_2 \ast_{\mu} w_1 \ast_{\mu} w_3}{\mu(g_2,e)\mu(g_2,g_1)} - \frac{w_1 \ast_{\mu}
w_2 \ast_{\mu} w_3}{\mu(e,g_2)\mu(g_2,g_1)} \\
&= -w_3 \ast_{\mu} w_2 \ast_{\mu} w_1 + w_3 \ast_{\mu} w_1 \ast_{\mu} w_2 + w_2 \ast_{\mu} w_1 \ast_{\mu} w_3 - w_1
\ast_{\mu} w_2 \ast_{\mu} w_3 \\
&= [v_3,[v_1,v_2]].\end{aligned}$$ This relation is shared by $\mathcal{C}(1)$ under the new generating set, which proves the second isomorphism.
We now move on to the remaining two isomorphisms. The algebras in the relevant families share three relations, two of which we have already shown are preserved under the cocycle twist. This is also true for the third relation, which as yet we have not encountered: $$\begin{aligned}
0 = w_3^2w_2 +w_2w_3^2 &= \frac{w_3 \ast_{\mu} w_3 \ast_{\mu} w_2}{\mu(g_1,g_1)\mu(e,g_2)} + \frac{w_2 \ast_{\mu} w_3
\ast_{\mu} w_3}{\mu(g_2,g_1)\mu(g_1g_2,g_1)} \\&= w_3 \ast_{\mu} w_3 \ast_{\mu} w_2 + w_2 \ast_{\mu} w_3 \ast_{\mu} w_3
\\ &= v_3^2 v_2 +v_2 v_3^2.\end{aligned}$$
Once again, it suffices to see what happens to the non-shared relation. In $\mathcal{E}(1,\gamma)$, where $\gamma = \pm
i$, one has $$\begin{aligned}
0 &= w_3w_2w_1 - w_1w_3w_2 +\gamma w_1w_2w_3 - \gamma w_2w_1w_3 \\
&= \frac{w_3 \ast_{\mu} w_2 \ast_{\mu} w_1}{\mu(g_1,g_2)\mu(g_1g_2,e)} - \frac{w_1 \ast_{\mu} w_3 \ast_{\mu}
w_2}{\mu(e,g_1)\mu(g_1,g_2)} + \gamma \frac{w_1 \ast_{\mu} w_2 \ast_{\mu} w_3}{\mu(e,g_2)\mu(g_2,g_1)} - \gamma
\frac{w_2 \ast_{\mu} w_1 \ast_{\mu} w_3}{\mu(g_2,e)\mu(g_2,g_1)} \\
&= -w_3 \ast_{\mu} w_2 \ast_{\mu} w_1 + w_1 \ast_{\mu} w_3 \ast_{\mu} w_2 + \gamma w_1 \ast_{\mu} w_2 \ast_{\mu} w_3 -
\gamma w_2 \ast_{\mu} w_1 \ast_{\mu} w_3 \\
&= -v_3v_2v_1 + v_1v_3v_2 +\gamma v_1v_2v_3 - \gamma v_2v_1v_3.\end{aligned}$$ This is the final relation in $\mathcal{E}(1,-\gamma)$ under the new generators, which proves the penultimate isomorphism.
We now twist the final relation of $\mathcal{G}(1,\gamma)$, where $\gamma=\frac{1 + i}{2}$ and so $\overline{\gamma}=\frac{1}{2 \gamma}$: $$\begin{aligned}
0 &= w_3 w_1 w_2 +w_3 w_2 w_1 +i w_1w_2w_3 + i w_2w_1w_3 \\
&= \frac{w_3 \ast_{\mu} w_1 \ast_{\mu} w_2}{\mu(g_1,e)\mu(g_1,g_2)} + \frac{w_3 \ast_{\mu} w_2 \ast_{\mu}
w_1}{\mu(g_1,g_2)\mu(g_1g_2,e)} +i\frac{w_1 \ast_{\mu} w_2 \ast_{\mu} w_3}{\mu(e,g_2)\mu(g_2,g_1)} + i \frac{w_2
\ast_{\mu} w_1 \ast_{\mu} w_3}{\mu(g_2,e)\mu(g_2,g_1)} \\
&= -w_3 \ast_{\mu} w_1 \ast_{\mu} w_2 - w_3 \ast_{\mu} w_2 \ast_{\mu} w_1 + i w_1 \ast_{\mu} w_2 \ast_{\mu}
w_3 + i w_2 \ast_{\mu} w_1 \ast_{\mu} w_3 \\
&= -v_3 v_1 v_2 - v_3 v_2 v_1 + i v_1 v_2 v_3 + i v_2 v_1 v_3.\end{aligned}$$ This is precisely the final relation of $\mathcal{G}(1,\overline{\gamma})$ under the new generators, which proves the last isomorphism in the statement of the theorem.
Combined with the fact that $\mathcal{A}(b,-1)$, $\mathcal{B}(b)$, $\mathcal{C}(b)$, $\mathcal{D}(b,b^4)$, $\mathcal{E}(b,\gamma)$ and $\mathcal{G}(b,\gamma)$ are Zhang twists of the respective algebras in the statement of Theorem \[thm: rogzhangmymain\] for any parameter $b \in k^{\times}$ [@rogalski2012regular §3], this result gives a partial description of such algebras up to cocycle twisting.
Twisting an algebra of Vancliff {#sec: vancliffql}
-------------------------------
In this section we will investigate cocycle twists of algebras studied in [@vancliff1994quadratic], which are defined for algebraically closed fields of characteristic not equal to 2. Their properties are strongly controlled by some associated geometry; corresponding to each algebra there is nonsingular quadric $Q$ and a line $L$ in ${\ensuremath{\mathbb{P}_{k}^{3}}}$, along with an automorphism $\sigma \in \text{Aut}(Q \cup L)$. The algebras split into two families depending upon whether $\sigma$ preserves or interchanges the rulings on $Q$.
We will focus on the algebras for which the automorphism preserves the two rulings. While there exist graded automorphisms in the other case, those which are diagonal with respect to the relations given in [@vancliff1994quadratic Lemma 1.3(b)] produce cocycle twists which are Zhang twists of the [$\mathbb{N}$]{}-grading. This will not be true for our examples, since the point scheme of the twists is only 1-dimensional, as proved in Proposition \[prop: vancliffptscheme\].
The algebras we shall twist will be denoted by $R(\alpha,\beta,\lambda)$, where $\alpha, \beta, \lambda \in k^{\times}$ are scalars satisfying $\lambda \neq \alpha \beta$. As we did in Chapter \[chap: sklyanin\] for Sklyanin algebras, we will omit the parameters if no ambiguity will arise. The defining relations of $R(\alpha,\beta,\lambda)$ are given in [@vancliff1994quadratic Lemma 1.3(a)]: $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqlrelns}
&x_2 x_1 =\alpha x_1 x_2,\;\; x_3 x_1 = \lambda x_1 x_3,\;\; x_4 x_1 = \alpha\lambda x_1 x_4, \;\; x_4x_3=\alpha x_3
x_4,\\ &x_4 x_2 = \lambda x_2 x_4,\;\; x_3x_2 - \beta x_2 x_3=(\alpha \beta - \lambda)x_1 x_4.
\end{aligned}\end{gathered}$$
Example 1.5 from [@vancliff1994quadratic] shows that taking $\alpha = q, \beta = 1$ and $\lambda = q^{-1}$ for some $q \in k^{\times}$ such that $q^2 \neq 1$, one obtains the coordinate ring of quantum $2 \times 2$ matrices, denoted by ${\ensuremath{\mathcal{O}}}_q(M_2(k))$.
The geometric data associated to $R$ lies inside $\mathbb{P}(R_1)={\ensuremath{\mathbb{P}_{k}^{3}}}$, namely the quadric $Q=V(x_1x_4+x_2x_3)$ and the line $L=V(x_1,x_4)$. The automorphism $\sigma$ is described in the proof of Lemma 1.3(a) op. cit.: $$\label{eq: vancliffqlaut}
{\sigma}|_{Q}= \begin{pmatrix}
\alpha \lambda & 0 & 0 & 0 \\
0 & \lambda & 0 & 0 \\
0 & 0 & \alpha & 0 \\
0 & 0 & 0 & 1
\end{pmatrix},\;\;\; {\sigma}|_{L}=\begin{pmatrix}
\beta & 0 \\
0 & 1
\end{pmatrix}.$$
To find ${\ensuremath{\mathbb{N}}}$-graded algebra automorphisms for which the given generators form a diagonal basis, one only needs to consider the relation $x_3x_2 - \beta x_2 x_3=(\alpha \beta - \lambda)x_1 x_4$; the other relations are fixed by *any* automorphism that acts diagonally on the generators. Suppose that an automorphism acts by $x_i \mapsto
\lambda_i x_i$ for some $\lambda_i \in k^{\times}$. From the relevant relation we see that $\lambda_2 \lambda_3=
\lambda_1 \lambda_4$ must hold. It is clear that there are many automorphisms that would satisfy this condition, but we will focus on the following action of $G = (C_2)^2 = \langle g_1,g_2 \rangle$ on $R$: $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqlact}
&g_1:\; x_1 \mapsto x_1,\;\; x_2 \mapsto x_2,\;\; x_3 \mapsto -x_3,\;\; x_4 \mapsto -x_4, \\
&g_2:\; x_1 \mapsto x_1,\;\; x_2 \mapsto -x_2,\;\; x_3 \mapsto x_3,\;\; x_4 \mapsto -x_4.
\end{aligned}\end{gathered}$$
Since $|G|=4$, our assumption on the characteristic of $k$ means that $\text{char}(k) \nmid |G|$ always hold. Under the isomorphism $G \cong G^{\vee}$ given by , the action in induces the following $G$-grading on the generators of $R$: $$\label{eq: vancliffqlgrad}
x_1 \in R_{e},\;\; x_2 \in R_{g_{1}},\;\; x_3 \in R_{g_{2}},\;\; x_4 \in R_{g_{1}g_{2}}.$$
We now give the relations of the cocycle twist of $R$ that we will study.
\[lem: vancliffrelns\] Consider the $G$-grading on $R(\alpha,\beta,\gamma)$ induced by the action of $G$ given in , and let $\mu$ be the 2-cocycle defined in . Then the associated cocycle twist $R(\alpha,\beta,\gamma)^{G,\mu}$ has defining relations $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqltwistrelns}
&v_2 v_1 =\alpha v_1 v_2,\;\; v_3 v_1 = \lambda v_1 v_3,\;\; v_4 v_1 = \alpha\lambda v_1 v_4, \;\; v_4v_3=-\alpha v_3
v_4,\\
&v_4 v_2 = -\lambda v_2 v_4,\;\; v_3v_2 + \beta v_2 v_3=(\alpha \beta - \lambda)v_1 v_4.
\end{aligned}\end{gathered}$$
Note that since $x_1 \in R_{e}$, the three commutation relations from involving $x_1$ are left invariant under a twist by $\mu$. Computations for the remaining three relations when twisted by this 2-cocycle are deferred to Appendix \[subsec: calcvancliff\]. Lemma \[lem: defrelns\] shows that the ideal of relations in the twist is generated by the six relations in .
We now highlight some of the properties that the cocycle twist possesses.
\[thm: vancliffasregular\] Suppose that $\alpha, \beta, \lambda \in k^{\times}$ and $\lambda \neq \alpha \beta$. Then $R(\alpha,\beta,\lambda)^{G,\mu}$ has the following properties:
- it is an iterated Ore extension over $k$;
- it is generated in degree 1 with Hilbert series $1/(1-t)^4$;
- it is noetherian;
- it is Cohen-Macaulay and Auslander regular of global dimension 4;
- it is AS-regular.
By adjoining the generators in the order $v_1,v_2,v_4,v_3$ to $k$, it is clear that the twist is an iterated Ore extension (in which the endomorphisms are in fact automorphisms). This has two consequences; firstly, it implies that $R(\alpha,\beta,\lambda)^{G,\mu}$ has Hilbert series $1/(1-t)^4$ and is generated in degree 1, while an application of [@goodearl2004introduction Theorem 2.6] shows that it is noetherian.
By Corollary 1.9 and Proposition 2.3 from [@vancliff1994quadratic], $R(\alpha,\beta,\lambda)$ is Cohen-Macaulay and Auslander regular of global dimension 4. These properties are preserved under twisting by Propositions \[prop: gldim\] and \[prop: cohenmac\]. Part (v) can then be proved by applying [@levasseur1992some Theorem 6.3] to $R(\alpha,\beta,\lambda)^{G,\mu}$.
We will drop the parameters from our notation in future since there will be no ambiguity, thus $R^{G,\mu}$ is our object of study.
Our focus now turns to geometry, beginning with a description of the point scheme of $R^{G,\mu}$ at the level of its closed points. This will show that such algebras do not belong to either of the families defined in [@vancliff1994quadratic], nor are Zhang twists of the ${\ensuremath{\mathbb{N}}}$-grading of such an algebra.
\[lem: pointsparameterisevcliff\] Let $\Gamma_2$ be the projective scheme determined by the multilinearisations of the defining relations of $R^{G,\mu}$ given in . Then such multilinearisations can be expressed in terms of matrices as $M \cdot \underline{v} = 0$, where $$\label{eq: vancliffqlmultilin}
M = \begin{pmatrix}
v_{21} & - \alpha v_{11} & 0 & 0 \\
v_{31} & 0 & -\lambda v_{11} & 0 \\
v_{41} & 0 & 0 & -\alpha \lambda v_{11} \\
0 & 0 & v_{41} & \alpha v_{31} \\
0 & v_{41} & 0 & \lambda v_{21} \\
0 & v_{31} & \beta v_{21} & -(\alpha \beta - \lambda)v_{11}
\end{pmatrix}\;\text{ and }\; \underline{v}= \begin{pmatrix}
v_{12} \\ v_{22} \\ v_{32} \\ v_{42}
\end{pmatrix}.$$ Furthermore, $\Gamma_2$ is the graph of the point scheme $\Gamma$ of $R^{G,\mu}$ under the associated automorphism $\phi$. Points in $\Gamma$ are precisely those points in [$\mathbb{P}_{k}^{3}$]{} for which the matrix $M$ has rank 3 when evaluated at them.
It is a routine calculation to verify that the matrices in and the equation $M \cdot
\underline{v} = 0$ give an alternative description of the multilinearisations of the relations in . This formulation will prove useful to us when proving Proposition \[prop: vancliffptscheme\].
To prove the second part of the lemma it would suffice to apply Theorem \[thm: pointschemenice\]. This is possible by the conclusion of Theorem \[thm: vancliffasregular\], thus $\Gamma_2$ is the graph of $\Gamma$ under the associated automorphism $\phi$.
Let us now address the final part of the lemma by supposing that $p \in \Gamma$. As $\Gamma_2$ is the graph of an automorphism, $p^{\phi}$ is the unique point $q$ for which $(p,q) \in \Gamma_2$. In terms of the matrix equation $M
\cdot \underline{v} = 0$, this means precisely that the rank of $M$ when evaluated at $p$ must be 3.
We are now in a position to describe the closed points of $\Gamma$. Rather than first finding the points of $\Gamma_2$ as we have done on previous occasions (in Lemma \[lem: ptschemecontains\] for example), we will find the points of $\Gamma$ directly.
\[prop: vancliffptscheme\] The point scheme $\Gamma$ of $R^{G,\mu}$ consists of the union of the following lines in [$\mathbb{P}_{k}^{3}$]{}: $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqlptscheme}
&L_{12}:\; v_1=v_2=0,\;\; L_{13}:\; v_1=v_3=0,\;\; L=L_{14}:\; v_1=v_4=0, \\
&L_{24}:\; v_2=v_4=0, \;\; L_{34}:\; v_3=v_4=0.
\end{aligned}\end{gathered}$$ Moreover, the associated automorphism $\phi$ is defined on each of these lines by $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqlptaut}
&{\phi}|_{L_{12}}=\begin{pmatrix} -\alpha & 0 \\ 0 & 1 \end{pmatrix},\;\;
{\phi}|_{L_{13}}=\begin{pmatrix} -\lambda & 0 \\ 0 & 1 \end{pmatrix},\;\;
{\phi}|_{L_{14}}=\begin{pmatrix} -\beta & 0 \\ 0 & 1 \end{pmatrix}, \\
&{\phi}|_{L_{24}}=\begin{pmatrix} \lambda & 0 \\ 0 & 1 \end{pmatrix}, \;
{\phi}|_{L_{34}}=\begin{pmatrix} \alpha & 0 \\ 0 & 1 \end{pmatrix},
\end{aligned}\end{gathered}$$ where these lines have been identified with [$\mathbb{P}_{k}^{1}$]{} in the obvious manner.
Consider the matrix $M$ in , where we now suppress the second subscript of the variables therein. By Lemma \[lem: pointsparameterisevcliff\] we are looking for points $p \in {\ensuremath{\mathbb{P}_{k}^{3}}}$ at which the matrix $M$ has rank 3. In particular, the $4 \times 4$ minors of $M$ must vanish. On computing these minors, one obtains the following equations – note that there are only 14 equations since one of the minors vanishes at all points of [$\mathbb{P}_{k}^{3}$]{}. $$\begin{gathered}
\begin{aligned}\label{eq: vancliffqlminors}
&m_1 := 2 \alpha^2 \lambda v_1^2 v_3 v_4,\;\;\;\;\;\, m_2:=2 \alpha \lambda^2 v_1^2 v_2 v_4,\;\;\,m_3:=2 \alpha \lambda
v_1 v_2 v_4^2, \\
&m_4:=-2 \alpha \beta \lambda v_1 v_2^2 v_4,\;\; m_5:=2 \alpha \lambda v_1 v_3 v_4^2, \;\;\;\; m_6:=2 \alpha \lambda v_1
v_3^2 v_4, \\
&m_7:=\alpha \lambda v_1^2((\lambda-\alpha \beta)v_1v_4+ (\lambda+\alpha \beta)v_2v_3), \\
&m_8:=\alpha v_1v_3((\lambda-\alpha \beta)v_1v_4-(\lambda+\alpha \beta)v_2 v_3), \\
&m_9:=\lambda v_1 v_2((\lambda-\alpha \beta)v_1v_4 - (\lambda+\alpha \beta)v_2v_3), \\
&m_{10}:=\alpha v_1v_4((\lambda-\alpha \beta)v_1v_4+(\lambda- \alpha \beta)v_2 v_3), \\
&m_{11}:=v_2 v_4((\lambda+\alpha \beta)v_2v_3-(\lambda-\alpha \beta)v_1v_4), \\
&m_{12}:=\lambda v_1v_4((\lambda-\alpha \beta)v_2v_3-(\lambda-\alpha \beta)v_1v_4), \\
&m_{13}:=v_3v_4((\lambda+\alpha \beta)v_2v_3-(\lambda-\alpha \beta)v_1 v_4), \\
&m_{14}:=v_4^2((\lambda+\alpha \beta)v_2 v_3-(\lambda-\alpha \beta)v_1 v_4).
\end{aligned}\end{gathered}$$
We will now analyse . Equations $m_1$ through $m_6$ vanish if either $v_1=0$, $v_4=0$ or $v_2=v_3=0$. It is clear by looking at $m_7$ or $m_{14}$ that the latter case can be subsumed into either of the former two cases. If $v_1=0$ then $m_{14}$ implies that one of the other generators must also vanish, in which case all of the equations vanish. When $v_4=0$ one can use $m_7$ to reach the same conclusion.
The points obtained from this argument are those on the lines in . It remains to see that the matrix $M$ has rank exactly 3 when evaluated at any of such point; this is clear from . We can then determine the behaviour of $\phi$ on each line in $\Gamma$ by solving the matrix equation $M \cdot
\underline{v} = 0$ when $M$ has been evaluated at a general point on the line. This analysis reveals that $\phi$ behaves as described in ; in particular, it preserves each of the lines in .
In future we will write $L_{14}$ for the line $L$ to fit in with the notation used in for the other lines in the point scheme of $R^{G,\mu}$.
\[cor: vancliffnotztwist\] Suppose that $\alpha, \beta, \lambda \in k^{\times}$ and $\lambda \neq \alpha \beta$. The twist $R(\alpha,\beta,\lambda)^{G,\mu}$ is not isomorphic as an ${\ensuremath{\mathbb{N}}}$-graded algebra to an algebra from one of the two families studied in [@vancliff1994quadratic]. Furthermore, it is not a Zhang twist of the ${\ensuremath{\mathbb{N}}}$-grading of such an algebra either.
Both graded isomorphisms and Zhang twists preserve the point scheme of an algebra (the latter via Theorem \[thm: ztwistgmodequiv\]). However, the algebras in the two families in [@vancliff1994quadratic] have a point scheme containing a quadric by Proposition 2.1 op. cit., therefore in particular are 2-dimensional. By Proposition \[prop: vancliffptscheme\] the point scheme of $R(\alpha,\beta,\lambda)^{G,\mu}$ is 1-dimensional, from which the result follows.
It is shown in [@vancliff1994quadratic Corollary 3.4] that for some normal degree 2 element $\Omega$ there is an isomorphism of $k$-algebras $$B(Q,\mathcal{M},\varsigma) \cong R/(\Omega).$$ The invertible sheaf $\mathcal{M}=j^{\ast}{\ensuremath{\mathcal{O}}}_{{\ensuremath{\mathbb{P}_{k}^{3}}}}(1)$ is associated to the embedding $j:Q \hookrightarrow
{\ensuremath{\mathbb{P}_{k}^{3}}}$. We will write $S:= B(Q,\mathcal{M},\varsigma)$ in future. In Vancliff’s notation $\varsigma=\sigma|_{Q}$, however we wish to distinguish this automorphism from that associated to the point scheme of $R$.
One can say more about the normal element governing this factor ring: Lemma 1.11(b) from [@vancliff1994quadratic] implies that $\Omega$ is the unique (up to scalar multiple) element which annihilates those point modules lying on $Q$, while not annihilating all of those on $L_{14}$. The explicit form of $\Omega$ is not used in Vancliff’s paper, however the following calculation implies that $\Omega=\alpha x_1 x_4 + x_2 x_3$ up to scalar: evaluating this element at a point $p=(p_1,p_2,p_3,p_4) \in Q$ gives $$\label{eq: vancliffqlnormal}
(\alpha x_1 x_4 + x_2 x_3)(p,p^{\sigma}) = \alpha x_1(p) x_4(p^{\sigma}) + x_2(p) x_3(p^{\sigma})=\alpha^2 \lambda
(p_1p_4+p_2 p_3)=0,$$ upon using . The only points on $L_{14}$ which are annihilated by $\Omega$ are $e_2$ and $e_3$, which are the only points in the intersection $Q \cap L_{14}$.
Notice that $\Omega$ is homogeneous with respect to the $G$-grading defined in and therefore by Lemma \[lem: defrelns\] one has $$\label{eq: vancliffqltwistiso}
S^{G,\mu}= (R/(\Omega))^{G,\mu} \cong R^{G,\mu}/(\Theta),$$ where $\Theta:=\alpha v_1 v_4 - v_2 v_3$, the element corresponding to $\Omega$ under the twist.
We now prove a lemma concerning point modules over $S^{G,\mu}$.
\[lem: vanclifffacptmods\] The point scheme $\Gamma'$ of $S^{G,\mu}$ consists of the four lines $L_{12}, L_{13},L_{24}$ and $L_{34}$.
Recall from Proposition \[prop: vancliffptscheme\] that the point scheme $\Gamma$ of $R^{G,\mu}$ consists of the four lines in the statement of the lemma together with $L_{14}$. Since $S^{G,\mu}$ is a factor ring, proving the lemma amounts to showing that $\Theta$ vanishes at points on $L_{12}, L_{13},L_{24}$ and $L_{34}$, but not at points of $L_{14}$ (other than points of intersection with the other lines).
Evaluating $\Theta$ at a point $p=(p_1,p_2,p_3,p_4) \in \Gamma$ gives $$\Theta(p,p^{\phi}) = \alpha v_1(p) v_4(p^{\phi}) - v_2(p) v_3(p^{\phi}).$$ One can now use , which describes the automorphism $\phi$, to see that $\Theta$ vanishes on the four lines given in the statement of the lemma but not on $L_{14} \setminus \{e_2,e_3\}$.
While both $S^{G,\mu}$ and $B^{G,\mu}$ are cocycle twists of twisted homogeneous coordinate rings, the former has a 1-dimensional family of point modules by Lemma \[lem: vanclifffacptmods\], while the latter has none. We know that there are fat point modules of multiplicity 2 over $B^{G,\mu}$, therefore it is natural to ask if $S^{G,\mu}$ possesses such modules too, and moreover if they arise in the same way. We will show that this is indeed the case, and in so doing explain the existence of the point modules over $S^{G,\mu}$. To allow us to use results from §\[sec: modules\] we will work with $R$ and $R^{G,\mu}$ rather than the factor rings $S$ and $S^{G,\mu}$ respectively.
The following result is suggestive because the locus it describes is precisely the point scheme of $S^{G,\mu}$.
\[lem: ramlocus\] The locus of points on $Q$ that belong to $G$-orbits containing fewer than 4 points is $L_{12} \cup L_{13} \cup L_{24}
\cup L_{34}$.
Firstly, note that we are in the situation described by Hypotheses \[hyp: genhypforfatpts\]; $R$ is generated in degree 1, $G$ is the Klein-four group and $\mu$ is the 2-cocycle defined in . Moreover, the action of $G$ on $R_1$ affords the regular representation by , while $R$ satisfies the hypotheses of Theorem \[thm: pointschemenice\] by Corollary 1.9 and Proposition 2.3 from [@vancliff1994quadratic]. Consequently, the action of $G$ on the point scheme $Q \cup L_{14}$ of $R$ is as given in .
We can now observe that, as in the proof of Lemma \[lem: orbit4points\], a point $p \in Q$ with three non-zero coordinates lies in a $G$-orbit containing 4 points. We are led to look for points on $Q$ for which at least two coordinates are zero, which is precisely the union of lines given in the statement of the lemma. Note that orbits on this union of lines all contain 2 points apart from the singleton orbits $[e_0],[e_1],[e_2]$ and $[e_3]$, which are fixed by the action of $G$.
We will refer to orbits containing fewer than 4 points as *degenerate*, whereas those containing 4 points will be described as *nondegenerate*. From the defining equation of $Q$ one can see that the nondegenerate orbits contain precisely those points for which at least three of their coordinates are non-zero.
In the next result we once again use Notation \[not: tilde\] to avoid confusion; point modules over $R$ will be denoted by $M_p$, while those over $R^{G,\mu}$ will be written $\widetilde{M}_p$.
\[prop: modulesramify\] Let $M_p$ be a point module over $R$. The right $M_2(R)$-module $M_p^2$ becomes an $R^{G,\mu}$-module upon restriction for which:
- if $p$ belongs to a nondegenerate orbit then $M_p^2$ is a fat point module of multiplicity 2 over $R^{G,\mu}$, where $M_p^2 \cong M_q^2$ if and only if $q \in [p]$;
- if $p$ belongs to a degenerate orbit containing 2 points then $M_p^2 \cong \widetilde{M}_{p'} \oplus
\widetilde{M}_{q'}$, where $[p'] = \{p',q'\}$. Furthermore, if $p \notin L_{14}$ then $[p] = [p']$;
- if $p=e_j$ then $M_p^2 \cong \widetilde{M}_p^2$.
As we saw in Lemma \[lem: ramlocus\], the action of $G$ on $Q \cup L_{14}$ is given in . The automorphism $\sigma$ which is associated to the point scheme was described in . By comparing the two actions on points one can see that they commute. Thus the action of $\sigma$ on $Q \cup L_{14}$ preserves (non)degeneracy of orbits.
Suppose now that a point $p$ lies in a nondegenerate orbit of $Q \cup L_{14}$. By definition of such an orbit, three coordinates of $p$ are non-zero and so we may apply Proposition \[prop: fatpoints\] to construct a fat point module $M_p^2$ associated to it. If there were an isomorphism between two such modules, $M_p^2 \cong M_q^2$ say, then Corollary \[cor: fatpointisoclasses\] implies that $q=p^g$, which completes the proof of (i).
For the remaining cases we will have to get our hands dirtier, therefore we recall from Lemma \[lem: matrixgens\] that the generators of $R^{G,\mu}$ as a subring of $M_2(R)$ are given by the matrices $$\label{eq: matrixembedding23}
v_1 = \begin{pmatrix} x_1 & 0 \\ 0 & x_1 \end{pmatrix},\; v_2 = \begin{pmatrix} x_2 & 0 \\ 0 & -x_2 \end{pmatrix},\; v_3
= \begin{pmatrix} 0 & x_3 \\ x_3 & 0 \end{pmatrix}, \;\;
v_4 = \begin{pmatrix} 0 & -x_4 \\ x_4 & 0 \end{pmatrix}.$$ Recall also that for a point module $M_p$ the action of the generators of $R$ on $(M_p)_n$ is governed by coordinates of $\sigma^n(p)$.
We now begin to prove (ii). Let $p \in Q$ belong to a degenerate orbit of size 2, thus $[p]= \{p,q\}$. The point module $M_p$ is governed by a right ideal $I_p$ in $R$ of the form $(x_i,x_j,\lambda x_k-x_l)$ for some $\lambda \in
k^{\times}$ and distinct $i,j,k,l \in \{1,2,3,4\}$. We will deal with each line in the locus described in Lemma \[lem: ramlocus\] separately.
First suppose that $p=(0,0,\omega,1) \in L_{12}$ for some $\omega \in k^{\times}$. Note that $\sigma^n(p)=(0,0,\alpha^n
\omega,1)$ for all $n \in {\ensuremath{\mathbb{N}}}$. The generators $v_1$ and $v_2$ annihilate the module, and for scalars $\eta, \zeta \in k$ one has $$(\eta m_n, \zeta m_n) \cdot v_3 = (\alpha^n \omega \zeta m_{n+1},\alpha^n \omega \eta m_{n+1}),\;\;
(\eta m_n, \zeta m_n) \cdot v_4 = (\zeta m_{n+1}, -\eta m_{n+1}).$$
Suppose that both $v_3$ and $v_4$ send this element to the same 1-dimensional subspace. One is forced to have either $\eta=0$ or $\zeta=0$. The submodules generated by $(m_0,m_0)$ and $(m_0,-m_0)$ are therefore point modules over $R^{G,\mu}$, which we will denote by $\widetilde{M}_{p'}$ and $\widetilde{M}_{q'}$ for some $p',q' \in \Gamma$ respectively. They span $M_p^2$ and do not intersect each other. To discover which points $p'$ and $q'$ are, observe that $v_3+(-1)^{n}\alpha^n \omega v_4$ annihilates $(\widetilde{M}_{p'})_n$ and $v_3-(-1)^{n}\alpha^n \omega v_4$ annihilates $(\widetilde{M}_{q'})_n$ for all $n \in {\ensuremath{\mathbb{N}}}$. From this one can see that there is a decomposition of right $R^{G,\mu}$-modules $$M_{p}^2 \cong (m_0,0)R^{G,\mu} \oplus (0,m_0)R^{G,\mu} \cong \widetilde{M}_{p^{g_{2}}} \oplus \widetilde{M}_{p}.$$
Now suppose that $p=(\omega,1,0,0) \in L_{34}$ for some $\omega \in k^{\times}$, in which case one has $\sigma^n(p)=(\alpha^n \omega,1,0,0)$ for all $n \in {\ensuremath{\mathbb{N}}}$. The generators $v_3$ and $v_4$ annihilate the module, and for scalars $\eta, \zeta \in k$ one has $$(\eta m_n, \zeta m_n) \cdot v_1 = (\alpha^n \omega \eta m_{n+1},\alpha^n \omega \zeta m_{n+1}),\;\;
(\eta m_n, \zeta m_n) \cdot v_2 = (\eta m_{n+1}, -\zeta m_{n+1}).$$
Once again, the disjoint submodules generated by $(m_0,0)$ and $(0,m_0)$ are point modules over $R^{G,\mu}$. We will denote these modules by $\widetilde{M}_{p'}$ and $\widetilde{M}_{q'}$ respectively, as we did for points on $L_{12}$. Observe that $v_1-\alpha^n \omega v_2$ annihilates $(\widetilde{M}_{p'})_n$ and $v_1+\alpha^n \omega v_2$ annihilates $(\widetilde{M}_{q'})_n$ for all $n \in {\ensuremath{\mathbb{N}}}$. Thus there is a decomposition of right $R^{G,\mu}$-modules $$M_{p}^2 \cong (m_0,0)R^{G,\mu} \oplus (0,m_0)R^{G,\mu} \cong \widetilde{M}_{p} \oplus \widetilde{M}_{p^{g_{2}}}.$$
In the remaining two cases there is a slightly different direct sum decomposition. Suppose that $p=(0,\omega,0,1) \in
L_{13}$ for some $\omega \in k^{\times}$. One has $\sigma^n(p)=(0,\lambda^n \omega,0,1)$ for all $n \in {\ensuremath{\mathbb{N}}}$. The generators $v_1$ and $v_3$ annihilate the module, and for scalars $\eta, \zeta \in k$ one has $$(\eta m_n, \zeta m_n) \cdot v_2 = (\lambda^n \omega \eta m_{n+1}, - \lambda^n \omega \zeta m_{n+1}),\;\;
(\eta m_n, \zeta m_n) \cdot v_4 = (\zeta m_{n+1}, -\eta m_{n+1}).$$
In this case one must have $\eta^2=\zeta^2$ to ensure that both generators map $(\eta m_n, \zeta m_n)$ into the same 1-dimensional vector space. One can see that the submodules generated by $(m_0,m_0)$ and $(m_0,-m_0)$ are disjoint point modules over $R^{G,\mu}$, which we will denote by $\widetilde{M}_{p'}$ and $\widetilde{M}_{q'}$ respectively, as above. Observe that $v_2-(-1)^n \lambda^n \omega v_4$ annihilates $(\widetilde{M}_{p'})_n$ and $v_2+(-1)^n\lambda^n \omega v_4$ annihilates $(\widetilde{M}_{q'})_n$ for all $n \in {\ensuremath{\mathbb{N}}}$. Thus there is a decomposition of right $R^{G,\mu}$-modules $$M_{p}^2 \cong (m_0,m_0)R^{G,\mu} \oplus (m_0,-m_0)R^{G,\mu} \cong \widetilde{M}_{p} \oplus \widetilde{M}_{p^{g_{1}}}.$$
Finally, suppose that $p=(\omega,0,1,0) \in L_{24}$ for some $\omega \in k^{\times}$. One has $\sigma^n(p)=(\lambda^n
\omega,0,1,0)$ for all $n \in {\ensuremath{\mathbb{N}}}$. The generators $v_2$ and $v_4$ annihilate the module, and for scalars $\eta, \zeta
\in k$ one has $$(\eta m_n, \zeta m_n) \cdot v_1 = (\lambda^n \omega \eta m_{n+1}, \lambda^n \omega \zeta m_{n+1}),\;\;
(\eta m_n, \zeta m_n) \cdot v_3 = (\zeta m_{n+1}, \eta m_{n+1}).$$
As for the line $L_{13}$, the submodules generated by $(m_0,m_0)$ and $(m_0,-m_0)$ are disjoint point modules over $R^{G,\mu}$, which we denote by $\widetilde{M}_{p'}$ and $\widetilde{M}_{q'}$ once again. Observe that $v_1- \lambda^n
\omega v_3$ annihilates $(\widetilde{M}_{p'})_n$ and $v_1+\lambda^n \omega v_3$ annihilates $(\widetilde{M}_{q'})_n$ for all $n \in {\ensuremath{\mathbb{N}}}$. Thus there is a decomposition of right $R^{G,\mu}$-modules $$M_{p}^2 \cong (m_0,m_0)R^{G,\mu} \oplus (m_0,-m_0)R^{G,\mu} \cong \widetilde{M}_{p} \oplus \widetilde{M}_{p^{g_{1}}}.$$
To complete the proof of part (ii) we must consider the points belonging to degenerate orbits of order 2 for which $p
\in L_{14}$. Suppose that $p = (0,\omega,1,0)$ for some $\omega \in k^{\times}$ (thus $p \in L_{14} \setminus Q$) and consider the restriction of the right $M_2(R)$-module $M_p^2$ to $R^{G,\mu}$. One has $\sigma^n(p)=(0,\beta^n\omega,1,0)$ for all $n \in {\ensuremath{\mathbb{N}}}$. The generators $v_1$ and $v_4$ annihilate the module, and for scalars $\eta, \zeta \in k$ one has $$(\eta m_n, \zeta m_n) \cdot v_2 = (\beta^n \omega \eta m_{n+1}, -\beta^n \omega \zeta m_{n+1}),\;\;
(\eta m_n, \zeta m_n) \cdot v_3 = (\zeta m_{n+1}, \eta m_{n+1}).$$
The submodules generated by $(m_0, i m_0)$ and $(m_0,-i m_0)$ can be seen to be disjoint point modules over $R^{G,\mu}$. Observe that $v_2+ (-\beta)^n i \omega v_3$ and $v_2 - (-\beta)^n i \omega v_3$ annihilate the degree $n$ pieces of these modules respectively. It follows that there is a decomposition of right $R^{G,\mu}$-modules $$M_{p}^2 \cong (m_0, i m_0)R^{G,\mu} \oplus (m_0,-i m_0)R^{G,\mu} \cong \widetilde{M}_{(0,- i \omega,1,0)} \oplus
\widetilde{M}_{(0,i \omega,1,0)}.$$
We must now address (iii) and the four points $p=e_j$ for $j=0,1,2,3$. Since in that case three of the generators of $R$ annihilate $M_p$, the $R^{G,\mu}$-submodules of $M_p^2$ generated by $(m_0,0)$ and $(0,m_0)$ are disjoint, span $M_p^2$ and are both isomorphic to $\widetilde{M}_p$.
Let us now use Proposition \[prop: modulesramify\] to consider the restriction of modules from $R^{G,\mu}$ to $S^{G,\mu}$. Recall that the point scheme of $S$ is the quadric $Q$, thus the modules over $R^{G,\mu}$ considered in the proposition that correspond to points on $Q$ can be restricted to $S^{G,\mu}$. In particular, note that Proposition \[prop: modulesramify\](i) implies that $S^{G,\mu}$ has a family of fat point modules of multiplicity 2 parameterised by an open subset of $Q^G$.
One can compare the behaviour described in Proposition \[prop: modulesramify\] with the work in [@lebruyn1995central], where Le Bruyn studies fat point modules over algebras that are finite over their centre. As stated in the introduction to that paper, the study of fat point modules of a certain multiplicity is equivalent to studying the ramification locus of a push-forward sheaf. In our situation, if one regards points in degenerate orbits as being ramified — which coincides with the meaning of the term for curves, see [@ueno2003algebraic Lemma 8.15] — then the behaviour of fat point modules is once again related to ramification.
The behaviour we have encountered in §\[subsubsec: geomdescrthcrtwist\] and §\[sec: vancliffql\] with regard to fat point modules over certain cocycle twists suggests the following questions.
\[ques: thcrtwist\] Let $T=B(X,{\ensuremath{\mathcal{L}}},\sigma)$ be a twisted homogeneous coordinate ring. Suppose that a finite abelian group $G$ acts by graded automorphisms on $T$, and let $T^{G,\mu}$ be a cocycle twist of the induced $G$-grading.
- Can $T^{G,\mu}$ be described geometrically?
- Can one construct fat point modules over $T^{G,\mu}$ via restriction of modules from $TG_{\mu}$, even when the twisted group algebra is not a matrix ring as in our examples?
- If there is a positive answer to (ii), is the decomposition of such a $T^{G,\mu}$-module into 1-critical modules — in the sense of a critical composition series — determined geometrically?
Twisting a graded skew Clifford algebra {#subsec: gradedskewclifford}
---------------------------------------
In this section we study graded skew Clifford algebras, which were introduced by Cassidy and Vancliff in [@cassidy2010generlizations]. Such algebras are — as the name suggests — generalisations of graded Clifford algebras. In [@nafari2011classifying] it was shown that almost all AS-regular algebras of dimension 3 generated by three degree 1 generators are either graded skew Clifford algebras or Ore extensions of such algebras (which are themselves AS-regular of dimension 2).
There is a recent corrigendum to [@cassidy2010generlizations], namely [@cassidy2013corrigendum], which contains modifications of several definitions and results. Where appropriate we will reference the corrigendum; for a description of what remains valid in the original paper, see the introduction to [@cassidy2013corrigendum].
We will assume throughout this section that $\text{char}(k)\neq 2$. Let $\mu \in M_n(k)$ be a *multiplicatively skew-symmetric matrix* with $\mu_{ii}=1$. This means that $\mu_{ij}=\mu_{ji}^{-1}$ for all $i,j=1,\ldots,n$. Such a matrix defines a skew polynomial ring $S$ on the generators $z_1,\ldots,z_n$ in a natural manner, with defining relations $z_jz_i=\mu_{ij}z_iz_j$. A matrix $M \in M_n(k)$ is *$\mu$-symmetric* if $M_{ij}=\mu_{ij}M_{ji}$ for all $i,j=1,\ldots,n$.
\[defn: gradedskewclifford\] Let $\mu$ be as above and $M_1,\ldots,M_n$ be $\mu$-symmetric matrices. The *graded skew Clifford algebra* $A=A(\mu,M_1,\ldots,M_n)$ associated to this data is a graded $k$-algebra with degree 1 generators $x_1,\ldots,x_n$ and degree 2 generators $y_1,\ldots,y_n$. The following conditions are satisfied:
- the relations are of the form $x_ix_j+\mu_{ij}x_jx_i=\sum_{k=1}^n (M_k)_{ij} y_k$ for all $i,j=1,\ldots,n$;
- there exists a *normalising sequence* $\{r_1,\ldots,r_n\}$ that spans $ky_1+\ldots+ky_n$ (see [@cassidy2010generlizations Definition 1.9(a)]).
Although it may appear that this algebra is not generated in degree 1, Lemma 1.13 op. cit. gives equivalent conditions such that $y_i \in (A_1)^2$ for all $i=1,\ldots,n$. One such condition is that the matrices $M_1,\ldots,M_n$ are linearly independent.
The main result regarding such algebras is Theorem 4.2 op. cit., which is correct as stated under the changes in the relevant definitions given by [@cassidy2013corrigendum Definition 2]. It relates to a condition on the quadric system determined by $q_k=zM_k z^T$ for $k=1,\ldots,n$, where $z=(z_1,\ldots,z_n)$. Suppose that this system is normalising, thus $Sq_k=q_kS$ for all $k=1,\ldots,n$, and satisfies one of the equivalent conditions of [@cassidy2013corrigendum Corollary 11]. Then the main theorem states that $A(\mu,M_1,\ldots,M_n)$ is an AS-regular domain of global dimension $n$.
Let us consider a concrete example which is AS-regular of dimension 4. The example we will study has already appeared in Theorem \[thm: new\] and is the algebra obtained by factoring the free algebra $k\{x_1,x_2,x_3,x_4\}$ by the ideal generated by the relations $$\begin{gathered}
\begin{aligned}\label{eq: cliffordeg}
&x_4x_1-ix_1x_4,\;\; x_3^2-x_1^2,\;\; x_3x_1-x_1x_3+x_2^2,\;\; x_3x_2-ix_2x_3,\;\; x_4^2-x_2^2,\\
&x_4x_2-x_2x_4+\gamma x_1^2.
\end{aligned} \end{gathered}$$ This algebra — which we denote by $A(\mu,M_1,\ldots,M_4)$ — is discussed in [@cassidy2010generlizations Example 5.1]. The corresponding multiplicatively skew-symmetric matrix is $$\label{eq: cliffordskew}
\mu=\begin{pmatrix} 1 & i & -1 & i \\ -i & 1 & i & -1 \\ -1 & -i & 1 & i \\ -i & -1 & -i & 1 \end{pmatrix},$$ with associated normalising quadric system $$\label{eq: cliffordnormal}
q_1=z_1z_2,\;\; q_2=z_3z_4,\;\; q_3=z_1^2+z_3^2+\gamma z_2 z_4,\;\; q_4=z_2^2+z_4^2+z_1 z_3.$$
One can calculate using $q_k=zM_k z^T$ that the corresponding $\mu$-symmetric matrices are $$\begin{gathered}
\begin{aligned}\label{eq: cliffordmatrices}
&M_1=\begin{pmatrix} 0 & \frac{1}{2} & 0 & 0 \\ -\frac{i}{2} & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0
\end{pmatrix},\;\;
M_2=\begin{pmatrix} 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & \frac{1}{2} \\ 0 & 0 & -\frac{i}{2} & 0
\end{pmatrix} \;\;, \\
&M_3=\begin{pmatrix} 1 & 0 & 0 & 0 \\ 0 & 0 & 0 & \frac{\gamma}{2} \\ 0 & 0 & 1 & 0 \\ 0 & -\frac{\gamma}{2} & 0 & 0
\end{pmatrix},\;\;
M_4=\begin{pmatrix} 0 & 0 & \frac{1}{2} & 0 \\ 0 & 1 & 0 & 0 \\ -\frac{1}{2} & 0 & 0 & 0 \\ 0 & 0 & 0 & 1
\end{pmatrix}.
\end{aligned} \end{gathered}$$
These matrices are easily seen to be linearly independent, which implies that the degree 2 generators can be written in terms of the degree 1 generators.
One equivalent condition in [@cassidy2013corrigendum Corollary 11] is that factoring out the elements in from $S$ produces a finite-dimensional algebra. The relations of $S$ imply that to prove this it suffices to show that sufficiently high powers of the generators vanish; any monomial can be rearranged into lexicographic order and will contain such a power if of high enough degree. One can verify that this is true in our case via equations in Appendix \[subsec: calcskewcliff\]. By [@cassidy2010generlizations Theorem 4.2] the algebra $A(\mu,M_1,\ldots,M_4)$ is therefore an AS-regular domain of global dimension 4.
We will now apply a cocycle twist to $A(\mu,M_1,\ldots,M_4)$. Consider the graded action of $G=(C_2)^2=\langle
g_1,g_2\rangle$, defined on the generators by $$\label{eq: gencliffordaction}
g_1: x_1 \mapsto x_1,\; x_2 \mapsto -x_2,\; x_3 \mapsto x_3,\; x_4 \mapsto -x_4, \;\;\; x_i^{g_{2}}=-x_i^{g_{1}}\text{
for }i=1,2,3,4.$$
Our assumption on the characteristic of $k$ means that $\text{char}(k) \nmid |G|$ will always hold. As usual, we use the isomorphism $G\cong G^{\vee}$ given by and the 2-cocycle $\mu$ defined in . To avoid ambiguity we relabel this cocycle by $\tau$ for the duration of this section.
\[lem: relnsskewclifford\] The cocycle twist $A(\mu,M_1,\ldots,M_4)^{G,\tau}$ can be presented as the factor of the free $k$-algebra $k\{v_1,v_2,v_3,v_4\}$ by the ideal generated by the relations $$\begin{gathered}
\begin{aligned}\label{eq: gencliffordnewrelns}
&v_4v_1+iv_1v_4,\;\; v_3^2-v_1^2,\;\; v_3v_1-v_1v_3+v_2^2,\;\; v_3v_2+iv_2v_3,\;\; v_4^2-v_2^2,\\
&v_4v_2-v_2v_4+\gamma v_1^2.
\end{aligned}\end{gathered}$$
The algebra $A(\mu,M_1,\ldots,M_4)$ is generated in degree 1, thus by the action of $G$ defined in and Remark \[rem: montfingenremark\] one can conclude that $A(\mu,M_1,\ldots,M_4)^{G,\tau}$ is also generated in degree 1. The computations needed to obtain the relations in from those in are given in in Appendix \[subsec: calcskewcliff\]. Lemma \[lem: defrelns\] confirms that these relations are in fact the defining relations in the twist.
In fact, the cocycle twist $A(\mu,M_1,\ldots,M_4)^{G,\tau}$ can also be described as a Zhang twist of the ${\ensuremath{\mathbb{N}}}$-grading on $A(\mu,M_1,\ldots,M_4)$, as the following proposition shows. One might expect this behaviour since $g_1g_2$ acts by a scalar, and using such an action was the key idea in the proof of Proposition \[prop: recoverztwist\].
\[prop: zcocycletwist\] Let $\phi$ denote the algebra automorphism by which the element $g_1 \in G$ acts. Then there is an isomorphism of $k$-algebras $$A(\mu,M_1,\ldots,M_4)^{G,\tau} \cong A(\mu,M_1,\ldots,M_4)^{{\ensuremath{\mathbb{N}}},\phi},$$ where the algebra on the right is the Zhang twist of the ${\ensuremath{\mathbb{N}}}$-grading on $A(\mu,M_1,\ldots,M_4)$ by $\phi$.
Consider the relations of $A(\mu,M_1,\ldots,M_4)$ in . Their behaviour under a Zhang twist by $\phi$ is given below: $$\begin{gathered}
\begin{aligned}\label{eq: ztwistskewcliffrelns}
&x_4x_1-ix_1x_4 = x_4 \ast_{\phi} x_1 + ix_1 \ast_{\phi} x_4,\;\;\;\;\;\; x_3^2-x_1^2 = x_3 \ast_{\phi} x_3 - x_1
\ast_{\phi} x_1,\\
&x_3x_2-ix_2x_3 = -x_3 \ast_{\phi} x_2 - i x_2 \ast_{\phi} x_3,\;\;\; x_4^2-x_2^2 = -x_4 \ast_{\phi} x_4 + x_2
\ast_{\phi} x_2,\\
&x_3x_1-x_1x_3+x_2^2 = x_3 \ast_{\phi} x_1 - x_1 \ast_{\phi} x_3 - x_2 \ast_{\phi} x_2,\\
&x_4x_2-x_2x_4+\gamma x_1^2 = - x_4 \ast_{\phi} x_2 + x_2 \ast_{\phi} x_4 +\gamma x_1 \ast_{\phi} x_1.
\end{aligned} \end{gathered}$$
To obtain the relations in $A(\mu,M_1,\ldots,M_4)^{G,\tau}$ from those in , one needs to rescale the generators by $x_2 \mapsto -x_2$ and $x_3 \mapsto -x_3$.
We now consider the question of whether $A(\mu,M_1,\ldots,M_n)^{G,\tau}$ is another graded skew Clifford algebra. At the end of [@cassidy2010generlizations Example 5.1] it is explained that some Zhang twists of their example correspond to other possible skew-symmetric matrices for the same normalising sequence. Since our example is a Zhang twist of this form by Proposition \[prop: zcocycletwist\], it seems likely that it would be associated to the same normalising quadric system. The relations in suggest that we would have $$\label{eq: multskwesymmetric}
\mu_{13}=-1,\;\; \mu_{14}=i,\;\; \mu_{24}=-1,\;\; \mu_{23}=-i,$$ in the associated multiplicatively skew-symmetric matrix if this were the case. We prove that the twist is indeed another graded skew Clifford algebra with the same normalising sequence, but a different skew-symmetric matrix.
\[prop: cliffordtwist\] For $\mu$ and $M_1,\ldots,M_n$ as in and , $$A(\mu,M_1,\ldots,M_n)^{G,\tau}\cong A(\mu',M_1,\ldots,M_n),$$ as $k$-algebras, where $$\label{eq: musystem}
\mu'=\begin{pmatrix} 1 & -i & -1 & i \\ i & 1 & -i & -1 \\ -1 & i & 1 & -i \\ -i & -1 & i & 1 \end{pmatrix}.$$ Moreover, $A(\mu,M_1,\ldots,M_n)^{G,\tau}$ is an AS-regular domain of global dimension 4.
From we already have several entries of a possible multiplicatively skew-symmetric matrix. The requirement that is still normalising allows us to calculate the rest of the entries as demonstrated in equations through in Appendix \[subsec: calcskewcliff\]. One obtains the matrix $\mu'$ given above.
We will show that $A(\mu',M_1,\ldots,M_n)$ satisfies the relations in . Using Definition \[defn: gradedskewclifford\](i) and the data from and , the relations in $A(\mu',M_1,\ldots,M_n)$ are $$\begin{gathered}
\begin{align*}&2x_1^2=y_3,\;\; 2x_2^2=y_4,\;\; 2x_3^2=y_3,\;\; 2x_4^2=y_4,\;\; x_1x_2-i x_2 x_1 = \frac{1}{2} y_1,\;\; x_1 x_3-x_3 x_1
= \frac{1}{2} y_4,\\
&x_1x_4-ix_4x_1=0,\;\; x_2x_3-i x_3 x_2 = 0,\;\; x_2 x_4-x_4 x_2 = \frac{\gamma}{2} y_3,\;\;
x_3x_4-ix_4x_3=\frac{1}{2}y_2.
\end{align*}\end{gathered}$$
One obtains the six relations from and some extra relations telling us how to write the remaining degree 2 generators in terms of the degree 1 generators. Thus one has a surjective map $A(\mu,M_1,\ldots,M_n)^{G,\tau} \rightarrow A(\mu',M_1,\ldots,M_n)$. Notice that the associated quadric system of $A(\mu',M_1,\ldots,M_n)$ is the same as for $A(\mu,M_1,\ldots,M_n)$, thus [@cassidy2010generlizations Theorem 4.2] implies that it must be an AS-regular domain of dimension 4. Since it is generated in degree 1 it must have the same Hilbert series as the twist, which implies that the surjection above is an isomorphism, completing the proof.
\[rem: twistotherproofcliff\] We could also have proved that the twist is AS-regular of dimension 4 using results from Chapter \[chap: cocycletwists\].
Recall from the proof of Theorem \[thm: new\] that $A(\mu,M_1,\ldots,M_n)$ can be considered as part of a 1-parameter family of algebras. This family can be denoted by $A(\gamma)$ in light of the final relation in . Cassidy and Vancliff state in [@cassidy2010generlizations Example 5.1] that the algebra $A(\gamma)$ has a 0-dimensional point scheme and a 1-dimensional line scheme for all $\gamma \in k^{\times}$. By Proposition \[prop: zcocycletwist\] the cocycle twist $A(\mu,M_1,\ldots,M_n)^{G,\tau}$ is also a Zhang twist of the ${\ensuremath{\mathbb{N}}}$-grading on $A(\mu,M_1,\ldots,M_n)$. It therefore follows from Theorem \[thm: ztwistgmodequiv\] that $A(\mu,M_1,\ldots,M_n)^{G,\tau}$ also has a 0-dimensional point scheme and a 1-dimensional line scheme.
Twisting a universal enveloping algebra {#subsec: homenvelopalg}
---------------------------------------
In this section we investigate cocycle twists of a universal enveloping algebra and its homogenisation. The relevant Lie algebra is $\mathfrak{sl}_2(k)$, where $k$ is some algebraically closed field of characteristic not equal to 2. Later in this section we will assume that $k={\ensuremath{\mathbb{C}}}$. After studying these twists we will discuss them in relation to the papers [@lebruyn1993homogenized] and [@lebruyn1993on]. The first of these papers contrasts the Lie-theoretic geometry of $\mathfrak{sl}_2({\ensuremath{\mathbb{C}}})$ with the geometry encoded in $\text{qgr}(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))$, while the latter generalises these ideas to the homogenised enveloping algebra of any finite-dimensional complex Lie algebra.
The standard presentation of $U(\mathfrak{sl}_2(k))$ is given by $k$-algebra generators $e,f,h$ subject to the relations $$ef-fe=[e,f]=h,\;\; he-eh=[h,e]=2e,\;\; hf-fh=[h,f]=-2f,$$ where $[-,-]$ denotes the Lie bracket on $\mathfrak{sl}_2(k)$.
As usual, to twist this algebra we need to first find some algebra automorphisms. Observe that if such an automorphism acts diagonally on the given generators and is defined by $$e \mapsto \lambda_1 e,\;\; f \mapsto \lambda_2 f,\;\; h \mapsto \lambda_3 h,$$ then we must have $\lambda_3=1$ and $\lambda_1 \lambda_2=1$. In addition to such scalar automorphisms, the maps defined by $$e \mapsto f,\;\; f \mapsto e,\;\; h \mapsto \pm h,$$ also define automorphisms, although the given generators do not form a diagonal basis.
We will consider the action of $G=(C_2)^2=\langle g_1,g_2\rangle$ given by $$\label{eq: homogenisedsl2nondiagact}
g_1:\; e \mapsto f,\;\; f \mapsto e,\;\; h \mapsto -h, \;\; g_2:\; e \mapsto -e,\;\; f \mapsto -f,\;\; h \mapsto h.$$
The change of generators $E=e+f,F=e-f,H=h$ produces a diagonal basis, and the relations of $U(\mathfrak{sl}_2(k))$ in terms of this new basis are $$\label{eq: homogenisedsl2newbasis}
EF-FE+2H,\;\; HE-EH-2F,\;\; HF-FH-2E.$$
Under the action of $G$ defined in and the isomorphism $G \cong G^{\vee}$ given by , the induced $G$-grading on the new generators of $U(\mathfrak{sl}_2(k))$ is $$\label{eq: homogenisedsl2nondiaggrad}
E \in U(\mathfrak{sl}_2(k))_{g_{1}},\;\; F \in U(\mathfrak{sl}_2(k))_{g_{1}g_{2}},\;\; H \in
U(\mathfrak{sl}_2(k))_{g_{2}}.$$
We can now give the relations of the cocycle twist that we will study henceforth.
Let $\mu$ be the 2-cocycle defined in . Then the cocycle twist $U(\mathfrak{sl}_2(k))^{G,\mu}$ obtained using the $G$-grading in has the following three defining relations: $$\label{eq: homogenisedsl2nondiagrelns}
E \ast_{\mu} F+F \ast_{\mu} E-2H,\;\; H \ast_{\mu} E+E \ast_{\mu} H-2F,\;\; H \ast_{\mu} F+F \ast_{\mu} H-2E.$$
These relations can be computed from those in , with the computations being given in Appendix \[subsec: calcenvelop\]. Lemma \[lem: defrelns\] implies that the ideal of relations in the twist is generated by precisely these three relations.
The homogenisation of $U(\mathfrak{sl}_2(k))$, denoted $U_h(\mathfrak{sl}_2(k))$, is the ${\ensuremath{\mathbb{N}}}$-graded algebra obtained by homogenising the defining relations of the enveloping algebra by a central generator, $t$ say. One immediate consequence of this is that $$\label{eq: dehomogenise}
U(\mathfrak{sl}_2(k))\cong U_h(\mathfrak{sl}_2(k))/(t-1),$$ while factoring out the ideal generated by $t$ produces a polynomial ring on three generators by the PBW Theorem [@krause2000growth Theorem 6.8].
The action of $G$ defined in can be extended to $U_h(\mathfrak{sl}_2(k))$ by letting $G$ act trivially on $t$. One can then twist the induced $G$-grading by the 2-cocycle $\mu$ to obtain $U_{h}(\mathfrak{sl}_2(k))^{G,\mu}$, whose defining relations are comprised of the homogenisations of those in : $$E \ast_{\mu} F+F \ast_{\mu} E-2H \ast_{\mu} t,\;\; H \ast_{\mu} E+E \ast_{\mu} H-2F \ast_{\mu} t,\;\; H \ast_{\mu} F+F
\ast_{\mu} H-2E \ast_{\mu} t,$$ together with the three additional relations $$t \ast_{\mu} E - E \ast_{\mu} t,\;\;\; t \ast_{\mu} F - F \ast_{\mu} t,\;\text{ and }\; t \ast_{\mu} H - H \ast_{\mu} t.$$
The following result is a twisted version of .
\[lem: isoassgr\] There is an isomorphism of $k$-algebras $$U(\mathfrak{sl}_2(k))^{G,\mu}\cong U_h(\mathfrak{sl}_2(k))^{G,\mu}/(t-1).$$
The result follows from the fact that $t-1$ is fixed by the action of $G$.
Let us now assume that $k={\ensuremath{\mathbb{C}}}$ and prove that $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$ has several good properties.
\[prop: homogeniseprops\] The cocycle twist $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$ is noetherian, Auslander regular of global dimension 4 and Cohen-Macaulay. Furthermore, it has Hilbert seres $1/(1-t)^4$ and is generated in degree 1.
As noted at the beginning of [@lebruyn1993homogenized §2], $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))$ has all of the properties mentioned in the statement of the proposition. The result follows by Remark \[rem: montfingenremark\] and an application of Lemma \[lem: hilbseries\], Corollary \[cor: uninoeth\] and Proposition \[prop: cohenmac\].
Now let us turn to the papers cited at the beginning of the section. It is remarked at the top of pg. 728 in [@lebruyn1993homogenized] that there is a dichotomy in the problem of finding linear modules over $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))$. In fact, this occurs in the more general situation of a finite dimensional complex Lie algebra $\mathfrak{g}$, its universal enveloping algebra $U(\mathfrak{g})$ and homogenisation $U_{h}(\mathfrak{g})$. Le Bruyn and Van den Bergh use the following fact in the proof of [@lebruyn1993on Theorem 2.2]: the homogenising generator must either act faithfully on a linear module over $U_{h}(\mathfrak{g})$ or annihilate it.
The $d$-linear modules over $U_{h}(\mathfrak{g})$ that are not annihilated by the homogenising generator come from 1-dimensional representations of Lie subalgebras of $\mathfrak{g}$ of codimension $d$, which are induced up to $U(\mathfrak{g})$ and then homogenised. Corollary 3.4 and Theorem 3.5 from [@lebruyn1993on] describe the case when $d=0$ in more detail. The structure of the point scheme of $U_{h}(\mathfrak{g})$ depends on whether $\mathfrak{g} = [\mathfrak{g},\mathfrak{g}]$ or not: if this is true then the point scheme contains an embedded component; otherwise, the point scheme is reduced.
We will now consider point modules over $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$, for which there is also a dichotomy. By definition such modules are 1-critical, therefore by [@levasseur1993modules Lemma 2.10] the generator $t$ either acts faithfully on a point module or annihilates it. Our final result describes the point modules over $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$.
The point modules of $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$ correspond to the three lines $t=E=0$, $t=F=0$ and $t=H=0$, as well as the additional points $$\label{eq: homsl2extrapts}
(1,1,1,1),\;\; (1,1,-1,-1),\;\; (1,-1,1,-1)\;\text{ and }\; (1,-1,-1,1).$$
By Proposition \[prop: homogeniseprops\], $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$ satisfies the hypotheses of Theorem \[thm: pointschemenice\]. That theorem implies that the graph of the point scheme of $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}$ under its associated automorphism arises as the scheme determined by the multilinearisations of the defining relations.
The matrix formulation of such multilinearisations (as in Lemma \[lem: pointsparameterisevcliff\]) can be used to see that the point scheme consists of points $(t,E,F,H) \in {\ensuremath{\mathbb{P}_{k}^{3}}}$ for which the matrix $$\label{eq: homsl2matrix}
\left(\begin{array}{cccc}
-2H & F & E & 0 \\
-2F & H & 0 & E \\
-2E & 0 & H & F \\
E & -t & 0 & 0 \\
F & -t & 0 & 0 \\
H & 0 & 0 & -t
\end{array}\right)$$ has rank precisely 3.
Consider those point modules which are annihilated by $t$, which enables us to set $t= 0$ in . Such point modules restrict to the factor ring $U_h(\mathfrak{sl}_2({\ensuremath{\mathbb{C}}}))^{G,\mu}/(t)$. This ring is isomorphic as a ${\ensuremath{\mathbb{C}}}$-algebra to $${\ensuremath{\mathbb{C}}}_{-1}[E,F,H] := {\ensuremath{\mathbb{C}}}\{E,F,H\}/(EF+FE,EH+HE,FH+HF).$$ It is well-known that any point module over this algebra is annihilated by one of the generators, with any point on one of the lines in the statement of the result giving rise to a point module.
By the dichotomy on point modules it remains to consider point modules on which $t$ acts faithfully. Thus we may assume that $t=1$ into . Under this condition, the only points at which the matrix in has rank 3 are precisely those stated in .
Calculations {#app: calc}
============
In this appendix we collect together some of the calculations that were omitted earlier in the thesis.
Calculations for Chapter \[chap: thcrtwist\] {#sec: calcthcr}
--------------------------------------------
### Additional proof of Proposition \[prop: bgmunoptmodules\] {#app: calcptmodnonproof}
We now give the more computational proof that the point scheme of $B^{G,\mu}$ is empty when $|\sigma| = \infty$.
The point modules over $B^{G,\mu}$ are precisely those over $A^{G,\mu}$ that are annihilated by both of the central elements $\Theta_1$ and $\Theta_2$. We give the necessary computations to show that the $\Theta_1$ does not vanish at any of the 20 points in the point scheme of $A^{G,\mu}$.
By Lemma \[lem: annGinvariant\], if an element that is fixed by $G$ annihilates a point module then it annihilates the other point modules in the same $G$-orbit. Thus it suffices to show that $\Theta_1$ does not vanish when evaluated at one point from each $G$-orbit.
In Lemma \[lem: gammaorbits\] we described the $G$-orbits of the point scheme of $A^{G,\mu}$; there are four singleton orbits consisting of a point of the form $e_j$ and four more orbits of order 4. Furthermore, Lemma \[lem: ptschemecontains\] describes the behaviour of these points under the automorphism $\phi$ that is associated to the point scheme. It is clear that $\Theta_1(e_j,e_j)\neq 0$ for $i=0,1,2,3$, therefore we can immediately dismiss the singleton orbits.
Now let $p=(1, i, -i ,-1)$, which will represent the remaining orbit of points fixed under $\phi$. Computing $\Theta_1(p,p)$, we find that $$\Theta_1(p,p) = -v_0(1)v_0(1)+v_1(i)v_1(i)+v_2(-i)v_2(-i)-v_3(-1)v_3(-1) =-4.$$
Now let $p = \left(1,-(\beta \gamma)^{-\frac{1}{2}},\gamma^{-\frac{1}{2}},\beta^{-\frac{1}{2}}\right)$, therefore $p^{\phi}=\left(1,-(\beta \gamma)^{-\frac{1}{2}},-\gamma^{-\frac{1}{2}},-\beta^{-\frac{1}{2}}\right)$. Evaluating $\Theta_1$ at $(p,p^{\phi})$ gives $$\begin{gathered}
\begin{aligned}\label{eq: 4sklyaninptmodeg1}
\Theta_1(p,p^{\phi}) =\; &-v_0(1)v_0(1)+v_1(-(\beta \gamma)^{-\frac{1}{2}})v_1(-(\beta
\gamma)^{-\frac{1}{2}})+v_2(\gamma^{-\frac{1}{2}})v_2(-\gamma^{-\frac{1}{2}})\\
&-v_3(\beta^{-\frac{1}{2}})v_3(-\beta^{-\frac{1}{2}})\\
=\; &-1 + \frac{1}{\beta \gamma}-\frac{1}{\gamma}+\frac{1}{\beta}.
\end{aligned}\end{gathered}$$ If were equal to zero then it could be rearranged to the form $(1-\beta)(\gamma+1)=0$. Both solutions correspond to choices of parameters that do not satisfy , which is not permitted.
Now let $p = \left(1,i \gamma^{-\frac{1}{2}},(\alpha \gamma)^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}}\right)$, thus $p^{\phi} =\left(1,-i \gamma^{-\frac{1}{2}},(\alpha \gamma)^{-\frac{1}{2}}, -i \alpha^{-\frac{1}{2}}\right)$. One has $$\begin{gathered}
\begin{aligned}\label{eq: 4sklyaninptmodeg2}
\Theta_1(p,p^{\phi} ) =\; &-v_0(1)v_0(1)+v_1(i \gamma^{-\frac{1}{2}})v_1(-i \gamma^{-\frac{1}{2}})+v_2((\alpha
\gamma)^{-\frac{1}{2}})v_2((\alpha \gamma)^{-\frac{1}{2}})\\
&-v_3(i \alpha^{-\frac{1}{2}})v_3(-i \alpha^{-\frac{1}{2}})\\
=\; &-1 + \frac{1}{\gamma}+\frac{1}{\alpha\gamma}-\frac{1}{\alpha}.
\end{aligned}\end{gathered}$$ As for the previous orbit, if were equal to zero then it could be rearranged to the form $(\alpha+1)(1-\gamma)=0$. Again, both solutions correspond to choices of parameters that we have excluded.
Finally, let $p = \left(1,\beta^{-\frac{1}{2}}, i \alpha^{-\frac{1}{2}},i(\alpha \beta)^{-\frac{1}{2}}\right)$, thus $p^{\phi} =\left(1,-\beta^{-\frac{1}{2}}, -i \alpha^{-\frac{1}{2}},i(\alpha \beta)^{-\frac{1}{2}}\right)$. Once again $$\begin{gathered}
\begin{aligned}\label{eq: 4sklyaninptmodeg3}
\Theta_1(p,p^{\phi}) =\; &-v_0(1)v_0(1)+v_1(\beta^{-\frac{1}{2}})v_1(-\beta^{-\frac{1}{2}})+v_2(i
\alpha^{-\frac{1}{2}})v_2(-i \alpha^{-\frac{1}{2}})\\
&-v_3(i(\alpha \beta)^{-\frac{1}{2}})v_3(i(\alpha \beta)^{-\frac{1}{2}})\\
=\; &-1 - \frac{1}{\beta}+\frac{1}{\alpha}+\frac{1}{\alpha \beta}.
\end{aligned}\end{gathered}$$ If were equal to zero then it could be rearranged to the form $(1-\alpha)(\beta+1)=0$, both of whose solutions lead to forbidden parameter triples.
Calculations for Chapter \[chap: othertwists\]
----------------------------------------------
### Relations in §\[sec: vancliffql\] {#subsec: calcvancliff}
The relations of $R(\alpha,\beta,\lambda)^{G,\mu}$ that are not preserved under twisting are calculated as follows: $$\begin{aligned}
0 &=x_4x_3 - \alpha x_3 x_4 = \frac{x_4 \ast_{\mu} x_3}{\mu(g_1g_2,g_2)} - \alpha \frac{x_3 \ast_{\mu}
x_4}{\mu(g_2,g_1g_2)} = -(x_4 \ast_{\mu} x_3 + \alpha x_3 \ast_{\mu} x_4). \\
0 &=x_4 x_2 - \lambda x_2 x_4 = \frac{x_4 \ast_{\mu} x_2}{\mu(g_1g_2,g_1)} - \lambda \frac{x_2 \ast_{\mu}
x_4}{\mu(g_1,g_1g_2)} = x_4 \ast_{\mu} x_2+ \lambda x_2 \ast_{\mu} x_4. \\
0 &=x_3x_2 - \beta x_2 x_3 - (\alpha \beta - \lambda)x_1 x_4 \\
&= \frac{x_3 \ast_{\mu} x_2}{\mu(g_2,g_1)} - \beta \frac{x_2 \ast_{\mu} x_3}{\mu(g_1,g_2)} - (\alpha \beta - \lambda)
\frac{x_1 \ast_{\mu} x_4}{\mu(e,g_1g_2)} \\
&= x_3 \ast_{\mu} x_2 + \beta x_2 \ast_{\mu} x_3 - (\alpha \beta - \lambda) x_1 \ast_{\mu} x_4. \end{aligned}$$
### Calculations for §\[subsec: gradedskewclifford\] {#subsec: calcskewcliff}
The relations of $A(\mu,M_1,\ldots,M_n)^{G,\tau}$ are computed as follows: $$\begin{gathered}
\begin{aligned}\label{eq: relnskewcliff}
0 &= x_4x_1-ix_1x_4= \frac{x_4 \ast_{\tau} x_1}{\tau(g_2,g_1)}-i \frac{x_1 \ast_{\tau} x_4}{\tau(g_1,g_2)} = x_4
\ast_{\tau} x_1 + i x_1 \ast_{\tau} x_4. \\
0 &= x_3^2-x_1^2 = \frac{x_3 \ast_{\tau} x_3}{\tau(g_1,g_1)}- \frac{x_1 \ast_{\tau} x_1}{\tau(g_1,g_1)} = x_3
\ast_{\tau} x_3 - x_1 \ast_{\tau} x_1. \\
0 &= x_3x_1-x_1x_3+x_2^2 = \frac{x_3 \ast_{\tau} x_1}{\tau(g_1,g_1)}- \frac{x_1 \ast_{\tau} x_3}{\tau(g_1,g_1)}
+\frac{x_2 \ast_{\tau} x_2}{\tau(g_2,g_2)} \\
&= x_3 \ast_{\tau} x_1 - x_1 \ast_{\tau} x_3 + x_2 \ast_{\tau} x_2. \\
0 &= x_3x_2-ix_2x_3 = \frac{x_3 \ast_{\tau} x_2}{\tau(g_1,g_2)}- i \frac{x_2 \ast_{\tau} x_3}{\tau(g_2,g_1)} = - x_3
\ast_{\tau} x_2 - i x_2 \ast_{\tau} x_3. \\
0 &= x_4^2-x_2^2 = \frac{x_4 \ast_{\tau} x_4}{\tau(g_2,g_2)}- \frac{x_2 \ast_{\tau} x_2}{\tau(g_2,g_2)} = x_4
\ast_{\tau} x_4 - x_2 \ast_{\tau} x_2. \\
0 &= x_4x_2-x_2x_4+\gamma x_1^2 = \frac{x_4 \ast_{\tau} x_2}{\tau(g_2,g_2)}- \frac{x_2 \ast_{\tau} x_4}{\tau(g_2,g_2)} +
\gamma \frac{x_1 \ast_{\tau} x_1}{\tau(g_1,g_1)} \\
&= x_4 \ast_{\tau} x_2 - x_2 \ast_{\tau} x_4 + \gamma x_1 \ast_{\tau} x_1.
\end{aligned}\end{gathered}$$
We now show that $S/(q_1,q_2,q_3,q_4)$ is finite-dimensional. As noted in §\[subsec: gradedskewclifford\], it suffices to show that sufficiently high powers of the generators vanish. The following calculations verify this: $$\begin{gathered}
\begin{aligned}\label{eq: 5thpowervanish1}
z_1^5 &=z_1(-z_3^2-\gamma z_2 z_4)z_1^2=-z_1z_3^2 z_1^2=(z_2^2+z_4^2)z_3z_1^2=0, \\
z_2^5 &=z_2(-z_4^2- z_1 z_3)z_2^2=-z_2z_4^2 z_2^2=\gamma^{-1}(z_1^2+z_3^2)z_4z_2^2=0, \\
z_3^5 &=z_3(-z_1^2-\gamma z_2 z_4)z_3^2=-z_3z_1^2 z_3^2=z_3 z_1(z_2^2+z_4^2)z_3=0, \\
z_4^5 &=z_4(-z_2^2-z_1 z_3)z_4^2=-z_4z_2^2 z_4^2=\gamma^{-1} z_4 z_2(z_1^2+z_3^2)z_4=0.
\end{aligned}\end{gathered}$$
Our final calculation in this section shows that the quadric system given in is still normalising for the multiplicatively skew-symmetric matrix in . Note that $q_1$ and $q_2$ are monomials and therefore are clearly normal in $S$, therefore we only need to check that $q_3$ and $q_4$ are normal. At first we only have the partial information given in , from which we will deduce the remaining entries. $$\begin{gathered}
\begin{aligned}\label{eq: normalisingsystem}
z_1 q_3 = z_1^3 +z_1 z_3^2 + \gamma z_1 z_2 z_4 &= (z_1^2+ \mu_{31}^2 z_3^2+ \mu_{21}\mu_{41} \gamma z_2 z_4) z_1 \\
&= (z_1^2+ z_3^2- \mu_{21} i \gamma z_2 z_4) z_1.
\end{aligned}\end{gathered}$$ This suggests that we must take $\mu_{21}=i=-\mu_{12}$ so that $z_1 q_3= q_3 z_1$. Continuing in the same manner we have $$\begin{gathered}
\begin{aligned}\label{eq: normalisingsystem1}
z_2 q_3 = z_2 z_1^2 +z_2 z_3^2 + \gamma z_2^2 z_4 &= (\mu_{12}^2 z_1^2+ \mu_{32}^2 z_3^2+ \mu_{42} \gamma z_2 z_4) z_2 =
-q_3 z_2, \\
z_3 q_3 = z_3 z_1^2 + z_3^3 + \gamma z_3 z_2 z_4 &= (\mu_{13}^2 z_1^2+ z_3^2+ \mu_{23}\mu_{43} \gamma z_2 z_4) z_3 \\
&= (z_1^2+ z_3^2-i \mu_{43} \gamma z_2 z_4) z_3.
\end{aligned}\end{gathered}$$ This suggests that we must take $\mu_{43}=i=-\mu_{34}$ so that $z_3 q_3= q_3 z_3$, completing the skew-symmetric matrix $\mu'$. The final calculation for $q_3$ is $$\begin{gathered}
\begin{aligned}\label{eq: normalisingsystem3}
z_4 q_3 &= z_4 z_1^2 +z_4 z_3^2 + \gamma z_4 z_2 z_4 = (\mu_{14}^2 z_1^2+ \mu_{34}^2 z_3^2+ \mu_{24} \gamma z_2 z_4) z_4
= -q_3 z_4.
\end{aligned}\end{gathered}$$
Checking that $q_4$ is normal is now straightforward: $$\begin{gathered}
\begin{aligned}\label{eq: normalisingsystem4}
z_1 q_4 &= z_1 z_2^2+z_1 z_4^2+z_1^2 z_3 = (\mu_{21}^2 z_2^2+\mu_{41}^2 z_4^2+ \mu_{31} z_1 z_3) z_1 = -q_4 z_1,\\
z_2 q_4 &= z_2^3+z_2 z_4^2+z_2 z_1 z_3 = (z_2^2+\mu_{42}^2 z_4^2+ \mu_{12}\mu_{32} z_1 z_3) z_2 = q_4 z_2,\\
z_3 q_4 &= z_3 z_2^2+z_3 z_4^2+z_3 z_1 z_3 = (\mu_{23}^2 z_2^2+\mu_{43}^2 z_4^2+ \mu_{13} z_1 z_3) z_3 = -q_4 z_3,\\
z_4 q_4 &= z_4 z_2^2+ z_4^3+z_4 z_1 z_3 = (\mu_{24}^2 z_2^2+ z_4^2+ \mu_{14}\mu_{34} z_1 z_3) z_4 = q_4 z_4,\\
\end{aligned}\end{gathered}$$
### Relations in §\[subsec: homenvelopalg\] {#subsec: calcenvelop}
The relations of $U(\mathfrak{sl}_2)^{G,\mu}$ are computed below. $$\begin{aligned}
\label{eq: relnenvelop}
0 &= EF-FE+2H = \frac{E\ast_{\mu}F}{\mu(g_1,g_1g_2)}-\frac{F\ast_{\mu}E}{\mu(g_1g_2,g_1)} +2H = -E\ast_{\mu}F-
F\ast_{\mu}E +2H. \\
0 &= HE-EH-2F = \frac{H\ast_{\mu}E}{\mu(g_2,g_1)}-\frac{E\ast_{\mu}H}{\mu(g_1,g_2)} -2F = H\ast_{\mu}E + E\ast_{\mu}H -2
F. \\
0 &= HF-FH-2E = \frac{H\ast_{\mu}F}{\mu(g_2,g_1g_2)}-\frac{F\ast_{\mu}H}{\mu(g_1g_2,g_2)} -2E = H\ast_{\mu}F +
F\ast_{\mu}H - 2E. \end{aligned}$$
Computer code {#app: comp}
=============
In this appendix we give the computer code used in several proofs earlier in the thesis. The two programs that we used in relation to the work in this thesis are Macaulay2 and Affine, a brief description of which follow.
Macaulay2 is a commutative algebra/algebraic geometry program that we use to prove results regarding the point and line schemes of some cocycle twists. More information on this software can be obtained from [@M2] or [@eisenbud2002computations].
Affine is a package for the computer algebra system Maxima [@maxima]. It can be used to perform Groebner basis-like computations with noncommutative algebras. More specifically, Affine implements the diamond lemma, an important result that originated in graph theory. It was first studied in the context of noncommutative algebras in [@bergman1978diamond]. We utilised Affine to study the behaviour of several examples of cocycle twists.
Code for §\[subsec: 4sklyanintwistandptscheme\] {#sec: sklyanincalcs}
-----------------------------------------------
The following Macaulay2 code is needed in Proposition \[prop: linescheme1dim\] to calculate the dimension of the line scheme of $A(\alpha,\beta,\gamma)^{G,\mu}$.
\[code: lineschemeagmu\]
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&
Code for §\[subsec: staffordalgs\] {#subsec: staffordcalcs1}
----------------------------------
The following two pieces of Macaulay2 code are used in the proof of Proposition \[prop: staffordptschemes\]. The first piece of code is used to determine the intersection of the scheme $\Gamma_2$ associated to $S_{\infty}^{G,\mu}(\alpha,\beta,\gamma)$ with the affine subscheme of ${\ensuremath{\mathbb{P}_{k}^{3}}} \times {\ensuremath{\mathbb{P}_{k}^{3}}}$ in which $v_{01},v_{02} \neq 0$.
\[code: lineschemeainfty1\]
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
& &\
&&
The second piece of code shows that there are no points lying in the intersection of $\Gamma_2$ with the locus where $v_{02}=0$ and the remaining coordinates are non-zero. We can therefore assume that $v_{01}=1$ and $v_{11}v_{12}v_{21}v_{22}v_{31}v_{32}=1$ by rescaling the two copies of ${\ensuremath{\mathbb{P}_{k}^{3}}}$.
\[code: lineschemeainfty2\]
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&
Finally, we give the Macaulay2 code used in the proof of Proposition \[prop: staffordlinescheme\], which is used to calculate the dimension of the line scheme of $S_{\infty}^{G,\mu}$.
\[code: lineschemeainfty\]
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&\
&&
Properties preserved under twisting {#app: preserve}
===================================
In this appendix we give a table summarising the properties that we show are preserved under cocycle twisting in §\[sec: preservation\]. The table includes a reference for each result and any relevant hypotheses.\
[| c | c | c |]{}
**Property**
&
**Reference**
&
**Hypotheses**
\
Hilbert series & Lemma \[lem: hilbseries\] & Hyp. \[hyp: gradedcase\]\
Strongly noetherian & Corollary \[cor: uninoeth\] &\
GK dimension & Proposition \[prop: gkdim\] & Hyp. \[hyp: generalcase\]\
Global dimension & Proposition \[prop: gldim\] &\
AS-Gorenstein & Proposition \[prop: asgor\] & Hyp. \[hyp: gradedcase\]\
AS-regular & Corollary \[cor: asreg\] & Hyp. \[hyp: gradedcase\], c.g. algebra\
Koszul & Proposition \[prop: koszul\] & Hyp. \[hyp: gradedcase\], quadratic algebra\
Cohen-Macaulay & &\
Auslander-Gorenstein & Proposition \[prop: cohenmac\] & Hyp. \[hyp: gradedcase\], noetherian algebra\
Auslander regular & &\
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Multitask learning, i.e. learning several tasks at once with the same neural network, can improve performance in each of the tasks. Designing deep neural network architectures for multitask learning is a challenge: There are many ways to tie the tasks together, and the design choices matter. The size and complexity of this problem exceeds human design ability, making it a compelling domain for evolutionary optimization. Using the existing state of the art soft ordering architecture as the starting point, methods for evolving the modules of this architecture and for evolving the overall topology or routing between modules are evaluated in this paper. A synergetic approach of evolving custom routings with evolved, shared modules for each task is found to be very powerful, significantly improving the state of the art in the Omniglot multitask, multialphabet character recognition domain. This result demonstrates how evolution can be instrumental in advancing deep neural network and complex system design in general.'
author:
- 'Jason Liang, Elliot Meyerson, and Risto Miikkulainen'
bibliography:
- 'references.bib'
title: Evolutionary Architecture Search For Deep Multitask Networks
---
<ccs2012> <concept> <concept\_id>10010147.10010257.10010293.10010294</concept\_id> <concept\_desc>Computing methodologies Neural networks</concept\_desc> <concept\_significance>500</concept\_significance> </concept> <concept> <concept\_id>10010147.10010178.10010219</concept\_id> <concept\_desc>Computing methodologies Distributed artificial intelligence</concept\_desc> <concept\_significance>300</concept\_significance> </concept> <concept> <concept\_id>10010147.10010178.10010224</concept\_id> <concept\_desc>Computing methodologies Computer vision</concept\_desc> <concept\_significance>100</concept\_significance> </concept> </ccs2012>
Introduction
============
In multitask learning (MTL) , a neural network is trained simultaneously to perform several different tasks at once [@Caruana:1998]. For instance, given an image as input, it can recognize the objects in it, identify the type of scene, and generate a verbal caption for it. Typically the early parts of the network are shared between tasks, and the later parts, leading to the different tasks, are separate [@Caruana:1998; @Collobert:2008; @Dong:2015; @Lu:2016; @Ranjan:2016]. The network is trained with gradient descent in all these tasks, and therefore the requirements of all tasks are combined in the shared parts of the network. The embeddings thus reflect the requirements of all tasks, making them more robust and general. Performance of a multitask network in each task can therefore exceed the performance of a network trained in only a single task.
Much of the research in deep learning in recent years has focused on coming up with better architectures, and MTL is no exception. As a matter of fact, architecture plays possibly an even larger role in MTL because there are many ways to tie the multiple tasks together. The best network architectures are large and complex, and have become very hard for human designers to optimize [@szegedy2015going; @szegedy:cvpr16; @zoph:arxiv16; @jaderberg2017population]
This paper develops an automated, flexible approach for evolving architectures, i.e. hyperparameters, modules, and module routing topologies, of deep multitask networks. A recent deep MTL architecture called soft ordering [@Meyerson:2018] is used as a starting point, in which a different soft sequence of modules is learned for each task. This paper extends this architecture in several ways. First, this paper proposes a novel algorithm for evolving *task specific routings* that create an unique routing between modules for each task. Second, *more general modules* with the same soft ordering architecture are evolved. Third, the general modules are evolved together with a *blueprint*, a shared routing for all tasks, that improves upon the soft ordering architecture. Fourth, as a capstone architecture, the task specific routing are evolved together with the general modules. These four approaches are evaluated in the Omniglot task [@Lake:2015] of learning to recognize characters from many different alphabets. A series of results confirms the intuition well: As a baseline, soft ordering performs significantly better in each task than single-task training (67% vs. 61% accuracy). Evolution of modules and topologies improves significantly upon soft ordering. Coevolution of modules and topologies together improves even more, and the capstone architecture turns out to be the best (at 88%). The results thus demonstrate three general points: evolutionary architecture search can make a large difference in performance of deep learning networks; MTL can improve performance of deep learning tasks; and putting these together results in a particularly powerful approach. In the future it can be applied to various problems in vision, language, and control, and in particular to domains with multimodal inputs and outputs. The rest of the paper is organized as follows: In Section \[sec:background\], previous work on deep MTL and neural architecture search are summarized. In Section \[sec:algorithms\], the key contribution of this paper, novel evolutionary algorithms for architecture search of multitask networks are described. Finally, in Section \[sec:experiments\] and Section \[sec:discussion\] experimental results on the Omniglot domain are presented and analyzed.
Background and Related Work {#sec:background}
===========================
Before introducing methods for combining them in Section \[sec:algorithms\], this section reviews deep MTL and neural architecture search.
![\[fig:soft\_order\] Example soft ordering network with three shared layers. Soft ordering learns how to use the same layers in different locations by learning a tensor $S$ of task-specific scaling parameters. $S$ is learned jointly with the $W_d$, to allow flexible sharing across tasks and depths. This architecture enables the learning of layers that are used in different ways at different depths for different tasks.](figures/soft_ordering-crop.pdf){width="\linewidth" height="1.1in"}
Deep Multitask Learning
-----------------------
MTL [@Caruana:1998] exploits relationships across problems to increase overall performance. The underlying idea is that if multiple tasks are related, the optimal models for those tasks will be related as well. In the convex optimization setting, this idea has been implemented via various regularization penalties on shared parameter matrices [@Argyriou:2008; @Evgeniou:2004; @Kang:2011; @Kumar:2012]. Evolutionary methods have also had success in MTL, especially in sequential decision-making domains [@Huizinga:2016; @Kelly:2017; @Jaskowski:2008; @Schrum:2016; @Snel:2010].
Deep MTL has extended these ideas to domains where deep learning thrives, including vision [@Bilen:2017; @Kaiser:2017; @Lu:2016; @Misra:2016; @Ranjan:2016; @Rebuffi:2017; @Yang:2017; @Zhang:2014], speech [@Huang:2013; @Huang:2015; @Kaiser:2017; @Seltzer:2013; @Wu:2015], natural language processing [@Collobert:2008; @Dong:2015; @Hashimoto:2016; @Kaiser:2017; @Liu:2015; @Luong:2016; @Zhang:2016], and reinforcement learning [@Devin:2016; @Jaderberg:2016; @Teh:2017]. The key design decision in constructing a deep multitask network is deciding how parameters such as convolutional kernels or weight matrices are shared across tasks. Designing a deep neural network for a single task is already a high-dimensional open-ended optimization problem; having to design a network for multiple tasks *and* deciding how these networks share parameters grows this search space combinatorially. Most existing approaches draw from the deep learning perspective that each task has an underlying feature hierarchy, and tasks are related through an *a priori* alignment of their respective hierarchies. These methods have been reviewed in more detail in previous work [@Meyerson:2018; @Ruder:2017]. Another existing approach adapts network structure by learning task hierarchies, though it still assumes this strong hierarchical feature alignment [@Lu:2016].
Soft ordering is a recent approach that avoids such an alignment by allowing shared layers to be used across different depths [@Meyerson:2018]. Through backpropagation, the joint model learns how to use each shared (potentially nonlinear) layer $W_d$ at each depth $d$ for the $t$-th task. This idea is implemented by learning a distinct scalar $s_{tdl}$ for each such location, which then multiplies the layer’s output. The final output at depth $d$ for the task is then the sum of these weighted outputs across layers, i.e., a *soft merge*. More generally, a soft merge is a learnable function given by $$\label{eq:soft_merge}
\text{softmerge}(\text{in}_1, \ldots, \text{in}_M) = \sum_{m=1..M} s_{m}\text{in}_m, \text{ with } \sum_{m=1..M} s_{m} = 1 \, ,$$ where the $\text{in}_m$ are a list of incoming tensors, $s_m$ are scalars trained simultaneously with internal layer weights via backpropagation, and the constraint that all $s_m$ sum to 1 is enforced via a softmax function. Figure \[fig:soft\_order\] shows an example soft ordering network.
More formally, given shared layers $W_1, W_2, \ldots, W_D$, the soft ordering model $\bm{y}_t = f(\bm{x}_t)$ for the $t$-th task $\{(\bm{y}_{ti}, \bm{x}_{ti})\}_{i=1}^N$ is given by $\bm{y}_t = \mathcal{D}_t(\bm{y}^D_t)$, where $\bm{y}^0_t = \mathcal{E}_t(\bm{x}_t)$ and $$\begin{aligned}
\bm{y}^{d}_t &= \text{softmerge}\big(W_1(\bm{y}^{d-1}_t), \ldots , W_D(\bm{y}^{d-1}_t)\big) \ \ \ \forall \ d \in 1..D\, , \label{eq:soft_order}\end{aligned}$$ where $\mathcal{E}_t$ is a task-specific encoder mapping the task input to the input of the shared layers, $\mathcal{D}_t$ is a task-specific decoder mapping the output of the shared layers to an output layer, e.g., classification. Although soft ordering allows flexible sharing across depths, layers are still only applied in a fixed grid-like topology, which biases and restricts the type of sharing that can be learned. This paper generalizes soft ordering layers to more general modules, and introduces evolutionary approaches to both design these modules and to discover how to assemble these modules into appropriate topologies for multitask learning. To our knowledge, this is the first paper that takes an evolutionary approach to deep MTL.
![Assembling networks for fitness evaluation in CoDeepNEAT. Modules from species specified in the blueprint are inserted into locations specified in the blueprint, forming a full network. This approach allows evolving repetitive and deep structures seen in many successful DNNs.[]{data-label="fg:assembling"}](figures/codeepneat.pdf){width="\linewidth"}
Architecture and Hyperparameter Search {#sc:nas}
--------------------------------------
As deep learning tasks and benchmarks become increasing complex, finding the right architecture becomes more important. In fact, the performance of many state of the art networks [@he:arxiv16; @szegedy2015going; @ng:arxiv15; @szegedy:cvpr16] depend mostly on novel and interesting architectural innovations. Unfortunately, discovering useful hyperparameters and architectures by hand is tedious and difficult; as a result, much research focuses on developing automated methods for doing it. Some promising methods for hyperparameter search include deep Bayesian optimization [@snoek:arxiv15] and CMA-ES [@loshchilov:arxiv16]. One unique approach uses reinforcement learning to develop an LSTM policy for generating appropriate network topologies and hyperparameters for a given task [@zoph:arxiv16].
One particular promising area of research is the use of evolutionary algorithms (EAs) for performing architecture search. Evolutionary methods are well suited for this kind of problems because they can be readily applied with no gradient information. Some of these approaches use a modified version of NEAT [@miikkulainen2017evolving; @real2017large], an EA for neuron-level neuroevolution [@stanley:ec02], for searching network topologies. Others rely on genetic programming [@suganuma2017genetic] or hierarchical evolution [@jaderberg2017population]. Along these lines, CoDeepNEAT [@miikkulainen2017evolving] combines the power of NEAT’s neural topology search with hierarchical evolution to efficiently discover architectures within large search spaces. Networks evolved using CoDeepNEAT have achieved good results in image classification and image captioning domains, outperforming popular hand-designed architectures. Consequently, CoDeepNEAT is one of the methods used in this paper for optimizing the topologies of the deep multitask networks. This paper extends CoDeepNEAT to coevolve modules and blueprints for deep multitask networks and also combines it with a novel and powerful EA for evolving task-specific architectures with shared modules.
CoDeepNEAT begins by initializing two populations, one of modules and one of blueprints, with minimal complexity. The blueprints and modules each contain at least one species and are evolved/complexified separately with a modified version of NEAT. An individual in the blueprint population is a directed acyclic graph (DAG) where each node contains a pointer to a particular module species. An individual in the module population is a DAG where each node represents a particular DNN layer and its corresponding hyperparameters (number of neurons, activation function, etc). As shown in Figure \[fg:assembling\], the modules are inserted into the blueprints to create a temporary population of assembled networks. Each individual in this population is then evaluated by training it on a supervised learning task, and assigning its performance as fitness. The fitnesses of the individuals (networks) are attributed back to blueprints and modules as the average fitness of all the assembled networks containing that blueprint or module. One of the advantages of CoDeepNEAT is that it is capable of discovering modular, repetitive structures seen in state of the art networks such as Googlenet and Resnet [@szegedy2015going; @szegedy:cvpr16; @he:arxiv16].
![The relationships of the five methods tested in this paper. The soft ordering method [@Meyerson:2018] is used as the starting point, extending it with CoDeepNEAT on the left and task-specific routing on the right. The CMTR on bottom right combines these two main ideas and performs the best.[]{data-label="fg:overview"}](figures/overview.pdf "fig:"){width="\linewidth"}\
**Given** fixed blueprint\
**Initialize** module population\
**Each** generation:\
**Assemble** MTL networks with modules\
**Randomly initialize all weights**\
**Train** each MTL network with backprop\
**Assign fitnesses** to modules\
**Update** module populations
**Initialize** blueprint/module populations\
**Each** generation:\
**Assemble** MTL networks with\
blueprints/modules\
**Randomly initialize all weights**\
**Train** each MTL network with backprop\
**Assign fitnesses** to modules and blueprints\
**Update** blueprint/module populations
**Given** set of modules\
**Initialize** topology population for each task\
**Randomly initialize all weights**\
**Each** meta-iteration:\
**Assemble** networks\
Jointly **train** all networks with backprop\
**Assign fitnesses** to topologies\
**Update** topology populations
**Initialize** module population\
**Each** generation:\
**Assemble** sets of modules\
**Train** sets of modules with CTR\
**Assign fitnesses** to modules\
**Update** module populations
Algorithms for Deep MTL Evolution {#sec:algorithms}
=================================
Figure \[fg:overview\] provides an overview of the methods tested in this paper in multitask learning. The foundation is (1) the original soft ordering, which uses a fixed architecture for the modules and a fixed routing (i.e. network topology) that is shared among all tasks. This architecture is then extended in two ways with CoDeepNEAT: (2) by coevolving the module architectures (CM), and (3) by coevolving both the module architectures and a single shared routing for all tasks using (CMSR). This paper also introduces a novel approach (4) that keeps the module architecture fixed, but evolves a separate routing for each task during training (CTR). Finally, approaches (2) and (4) are combined into (5), where both modules and task routing are coevolved (CMTR). Figure \[fig:algs\] gives high-level algorithmic descriptions of these methods, which are described in detail below.
Coevolution of Modules {#subsec:cm}
----------------------
In Coevolution of Modules (CM), CoDeepNEAT is used to search for promising module architectures, which then are inserted into appropriate positions to create an *enhanced soft ordering* network. The evolutionary process works as follows:
1. \[CM:zero\] CoDeepNEAT initializes a population of modules $MP$. The blueprints are not used.
2. Modules are randomly chosen from each species in $MP$, grouped into sets $M$ and are assembled into enhanced soft ordering networks.
3. Each assembled network is trained/evaluated on some task and its performance is returned as fitness.
4. Fitness is attributed to the modules, and NEAT evolutionary operators are applied to evolve the modules.
5. The proess is repeated from step \[CM:zero\] until CoDeepNEAT terminates, i.e. no further progress is observed for a given number of generations.
Unlike in soft ordering [@Meyerson:2018], the number of modules and the depth of the network are not fixed but are evolved as global hyperparameters by CoDeepNEAT (however the layout is still a grid-like structure). Since the routing layout is fixed, the blueprint population of CoDeepNEAT, which determines how the modules are connected, is not used. Thus one key operation in the original CoDeepNEAT, i.e. inserting modules into each node of the blueprint DAG, is skipped; only the module population is evolved.
To assemble a network for fitness evaluation, an individual is randomly chosen from each species in the module population to form an ordered set of distinct modules $M$. The hyperparameters evolved in each of the module’s layers include the activation function, kernel size, number of filters, L2 regularization strength and output dropout rate. In addition, CoDeepNEAT also coevolves global hyperparameters that are relevant to the entire assembled network as a whole; these include learning rate, the number of filters of the final layer of each module, and the weight initialization method. Evolvable hyperparameters in each node include the activation function, kernel size, number of filters, L2 regularization strength and output dropout rate. The modules are then transformed into actual neural networks by replacing each node in the DAG with the corresponding layer. To ensure compatibility between the inputs and outputs of each module, a linear $1 \times 1$ convolutional layer (number of filters determined by a global hyperparameter), followed by a max-pooling layer (provided that the feature map before pooling is at least $4 \times 4$) is included as the last layer in each module.
The modules are then inserted into the soft ordering network. The architecture of the network is interpreted as a grid of $K \times D$ slots, where $d$ indicates the depth of the network and the slots with the same $k$ value have the same module topology. For each available slot $T_{kd}$, the corresponding module $M_k$ is inserted. If $k > |M|$, then $M_{k \bmod |M|}$ is inserted instead.
Finally, each module in a particular slot has the potential to share its weights with modules that have the same architecture and are located in other slots of the blueprint. Flag $F_k$ in each module indicates whether or not the module’s weights are shared. This flag is evolved as part of the module genotype in CoDeepNEAT. Also, there is also global flag $F_d$ for each depth of the soft ordering network. If the $M_k$ is placed in $T_{kd}$ and both $F_k$ and $F_d$ are turned on, then the module is able to share its weights with any other $M_k$ whose slot have both flags turned on as well. Such an arrangement allows each slot to have sharing enabled and disabled independently.
The assembled network is attached to separate encoders and decoders for each task and trained jointly using a gradient-based optimizer. Average performance over all tasks is returned as fitness back to CoDeepNEAT. That fitness is assigned to each of the modules in the assembled network. If a module is used in multiple assembled networks, their fitnesses are averaged into module fitness. After evaluation is complete, standard NEAT mutation, crossover, and speciation operators are applied to create the next generation of the module population [@stanley:ec02].
Coevolution of Modules/Shared Routing {#subsec:cmsr}
-------------------------------------
Coevolution of Modules and Shared Routing (CMSR) extends CM to include blueprint evolution. Thus, the routing between various modules no longer follows the fixed grid-like structure, but instead an arbitrary DAG. Each node in the blueprint genotype points to a particular module species. During assembly, the blueprints are converted into deep multitask networks as follows:
1. \[CMSR:zero\] For each blueprint in the population, an individual module is randomly chosen from each species.
2. Each node in the blueprint is then replaced by the module from the appropriate species.
3. If a module has multiple inputs from previous nodes in the blueprint, the inputs are soft merged first [@Meyerson:2018].
4. The process is repeated from step \[CMSR:zero\] until reaching a target number of assembled networks.
As in CM, each node in the blueprint has a flag $F_i$ that indicates whether node $N_i$ should be shared or not. If two nodes are replaced by the same module and if both nodes have the sharing flag turned on, then the two modules will share weights. Such an arrangement allows each node to evolve independently whether to share weights or not. The training procedures for both CM and CMSR are otherwise identical. After fitness evaluation, the fitness is assigned to both blueprints and modules in the same manner as with CM. To accelerate evolution, the blueprint population is not initialized from minimally connected networks like the modules, but from randomly mutated networks that on average have five nodes.
Coevolution of Task Routing {#subsec:ctr}
---------------------------
This section introduces Coevolution of Task Routing (CTR), a multitask architecture search approach that takes advantage of the dynamics of soft ordering by evolving task-specific topologies instead of a single blueprint.
### Overview
Like in soft ordering, in CTR there are $K$ modules whose weights are shared everywhere they are used across all tasks. Like in blueprint evolution, CTR searches for the best ways to assemble modules into complete networks. However, unlike in blueprint evolution, CTR searches for a distinct module routing scheme *for each task*, and trains a *single set of modules* throughout evolution. Having a distinct routing scheme for each task makes sense if the shared modules are seen as a set of building blocks that are assembled to meet the differing demands of different problems. Training a single set of modules throughout evolution then makes sense as well: As modules are trained in different locations for different purposes during evolution, their functionality should become increasingly general, and it should thus become easier for them to adapt to the needs of a new location. Such training is efficient since the core structure of the network need not be retrained from scratch at every generation. In other words, CTR incurs no additional iterations of backpropagation over training a single fixed-topology multitask model. Because of this feature, CTR is related to PathNet [@Fernando:2017], which evolves pathways through modules as those modules are being trained. However, unlike in PathNet, in CTR distinct routing schemes are coevolved across tasks, modules can be applied in any location, and module usage is adapted via soft merges.
CTR operates a variant of a $(1+1)$ evolutionary strategy ($(1+1)$-ES) for each task. Separate ES for each task is possible because an evaluation of a multitask network yields a performance metric *for each task*. The $(1+1)$-ES is chosen because it is efficient and sufficiently powerful in experiments, though it can potentially be replaced by any population-based method. To make it clear that a single set of modules is trained during evolution, and to disambiguate from the terminology of CoDeepNEAT, for CTR the term *meta-iteration* is used in place of *generation*.
### Algorithm Description
Each individual constitutes a module routing scheme for a particular task. At any point in evolution, the $i$th individual for the $t$th task is represented by a tuple $(\mathcal{E}_{ti}, G_{ti}, \mathcal{D}_{ti})$, where $\mathcal{E}_{ti}$ is an encoder, $G_{ti}$ is a DAG, which specifies the module routing scheme, and $\mathcal{D}_{ti}$ is a decoder. The complete model for an individual is then given by $$\bm{y}_t = \big(\mathcal{D}_{ti} \circ \mathcal{R}\big(G_{ti}, \big\{\mathcal{M}_k\big\}_{k=1}^{K}\big) \circ \mathcal{E}_{ti}\big)(\bm{x}_t) \, ,$$ where $\mathcal{R}$ indicates the application of the shared modules $\mathcal{M}_k$ based on the DAG $G_{ti}$. Note that $\circ$ denotes function composition and $\mathcal{E}_{ti}$, and $\mathcal{D}_{ti}$ can be any neural network functions that are compatible with the set of shared modules. In the experiments in this paper, each $\mathcal{E}_{ti}$ is an identity transformation layer, and each $\mathcal{D}_{ti}$ is a fully connected classification layer.
$G_{ti}$ is a DAG, whose single source node represents the input layer for that task, and whose single sink node represents the output layer, e.g., a classification layer. All other nodes either point to a module $\mathcal{M}_k$ to be applied at that location, or a parameterless adapter layer that ensures adjacent modules are technically compatible. In the experiments in this paper, all adapters are $2\times2$ max-pooling layers. Whenever a node of $G_{ti}$ has multiple incoming edges, their contents are combined in a learned soft merge (Eq. \[eq:soft\_merge\]). The algorithm begins by initializing the shared modules $\{\mathcal{M}_k\}_{k=1}^{K}$ with random weights. Then, each champion $(\mathcal{E}_{t1}, G_{t1}, \mathcal{D}_{t1})$ is initialized, with $\mathcal{E}_{t1}$ and $\mathcal{D}_{t1}$ initialized with random weights, and $G_{t1}$ according to some graph initialization policy. For example, the initialization of $G_{t1}$ can be minimal or random. In the experiments in this paper, $G_{t1}$ is initialized to reflect the classical deep multitask learning approach, i.e., $$\mathcal{E}_{t1} \rightarrow \mathcal{M}_1 \rightarrow \mathcal{M}_2 \rightarrow \ldots \rightarrow \mathcal{M}_K \rightarrow \mathcal{D}_{t1} \, ,$$ with adapters added as needed.
At the start of each meta-iteration, a challenger $(\mathcal{E}_{t2}, G_{t2}, \mathcal{D}_{t2})$ is generated by mutating the $t$th champion as follows (the insertion of adapters is omitted for clarity):
1. The challenger starts as a copy of the champion, including learned weights, i.e., $(\mathcal{E}_{t2}, G_{t2}, \mathcal{D}_{t2}) \coloneqq (\mathcal{E}_{t1}, G_{t1}, \mathcal{D}_{t1})$.
2. A pair of nodes $(u, v)$ is randomly selected from $G_{t2}$ such that $v$ is an ancestor of $u$.
3. A module $\mathcal{M}_k$ is randomly selected from $\{\mathcal{M}_k\}_{k=1}^{K}$.
4. A new node $w$ is added to $G_{t2}$ with $\mathcal{M}_k$ as its function.
5. New edges $(u, w)$ and $(w, v)$ are added to $G_{t2}$.
6. The scalar weight of $(w, v)$ is set such that its value after the softmax is some $\alpha \in (0, 1)$. To initially preserve champion behavior, $\alpha$ is set to be small. I.e., if $s_1, \ldots, s_m$ are the scales of the existing inbound edges to $v$, $s_{m+1}$ is the initial scale of the new edge, and $s_{\max} = \max(s_1, \ldots, s_m)$ then $$s_{m+1} = \ln\big(\frac{\alpha}{1-\alpha}\sum_{j=1..m} e^{s_j - s_{\max}}\big) + s_{\max} \, .$$
After challengers are generated, all champions and challengers are trained jointly for $M$ iterations with a gradient-based optimizer. Note that the scales of $G_{t1}$ and $G_{t2}$ diverge during training, as do the weights of $\mathcal{D}_{t1}$ and $\mathcal{D}_{t2}$. After training, all champions and challengers are evaluated on a validation set that is disjoint from the training data. The fitness for each individual is its performance for its task on the validation set. In this paper, accuracy is the performance metric. If the challenger has higher fitness than the champion, then the champion is replaced, i.e.,$(\mathcal{E}_{t1}, G_{t1}, \mathcal{D}_{t1}) \coloneqq (\mathcal{E}_{t2}, G_{t2}, \mathcal{D}_{t2})$. After selection, if the average accuracy across all champions is the best achieved so far, the entire system is checkpointed, including the states of the modules. After evolution, the champions and modules from the last checkpoint constitute the final trained model, and are evaluated on a held out test set.
### An Ecological Perspective
More than most evolutionary methods, this algorithm reflects an artificial ecology. The shared modules can be viewed as a shared finite set of environmental resources that is constantly exploited and altered by the actions of different tasks, which can correspond to different species in an environment. Within each task, individuals compete and cooperate to develop mutualistic relationships with the other tasks via their interaction with this shared environment. A visualization of CTR under this perspective is shown in Figure \[fig:ctr\_diagram\].
![\[fig:ctr\_diagram\] This figure shows an instance of CTR with three tasks and four modules that are shared across all tasks. Each individual assembles the modules in different ways. Through gradient-based training, individuals exploit the shared resources to compete within a task, and over time must develop mutualistic relationships with other tasks via their use of the shared modules.](figures/CTR_diagram-crop.pdf "fig:"){width="\linewidth"}\
Importantly, even if a challenger does not outperform its champion, its developmental (learning) process still affects the shared resources. This perspective suggests a more optimistic view of evolution, in which individuals can have substantial positive effects on the future of the ecosystem even without reproducing.
Coevolution of Modules and Task Routing {#subsec:cmtr}
---------------------------------------
Both CM and CTR improve upon the performance of the original soft ordering baseline. Interestingly, these improvements are largely orthogonal, and they can be combined to form an even more powerful algorithm called Coevolution of Modules and Task Routing (CMTR). Since evolution in CTR occurs during training and is highly computational efficient, it is feasible to use CoDeepNEAT as an outer evolutionary loop to evolve modules. To evaluate and assign fitness to the modules, they are passed on to CTR (the inner evolutionary loop) for evolving and assembling the task specific routings. The performance of the final task-specific routings is returned to CoDeepNEAT and attributed to the modules in the same way as in CM: Each module is assigned the mean of the fitnesses of all the CTR runs that made use of that module. Another way to characterize CMTR is that it overcomes the weaknesses in both CM and CTR: CM’s inability to create a customized routing for each task and CTR’s inability to search for better module architectures.
CMTR’s evolutionary loop works as follows:
1. \[CMTR:zero\] CoDeepNEAT initializes a population of modules $MP$. The blueprints are not used.
2. Modules are randomly chosen from each species in $MP$ and grouped together into sets of modules $M_k$.
3. Each set of modules $M_k$ is given to CTR, which assembles the modules by evolving task-specific routings. The performance of the evolved routings on a task is returned as fitness.
4. Fitness is attributed to the modules, and NEAT’s evolutionary operators applied to evolve the modules.
5. The process repeats from step \[CMTR:zero\] until CoDeepNEAT terminates, i.e. no improvement for a given number of generations.
One difference between CMTR and CM is that each module’s final convolutional layer has additional evolvable hyperparameters such as kernel size, activation function, and output dropout rate. Preliminary experiments suggested that the relatively complex routings in CMTR (when compared to CM and CMSR) require more complex final layers as well, thus evolving the complexity of the final layer is optimal. Like in CTR, the weights between modules are always shared in CMTR. If modules with completely new weights are added to the task routings, they have to be trained from scratch and may even hurt performance, whereas adding a module with already partially trained weights does not. In addition, as the routings evolved by CTR are much larger than those discovered by CM and CMSR, disabling or evolving weight sharing significantly bloats the total number of weight parameters and slows training significantly.
Experiments {#sec:experiments}
===========
This section details experiments comparing the five methods in the Omniglot MTL domain.
Omniglot Character Recognition
------------------------------
The Omniglot dataset consists of 50 alphabets of handwritten characters [@Lake:2015], each of which induces its own character recognition task. There are 20 instances of each character, each a $105 \times 105$ black and white image. Omniglot is a good fit for MTL, because there is clear intuition that knowledge of several alphabets will make learning another one easier. Omniglot has been used in an array of settings: generative modeling [@Lake:2015; @Rezende:2016], one-shot learning [@Koch:2015; @Lake:2015; @Shyam:2017], and deep MTL [@Bilen:2017; @Maclaurin:2015; @Meyerson:2018; @Rebuffi:2017; @Yang:2017]. Previous deep MTL approaches used random training/testing splits for evaluation [@Bilen:2017; @Meyerson:2018; @Yang:2017]. However, with model search (i.e. when the model architecture is learned as well), a validation set separate from the training and testing sets is needed. Therefore, in the experiments in this paper, a fixed training/validation/testing split of 50%/20%/30% is introduced for each task. Because training is slow and increases linearly with the number of tasks, a subset of 20 tasks out of the 50 possible is used in the current experiments. These tasks are trained in a fixed random order. Soft ordering is the current state-of-the-art method in this domain [@Meyerson:2018]. The experiments therefore use soft ordering as a starting point for designing further improvements.
Experimental Setup
------------------
For CoDeepNEAT fitness evaluations, all networks are trained using Adam [@Kingma:2014] for 3000 iterations over the 20 alphabets; for CTR, the network is trained for 120 meta-iterations (30,000 iterations). Each iteration is equivalent to one full forward and backward pass through the network with a single example image and label chosen randomly from each task. The fitness assigned to each network is the average validation accuracy across the 20 tasks after training.
For CM and CMSR, CoDeepNEAT is initialized with approximately 50 modules (in four species) and 20 blueprints (in one species). For CMTR, a smaller module population of around 25 (in two species) is found to be beneficial in reducing noise since each module is evaluated more often. During each generation, 100 networks are assembled from modules and/or blueprints for evaluation. The global and layer-specific evolvable hyperparameters are described in Section \[sec:algorithms\]. With CoDeepNEAT, the evaluation of assembled networks is distributed over 100 separate EC2 instances with a K80 GPU in AWS. The average time for training is usually around 1-2 hours depending on the network size. With CTR, because it is a $(1+1)$ evolutionary strategy with a small population size, it is sufficient to run the algorithm on a single GPU.
Because the fitness returned for each assembled network is noisy, to find the best assembled CoDeepNEAT network, the top 50 highest fitness networks from the entire history of the run are retrained for 30,000 iterations. For the CM and CMSR experiments, decaying the learning rate by a factor of 10 after 10 and 20 epochs of training gave a moderate boost to performance. Similar boost is not observed for CTR and CMTR and therefore learning rate is not decayed for them. To evaluate the performance of the best assembled network on the test set (which is not seen during evolution or training), the network is trained from scratch again for 30,000 iterations. For CTR and CMTR, this is equivalent to training for 120 meta-iterations. During training, a snapshot of the network is taken at the point of highest validation accuracy. This snapshot is then evaluated and the average test accuracy over all tasks returned.
Results
-------
![Comparison of fitness (validation accuracy after partial training for 3000 iterations) over generations of single runs of CM, CMSR, and CMTR. Solid lines show the fitness of best assembled network and dotted line show the mean fitness. All methods reach a similar fitness, but CMTR is the fastest and CMSR the slowest.[]{data-label="fg:evolution_plot"}](figures/evolution_plot.pdf "fig:"){width="\linewidth"}\
![Comparison of fitness over number of meta-iterations of training for CTR and CMTR. Evolution discovers modules that leverage the available training better, forming a synergy of the two processes.[]{data-label="fg:evolution_ctr_plot"}](figures/evolution_ctr_plot.pdf "fig:"){width="\linewidth"}\
![Comparison of fitness over generations of CM with disabling, enabling, and evolving module weight sharing. No sharing is better than forced sharing, but evolvable sharing outperforms them both, validating the approach.[]{data-label="fg:evolution_plot_sharing"}](figures/evolution_plot_sharing.pdf "fig:"){width="\linewidth"}\
**Algorithm** **Val Accuracy (%)** **Test Accuracy (%)**
------------------------------------ ---------------------- -----------------------
1\. Single Task [@Meyerson:2018] 63.59 (0.53) 60.81 (0.50)
2\. Soft Ordering [@Meyerson:2018] 67.67 (0.74) 66.59 (0.71)
3\. CM 80.38 (0.36) 81.33 (0.27)
4\. CMSR 83.69 (0.21) 83.82 (0.18)
5\. CTR 82.48 (0.21) 82.36 (0.19)
6\. CMTR **88.20** (1.02) **87.82** (1.02)
: Average validation and test accuracy over 20 tasks for each algorithm. CMTR performs the best as it combines both module and routing evolution. Pairwise *t*-tests show all differences are statistically significant with $p<0.05$.[]{data-label="fg:fitness"}
Figure \[fg:evolution\_plot\] demonstrates how the best and mean fitness improves for CM, CMSR, and CMTR in the CoDeepNEAT outer loop where module/blueprint coevolution occurs. All three algorithms converge roughly to the same final fitness value, which is around 78% validation accuracy. CMTR converges the fastest, followed by CM, and lastly CMSR. This result is expected since the search space of CMTR is the smallest (only the modules are evolved with CoDeepNEAT), larger for CM (evolution of modules and weight sharing), and largest for CMSR (evolution of modules, blueprints, and weight sharing). Although CM, CMSR, and CMTR converge to the same fitness in evolution, CMTR achieves better final performance because training occurs via CTR. Figure \[fg:evolution\_ctr\_plot\] compares how fitness (i.e. average validation accuracy) improves for CTR (using the default modules) and CMTR (using the best evolved modules discovered by CMTR) during training, averaged over 10 runs. Interestingly, while CTR improves faster in the first 10 meta-iterations, it is soon overtaken by CMTR, demonstrating how evolution discovers modules that leverage the available training better.
One open question is how much sharing of weights between modules affects the performance of the assembled network. Although disabling weight sharing is not optimal for CTR due to the complexity of the routing, both CM and CMSR may benefit since their routing topologies are much smaller (minimizing the effects of parameter bloat). Figure \[fg:evolution\_plot\_sharing\] compares the effect of enabling, disabling, and evolving weight sharing with CM. Interestingly, disabling weight sharing leads to better performance than enabling it, but evolving it is best. Thus, the design choice of evolving sharing in CM and CMSR is vindicated. An analysis of the architecture of the best assembled networks shows that weight sharing in particular locations such as near the output decoders is a good strategy.
Table \[fg:fitness\] shows the validation and test accuracy for the best evolved network produced by each method, averaged over 10 runs. The best-performing methods are highlighted in bold and standard error for the 10 runs is shown in parenthesis. In addition, performance of the baseline methods are shown, namely (1) a hand-designed single-task architecture, i.e. where each task is trained and evaluated separately, and (2) the soft ordering network architecture [@Meyerson:2018]. Indeed the methods improve upon the baseline according to increasing complexity: Evolving modules and evolving topologies is significantly better than the baselines, and evolving both is significantly better than either alone. CMTR, the combination of CoDeepNEAT and routing evolution, combines the advantages of both and performs the best.
The best networks have approximately three million parameters. Figure \[fg:vis\] visualizes one of the best performing modules from the CMTR experiment, and sample routing topologies evolved for the different alphabets. Because the CoDeepNEAT outer loop is based on two species, the four modules passed to the CTR inner loop consist of two different designs (but still separate weights). Thus, evolution has discovered that a combination of simple and complex modules is beneficial. Similarly, while the routing topologies for some alphabets are simple, others are very complex. Moreover, similar topologies emerge for similar alphabets (such as those that contain prominent horizontal lines, like Gurmukhi and Manipuri). Also, when evolution is run multiple times, similar topologies for the same alphabet result. Such useful diversity in modules and routing topologies, i.e. structures that complement each other and work well together, would be remarkably difficult to develop by hand. However, evolution discovers them consistently and effectively, demonstrating the power of the approach.
![Structures of the best modules and routing topologies discovered by evolution. The two species in CMTR evolve very different modules: one simple and one complex. The thick boxes represent convolutional, medium max pooling, and thin dropout layers, with hyperparameters listed on the left. The routing topologies represent a range from simple to complex; similar alphabets have similar topologies, and the structure is consistently found. Such useful diversity would be difficult to develop by hand, demonstrating the power of evolution in designing complex systems.[]{data-label="fg:vis"}](figures/2047_0.pdf){width="\linewidth"}
![Structures of the best modules and routing topologies discovered by evolution. The two species in CMTR evolve very different modules: one simple and one complex. The thick boxes represent convolutional, medium max pooling, and thin dropout layers, with hyperparameters listed on the left. The routing topologies represent a range from simple to complex; similar alphabets have similar topologies, and the structure is consistently found. Such useful diversity would be difficult to develop by hand, demonstrating the power of evolution in designing complex systems.[]{data-label="fg:vis"}](figures/2047_1.pdf){width="\linewidth"}
![Structures of the best modules and routing topologies discovered by evolution. The two species in CMTR evolve very different modules: one simple and one complex. The thick boxes represent convolutional, medium max pooling, and thin dropout layers, with hyperparameters listed on the left. The routing topologies represent a range from simple to complex; similar alphabets have similar topologies, and the structure is consistently found. Such useful diversity would be difficult to develop by hand, demonstrating the power of evolution in designing complex systems.[]{data-label="fg:vis"}](figures/angelic.pdf){width="\linewidth"}
![Structures of the best modules and routing topologies discovered by evolution. The two species in CMTR evolve very different modules: one simple and one complex. The thick boxes represent convolutional, medium max pooling, and thin dropout layers, with hyperparameters listed on the left. The routing topologies represent a range from simple to complex; similar alphabets have similar topologies, and the structure is consistently found. Such useful diversity would be difficult to develop by hand, demonstrating the power of evolution in designing complex systems.[]{data-label="fg:vis"}](figures/gurmukhi.pdf "fig:"){width="0.45\linewidth"} ![Structures of the best modules and routing topologies discovered by evolution. The two species in CMTR evolve very different modules: one simple and one complex. The thick boxes represent convolutional, medium max pooling, and thin dropout layers, with hyperparameters listed on the left. The routing topologies represent a range from simple to complex; similar alphabets have similar topologies, and the structure is consistently found. Such useful diversity would be difficult to develop by hand, demonstrating the power of evolution in designing complex systems.[]{data-label="fg:vis"}](figures/manipuri.pdf "fig:"){width="0.45\linewidth"}
Discussion and Future Work {#sec:discussion}
==========================
The experiments show that MTL can improve performance significantly across tasks, and that the architecture used for it matters a lot. Multiple ways of optimizing the architecture are proposed in this paper and the results lead to several insights.
First, modules used in the architecture can be optimized and the do end up different in a systematic way. Unlike in the original soft ordering architecture, evolution in CM, CMSR, and CMTR results in discovery of a wide variety of simple and complex modules, and they are often repeated in the architecture. Evolution thus discovers a useful set of building blocks that are diverse in structure. Second, the routing of the modules matter as well. In CMSR, the shared but evolvable routing allows much more flexibility in how the modules can be reused and extends the principals that makes soft ordering useful. The power of CTR and CMTR is from evolving different topologies for different tasks, and tie the tasks together by sharing the modules in them. In addition, sharing components (including weight values) in CMTR is crucial to its performance. If indeed the power from multitasking comes from integrating requirements of multiple tasks, this integration will happen in the embeddings that the modules form, so it makes sense that sharing plays a central role. Third, compared to the CTR and CMTR, CM and CMSR have evolved away from sharing of module weights, despite the fact that module architectures are often reused in the network. This result makes sense as well: because the topology is shared in this approach, the differentiation between tasks comes from differentiated modules. Such an approach is an opposite way to solve the problem. Even though it is an effective approach as well, it is not quite as powerful as differentiated topologies and shared modules.
There are several directions for future work. The proposed algorithms can be extended to many applications that lend themselves to the multitask approach. For instance, it will be interesting to see how it can be used to find synergies in different tasks in vision, and in language. Further, as has been shown in related work, the tasks do not even have to be closely related to gain the benefit from MTL. For instance, object recognition can be paired with caption generation. It is possible that the need to express the contents of an image in words will help object recognition, and vice versa. Discovering ways to tie such multimodal tasks together should be a good opportunity for evolutionary optimization, and constitutes a most interesting direction for future work.
Conclusion {#sec:conclusion}
==========
This paper presents a family of EAs for optimizing the architectures of deep multitask networks. They extend upon previous work which has shown that carefully designed routing and sharing of modules can significantly help multitask learning. The power of the proposed algorithms is shown by achieving state of the art performance on a widely used multitask learning dataset and benchmark.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It has been shown that for every prime number $p$, the pseudovariety $\mathsf{G}_p$ of all finite $p$-groups is tame with respect to an implicit signature containing the canonical implicit signature. In this paper we generalize this result and we show that the pseudovariety of all finite nilpotent groups is tame but it is not completely tame.'
address: 'Centro de Matemática e Departamento de Matemática, Faculdade de Ciências, Universidade do Porto, Porto, Portugal'
author:
- Khadijeh Alibabaei
title: The pseudovariety of all nilpotent groups is tame
---
**Keywords.** [relatively free profinite semigroup, pseudovariety of semigroups, system of equations, implicit signature, completely tame, completely reducible, rational constraint, $\sigma$-full, weakly reducible.]{}
Introduction
============
By a *pseudovariety* we mean a class of semigroups which is closed under taking subsemigroups, finite direct products, and homomorphic images. A pseudovariety is said to be *decidable* if there is an algorithm to test membership of a finite semigroup; otherwise, the pseudovariety is said to be *undecidable*. Eilenberg [@Eilenberg:1974] established a correspondence between varieties of rational languages and pseudovarieties of finite semigroups which translates problems in language theory into the decidability of pseudovarieties of semigroups. In general the decidability of pseudovarieties is not preserved by many operations on pseudovarieties such as semidirect product, join and Mal’cev product (). Almeida and Steinberg introduced a refined version of decidability called *tameness* . The tameness property requires the *reducibility* property which is a generalization of the notion of inevitability that Ash introduced to prove the type II conjecture of Rhodes [@Ash:1991].
There are various results using tameness of pseudovarieties to establish the decidability of pseudovarieties obtained by application of the operations of semidirect product, Mal’cev product and join .
Also there are connections between tameness and geometry and model theory . So, it is worth finding more examples of tame pseudovarieties.
It has been established that for every prime numbers $p$, the pseudovariety $\mathsf{G}_p$ of all finite $p$-groups is tame [@Jorge:2002]. Using this result, in this paper we show that the pseudovariety $\mathsf{G}_{nil}$ of all finite nilpotent groups is tame with respect to an enlarged implicit signature $\sigma$. Since the free $\sigma$-subalgebra generated by a finite alphabet $A$ is not any more the free group, in section 3, we prove the word problem is decidable in this free $\sigma$-subalgebra, meaning that there is an algorithm to decide whether two elements of this $\sigma$-algebra represent the same element.
In the last section, we show that the pseudovariety $\mathsf{G}_{nil}$ is $\sigma$-reducible if and only of for all prime number $p$, the pseudovariety $\mathsf{G}_p$ is $\sigma$-reducible. This theorem yields as a corollary that the pseudovariety $\mathsf{G}_{nil}$ is tame with respect to the systems of equations associated to finite directed graphs but is not completely tame.
Preliminaries
=============
A *topological semigroup* is a semigroup $S$ endowed with a topology such that the basic semigroup multiplication $S \times S \rightarrow S$ is continuous. We say that a topological semigroup $S$ is $A$-*generated* if there is a mapping $\varphi : A \rightarrow S$ such that $\varphi(A)$ generates a dense subsemigroup of $S$.
A *pseudovariety of semigroups* is a class of finite semigroups closed under taking subsemigroups, homomorphic images, and direct products. Given a pseudovariety $\mathsf{V}$ of semigroups, by a pro-$\mathsf{V}$ semigroup $S$ we mean a compact, zero-dimensional semigroup which is residually in $\mathsf{V}$, that is for every two distinct points $s, t \in S$, there exists a continuous homomorphism $\varphi : S \rightarrow T$ into some member $T\in \mathsf{V}$ such that $\varphi(s) \neq \varphi(t)$.
For a finite set $A$ in the variety generated by $\mathsf{V}$, we denote by $\overline{\Omega}_A\mathsf{V}$ the free pro-$\mathsf{V}$ semigroup. The free pro-$\mathsf{V}$ semigroup has the universal property in variety of pro-$\mathsf{V}$ semigroups in the sense that for every mapping $\varphi : A \rightarrow S$ into a pro-$\mathsf{V}$ semigroup $S$, there exists a unique continuous homomorphism $\hat{\varphi} : \overline{\Omega}_A\mathsf{V}\rightarrow S$ such that the following diagram commutes: $$\xymatrix{
A
\ar[r]^(.4)\iota
\ar[d]_\varphi
&
\overline{\Omega}_A\mathsf{V}
\ar[ld]^{\hat\varphi}
\\
S
}$$ For an $A$-generated pro-$\mathsf{V}$ semigroup $S$, we view $S^A$ both as a direct power of $S$ and as the set of all functions from the set $A$ to $S$. To each element $w\in \overline{\Omega}_A\mathsf{V}$, we may associate an $A$-ary operation $w_S:S^A\rightarrow S$: for every $\varphi\in S^A$, by the universal property of $\overline{\Omega}_A\mathsf{V}$, there is a unique extension $\hat\varphi:\overline{\Omega}_A\mathsf{V}\rightarrow S$. Define $w_S(\varphi)=\hat\varphi(w)$. It is easy to see that for every continuous homomorphism $f : S \rightarrow T$ between pro-$\mathsf{V}$ semigroups, the following diagram commutes: $$\xymatrix{
S^A
\ar[r]^{w_S}
\ar[d]_{f\circ-}
&
S
\ar[d]^{f}
\\
T^A
\ar[r]^{w_T}
&
T
}$$ Operations with that property are called $A$-*ary implicit operations*. The element $w\in \overline{\Omega}_A\mathsf{V}$ is completely determined by the implicit operation $(w_S)_{S\in \mathsf{V}}$ [@Jorge:2002]. Note that, the elements of $A$ correspond to the component projects.
We say that an implicit operation $w\in \overline{\Omega}_A\mathsf{V}$ is *computable* if there is an algorithm which given $S\in\mathsf{V}$ and $\varphi\in S^A$, output the value $w_S(\varphi)$.
An $A$-*ary implicit operator* on a pro-$\mathsf{V}$ semigroup $S$ is a transformation $f:S^A\rightarrow S^A$ of $S^A$ to itself whose components $f_i:S^A\rightarrow S$ determined an $A$-ary implicit operation. The set of all $A$-ary implicit operators on a pro-$\mathsf{V}$ semigroup $S$ is denoted by $\mathcal{O}_A(S)$. The set $\mathcal{O}_A(S)$ is a monoid under composition.
[@Jorge:2002 Proposition 2.2] There is a natural topology on $\mathcal{O}_A(S)$ such that the correspondence $$S\rightarrow \mathcal{O}_A(S)$$ defines a functor from the category of pro-$\mathsf{V}$ semigroups with onto continuous homomorphisms as morphisms into the category of profinite monoids.
For $n$-ary implicit operations $w_1,\ldots,w_n\in \overline{\Omega}_n\mathsf{V}$ and a pro-$\mathsf{V}$ semigroup $S$, denote by $( w_1,\ldots,w_n)$ the implicit operator $$\begin{aligned}
S^n&\rightarrow S^n\\
(s_1,\ldots,s_n)&\mapsto((w_1)_S(s_1,\ldots,s_n),\ldots,(w_n)_S(s_1,\ldots,s_n)).
\end{aligned}$$ Denote composition of operators by concatenation: $ ( v_1,\ldots,v_n)( w_1,\ldots,w_n)$ has component $i$ determined by the operation $v_i( w_1,\ldots,w_n)$.
Recall that, for an element $v$ of a finite semigroup $V$, $v^\omega$ denotes the unique idempotent power of $v$. This defines a unary implicit operation $x \mapsto x^\omega$ on finite semigroups (and similarly on finite monoids) which therefore has a natural interpretation on each profinite monoid of the form $\mathcal{O}_n(S)$. Note that $x^\omega$ is the limit of the sequence $\{x^{n!}\}_n$. We denote by $a_j\circ(w_1,\ldots,w_n)^\omega$, the $j$-th component of the $\omega$-power of the operator $(w_1,\ldots,w_n)$.
[@Jorge:2002 Corollary 2.5] Let $w_1,\ldots, w_n\in \overline{\Omega}_n\mathsf{V}$. Then each component of $( w_1,\ldots,w_n)^\omega$ is also a member of $\overline{\Omega}_n\mathsf{V}$. Moreover, if $w_i$ are computable operations, then so is each $a_j\circ( w_1,\ldots,w_n)^\omega$.
Let $\mathsf{S}$ be the pseudovariety of all finite semigroups. The elements of $(\overline{\Omega}_A\mathsf{S})^1$, over arbitrary finite alphabets $A$, are called *pseudowords*. A *pseudoidentity* is a formal equality of the form $u = v$ with $u, v \in \overline{\Omega}_A\mathsf{S}$ for some finite alphabet $A$. For a pseudovariety $\mathsf{V}$ of semigroups, we denote by $\psi_\mathsf{V}$ the unique continuous homomorphism $\overline{\Omega}_A\mathsf{S}\rightarrow \overline{\Omega}_A\mathsf{V}$ which restricts to the identity on $A$. We say that $\mathsf{V}$ satisfies the pseudoidentity $u = v$ with $u, v \in \overline{\Omega}_A\mathsf{S}$ if $\psi_\mathsf{V}(u)=\psi_\mathsf{V}(v)$.
By an *implicit signature* we mean a set of pseudowords including multiplication. An important example is given by the canonical signature $\kappa$ consisting of the multiplication and the unary operation $x \mapsto x^{\omega-1}$ which, to an element $s$ of a finite semigroup with $n$ elements, associates the inverse of $s^{1+n!}$ in the cyclic subgroup generated by this power.
Let $\sigma$ be an implicit signature. Under the natural interpretation of the elements of $\sigma$, every profinite semigroup may be viewed as a $\sigma$-algebra in the sense of universal algebra. The $\sigma$-subalgebra of $\overline{\Omega}_{A}\mathsf{V}$ generated by $A$ is denoted by ${\Omega}_{A}^{\sigma}\mathsf{V}$ and it is freely generated by $A$. We say that ${\Omega}_{A}^{\sigma}\mathsf{V}$ has *decidable word problem* if there is an algorithm to decide whether two pseudowords $u, v\in\overline{{\Omega}}_{A}\mathsf{S}$ with $\psi_{\mathsf{V}}(u),\psi_{\mathsf{V}}(v)\in {\Omega}_{A}^{\sigma}\mathsf{V}$ represent the same implicit operation in ${\Omega}_{A}^{\sigma}\mathsf{V}$.
For every subset $L$of $\overline{\Omega}_A\mathsf{S}$ and $\mathsf{V}$ be a pseudovariety, we denote by $Cl(L)$, $Cl_{\mathsf{V}}(L)$, and $Cl_{\sigma,\mathsf{V}}(L)$, the closure of $L$ in $\overline{\Omega}_{A}\mathsf{S}$, $\overline{\Omega}_{A}\mathsf{V}$, and ${\Omega}^{\sigma}_{A}\mathsf{V}$, respectively.
As it is mentioned in the introduction, the property tameness requires the property reducibility. To define reducibility we need a system of equations. Let $X$ and $P$ be disjoint finite sets, whose elements will be the *variables* and the *parameters* of the system, respectively. Consider the following system of equations: $$\label{eq32}
\begin{aligned}
u_i=v_i&&&&&&&& (i=1\ldots,m),
\end{aligned}$$ where $u_i$ and $v_i$ are pseudowords of $\overline{\Omega}_{X\cup P}\mathsf{S}$. We also fix a finite set $A$ and for every $x \in X$, we choose a rational subset $L_x\subseteq A^{*}$. For every parameter $p \in P$, we associate an element $w_p\in\overline{\Omega}_{A}\mathsf{S}$. A *solution* of the system modulo $\mathsf{V}$ satisfying the constraints is a function $\delta:{X\cup P}\rightarrow \overline{\Omega}_{A}\mathsf{S}$ satisfying the following conditions:
1. $\delta(x)\in Cl(L_x)$.
2. $\delta(p)= w_p$.
3. $\mathsf{V}$ satisfies the pseudoidentities $\delta(u_i)=\delta(v_i)$ ($i=1\ldots,m$).
\[theorem11\][@Jorge:1994 Theorem 5.6.1] Let $\mathsf{V}$ be a pseudovariety of finite semigroups. The following conditions are equivalent for a finite system $\Sigma$ of equations with rational constraints over the finite alphabet $A$:
- $\Sigma$ has a solution modulo every $A$-generated semigroup in $\mathsf{V}$
- $\Sigma$ has a solution modulo $\overline{\Omega}_A\mathsf{V}$.
Let $\sigma$ be an implicit signature. Consider a system of the form , with constraints $L_x\subseteq A^*$ and $w_p\in {\Omega}^{\sigma}_{A}\mathsf{S}$ ($x\in X$ and $p\in P$) where $u_i$ and $v_i$ are $\sigma$-terms in ${\Omega}^{\sigma}_{X\cup P}\mathsf{S}$. Assume that this system has a solution modulo $\mathsf{V}$. A pseudovariety $\mathsf{V}$ is said to be *$\sigma$-reducible* for this system if it has a solution $\delta:{X\cup P}\rightarrow {\Omega}_{A}^\sigma \mathsf{S}$ modulo $\mathsf{V}$. We say that $\mathsf{V}$ is *completely $\sigma$-reducible* if it is $\sigma$-reducible for every such system.
Let $\mathsf{V}$ be a pseudovariety. If $\mathsf{V}$ is $\sigma$-reducible respect to the systems of equations of $\sigma$-terms without parameters, then $\mathsf{V}$ is completely $\sigma$-reducible.
We say that a pseudovariety $\mathsf{V}$ is *$\sigma$-tame* with respect to a class $\mathfrak{C}$ of systems of equations if the following conditions hold:
- for every system of equations in $\mathfrak{C}$, the pseudovariety $\mathsf{V}$ is $\sigma$-reducible;
- the word problem is decidable in $\Omega_A^{\sigma}\mathsf{V}$;
- $\mathsf{V}$ is recursively enumerable, in the sense that there is some Turing machine which outputs successively representatives of all the isomorphism classes of members of $V$.
We say that $\mathsf{V}$ is *completely $\sigma$-tame* if $V$ is completely $\sigma$-reducible. Some important tameness results are as follows:
- The pseudovariety $\mathsf{G}$ of all finite groups is $\kappa$-tame with respect to systems of equations associated with finite directed graphs . It follows from results of Coulbois and Khélif that $\mathsf{G}$ is not completely $\kappa$-tame .
- For a prime number $p$, the pseudovariety $\mathsf{G}_p$ of all finite $p$-groups is tame with respect to the systems of equations associated with finite directed graphs, but not $\kappa$-tame [@Jorge:2002].
- The pseudovariety $\mathsf{Ab}$ of all finite abelian groups is completely $\kappa$-tame .
Consider a system of equations with constraints $L_x\subseteq A^*$ and $w_p\in {\Omega}^{\sigma}_{A}\mathsf{S}$ ($x\in X$ and $p\in P$) where $u_i$ and $v_i$ are $\sigma$-terms in ${\Omega}^{\sigma}_{X\cup P}\mathsf{S}$. A pseudovariety $\mathsf{V}$ is said to be *weakly $\sigma$-reducible* with respect to this system if in the case it has a solution modulo $\mathsf{V}$, then there is a solution $\delta:{X\cup P}\rightarrow \overline{\Omega}_{A} \mathsf{S}$ modulo $\mathsf{V}$ which satisfies the conditions $\psi_{\mathsf{V}}(\delta(x))\in {\Omega}_{A}^\sigma \mathsf{V}$ ($x\in X$). It is obvious that if a pseudovariety $\mathsf{V}$ is $\sigma$-reducible, then it is weakly $\sigma$-reducible but the converse is not true. For a prime number $p$, the pseudovariety $\mathsf{G}_p$ of all finite $p$ groups is weakly $\kappa$-reducible but it is not $\kappa$-reducible .
We say that a pseudovariety $\mathsf{V}$ is $\sigma$-*full* if for every rational language $L \subseteq A^*$, the set $\psi_\mathsf{V}(Cl_{\sigma}(L))$ is closed in ${\Omega}_{A}^{\sigma}\mathsf{V}$.
\[proposition3\] Every $\sigma$-full weakly $\sigma$-reducible pseudovariety is $\sigma$-reducible.
Let $x\in \overline{\Omega}_A\mathsf{S}$ and fix $n\in \mathbb{N}$. We denote by $x^{n^\omega}$, the pseudoword $$\lim_{k\to \infty} x^{n^{k!}}.$$
Let $w_1,\ldots,w_k$ be group words. We denote by $M(w_1,\ldots,w_k)$ the $k\times |A|$ matrix whose $(i,j)$-entry is the sum of the exponents in the occurrences of the letter $a_j\in A$ in the word $w_i$.
[@Jorge:2002] Let $p$ be a prime number and $\sigma_p$ be the set of all implicit signature obtained by adding to the canonical signature $\kappa$ all implicit operations of the form $$a_j\circ(w_1,\ldots, w_n)^\omega$$ with $j = 1,\ldots, n$, $w_i$ are $\kappa$-terms such that the subgroup $\left<w_1,\ldots,w_n\right>$ is $\mathsf{G}_p$-dense. Then $\mathsf{G}_{p}$ is $\sigma_p$-tame.
We use the implicit operations in the preceding theorem and we prove the following theorem:
Let $\sigma$ be the set of all implicit signature obtained by adding to the canonical signature $\kappa$ all implicit operations of the form $$\left(a_j\circ(w_1,\ldots, w_n)^{\omega}\right)^{m^\omega} \ \ \ (m\in \mathbb{N})$$ with $j = 1,\ldots, n$, $w_i$ are $\kappa$-terms such that $\det{M(w_1,\ldots,w_n)}\ne0$ and every prime number $p$ dividing $\det{M(w_1,\ldots,w_n)}$ also divides $m$. Then $\mathsf{G}_{nil}$ is $\sigma$-tame.
Note that by , for a prime number $p$ and $\kappa$-words $w_1,\ldots, w_n$, the subgroup $\left<w_1,\ldots,w_n\right>$ is $\mathsf{G}_p$-dense if and only if $\det{M(w_1,\ldots,w_n)}\nequiv0 \pmod{p}$ .
The pseudovariety TEXT is TEX-tame
==================================
The word problem is decidable in TEXT
-------------------------------------
We use the following lemmas to reduced the word problem in $\Omega_A^{\sigma}\mathsf{G}_{nil}$ to the the word problem in the free group generated by $A$.
\[lemma53\] Let $u,v\in \overline{\Omega}_A\mathsf{S}$. Then the pseudoidentity $u=v$ is valid in $\mathsf{G}_{nil}$ if and only if, for every prime number $p$, the pseudoidentity $u=v$ is valid in $\mathsf{G}_p$.
Since every $p$-group is a nilpotent group, if $\mathsf{G}_{nil}$ satisfies the pseudoidentity $u=v$, then for every prime number $p$, $\mathsf{G}_p$ satisfies the pseudoidentity $u=v$.
The converse follows from fact that every finite nilpotent group is isomorphic to the direct product of its $p$-Sylow subgroups.
We denote by $\mathds{P}$, the set of all prime numbers.
\[lemma54\] Fix $p\in \mathds{P}$. The pseudoidentity $x^{n^\omega}=x$ is valid in $\mathsf{G}_p$ if $n$ is not divisible by $p$ and the pseudoidentity $x^{n^\omega}=1$ is valid in $\mathsf{G}_p$ otherwise.
The result follows from the elementary Euler congruence theorem.
\[th1\] For every pseudoword $u\in \Omega_A^{\sigma}\mathsf{S}$, there is a computable cofinite subset $S(u)$ of $\mathds{P}$ such that the following properties hold:
- there is a computable $\kappa$-word $w_0$ such that for every $p\in S(u)$, the pseudoidentity $u=w_0$ is valid in $\mathsf{G}_p$;
- for every $p_i\in\{p_1,\ldots,p_r\}=\mathds{P}\setminus S(u)$, there is a computable $\kappa$-word $w_i$ such that the pseudoidentity $u=w_i$ is valid in $\mathsf{G}_{p_i}$ $(1\leq i\leq r)$.
In particular, for every prime number $p$, we have $\Omega^{\sigma}_A\mathsf{G}_{p}=\Omega^{\kappa}_A\mathsf{G}_{p}$.
Since every $u\in \Omega_A^{\sigma}\mathsf{S}$ is constructed from the letters in $A$ using a finite number of times the operations in $\sigma$ and the intersection of a finite number of cofinite sets is cofinite, it is enough to show that the statement of the theorem holds for every pseudoword $u\in \sigma\setminus \kappa$.
Consider the following pseudoword in $\sigma\setminus \kappa$ $$u= \left(a_j\circ(w_1,\ldots, w_n)^{\omega}\right)^{m^\omega} \ \ \ (m\in \mathbb{N})$$ where $j = 1,\ldots, n$, $w_i$ are $\kappa$-terms such that $\det{M(w_1,\ldots,w_n)}\ne0$ and every prime number $p$ dividing $\det{M(w_1,\ldots,w_n)}$ also divides $m$. Let $$S(u)=\mathds{P}\setminus\{p\mid p \ \text{divides} \ m\}.$$ If $p$ lies in $S(u)$, then by , the subgroup $\left<w_1,\ldots,w_n\right>$ is $\mathsf{G}_p$-dense in $\overline{\Omega}_A\mathsf{G}_p$. Hence, by [@Jorge:2002 Lemma 6.5], the following pseudoidentity holds in $\mathsf{G}_p$ $$a_j\circ(w_1,\ldots, w_n)^{\omega}=a_j.$$ Consider $w_0=a_j$.
Otherwise, $p$ divides $m$ and therefore by Lemma \[lemma54\], the pseudoidentity $$\left(a_j\circ(w_1,\ldots, w_n)^{\omega}\right)^{m^\omega}=1$$ holds in $\mathsf{G}_p$.
The word problem is decidable in $\Omega^{\sigma}_A\mathsf{G}_{nil}$.
Let $u,v\in \Omega^{\sigma}_A\mathsf{S}$. Then by Lemma \[lemma53\], the pseudoidentity $u=v$ holds in $\mathsf{G}_{nil}$ if and only if for every prime number $p$, the pseudoidentity $u=v$ holds in $\mathsf{G}_{p}$.
Since $\mathsf{G}_{p}$ is a pseudovariety of finite groups, the pseudoidentity $u=v$ is valid in $\mathsf{G}_{p}$ if and only if the pseudoidentity $uv^{\omega-1}=1$ is valid in $\mathsf{G}_{p}$. Now the result follows from the preceding lemma and the fact that the word problem is decidable in the free group.
The pseudovariety TEXT is TEXT-reducible
----------------------------------------
We denote by $FG(A)$, the free group over $A$ and for a finitely generated (f.g.) subgroup $H$ of a free group, denote by $\mathds{P}(H)$, the set of all prime numbers $p$ such that $H$ is $\mathsf{G}_p$-closed.
Let $H$ be a f.g. subgroup of $FG(A)$. The set $\mathds{P}(H)$ is a finite or a cofinite subset of $\mathds{P}$, and it is effectively computable.
\[lemma51\] Let $H$ be a f.g. subgroup of $FG(A)$. Then there is a cofinite subset $S(H)$ of $\mathds{P}$ and a f.g. subgroup $K$ of $FG(A)$ such that for every $p\in S(H)$, $Cl_{\mathsf{G}_p,\kappa}(H)=K$.
Let $K_1,\ldots,K_m$ be the set of all overgroups of the subgroup $H$ (i.e., the automaton of $K_i$ is a quotient of the automaton of $H$). For every prime number $p$, there is $i$ ($1\leq i\leq m$) such that $Cl_{\mathsf{G}_p,\kappa}(H)=K_i$ . Since the number of overgroups of $H$ is a finite set, there is $K_{j}$ and an infinite subset $S$ of $\mathds{P}$ such that for every $p\in S$, we have $Cl_{\mathsf{G}_p,\kappa}(H)=K_{j}$ $(1\leq j\leq m)$. If $S$ is cofinite, then we are done. For every $p\in S$, we have $$K_{j}\subseteq Cl_{\mathsf{G}_p,\kappa}(K_{j})\subseteq Cl_{\mathsf{G}_p,\kappa}(Cl_{\mathsf{G}_p,\kappa}(H))= Cl_{\mathsf{G}_p,\kappa}(H)=K_{j}$$ and hence, $S\subseteq \mathds{P}(K_{j})$. Therefore, by the preceding proposition, $\mathds{P}(K_{j})$ is a cofinite subset of $\mathds{P}$.
Suppose that there is an overgroup $L$ of $H$ properly contained in $K_j$ such that $\mathds{P}(L)$ is a cofinite subset of $\mathds{P}$. Then for every $p\in \mathds{P}(L)$, we have $$Cl_{\mathsf{G}_p,\kappa}(H)\subseteq Cl_{\mathsf{G}_p,\kappa}(L)=L\varsubsetneq K_j.$$ Hence, the set of all prime numbers $q$ such that $Cl_{\mathsf{G}_q,\kappa}(H)=K_j$ is contained in the set $\mathds{P}\setminus \mathds{P}(L)$. Since we assume that $\mathds{P}(L)$ is a cofinite subset of $\mathds{P}$, the set $\mathds{P}\setminus \mathds{P}(L)$ is a finite set which contradicts the choice of $K_j$.
So, for every overgroup $L$ of $H$ properly contained in $K_{j}$, $\mathds{P}(L)$ is a finite set. Let $K=K_{j}$ and $$S(H)=\mathds{P}(K)\setminus\bigcup_{\mathclap{\substack{L\ \text{overgroup of}\ H \\
L\subset K}}} \mathds{P}(L).$$ As the number of over groups of $H$ is finite, $S(H)$ is a cofinite subset of $\mathds{P}$. For every $p\in S(H)$, we have $$Cl_{\mathsf{G}_p,\kappa}(H)\subseteq Cl_{\mathsf{G}_p,\kappa}(K)=K,$$ hence, $Cl_{\mathsf{G}_p,\kappa}(H)$ is a $\mathsf{G}_p$-closed overgroup of $H$ contained in $K$, by the choice of $p$, it follows that $Cl_{\mathsf{G}_p,\kappa}(H)=K$.
The following two propositions are the main tools to show that the pseudovariety $\mathsf{G}_{nil}$ is $\sigma$-reducible.
\[lemma52\] Let $H_1,\ldots, H_t$ be f.g. subgroups of the free group and fix the cofinite sets $S(H_i)$ and the subgroups $K_i$ as in the preceding lemma. Let $w\in K_1\ldots K_t$. Then there are $u_i\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_i)$ and cofinite subsets $S_1(H_i)$ of $\mathds{P}$ contained in $S(H_i)$ such that, if $p\in \displaystyle\bigcap_{i=1}^tS_1(H_i)$, then the pseudoidentity $w=u_1\ldots u_t$ is valid in $\mathsf{G}_p$ and the pseudoidentity $u_i=1$ is valid in $\mathsf{G}_p$ otherwise.
Since, for every prime number $p$, $\mathsf{G}_p$ is an extension-closed pseudovariety, by , there are $\kappa$-words $w_{1,i},\ldots,w_{s_i,i}$ such that $K_i=\left<w_{1,i}\ldots,w_{s_i,i}\right>$ and $rk(K_i)\leq rk(H_i)$ ($1\leq i \leq t$).
Fix $q\in \displaystyle\bigcap_{i=1}^tS(H_i)$. As $K_i$ contains $H_i$, we can rewrite the generator $h$ of $H_i$ as a reduced word $h'$ in terms of generators of $K_i$. Since $Cl_{\mathsf{G}_q,\kappa}(H_i)=K_i$, by $H_i$ is $\mathsf{G}_q$-dense in the pro-$\mathsf{G}_q$ topology on $K_i$. Hence, by [@Jorge:2002 Proposition 5.2], there is a subset $\{h_{1,i}',\ldots,h_{s_{i},i}'\}$ of generators of $H_i$ such that the subgroup $H_i'$ generated by $\{h_{1,i}',\ldots,h_{s_{i},i}'\}$ is $\mathsf{G}_q$-dense in the pro-$\mathsf{G}_q$ topology on $K_i$. Let $M(h_{1,i}',\ldots,h_{s_i,i}')$ be the $s_i\times s_i$ matrix whose $(k,j)$-entry is the number of occurrences of $w_{j,i}$ in $h_{k,i}'$. Then by , we have $\det{ M(h_{1,i}',\ldots,h_{s_i,i}')}\nequiv 0 \pmod{q}$ and also for every prime number $p$ not dividing $\det{M(h_{1,i}',\ldots,h_{s_i,i}')}$, the subgroup $H_i'$ is $p$-dense in the pro-$\mathsf{G}_p$ topology on $K_i$. Since for every $p\in S(H_i)$, $K_i$ is $\mathsf{G}_p$-closed, by for every $p\in S(H_i)$ not dividing $\det{M(h_{1,i}',\ldots,h_{s_i,i}')}$ we have $Cl_{\mathsf{G}_p,\kappa}(H_i')=Cl_{\mathsf{G}_p,K_i}(H_i')=K_i$, where $Cl_{\mathsf{G}_p,K_i}(H_i')$ is the $p$-closure of $H_i'$ in the pro-$\mathsf{G}_p$ topology on $K_i$. Hence, by the proof of [@Jorge:2002 Theorem 6.1], for every $p\in S(H_i)$ not dividing $\det{M(h_{1,i}',\ldots,h_{s_i,i}')}$, the following pseudoidentity holds in $\mathsf{G}_p$ $$\label{01}
w_{j,i}=a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega$$ where $h_{k,i}$ corresponds to $h_{k,i}'$ as a word in terms of the alphabet $A$. Note that $a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega$ belongs to $Cl(H_i)\subseteq \overline{\Omega}_A\mathsf{S}$.
Let $$S_1(H_i)=S(H_i)\setminus\{p\mid \det{M(h_{1,i}',\ldots,h_{s_i,i}')} \equiv0 \pmod{p}\}.$$ As $\det{M(h_{1,i}',\ldots,h_{s_i,i}')}\neq 0$, the sets $S_1(H_i)$ are cofinite subsets of $\mathds{P}$.
Let $P_0=\{p_1,\ldots,p_r\}=\mathds{P}\setminus \bigcap_{i=1}^rS_1(H)$ and $n_0=p_1\ldots p_r$. Consider the following pseudowords: $$u_{j,i}= \left(a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega\right)^{{n_0}^\omega}\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_i).$$ If $p\in \mathds{P}\setminus P_0$, by Lemma \[lemma54\], the following pseudoidentities are valid in $\mathsf{G}_{p}$: $$\begin{aligned}
u_{j,i}&= \left(a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega\right)^{{n_0}^\omega}=a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega\stackrel{\eqref{01}}{=}w_{j,i}.
\end{aligned}$$ Otherwise, the following pseudoidentities are valid in $\mathsf{G}_{p}$ ($p\in P_0$): $$\begin{aligned}
u_{j,i}&=\left(a_j\circ(h_{1,i},\ldots,h_{s_i,i})^\omega\right)^{{n_0}^\omega}=1.
\end{aligned}$$ So, we showed that for every generator $w_{k,i}$ of $K_{i}$, there are $u_{k,i}\in \Omega_A^{\sigma}\mathsf{S} \cap Cl(H_i)$ such that, if $p\in \bigcap_{i=1}^rS_1(H_i)$, then the pseudoidentity $u_{k,i}=w_{k,i}$ is valid in $\mathsf{G}_{{p}}$ and the pseudoidentity $u_{k,i}=1$ is valid in $\mathsf{G}_{{p}}$ otherwise. Hence, every $w\in K_i$ has this property ($1\leq i\leq t$).
Let $w\in K_1\ldots K_t$. Then there are $w_i\in K_i$ such that $w=w_1\ldots w_t$. By the preceding paragraph, there are $u_{i}\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_i)$ such if $p\in \bigcap_{i=1}^rS_1(H_i)$, then the pseudoidentity $u_{i}=w_i$ is valid in $\mathsf{G}_{{p}}$ and the pseudoidentity $u_{i}=1$ is valid in $\mathsf{G}_{{p}}$ otherwise. Let $v=u_1\ldots u_t\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_1\ldots H_t).$
If $p\in \bigcap_{i=1}^rS_1(H_i)$, then the pseudoidentity $v=w$ is valid in $\mathsf{G}_{{p}}$ and the pseudoidentity $v=1$ is valid in $\mathsf{G}_{{p}}$ otherwise.
\[proposition2\] Let $H_1,\ldots, H_t$ be f.g. subgroups of the free group and fix $S_1(H_i)$ as in the preceding proposition. Then for every $p$ in the set $$\mathds{P}\setminus \displaystyle\bigcap_{i=1}^t S_1(H_i)=\{p_1\ldots,p_r\}$$ and every $w\in Cl_{\mathsf{G}_{p},\kappa}(H_1\ldots H_t)$, there are $u_i\in\Omega_A^{\sigma}\mathsf{S}\cap Cl(H_i)$ such that the pseudoidentity $u_{i}=1$ is valid in $\mathsf{G}_q$ for every $q\in \mathds{P}\setminus \{p\}$ and the pseudoidentity $u_1\ldots u_t=w$ is valid in $\mathsf{G}_{p}$.
By , for every prime number $p$, we have $$\label{eqc}
Cl_{\mathsf{G}_{p},\kappa}(H_1\ldots H_t)=Cl_{\mathsf{G}_{p},\kappa}(H_1)\ldots Cl_{\mathsf{G}_{p},\kappa}(H_t).$$
Let $P_0=\{p_1,\ldots,p_r\}$ and $K_{p_i,j}=Cl_{\mathsf{G}_{p_i},\kappa}(H_j)$ ($1\leq i\leq r$ and $1\leq j\leq t$). There are $\kappa$-words $w_{1,p_i,j},\ldots, w_{s_{p_i,j},p_i,j}$ such that $$K_{p_i,j}=\left<w_{1,p_i,j},\ldots, w_{s_{p_i,j},i,j}\right>.$$ By the proof of [@Jorge:2002 Theorem 6.1], there is a subset $\{h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j}\}$ of the generators of $H_j$ such that $$Cl_{\mathsf{G}_{p_i},\kappa}(\left<h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j}\right>)= Cl_{\mathsf{G}_{p_i},\kappa}(H_j)=K_{p_i,j},$$ and the pseudoidentity $$\label{eq23}
w_{k,p_i,j}=a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\ \ \ \ (1\leq k\leq s_{p_i,j})$$ is valid in $\mathsf{G}_{p_i}$. We consider the following finite subsets of $\mathds{P}$: $$\begin{aligned}
P_i=&\bigcup_{j=1}^{t}\{p\mid p \ \text{divides}\ \det{M(h_{1,p_i,j}',\ldots,h_{s_{p_i,j},p_i,j}')}\} &&&&(1\leq i\leq r)
\end{aligned}$$ where $h_{k,p_i,j}'$ is generator $ h_{k,p_i,j}$ written in terms of generators of $K_{p_i,j}$ ($1\leq k\leq s_{p_i,j}$). Note that, since $K_{p_i,j}$ is a $p_i$-closed subgroup and $$Cl_{\mathsf{G}_{p_i},\kappa}(\left<h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j}\right>)=K_{p_i,j},$$ by the subgroup $\left<h_{1,p_i,j}',\ldots,h_{s_{p_i,j},p_i,j}'\right>$ is $p_i$-dense in the pro-$\mathsf{G}_{p_i}$ topology on $K_{p_i,j}$ and, therefore, $p_i$ does not belong to $P_i$. Consider the following natural numbers: $$\begin{aligned}
n_0=&p_1\ldots p_r,\\
n_i=&\prod_{j=0}^r\prod_{p\in P_j\setminus \{p_i\}} p &&&&&&&&(1\leq i\leq r).
\end{aligned}$$ The natural numbers $n_i$ ($1\leq i\leq r$) satisfy the following properties:
1. The prime number $p_i$ does not divide $n_i$.
2. Fix $p_i\in P_0$. For every $p_j\in P_0$ ($j\neq i$), $p_j$ divides $n_i$, because $p_j$ belongs to the set $P_0\setminus \{p_i\}$.
3. Since we have $p_i\notin P_i$, every prime number $p\in P_i$ divides $n_i$.
4. For every prime number $p$ in $\left(P_1\cup\ldots\cup P_r\right)\setminus P_0$, $p$ divides $n_i$ ($1\leq i\leq r$), because there is $j$ such that $p\in P_j$ and since $p$ does not belong to $P_0$, $p$ is in $P_j\setminus \{p_i\}$.
For every $i$, $j$, and $k$ ($1\leq j\leq s_{p_i,j}$, $1\leq i\leq r$, and $1\leq j\leq t$), we let $$\begin{aligned}
u_{k,p_i,j}&= \left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{{n_i}^\omega}\\
&\left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{(\omega-1){(p_in_i)}^\omega}.
\end{aligned}$$ Note that $u_{k,p_i,j}$ belongs to $ \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_j)$. We claim that the pseudoidentity $u_{k,p_i,j}=w_{k,p_i,j}$ is valid in $\mathsf{G}_{p_i}$ and for every $p\in \mathds{P}\setminus\{p_i\}$ and the pseudoidentity $u_{k,p_i,j}=1$ is valid in $\mathsf{G}_p$ ($1\leq k\leq s_{p_i,j}$ and $1\leq j\leq t$).
It remains to establish the claim. Consider the following cases:
- Let $p=p_i$. By the property (1) of $n_i$, $p_i$ does not divides $n_i$. Hence, we have the following pseudoidentities in $\mathsf{G}_{p_i}$ $$\begin{aligned}
u_{k,p_i,j}=& \left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{{n_i}^\omega} \\ &\left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{(\omega-1){(p_in_i)}^\omega}\\
=& \left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)1 \stackrel{\eqref{eq23}}{=}w_{k,p_i,j}.
\end{aligned}$$
- Consider either $p\in P_0\setminus \{p_i\}$ or $p\in (P_1\cup\ldots\cup P_r)\setminus P_0$ . By the property (2) and (4) of $n_i$, $p$ divides $n_i$. Hence, we have the following pseudoidentities in $\mathsf{G}_{p}$ $$\begin{aligned}
\left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{{n_i}^\omega} =&1,\\ \left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{(\omega-1){(p_in_i)}^\omega}=&1.
\end{aligned}$$ So, the pseudoidentity $u_{k,p_i,j}=1$ is valid in $\mathsf{G}_p$.
- Consider $p\in \mathds{P}\setminus (P_0\cup\ldots\cup P_r) $. By the choice of $n_i$, $p$ does not divide $p_in_i$ and, therefore, does not divide $n_i$. Hence, we have the following pseudoidentities in $\mathsf{G}_{p}$ $$\begin{aligned}
u_{k,p_i,j}=& \left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{{n_i}^\omega}\\ &\left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{(\omega-1){(p_in_i)}^\omega}\\
=& \left( a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)\\
&\left(a_k\circ(h_{1,p_i,j},\ldots,h_{s_{p_i,j},p_i,j})^\omega\right)^{\omega-1}\\=&1.
\end{aligned}$$ This completes the proof of the claim.
We showed for every generator $w_{k,p_i,j}$ of $Cl_{\mathsf{G}_{p_i},\kappa}(H_j)=K_{p_i,j}$, there are $u_{k,p_i,j}\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_j)$ such that the pseudoidentity $u_{k,p_i,j}=w_{k,p_i,j}$ is valid in $\mathsf{G}_{{p}_i}$ and the pseudoidentity $u_{k,p_i,j}=1$ is valid in $\mathsf{G}_{{p}}$ ($ p\in \mathds{P}\setminus \{p_i\}$). Hence, every $w\in Cl_{\mathsf{G}_{p_i},\kappa}(H_j)$ has this property.
Fix $p\in P_0$ and let $w\in Cl_{\mathsf{G}_p,\kappa}(H_1\ldots H_t)$. By , there are $w_{j}\in Cl_{\mathsf{G}_p,\kappa}(H_j)$ such that $w=w_1\ldots w_t$. By the preceding paragraph, there are $u_{j}\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_j)$ such that the pseudoidentity $u_{j}=w_j$ is valid in $\mathsf{G}_{{p}}$ and the pseudoidentity $u_{j}=1$ is valid in $\mathsf{G}_{{q}}$ ($q\in \mathds{P}\setminus \{p\}$). Let $$v=u_{1}\ldots u_{t}\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(H_1\ldots H_t).$$ Then the equality $v=w$ is valid in $\mathsf{G}_{p}$ and the equality $v=1$ is valid in $\mathsf{G}_{{q}}$ ($ q\in \mathds{P}\setminus \{p\}$).
\[cor1\] For every prime number $p$, the pseudovariety $\mathsf{G}_p$ is $\sigma$-reducible with respect to the systems of equations associated with finite directed graphs.
By Theorem \[th1\], for a rational subset $L$ of $A^*$ we have $$Cl_{\mathsf{G}_p,\sigma}(\psi_{\mathsf{G}_p}(L))=Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L)).$$ Since the pseudovariety $\mathsf{G}_p$ is weakly $\kappa$-reducible with respect to the systems of equations associated with finite directed graphs [@Steinberg:1998a], by Theorem \[proposition3\] we just need to show that $\mathsf{G}_p$ is $\sigma$-full.
By definition of $\sigma$-full pseudovariety, it is enough to show that for a rational subset of $A^*$: $$\label{eq31}
\psi_{\mathsf{G}_p}(Cl_\sigma(L))\subseteq Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L)),$$ or equivalently, it is enough to show that $$\label{eq40}
w\in Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))\Rightarrow Cl(L)\cap \Omega_A^\sigma \mathsf{S}\cap (\psi^{-1}(w))\neq \emptyset$$ The proof is similar to the proof of [@Jorge:2002 Theorem 6.1]. Since any rational subset of the free semigroup can be obtained by taking a finite number of finite subsets of the free semigroup and applying the union, product and the plus operation $L\rightarrow L^{+}$ a finite number of times, it is enough to show that these operations preserve this property. As it is mentioned in the proof of [@Jorge:2002 Theorem 6.2], for the rational subsets $L$ and $K$, we have
1. $Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))=\psi_{\mathsf{G}_p}(L)$ if $L$ is finite;
2. $Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(LK))=Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(K))$;
3. $Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L\cup K))=Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))\cup Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))$
4. $Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L)^+)= Cl_{\mathsf{G}_p,\kappa}(\left<\psi_{\mathsf{G}_p}(L)\right>)$
If a language $L$ is finite, then any $w$ in $Cl_{\mathsf{G}_p,\kappa}(\psi_{\mathsf{G}_p}(L))$ is also an element of $Cl(L)\cap \Omega_A^\sigma \mathsf{S}\cap (\psi^{-1}(w))$ and so finite languages satisfy . Suppose that $L_1$ and $L_2$ are rational subsets of $\Omega_A\mathsf{S}$ satisfy Property . By property (3), at least one of the sets $Cl(L_1)\cap \Omega_A^\sigma \mathsf{S}\cap (\psi^{-1}(w))$ and $Cl(L_2)\cap \Omega_A^\sigma \mathsf{S}\cap (\psi^{-1}(w))$ is nonempty and, therefore, so is their union.
Let $u\in \psi_{\mathsf{G}_p}(Cl(L_1L_2))\cap \Omega_A^{\sigma}\mathsf{G}_p$. By property (2), there are $$u_i\in Cl_{\kappa,\mathsf{G}_p}( \psi_{\mathsf{G}_p}(L_i))\ \ (i=1,2)$$ such that $u=u_1u_2$. By the induction hypotheses, there are $w_i\in \psi_{\mathsf{G}_p}^{-1}(u_i)\cap\Omega_A^{\sigma}\mathsf{S}\cap Cl(L_i)$. Let $w=w_1w_2\in \Omega_A^{\sigma}\mathsf{S}\cap Cl(L_1)Cl(L_2)\subseteq\Omega_A^{\sigma}\mathsf{S}\cap Cl(L_1L_2)$. Then we have $$\psi_{\mathsf{G}_p}(w)=\psi_{\mathsf{G}_p}(w_1w_2)=\psi_{\mathsf{G}_p}(w_1)\psi_{\mathsf{G}_p}(w_2)=u_1u_2=u.$$ Thus, $w$ belongs to $\Omega_A^{\sigma}\mathsf{S}\cap Cl(L_1L_2)\cap \psi_{\mathsf{G}_p}^{-1}(u)$.
Let $L$ be a rational subset satisfies the property . By [@Jorge:2002 Lemma 6.6], the is a finite subset $Y$ of $\kappa$ words, such that $\left<L\right>=\left<Y\right>$ and $Cl(Y)\subseteq CL(L)$. Hence, to show that a rational language of the form $L^+$ satisfies , it is enough to establish that, for any finite subset $Y$ of $\kappa$ words the following property holds: $$w\in Cl_{\mathsf{G}_p,\kappa}(\left<Y\right>)\Rightarrow Cl(Y^+)\cap \Omega_A^\sigma \mathsf{S}\cap (\psi^{-1}(w))\neq \emptyset$$ The result follows from the Propositions \[lemma52\] and \[proposition2\] by $t=1$.
Let $\mathcal{C}$ be a class of systems of equations. For every prime number $p$, the pseudovariety $\mathsf{G}_p$ is $\sigma$-reducible with respect to the systems of equations in $\mathcal{C}$ if and only if the pseudovariety $\mathsf{G}_{nil}$ is $\sigma$-reducible with respect to the systems of equations in $\mathcal{C}$
Let $\mathsf{V}$ and $ \mathsf{W}$ be pseudovarieties of semigroups such that $\mathsf{V}\subseteq \mathsf{W}$. If a pseudoidentity is valid in $\mathsf{W}$, then it is valid in $\mathsf{V}$. Hence, if a equation has solution $\delta$ modulo $\mathsf{W}$, then $\delta$ is a solution modulo $\mathsf{V}$. Conversely, let $X$ be a finite set of variable and let $k$ be the cardinality of $X$. Consider a system of equations in $\mathcal{C}$ of the form $$\label{1}
u_i=v_i \ \ \ (i=1,\ldots,m),$$ with rational constraints $L_x\subseteq (\Omega_{A}\mathsf{S})^1$ and $u_i,v_i\in\Omega_{X}^\sigma \mathsf{S}$. Suppose that this system has the solution $\delta:X\rightarrow \overline{\Omega}_A\mathsf{S}$ modulo $\mathsf{G}_{nil}$. Since $L_x$ is a rational subset of $(\Omega_{A}\mathsf{S})^1$, $L_x$ is a finite union of sets of the form $R_0^{*}w_0R_1^{*}\ldots w_tR_t^{*}$, where $R_i$ are rational subsets of $(\Omega_{A}\mathsf{S})^1$ and $w_i\in (\Omega_{A}\mathsf{S})^1$. Hence, for every $x\in X$, there is a simple rational subset $$L_x'=R_{0,x}^{*}u_{0,x}R_{1,x}^{*}\ldots u_{t_x-1,x}R_{t_x,x}^{*}\subseteq L_x$$ such that $\delta(x)\in Cl_{\mathsf{G}_{nil}}(L_x')$. So, the system with constraints $L'_x\subseteq \Omega_{A}\mathsf{S}$ has the solution $\delta$ modulo $\mathsf{G}_{nil}$ and, therefore, for every prime number $p$, the system with constraints $L'_x\subseteq \Omega_{A}\mathsf{S}$ has a solution modulo $\mathsf{G}_{p}$. We show that the system with constraints $L_x'$ ($x\in X$) has a solution $\delta':X\rightarrow {\Omega}^\sigma_A\mathsf{S}$ modulo $\mathsf{G}_{nil}$.
By [@Jorge:2002 Lemma 6.6], there are finite subsets $Y_{i,x}$ of $\Omega_A^\kappa\mathsf{S}$ such that $$Cl(Y_{i,x}^{*})\subseteq Cl(R_{i,x}^{*}),$$ and the subgroup generated by $Y_{i,x}$ is equal to the subgroup generated by $R_{i,x}$ in the free group ($x\in X$ and $1\leq i\leq t_x$). Hence, we have $$\label{eq26}
Cl_{\mathsf{G}_{p}}(R_{0,x}^{*}u_{0,x}R_{1,x}^{*}\ldots u_{t_x-1,x}R_{t_x,x}^{*})=Cl_{\mathsf{G}_{p}}(Y_{0,x}^{*}u_{0,x}Y_{1,x}^{*}\ldots u_{t_x-1,x}Y_{t_x,x}^{*}).$$
Let $S_1(\left<Y_{i,x}\right>)$ be as in Proposition \[lemma52\] and $S=\displaystyle\bigcap_{x\in X}\displaystyle\bigcap_{i=1}^{t_x}S_1(\left<Y_{i,x}\right>)$. Fix $q\in S$. Since the system with constraints $L'_x\subseteq \Omega_{A}\mathsf{S}$ has a solution modulo $\mathsf{G}_{q}$ and $\mathsf{G}_{q}$ is $\sigma$-reducible, there is a solution $\delta_0:X\rightarrow{{\Omega}}^\sigma_A\mathsf{S}$ modulo $\mathsf{G}_q$. Hence, for every $x\in X$, $\delta_0(x)$ lies in $Cl_{\mathsf{G}_{q},\sigma}(L_x')=Cl_{\mathsf{G}_{q},\kappa}(L_x')$. For every $p\in S$, we have $$\begin{aligned}
Cl_{\mathsf{G}_{p},\kappa}(L_x')&=Cl_{\mathsf{G}_{p},\kappa}\left(R_{0,x}^{*}u_{0,x}R_{1,x}^{*}\ldots u_{t_x-1,x}R_{t_x,x}^{*}\right)\\
&=Cl_{\mathsf{G}_{p},\kappa}(R_{0,x}^{*})u_{0,x}Cl_{\mathsf{G}_{p},\kappa}(R_{1,x}^{*})\ldots
u_{t_x-1,x}Cl_{\mathsf{G}_{p},\kappa}(R_{t_x,x}^{*})\\
&=Cl_{\mathsf{G}_{p},\kappa}(\left<R_{0,x}\right>)u_{0,x}Cl_{\mathsf{G}_{p},\kappa}(\left<R_{1,x}\right>)\ldots
u_{t_x-1,x}Cl_{\mathsf{G}_{p},\kappa}(\left<R_{t_x,x}\right>)\\
&=Cl_{\mathsf{G}_{p},\kappa}(\left<Y_{0,x}\right>)u_{0,x}Cl_{\mathsf{G}_{p},\kappa}(\left<Y_{1,x}\right>)\ldots
u_{t_x-1,x}Cl_{\mathsf{G}_{p},\kappa}(\left<Y_{t_x,x}\right>)\\
&=Cl_{\mathsf{G}_{q},\kappa}(\left<Y_{0,x}\right>)u_{0,x}Cl_{\mathsf{G}_{q},\kappa}(\left<Y_{1,x}\right>)\ldots
u_{t_x-1,x}Cl_{\mathsf{G}_{q},\kappa}(\left<Y_{t_x,x}\right>)\\
&=Cl_{\mathsf{G}_{q},\kappa}\left(Y_{0,x}^{*}u_{0,x}Y_{1,x}^{*}\ldots u_{t_x-1,x}Y_{t_x,x}^{*}\right)\\
&=Cl_{\mathsf{G}_{q},\kappa}(L_x').
\end{aligned}$$ So, for every $p\in S$, $\delta_0(x)$ lies in $Cl_{\mathsf{G}_{p},\kappa}(L_x')$.
By Proposition \[lemma52\], there are $v_{i,x}\in Cl(Y_{i,x}^{*})\cap \Omega_A^{\sigma}\mathsf{S}$ such that if $p\in \mathds{P}\setminus S$, then the pseudoidentity $v_{i,x}=1$ is valid in $\mathsf{G}_p$ and the pseudoidentity $$\label{eq17}
v_{0,x}u_{0,x}v_{1,x}\ldots u_{t_x-1,x}v_{t_x,x}=\delta_0(x).$$ is valid in $\mathsf{G}_p$ otherwise ($x\in X$ and $1\leq i\leq t_x$). Note that since $Cl(Y_{i,x}^{*})\subseteq Cl(R_{i,x}^{*})$, the pseudowords $v_{0,x}u_{0,x}v_{1,x}\ldots u_{t_x-1,x}v_{t_x,x}$ lie in $Cl(L_x')$.
Let $P=\{p_1,\ldots,p_r\}=\mathds{P}\setminus S$. For every $p_j\in P$, there is a solution $\delta_j:X\rightarrow{{\Omega}}^\sigma_A\mathsf{S}$ modulo $\mathsf{G}_{p_j}$ such that for every $x\in X$, $\delta_j(x)$ lies in $Cl_{\mathsf{G}_{p_j},\kappa}(L_x')$.
By Proposition \[lemma52\], there are $v_{i,x,p_j}\in Cl(Y_{i,x}^*)\cap \Omega_A^{\sigma}\mathsf{S}$ such that if $p=p_j$, then the equality $$\label{eq18}
v_{0,i,p_j}u_{0,x}v_{1,x,p_j}\ldots u_{t_x-1,x}v_{t_x,x,p_j}=\delta_j(x)$$ holds in $\mathsf{G}_p$ and the equality $v_{i,x,p_j}=1$ is valid in $\mathsf{G}_p$ otherwise ($x\in X$, $1\leq i\leq t_x$, $1\leq j\leq r$). Note that since $Cl(Y_{i,x}^*)\subseteq Cl(R_{i,x}^*)$, the pseudowords $$v_{0,x,p_j}u_{0,x}v_{1,x,p_j}\ldots u_{t_x-1,x}v_{t_x,x,p_j}$$ lie in $Cl(L_x')$.
We define the function $\delta':X\rightarrow{\Omega}_A^{\sigma}\mathsf{S}$ with $$\begin{aligned}
\delta'(x)&=\left(v_{0,x}v_{0,x,p_1}\ldots v_{0,x,p_r}\right)u_{0,x}\left(v_{1,x}v_{1,x,p_1}\ldots v_{1,x,p_r}\right)\\
&u_{t_x-1,x}\left(v_{t_x,x}v_{t_x,x,p_1}\ldots v_{t_x,x,p_r}\right).
\end{aligned}$$ We claim that $\delta'$ is a solution of the system modulo $\mathsf{G}_{nil}$.
It remains to establish the claim. Note that since a closure of a subsemigroup is again a subsemigroup, the pseudoword $v_{i,x}v_{i,x,p_1}\ldots v_{i,x,p_r}$ lies in $Cl(Y_{i,x}^*)$ and so, $\delta'(x)$ lies in $Cl(L_x')$. We show that the pseudovariety $\mathsf{G}_{nil}$ satisfies the pseudoidentities $\delta'(u_i)=\delta'(v_i)$ ($1\leq i\leq m$). By Lemma \[lemma53\], it is enough to show that for every prime number $p$, the pseudovariety $\mathsf{G}_{p}$ satisfies the pseudoidentities $\delta'(u_i)=\delta'(v_i)$ ($1\leq i\leq m$).
First we show that
1. if $p\in S$, then for every $x\in X$ the pseudoidentity $$\label{eq19}
\delta'(x)=\delta_0(x)$$ is valid in $\mathsf{G}_p$;
2. if $p=p_m\in \mathds{P}\setminus S$, then for every $x\in X$ the pseudoidentity $$\label{eq22}
\delta'(x)=\delta_m(x)$$ is valid in $\mathsf{G}_{p_m}$.
We consider the following cases:
- Let $p\in S$. Then for every $x\in X$, the following pseudoidentities hold in $\mathsf{G}_p$: $$\begin{aligned}
v_{i,x,p_j}=1&&&&&&(1\leq i\leq t_{x} \ \text{and} \ 1\leq j\leq r), \\
\end{aligned}$$ Hence, the following pseudoidentities hold in $\mathsf{G}_p$: $$\begin{aligned}
\delta'(x)&=v_{0,x}u_{0,x}v_{1,x}u_{t_{x}-1,x}v_{t_{x},x}\stackrel{\eqref{eq17}}{=}
\delta_0(x).
\end{aligned}$$
- Let $p=p_m\in \mathds{P}\setminus S$. Then the following pseudoidentities hold in $\mathsf{G}_p$: $$\begin{aligned}
v_{i,x,p_j}=v_{s,x}=1&&&&&&(1\leq i,s\leq t_{x} \ \text{and} \ j\neq m).
\end{aligned}$$ Hence, the following pseudoidentities hold in $\mathsf{G}_{p_m}$: $$\begin{aligned}
\delta'(x)&=v_{0,x,p_m}u_{0,x}v_{1,x,p_m}u_{t_{x}-1,x}v_{t_{x},x,p_m}\stackrel{\eqref{eq18}}{=}
\delta_m(x).
\end{aligned}$$
Now we show that for every prime number $p$, the pseudovariety $\mathsf{G}_{p}$ satisfies the pseudoidentities $\delta'(u_i)=\delta'(v_i)$ ($1\leq i\leq m$). We consider the two following cases:
- Let $p\in S$. Since $\delta_0$ is the solution of the system modulo $\mathsf{G}_p$, $\delta_0(x)=\delta'(x)$ ($x\in X$), and $u_i,v_i\in \Omega_X^\sigma\mathsf{S}$, the pseudovariety $\mathsf{G}_p$ satisfies the pseudoidentities $\delta'(u_i)=\delta'(v_i)$ ($1\leq i \leq m$).
- consider $p_j\in \mathds{P}\setminus S$. Since $\delta_j$ is the solution of the system modulo $\mathsf{G}_{p_j}$, $\delta_j(x)=\delta'(x)$ ($x\in X$), and $u_i,v_i\in \Omega_X^\sigma\mathsf{S}$, the pseudovariety $\mathsf{G}_{p_j}$ satisfies the pseudoidentities $\delta'(u_i)=\delta'(v_i)$ ($1\leq i \leq m$).
This proves the theorem.
The pseudovariety $\mathsf{G}_{nil}$ is $\sigma$-reducible with respect to the systems of equations associated with finite directed graphs.
The result follows from Corollary \[cor1\] and the preceding theorem.
The pseudovariety $\mathsf{G}_{nil}$ is not completely $\sigma$-reducible.
By the preceding Theorem, it is enough to show that for some prime number $p$, the pseudovariety $\mathbf{G}_p$ is not completely $\sigma$-reducible. Let $p$ be a odd prime number, $A=\{a,b\}$ and consider the following equation $$\label{eq30}
[x^2a,y^{-1}z^2by]=1$$ It has been shown that the equation does not have solution in the free group . We consider the following constraints:
- $L_x=\{a\}^*$;
- $L_z=\{b\}^*$;
- $L_y=A^*$.
Let $p$ be an odd prime number. We find a solution of the equation with the above constraints modulo $\mathsf{G}_p$. The pseudowords $x=(a^{\omega-1})^{2^{\omega-1}}$, $z=(b^{\omega-1})^{2^{\omega-1}}$ and $y=1$ are a solutions of this equation modulo $\mathsf{G}_p$ because the pseudoidentities $(a^{\omega-1})^{2^{\omega}}=a^{\omega-1}$ and $(b^{\omega-1})^{2^{\omega}}=b^{\omega-1}$ hold in $\mathsf{G}_p$.
Since $\mathsf{G}_p$ is $\sigma$-full, $\mathsf{G}_p$ is $\sigma$-reducible for the equation $\eqref{eq30}$ if and only if it has a solution $\delta:\overline{\Omega}_X\mathsf{S}\rightarrow\overline{\Omega}_A \mathsf{S}$ such that $\psi_{\mathsf{G}_p}(\delta(u))=1$ and $\psi_{\mathsf{G}_p}(\delta(x))\in {\Omega}_A^\sigma \mathsf{G}_p$. But as $\Omega_A^\sigma \mathsf{G}_p=\Omega_A^\kappa \mathsf{G}_p$, the equation must have a solution in the free group which is contradiction.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work is part of the author’s preparation of a doctoral thesis under the supervision of Prof. Jorge Almeida, whose advice is gratefully acknowledged.
It was partially supported by the FCT Docoral Grant with reference (SFRH/ BD/98202/2013). It was also partially supported by CMUP (UID /MAT/00144/ 2013), which is funded by FCT (Portugal) with national (MEC) and European structural funds (FEDER), under the partnership agreement PT2020.
[10]{}
D. Albert, R. Baldinger, and J. Rhodes, *Undecidability of the identity problem for finite semigroups*, J. Symbolic Logic **57** (1992), 179–192.
J. Almeida, *Finite semigroups and universal algebra*, Series in Algebra, vol. 3, World Scientific Publishing Co., Inc., River Edge, NJ, 1994, Translated from the 1992 Portuguese original and revised by the author.
[to3em]{}, *Dynamics of implicit operations and tameness of pseudovarieties of groups*, Trans. Amer. Math. Soc. **354** (2002), 387–411.
J. Almeida, A. Azevedo, and M. Zeitoun, *Pseudovariety joins involving [$J$]{}-trivial semigroups*, Internat. J. Algebra Comput. **9** (1999), 99–112.
J. Almeida, J. C. Costa, and M. L. Teixeira, *Semidirect product with an order-computable pseudovariety and tameness*, Semigroup Forum **81** (2010), 26–50.
J. Almeida, J. C. Costa, and M. Zeitoun, *Tameness of pseudovariety joins involving [R]{}*, Monatsh. Math. **146** (2005), 89–111.
[to3em]{}, *Complete reducibility of systems of equations with respect to [$R$]{}*, Port. Math. (N.S.) **64** (2007), 445–508.
J. Almeida and M. Delgado, *Sur certains systèmes d’équations avec contraintes dans un groupe libre*, Portugal. Math. **56** (1999), 409–417.
[to3em]{}, *Sur certains systèmes d’équations avec contraintes dans un groupe libre—addenda*, Portugal. Math. **58** (2001), 379–387.
[to3em]{}, *Tameness of the pseudovariety of abelian groups*, Internat. J. Algebra Comput. **15** (2005), 327–338.
J. Almeida and B. Steinberg, *On the decidability of iterated semidirect products with applications to complexity*, Proc. London Math. Soc. (3) **80** (2000), 50–74.
C. J. Ash, *Inevitable graphs: a proof of the type [II]{} conjecture and some related decision procedures*, Internat. J. Algebra Comput. **1** (1991), 127–146.
T. Coulbois and A. Khélif, *Equations in free groups are not finitely approximable*, Proc. Amer. Math. Soc. **127** (1999), 963–965.
S. Eilenberg, *Automata, languages and machines*, vol. A, Academic Press, New York, 1974.
R. Gitik, *On the profinite topology on negatively curved groups*, J. Algebra **219** (1999), 80–86.
R. Gitik and E. Rips, *On separability properties of groups*, Internat. J. Algebra Comput. **5** (1995), 703–717.
B. Herwig and D. Lascar, *Extending partial automorphisms and the profinite topology on free groups*, Trans. Amer. Math. Soc. **352** (2000), 1985–2021.
S. Margolis, M. Sapir, and P. Weil, *Closed subgroups in pro-[$\bold V$]{} topologies and the extension problem for inverse automata*, Internat. J. Algebra Comput. **11** (2001), 405–445.
J. Rhodes, *Undecidability, automata and pseudovarieties of finite semigroups*, Internat. J. Algebra Comput. **9** (1999), 455–473.
L. Ribes and P. A. Zalesski[ĭ]{}, *The pro-$p$ topology of a free group and algorithmic problems in semigroups*, Internat. J. Algebra Comput. **4** (1994), 359–374.
B. Steinberg, *Inevitable graphs and profinite topologies: some solutions to algorithmic problems in monoid and automata theory, stemming from group theory*, Internat. J. Algebra Comput. **11** (2001), 25–71.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The *threshold degree* of a function $f\colon\zoon\to\moo$ is the least degree of a real polynomial $p$ with $f(x)\equiv\sign p(x).$ We prove that the intersection of two halfspaces on $\zoon$ has threshold degree $\Omega(n),$ which matches the trivial upper bound and completely answers a question due to Klivans (2002). The best previous lower bound was $\Omega(\sqrt n).$ Our result shows that the intersection of two halfspaces on $\zoon$ only admits a trivial $2^{\Theta(n)}$-time learning algorithm based on sign-representation by polynomials, unlike the advances achieved in PAC learning DNF formulas and read-once Boolean formulas. The proof introduces a new technique of independent interest, based on Fourier analysis and matrix theory.'
author:
- 'Alexander A. Sherstov$^*$'
bibliography:
- |
%
/Users/sasha/bib/general.bib
- |
%
/Users/sasha/bib/fourier.bib
- |
%
/Users/sasha/bib/cc.bib
- |
%
/Users/sasha/bib/learn.bib
- |
%
/Users/sasha/bib/my.bib
title: |
Optimal Bounds for Sign-Representing the Intersection\
of Two Halfspaces by Polynomials
---
[^1]
Introduction {#sec:intro}
============
A well-studied notion in computational learning theory is that of a *perceptron*. This term stands for the representation of a given Boolean function $f\colon\zoon\to\moo$ in the form $
f(x) \equiv \sign p(x)
$ for a real polynomial $p$ of some degree $d.$ The least degree $d$ for which $f$ admits such a representation is called the *threshold degree* of $f,$ denoted $\degthr(f).$ In other words, $\degthr(f)$ is the least degree of a real polynomial that agrees with $f$ in sign. Perceptrons are appealing from a learning standpoint because they immediately lead to efficient learning algorithms. In more detail, let $f\colon\zoon\to\moo$ be an unknown function of threshold degree $d.$ Then by definition, $f$ has a representation of the form $$\begin{aligned}
f(x) \equiv \sign {\left(
\sum_{\substack{\abs{S}\leq d}}
^{\phantom{\abs{S}\leq d}}
\lambda_S \prod_{i\in S} x_i\right)} \end{aligned}$$ for some reals $\lambda_S$ and is thus a halfspace in $N={n\choose 0}+{n\choose 1}+\cdots+{n\choose d}$ dimensions. As a result, $f$ can be PAC learned in time polynomial in $N,$ using any of a variety of halfspace learning algorithms. (Throughout this paper, the term “PAC learning” refers to Valiant’s standard model [@valiant84pac] of learning under arbitrary distributions.)
The study of perceptrons dates back forty years to the seminal monograph of Minsky and Papert [@minsky88perceptrons], who examined the threshold degree of several common functions. Today, the perceptron-based approach yields the fastest known PAC learning algorithms for several concept classes. One such is the class of DNF formulas of polynomial size, posed a challenge in Valiant’s original paper [@valiant84pac] and extensively studied over the past two decades. The fastest known algorithm for PAC learning DNF formulas runs in time $\exp\{\tilde O(n^{1/3})\}$ and is due to Klivans and Servedio [@KS01dnf]. Specifically, the authors of [@KS01dnf] prove an upper bound of $O(n^{1/3}\log n)$ on the threshold degree of polynomial-size DNF formulas, which essentially matches a classical lower bound of $\Omega(n^{1/3})$ due to Minsky and Papert [@minsky88perceptrons].
Another success story of the perceptron-based approach is the concept class of Boolean formulas, i.e., Boolean circuits with fan-out $1$ at every gate. O’Donnell and Servedio [@odonnell03degree] proved an upper bound of $\sqrt s\log^{O(d)} s$ on the threshold degree of Boolean formulas of size $s$ and depth $d,$ giving the first subexponential algorithm for a family of formulas of superconstant depth. This upper bound on the threshold degree was improved to $s^{0.5+o(1)}$ for any depth $d$ by Ambainis et al. [@ACRSZ07nand], building on a quantum query algorithm of Farhi et al. [@FGG08nand]. More recently, Lee [@lee09formulas] sharpened the upper bound to $O(\sqrt s),$ which is tight. This line of research gives the fastest known algorithm for PAC learning Boolean formulas.
Another extensively studied problem in computational learning theory, and the subject of this paper, is the problem of learning *intersections of halfspaces*, i.e., conjunctions of functions of the form $f(x)=\sign(\sum \alpha_ix_i - \theta)$ for some reals $\alpha_1,\dots,\alpha_n,\theta.$ While solutions are known to several restrictions of this problem [@blum-kannan97intersection-of-halfspaces; @KwekPitt:98; @Vempala:97; @arriaga98proj; @KOS:02; @KlivansServedio:04coltmargin; @KLT09intersections-of-halfspaces], no algorithm has been discovered for PAC learning the intersection of even two halfspaces in time faster than $2^{\Theta(n)}.$ Progress on proving hardness results has also been scarce. Indeed, all known hardness results [@blum92trainingNN; @ABFKP:04; @focs06hardness; @khot-saket08hs-and-hs] either require polynomially many halfspaces or assume *proper* learning. In particular, we are not aware of any representation-independent hardness results for PAC learning the intersection of $O(1)$ halfspaces.
Our Results {#our-results .unnumbered}
-----------
Since the perceptron-based approach yields the fastest known algorithms for PAC learning DNF formulas and read-once Boolean formulas, it is natural to wonder whether it can yield any nontrivial results for the intersection of two halfspaces. Letting $D(n)$ stand for the maximum threshold degree over all intersections of two halfspaces on $\zoon,$ the question becomes whether $D(n)$ is a nontrivial (sublinear) function of the dimension $n.$ This question has been studied by several authors, as summarized in Table \[tab:history\]. Forty years ago, Minsky and Papert [@minsky88perceptrons] used a compactness argument to show that $D(n)=\omega(1),$ the function in question being the intersection of two majorities on disjoint sets variables. O’Donnell and Servedio [@odonnell03degree] studied the same function using a rather different approach and thereby proved that $D(n)=\Omega(\log n/\log\log n).$ No nontrivial upper bounds on $D(n)$ being known, Klivans [@klivans-thesis §7] formally posed the problem of proving a lower bound substantially better than $\Omega(\log n)$ or an upper bound of $o(n).$
*Result* *Reference*
---------------------------------- ------------------------
$D(n)=\omega(1)$ [@minsky88perceptrons]
$D(n)=\Omega(\log n/\log\log n)$ [@odonnell03degree]
$D(n)=\Omega(\sqrt n)$ [@sherstov09hshs]
$D(n)=\Theta(n)$ this paper
: Lower bounds for the intersection of two halfspaces.[]{data-label="tab:history"}
It was recently shown in [@sherstov09hshs] that $D(n)=\Omega(\sqrt n),$ solving Klivans’ problem and ruling out an $n^{o(\sqrt n)}$-time PAC learning algorithm based on perceptrons. It is clear, however, that a PAC learning algorithm for the intersection of two halfspaces in time $n^{\Theta(\sqrt n)}$ would still be a breakthrough in computational learning theory, comparable to the advances in the study of DNF formulas and read-once Boolean formulas. The main contribution of this paper is to prove that $D(n)=\Omega(n),$ which matches the trivial upper bound and definitively rules out the perceptron-based approach for learning the intersection of two halfspaces in nontrivial time.
\[thm:main\] For $n=1,2,3,\dots,$ let $D(n)$ denote the maximum threshold degree of a function of the form $f(x)\wedge g(x),$ where $f,g\colon\zoon\to\moo$ are halfspaces. Then $$\begin{aligned}
D(n) = \Theta(n).\end{aligned}$$
To be more precise, we give a randomized algorithm which with probability at least constructs two halfspaces on $\zoon$ whose intersection has threshold degree $\Theta(n).$ In Section \[sec:main\], we develop several refinements of Theorem \[thm:main\]. For example, we show that the intersection of two halfspaces on $\zoon$ requires a perceptron with $\exp\{\Theta(n)\}$ monomials, i.e., does not have a sparse sign-representation. We also give an essentially tight lower bound on the threshold degree of the intersection of a halfspace and a majority function, improving quadratically on the previous bound in [@sherstov09hshs].
In summary, unlike DNF formulas and read-once Boolean formulas, the intersection of two halfspaces does not admit a nontrivial sign-representation. Apart from computational learning theory, lower bounds on the threshold degree have played a key role in several works on circuit complexity [@paturi-saks94rational; @siu-roy-kailath94rational; @krause94depth2mod; @KP98threshold; @sherstov07ac-majmaj], Turing complexity classes [@aspnes91voting; @beigel91rational; @beigel94perceptrons], and communication complexity [@sherstov07ac-majmaj; @sherstov07quantum; @sherstov07symm-sign-rank; @RS07dc-dnf]. For this reason, we consider Theorem \[thm:main\] and the techniques used to obtain it to be of interest outside of computational learning.
Theorem \[thm:main\] and much previous work suggest that the nature of a PAC learning problem changes significantly when, instead of Valiant’s original arbitrary-distribution setting, one considers learning with respect to restricted distributions. For example, the uniform distribution on the sphere $\mathbb{S}^{n-1}$ or hypercube $\zoon$ allows the use of tools other than sign-representing polynomials, such as Fourier analysis. In particular, polynomial-time algorithms are known for the uniform-distribution learning of intersections of a constant number of halfspaces on the sphere [@blum-kannan97intersection-of-halfspaces; @Vempala:97] and hypercube [@KOS:02]. Furthermore, if membership queries are allowed, DNF formulas are known to be learnable in polynomial time with respect to the uniform distribution on the hypercube [@jackson94sieve].
Our Techniques {#our-techniques .unnumbered}
--------------
Let $f\wedge f$ denote the conjunction of two copies of a given Boolean function $f,$ each on an independent set of variables. It was shown in [@sherstov09hshs] that the threshold degree of $f\wedge f$ equals, up to a small multiplicative constant, the least degree of a rational function $R$ with $\|f-R\|_\infty\leq 1/3.$ With this characterization in hand, the equality $\degthr(f\wedge
f)=\Theta(\sqrt n)$ was derived in [@sherstov09hshs] by solving the rational approximation problem for the halfspace $$\begin{aligned}
f(x) = \sign{\left(1 +
\sum_{i=1}^{\sqrt n} \sum_{j=1}^{\sqrt n} 2^i x_{ij}\right)}.\end{aligned}$$ Unfortunately, the $\Theta(\sqrt n)$ barrier is fundamental to the analysis in [@sherstov09hshs]. To prove that in fact $D(n)=\Theta(n),$ we pursue a rather different approach.
The intuition behind our work is as follows. Let $\alpha_1,\alpha_2,\dots,\alpha_n$ be given nonzero integers, and let $f\colon\zoon\to\moo$ be a given Boolean function such that $f(x)$ is completely determined by the sum $\sum \alpha_ix_i.$ When approximating $f$ pointwise by polynomials and rational functions of a given degree, can one restrict attention to those approximants that are, like $f,$ functions of the sum $\sum\alpha_ix_i$ alone rather than the individual bits $x_1,x_2,\dots,x_n$? If true, this claim would dramatically simplify the analysis of the threshold degree of $f$ by reducing it to a univariate question. Minsky and Papert [@minsky88perceptrons] showed that the claim is indeed true in the highly special case $\alpha_1=\alpha_2=\cdots=\alpha_n.$ For the purposes of this paper, however, the nonzero coefficients $\alpha_1,\alpha_2,\dots,\alpha_n$ must be of increasing orders of magnitude and in particular must satisfy $$\begin{aligned}
\max_{i,j} \ABS{\frac{\alpha_i}{\alpha_j}}>\exp\{\Omega(n)\}.\end{aligned}$$ Minsky and Papert’s argument breaks down completely in this setting, and with good reason: coefficients $\alpha_1,\dots,\alpha_n$ are easily constructed [@beigel94perceptrons] for which the passage to univariate approximation increases the degree requirement from $1$ to $n.$
To overcome this difficulty, we use techniques from Fourier analysis and matrix perturbation theory. Specifically, we define an appropriate distribution on $n$-tuples $(\alpha_1,\dots,\alpha_n)$ and study the behavior of the sum $\sum\alpha_ix_i$ as the vector $x$ ranges over $\zoon.$ We prove that for a typical $n$-tuple $(\alpha_1,\dots,\alpha_n$) and any collection of sums $S\subset\ZZ$ of interest, the subset $X_S\subset\zoon$ that induces the sums in $S$ is highly random in that membership in $X_S$ is uncorrelated with any polynomial of degree up to $\Theta(n).$ With some additional work, this allows the sought passage to a univariate question. In particular, we are able to prove the existence of a halfspace $f\colon\zoon\to\moo$ such that any multivariate rational approximant for $f$ gives a univariate rational approximant for the sign function on $\{\pm1,\pm2,\pm3,\dots,\pm2^{\Theta(n)}\}$ with the same degree and error. The univariate question being well-understood, we infer that $f$ requires a rational function of degree $\Omega(n)$ for pointwise approximation within $1/3$ and hence $\degthr(f\wedge
f)\geq\Omega(n)$ by the characterization from [@sherstov09hshs].
Preliminaries {#sec:prelim}
=============
Notation. {#notation. .unnumbered}
---------
We will view Boolean functions as mappings $X\to\zoo$ or $X\to\moo$ for some finite set $X,$ where the output value $1$ corresponds to “true” in the former case and “false” in the latter. We adopt the following standard definition of the sign function: $$\begin{aligned}
\sign x =
{\begin{cases}
-1, &x<0, \\
0, &x=0, \\
1, &x>0.
\end{cases}}\end{aligned}$$ The complement of a set $S$ is denoted $\overline S.$ We denote the symmetric difference of sets $S$ and $T$ by $S\oplus T = (S\cap \overline T) \cup (\overline S
\cap T).$ For a finite set $X,$ the symbol $\Pcal(X)$ denotes the family of all $2^{\abs{X}}$ subsets of $X.$ For functions $f,g\colon X\to\Re$ on a finite set $X,$ we use the notation $$\begin{aligned}
\langle f, g\rangle = \frac1{\abs X} \sum_{x\in X} f(x) g(x).\end{aligned}$$ We let $\log x$ stand for the logarithm of $x$ to the base $2.$ The binary entropy function $H\colon[0,1]\to[0,1]$ is given by $H(p) = -p\log p -
(1-p)\log (1-p)$ and is strictly increasing on $[0,1/2].$ The following bound is well known [@jukna01extremal p. 283]: $$\begin{aligned}
\sum_{i=0}^k {n\choose i} \leq 2^{H(k/n)n}, &&
k=0,1,2,\dots,\lfloor n/2\rfloor.
\label{eqn:entropy-bound}\end{aligned}$$ For elements $x,y$ of a given set, we use the Kronecker delta $$\begin{aligned}
\delta_{x,y}
={\begin{cases} 1, & x=y, \\
0, &x\ne y.
\end{cases}}\end{aligned}$$ The symbol $P_k$ stands for the family of all univariate real polynomials of degree up to $k.$ The majority function $\MAJ_n\colon\zoon\to\moo$ has the usual definition: $$\begin{aligned}
\MAJ_n(x)
={\begin{cases}-1, & x_1+x_2+\cdots+x_n > n/2, \\
1, & \text{otherwise}.\end{cases}}\end{aligned}$$
Fourier transform. {#fourier-transform. .unnumbered}
------------------
Consider the vector space of functions $\zoon\to\Re,$ equipped with the inner product $$\begin{aligned}
\langle f,g\rangle = 2^{-n} \sum_{x\in\zoon}f(x)g(x).\end{aligned}$$ For $S\subseteq\oneton,$ define $\chi_S\colon \zoon\to\moo$ by $\chi_S(x) =(-1)^{\sum_{i\in S} x_i}.$ Then $\{\chi_S\}_{S\subseteq\oneton}$ is an orthonormal basis for the inner product space in question. As a result, every function $f\colon \zoon\to\Re$ has a unique representation of the form $$\begin{aligned}
f=\sum_{S\subseteq\oneton} \hat f(S)\chi_S,\end{aligned}$$ where $\hat
f(S)=\langle f,\chi_S\rangle$. The reals $\hat f(S)$ are called the *Fourier coefficients of $f.$* The orthonormality of $\{\chi_S\}$ immediately yields *Parseval’s identity*: $$\begin{aligned}
\sum_{S\subseteq\oneton} \hat f(S)^2
= \langle f,f\rangle = \Exp_{x\in\zoon}[f(x)^2]. \label{eqn:parsevals}\end{aligned}$$
Matrices. {#matrices. .unnumbered}
---------
The symbol $\Re^{m\times n}$ refers to the family of all $m\times n$ matrices with real entries. A matrix $A\in\Re^{n\times n}$ is called *strictly diagonally dominant* if $$\begin{aligned}
\abs{A_{ii}} >
\sum_{\substack{j=1\\j\ne i}}^n \abs{A_{ij}}, &&i=1,2,\dots,n.\end{aligned}$$ A well-known result in matrix perturbation theory, due to Gershgorin [@gershgorin31disks], states that the eigenvalues of a matrix lie in the union of certain disks in the complex plane centered around the diagonal entries of the matrix. We will need the following very special case, which corresponds to showing that the eigenvalues are all nonzero.
\[thm:gershgorin\] Let $A\in\Re^{n\times n}$ be strictly diagonally dominant. Then $A$ is nonsingular.
Fix a nonzero vector $x\in\Re^n$ and choose $i$ such that $\abs{x_i}=\| x\|_\infty.$ Then by strict diagonal dominance, $$\begin{aligned}
\abs{(Ax)_i} = \ABS{\sum_{j=1}^n A_{ij}x_j}
\geq \abs{A_{ii}} \|x\|_\infty -
\sum_{\substack{j=1\\j\ne i}}^n \abs{A_{ij}} \|x\|_\infty
>0,\end{aligned}$$ so that $Ax\ne 0.$
Rational approximation. {#rational-approximation. .unnumbered}
-----------------------
The degree of a rational function $p(x)/q(x),$ where $p$ and $q$ are polynomials on $\Re^n,$ is the maximum of the degrees of $p$ and $q.$ Consider a function $f\colon X\to\moo,$ where $X\subseteq \Re^n.$ For $d\geq0,$ define $$\begin{aligned}
R(f,d) \,= \,\inf_{\rule{0pt}{7pt}p,q} \,\sup_{x\in X}
\left\lvert f(x) - \frac{p(x)}{q(x)} \right\rvert,\end{aligned}$$ where the infimum is over multivariate polynomials $p$ and $q$ of degree up to $d$ such that $q$ does not vanish on $X.$ In words, $R(f,d)$ is the least error in an approximation of $f$ by a multivariate rational function of degree up to $d.$ A closely related quantity is $$\begin{aligned}
R^+(f,d) \,= \,\inf_{\rule{0pt}{7pt}p,q} \,\sup_{x\in X}
\left\lvert f(x) - \frac{p(x)}{q(x)} \right\rvert,\end{aligned}$$ where the infimum is over multivariate polynomials $p$ and $q$ of degree up to $d$ such that $q$ is positive on $X.$ These two quantities are related in a straightforward way: $$\begin{aligned}
R^+(f,2d) \leq R(f,d) \leq R^+(f,d). \end{aligned}$$ The second inequality here is trivial. The first follows from the fact that every rational approximant $p(x)/q(x)$ of degree $d$ gives rise to a degree-$2d$ rational approximant with the same error and a positive denominator, namely, $\{p(x)q(x)\}/q(x)^2.$
The infimum in the definitions of $R(f,d)$ and $R^+(f,d)$ cannot in general be replaced by a minimum [@rivlin-book], even when $X$ is finite subset of $\Re.$ This contrasts with the more familiar setting of a finite-dimensional normed linear space, where least-error approximants are guaranteed to exist.
For $S\subseteq\Re,$ we let $$\begin{aligned}
R^+(S,d) = \inf_{\rule{0pt}{7pt}p,q}\, \sup_{x\in S}
\left\lvert \sign x - \frac{p(x)}{q(x)} \right\rvert,\end{aligned}$$ where the infimum ranges over $p,q\in P_d$ such that $q$ is positive on $S.$ The study of the rational approximation of the sign function dates back to seminal work by Zolotarev [@zolotarev1877rational] in the late 19th century. A much later result due to Newman [@newman64rational] gives highly accurate estimates of $R^+([-n,-1]\cup[1,n],d)$ for all $n$ and $d.$ Newman’s work in particular provides upper bounds on $R^+(\{\pm1,\pm2,\dots,\pm n\}, d),$ which in [@sherstov09hshs] were sharpened and complemented with matching lower bounds to the following effect:
\[thm:R+sign\] Let $n,d$ be positive integers, $R=R^+(\{\pm1,\pm2,\dots,\pm n\},d).$ For $1\leq d\leq\log n,$ $$\begin{aligned}
\exp{\left\{-\Theta{\left(\frac1{n^{1/(2d)}}\right)}\right\}}
\leq R < \exp{\left\{-\frac1{n^{1/d}}\right\}}. \end{aligned}$$ For $\log n < d <n,$ $$\begin{aligned}
R=\exp{\left\{-\Theta{\left(\frac{d}{\log (2n/d)}\right)}\right\}}.\end{aligned}$$ For $d\geq n,$ $$\begin{aligned}
R=0.\end{aligned}$$
Theorem \[thm:R+sign\] has the following corollary [@sherstov09hshs Thm. 1.7], in which we adopt the notation $\rdeg_\eps(f)= \min\{d :
R^+(f,d)\leq\epsilon\}.$
\[thm:approx-maj\] Let $\MAJ_n\colon \zoon\to\moo$ denote the majority function. Then $$\begin{aligned}
\rdeg_\epsilon(\MAJ_n)=
{\begin{cases}
\displaystyle
\Theta{\left(
\log {\left\{\frac{2n}{\log(1/\epsilon)}\right\}}
\cdot
\log \frac1\epsilon
\right)},
&\qquad 2^{-n}<\epsilon<1/3,\\
\rule{0mm}{10mm}
\displaystyle
\Theta{\left(1 + \frac{\log n}{\log\{1/(1-\epsilon)\}}\right)},
&\qquad 1/3\leq \epsilon<1.
\end{cases}}\end{aligned}$$
Threshold degree. {#threshold-degree. .unnumbered}
-----------------
Let $f\colon X\to\moo$ be a given Boolean function, where $X\subset \Re^n$ is finite. The *threshold degree* of $f,$ denoted $\degthr(f),$ is the least degree of a polynomial $p(x)$ such that $f(x)\equiv\sign p(x).$ The term “threshold degree” appears to be due to Saks [@saks93slicing]. Equivalent terms in the literature include “strong degree” [@aspnes91voting], “voting polynomial degree” [@krause94depth2mod], “polynomial threshold function degree” [@odonnell03degree], and “sign degree” [@buhrman07pp-upp].
Given functions $f\colon X\to\moo$ and ${g}\colon
Y\to\moo,$ we let the symbol $f\wedge{g}$ stand for the function $X\times Y\to\moo$ given by $(f\wedge
{g})(x,y)=f(x)\wedge {g}(y).$ Note that in this notation, $f$ and $f\wedge f$ are completely different functions, the former having domain $X$ and the latter $X\times
X.$ An elegant observation, due to Beigel et al. [@beigel91rational], relates the notions of sign-representation and rational approximation for conjunctions of Boolean functions.
\[thm:rational-is-possible\] Let $f\colon X\to\moo$ and ${g}\colon Y\to\moo$ be given functions, where $X,Y\subseteq\Re^n.$ Let $d$ be an integer with $R^+(f,d) + R^+({g},d)<1.$ Then $$\begin{aligned}
\degthr(f\wedge {g}) \leq 2d.\end{aligned}$$
Consider rational functions $p_1(x)/q_1(x)$ and $p_2(y)/q_2(y)$ of degree at most $d$ such that $q_1$ and $q_2$ are positive on $X$ and $Y,$ respectively, and $$\begin{aligned}
\sup_{X} \left| f(x) - \frac{p_1(x)}{q_1(x)}\right| +
\sup_{Y} \left| {g}(y) - \frac{p_2(y)}{q_2(y)}\right| < 1. \end{aligned}$$ Then $$\begin{aligned}
f(x)\wedge {g}(y) \equiv \sign\{1 + f(x)+{g}(y)\}
\equiv \sign{\left\{ 1 + \frac{p_1(x)}{q_1(x)} +
\frac{p_2(y)}{q_2(y)} \right\}}.\end{aligned}$$ Multiplying the last expression by the positive quantity $q_1(x)q_2(y)$ gives $f(x)\wedge {g}(y) \equiv \sign\{ q_1(x)q_2(y) + p_1(x)q_2(y)
+ p_2(y)q_1(x)\}.$
We will also need a converse to Theorem \[thm:rational-is-possible\], proved in .
\[thm:sherstov-degthr-R\] Let $f\colon X\to\moo$ and ${g}\colon Y\to\moo$ be given functions, where $X,Y\subset\Re^n$ are arbitrary finite sets. Assume that $f$ and $g$ are not identically false. Let $d=\degthr(f\wedge g).$ Then $$\begin{aligned}
R^+(f,4d) + R^+(g,2d) < 1.\end{aligned}$$
Symmetric functions. {#symmetric-functions. .unnumbered}
--------------------
Let $S_n$ denote the symmetric group on $n$ elements. For $\sigma\in S_n$ and $x\in\zoon$, we denote $\sigma
x=(x_{\sigma(1)},\ldots,x_{\sigma(n)})\in\zoon.$ For $x\in\zoon,$ we define $|x|= x_1+x_2+\cdots+x_n.$ A function $\phi\colon \zoon\to\Re$ is called *symmetric* if $\phi(x) = \phi(\sigma x)$ for every $x\in\zoon$ and every $\sigma\in S_n.$ Equivalently, $\phi$ is symmetric if $\phi(x)$ is uniquely determined by $|x|.$ Symmetric functions on $\zoon$ are intimately related to univariate polynomials, as borne out by Minsky and Papert’s *symmetrization argument* [@minsky88perceptrons]:
Let $\phi\colon \zoon\to\Re$ be a polynomial of degree $d.$ Then there is a polynomial $p\in P_d$ such that $$\begin{aligned}
\Exp_{\sigma\in S_n} [\phi(\sigma x)] = p(|x|),
&&x\in\zoon.\end{aligned}$$
We will need the following consequence of Minsky and Papert’s technique for rational functions, pointed out in [@sherstov09hshs Prop. 2.7].
\[prop:symm-rational\] Let $n_1,\dots,n_k$ be positive integers. Consider a function $F\colon\zoo^{n_1}\times\cdots\times\zoo^{n_k}\to\moo$ such that $F(x_1,\dots,x_k)\equiv
f(\abs{x_1},\dots,\abs{x_k})$ for some $f\colon\{0,1,\dots,n_1\}\times\cdots\times\{0,1,\dots,n_k\}\to\moo.$ Then for all $d,$ $$\begin{aligned}
R^+(F,d) = R^+(f,d).\end{aligned}$$
Analysis of Random Halfspaces {#sec:analysis-of-random-hs}
=============================
In this section, we prove a certain structural property of random halfspaces. Specifically, we will fix integers $w_1,w_2,\dots,w_n$ at random from a suitable range and analyze the sum $$\begin{aligned}
\sum_{i=1}^n w_ix_i\end{aligned}$$ as $x$ ranges over $\zoon.$ Our objective will be to show that, for a typical choice of the weights $w_1,w_2,\dots,w_n,$ the distribution of this sum modulo $2^{\Theta(n)}$ is highly random. More precisely, we will show that the subset $X_s\subset\zoon$ that induces any particular sum $s$ modulo $2^{\Theta(n)}$ is relatively large and that membership in $X_s$ is almost uncorrelated with any polynomial of low degree. We start with a technical lemma.
\[lem:random-xor\] Let $f,g\colon\zoon\to\zoo$ be given functions. Fix an integer $k$ with $0\leq k\leq n/2.$ For a set $S\subseteq\oneton,$ define $F_S\colon\zoon\to\zoo$ by $$\begin{aligned}
F_S(x) = f(x) \wedge \left(g(x)\oplus \bigoplus_{i\in S}
x_i\right).\end{aligned}$$ Fix a real $\zeta >0.$ Then with probability at least $1-2^{-n+H(k/n)n +2\zeta n}$ over a uniformly random choice of $S\in\Pcal(\oneton),$ one has $$\begin{aligned}
\ABS{\hat F_S(T) - \frac12 \hat f(T)}
\leq 2^{-\zeta n-1}, && |T|\leq k.\end{aligned}$$
Define $\phi\colon\zoon\to[-1/2,1/2]$ by $\phi(x)=f(x)g(x) - \frac12 f(x).$ Define $\Scal\subseteq\Pcal(\oneton)$ by $\Scal = \{ S : \abs{\hat \phi(S)} \geq 2^{-\zeta n-1}\}.$ By Parseval’s identity (\[eqn:parsevals\]), $$\begin{aligned}
\abs{\Scal} \leq 4^{\zeta n}.
\label{eqn:A-B-bound}\end{aligned}$$ Since $F_S(x) = \frac12 f(x) + (-1)^{\sum_{i\in S}x_i}\phi(x),$ we have $$\begin{aligned}
\ABS{\hat F_S(T) - \frac12 \hat f(T)}
=\abs{\hat \phi(S\oplus T)},
&&S,T\subseteq\oneton.
\label{eqn:F-S-spectrum}\end{aligned}$$
For a uniformly random $S\in\Pcal(\oneton),$ the set $\{S\oplus
T: \abs{T}\leq k\}$ contains any fixed element of $\Pcal(\oneton)$ with probability $2^{-n}\sum_{i=0}^k{n\choose i}.$ By the union bound, we infer that $$\begin{aligned}
\Prob_S[\{S\oplus T:\abs{T}\leq k\}\cap \Scal \ne
\varnothing]
\leq \abs{\Scal}\,2^{-n}\sum_{i=0}^k {n\choose i},\end{aligned}$$ which in view of (\[eqn:entropy-bound\]) and (\[eqn:A-B-bound\]) is bounded from above by $2^{-n+H(k/n)n+2\zeta n}.$ This observation, along with (\[eqn:F-S-spectrum\]), completes the proof.
Using Lemma \[lem:random-xor\] and induction, we now obtain a key intermediate result.
\[lem:resheto\] Fix an integer $k\geq0$ and reals $\epsilon,\zeta\in(0,1/2).$ Choose sets $S_0,S_1,\dots,S_k\in\Pcal(\oneton)$ uniformly at random. Fix any integer $s$ and define $f\colon\zoon\to\zoo$ by $$\begin{aligned}
f(x)=1\qquad\Leftrightarrow\qquad \sum_{i=0}^k 2^i\sum_{j\in
S_i} x_j\equiv s \pmod{2^{k+1}}.
\label{eqn:f-modular}\end{aligned}$$ Then with probability at least $1-(k+1)2^{-n+H(\epsilon)n+2\zeta n}$ over the choice of $S_0,S_1,\dots,S_k,$ one has $$\begin{aligned}
&\ABS{\hat f(T) - \frac{\delta_{T,{\varnothing}}}{2^{k+1}}} \leq
2^{-\zeta n},
&&\abs{T}\leq \epsilon n.
\label{eqn:f-low-order}\end{aligned}$$
In view of the modular counting in (\[eqn:f-modular\]), one may assume that $0\leq s<2^{k+1}$ and therefore $s=\sum_{i=0}^k 2^ib_i$ for some $b_0,b_1,\dots,b_k\in\zoo.$ The proof of the lemma is by induction on $k$ for a fixed $s.$
The base case $k=0$ corresponds to $f(x) = \frac12 + \frac12(-1)^{b_0}\chi_{S_0}(x).$ One obtains (\[eqn:f-low-order\]) by conditioning on the event $|S_0|>\epsilon n,$ which in view of (\[eqn:entropy-bound\]) occurs with probability no smaller than $1-2^{-n + H(\epsilon)n}.$
We now consider the inductive step. Define $f'\colon\zoon\to\zoo$ by $$\begin{aligned}
f'(x)=1
\qquad\Leftrightarrow\qquad
\sum_{i=0}^{k-1} 2^i\sum_{j\in S_i} x_j\equiv
\sum_{i=0}^{k-1} 2^ib_i\pmod{2^{k}}.\end{aligned}$$ Let $E_1$ be the event, over the choice of $S_0,\dots,S_{k-1},$ that $\abs{\widehat
{f'}(T)-2^{-k}\delta_{T,{\varnothing}}} \leq
2^{-\zeta n}$ for $\abs{T}\leq \epsilon n.$ By the inductive hypothesis, $$\begin{aligned}
\Prob[E_1] \geq 1-k2^{-n+H(\epsilon)n + 2\zeta n}.
\label{eqn:inductive-assumption}\end{aligned}$$ Let $E_2$ be the event, over the choice of $S_0,\dots,S_k,$ that $\abs{\hat f(T) - \frac12 \widehat{f'}(T)}
\leq 2^{-\zeta n-1}$ for $\abs{T}\leq \epsilon n.$ In this terminology, it suffices to show that $$\begin{aligned}
\Prob[E_1\wedge E_2] \geq 1- (k+1)2^{-n+H(\epsilon)n +2\zeta n}.
\label{eqn:claim-rephrased}\end{aligned}$$ Observe that $$\begin{aligned}
f(x) = f'(x)\wedge
{\left(g(x)\oplus \bigoplus_{i\in S_k}^{\phantom{a}} x_i\right)},\end{aligned}$$ where $g\colon\zoon\to\zoo$ is the function such that $g(x)=1$ if and only if $b_k$ is the $(k+1)$st least significant bit of the integer $\sum_{i=0}^{k-1} 2^i
\sum_{j\in S_i} x_j.$ As a result, Lemma \[lem:random-xor\] shows that $\Prob[E_2]\geq 1-2^{-n+H(\epsilon)n +2\zeta n}.$ This bound, along with (\[eqn:inductive-assumption\]), settles (\[eqn:claim-rephrased\]) and thereby completes the induction.
We have reached the main result of this section.
\[thm:RESHETO\] Fix an integer $k\geq0$ and reals $\epsilon,\zeta\in(0,1/2).$ Choose integers $w_1,w_2,\dots,w_n$ uniformly at random from $\{0,1,\dots,2^{k+1}-1\}.$ For $s\in\ZZ,$ define $f_s\colon\zoon\to\zoo$ by $$\begin{aligned}
f_s(x)=1\qquad\Leftrightarrow\qquad \sum_{i=1}^n w_ix_i\equiv
s\pmod{2^{k+1}}.
\label{eqn:f_s}\end{aligned}$$ Then with probability at least $1-(k+1)2^{-n+H(\epsilon)n+2\zeta n+k+1}$ over the choice of $w_1,w_2,\dots,w_n,$ one has $$\begin{aligned}
&\ABS{\hat f_s(T) - \frac{\delta_{T,{\varnothing}}}{2^{k+1}}} \leq
2^{-\zeta n},
&&\abs{T}\leq \epsilon n, \quad s\in\ZZ.\end{aligned}$$
In view of the modular counting in (\[eqn:f\_s\]), it suffices to prove the theorem for $s\in\{0,1,\dots,2^{k+1}-1\}.$ The functions $f_s$ have the following equivalent definition: pick sets $S_0,S_1,\dots,S_k\in\Pcal(\oneton)$ uniformly at random and define $$\begin{aligned}
f_s(x)=1\qquad\Leftrightarrow\qquad \sum_{i=0}^k2^i\sum_{j\in
S_i} x_j \equiv s\pmod{2^{k+1}}.\end{aligned}$$ The proof is now complete by Lemma \[lem:resheto\] and the union bound over $s.$
Zeroing out Correlations by a Change of Distribution {#sec:change-of-distribution}
====================================================
Recall the setting of the previous section, where we fixed integers $w_1,w_2,\dots,w_n$ at random from a suitable range and analyzed the sum $
\sum_{i=1}^n w_ix_i
$ as $x$ ranged over $\zoon.$ We showed that the subset $X_s\subset\zoon$ that induces any particular sum $s$ modulo $2^{\Theta(n)}$ is relatively large and that membership in $X_s$ has *almost* zero correlation with any given polynomial of low degree. For the purposes of this paper, the correlations with low-degree polynomials need to be *exactly* zero. In this section we show that, with respect to a suitable distribution $\mu_s$ on each $X_s,$ membership in $X_s$ will indeed have zero correlation with any low-degree polynomial.
A starting point in our discussion is a general statement on zeroing out the correlations of given Boolean functions $\chi_1,\chi_2,\dots,\chi_k$ with another Boolean function $f.$ Recall that for functions $f,g\colon X\to\Re$ on a finite set $X,$ we use the notation $$\begin{aligned}
\langle f, g\rangle = \frac1{\abs X} \sum_{x\in X} f(x) g(x).\end{aligned}$$
\[thm:distribution-by-inversion\] Let $f,\chi_1,\dots,\chi_k\colon X\to\moo$ be given functions on a finite set $X.$ Suppose that $$\begin{aligned}
&\sum_{\substack{i=1}}^{k} \abs{\langle f,\chi_i\rangle}
<\frac12,
\label{eqn:f-chi-bounded}\\
&\sum_{\substack{j=1\\j\ne i}}^{k} \abs{\langle \chi_i,\chi_j\rangle} \leq
\frac12, &&i=1,2,\dots,k.
\label{eqn:diag-dominance}\\
\intertext{Then there exists a probability distribution $\mu$ on $X$ such
that}
&\Exp_\mu\,[f(x)\chi_i(x)] = 0, &&i=1,2,\dots,k.
\nonumber\end{aligned}$$
A comment is in order on the hypothesis of Theorem \[thm:distribution-by-inversion\]. The theorem states that if $\chi_1,\chi_2,\dots,\chi_k$ each have a small correlation with $f$ and, in addition, have small pairwise correlations, then a distribution exists with respect to which $f$ is completely uncorrelated with $\chi_1,\chi_2,\dots,\chi_k.$ The latter part of the hypothesis, namely the requirement (\[eqn:diag-dominance\]) of small pairwise correlations for $\chi_1,\chi_2,\dots,\chi_k,$ may seem unnecessary at first. In actuality, it is vital. Exponential lower bounds on the weights of linear perceptrons [@myhill-kautz61; @siu91small-weights] imply, by linear programming duality, the existence of functions $f,\chi_1,\chi_2,\dots,\chi_k\colon X\to\moo$ such that $\abs{\langle f,\chi_i\rangle} =
\exp\{-\Theta(k)\},$ $i=1,2,\dots,k,$ and yet $$\begin{aligned}
f(x) \equiv \sign {\left(\sum_{i=1}^k \alpha_i\chi_i(x)\right)}
\label{eqn:linear-comb}\end{aligned}$$ for some fixed reals $\alpha_1,\dots,\alpha_k.$ In this construction, the correlation of $f$ with each $\chi_i$ is small, in fact exponentially smaller than what is assumed in Theorem \[thm:distribution-by-inversion\]; nevertheless, the representation (\[eqn:linear-comb\]) rules out a distribution $\mu$ with respect to which $f$ could have zero correlation with each $\chi_i,$ for such a distribution $\mu$ would have to obey $$\begin{aligned}
0<\Exp_\mu\left[\left|\sum_{i=1}^k
\alpha_i\chi_i(x)\right|\right]=\Exp_\mu\left[f(x)\sum_{i=1}^k
\alpha_i\chi_i(x)\right]=\sum_{i=1}^k\alpha_i\Exp_\mu[f(x)\chi_i(x)]=0.\end{aligned}$$
Consider the linear system $$\begin{aligned}
M\alpha = \gamma
\label{eqn:lin-system}\end{aligned}$$ in the unknown $\alpha\in\Re^k,$ where $M=[\langle \chi_i,\chi_j\rangle]_{i,j}$ is a matrix of order $k$ and $\gamma = (\langle f,\chi_1\rangle, \dots, \langle f,\chi_k\rangle)\in\Re^k.$ Then (\[eqn:diag-dominance\]) shows that $M$ is strictly diagonally dominant and hence nonsingular by Theorem \[thm:gershgorin\]. Fix the unique solution $\alpha$ to the system (\[eqn:lin-system\]). Then $2\abs{\alpha_i} - \sum_{j=1}^{k} \abs{\alpha_j\langle
\chi_i,\chi_j\rangle} \leq \abs{\langle f,\chi_i\rangle}$ for $i=1,2,\dots,k.$ Summing these $k$ inequalities, we obtain $$\begin{aligned}
2\sum_{i=1}^k \abs{\alpha_i} - \sum_{j=1}^{k}\abs{\alpha_j}
\sum_{i=1}^k\abs{\langle \chi_i,\chi_j\rangle} \leq
\sum_{i=1}^k\abs{\langle f, \chi_i\rangle}, \end{aligned}$$ which in view of (\[eqn:f-chi-bounded\]) and (\[eqn:diag-dominance\]) shows that $\sum_{i=1}^k\abs{\alpha_i}<1.$ Therefore, the function $\mu\colon X\to\Re$ given by $$\begin{aligned}
\mu(x) = \epsilon{\left(
1 - f(x)\sum_{i=1}^k \alpha_i \chi_i(x)
\right)}\end{aligned}$$ is a probability distribution on $X$ for a suitable normalizing factor $\epsilon>0.$ At last, $$\begin{aligned}
\Exp_{\mu}\,[f(x)\chi_i(x)] =
\epsilon \abs{X} \left(
\langle f,\chi_i\rangle - \sum_{j=1}^k \alpha_j\langle
\chi_i,\chi_j\rangle \right) = 0,\end{aligned}$$ where the final equality holds by (\[eqn:lin-system\]).
We are now in a position to prove the main result of this section.
\[thm:mu\_s-constructed\] Let $\alpha>0$ be a sufficiently small absolute constant. Choose integers $w_1,w_2,\dots,w_n$ uniformly at random from $\{0,1,\dots,2^{\lfloor\alpha n\rfloor+1}-1\}.$ For $s\in\ZZ,$ define $$\begin{aligned}
\label{eqn:X_s}
X_s = \left\{
x\in\zoon : \sum_{i=1}^n w_ix_i \equiv s \pmod{2^{\lfloor\alpha
n\rfloor+1}}\;
\right\}. \end{aligned}$$ Then with probability at least $1-\e^{-n/3}$ over the choice of $w_1,w_2,\dots,w_n,$ there is a distribution $\mu_s$ on $X_s$ $($for each $s)$ such that $$\begin{aligned}
\label{eqn:mu_s-equivalence}
\Exp_{\mu_s}\, [p(x)] = \Exp_{\mu_t}\, [p(x)]\end{aligned}$$ for any $s,t\in\ZZ$ and any polynomial $p$ of degree at most $\lfloor \alpha n\rfloor.$
Let $\alpha>0$ be sufficiently small. We will assume throughout the proof that $n\geq 1/\alpha,$ the theorem being trivial otherwise. Set $\epsilon=2\alpha,$ $\zeta=1/5,$ and $k=\lfloor \alpha n\rfloor$ in Theorem \[thm:RESHETO\]. Then with probability at least $1-\e^{-n/3}$ over the choice of $w_1,w_2,\dots,w_n,$ one has $$\begin{aligned}
&\ABS{\hat f_s(T) - \frac{\delta_{T,{\varnothing}}}{2^{\lfloor \alpha
n\rfloor+1}}} \leq
2^{-n/5},
&&\abs{T}\leq 2\alpha n,\quad s\in\ZZ,
\label{eqn:f_s-guarantee}\end{aligned}$$ where $f_s\colon\zoon\to\zoo$ is given by $f_s(x)=1\Leftrightarrow x\in X_s.$ It follows that for each $s,$ $$\begin{aligned}
\label{eqn:X_s-size}
\abs{X_s} = 2^n\hat f_s({\varnothing}) \geq
2^n(2^{-\lfloor\alpha n\rfloor-1}-2^{-n/5}).\end{aligned}$$
For $f,g\colon\zoon\to\Re,$ we will write $\langle f,g\rangle_{X_s} = \abs{X_s}^{-1}\sum_{x\in X_s}f(x)g(x).$ Let $\Scal\subset\Pcal(\oneton)$ be the system of nonempty subsets of at most $\alpha n$ elements. Fix any $T\in \Scal.$ Then for each $s,$ $$\begin{aligned}
\sum_{\substack{S\in\Scal\\S\ne T}} \abs{\langle \chi_S,\chi_T\rangle_{X_s}}
=\frac{2^{n}}{\abs{X_s}}
\sum_{\substack{S\in\Scal\\S\ne T}} \abs{\hat f_s(S\oplus T)}
\leq \frac{2^{n}}{\abs{X_s}} \cdot\abs{\Scal}\, 2^{-n/5} <\frac12,
\label{eqn:cross-correlations}\end{aligned}$$ where the final two inequalities follow from (\[eqn:entropy-bound\]), (\[eqn:f\_s-guarantee\]), and (\[eqn:X\_s-size\]). Similarly, for each $s,$ $$\begin{aligned}
\sum_{S\in\Scal} \abs{\langle f_s,\chi_S\rangle_{X_s}}
=\frac{2^{n}}{\abs{X_s}}
\sum_{S\in\Scal} \abs{\hat f_s(S)}
\leq \frac{2^{n}}{\abs{X_s}} \cdot\abs{\Scal}\, 2^{-n/5} <\frac12.
\label{eqn:correlations-with-f_s}\end{aligned}$$ In view of (\[eqn:cross-correlations\]) and (\[eqn:correlations-with-f\_s\]), Theorem \[thm:distribution-by-inversion\] provides a distribution $\mu_s$ on $\zoon$ that is supported on $X_s$ and obeys $\hat\mu_s(S) = 0$ for $S\in\Scal.$ Since $\mu_s$ is a probability distribution, we additionally have $\hat\mu_s({\varnothing})=2^{-n}$ for all $s.$ In particular, the distributions $\mu_s$ have identical Fourier spectra up to coefficients of order $\alpha n,$ which is another way of stating (\[eqn:mu\_s-equivalence\]).
Reduction to a Univariate Problem {#sec:univariate-reduction}
=================================
Recall from the Introduction that the crux of our proof is to establish the existence of a halfspace $f\colon\zoon\to\moo$ that requires a rational function of degree $\Theta(n)$ for pointwise approximation within $1/3.$ The purpose of this section is to reduce this task, for a suitably chosen random halfspace, to a univariate problem. The univariate problem pertains to the uniform approximation of the sign function on the set $\{\pm1,\pm2,\pm3,\dots,\pm2^{\Theta(n)}\}$ and has been solved in previous work. Key to this univariate reduction will be the construction of probability distributions in the previous two sections.
\[thm:univariate-reduction-symm\] Put $k=\lfloor\alpha n\rfloor,$ where $\alpha>0$ is the absolute constant from Theorem . Choose $w_1,w_2,\dots,w_n$ uniformly at random from $\{0,1,\dots,2^{k+1}-1\}.$ Define $f\colon\zoon\times\{0,1,2,\dots,n\}\to\moo$ by $$\begin{aligned}
f(x,t) = \sign \left( \frac12 + \sum_{i=1}^n w_ix_i - 2^{k+1}t\right).\end{aligned}$$ Then with probability at least $1-\e^{-n/3}$ over the choice of $w_1,w_2,\dots,w_n,$ one has $$\begin{aligned}
R^+(f,d)\geq R^+(\{{\pm1,\pm2,\pm 3,\dots,\pm2^k}\},d), && d=0,1,\dots,k.\end{aligned}$$
For $s={\pm1,\pm2,\pm 3,\dots,\pm2^k},$ define $X_s\subseteq\zoon$ by (\[eqn:X\_s\]). Then by Theorem \[thm:mu\_s-constructed\], with probability at least $1-\e^{-n/3}$ there is a distribution $\mu_s$ on $X_s$ for each $s$ such that $$\begin{aligned}
\label{eqn:mu_s-equivalence-restated}
\Exp_{\mu_s}\, [p(x)] = \Exp_{\mu_r}\, [p(x)]\end{aligned}$$ for any $s,r\in\{{\pm1,\pm2,\pm 3,\dots,\pm2^k}\}$ and any polynomial $p$ of degree no greater than $k.$ In the remainder of the proof, we will work with a fixed choice of weights $w_1,w_2,\dots,w_n$ for which the described distributions $\mu_s$ exist.
Suppose that $R^+(f,d)<\epsilon$ where $0<\epsilon<1$ and $0\leq d\leq k.$ Then there are degree-$d$ polynomials $p,q$ on $\Re^n\times \Re$ such that on the domain of $f,$ $$\begin{aligned}
0<(1-\epsilon)q(x,t)\leq p(x,t)f(x,t)\leq (1+\epsilon)q(x,t).
\label{eqn:multivariate-approximant}\end{aligned}$$ On the support of $\mu_s$ (for $s={\pm1,\pm2,\pm 3,\dots,\pm2^k}$), the linear form $$\begin{aligned}
\ell(x,s) = 2^{-k-1}{\left(\sum_{i=1}^nw_ix_i-s\right)}\end{aligned}$$ obeys $\ell(x,s)\in\{0,1,2,\dots,n\}$ and $f(x,\ell(x,s))=\sign s.$ Letting $t=\ell(x,s)$ in (\[eqn:multivariate-approximant\]) and passing to expectations, $$\begin{aligned}
0<\Exp_{x\sim\mu_s}
\left[q(x,\ell(x,s))\right]
(1-\epsilon)
\leq &\Exp_{x\sim\mu_s}
\left[p(x,\ell(x,s))\right]\sign s\\
\leq &\Exp_{x\sim\mu_s}
\left[q(x,\ell(x,s))\right]
(1+\epsilon).\end{aligned}$$ It follows from (\[eqn:mu\_s-equivalence-restated\]) that $
\Exp_{\mu_s} [p(x,\ell(x,s))] = P(s)$ and $\Exp_{\mu_s}[q(x,\ell(x,s))] = Q(s)$ for some $P,Q\in P_d$ and all $s.$ As a result, $R^+(\{{\pm1,\pm2,\pm 3,\dots,\pm2^k}\},d)\leq\epsilon,$ the approximant in question being $P/Q.$
It remains to rewrite the previous theorem in terms of functions on the hypercube $\zoo^{2n}$ rather than the set $\zoon\times\{0,1,2,\dots,n\}.$
\[thm:univariate-reduction\] Put $k=\lfloor\alpha n\rfloor,$ where $\alpha>0$ is the absolute constant from Theorem . Choose $w_1,w_2,\dots,w_n$ uniformly at random from $\{0,1,\dots,2^{k+1}-1\}.$ Define $f\colon\zoo^{2n}\to\moo$ by $$\begin{aligned}
f(x) = \sign \left( \frac12 + \sum_{i=1}^n w_ix_i -
2^{k+1}\sum_{i=n+1}^{2n} x_i\right).\end{aligned}$$ Then with probability at least $1-\e^{-n/3}$ over the choice of $w_1,w_2,\dots,w_n,$ one has $$\begin{aligned}
R^+(f,d)\geq R^+(\{{\pm1,\pm2,\pm 3,\dots,\pm2^k}\},d), && d=0,1,\dots,k.\end{aligned}$$
Immediate from Proposition \[prop:symm-rational\] and Theorem \[thm:univariate-reduction-symm\].
Main Result and Generalizations {#sec:main}
===============================
We now combine the newly obtained result on rational approximation with known results from Section \[sec:prelim\] to prove the main theorem of this work.
\[thm:main-detailed\] Fix sufficiently small absolute constants $\alpha>0$ and $\beta=\beta(\alpha)>0.$ Choose integers $w_1,w_2,\dots,w_n\in
\{0,1,\dots,2^{\lfloor\alpha n\rfloor+1}-1\}$ uniformly at random. Then with probability at least $1-\e^{-n/3},$ the function $f\colon \zoo^{2n}\to\moo$ given by $$\begin{aligned}
f(x) = \sign\left(\frac12 + \sum_{i=1}^n w_ix_i -
2^{\lfloor\alpha n\rfloor +1}\sum_{i=n+1}^{2n} x_i\right)\end{aligned}$$ obeys $$\begin{aligned}
\degthr(f\wedge f) \geq \lfloor \beta n\rfloor.
\label{eqn:main-beta}\end{aligned}$$
Theorem \[thm:univariate-reduction\] shows that with probability at least $1-\e^{-n/3}$ over the choice of $w_1,w_2,\dots,w_n,$ one has $$\begin{aligned}
R^+(f,d)\geq
R^+(S,d),
&& d=0,1,\dots,\lfloor\alpha n\rfloor,
\label{eqn:univariate-reduction}\end{aligned}$$ where $S=\{\pm1,\pm2,\pm3,\dots,\pm2^{\lfloor\alpha n\rfloor}\}$ and $\alpha>0$ is the absolute constant from Theorem \[thm:mu\_s-constructed\]. In the remainder of the proof, we will condition on this event.
Suppose now that $\degthr(f\wedge f)<\lfloor \beta n\rfloor,$ where $\beta$ is a constant to be chosen later subject to $0<\beta<\alpha/4.$ Then Theorem \[thm:sherstov-degthr-R\] implies that $R^+(f,\lfloor4\beta n\rfloor)<1/2,$ which in view of (\[eqn:univariate-reduction\]) leads to $R^+(S,\lfloor4\beta
n\rfloor) <1/2.$ The last inequality violates Theorem \[thm:R+sign\] for small enough $\beta>0.$ Thus, (\[eqn:main-beta\]) holds for $\beta$ small enough.
Recall that the technical crux of this paper is an optimal lower bound for the rational approximation of a halfspace. We will have occasion to appeal to this result again, and for this reason we formulate it as a theorem in its own right.
\[thm:R-hn\] A family of halfspaces $h_n\colon\zoon\to\moo,$ $n=1,2,3,\dots,$ exists such that $$\begin{aligned}
R^+(h_n,d) = 1-\exp{\left\{-\Theta{\left(\frac nd\right)}\right\}}
, \qquad d=1,2,\dots,\Theta(n).
\label{eqn:R-hn}\end{aligned}$$
The lower bound in (\[eqn:R-hn\]) is immediate from Theorem \[thm:univariate-reduction\] and the univariate lower bounds in Theorem \[thm:R+sign\].
Next, every halfspace $h_n\colon\zoon\to\moo$ constructed in Theorem \[thm:univariate-reduction\] trivially obeys $R^+(h_n,1)<1-\exp\{-\Theta(n)\}.$ For $0<\xi<1,$ Newman’s classical work [@newman64rational] shows that $R^+([-1,-\xi]\cup[\xi,1],d)\leq 1-\xi^{\Theta(1/d)},$ whence by composition of the approximants one obtains the upper bound in (\[eqn:R-hn\]).
Mixed intersection. {#mixed-intersection. .unnumbered}
-------------------
Theorem \[thm:main-detailed\] shows that the intersection of two halfspaces has the asymptotically highest threshold degree. At the same time, Beigel et al. [@beigel91rational] showed that the intersection of a constant number of majority functions on $\zoon,$ which are particularly simple halfspaces, has threshold degree $O(\log n).$ We now derive a lower bound of $\Omega(\sqrt{n \log n})$ on the threshold degree of the intersection of a halfspace and a majority function, which improves quadratically on the previous bound in [@sherstov09hshs] and essentially matches the upper bound, $O(\sqrt n\log n),$ given below in Remark \[rem:h-majn\].
\[thm:hmajn\] A family of halfspaces $h_n\colon\zoon\to\moo,$ $n=1,2,3,\dots,$ exists such that $$\begin{aligned}
\degthr(h_n\wedge \MAJ_n) = \Theta(\sqrt {n\log n}).
\label{eqn:degthr-h-maj}\end{aligned}$$
The lower bound in (\[eqn:degthr-h-maj\]) is immediate from Theorems \[thm:approx-maj\], \[thm:sherstov-degthr-R\], and \[thm:R-hn\]. The upper bound in (\[eqn:degthr-h-maj\]) is immediate from Theorems \[thm:approx-maj\], \[thm:rational-is-possible\], and \[thm:R-hn\].
\[rem:h-majn\] The construction of Theorem \[thm:hmajn\] is essentially best possible in that every sequence of halfspaces $h_n\colon\zoon\to\moo,$ $n=1,2,3,\dots,$ obeys $$\begin{aligned}
\degthr(h_n\wedge\MAJ_n)=O(\sqrt n\log n).
\label{eqn:general-hmajn-upper}\end{aligned}$$ To derive this upper bound, recall that $R^+(h_n,1)<1-\exp\{-\Theta(n\log n)\}$ for every halfspace $h_n\colon\zoon\to\moo,$ by a classical result due to Muroga [@muroga71threshold]. Since $R^+([-1,-\xi]\cup[\xi,1],d)<1-\xi^{\Theta(1/d)}$ for $0<\xi<1$ by Newman [@newman64rational], we obtain by composition of approximants that $R^+(h_n,d)<1-\exp\{-\Theta(\{n\log n\}/d)\}.$ This settles (\[eqn:general-hmajn-upper\]) in view of Theorems \[thm:approx-maj\] and \[thm:rational-is-possible\].
Threshold density. {#threshold-density. .unnumbered}
------------------
In addition to threshold degree, several other complexity measures are of interest when sign-representing Boolean functions by real polynomials. One such complexity measure is *density,* i.e., the least $k$ for which a given function can be sign-represented by a linear combination of $k$ parity functions. Formally, for a given function $f\colon\zoon\to\moo,$ the *threshold density* $\dns(f)$ is the minimum size $\abs{\Scal}$ of a family $\Scal\subseteq\Pcal(\oneton)$ such that $$\begin{aligned}
f(x) \equiv \sign{\left(\sum_{S\in\Scal}^{\phantom{S\in\Scal}}
\lambda_S \chi_S(x)\right)}\end{aligned}$$ for some reals $\lambda_S,$ $S\in\Scal.$ It is clear from the definition that $\dns(f)\leq 2^n$ for all functions $f\colon\zoon\to\moo,$ and we will show that the intersection of two halfspaces on $\zoon$ has threshold density $2^{\Theta(n)}.$
To this end, we recall an elegant technique for converting Boolean functions with high threshold degree into Boolean functions with high threshold density, due to Krause and Pudl[á]{}k [@krause94depth2mod Prop. 2.1]. Their construction sends a function $f\colon\zoon\to\moo$ to the function $f^{\text{\rm KP}}\colon(\zoon)^3\to\moo$ given by $$\begin{aligned}
f^{\text{\rm KP}}(x,y,z) =
f(\dots, (\overline{z_i}\wedge x_i)\vee(z_i\wedge y_i), \dots).\end{aligned}$$
\[thm:degree-length\] Every function $f\colon\zoon\to\moo$ obeys $$\begin{aligned}
\dns(f^{\text{\rm KP}}) \geq 2^{\degthr(f)}.\end{aligned}$$
We are now in a position to obtain the desired density results.
\[thm:dns\] A family of halfspaces $h_n\colon\zoon\to\moo,$ $n=1,2,3,\dots,$ exists such that $$\begin{aligned}
\dns(h_n\wedge h_n) &\geq \exp\{\Theta(n)\},
\label{eqn:hh}\\
\dns(h_n\wedge \MAJ_n) &\geq \exp\{\Theta(\sqrt {n\log n})\}.
\label{eqn:hmajn}\end{aligned}$$
The parity of several parity functions is another parity function. As a result, $$\begin{aligned}
\max_{h_n}\{\dns(h_n\wedge h_n)\} \geq \max_F \{\dns(F\wedge F)\},
\label{eqn:h-F}\end{aligned}$$ where the maximum on the left is over all halfspaces $h_n\colon\zoon\to\moo$ and the maximum on the right is over arbitrary functions $F\colon\zoo^m\to\moo$ (for arbitrary $m$) such that $\dns(F)\leq n.$ For each $n=1,2,3,\dots,$ Theorem \[thm:main-detailed\] ensures the existence of a halfspace $f_n\colon\zoon\to\moo$ with $\degthr(f_n\wedge
f_n)\geq\Omega(n).$ By Theorem \[thm:degree-length\], the function $(f_n\wedge f_n)^{\text{\rm KP}}={f_n}^{\text{\rm KP}}\wedge {f_n}^{\text{\rm KP}}$ has threshold density $\exp\{\Omega(n)\}.$ Since $\dns({f_n}^{\text{\rm KP}})\leq
4n+1,$ the right member of (\[eqn:h-F\]) is at least $\exp\{\Omega(n)\}.$
This completes the proof of (\[eqn:hh\]). The proof of (\[eqn:hmajn\]) is closely analogous, with Theorem \[thm:hmajn\] used instead of Theorem \[thm:main-detailed\].
The lower bounds in Theorem \[thm:dns\] are essentially optimal. Specifically, (\[eqn:hh\]) is tight for trivial reasons, whereas the lower bound (\[eqn:hmajn\]) nearly matches the upper bound of $\exp\{\Theta(\sqrt n\log^2n)\}$ that follows from (\[eqn:general-hmajn-upper\]).
We also note that Theorem \[thm:degree-length\] readily generalizes to linear combinations of conjunctions rather than parity functions. In other words, if a function $f\colon\zoon\to\moo$ has threshold degree $d$ and $f^{\text{\rm KP}}(x,y,z)\equiv \sign(\sum_{i=1}^N\lambda_i
T_i(x,y,z))$ for some conjunctions $T_1,\dots,T_N$ of the literals $x_1,y_1,z_1,\dots,x_n,y_n,z_n,$ $\neg x_1,\neg
y_1,\neg z_1,\dots,\neg x_n,\neg y_n,\neg z_n,$ then $N\geq
2^{\Omega(d)}.$ With this remark in mind, Theorem \[thm:dns\] and its proof readily carry over to this alternate definition of density.
Acknowledgments {#acknowledgments .unnumbered}
===============
The author is thankful to Adam Klivans, Ryan O’Donnell, Rocco Servedio, and the anonymous reviewers for their feedback on this manuscript.
[^1]: $^*$ Microsoft Research, Cambridge, MA 02142. Email: [[email protected]]([email protected]).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, three topics on semi-selfdecomposable distributions are studied. The first one is to characterize semi-selfdecomposable distributions by stochastic integrals with respect to Lévy processes. This characterization defines a mapping from an infinitely divisible distribution with finite $\log$-moment to a semi-selfdecomposable distribution. The second one is to introduce and study a Langevin type equation and the corresponding Ornstein-Uhlenbeck type process whose limiting distribution is semi-selfdecomposable. Also, semi-stationary Ornstein-Uhlenbeck type processes with semi-selfdecomposable distributions are constructed. The third one is to study the iteration of the mapping above. The iterated mapping is expressed as a single mapping with a different integrand. Also, nested subclasses of the class of semi-selfdecomposable distributions are considered, and it is shown that the limit of these nested subclasses is the closure of the class of semi-stable distributions.'
address:
- 'Makoto Maejima: Department of Mathematics, Keio University, 3-14-1, Hiyoshi, Kohoku-ku, Yokohama 223-8522, Japan'
- 'Yohei Ueda: Department of Mathematics, Keio University, 3-14-1, Hiyoshi, Kohoku-ku, Yokohama 223-8522, Japan'
author:
- Makoto Maejima
- Yohei Ueda
title: |
Stochastic integral characterizations of semi-selfdecomposable distributions and\
related Ornstein-Uhlenbeck type processes
---
Introduction
============
Let $I({{\mathbb R^d}})$ be the class of all infinitely divisible distributions on ${{\mathbb R^d}}$ and let $\{X_t^{(\mu)},t\geq 0\}$ be an ${{\mathbb R^d}}$-valued Lévy process with $\mu\in I({{\mathbb R^d}})$ as its distribution at time 1. Many subclasses of $I({{\mathbb R^d}})$ have recently been investigated in many aspects. Among those, there are characterizations of those classes in terms of stochastic integrals with respect to Lévy processes. In such cases, we define mappings $$\Phi_f(\mu) = {\mathcal L}\left ( {\int_0^{\infty}}f(t)dX_t^{(\mu)}\right ) ,\quad \mu\in\mathfrak D(\Phi_f)\subset I({{\mathbb R^d}})$$ for nonrandom measurable functions $f\colon [0,\infty)\rightarrow {\mathbb{R}}$, where ${\mathcal L}(X)$ is the law of a random variable $X$ and $\mathfrak D (\Phi_f)$ is the domain of a mapping $\Phi_f$ that is the class of $\mu\in I({{\mathbb R^d}})$ for which ${\int_0^{\infty}}f(t)dX_t^{(\mu)}$ is definable. For the definition of stochastic integrals with respect to Lévy processes of nonrandom measurable functions, see the next section. When we consider the composition of two mappings $\Phi_f$ and $\Phi_g$, denoted by $\Phi_g\circ\Phi_f$, the domain of $\Phi_g\circ\Phi_f$ is $\mathfrak D (\Phi_g\circ \Phi_f) =
\{ \mu\in I({{\mathbb R^d}}) : \mu\in \mathfrak D (\Phi_f)\,\,
\text{and}\,\, \Phi_f(\mu)\in \mathfrak D (\Phi_g)\}$. Once we define such a mapping, we can characterize a subclass of $I({{\mathbb R^d}})$ as the range of $\Phi_f$, $\mathfrak R (\Phi_f)$, say. Among such classes, there are the Jurek class, the class of selfdecomposable distributions, the Goldie-Steutel-Bondesson class, the Thorin class, the class of generalized type $G$ distributions and so on. (For details on these, see, e.g., @MaejimaSato2009.) Also, by iterating a mapping $\Phi_f$, we can define a sequence of nested subclasses $\mathfrak R (\Phi_f^m), m\in{\mathbb{N}}$, where $\Phi_f^m$ is the $m$ times composition of the mapping $\Phi_f$ itself.
The class of selfdecomposable distributions, denoted by $L({{\mathbb R^d}})$, has the longest history in the study of subclasses of $I({{\mathbb R^d}})$. Let ${\widehat}\mu (z), z\in{{\mathbb R^d}}$, be the characteristic function of $\mu$. $\mu\in I({{\mathbb R^d}})$ is said to be selfdecomposable if for any $b>1$, there exists a distribution $\rho_b$ such that ${\widehat}\mu (z) = {\widehat}\mu (b^{-1}z){\widehat}{\rho}_b (z)$. This $\rho_b$ automatically belongs to $I({{\mathbb R^d}})$. $\mu\in L({{\mathbb R^d}})$ is also a limiting distribution of normalized partial sums of independent random variables under infinitesimal condition, and has the stochastic integral representation with respect to a Lévy process, which is ${\mathcal L}\left ({\int_0^{\infty}}e^{-t}dX_t\right)$ with ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$, where $I_{\log}({{\mathbb R^d}}) = \{ \mu\in I({{\mathbb R^d}}) \colon \int _{{{\mathbb R^d}}} \log ^+|x| \mu (dx) <\infty \}$, $\log^+|x|= (\log|x|)\vee 0$, and $|x|$ is the Euclidean norm of $x\in{{\mathbb R^d}}$. Furthermore, $\mu\in L({{\mathbb R^d}})$ is the limiting distribution of the solution of a Langevin equation with Lévy noise. More precisely, let $\{X_t, t\ge 0\}$ be a Lévy process on ${{\mathbb R^d}}$, $c\in{\mathbb{R}}$, and let $M$ be an ${{\mathbb R^d}}$-valued random variable. The Langevin equation is $$\label{lang}
Z_t = M + X_t -c \int_0^t Z_sds, \quad t\geq 0,$$ and the following is known, (see, e.g., @Sato's_book2003). $$Z_t = e^{-ct}M + e^{-ct}\int_0^te^{cs}dX_s, \quad t\ge 0,$$ is an almost surely unique solution of , and if $c>0$, $E\left[\log ^+|X_1|\right ] <\infty$, and $M$ is independent of $\{X_t\}$, then ${\mathcal L}(Z_t)\to \mu \in L({{\mathbb R^d}})$ as $t\rightarrow \infty $. We also know that, for a fixed $c>0$, the equation $$\label{lang_stationary}
Z_t-Z_s=X_t-X_s-c\int_s^t Z_udu, \quad -\infty<s\leq t<\infty,$$ has an almost surely unique stationary solution $$Z_t=\int_{-\infty}^te^{-c(t-u)}dX_u,\quad t\in{\mathbb{R}},$$ where $\{X_t,t\in{\mathbb{R}}\}$ is a h.-i.s.r.m.-process (whose precise definition is given in Section 2) satisfying $E\left[\log ^+|X_1|\right ] <\infty$. (See, e.g., @Sato's_book2003 or @MaejimaSato2003.) This stationary solution fulfills that ${\mathcal L}(Z_t)={\mathcal L}\left(\int_0^\infty e^{-cu}dX_u\right)
\in L({{\mathbb R^d}})$ for all $t\in{\mathbb{R}}$. Also, it is recognized that some selfdecomposable distributions on ${\mathbb{R}}$ are very important in the area of mathematical finance, (see @Yor2007).
In @MaejimaNaito1998, the concept of the selfdecomposability was extended to the semi-selfdecomposability. Here, $\mu\in I({{\mathbb R^d}})$ is called [*semi-selfdecomposable*]{} if there exist $b>1$ and $\rho\in I({{\mathbb R^d}})$ such that ${\widehat}\mu (z) = {\widehat}\mu (b^{-1}z){\widehat}\rho(z)$. We call this $b$ a span of $\mu$, and we denote the class of all semi-selfdecomposable distributions with span $b$ by $L(b, {{\mathbb R^d}})$. From the definitions, $L(b,{{\mathbb R^d}})\supsetneqq L({{\mathbb R^d}})$ and $L({{\mathbb R^d}}) = \bigcap _{b>1}L(b,{{\mathbb R^d}})$. $\mu\in L(b,{{\mathbb R^d}})$ is also realized as a limiting distribution of normalized partial sums of independent random variables under infinitesimal condition when the limit is taken through a geometric subsequence. A typical example is a semi-stable distribution, where $\mu\in I({{\mathbb R^d}})$ is said to be semi-stable with span $b$ if there exist $a>1$ and $c\in{{\mathbb R^d}}$ satisfying ${\widehat}\mu(z)^a={\widehat}\mu(bz)e^{i{\langle}c,z{\rangle}}$. Recently, several natural examples of semi-selfdecomposable distributions have appeared in the literature. We will mention some of them in the next section.
In @MaejimaSato2003, they gave a stochastic integral characterization of $\mu\in L(b,{{\mathbb R^d}})$ in terms of, not Lévy process, but natural semi-Lévy process. Here a semi-Lévy process with period $p>0$ is an additive process with periodically stationary increments with period $p$ and natural additive process was defined in @Sato2004 as semimartingale additive process in terms of the Lévy-Khintchine triplet. Namely, they showed that for each $b>1$, $\mu\in L(b,{{\mathbb R^d}})$ if and only if $\mu={\mathcal L}\left( {\int_0^{\infty}}e^{-t}dX_t\right )$, where $\{X_t\}$ is a semi-Lévy process with period $p=\log b$ and ${\mathcal L}(X_p)\in I_{\log}({{\mathbb R^d}})$. Our first topic of this paper is to give a stochastic integral characterization of $\mu\in L(b,{{\mathbb R^d}})$ in terms of Lévy process. If all natural semi-Lévy processes can be expressed as stochastic integrals with respect to Lévy processes, this problem is trivial from a result in @MaejimaSato2003 just mentioned now. However, as we will see in Example \[example\] later, it is not the case. Once we could solve this problem, we would define a mapping $\Phi_b$ from $\mathfrak D (\Phi_b)$ into $I({{\mathbb R^d}})$ and we can enjoy many stories similar to those about $L({{\mathbb R^d}})$. For instance, we can characterize $L(b,{{\mathbb R^d}})$ as the range of the mapping $\Phi_b$, that is, $L(b, {{\mathbb R^d}}) = \Phi_b(I_{\log}({{\mathbb R^d}}))$.
Our second topic is to construct and study a Langevin type equation and the corresponding Ornstein-Uhlenbeck type processes related to semi-selfdecomposable distributions, which are analogies of and in the case of selfdecomposable distributions, not in terms of semi-Lévy processes given in @MaejimaSato2003, but in terms of Lévy processes. Namely, we introduce a Langevin type equation and give its unique solution, which we call an Ornstein-Uhlenbeck type process. We then show that the limit of the Ornstein-Uhlenbeck type process exists in law, when the noise process has finite $\log$-moment, and the limiting distribution is semi-selfdecomposable. We also construct semi-stationary Ornstein-Uhlenbeck type process whose marginal distributions are semi-selfdecomposable.
Our third topic is to look for the ranges of the iterated mappings $\Phi_b^m$ and its limit. In @MaejimaNaito1998, the nested subclasses of $L(b,{{\mathbb R^d}})$, $L_m(b,{{\mathbb R^d}}), m\in{\mathbb{Z}}_+$, are defined as follows: $\mu \in L_m(b,{{\mathbb R^d}})$ if and only if there exists $\rho\in L_{m-1}(b,{{\mathbb R^d}})$ such that ${\widehat}\mu (z) = {\widehat}\mu (b^{-1}z){\widehat}\rho (z)$, where $L_0(b, {{\mathbb R^d}}) = L(b,{{\mathbb R^d}})$. Then we will show $$\label{nested_subclass_of_ssd}
L_m(b, {{\mathbb R^d}}) = \Phi^{m+1} _b\left(I_{\log ^{m+1}}({{\mathbb R^d}})\right),\quad m\in {\mathbb{Z}}_+,$$ where $I_{\log^{m+1}}({{\mathbb R^d}}) = \{ \mu\in I({{\mathbb R^d}}) \colon \int _{{{\mathbb R^d}}}
(\log ^+|x|)^{m+1} \mu (dx) <\infty \}$. The relation implies that the limit of these nested subclasses is the closure of the class of semi-stable distributions, where the closure is taken under convolution and weak convergence.
Organization of this paper is the following. In Section 2, we explain some notation and give preliminaries and some examples of semi-selfdecomposable distributions. In Section 3, the first topic is considered. In Sections 4–6, we study the second topic. Finally, in Section 7, we treat the third topic.
Notation, preliminaries and examples
====================================
In this section, we explain necessary notation, and give some preliminaries and examples.
Let $J$ be ${\mathbb{R}}$ or $[0,\infty)$, and ${\mathcal{B}}_J^0$ the class of all bounded Borel sets in $J$. An ${{\mathbb R^d}}$-valued independently scattered random measure (abbreviated as i.s.r.m.) $X=\{X(B),B\in{\mathcal{B}}_J^0\}$ is said to be homogeneous if ${\mathcal L}(X(B))={\mathcal L}(X(B+a))$ for all $B\in{\mathcal{B}}_J^0$ and $a\in{\mathbb{R}}$ satisfying $B+a\in{\mathcal{B}}_J^0$. See @MaejimaSato2003 and @Sato2004 [@Sato2006a], for the definition and deep study of stochastic integrals of nonrandom measurable functions $f\colon J\to{\mathbb{R}}$ with respect to ${{\mathbb R^d}}$-valued i.s.r.m.’s $X$, denoted by $\int_B f(s)X(ds), B\in{\mathcal{B}}_J^0$. For a fixed $t_0\in J$, we use the symbol $$\int_{t_0}^t f(s)X(ds)=
\begin{cases}
\int_{(t_0,t]} f(s)X(ds),&\text{for }t\in (t_0,\infty),\\
0,&\text{for }t=t_0,\\
-\int_{(t,t_0]} f(s)X(ds),&\text{for }t\in J\cap(-\infty,t_0),
\end{cases}$$ which is understood to be a càdlàg modification, (see Remark 3.16 of @MaejimaSato2003). If $\{X_t,t\geq 0\}$ is a Lévy process, then there exists a unique ${{\mathbb R^d}}$-valued homogeneous i.s.r.m.$X$ over $[0,\infty)$ satisfying $X_t=X([0,t])$ a.s.for each $t\geq 0$. Then $\int_0^tf(s)dX_s$ is defined by $\int_0^tf(s)X(ds)$ for $t\in[0,\infty)$. The improper stochastic integral ${\int_0^{\infty}}f(s)dX_s$ is defined as the limit in probability of $\int_0^tf(s)dX_s$ as $t\to\infty$ whenever the limit exists. See also @Sato2006b. In this paper, we say that a stochastic process $\{X_t,t\in{\mathbb{R}}\}$ on ${{\mathbb R^d}}$ is a *h.-i.s.r.m.-process* if there exists an ${{\mathbb R^d}}$-valued homogeneous i.s.r.m.$X$ over ${\mathbb{R}}$ such that $X_t=\int_0^tX(du),t\in{\mathbb{R}}$. We define a stochastic integral $\int_s^tf(u)dX_u,-\infty<s\leq t<\infty$ of a nonrandom measurable function $f\colon{\mathbb{R}}\rightarrow {\mathbb{R}}$ with respect to this process $\{X_t, t\in{\mathbb{R}}\}$ by $\int_s^tf(u)X(du)$. See also @Sato's_book2003 and @MaejimaSato2003. The improper stochastic integral $\int_{-\infty}^tf(u)dX_u$ is defined as the limit in probability of $\int_s^tf(u)dX_u$ as $s\rightarrow -\infty$, provided that this limit exists. If the improper stochastic integral $\int_{-\infty}^tf(u)dX_u$ is definable for $t\in{\mathbb{R}}$, then we regard it as a càdlàg process, since such a modification always exists.
Throughout this paper, we use the Lévy-Khintchine representation of the characteristic function of $\mu\in I({\mathbb{R}}^d)$ in the following way: $$\widehat{\mu}(z)=\exp\left\{-\frac 12 {\langle}z,Az{\rangle}+i{\langle}\gamma,z{\rangle}+\int_{{\mathbb R^d}}\left(e^{i{\langle}z,x{\rangle}}-1-\frac{i{\langle}z,x{\rangle}}{1+|x|^2}\right)\nu(dx)\right\},\quad z\in{{\mathbb R^d}},$$ where ${\langle}\cdot,\cdot{\rangle}$ denotes Euclidean inner product on ${{\mathbb R^d}}$, $A$ is a nonnegative-definite symmetric $d\times d$ matrix, $\gamma\in{{\mathbb R^d}}$, and $\nu$ is a measure, called Lévy measure, satisfying $\nu(\{0\})=0$ and $\int_{{\mathbb R^d}}(|x|^2\wedge 1)\nu(dx)<\infty$. We call $(A,\nu,\gamma)$ the Lévy-Khintchine triplet of $\mu$ and we write $\mu=\mu_{(A,\nu,\gamma)}$ when we want to emphasize the Lévy-Khintchine triplet. $C_{\mu}(z),z\in{{\mathbb R^d}}$, denotes the cumulant function of $\mu\in I({{\mathbb R^d}})$, that is, $C_{\mu}(z)$ is the unique continuous function satisfying ${\widehat}\mu(z) = e^{C_{\mu}(z)}$ and $C_{\mu}(0)=0$. When a random variable $X$ has the distribution $\mu$, we sometime write $C_X(z)$ for $C_{\mu}(z)$.
We also use the polar decomposition of the Lévy measure $\nu$ of $\mu\in I({{\mathbb R^d}})$ with $0<\nu({\mathbb{R}}^d)\le\infty$. There exist a measure ${\lambda}$ on $S:=\{x\in{{\mathbb R^d}}\colon|x|=1\}$ with $0<{\lambda}(S)\le\infty$ and a family $\{\nu_{\xi}, \xi\in S\}$ of measures on $(0,\infty)$ such that $\nu_{\xi}(B)$ is measurable in $\xi$ for each $B\in\mathcal B((0,\infty))$, $0<\nu_{\xi}((0,\infty))\le\infty$ for each $\xi\in S$ and $$\begin{aligned}
\label{polar}
\nu(B)=\int_S {\lambda}(d\xi)\int_0^{\infty} \1_B(r\xi)\nu_{\xi}(dr),\quad
B\in \mathcal B (\mathbb R^d \setminus \{ 0\}).\end{aligned}$$ Here ${\lambda}$ and $\{\nu_{\xi}\}$ are uniquely determined by $\nu$ up to multiplication of measurable functions $c(\xi)$ and $c(\xi)^{-1}$, respectively, with $0<c(\xi)<\infty$. We say that $\nu$ has the polar decomposition $({\lambda},\nu_{\xi})$, and ${\lambda}$ and $\nu_{\xi}$ are called the spherical and the radial components of $\nu$, respectively. (See, e.g., @Barndorff-NielsenMaejimaSato2006, Lemma 2.1.)
Recently, several natural examples of semi-selfdecomposable distributions have appeared in the literature. In @Watanabe2002, he showed that the distribution of a certain supercritical branching process and the first hitting time of Brownian motion starting at the origin on the unbounded Sierpinski gasket on $\mathbb R^2$ are both semi-selfdecomposable. Also, let $\{N_t,t\ge 0\}$ be a Poisson process and $\{X_t\}$ a Lévy process on ${{\mathbb R^d}}$ independent of $\{N_t\}$. Suppose $E\left [\log^+|X_1|\right ] <\infty$ and $b>1$. Then ${\mathcal L}\left ({\int_0^{\infty}}b^{-N_{t-}}dX_t\right )$ is semi-selfdecomposable with span $b$. (Theorem 3.2 of @KondoMaejimaSato2006.) In a recent paper by @LindnerSato2009, we can also find several examples of semi-selfdecomposable distributions with the form ${\mathcal L}\left ({\int_0^{\infty}}c^{-N_{t-}}dX_t\right )$, where $\{(N_t, X_t)\}$ is a bivariate compound Poisson process with Lévy measure concentrated on the three points $(1,0), (0,1)$ and $(1,1)$. Another recent example is found by @Pacheco-Gonzalez2009 in some financial modeling. These indicate introducing of semi-selfdecomposable distributions allows us more flexibility in stochastic modeling.
Stochastic integral characterizations of semi-selfdecomposable distributions
============================================================================
As we mentioned in Introduction, in this section, we introduce a mapping from a subset of $I({{\mathbb R^d}})$ into $I({{\mathbb R^d}})$, by which semi-selfdecomposable distributions can be characterized.
Let $b>1$ and $\mu\in I({\mathbb{R}}^d)$. Define a mapping $\Phi_b$ by $$\label{def_of_Phi_b}
\Phi_b(\mu):= {\mathcal L}\left(\int_0^\infty b^{-[t]}dX_t^{(\mu)}\right),$$ provided that this improper stochastic integral is definable, where $[x]$ denotes the largest integer not greater than $x\in{\mathbb{R}}$.
The domain of the mapping $\Phi_b$, where the improper stochastic integral in is definable, is given as follows by Theorem 2.4 of @Sato2006.
\[domain\_of\_Phib\] $\mathfrak{D}(\Phi_b)=I_{\log}({{\mathbb R^d}})$.
We start with the following theorem.
\[equivalence\_between\_decomposability\_and\_mapping\] Fix any $b>1$. Let $\mu$ and $\rho$ be distributions on ${{\mathbb R^d}}$. Then, $$\label{relation_c.f.}
\rho\in I({{\mathbb R^d}})\quad\text{and}\quad \widehat{\mu}(z)=\widehat{\mu}(b^{-1}z)\widehat{\rho}(z)$$ if and only if $$\label{mapping}
\rho\in I_{\log}({{\mathbb R^d}})\quad\text{and}\quad \mu=\Phi_b(\rho).$$
To show the “if" part, suppose . Note that $$\label{product}
\Phi_b(\rho) = {\mathcal L}\left ({\int_0^{\infty}}b^{-[t]}dX_t^{(\rho)}\right ) = {\mathcal L}\left (\sum _{j=0}^{\infty}b^{-j} \left(X_{j+1}^{(\rho)} - X_{j}^{(\rho)}\right)\right ).$$ Then $${\widehat}\mu(z) = \prod_{j=0}^{\infty}{\widehat}\rho (b^{-j}z) = \prod_{j=1}^{\infty}{\widehat}\rho (b^{-j}z)\times {\widehat}\rho (z)
= \prod_{k=0}^{\infty}{\widehat}\rho (b^{-k}(b^{-1}z))\times {\widehat}\rho (z) = {\widehat}\mu (b^{-1}z){\widehat}\rho(z),$$ which concludes .
We next show the “only if" part. Assume . Then as can be seen in @Wolfe1983, we have $${\widehat}\mu(z) = {\widehat}\mu (b^{-1}z){\widehat}\rho(z) = {\widehat}\mu (b^{-2}z){\widehat}\rho(b^{-1}z){\widehat}\rho(z)
= \cdots = {\widehat}\mu(b^{-n}z)\prod_{j=0}^{n-1}{\widehat}\rho (b^{-j}z),$$ for all $n\in{\mathbb{N}}$. Hence it follows that $\prod_{j=0}^{\infty}{\widehat}\rho (b^{-j}z)$ exists and equals ${\widehat}\mu(z)$, which implies $\rho\in I_{\log}({{\mathbb R^d}})$ by @Wolfe1983. Then $\Phi_b(\rho)$ is definable and satisfies . Thus we have $\Phi_b(\rho)=\mu$, which yields .
Theorem \[equivalence\_between\_decomposability\_and\_mapping\] yields the following.
\[s.s.d.\] Fix any $b>1$. Then, the range $\mathfrak R (\Phi_b)$ is the class of all semi-selfdecomposable distributions with span $b$ on ${{\mathbb R^d}}$, namely, $$\Phi_b\left(I_{\log}({{\mathbb R^d}})\right)=L(b,{\mathbb{R}}^d).$$
The injectivity of the mapping $\Phi_b$ is shown as follows.
\[ssd\_injective\] For each $b>1$, the mapping $\Phi_b$ is injective.
Let $\rho_1,\rho_2\in I_{\log}({{\mathbb R^d}})$ and $\mu=\Phi_b(\rho_1)=\Phi_b(\rho_2)$. Then, Theorem \[equivalence\_between\_decomposability\_and\_mapping\] yields that $$\widehat{\mu}(z)=\widehat{\mu}(b^{-1}z)\widehat{\rho}_1(z)=\widehat{\mu}(b^{-1}z)\widehat{\rho}_2(z).$$ Since $\widehat{\mu}(b^{-1}z)\neq 0$ for all $z\in{{\mathbb R^d}}$ by the infinite divisibility of $\mu$, it follows that $\widehat{\rho}_1(z)=\widehat{\rho}_2(z)$.
As mentioned in Introduction, if $\mu\in L({{\mathbb R^d}})$, then there exists $\mu_0\in I_{\log}({{\mathbb R^d}})$ such that $\mu={\mathcal L}\left ( {\int_0^{\infty}}e^{-t}dX_t^{(\mu_0)}\right )$, and it is known that this $\mu_0$ is uniquely determined by $\mu$. We have just shown that if $\mu\in L(b, {{\mathbb R^d}})$, then there exists $\mu_b\in I_{\log}({{\mathbb R^d}})$ such that $\mu={\mathcal L}\left ( {\int_0^{\infty}}b^{-[t]}dX_t^{(\mu_b)}\right )$, and the uniqueness of $\mu_b$ is assured by Proposition \[ssd\_injective\]. If $\mu\in L({{\mathbb R^d}})$, then $\mu\in L(b,{{\mathbb R^d}})$ for any $b>1$. Then it is natural to ask what relation there is between $\mu_0$ and $\mu_b$ with $b>1$. We answer this question below. Fix $b>1$. Let $\mu=\mu_{(A,\nu,\gamma)}$, $\mu_0={\mu_0}_{(A_0,\nu_0,\gamma_0)}$ and $\mu_b={\mu_b}_{(A_b,\nu_b,\gamma_b)}$. Denote the polar decompositions of $\nu$, $\nu_0$ and $\nu_b$ by $(\lambda, \nu_{\xi})$, $(\lambda_0,\nu_{0,\xi})$ and $(\lambda_b,\nu_{b,\xi})$, respectively. Note that $\mu\in L({{\mathbb R^d}})$ if and only if $$\nu_{\xi}(dr)=\frac{k_\xi(r)}{r}dr,\quad r>0,$$ where $k_\xi(r)$ is a nonnegative function, which is measurable in $\xi$, and is nonincreasing and right-continuous in $r$. (See @Sato's_book1999, Theorem 15.10.) We have $$\begin{aligned}
\label{A_0}A&=\int_0^\infty e^{-2t}A_0dt=2^{-1}A_0,\\
\notag\gamma&=\int_0^\infty e^{-t}dt\left\{\gamma_0+\int_{{\mathbb R^d}}x\left(\frac{1}{1+e^{-2t}|x|^2}-\frac{1}{1+|x|^2}\right)
\nu_0(dx)\right\}\\
\label{gamma_0}&=\gamma_0+\int_{{{\mathbb R^d}}\setminus\{0\}}x\left(\frac{\arctan |x|}{|x|}-\frac{1}{1+|x|^2}\right)\nu_0(dx),\end{aligned}$$ and it follows from Theorem 41 (ii) of @Sato's_book2003 that $\lambda_0=\lambda$ and $\nu_{0,\xi}(dr)=-dk_\xi(r)$, up to multiplication of positive finite measurable functions $c(\xi)$ and $c(\xi)^{-1}$. On the other hand, Theorem \[equivalence\_between\_decomposability\_and\_mapping\] yields that $\mu_b$ is an infinitely divisible distribution satisfying $\widehat\mu(z)=\widehat\mu(b^{-1}z)\widehat\mu_b(z)$. Therefore $$\begin{aligned}
\label{A_b}A_b&=\left(1-b^{-2}\right)A,\\
\label{gamma_b}\gamma_b&=\left(1-b^{-1}\right)\gamma-\int_{{{\mathbb R^d}}}x\left(\frac{1}{1+|x|^2}-\frac{1}{1+|bx|^2}\right)\nu(b\,dx),\end{aligned}$$ and $$\begin{aligned}
\nu_b(B)&=\nu(B)-\nu(bB)=\int_S\lambda(d\xi)\int_0^\infty\1_B(r\xi)\frac{k_\xi(r)-k_\xi(br)}{r}dr\\
&=\int_S\lambda(d\xi)\int_0^\infty\1_B(r\xi)\frac{\nu_{0,\xi}\left((r,br]\right)}{r}dr,\qquad B\in\mathcal B_0({{\mathbb R^d}}).\end{aligned}$$ Then, it follows that $$\label{nu_0_and_nu_b}
\lambda_b=\lambda_0=\lambda\quad\text{and}\quad\nu_{b,\xi}(dr)=\frac{\nu_{0,\xi}\left((r,br]\right)}{r}dr=
\frac{k_\xi(r)-k_\xi(br)}{r}dr\ \lambda\text{-a.e.}\,\,\xi\in S,$$ up to multiplication of positive finite measurable functions $c(\xi)$ and $c(\xi)^{-1}$. One can see the relation between $\mu_0$ and $\mu_b$ by , , , and .
As also mentioned in Introduction, @MaejimaSato2003 characterized semi-selfdecomposable distributions by stochastic integrals with respect to natural semi-Lévy processes. The following theorem is another version of Corollary \[s.s.d.\] in this paper and Corollary 5.4 of @MaejimaSato2003, and connects them. For $b>1$, let $G_b$ denote the totality of bounded periodic measurable functions with period $\log b$.
\[period\] Fix any $b>1$. Then we have $$L(b,{\mathbb{R}}^d)=\left\{{\mathcal L}\left(\int_0^\infty e^{-t}g(t)dX_t^{(\mu)}\right)\colon g\in G_b \text{ and }
\mu\in I_{\log}({{\mathbb R^d}})
\right\}.$$
Let $\widetilde\mu\in L(b,{{\mathbb R^d}})$. Then Corollary \[s.s.d.\] yields that $\widetilde\mu=\Phi_b(\mu)$ for some $\mu\in I_{\log}({{\mathbb R^d}})$. If we let $$g(t):=b^{\frac{t}{\log b}-\left[\frac{t}{\log b}\right]},$$ then $g\in G_b$ and $e^{-t}g(t)=b^{-[t/\log b]}$. It follows that $$\widetilde\mu=\Phi_b(\mu)
={\mathcal L}\left(\int_0^\infty b^{-[t/\log b]}dX_{t/\log b}^{(\mu)}\right)
={\mathcal L}\left(\int_0^\infty e^{-t}g(t)dX_t^{\left(\mu^{1/\log b}\right)}\right),$$ where for $p>0$, $\mu^p$ is an infinitely divisible distribution with characteristic function $\widehat{\mu}(z)^p$.
Conversely, suppose that $\widetilde\mu={\mathcal L}\left(\int_0^\infty e^{-t}g(t)dX_t^{\left(\mu\right)}\right)$ for $g\in G_b$ and $\mu\in I_{\log}({{\mathbb R^d}})$. Putting $Y_t:=\int_0^tg(s)dX_s^{(\mu)}$, we have $\widetilde\mu={\mathcal L}\left(\int_0^\infty e^{-t}dY_t\right)$ due to Theorem 4.6 of @Sato2004, and we see that $\{Y_t\}$ is a natural semi-Lévy process with period $\log b$. Moreover, we have $E\left[\log^+|Y_{\log b}|\right]<\infty$ since the Lévy measure of ${\mathcal L}(Y_{\log b})$ denoted by $\widetilde \nu$ satisfies that $$\begin{aligned}
\int_{|x|>1}\log|x|\widetilde\nu(dx)&=\int_0^{\log b}ds\int_{{\mathbb R^d}}\log^+|g(s)x|\nu(dx)\\
&\leq \log b\int_{{\mathbb R^d}}\log^+\biggl(\sup_{s\in[0,\log b]}|g(s)||x|\biggr)\nu(dx)<\infty,\end{aligned}$$ where $\nu$ is the Lévy measure of $\mu$. Then Corollary 5.4 of @MaejimaSato2003 implies that $\widetilde\mu\in L(b,{{\mathbb R^d}})$.
In the proof of Theorem \[period\], $Y_t = \int_0^tg(s)dX_s^{(\mu)}$ with $g\in G_b$ and $\mu\in I_{\log}({{\mathbb R^d}})$ is proved to be a natural semi-Lévy process. However, any natural semi-Lévy process is not necessarily expressed in this way as is shown in the following example.
\[example\] We claim that not all natural semi-Lévy processes can be expressed as stochastic integrals with respect to Lévy processes. Fix an arbitrary $p>0$. Let $\varphi\colon [0,p]\rightarrow [0,\infty)$ be a nondecreasing function satisfying $\varphi(0)=0$ which is continuous but not absolutely continuous, for example, Cantor’s function on $[0,p]$. Suppose that $\mu\in I({{\mathbb R^d}})\setminus\{\delta_0\}$. Then there exists a semi-Lévy process $\{Y_t,t\geq 0\}$ with period $p$ such that $C_{Y_t}(z)={\varphi(t)}C_{\mu}(z)$ for $z\in{{\mathbb R^d}}$ and $t\in [0,p]$, due to Proposition 2.2 of @MaejimaSato2003. Furthermore, it follows from the monotonicity of $\varphi$ that $\{Y_t\}$ is natural. However, we cannot express $\{Y_t\}$ in the form that $Y_t=\int_0^tf(s)dX_s$ for any measurable function $f$ and any Lévy process $\{X_t\}$. Indeed, if $\int_0^tf(s)dX_s$ is definable for all $t\in[0,\infty)$, then $C_{\int_0^tf(s)dX_s}(z)=\int_0^tC_{X_1}(f(s)z)ds$ which is absolutely continuous in $t$, although $C_{Y_t}(z)$ is not absolutely continuous in $t$ by the property of $\varphi$.
A Langevin type equation
========================
The purpose of this and the following two sections is to find a Langevin type equation like or related to semi-selfdecomposable distributions. The ideas of proofs of the results below come from Sections 2.2 and 2.4 of @Sato's_book2003 and @MaejimaSato2003.
For this purpose, we first consider the following *Langevin type equation*: $$\label{Langevin_t_0}
Z_t = M + X_{[ct]/c}-X_{[ct_0]/c} -(b-1) \int_{t_0}^{t} Z_sd[cs], \quad t\geq t_0,$$ where $t_0\in{\mathbb{R}}$, $c>0$, $b>1$, $\{X_t, t\in{\mathbb{R}}\}$ is a h.-i.s.r.m.-process on ${{\mathbb R^d}}$, $M$ is an ${{\mathbb R^d}}$-valued random variable, and $\int_0^{t} Z_sd[cs]$ is as follows: $d[cs]$ denotes the Lebesgue-Stieltjes measure associated with $s\mapsto [cs]$, which is equal to $\sum_{k\in{\mathbb{Z}}}\delta_{k/c}(ds)$, and $\int_{(\alpha,\beta]} f(s)d[cs]$, written as $\int_\alpha^\beta f(s)d[cs]$, exists for $-\infty<\alpha<\beta<\infty$ and any measurable function $f$, since it is a finite sum in fact. A stochastic process $\{Z_t\}$ is said to be a *solution* of the Langevin type equation or an *Ornstein-Uhlenbeck type process* generated by $\{X_t\}$, $b$ and $c$ starting from $Z_{t_0}=M$ if sample paths of $\{Z_t\}$ are right-continuous with left limits and $\{Z_t\}$ satisfies almost surely. We claim that $$\label{solution_t_0}
Z_t=b^{-([ct]-[ct_0])}M+b^{-[ct]}\int_{[ct_0]/c}^{[ct]/c}b^{[cs]}dX_s, {\quad}t\ge t_0,$$ is an almost surely unique solution of .
\[thm\_Langevin\_t\_0\] Suppose that $\{X_t, t\in{\mathbb{R}}\}$ is a h.-i.s.r.m.-process on ${{\mathbb R^d}}$, $t_0\in{\mathbb{R}}$, $c>0$, $b>1$, and $M$ is an ${{\mathbb R^d}}$-valued random variable. Then, $\{Z_t\}$ in is an almost surely unique solution of the equation .
If we define $\{Z_t\}$ by , then it is a càdlàg process. If $t_0\leq t<t_0+1/c$, then $Z_t=M$ and it satisfies . Let $t\geq t_0+1/c$. Then, it follows that $$\begin{aligned}
(b&-1)\int_{t_0}^{t} Z_sd[cs]\\
&=(b-1)\sum_{k=[ct_0]+1}^{[ct]}Z_{k/c}
=(b-1)\sum_{k=[ct_0]+1}^{[ct]}\left\{b^{-(k-[ct_0])}M+b^{-k}\int_{[ct_0]/c}^{k/c}b^{[cs]}dX_s\right\}\\
&=\left(1-b^{-([ct]-[ct_0])}\right)M +(b-1)\sum_{k=[ct_0]+1}^{[ct]}b^{-k}\sum_{\ell=[ct_0]+1}^kb^{\ell-1}
\left(X_{{\ell}/{c}}-X_{{(\ell-1)}/{c}}\right)\\
&=\left(1-b^{-([ct]-[ct_0])}\right)M +(b-1)\sum_{\ell=[ct_0]+1}^{[ct]}b^{\ell-1}\left(X_{{\ell}/{c}}
-X_{({\ell-1)}/{c}}\right)
\sum_{k=\ell}^{[ct]}b^{-k}\\
&=\left(1-b^{-([ct]-[ct_0])}\right)M +\sum_{\ell=[ct_0]+1}^{[ct]}\left(1-b^{-[ct]+\ell-1}\right)\left(X_{{\ell}/{c}}
-X_{{(\ell-1)}/{c}}\right)\\
&=\left(1-b^{-([ct]-[ct_0])}\right)M +X_{[ct]/c}-X_{[ct_0]/c}-b^{-[ct]}\int_{[ct_0]/c}^{[ct]/c}b^{[cs]}dX_s\\
&=M+X_{[ct]/c}-X_{[ct_0]/c}-Z_t.\end{aligned}$$ This yields .
It remains to prove the uniqueness of the solution of . Suppose that two $\{Z^{(1)}_t\}$ and $\{Z^{(2)}_t\}$ are the solutions of . Setting $Z_t:=Z^{(1)}_t-Z^{(2)}_t$, we have $$\label{uniqueness}
Z_t=-(b-1)\int_{t_0}^tZ_sd[cs]\qquad\text{for } t\geq t_0,\quad\text{a.s.}$$ Let us show that $$\label{uniqueness_induction}
Z_t=0,\qquad \text{for }\,\,\left (t_0\vee \frac{[ct_0]+n-1}{c}\right )\leq t<\frac{[ct_0]+n}{c},\quad\text{a.s.}$$ for any $n\in{\mathbb{N}}$ by induction. is true for $n=1$, since the right-hand side of is zero for $t_0\leq t<([ct_0]+1)/c$. Assume that holds for $n=1,2,\dots,m$. Then, for $([ct_0]+m)/c\leq t<([ct_0]+m+1)/c$, can be reduced to that $$Z_t=-(b-1)\int_{([ct_0]+m-1)/c}^{([ct_0]+m)/c}Z_sd[cs]=-(b-1)Z_{([ct_0]+m)/c},$$ which implies that $X_{([ct_0]+m)/c}=0$ by letting $t=([ct_0]+m)/c$ and thus $Z_t=0$ for $([ct_0]+m)/c\leq t<([ct_0]+m+1)/c$. Hence is true for $n=m+1$. Therefore it holds with probability one that for any $t\geq 0$, $Z_t=Z^{(1)}_t-Z^{(2)}_t=0$.
Limiting distributions of Ornstein-Uhlenbeck type processes
===========================================================
In this section, we study the Langevin type equation with $t_0=0$: $$\label{Langevin}
Z_t = M + X_{[ct]/c} -(b-1) \int_0^{t} Z_sd[cs], \quad t\geq 0,$$ where $\{X_t,t\geq 0\}$ is a Lévy process on ${{\mathbb R^d}}$, $c>0$, $b>1$, and $M$ is an ${{\mathbb R^d}}$-valued random variable. Theorem \[thm\_Langevin\_t\_0\] yields that $$\label{solution}
Z_t=b^{-[ct]}M+b^{-[ct]}\int_0^{[ct]/c}b^{[cs]}dX_s, \quad t\geq 0,$$ is an almost surely unique solution of . We show that if $M$ is independent of $\{X_t\}$ and ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$, then the limit of ${\mathcal L}(Z_t)$ exists as $t\rightarrow \infty $ and the limiting distribution is semi-selfdecomposable with span $b$.
\[limiting\_distribution\] Suppose that $\{X_t, t\geq 0\}$ is a Lévy process on ${{\mathbb R^d}}$, $c>0$, $b>1$, and $M$ is an ${{\mathbb R^d}}$-valued random variable independent of $\{X_t\}$. Then, ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$ if and only if $Z_t$ in converges in law to some ${{\mathbb R^d}}$-valued random variable as $t\rightarrow \infty $. This limit $\lim_{t\rightarrow \infty }{\mathcal L}(Z_t)$ is equal to ${\mathcal L}\left(\int_0^\infty b^{-[cs]-1}dX_s\right)$ which is semi-selfdecomposable with span $b$ and does not depend on $M$. Furthermore, if we let ${\mathcal L}(M):=\lim_{t\rightarrow \infty }{\mathcal L}(Z_t)$, then ${\mathcal L}(Z_t)={\mathcal L}(M)$ for all $t\geq 0$.
In the rest of the paper, we write ${\widehat}{{\mathcal L}}(X)(z) $ for the characteristic function of ${\mathcal L}(X)$ for notational simplicity. It follows that $$\begin{aligned}
\notag \widehat{{\mathcal L}}(Z_t)(z)
&=\widehat{{\mathcal L}}(M)\left(b^{-[ct]}z\right)\exp\left\{\int_0^{[ct]/c}C_{X_1}
\left(b^{\left[cs-[ct]\right]}z\right)ds\right\}\\
\notag&=\widehat{{\mathcal L}}(M)\left(b^{-[ct]}z\right) \exp\left\{\int_0^{[ct]/c}
C_{X_1}\left(b^{[-cu]}z\right)du\right\}\\
\label{sol_cf}&=\widehat{{\mathcal L}}(M)\left(b^{-[ct]}z\right) \exp\left\{\int_0^{[ct]/c}
C_{X_1}\left(b^{-[cu]-1}z\right)du\right\}.\end{aligned}$$ Note that $\widehat{{\mathcal L}}(M)\left(b^{-[ct]}z\right)$ tends to $1$ as $t\rightarrow \infty $.
Assume ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$. Then, tends to the cumulant function of the improper stochastic integral $\int_0^\infty b^{-[cu]-1}dX_u$, whose law is semi-selfdecomposable with span $b$ due to Corollary \[s.s.d.\]. Also, $\lim_{t\rightarrow \infty }{\mathcal L}(Z_t)={\mathcal L}\left(\int_0^\infty b^{-[cu]-1}dX_u\right)$ does not depend on $M$ and if we set ${\mathcal L}(M):={\mathcal L}\left(\int_0^\infty b^{-[cu]-1}dX_u\right)$, then is $$\begin{aligned}
C_{Z_t}(z)
& =\int_0^{\infty}C_{X_1}\left(b^{-\left[cu+[ct]\right]-1}z\right)du
+\int_0^{[ct]/c}C_{X_1}\left(b^{-[cu]-1}z\right)du\\
&=\int_{[ct]/c}^{\infty}C_{X_1}\left(b^{-[cv]-1}z\right)dv
+\int_0^{[ct]/c}C_{X_1}\left(b^{-[cu]-1}z\right)du\\
&=\int_0^{\infty}C_{X_1}\left(b^{-[cu]-1}z\right)du,\end{aligned}$$ which yields that ${\mathcal L}(Z_t)={\mathcal L}\left(\int_0^\infty b^{-[cu]-1}dX_u\right)$ for all $t\geq 0$.
Next assume that $Z_t$ converges in law as $t\rightarrow \infty $, namely, $\int_0^{n/c}C_{X_1}
\left(b^{-[cu]-1}z\right)du$ tends to the cumulant function of some $\mu\in I({{\mathbb R^d}})$ with Lévy measure $\nu$ as $n\rightarrow \infty $. Denoting the Lévy measure of ${\mathcal L}(X_1)$ by $\nu_{X_1}$, we have $$\lim_{k\rightarrow \infty}\int_0^{n_k/c}du\int_{{\mathbb R^d}}\left(|b^{-[cu]-1}x|^2\wedge 1\right)\nu_{X_1}(dx)=
\int_{{\mathbb R^d}}(|x|^2\wedge 1)\nu(dx)$$ for some subsequence, due to the proof of Theorem 8.7 in @Sato's_book1999. It follows from the monotone convergence theorem that $\int_0^{\infty}du\int_{{\mathbb R^d}}\left(|b^{-[cu]-1}x|^2\wedge 1\right)\nu_{X_1}(dx)<\infty$, which implies ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$ by virtue of Lemma 2.7 of @Sato2006.
Theorem \[limiting\_distribution\] yields that if ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$, then $Z_t$ in converges in law as $t\rightarrow \infty $. Then, it might be natural to ask whether or not $Z_t$ converges in probability as $t\rightarrow \infty $. The following proposition is the answer.
Suppose that $\{X_t, t\geq 0\}$ is a Lévy process on ${{\mathbb R^d}}$, $c>0$, $b>1$, and $M$ is an ${{\mathbb R^d}}$-valued random variable independent of $\{X_t\}$. If ${\mathcal L}(X_1)$ is not any $\delta$-distribution, then $Z_t$ in does not converge in probability as $t\rightarrow \infty $.
Suppose that $\mu:={\mathcal L}(X_1)$ is not any $\delta$-distribution. Then, Lemma 13.9 of @Sato's_book1999 yields that $\left|\widehat{\mu}(z_0)\right|<1$ for some $z_0\in{{\mathbb R^d}}$. As $$Z_t-Z_{t-1/c}=b^{-[ct]}\int_{[ct-1]/c}^{[ct]/c}b^{[cs]}dX_s+(b^{-[ct]}-b^{-[ct-1]})
\left(M+\int_0^{[ct-1]/c}b^{[cs]}dX_s\right)$$ by , we have $$\begin{aligned}
&\left|\widehat{{\mathcal L}} \left(Z_t-Z_{t-1/c}\right)(z)\right|\leq \left|\widehat{{\mathcal L}}\left(b^{-[ct]}
\int_{[ct-1]/c}^{[ct]/c}b^{[cs]}dX_s\right)(z)\right|\\
&\qquad\qquad=\left|\widehat{{\mathcal L}}\left(b^{-1}\left(X_{[ct]/c}-X_{[ct-1]/c}\right)\right)(z)\right|
=\left|\widehat{\mu}(b^{-1}z)\right|^{1/c}\end{aligned}$$ for any $t\geq 1/c$ and $z\in{{\mathbb R^d}}$. This yields that for all $t\ge 1/c$, $$\left|\widehat{{\mathcal L}} \left(Z_t-Z_{t-1/c}\right)(bz_0)\right|\leq
\left|\widehat{\mu}(z_0)\right|^{1/c}<1.$$ Then $Z_t-Z_{t-1/c}$ does not tends to zero in probability as $t\rightarrow \infty$. Thus $Z_t$ does not converge in probability as $t\rightarrow \infty $.
The following remark is about the relation between the Langevin type equation and the mapping $\Phi_b$.
\[relation\_between\_Langevin\_and\_mapping\] Fix $b>1$ and $c>0$. Let $\mu\in I_{\log}({{\mathbb R^d}})$. Consider the limiting solution in law $\lim_{t\rightarrow \infty }{\mathcal L}(Z_t)$ of the Langevin type equation with a Lévy process $\{X_t\}$ satisfying ${\mathcal L}(b^{-1}X_{1/c})=\mu$ and an ${{\mathbb R^d}}$-valued random variable $M$ independent of $\{X_t\}$. It follows from Theorem \[limiting\_distribution\] that $$\begin{aligned}
\Phi_b(\mu)&
={\mathcal L}\left(\int_0^\infty b^{-[t]}dX_t^{(\mu)}\right)
={\mathcal L}\left(\int_0^\infty b^{-[t]}dX_t^{\left({\mathcal L}(b^{-1}X_{1/c})\right)}\right)\\
&={\mathcal L}\left(\int_0^\infty b^{-[t]}d(b^{-1}X_{t/c})\right)
={\mathcal L}\left(\int_0^\infty b^{-[cs]-1}dX_s\right)=\lim_{t\rightarrow \infty }{\mathcal L}(Z_t).\end{aligned}$$ Thus the mapping $\Phi_b$ can be defined also as the limiting distribution of the solution of the Langevin type equation .
We conclude this section with the Markov property of our Ornstein-Uhlenbeck type processes.
Suppose that $\{X_t, t\geq 0\}$ is a Lévy process on ${{\mathbb R^d}}$, $c>0$, $b>1$, and $M$ is an ${{\mathbb R^d}}$-valued random variable independent of $\{X_t\}$. Then, the process $\{Z_t\}$ in is a Markov process satisfying $$\label{Markov}
P\left(Z_t\in B\mid Z_s=x\right)=P\left(b^{-([ct]-[cs])}x+\int_0^{([ct]-[cs])/c}b^{-[cu]-1}dX_u\in B\right)$$ for $0\leq s\leq t$ and $B\in{\mathcal{B}}({{\mathbb R^d}})$.
Since $$Z_t=b^{-([ct]-[cs])}Z_s+b^{-[ct]}\int_{[cs]/c}^{[ct]/c}b^{[cu]}dX_u,$$ we can easily see the Markov property of $\{Z_t\}$ by virtue of the independent increment property of the Lévy process $\{X_t\}$. is shown as follows: $$\begin{aligned}
E&\left[e^{i{\langle}z,Z_t{\rangle}}\bigm| Z_s=x\right]\\
&=\exp\left\{i{\langle}z,b^{-([ct]-[cs])}x{\rangle}+\int_{[cs]/c}^{[ct]/c}C_{X_1}\left (b^{[cu-[ct]]}z\right )du\right\}\\
&=\exp\left\{i{\langle}z,b^{-([ct]-[cs])}x{\rangle}+\int_{0}^{([ct]-[cs])/c}C_{X_1}\left (b^{[-cv]}z\right)du\right\}\\
&=\exp\left\{i{\langle}z,b^{-([ct]-[cs])}x{\rangle}+\int_{0}^{([ct]-[cs])/c}C_{X_1}\left (b^{-[cv]-1}z\right )du\right\}.\end{aligned}$$
Semi-stationary Ornstein-Uhlenbeck type processes having semi-selfdecomposable distributions
============================================================================================
This section is concerned with the following *Langevin type equation* which has similar properties to : $$\label{Langevin_semi-stationary}
Z_t-Z_s =X_{[ct]/c}-X_{[cs]/c} -(b-1) \int_s^{t} Z_ud[cu], \quad -\infty<s\leq t<\infty,$$ where $c>0$, $b>1$, and $\{X_t,t\in{\mathbb{R}}\}$ is a h.-i.s.r.m.-process on ${{\mathbb R^d}}$. A stochastic process $\{Z_t\}$ is said to be a *solution* of the Langevin type equation or an *Ornstein-Uhlenbeck type process* generated by $\{X_t\}$, $b$ and $c$ if sample paths of $\{Z_t\}$ are right-continuous with left limits and $\{Z_t\}$ satisfies almost surely.
We show that if $E[\log^+|X_1|]<\infty$, then $\int_{-\infty}^tb^{[cu]}dX_u$ is definable for each $t\in{\mathbb{R}}$ and $$\label{solution_semi-stationary}
Z_t=b^{-[ct]}\int_{-\infty}^{[ct]/c}b^{[cu]}dX_u, \quad t\in{\mathbb{R}},$$ is an almost surely unique semi-stationary solution of , where the *semi-stationarity* of $\{Z_t\}$ means $\{Z_{t+p}\}{\overset{\mathrm d}{=}}\{Z_t\}$ for a fixed $p>0$. Here ${\overset{\mathrm d}{=}}$ stands for equality in all finite-dimensional distributions. This $p$ is called the *period* of the semi-stationary process $\{Z_t\}$.
To prove this, we prepare two lemmas.
\[lemma\_semi-stationary\_Langevin\] Let $\{X_t,t\in{\mathbb{R}}\}$ be a h.-i.s.r.m.-process on ${{\mathbb R^d}}$. Suppose that $b>1$ and $c>0$. Then, the following three statements are equivalent:
1. ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$,
2. $\int_{-\infty}^0b^{[ct]}dX_t$ is definable,
3. there exists an Ornstein-Uhlenbeck type process $\{Z_t\}$ generated by $\{X_t\}$, $b$ and $c$ satisfying ${\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}b^{[ct]}Z_t=0$, where ${\mathop{\mbox{\rm p-$\lim$}}}$ stands for limit in probability.
If [(iii)]{} holds, then $\{Z_t\}$ with the properties in [(iii)]{} is almost surely unique and expressed as .
We first show that (i) and (ii) are equivalent. Theorem 2.4 of @Sato2006 yields that ${\mathcal L}(X_1)={\mathcal L}\left(-X_{(-1)-}\right)\in I_{\log}({{\mathbb R^d}})$ if and only if $\int_0^\infty b^{[-cu]}d\left(-X_{(-u)-}\right)$ is definable. Lemma 4.8 of @MaejimaSato2003 implies that $\int_0^\infty b^{[-cu]}d\left(-X_{(-u)-}\right)$ is definable if and only if $\int_{-\infty}^0b^{[cu]}dX_u$ is definable. Thus (i) and (ii) are equivalent.
We next show that (ii) implies (iii). Assume that (ii) holds. Then, $\{Z_t\}$ in is definable. It satisfies due to Theorem \[thm\_Langevin\_t\_0\] by letting $t_0=s$ and $Z_{t_0}=M=b^{-[cs]}\int_{-\infty}^{[cs]/c}b^{[cu]}dX_u$. It follows from and (ii) that $${\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}b^{[ct]}Z_t={\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}\int_{-\infty}^{[ct]/c}b^{[cu]}dX_u=0.$$
Finally, we show that (iii) implies (i), the uniqueness of $\{Z_t\}$ in (iii), and the expression . If (iii) holds, then $\{Z_t\}$ in (iii) satisfies . Theorem \[thm\_Langevin\_t\_0\] yields that for each $s\in {\mathbb{R}}$, with probability one, $$Z_t=b^{-([ct]-[cs])}Z_s+b^{-[ct]}\int_{[cs]/c}^{[ct]/c}b^{[cu]}dX_u,\qquad\text{for } t\ge s,$$ namely, with probability one, $$b^{[ct]}Z_t-b^{[cs]}Z_s=\int_{[cs]/c}^{[ct]/c}b^{[cu]}dX_u,\qquad\text{for } t\ge s.$$ By letting $s\rightarrow -\infty$, it follows from (iii) that for each $t\in{\mathbb{R}}$, $$b^{[ct]}Z_t={\mathop{\mbox{\rm p-$\lim$}}}_{k\rightarrow \infty}\int_{-k/c}^{[ct]/c}b^{[cu]}dX_u,\qquad\text{a.s.}$$ By a similar argument to that in the proof of Theorem \[limiting\_distribution\], we have ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$. Thus (i) holds. Then (ii) holds and it follows that for each $t\in{\mathbb{R}}$, $$b^{[ct]}Z_t=\int_{-\infty}^{[ct]/c}b^{[cu]}dX_u,\qquad\text{a.s.}$$ However, since the both sides of the equation above have càdlàg paths, we have, with probability one, $$b^{[ct]}Z_t=\int_{-\infty}^{[ct]/c}b^{[cu]}dX_u,\qquad\text{for all }t\in{\mathbb{R}}.$$ This yields the almost sure uniqueness of $\{Z_t\}$ in (iii), and the expression .
\[semi-stationarity\_iff\_mildness\] Let $\{X_t,t\in{\mathbb{R}}\}$ be a h.-i.s.r.m.-process on ${{\mathbb R^d}}$, $b>1$ and $c>0$. Suppose that $\{Z_t\}$ is an Ornstein-Uhlenbeck type process generated by $\{X_t\}$, $b$ and $c$. Then, $\{Z_t\}$ is semi-stationary if and only if ${\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}b^{[ct]}Z_t=0$. Let these conditions be fulfilled. Then, semi-stationary process $\{Z_t\}$ has a period $1/c$. Moreover, $\{Z_t\}$ with the properties above is almost surely unique and expressed as .
We first show the “only if" part. Suppose that $\{Z_t\}$ is a semi-stationary Ornstein-Uhlenbeck type process generated by $\{X_t\}$, $b$ and $c$ with period $p>0$. Since $\{Z_t\}$ has càdlàg paths, for any sequence $\{t_n,n\in{\mathbb{N}}\}\subset[0,p]$, there exists its subsequence $\{t_{n_k},k\in{\mathbb{N}}\}$ satisfying $Z_{t_{n_k}}$ converges almost surely to some ${{\mathbb R^d}}$-valued random variable as $k\to\infty$. This implies the relative compactness of $\{{\mathcal L}(Z_t)\colon t\in[0,p]\}$ which is equal to $\{{\mathcal L}(Z_t)\colon t\in{\mathbb{R}}\}$ by the semi-stationarity of $\{Z_t\}$. Hence $\{{\mathcal L}(Z_t)\colon t\in{\mathbb{R}}\}$ is tight by Prohorov’s theorem. Then, it follows that for any $\varepsilon>0$, $$P\left(|b^{[ct]}Z_t|>\varepsilon\right)\leq \sup_{s\in{\mathbb{R}}}P\left(|Z_s|>b^{-[ct]}\varepsilon\right)\to 0,\quad\text{as }t\to -\infty.$$ Thus ${\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}b^{[ct]}Z_t=0$.
We next show the “if" part. Suppose that $\{Z_t\}$ is an Ornstein-Uhlenbeck type process generated by $\{X_t\}$, $b$ and $c$ satisfying ${\mathop{\mbox{\rm p-$\lim$}}}_{t\rightarrow -\infty}b^{[ct]}Z_t=0$. Then $\{Z_t\}$ has the form due to Lemma \[lemma\_semi-stationary\_Langevin\]. Let $-\infty<t_1<t_2<\dots<t_n<\infty$. Then, for each $j=2,3,\dots,n$, we have $$\begin{aligned}
b^{[ct_j]}Z_{t_j+1/c}-b^{[ct_{j-1}]}Z_{t_{j-1}+1/c}
&=b^{-1}\int_{[ct_{j-1}+1]/c}^{[ct_j+1]/c}b^{[cu]}dX_u\\
&=\int_{[ct_{j-1}]/c}^{[ct_j]/c}b^{[cv]}d(X_{v+1/c}-X_{1/c}),\end{aligned}$$ which is equal in law to $$\int_{[ct_{j-1}]/c}^{[ct_j]/c}b^{[cv]}dX_v
=b^{[ct_j]}Z_{t_j}-b^{[ct_{j-1}]}Z_{t_{j-1}}.$$ Since $\{b^{[ct]}Z_{t+1/c}\}$ and $\{b^{[ct]}Z_t\}$ have independent increment property due to the expression , it follows that $\{b^{[ct]}Z_{t+1/c}\}{\overset{\mathrm d}{=}}\{b^{[ct]}Z_t\}$, which yields $\{Z_{t+1/c}\}{\overset{\mathrm d}{=}}\{Z_t\}$. This is the semi-stationarity of $\{Z_t\}$ with period $1/c$.
The almost sure uniqueness of $\{Z_t\}$ follows from Lemma \[lemma\_semi-stationary\_Langevin\].
Now, we prove the following theorem on the relation between semi-stationary Ornstein-Uhlenbeck type processes and semi-selfdecomposable distributions.
\[thm\_semi-stationary\_Langevin\] Suppose that $\{X_t, t\in{\mathbb{R}}\}$ is a h.-i.s.r.m.-process on ${{\mathbb R^d}}$, $c>0$, and $b>1$. Then, ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$ if and only if has a semi-stationary solution. If ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$, $\{Z_t\}$ in is an almost surely unique semi-stationary solution of and it satisfies ${\mathcal L}(Z_t)={\mathcal L}\left(\int_0^\infty b^{-[cu]-1}dX_u\right)\in L(b,{{\mathbb R^d}})$ for all $t\in {\mathbb{R}}$. In this case, the semi-stationary process $\{Z_t\}$ has a period $1/c$.
The most parts of this theorem have already been proved in Lemmas \[lemma\_semi-stationary\_Langevin\] and \[semi-stationarity\_iff\_mildness\]. It remains to show the statement that if ${\mathcal L}(X_1)\in I_{\log}({{\mathbb R^d}})$, then ${\mathcal L}(Z_t)
={\mathcal L}\left(\int_0^\infty b^{-[cu]-1}dX_u\right)\in L(b,{{\mathbb R^d}})$ for all $t\in {\mathbb{R}}$. However, it follows that for all $t\in{\mathbb{R}}$, $$\begin{aligned}
C_{Z_t}(z)&=\int_{-\infty}^{[ct]/c}C_{X_1}\left(b^{\left[cu-[ct]\right]}z\right)du
=\int_0^\infty C_{X_1}\left(b^{[-cv]}z\right)dv\\
&=\int_0^\infty C_{X_1}\left(b^{-[cv]-1}z\right)dv
=C_{\int_0^\infty b^{-[cv]-1}dX_v}(z),\end{aligned}$$ which is the cumulant function of some semi-selfdecomposable distribution with span $b$, due to Corollary \[s.s.d.\].
The following remark, which is similar to Remark \[relation\_between\_Langevin\_and\_mapping\], is about the relation between the Langevin type equation and the mapping $\Phi_b$.
Fix $b>1$ and $c>0$. Let $\mu\in I_{\log}({{\mathbb R^d}})$. Consider the almost surely unique semi-stationary Ornstein-Uhlenbeck type process $\{Z_t\}$ generated by a h.-i.s.r.m.-process $\{X_t\}$ satisfying ${\mathcal L}(b^{-1}X_{1/c})=\mu$, and $b$ and $c$. Theorem \[thm\_semi-stationary\_Langevin\] and the same calculations as those in Remark \[relation\_between\_Langevin\_and\_mapping\] yield that $$\Phi_b(\mu)={\mathcal L}(Z_t),\quad\text{for all }t\in{\mathbb{R}}.$$ Hence the mapping $\Phi_b$ can be defined also as the distribution of the semi-stationary solution of the Langevin type equation .
Nested subclasses of $L(b,{{\mathbb R^d}})$ given by iterating the mapping $\Phi_b$
===================================================================================
We now go back to the mapping $\Phi_b$ itself again. The iterated mapping of $\Phi_b$ can be expressed by one stochastic integral as follows.
\[iteration\] Suppose $m\in{\mathbb{Z}}_+$. The domain of $\Phi_b^{m+1}$ is $$\label{iteration domain}
\mathfrak{D}(\Phi_b^{m+1})=I_{\log^{m+1}}({\mathbb{R}}^d).$$ Let $$f_m(u):= \int_0^u\binom{[v]+m}{m}dv,$$ and let $f_m^*$ be its inverse function. Then $$\Phi_b^{m+1}(\mu)={\mathcal L}\left(\int_0^\infty b^{-[f_m^*(t)]}dX_t^{(\mu)}\right),\quad \text{for }
\mu\in I_{\log^{m+1}}({\mathbb{R}}^d).$$
If we let $$\widetilde{f}_m(u):= \int_0^u\frac{v^m}{m!}dv,$$ then its inverse function is $\widetilde{f}_m^*(t)=\{(m+1)!\,t\}^{\frac{1}{m+1}}$ and $e^{-\widetilde{f}_m^*(t)}$ is the integrand of the stochastic integral of the iteration of the mapping in the case of $L({\mathbb{R}}^d)$, (see Remark 58 of @Sato's_book2003). One can see the difference between $f_m$ and $\widetilde{f}_m$ by $$f_m(u)= \int_0^u\binom{[v]+m}{m}dv=\int_0^u\frac{([v]+1)([v]+2)\cdots ([v]+m)}{m!}dv.$$
We prove the statement by induction. If $m=0$, the assertion is true by the definition of $\Phi_b$ and Proposition \[domain\_of\_Phib\]. Assume that the assertion is true for $0,1,\dots,m-1$ in place of $m$. Let $$\widetilde{\Phi}_{b,m+1}(\mu):={\mathcal L}\left(\int_0^\infty b^{-[f_m^*(t)]}dX_t^{(\mu)}\right).$$ Then $\mathfrak{D}^0(\widetilde{\Phi}_{b,m+1})=\mathfrak{D}(\widetilde{\Phi}_{b,m+1})=I_{\log^{m+1}}({\mathbb{R}}^d)$ due to Proposition 4.3 of @Sato2006b, where $\mathfrak{D}^0(\Phi_f)$ denotes the set of all $\mu\in I({{\mathbb R^d}})$ satisfying $\int_0^\infty|C_{\mu}(f(t)z)|dt<\infty$. If $\mu\in I_{\log^{m+1}}({\mathbb{R}}^d)\subset I_{\log^{m}}({\mathbb{R}}^d)$, then $\Phi_b^{m}(\mu)={\mathcal L}\left(\int_0^\infty b^{-[f_{m-1}^*(t)]}dX_t^{(\mu)}\right)$ by the assumption of induction, and $$\begin{aligned}
\notag
\int_0^\infty&\left|C_{\Phi_b^{m}(\mu)}(b^{-[t]}z)\right|dt\\
\notag&\leq \int_0^\infty dt\int_0^\infty \left|C_{\mu}\left(b^{-[f_{m-1}^*(s)]-[t]}z\right)\right|ds\\
\notag&=\int_0^\infty dt\int_0^\infty \left|C_{\mu}\left(b^{-\left[f_{m-1}^*(s)+[t]\right]}z\right)\right|ds\\
\notag&=\int_0^\infty dt\int_{[t]}^\infty \left|C_{\mu}(b^{-[u]}z)
\right|\binom{[u]{-}[t]{+}m{-}1}{m{-}1}du\\
\notag&=\int_0^\infty\left|C_{\mu}(b^{-[u]}z)\right|du\int_{0}^{[u]+1}
\binom{[u]{-}[t]{+}m{-}1}{m{-}1}dt\\
\notag&=\int_0^\infty\left|C_{\mu}(b^{-[u]}z)\right|\binom{[u]+m}{m}du\\
\label{Phi^m}&=\int_0^\infty\left|C_{\mu}(b^{-[f_m^*(t)]}z)\right|dt<\infty,\end{aligned}$$ since $\mu\in I_{\log^{m+1}}({\mathbb{R}}^d)=\mathfrak{D}^0(\widetilde{\Phi}_{b,m+1})$. Note that we have used above the formula $$\int_0^{n-k+1}\binom{n-[t]}{k}dt
=\sum_{j=0}^{n-k}\binom{n-j}{k}
=\binom{n+1}{k+1}\quad\text{for }k\leq n.$$ Hence $I_{\log^{m+1}}({\mathbb{R}}^d)\subset \mathfrak{D}(\Phi_b^{m+1})$. By similar calculations to , we have $$\int_0^\infty C_{\Phi_b^{m}(\mu)}(b^{-[t]}z)dt=\int_0^\infty C_{\mu}(b^{-[f_m^*(t)]}z)dt,
\quad\text{for }\mu\in I_{\log^{m+1}}({{\mathbb R^d}}),$$ where the use of Fubini’s theorem is permitted by the finiteness of . Thus $$\Phi_b^{m+1}(\mu)=\widetilde{\Phi}_{b,m+1}(\mu),\quad\text{for }\mu\in I_{\log^{m+1}}({{\mathbb R^d}}).$$ To conclude , it remains to prove that $\mathfrak{D}(\Phi_b^{m+1})\subset I_{\log^{m+1}}({{\mathbb R^d}})$. If $\mu\in I({{\mathbb R^d}})$ with Lévy measure $\nu$ satisfies $\mu\notin I_{\log^{m+1}}({{\mathbb R^d}})$, there exists $n\in\{0,1,\dots,m\}$ such that $\mu\in I_{\log^{n}}({{\mathbb R^d}})\setminus I_{\log^{n+1}}({{\mathbb R^d}})$ (consider $I_{\log^{0}}({{\mathbb R^d}})$ to be $I({{\mathbb R^d}})$). If $n=0$, $\mu\notin I_{\log}({{\mathbb R^d}})=\mathfrak{D}(\Phi_b)$ and thus $\mu\notin \mathfrak{D}(\Phi_b^{m+1})$. Suppose $n\geq 1$. Then $\Phi_b^n(\mu)$ is definable and equal to $\widetilde{\Phi}_{b,n}(\mu)$ by the assumption of induction. Denoting the Lévy measure of $\Phi_b^n(\mu)$ by $\nu_n$, we have $$\begin{aligned}
\int_{|x|>1}\log_b|x|&\nu_n(dx)=\int_0^\infty dt\int_{{\mathbb R^d}}\log_b^+ \left|b^{-[f_{n-1}^*(t)]}x\right|\nu(dx)\\
&=\int_{|x|>1}\nu(dx)\int_0^{f_{n-1}([\log_b|x|]+1)} \left(\log_b|x|-[f_{n-1}^*(t)]\right)dt\\
&\geq \int_{|x|>1}\nu(dx)\int_0^{(\log_b|x|)^n/n!} \left\{\log_b|x|-(n!\,t)^{\frac{1}{n}}\right\}dt\\
&=\int_{|x|>1}\frac{(\log_b|x|)^{n+1}}{(n+1)!}\nu(dx)=\infty.\end{aligned}$$ Thus, $\Phi_b^n(\mu)\notin I_{\log}({{\mathbb R^d}})=\mathfrak{D}(\Phi_b)$ and hence $\Phi_b^{m+1}(\mu)$ is not definable. Therefore $\mathfrak{D}(\Phi_b^{m+1})\subset I_{\log^{m+1}}({{\mathbb R^d}})$.
Recall that the definition of nested subclasses of $L(b,{{\mathbb R^d}})$ in @MaejimaNaito1998 mentioned in Introduction. The following theorem shows that $L_m(b,{{\mathbb R^d}})$ can be realized as $\mathfrak R (\Phi_b^{m+1})$.
Suppose that $b>1$ and $m\in{\mathbb{Z}}_+$. Then, $$\Phi_b^{m+1}\left(I_{\log^{m+1}}({{\mathbb R^d}})\right)=L_m(b,{{\mathbb R^d}}).$$
Let us show the statement by induction. If $m=0$, the assertion is Corollary \[s.s.d.\]. Assume that the assertion is true for $m{-}1$ in place of $m$.
We first show that $\Phi_b^{m+1}\left(I_{\log^{m+1}}({{\mathbb R^d}})\right)\supset L_m(b,{{\mathbb R^d}})$. If $\mu\in L_m(b,{{\mathbb R^d}})$, there exists $\rho\in L_{m-1}(b,{{\mathbb R^d}})$ satisfying $\widehat{\mu}(z)=\widehat{\mu}(b^{-1}z)\widehat{\rho}(z)$. The assumption of induction implies that $\rho=\Phi_b^m(\rho_0)$ for some $\rho_0\in\mathfrak{D}(\Phi_b^m)$. On the other hand, Theorem \[equivalence\_between\_decomposability\_and\_mapping\] yields that $\rho\in \mathfrak{D}(\Phi_b)$ and $\mu=\Phi_b(\rho)$. Hence $\rho_0\in \mathfrak{D}(\Phi_b^{m+1})$ and $\mu=\Phi_b^{m+1}(\rho_0)$. Thus $\mu\in\Phi_b^{m+1}\left(I_{\log^{m+1}}({{\mathbb R^d}})\right)$.
To show the converse inclusion of two sets, suppose $\mu\in\Phi_b^{m+1}\left(I_{\log^{m+1}}({{\mathbb R^d}})\right)$. Then, $\mu=\Phi_b^{m+1}(\rho)=\Phi_b\left(\Phi_b^{m}(\rho)\right)$ for some $\rho\in\mathfrak{D}(\Phi_b^{m+1})$. The assumption of induction implies that $\Phi_b^{m}(\rho)\in L_{m-1}(b,{{\mathbb R^d}})$, and Theorem \[equivalence\_between\_decomposability\_and\_mapping\] yields that $\widehat{\mu}(z)=\widehat{\mu}(b^{-1}z)\widehat{\Phi_b^{m}(\rho)}(z)$. Thus $\mu\in L_m(b,{{\mathbb R^d}})$.
Let $L_{\infty}(b,{{\mathbb R^d}})=\bigcap_{m=0}^\infty L_m(b,{{\mathbb R^d}})$. In @MaejimaSatoWatanabe2000, they studied $L_{\infty}(b,{{\mathbb R^d}})$ in the more general setting of operator semi-selfdecomposable distributions and as a special case, they proved that $
L_{\infty}(b,{{\mathbb R^d}})=\overline{\mathit{SS}(b,{{\mathbb R^d}})},
$ where $\mathit{SS}(b,{{\mathbb R^d}})$ is the class of all semi-stable distributions with span $b$ and $\overline{\mathit{SS}(b,{{\mathbb R^d}})}$ denotes the closure of $\mathit{SS}(b,{{\mathbb R^d}})$ taken under convolution and weak convergence. What we want to emphasize here is that we have characterized $L_m(b,{{\mathbb R^d}})$ as the range of the mapping $\Phi_b^{m+1}$, and so we can conclude the following. Note that since $L_m(b,{{\mathbb R^d}})\supset L_{m+1}(b,{{\mathbb R^d}})$, $\lim_{m\rightarrow \infty }L_m(b,{{\mathbb R^d}})=L_\infty(b,{{\mathbb R^d}})$.
\[closure\_of\_ss\] $$\lim_{m\rightarrow \infty }\Phi_b^{m+1}\left(I_{\log^{m+1}}({{\mathbb R^d}})\right)=L_{\infty}(b,{{\mathbb R^d}})=
\overline{\mathit{SS}(b,{{\mathbb R^d}})}.$$
In @MaejimaSato2009, they proved that the limits of nested classes of several classes in $I({{\mathbb R^d}})$ are identical with $L_{\infty}({{\mathbb R^d}})$, which is known to be the same as the closure of the class of all stable distributions on ${{\mathbb R^d}}$, $\overline{S({{\mathbb R^d}})}$, say. Then a natural question arose. Can we find mappings by which, as the limit of iteration, we get a larger or a smaller class than $\overline{S({{\mathbb R^d}})}$? It is easy to see that $L_{\infty}(b,{{\mathbb R^d}})\supsetneqq L_{\infty}({{\mathbb R^d}})$ so that $\overline{\mathit{SS}(b,{{\mathbb R^d}})}
\supsetneqq \overline{S({{\mathbb R^d}})}$. @Sato20072008 constructed mappings producing a class smaller than $\overline{S({{\mathbb R^d}})}$. Corollary \[closure\_of\_ss\] shows that a mapping $\Phi_b$ produces a larger class than $\overline{S({{\mathbb R^d}})}$ by iteration as a limit.
[**Acknowledgment.**]{} The authors would like to thank Ken-iti Sato for his valuable comments.
[18]{} natexlab\#1[\#1]{}url \#1[[\#1]{}]{}
Barndorff-Nielsen, O.E., Maejima, M. and Sato, K.: Some classes of multivariate infinitely divisible distributions admitting stochastic integral representations, (2006) 1–33.
Carr, P., Geman, H., Madan, D.B. and Yor, M.: (2007) 31–57.
Kondo, H., Maejima, M. and Sato, K.: (2006) 291–303.
Lindner, A. and Sato, K.: (2009) 250–274.
Maejima, M. and Naito, Y.: Semi-selfdecomposable distributions and a new class of limit theorems, (1998) 13–31.
Maejima, M. and Sato, K.: Semi-[L]{}[é]{}vy processes, semi-selfsimilar additive processes, and semi-stationary [O]{}rnstein-[U]{}hlenbeck type processes, (2003) 609–639.
Maejima, M. and Sato, K.: The limits of nested subclasses of several classes of infinitely divisible distributions are identical with the closure of the class of stable distributions, (2009) 119–142.
Maejima, M., Sato, K. and Watanabe, T.: Completely operator semi-selfdecomposable distributions, (2000) 235–253.
Pacheco-González, C.G.: , , 2009.
Rocha-Arteaga, A. and Sato, K.: , 2003.
Sato, K.: , Cambridge University Press, Cambridge, 1999.
Sato, K.: Stochastic integrals in additive processes and application to semi-[L]{}[é]{}vy processes, (2004) 211–236.
Sato, K.: (2006) 825–851.
Sato, K.: (2006) 23–39.
Sato, K.: (2006) 47–87.
Sato, K.: , 2007–2008.
Watanabe, T.: (2002) 631–665.
Wolfe, S.J.: (1983) 534–538.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present measurements of the $t \bar t$ production cross section in b-tagged lepton + jets events from $p \bar p$ collisions at 1.96 TeV using the CDF detector at Fermilab. B-jets are tagged with either a secondary vertex algorithm, or a soft lepton tagger that identifies muons from B hadron semileptonic decays. With Tevatron Run II data, we estimate the $t \bar t$ signal fraction in two different ways: by estimating the various background contributions, and by fitting directly the leading jet transverse energy spectrum for the signal and background contributions. A subset of the sample, with two secondary vertex tagged jets, yields a production cross section consistent with the inclusive measurements. Results are consistent with a Standard Model $t \bar t$ signal and current measurements of the top quark mass.'
address: |
Lawrence Berkeley National Laboratory\
1 cyclotron road, Bldg 50B-5239\
Berkeley CA 94720\
author:
- Henri Bachacou
title: 'Measurements of the $t \bar t$ production cross section at the Tevatron Run II CDF experiment using b-tagging'
---
Introduction
============
The Tevatron (Run II) collides protons and anti-protons head-on at a center-of-mass energy of 1.96 TeV. In such collisions, the Standard Model (SM) predicts a $t\bar t$ production cross section of $\sigma_{t \bar t} = 6.7^{+0.7}_{-0.9}$ pb at $m_{\rm t} = 175 \,{\rm GeV}/c^2$ [@theory]. Top quarks are expected to decay almost exclusively to a W boson and a b quark. When one W decays leptonically, the $t \bar t$ event contains a high transverse momentum lepton, missing energy from the unrecorded neutrino, and 4 high transverse momentum jets, 2 of which originate from b quarks. We use this decay channel to measure the total $t\bar t$ production cross section. A deviation from the predicted value would be an indication of new physics either in the production mechanism or in the top decay. We select events with an isolated electron E$_T$ (muon P$_T$) greater than 20 GeV, missing E$_T>$20 GeV and at least 3 jets with E$_T>$15 GeV and $|\eta| < 2.0$. Finally, we require at least one jet in the event to be identified as a heavy flavor jet, either using a secondary vertex algorithm (SECVTX), or a soft lepton tagger (SLT) that identifies muons from B hadron semileptonic decays. The analyzes using SECVTX (resp. SLT) are based on $162\,\mathrm{pb}^{-1}$ (resp. $194\,\mathrm{pb}^{-1}$) of data. The CDF detector is described in detail elsewhere [@CDF].
\[sec:secvtx\] Measurement with secondary vertex b-tagging.
===========================================================
We optimize the event selection by requiring that the total transverse energy in the event ($H_T$, the scalar sum of all jets $E_T$, lepton $p_T$, and missing $E_T$) be larger than 200 GeV. The SECVTX algorithm selects tracks within the jet with large impact parameter to reconstruct secondary vertices. Jets containing a secondary vertex more than $3\sigma$ away form the primary vertex (in the plane transverse to the beam) are identified as b-jets. After tuning the simulation on a control sample, the efficiency for tagging at least one jet in a $t \bar t$ event that passes all other selection requirements is ($53 \pm 4$) %. The main sources of background are W + Heavy Flavor events, W + light jets events where one jet is wrongly tagged, and QCD events that fake a W signal; they are estimated with techniques that use both Monte Carlo and data control samples. We expect $13.8\pm2.0$ background events and observe 48 events in the data; we measure a cross section of $5.6^{+1.2}_{-1.1}\rm{(stat.)}^{+1.0}_{-0.7}\rm{(syst.)}\,\rm{pb}$. Fig. \[fig:njets\_tags\] shows the number of candidate events vs jet multiplicity together with the expected background contributions. Fig. \[fig:ht\] shows the $H_T$ variable distribution of the candidates compared to the expected background and $t \bar t$ signal (normalized to 6.7 pb).
The sub-sample of events with at least two tagged jets contains 8 events, compared to an expected background of $0.9\pm0.2$ events, from which we measure a cross section of $5.4 \pm 2.2 \rm{(stat.)} \pm 1.1 \rm{(syst.)}\,\rm{pb}$.
Measurement with SECVTX using a kinematic fit.
==============================================
Instead of explicitly evaluating the contribution to the sample from backgrounds, one can extract the $t \bar t$ fraction by fitting some kinematic variable in the data. The leading jet $E_T$ variable was chosen for this purpose. Template shapes for the background are evaluated from the data; the template shape for $t \bar t$ is from Monte Carlo. The fit (Fig. \[fig:fit\]) measures a $t \bar t$ fraction of $(67^{+13}_{-16})$ %, leading to a cross section of $6.0^{+1.5}_{-1.8}\rm{(stat.)} \pm 0.8 \rm{(syst.)}\,\rm{pb}$.
Measurement with soft muon b-tagging.
=====================================
The muon SLT algorithm matches tracks in the central drift chamber with segments in the muon chambers. It uses a global $\chi^2$ built from the matching distributions, to define a pseudo-likelihood variable, $L$, that separates muon candidates from background. A jet is considered “tagged” if it contains an SLT muon with P$_T >$ 3 GeV/c, with $L <$3.5 and within $\Delta R <$0.6 of the jet axis. Efficiency and fake rate are measured on control samples. Backgrounds are estimated with techniques similar to Sec. \[sec:secvtx\]. We expect $11.6\pm1.5$ background events, and observe 20, and we measure a cross section of $4.2^{+2.9}_{-1.9}\rm{(stat.)} \pm 1.4 \rm{(syst.)}\,\rm{pb}$. Fig. \[fig:slt\] shows the jet multiplicity of the candidates compared to the expected background.
[0]{} M. Cacciari, et al., [*JHEP*]{} [**404**]{}, 68 (2004). F. Abe, et al., [*Nucl. Instrum. Methods Phys. Res.*]{} A [**271**]{}, 387 (1988); D. Amidei, et al., [*Nucl. Instrum. Methods Phys. Res.*]{} A [ **350**]{}, 73 (1994); F. Abe, et al., [*Phys. Rev.*]{} D [**52**]{}, 4784 (1995); P. Azzi, et al., [*Nucl. Instrum. Methods Phys. Res.*]{} A [ **360**]{}, 137 (1995); [*The CDFII Detector Technical Design Report*]{}, Fermilab-Pub-96/390-E
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a method to simulate color, 3-dimensional images taken with a space-based observatory by building off of the established [*shapelets*]{} pipeline. The simulated galaxies exhibit complex morphologies, which are realistically correlated between, and include, known redshifts. The simulations are created using galaxies from the 4 optical and near-infrared bands (*B*, *V*, *i* and *z*) of the Hubble Ultra Deep Field (UDF) as a basis set to model morphologies and redshift. We include observational effects such as sky noise and pixelization and can add astronomical signals of interest such as weak gravitational lensing. The realism of the simulations is demonstrated by comparing their morphologies to the original UDF galaxies and by comparing their distribution of ellipticities as a function of redshift and magnitude to wider HST COSMOS data. These simulations have already been useful for calibrating multicolor image analysis techniques and for better optimizing the design of proposed space telescopes.'
address:
- 'University of California, Berkeley, CA 94720'
- 'Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA 91109'
- 'California Institute of Technology, 1200 E. California Blvd., Pasadena, CA 91125'
- 'Royal Observatory, Blackford Hill, Edinburgh EH9 3HJ, UK'
- 'University of California, Riverside, CA 92521'
author:
- Matt Ferry
- Jason Rhodes
- Richard Massey
- Martin White
- Dan Coe
- Bahram Mobasher
title: 'Color, 3D simulated images with shapelets'
---
,
,
,
,
,
galaxies: fundamental parameters, statistics ,methods: statistical ,image simulations ,gravitational lensing
Introduction
============
As astronomical surveys become deeper and wider, analysis techniques correspondingly become more complex and demanding. To calibrate these methods, extensive work has already been invested in the simulation of monochromatic astronomical imaging. Simulation packages have been developed to incorporate a semi-analytic model of galaxy number counts and evolution [@EWBM], or to mimic the properties of real observations [@MRCB]. However, there are currently no packages able to create correlated images across several bands.
Multi-band image simulations are firstly useful to develop and calibrate analysis methods that use multicolor data. Many measurements in astronomy (for example photometry, astrometry and shape measurement) are “inverse problems,” where variation in a signal is easy to introduce but difficult to measure, usually due to complications involving observational seeing and noise. Simulated data provide the best way to calibrate such methods because these variables can be controlled. A known astronomical signal can be inserted into simulated data, and the accuracy of a method can be judged by examining any errors in its recovery.
One example is the measurement of weak gravitational lensing. In weak lensing, light from background galaxies is lensed by foreground matter distributions, causing a shear (distortion) of the background galaxies’ shapes. The distortion is easy to add during the construction of simulated data. Although the lensing signal is achromatic, color simulations can be used to test sophisticated measurement methods that take advantage of
- the increased number of shear-measurable galaxies if some galaxies are only sufficiently bright in certain bands,
- reduced [*noise*]{} on shear measurement (by $\sqrt{N}$) if the intrinsic shapes of galaxies are uncorrelated between $N$ bands, and
- reduced systematic [*bias*]{} on shear measurement if the intrinsic shapes of galaxies are correlated between bands [@JainJarvis].
One common challenge in weak lensing measurement is the deconvolution of galaxy shapes from the instrumental point-spread function (PSF). Since the PSF is different in each band, PSF-dependent biases will be averaged out by looking at multiple bands. Conversely, biases inherent to a method will not be ameliorated. Developing multicolor analysis techniques to exploit these tricks requires multicolor simulations.
Multi-band image simulations are also useful to optimize the design and improve the science case for planned, multi-band imaging surveys such as SNAP [@Aldering] or Euclid [@DUNE]. These surveys require multiple bands in order to observe different types of galaxies, observe objects typically obscured in other bands, observe objects out to different redshifts, and most importantly to obtain photometric redshifts for galaxies. Engineering requirements for the design of these instruments can be derived via image simulations by measuring the (often complex and subtle) effects of engineering parameters on scientific return [@High]. Predictions for the scientific return of a given mission can be similarly estimated [@WLII; @WLIII].
A full demonstration of the potential gains in multicolor shear measurement, or a full optimization of a future space-based lensing mission is beyond the scope of this paper. The purpose of this paper is to present a method for simulating deep, multi-color space-based images with correlated morphologies and redshifts. The simulation pipeline we present here will serve as a basis for performing the optimization of both shear measurement techniques and future space missions in future papers.
Our simulation pipeline generalizes the single-color method of [@MRCB], representing complex galaxy morphologies as “shapelets” [@Cartesian; @Polar]. Shapelet-based simulations are already widely used for weak lensing. The Shear TEsting Program (STEP) used similar simulated data to test and improve shape measurement and PSF correction methods [@Mass2006]. Our generalization to multi-band, 3-dimensional simulations thus increases the realism and utility of a well established technique.
This paper is organized as follows. In §2 we give a brief review of shapelets and how they can be used to generate simulated images. In §3 we present the methodology by which we create multi-band, 3-dimensional simulations. In §4 we test the realism of our simulations through comparison to the real HST data. Lastly, in §5, we discuss the conclusions and summarize our findings.
Background {#sec:back}
==========
Shapelets, or 2-dimensional Gaussian-weighted Laguerre polynomials, form a complete, orthonormal basis able to represent any localized image, including a galaxy shape, in a relatively small number of coefficients. Any image $f(\underline{x})$ can be represented as a linear combination of shapelet basis functions $\chi_{nm}(\underline{x};\beta)$: $$\label{eq:shape}
f(\underline{x}) = \sum_n \sum_m f_{nm} \chi_{nm} (\underline{x};\beta),$$
where $\underline{x}$ is the pixel position, $f_{nm}$ are the shapelet coefficients, and $\beta$ is a characteristic size [@Polar]. Shapelets also simplify the practical processes of image convolution and deconvolution. In real space, convolution is an expensive process with computation time scaling with the square of the number of pixels. With shapelets, convolution becomes a computationally inexpensive matrix operation [@Cartesian], and deconvolution merely requires a matrix inversion [@BR]. This is advantageous when it is necessary to deconvolve with a PSF and re-convolve with a different PSF.
Shapelet coefficients thus form a multi-dimensional parameter space that describes a galaxy. In general, any possible galaxy morphology can be thought of as a point in this multi-dimensional parameter space. When the shapelet coefficients of a set of observed galaxies are placed in this space, various correlations emerge. Different directions in parameter space correspond to characteristics of the galaxy such as size, ellipticity, or the number of spiral arms. A classic example of this effect is the Hubble tuning-fork diagram [@Hubble; @Sandage; @deV], which parametrizes galaxies’ ellipsoid, bulge/disk ratio, and how tightly wound the spiral arms are. The shapelets method increases the dimensionality of the parametrization with axes corresponding to galaxies’ magnitude, size ($\beta$), and polar shapelet coefficients [@MRCB].
Real galaxy morphologies only occupy a small region of this multi-dimensional parameter space. Most regions of parameter space, corresponding to random shapelet coefficients, do not produce an image that resembles a galaxy. To manufacture useful simulations, it is essential to map the region corresponding to morphologically realistic galaxies. This region will constitute a probability density function (PDF), from which we will be able to draw simulated galaxy images. To acquire the PDF, we begin with a sample of real galaxies. Of course, the PDF is only noisily sampled by this finite set of galaxies, so we smooth it to obtain an approximation to the true, underlying PDF.
Methodology
===========
In this section we present the methodology used to create the simulations. The process can be summarized in two steps: (1) shapelet catalog creation, and (2) image constitution.
Shapelet Catalog Creation
-------------------------
Since our goal is to simulate multicolor images, it is necessary to start from real, multicolor data. The best data for this are the Hubble Ultra-Deep Field (UDF) images. In this field, there are 8049 galaxies with a detection signal to noise ratio of at least 10 in any one band. Photometric redshifts for each galaxy are publicly available in the Coe *et al* [@photo-z] catalog.[^1]
We use the program *shex* from the Shapelets software package[^2] to decompose all of these galaxies, in all observed bands, into a linear combination of shapelet basis functions. We run *shex* up to a maximum radial oscillation value, or <span style="font-variant:small-caps;">n\_max</span> in the basis functions, of 20 in order to optimize decomposition. Although this is computationally expensive (<span style="font-variant:small-caps;">n\_max</span> = 20 corresponds to 231 coefficients), we are assured that the large objects are well modeled. This algorithm automatically copes with the varying pixel scale between optical and near-infrared imaging. To maximize the efficiency of the shapelet model, we iterate the center of each decomposition on the pixel grid, the maximum order $n_{\rm max}$, and the scale size $\beta$ of each decomposition independently in each band, using the algorithm discussed in [@Polar]. This number is recorded for later image reconstruction. We store catalogs of galaxy shapes in their raw form as well as deconvolved from the UDF PSF as modeled by the stars in the field.
To model $n$-band imaging, we thus increase the dimensionality of the shapelet parameter space $n$-fold. For example, while a bright object may be uniquely described in one band by 233 coefficients (including magnitude, size and redshift) if <span style="font-variant:small-caps;">n\_max</span>=20, it is now described by 932 coefficients. Though in the UDF $n=4$, our simulation software is not limited to this number and could accept an input catalog of galaxies with more bands in the future. The UDF galaxies in this highly dimensional parameter space automatically contain the correlations between shapelet coefficients necessary to produce realistic galaxy images.
We then smooth the finite number of points in shapelet parameter space, using an *Epanechnikov* kernel [@Epanech] with a different smoothing length, $\lambda_i$, for each parameter. Note that, if we choose $\lambda_i$ to be too small, the galaxy appears nearly unchanged, and we shall simply reproduce UDF galaxies in the simulated images; if we choose it to be too large, the galaxy is not realistic. Following the established smoothing scheme explored by [@MRCB], we smooth the complex polar shapelet coefficients in modulus and phase space, setting $\lambda_i=15^{\circ}$ for phases and the mean separation between nearest neighbors in that dimension for moduli. To perturb the galaxy redshifts slightly, we set the redshift smoothing length to be $\frac{1+z}{m}$, where $m$ is a free parameter. We choose to smooth over $1+z$ since it is used more frequently in determining cosmological parameters. We also choose $m$ to be 6 as a reasonable limit to conservative smoothing. If $m$ is chosen to be lower than 6, the high redshift objects could get smoothed to an unrealistically high redshift.
Image Creation {#sec:im_create}
--------------
For each simulated image, we generate a sufficient number of new galaxies that their density in the simulated image reproduces that in the UDF. In practice, rather than pixelating and drawing from the smoothed PDF, we use an equivalent Monte-Carlo bootstrap technique [@MRCB]. For each new galaxy, an original UDF galaxy is selected at random and perturbed in shapelet space, within the smoothing kernel, to create a new galaxy. We also append a mock catalog of photometric redshifts to these new galaxies. These redshifts are slightly perturbed from the original galaxy’s observed redshift via the same smoothing process, wile their distribution still follows the observed distribution in the UDF.
At this stage, a known weak lensing shear signal can be added to the objects. Similar effects could also be added to simulate, e.g. proper motions, photometric variability, or supernovae. The galaxy is finally convolved in shapelet space with the desired point spread function.
Once new objects are created, they are formed into a multi-band shape catalog to be arranged into new images. The objects are first re-composed into pixelated postage stamps, then placed into large, empty arrays. The placement of objects is done such that the object appears at the same (RA, Dec) position in each band, in flux units of photons per second per pixel, and with (for the sake of this paper) a constant pixel scale of $0.03$ arcsec per pixel. Together with the mock photometric redshift catalog, a 3-dimensional, color simulation is thus created.
The images are made realistic by adding both a sky background and shot noise. Figure $\ref{fig:sample}$ shows an example image. The simulations could also be made more realistic by adding cosmic rays, variable sky background/read noise, or charge transfer inefficiency trailing. The background noise could also be smoothed with a small kernel to approximate the effects of the `DRIZZLE` routine on stacked data from multiple, dithered exposures [@Drizzle]. On the one hand, `DRIZZLE` allows for a sharper pixel scale and correction for geometric distortions. On the other hand it produces correlated pixel noise. We have not enabled these options in the standard images studied in this paper, but intend to explore their effects in future work.
![Sample correlated image to UDF depth. The bands shown are, clockwise from top-left, *B* (top-left), *V* (top-right), *i* (bottom-left) and *z* (bottom-right). The images are displayed on the same logarithmic scale with the same contrast.[]{data-label="fig:sample"}](sims.ps)
Results
=======
We examine the morphological realism of our simulations by “blind-testing” them against the original UDF. We also create a “$\delta$-*function image*,” whereby the smoothing length ($\lambda_i$) for each coefficient is set to zero (ie, objects are not perturbed). This allows us to separate the morphological characteristics due to shapeletization and due to smoothing in shapelet space, the results of which are found in Table $\ref{tab:morph}$. Our tests are similar to those done in [@MRCB]. We first consider general comparisons using photometry and size from `SExtractor`. We then examine more advanced morphological tests.
We initially test our perturbed simulations against the UDF with a size-magnitude plot by plotting <span style="font-variant:small-caps;">fwhm</span> vs. AB magnitude. The i-band plot is given as a representative sample, as shown in Figure $\ref{fig:i_fwhm_v_mag}$. The distribution should be the same for the simulations as for the original UDF. We use `SExtractor` parameters <span style="font-variant:small-caps;">fwhm\_image</span> and <span style="font-variant:small-caps;">mag\_best</span>. We also test the simulations’ realism through a magnitude histogram, a histogram of ellipticity components $e_1$ and $e_2$, and a histogram of <span style="font-variant:small-caps;">fwhm</span>. Since one can always increase the number of simulated galaxies in an image, the histograms are normalized to the same area to better observe the simulations’ realism. We define the ellipticity components $e_1$ and $e_2$ to be
$$\left( e_1 \atop e_2 \right) \equiv {a^2-b^2 \over a^2+b^2} \left(\cos2\theta \atop \sin2\theta \right),$$
where $a$, $b$, and $\theta$ are the `SExtractor` parameters <span style="font-variant:small-caps;">a\_image</span>, <span style="font-variant:small-caps;">b\_image</span>, and <span style="font-variant:small-caps;">theta\_image</span>, namely the major and minor axes and the angle between the major axis and the horizontal. This convention is generally adopted in weak lensing.
The plots discussed above reveal a strong morphological agreement between the simulated images and the originals. There is a good representation of objects from AB magnitude 22 to 28 for all bands up to a normalization factor. The ellipticity histograms show a very strong representation of objects with $e_1$ and $e_2$ between $\pm 0.8$. Beyond this, objects are not as well represented on account of the difficulty in representing highly elliptical objects as shapelets. Lastly, the <span style="font-variant:small-caps;">fwhm</span> histogram as well shows a strong agreement across all sizes.
A more demanding test is provided by the morphological classification parameters asymmetry (A), concentration (C), and clumpiness (S). These morphology characteristics have been developed in [@BJC; @CGW; @C]. The *CAS* parameters are defined in this work slightly differently than in [@C]. We define the asymmetry, concentration, and clumpiness to be
$$A \equiv \frac{\sum |I_{x,y} - I_{x,y}^{180}|}{\sum |I_{x,y}|}$$
$$C \equiv 5 \times \log_{10}\left(\frac{r_{80}}{r_{20}}\right)$$
$$S(\sigma) \equiv 10 \times \frac{\Sigma_{xy} |I_{x,y} - I_{x,y}^{\sigma} |}{\Sigma_{xy} I_{x,y}},$$
where $I_{x,y}$ is the flux intensity at a given pixel, $I_{x,y}^{180}$ is the intensity at the point $180^{\circ}$ around the origin, $r_{80}$ and $r_{20}$ are radii containing 80$\%$ and 20$\%$ of the flux respectively, and $I_{x,y}^{\sigma}$ is the intensity after the image is smoothed with a Gaussian kernel of width $\sigma$. The definitions above do not include a correction for the background as the more typical versions do. This is noted, but the error should average to zero when many galaxies are used given that the noise characteristics are the same for both the UDF and the simulated images. We use a smoothing width $\sigma$ for $S$ to be 5 pixels. We also use the Petrosian Radius (R), defined to be the radius where the surface brightness at that radius is equal to 20$\%$ of the surface brightness integrated within that radius [@Petrosian]. Massey *et al* (2004) demonstrated that the monochromatic shapelet image simulations are consistent with real data by plotting A vs. C, A vs. S, and R vs. C for the simulations and for the real data. Another test is to check the mean and RMS values for A,C, and S for the simulations against the original UDF. It was found that the simulations relative to the Hubble Deep Field (HDF) demonstrated a roughly equal concentration while showing a lower asymmetry and clumpiness [@Polar]. Though the objects in the original simulations from the HDF have this discrepancy, it is relatively small, and it is concluded that the HDF simulations are realistic [@MRCB]. We demonstrate a similar recovery here.
Examining the *CAS* plots, we see generally a strong agreement between the simulated images and the originals. One will notice the spread in Petrosian Radius for the simulated images. This can be explained by the shapeletization of objects. We have chosen to optimize the shapelet decomposition of objects to completely model the wings of galaxies, but at the expense of sometimes truncating their central cusps. This causes an increase in the Petrosian Radius, but should not present a large problem to methods such as weak lensing since the shearing (and then PSF smearing) happens later; this is just a small change in the intrinsic shape of the individual object which varies far more than the shear signal anyway.
In computing our results, we rejected any major outliers in the simulated images, for there were, on occasion, hugely asymmetric objects with unrealistically high concentration and clumpiness indices. These objects were very large galaxies that had not properly decomposed into shapelets. These objects were flagged and not included in subsequent simulations. We also rejected any asymmetry, clumpiness, or concentration measurements in any band where a galaxy was so faint that the $CAS$ routines failed to converge. We also set the limiting magnitude for these statistics to be 28 as computed by `SExtractor` as a balance between believable measurements and sufficiently deep galaxies. The results from the i-band simulated images are presented as a representative sample in Figures $\ref{fig:i_fwhm_v_mag}$ through $\ref{fig:i_morph}$. A summary of results for all bands is presented in Table $\ref{tab:morph}$. The overall agreement is quite good with very little deviation from the original UDF.
As we shall use these simulations for weak lensing, of particular interest is the intrinsic ellipticity variance ($\sigma_\gamma$), as a function of magnitude and redshift for shear-measurable galaxies. We include $\sigma_\gamma(z,mag)$ in Figure $\ref{fig:sig_gamma}$. We calculate the shear of a galaxy with the weak lensing measurement method of *Rhodes, Refregier, and Groth* (hereafter RRG) [@RRG]. This is a mature method developed specifically for space-based weak lensing measurements and with thorough testing during analysis of many Hubble Space Telescope images: including the Groth Strip [@gs], the Medium Deep Survey [@mds], the STIS Parallel Survey [@stis_ps], and the COSMOS 2 Square Degree Survey [@cosmos]. This method was also used for testing during the development of the monochromatic shapelets image simulation pipeline described in [@MRCB]. We run the RRG pipeline on the simulated images exactly as we have run it on real HST data and make similar cuts on objects in order to obtain the most representative results possible.
We define a shear-measurable galaxy to be one that passes several cuts. We first discard faint galaxies with S/N less than 10, where the S/N is defined as the the ratio of the `SExtractor` parameters `flux_auto` to `fluxerr_auto`. We also remove galaxies with ellipticity $|e|>2$, after correction for the PSF [@cut]. Note that, in the presence of image noise, especially during the PSF correction stage, it is possible for a moment-based shape measurement method to produce a non-physical ellipticity $|e|>1$. This ellipticity cut also implicitly removes galaxies for which an iterative centroiding process in RRG failed to converge. This includes objects for which there was a large shift away from the initial position detected by `SExtractor`. They are usually blended objects or close pairs, for which an accurate shape measurement would be impossible anyway. We finally discard objects with size $d_{RRG} = \sqrt{\frac{1}{2}(I_{xx}+I_{yy})}$ (where $I_{xx}$ and $I_{yy}$ are the weighted second order moments) smaller than 1.2 times that of the PSF. It is important that the cuts we make on the galaxies useful for lensing be as realistic as possible. Given the long history of the RRG method in space-based weak lensing measurements we feel that running the RRG pipeline on the simulated images is the best way to make these cuts realistic.
In Fig. $\ref{fig:sig_gamma}$, we compare the RMS shear for our simulations with real data from COSMOS [@Alexie]. The high agreement within our sample error is indicative of the simulations’ realism. Running more simulations lowers the statistical error bars, but we are limited by sample variance due to the limited number of galaxies in the UDF. The high error bars seen in the figure reflect this uncertainty.
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
![<span style="font-variant:small-caps;">fwhm</span> vs. <span style="font-variant:small-caps;">mag</span> for i-band real UDF image (left) and simulated image (right).[]{data-label="fig:i_fwhm_v_mag"}](i_real_fwhm_v_mag_pert.ps "fig:") ![<span style="font-variant:small-caps;">fwhm</span> vs. <span style="font-variant:small-caps;">mag</span> for i-band real UDF image (left) and simulated image (right).[]{data-label="fig:i_fwhm_v_mag"}](i_sim_fwhm_v_mag_pert.ps "fig:")
---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Histograms of magnitudes (top-left), ellipticity $e$ (top-right), <span style="font-variant:small-caps;">fwhm</span> (bottom-left), and Petrosian Radius (bottom-right) for the i-band. Dashed lines refer to the original UDF while the solid lines refer to the simulated images.[]{data-label="fig:i_hist"}](i_sim_loghist_mag_pert.ps "fig:") ![Histograms of magnitudes (top-left), ellipticity $e$ (top-right), <span style="font-variant:small-caps;">fwhm</span> (bottom-left), and Petrosian Radius (bottom-right) for the i-band. Dashed lines refer to the original UDF while the solid lines refer to the simulated images.[]{data-label="fig:i_hist"}](i_sim_ellip_pert.ps "fig:")
![Histograms of magnitudes (top-left), ellipticity $e$ (top-right), <span style="font-variant:small-caps;">fwhm</span> (bottom-left), and Petrosian Radius (bottom-right) for the i-band. Dashed lines refer to the original UDF while the solid lines refer to the simulated images.[]{data-label="fig:i_hist"}](i_sim_hist_fwhm_pert.ps "fig:") ![Histograms of magnitudes (top-left), ellipticity $e$ (top-right), <span style="font-variant:small-caps;">fwhm</span> (bottom-left), and Petrosian Radius (bottom-right) for the i-band. Dashed lines refer to the original UDF while the solid lines refer to the simulated images.[]{data-label="fig:i_hist"}](pr.ps "fig:")
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
-------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------
![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_real_avc_pert.ps "fig:") ![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_real_avs_pert.ps "fig:") ![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_real_rvc_pert.ps "fig:")
![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_sim_avc_pert.ps "fig:") ![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_sim_avs_pert.ps "fig:") ![Morphology plots for i-band real UDF (top) and simulated images (bottom). []{data-label="fig:i_morph"}](i_sim_rvc_pert.ps "fig:")
-------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------
-- --
-- --
Filter Image $<$A$>$ rms A $<$C$>$ rms C $<$S$>$ rms S rms e
-------- ------------------- --------- ------- --------- ------- --------- ------- -------
B Real UDF 0.82 0.91 4.66 4.72 2.83 3.34 0.45
$\delta$-function 0.88 0.97 4.62 4.69 2.96 3.47 0.44
Perturbed 0.92 1.01 4.60 4.69 3.28 3.74 0.44
V Real UDF 0.88 1.02 4.87 4.94 2.73 3.46 0.43
$\delta$-function 0.89 0.97 4.81 4.88 2.68 3.06 0.42
Perturbed 0.98 1.07 4.84 4.95 3.27 3.74 0.43
i Real UDF 0.82 1.03 4.88 4.96 2.63 3.45 0.42
$\delta$-function 0.88 0.97 4.82 4.88 2.77 3.11 0.42
Perturbed 0.95 1.04 4.81 4.93 3.27 3.70 0.43
z Real UDF 0.75 0.87 4.72 4.80 2.76 3.19 0.43
$\delta$-function 0.88 0.95 4.80 4.91 3.05 3.39 0.41
Perturbed 0.90 1.03 4.70 4.82 3.17 3.53 0.43
: Final results table comparing morphologies of the UDF with the simulated images. The $\delta$-function images refer to simulated images created without perturbing the shapelet coefficients ($\lambda_i$). The perturbed images were created by the smoothing method discussed in §$\ref{sec:im_create}$. For all the images, the limiting AB magnitude was 28.[]{data-label="tab:morph"}
Conclusions
===========
We presented a method to create an arbitrary amount of 3-dimensional, color, simulated, unique deep space images. The simulations are created by perturbing a galaxy’s polar shapelet coefficients in such a way as to create unique but realistic objects. The previous simulation pipeline has been expanded to correlate morphologies across four wavelength bands and now also include a redshift distribution. Though we currently use four wavelength bands, our simulation pipeline is flexible enough to include an arbitrary number of colors, should such a data set become available and useful.
Our simulations were tested by comparing them to the original UDF images. They were found to have similar morphologies to the original galaxies. Additionally, the weak lensing cosmological properties of the simulations were tested against COSMOS HST data. Within reasonable error, our simulations were found to be consistent.
Acknowledgements {#acknowledgements .unnumbered}
================
The authors thank the Caltech SURF program, the Berkeley Physics Undergraduate Research Scholars program, and the Jet Propulsion Laboratory, run under a contract with NASA by Caltech, for their support. We would also like to thank Richard Ellis for further support and enthusiasm regarding the project and Alexandre Refregier for his ideas and enthusiasm. The project would not have gotten started on the right path were it not for Will High’s help. Thanks also to Peter Capak for useful insights in simultaneous detection with `SExtractor`. Dave Johnston provided the PSF for SNAP and was helpful in calibrating the measurements of $\sigma_\gamma$. Thanks also to Alexie Leauthaud for help in shear measurement. JR and MF were supported in part by NASA grant BEFS-399131.02.02.01.07.
[99.]{} Erben T., van Waerbeke L., Bertin E., Mellier Y. & Schneider P. 2001, A&A, 366, 717
Jarvis M. and Jain B. 2008, JCAP, 01, 003
Massey R., Refregier A., Conselice C., Bacon D., 2004 MNRAS 349, 214
Aldering, SNAP Collaboration SPIE, 4835
Refregier A. et al, SPIE, 6265
High F., Ellis R., Massey R., Rhodes J., Lamoureux J., SNAP Collaboration 2004AAS, 205.6502H
Massey R., Rhodes J., Refregier A., Albert J., Bacon D., Bernstein G., Ellis R., Jain B., McKay T., Perlmutter S., Taylor A., 2004, AJ, 127, 3089
Refregier A., et al. 2004, AJ, 127, 3102
Refregier A. 2003 MNRAS, 338, 35
Massey R., & Refregier A. 2005 MNRAS, 363, 197
Massey et al, 2007, MNRAS 376, 13
Refregier A. & Bacon D., 2003 MNRAS, 338, 48
Hubble E. 1926, ApJ, 64, 321
Sandage A. 1961, “The Hubble Atlas of Galaxies” (Carnegie Inst. Washington Publ. 618) (Washington: Carnegie Inst.)
de Vaucouleurs G. 1959, Hand. Physik, 53, 275
Szalay A., Connolly A., & Szokoly G. 1999, AJ, 117, 68
Bertin E. & Arnouts S. 1996, A&AS, 117, 393
Coe D., Ben[í]{}tez N., S[á]{}nchez S., Jee M., Bouwens R., Ford H. 2006, AJ, 132, 926
Refregier A. & Bacon D. 2003, MNRAS, 338, 48
Epanechnikov, V. 1969, “Nonparametric Estimation of a Multi-variate Probability Density,” Theory of Probability and Its Applications, 14, 153
Silverman B. 1986, “Density Estimation for Statistics” and Data Analysis (Chapman and Hall, London)
Fruchter A. & Hook R. 2002, PASP, 114, 144
Bershady M., Jangren A. & Conselice C. 2000, AJ, 119, 2645
Conselice C., Gallagher J. & Wyse R. 2002, AJ, 123, 2246
Conselice C. 2003, ApJS, 147, 1
Petrosian V. 1976, ApJ, 209, 1L
Rhodes J., Refregier A., & Groth E. 2000, ApJ, 536, 79
Rhodes J., Refregier A., Groth E. 2001, ApJ, 552L, 85
Refregier A., Rhodes J., Groth E. 2002, ApJ, 572L, 131
Rhodes J., Refregier A., Collins N., Gardner J., Groth E., Hill R. 2004, ApJ, 605, 29
Massey R., *et al* 2007, ApJS 172, 239
Rhodes J., Refregier A., Groth E. 2001, ApJ, 552L, 85
Hoekstra H., Mellier Y., van Waerbeke L., Sembolini E., Fu L., Hudson M., Parker L., Tereno I., Benabed K. 2006, ApJ, 647, 116.
Rhodes J., Refregier A., Massey R., Albert J., Bacon D., Bernstein G., Ellis R., Lampton M., Kim A., McKay T., Perlmutter S., SNAP collaboration 2004, Astropart. Phys. 20, 377
Leauthaud A., *et al* 2007, ApJS 172, 219
[^1]: Available for download at .
[^2]: Version 2.1$\beta$, available at .
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is inevitable to train large deep learning models on a large-scale cluster equipped with accelerators system. Deep gradient compression would highly increase the bandwidth utilization and speed up the training process but hard to implement on ring structure. In this paper, we find that redundant gradient and gradient staleness has negative effect on training. We have observed that in different epoch and different steps, the neural networks focus on updating different layers and different parameters. In order to save more communication bandwidth and preserve the accuracy on ring structure, which break the restrict as the node increase, we propose a new algorithm to measure the importance of gradients on large-scale cluster implementing ring all-reduce based on the size of the ratio of parameter calculation gradient to parameter value. Our importance weighted pruning approach achieved $64\times$ and $58.8\times$ of gradient compression ratio on AlexNet and ResNet50 on ImageNet. Meanwhile, in order to maintain the sparseness of the gradient propagation, we randomly broadcast the index of important gradients on each node. While the remaining nodes are ready for the index gradient and perform all-reduce update. This would speed up the convergence of the model and preserve the training accuracy.'
author:
- |
Zehua Cheng,Zhenghua Xu$^*$[^1]\
*State Key Laboratory of Reliability and Intelligence of Electrical Equipment*\
*Hebei University of Technology*\
`[email protected],[email protected]`\
bibliography:
- 'main.bib'
title: Bandwidth Reduction using Importance Weighted Pruning on Ring AllReduce
---
Introduction
============
Recently, deep learning have driven tremendous progress in computer vision, natural language processing, and forecasting. In order to have better performance, deeper and bigger models [@szegedy2015going] has trained on larger dataset which large-scale distributed training improves the productivity of training deeper and larger models [@jin2016scale; @dean2012large; @coates2013deep; @chilimbi2014project; @shi_modeling_2018].
However, training on large-scale dataset on distributed system also brought up some split-new problems. The extensive gradients and parameter synchronization extend communication time, which has led to the need for high-end bandwidth solutions for building distributed systems([@li2014scaling; @chen2016revisiting; @lian2015asynchronous; @sra2015adadelay]). However, to improve the training rate, the bandwidth and computing resource utilization are the major challenge for training deep learning model on large-scale distributed system. Synchronous stochastic gradient descent [@das2016distributed] is the most popular approach to scaling ImageNet([@deng2009imagenet]) training. However, to scale synchronous stochastic gradient descent required to increase the batch size requiring faster bandwidth. Many large-scale HPC solutions require Infiniband[@pfister2001introduction] to guarantee the communication bandwidth. Deep gradient compression ([@lin_deep_2017]) has significantly reduce the communication baxndwidth based on traditional distribution cluster by introducing momentum correction, local gradient clipping, momentum factor masking and warm-up training. However, deep gradient compression adopted on centralized deployment mode which the bottleneck is the high communication cost on the central nodes. [@gibiansky2017bringing] has proved that ring all-reduce approach came to have better performance on utilize the bandwidth resources. However, deep gradient compression would restricted by the increasingly dense gradient when the number of training nodes grows on ring structure and sorting according to the size of the gradient results in slower convergence rate. Many methodologies have proved to have weak performance on ring structure which restricted the development of large-scale deep learning training.
In this work, we propose importance weighted pruning on ring all-reduce structure to highly utilize the bandwidth and preserve the training accuracy. We build our large-scale distributed training cluster without using Infiniband and high-end GPUs(we only use NVIDIA GTX 1080ti) which have saved a lot of cost. We have observed that in different epoch and different step, the neural network focuses on updating different layers and different parameters. We further propose layer-wise method to determine the importance of gradient of each layer and send the selected influential gradient to update, leaving gradients accumulation locally to highly utilize the bandwidth. During the training process, as the learning rate continues to decrease, the judgment of the importance of the gradient is getting higher.
Related Work
============
Researchers have proposed many approaches to accelerate distributed deep learning training process. Distributed synchronous Stochastic Gradient Descent (SSGD) is commonly adopted solution for parallelize the tasks across machines. Message exchange algorithm would also speed up training process by make full use of the communication resources. [@you2017scaling] utilizing the characteristics of deep learning training process, using different learning rate for different layers based on the norm of the weights($||w||$) and the norm of the gradients($||\nabla w||$) for large batch training. The ratio of weights and gradients $(\frac{||w||}{||\nabla w||})$ varies significantly for different layers. But it is get a equal learning rate on the same layer and it doesn’t distinguish the different learning rate for the same layers because the ratio of weights and gradients ($\frac{||w||}{||\nabla w||}$) varies significantly for same layers.
**Deep Gradient Compression** Deep gradient compression implementing momentum correction and local gradient pruning on top of gradient sparseness to maintain model performance and save communication bandwidth. However, [@lin_deep_2017] copies in computing workers are trained in parallel by applying different subsets of data on centralized parameter server. Centralized distributed cluster restricted by the number of the nodes growth in which case indicated the limit of the training rate. However, implementing deep gradient compression on large-scale ring reduce structure would restricted by the node increases. Because as the number of ring nodes increases, the gradient on each node becomes denser as the ring reduce is performed, so network bandwidth cannot be saved. So if we took the top1% gradient on each node. As the node passes gradients to another, the worst case is that the top $k$ gradient is 2%. As the number of nodes increases, the gradient carried by the nodes will continue become denser. Therefore, we believe that the deep gradient compression lost the meaning of spreading the sparse gradient.
**Gradient Quantization and Sparsification** Quantizing the gradients to low-precision values reduce the communication bandwidth by cutting down the whole model which usually sacrifice a lot of precision ,like [@wen2017terngrad]. [@seide20141] proposed 1-bit SGD to reduce gradients transfer data size and achieved 10× speedup in traditional speech applications. [@strom2015scalable] proposed threshold quantization to only send gradients larger than a predefined constant threshold, but the predefined threshold is hard to determine.
**Message Exchange Algorithm** Cluster computing resources are restricted by bandwidth and updating strategies. Ring Allreduce([@gibiansky2017bringing]) is an algorithm with constant communication cost. The limit of the system is determined only by the slowest connection between GPUs, which is an ideal solution for large model with large amounts of data exchange. The algorithm proceeds an allgather operation after the scatter-reduce operation. Scatter-reduce operation would exchange the data and every GPU would ends up with a chunk of the final result. Allgather step will exchange those chunks ending up with the complete final result. [@zhao2013butterfly] improve the cluster performance with incremental update to solve the communication overhead through the network which has better performance on the traditional machine learning solutions like svm and logistic regression, but models like ResNet[@he2016deep] does not have better performance comparing to single ring allreduce approaches.
System Implementation
=====================
Ring Structure
--------------
The traditional parameter sever mode has two parts of communication. The node should first propagate the data to the parameter server to perform operations like sum and average([@li2014scaling; @li2014communication]). Then the parameter server broadcasts the processed gradient to the node server. This process limited by the increase of the number of node where both the calculation and communication bandwidth are highly required. The ring structure has less communication operations than the conventional structure.
![Algorithm Topology Structure[]{data-label="figure:algo_topology"}](figures/algorithm.pdf){width="6cm"}
Fig \[figure:algo\_topology\] has presented the topology of the traditional structure and ring structure. For the ring structure, since there is no central node, we randomly select serveral number of nodes to get the gradient importance to ensure the sparseness of the gradient.
Importance Weighted Pruning
---------------------------
Deep gradient compression([@lin_deep_2017]) has reduce the communication bandwidth by cutting down redundant gradients. [@lin_deep_2017] considered the sparse update strategy as important gradients. However, we have observed that the redundant gradients would also make an adverse effort on training, which called the staleness effect. Utilizing the characteristics of deep learning training process seems to have better performance on speed up training rate[@you2017scaling]. We solve this problem by sending the important gradients which depending on how much it can change the weight. And we set a threshold to determine which gradients should be transmitted. But a simple important gradient pruning can not adapted to the ring reduce. As the gradient reduce in ring, the sparseness of the gradient will continue to drop. To solve this problem, we randomly choose a node and broadcast its important gradient index, then reduce the important grad on each node.
Distributed training with vanilla momentum SGD on $N$ training nodes follows, $$g_{t} = mg_{t-1} + \sum_{k=1}^{N}(\bigtriangledown_{k,t}), \ \ \omega_{t+1} = \omega_{t} -\eta g_{t}$$ where $m$ is the momentum, $N$ is the number of training nodes, and $\bigtriangledown_{k,t}\ =\ \frac{1}{NB}\bigtriangledown f(x, \omega_{k,t})$, $B$ is the batchsize, $x$ is the data sampled from $\chi$ at iteration $t$, $\eta$ is the learning rate.
After $T$ iterations, the change for $i$-th layer weight $w^{(i)}$ could be changed as follows, $$\omega_{t+T}^{(i)}\ =\ \omega_{t}^{(i)} - \eta (... + (\sum_{\tau =0}^{T-2} m^{\tau})\bigtriangledown_{k, t+1}^{(i)} + (\sum_{\tau =0}^{T-1} m^{\tau})\bigtriangledown_{k, t}^{(i)})$$
In our approach, the important gradients will be transmitted and the residual will accumulated local with momentum. So, the momentum SGD with important gradient is changed as follow, $$s_{t} = sparse(mg_{t-1} + g_{t}), \ \ \tilde{\omega}_{t+1} = \omega_{t} -\eta s_{t}$$
As the update function shows, the parameter becomes differ, but in our algorithm, we choose the gradient which change the parameter greatly. So we could assume $\omega_{t+1} \approx \tilde{\omega}_{t+1}$, and this assumption would yield that $s_{t} \approx g_{t}$, follow this assumption, consider the $i$-th layer weight update after T iterations, the change in $\omega^{(i)}$ will as same as in Equation 2.
Random Gradient Selection
-------------------------
Gradient staleness can slow down convergence and degrade model performance. In our experiment, most of the parameters are updated between 100-300 steps. The dated gradient will lead to errors in the direction of parameter update. Therefore, the gradient not greater than threshold have a certain probability to update, which can resist the influence of gradient staleness to some extent.
$$P(update)=\frac{gradient\;importance}{threshold}$$
Layer-wise Solution
-------------------
In large-scale distributed cluster training, we have observed that not every layer contributes the same contribution, which explain the false frozen layer phenomena that after serveral training epochs, the parameters in some particular layer do not update. The mean and variance of the importance of each layer of gradient have certain differences. Determine the importance of each layer of gradients and send important gradients out for updates. Unimportant gradient local accumulation.
$$threshold=
\left\{
\begin{array}{lr}
\alpha_{epoch}+\beta_{epoch}\times \frac{var}{mean}(\frac{var}{mean}>C) \\
\alpha_{epoch}-\beta_{epoch}\times \frac{var}{mean}(\frac{var}{mean}>C)
\end{array}
\right.$$
{width="8cm"}
{width="8cm"}
Choosing $\frac{var}{mean}$ as the factor to hyperparameter to determine the threshold is because we observe that the larger the variance of the gradient importance, the more disordered the gradient and the farther away from the normal distribution, which is not conducive to convergence. so it is necessary to increase the threshold of the gradient importance. When it means very large, the gradient of this layer is very important, and it is beneficial for the update of the parameters, so the threshold needs to be lowered. At this point, introduce an appropriate threshold can speed up the convergence of the model which determined by mean and variance of the gradient importance.
{width="8cm"}
We can infer that the $\alpha_{epoch}$ hyperparameter is related to epoch at the epoch perspective. $\alpha_{epoch}$ can be set to a constant within a certain epoch interval, for example, $\alpha_{epoch}$ is a constant in epoch 0 to 20 interval. However, epoch 20 variance/mean ratio changes drastically which should slightly adjust the degree of compression. Using $-\beta_{epoch}\times\frac{variance}{mean}(\frac{variance}{mean}<C)$ to cut down the threshold otherwise perform $+\beta_{epoch}\times\frac{variance}{mean}(\frac{variance}{mean}<C)$ to add up the threshold to compress as many gradients as possible.
Experiments
===========
Experiments Settings
--------------------
We validate our approach on image classification task. We have tested ResNet50 and AlexNet on ImageNet and ResNet101 on CIFAR10. In this process, we has implemented warm-up training and local gradient clip. In our experiments, we regulated the importance of the threshold in different steps. We have set our threshold to 0.005, 0.01, 0.05, 0.1. We evaluate the reduction in the network bandwidth by the gradient compression ratio based on the work of [@lin_deep_2017].
$$Gradient Compression Ratio = \frac{size[encode(sparse(G^k))]}{size[G^k]}$$
Because some gradients cannot be updated, in our method, if the importance of the gradient is less than the threshold, we will let the gradient be updated with a certain probability. This probability is positively related to the importance of the gradient.
In our experiments, we have analysis the variance of importance of the hierarchical gradients distribution. In order to prevent the gradient staleness, we determine the probability of updating the gradient
Model Result and anlysis
------------------------
We have tested resnet50 on ImageNet dataset on our experiment environment which Adopting ring-reduce network topology structure that has 96 nodes. Each node has only one GTX1080ti.
{width="7cm"}
{width="7cm"}
[llll]{} &&&\
\
AlexNet &Baseline &58.17% &$1\times$\
&TernGrad&57.28%& $8\times$\
&Fix Threshold&57.98%&$\bf{64\times}$\
&Layerwise Threshold&**58.19%** &$53\times$\
Resnet50 &Baseline &76.09% &$1\times$\
&Fix Threshold &76.26% &$58.8\times$\
&**Layerwise Threshold**&**76.31%** &$\bf{47.6\times}$\
We can observe that our proposed method can guarantee the speed increase under the condition of ensuring a certain precision. In the compression efficiency for the gradient, we can achieve an improvement of about 58 times, and in the case of layering, It can also achieve a 47.6 times improvement.
We also observed that [@lin_deep_2017] was performed in our experimental environment, and when performing loop propagation, the original sparse gradient would become denser.
Network bandwidth analysis
--------------------------
In our approach, we have significantly reduce the bandwidth communication. In Figure \[fig:baseline\_bandwidth\], we have observed that in the traditional structure, the Networks I/O bandwidth is close to full load. And easily tell the difference between Figure \[fig:baseline\_bandwidth\] and Figure \[fig:exp\_bandwidth\], we have significantly reduce the Networks I/O.
We can guaranteed the effective communication speed on Gigabit with our importance weighted pruning strategy.
Conclusion
==========
We propose a new algorithm for measuring the importance of gradients on large-scale cluster implementing ring all-reduce based on the size of the ratio of parameter calculation gradient to parameter value which has speed up the training process and persevering inference accuracy. We made a large-scale deep learning training cluster without using Infiniband. The redundant gradient takes up bandwidth and affect the training process. Our importance weighted pruning on ring all-reduce solution would solve adverse effects of redundant gradient and fully utilize the bandwidth without deficiency when the number of nodes grows.
[^1]: $^*$Corresponding Author
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A first order formulation for the Maxwell field in five dimensions is dimentionally reduced using the Randall-Sundrum mechanism. We will see that massive photons can not be localized on the brane.'
---
\
\
[ Instituto de Física, Universidad Nacional Autónoma de México\
Apdo. Postal 20-364, 01000 México D. F. México]{}\
and
[*Centro de Astrofísica Teórica, Departamento de Física, Facultad de Ciencias, Universidad de los Andes,\
Mérida, 5101,Venezuela.*]{}\
email: [email protected]. [email protected]
In the Randall-Sundrum mechanism[@rs], gravitons and scalar fields in $D = d + 1$ dimensions described by the Einstein-Hilbert and Klein-Gordon actions respectively, lead to gravitons and scalar fields localized on a d-dimensional brane. But, the usual (second order) Maxwell action can not yield photons trapped on the brane[@kss]. A solution of this dilemma was achieved in ref.[@lp] where the photons localized on the brane in four dimensions come from two antisymmetric tensors in five dimensions, satisfying a first order self-dual action[@tpvn]. Odd dimensional self-dual actions can be obtained from Kaluza-Klein dimentional reduction [@nvv]. Specifically, Lu and Pope[@lp] found that the bosonic sector of ungauged $N = 2$, $D = 4$ supergravity can be obtained from $N = 4$, $D = 5$ gauged supergravity. The supersymmetric extension was considered by Duff, et al.[@dls] . The generalization for higher antisymmetric fields (p-forms) gives some negative results. Indeed, if we consider the Randall-Sundrum ansatz for the metric $$ds^2 = e^{-2k|z|}g_{mn(x)}dx^m dx^n + dz^2 ,$$ where $m,n = 0,1,...,d$, together with the natural anzatz for a p-form $$A_{M_{1}...M_{p}(x,z)} = A_{M_{1}...M_{p}(x)} ,$$ then, the p-form is localized on the brane if $$p < \frac{d - 2}{2} .$$ For $d = 4$ (five dimensions), we have $p < 1$ and only scalar fields ($0$-form) can be trapped on the brane. But, a scalar field is dual to a $3$-form in five dimensions, which is not localized on the brane if the anzatz (2) is considered. In consequence, apparently, the Randall-Sundrum mechanism can not explain the dual equivalence of $p$-forms. However, Duff and Liu [@dl] have solved this dilemma. They found that the right anzatz for $p$-forms is given by $$A_{m_{1}...m_{p-1}z(x,z)} =
e^{-2(p - \frac{d}{2})k|z|} A_{m_{1}...m_{p-1}(x)},
\quad A_{m_{1}...m_{p}(x,z)} = 0 .$$ With this anzatz, the duality between $p$-form and $(D - p -2)$-form in the bulk implies the duality between $(p - 1)$-form and $(D -(p-1) -2)$-form on the brane. The criterion for consistency in this case is $$p > \frac{d}{2} .$$ If $d = 4$, we see that photons are excluded too. Thus, in five dimensions, p-forms with $p = 0$ and $p > 2$ can be trapped on the brane, ruled out the possibility of bounding photons (1-form) and its dual partner (2-form) on the brane[@dl]. The consistency with the Einstein equation will impose additional restrictions [@dl]. These results are based on the consideration of a second order action for p-forms. If we try to modify the ansatz (2) for vectors, including an explicit dependence with $z$ in the following way: $A_{M} = (e^{-ak|z|}A_{m}, 0)$[@dls], then the action for massive vectors is obtained. In this note, we will see that not only massless vector fields but massive vector fields can not be localized on the brane in the framework of Randall-Sundrum dimentional reduction.
Let us start with the following first order action in five dimensions $$S= -\frac{1}{12}\int d^4 x dz [\epsilon^{MNPQR}H_{MNP}F_{QR} + \sqrt{-g} g^{MQ}
g^{NR}g^{PS}H_{MNP}H_{QRS}],$$ where $F_{MN} = \partial_{M}A_{N} - \partial_{N}A_{M}$. $H_{MNP}$ and $A_{M}$ are independent fields. We can introduce the dual of $H_{MNP}$ ($= \frac{1}{2}\sqrt{-g}\epsilon_{MNPQR}f^{QR}$) and (6) becomes the usual first order formulation for the Maxwell action. Their equations of motion are $$H^{MNP} = -\frac{1}{2\sqrt{-g}}\epsilon^{MNPQR}F_{QR}$$ and $$\epsilon^{MNPQR}\partial_{N}H_{PQR} = 0 .$$
Substituting eq. (7) into action, we obtain the second order Maxwell action. On the other hand, eq. (8), can be solved (locally) $$H_{MNP} = \partial_{M}B_{NP} + \partial_{N}B_{PM} + \partial_{P}B_{MN}$$ and substituting in the action, it becomes the action for an antisymmetric field ($B_{MN}$). In other words, the action (6) show us the dual equivalence between $A_{M}$ and $B_{MN}$ in five dimensions.
Now, we apply the Randall-Sundrum anzatz for the metric (eq. (1)). The ansatz for the $H_{MNP}$ field is $$H_{mnz(x,z)} = e^{-k|z|}b_{mn(x)}, \quad H_{mnp(x,z)} = 0$$ and for the vector field $$A_{m(x,z)} = e^{-k|z|}A_{m(x)}, \quad A_{z(x,z)} = 0 .$$
The following reduced action is obtained $$S = \int dz e^{-2k|z|} \int d^4 x [-\frac{1}{4}\epsilon^{mnpq}b_{mn}F_{pq} -
\frac{1}{4}\sqrt{-g_{4}}b_{mn}b^{mn} ].$$ The (auxiliary) field $b_{mn}$ can be eliminated using its equation of motion $$b^{mn} = -\frac{1}{2\sqrt{-g_{4}}}\epsilon^{mnpq}F_{pq}$$ and the Maxwell action trapped on the brane is obtained $$S = \int dz e^{-2k|z|} \int d^4 x [ -\frac{1}{4}\sqrt{-g_{4}}F_{mn}F^{mn} ] .$$ Although the Maxwell action appear on the brane, the ansatz is not consistent with the Einstein equation with cosmological constant ($\Lambda = -6k^{2}$). Indeed, we have the Einstein equation $$R_{MN} - \frac{1}{2}g_{MN}R = g_{MN}\Lambda + T_{MN}$$ where $$T_{MN} = -\frac{1}{2}g^{PR}g^{QS}H_{MPQ}H_{NRS} + \frac{1}{12}
g_{MN}H_{PQR}H^{PQR} .$$ Substituting the ansatz (1) and (10), we find for the $mn$ components $$R_{4mn} - \frac{1}{2}g_{mn}R_{4} = g^{pq}F_{mp(A)}F_{nq(A)} - \frac{1}{4}
g_{mn}g^{pr}g^{qs}F_{pq(A)}F_{rs(A)}$$ where we have used eq. (13). Note that both sides of eq. (17) depend only of the coordinates $x$. Taking trace, we have $R_{4} = 0$. Then, the $zz$ component leads to an inconsistency $$F_{mn}F^{mn} = 0.$$
Then, there is no consistency with the Einstein equation. This situation is similar to what happen in the usual Kaluza-Klein dimentional reduction if the zz component of the metric is $g_{zz} = 1$, instead of having $g_{zz} = \phi$. But the Randall-Sundrum ansatz does not admit any scalar field. Moreover, Duff and Liu [@dl] have shown that massless scalar and third rank antisymmetric fields, in five dimensions are the p-forms compatible with the Randall-Sundrum dimentional reduction when the coupling with gravity is considered. Furthermore, the ansatz $H_{mnp(x,z)} = 0$ is inconsistent with the equations of motion (7) and (8). Then, we must modify the anzatz for $H_{mnp(x,z)}$. This will lead to the action of a massive vector field on the brane, but the same inconsistency in the Einstein equations will be present. For instance, we choose
$$H_{mnp(x,z)} = e^{-2k|z|}(\partial_{m}B_{np(x)} + \partial_{n}B_{pm(x)} +
\partial_{p}B_{mn(x)}) \equiv e^{-2k|z|} h_{mnp(x)},$$
where $B_{mn(x)}$ is a second rank antisymmetric field. The following reduced action on the brane is obtained
$$S = \int d^4 x dz e^{-2k|z|}[-\frac{1}{4}\sqrt{-g}F_{mn}F^{mn}
-\frac{1}{12}h_{mnp}h^{mnp} - \frac{1}{4}\mu_{z}\epsilon^{mnpq}B_{mn}F_{pq}]$$
where $\mu_{z} = ke^{-k|z|}$ is a mass parameter. We identify the Cremmer-Sherk action [@cs] on the brane, which describes massive vector fields in a gauge invariant way. This action can be obtained from Kaluza-Klein dimentional reduction [@adel]
The mn components of the Einstein equation are now $$\begin{aligned}
R_{4mn} - \frac{1}{2}g_{mn}R_{4} &=& g^{pq}F_{mp(A)}F_{nq(A)} - \frac{1}{4}
g_{mn}g^{pr}g^{qs}F_{pq(A)}F_{rs(A)} \\ \nonumber
&-& \frac{1}{2}g^{pr}g^{qs}h_{mpq}h_{nrs} + \frac{1}{12}g_{mn}h_{pqr}h^{pqr}.\end{aligned}$$
Taking trace, the following value of the scalar curvature is obtained $$R_{4} = \frac{1}{6}h_{mnp}h^{mnp}$$
Using this value of $R_{4}$, the zz component of the Einstein equation yields the same inconsistency found previously, i.e., $F_{mn}F^{mn} = 0$. The same result is obtained if another ansatz for $H_{mnp}$ is considered, e.g., $H_{mnp} = f_{(z)}\sqrt {-g} \epsilon_{mnpq} A^{q}$.
Summarizing, we have considered a first order formulation for the Maxwell action in five dimensions and we have seen that massive photons can not be localized on the brane in a consistent way, using the Randall-Sundrum mechanism.
[ACKNOWLEDGEMENTS]{}
The author would like to thank to Marti Ruiz Altaba for his hospitality at Instituto de Física de la Universidad Nacional Autónoma de México. Also, the author thanks Conicit-Venezuela for financial support.
[60]{}
L. Randall and R. Sundrum, Phys. Rev. Lett. [**83**]{} (1999) 4690.
N. Kaloper, E. Silverstein and L. Susskind, [**hep-th/0006192**]{}; A. Pomarol, Phys. Lett. [**B486**]{} (2000) 153 and Phys. Rev. Lett. [**85**]{} (2000) 4004.
H. Lu and C. N. Pope [**hep-th/0008050**]{}.
P. K. Townsend, K. Pilch and P. van Nieuwenhuizen, Phys. Lett. [**B136**]{} (1984) 38; S. Deser and R. Jackiw, Phys. Lett. [**B139**]{} (1984) 371.
H. Nastase, D. Vaman and P. van Nieuwenhuizen. Nucl. Phys. [**B581**]{} (2000) 179.
M. J. Duff, J. T. Liu and W. A. Sabra [**hep-th/0009212**]{}.
M. J. Duff and J. T. Liu [**hep-th/0010171**]{}.
E. Cremmer and J. Scherk Nucl. Phys. [**B72**]{} (1974) 117.
A. Khoudeir Phys.Rev. [**D59**]{} (1999) 027702.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a definition of equivariant (with respect to an Iwahori subgroup) $K$-theory of the formal power series model ${\mathbf{Q}_{G}}$ of semi-infinite flag manifold and prove the Pieri-Chevalley formula, which describes the product, in the $K$-theory of ${\mathbf{Q}_{G}}$, of the structure sheaf of a semi-infinite Schubert variety with a line bundle (associated to a dominant integral weight) over ${\mathbf{Q}_{G}}$. In order to achieve this, we provide a number of fundamental results on ${\mathbf{Q}_{G}}$ and its Schubert subvarieties including the Borel-Weil-Bott theory, whose special case is conjectured in [@BF14c]. One more ingredient of this paper besides the geometric results above is (a combinatorial version of) standard monomial theory for level-zero extremal weight modules over quantum affine algebras, which is described in terms of semi-infinite Lakshmibai-Seshadri paths. In fact, in our Pieri-Chevalley formula, the positivity of structure coefficients is proved by giving an explicit representation-theoretic meaning through semi-infinite Lakshmibai-Seshadri paths.'
author:
- |
Syu Kato\
Department of Mathematics, Kyoto University,\
Oiwake Kita-Shirakawa, Sakyo, Kyoto 606-8502, Japan\
(e-mail: [[email protected]]{})\
Satoshi Naito\
Department of Mathematics, Tokyo Institute of Technology,\
2-12-1 Oh-okayama, Meguro-ku, Tokyo 152-8551, Japan\
(e-mail: [[email protected]]{})\
and\
Daisuke Sagaki\
Institute of Mathematics, University of Tsukuba,\
1-1-1 Tennodai, Tsukuba, Ibaraki 305-8571, Japan\
(e-mail: [[email protected]]{})
title: |
**Equivariant $K$-theory of\
semi-infinite flag manifolds\
and Pieri-Chevalley formula[^1]**
---
Introduction. {#sec:intro}
=============
Let $G$ be a connected and simply-connected simple algebraic group over ${\mathbb{C}}$, and let $X$ be the flag variety of $G$. The torus-equivariant Grothendieck group $K_H ( X )$ of $X$ affords rich structures from the perspective of geometry and representation theory. One of the highlights there is the positivity of the structure constants of the products among natural classes (called the Schubert classes; see Anderson-Griffeth-Miller [@AGM], and Baldwin-Kumar [@BK]), which serves as a basis of its interaction with the eigenvalue problems [@Kly98] and Gaudin models [@MTV]. There is a variant of this theme (called the Pieri-Chevalley formula), namely the structure constants of the products between Schubert classes and (ample) line bundles in $K_H ( X )$, which is also known to be positive by Mathieu [@Mat00] and Brion [@Br02].
Pittie and Ram [@PR99] initiated a program to describe such a positive structure constant by relating them with the standard monomial theory (SMT for short). In particular, they gave an explicit meaning of each structure coefficient in the Pieri-Chevalley formula in terms of Lakshmibai-Seshadri paths (LS paths for short; see, e.g., [@Lit95]), which carries almost all information about simple $G$-modules. Their program is subsequently completed by Littelmann-Seshadri [@LiSe03] and Lenart-Shimozono [@LeSh14] (see also Lenart-Postnikov [@LeP07]).
Peterson [@Pet97] noticed that the quantum $K$-theory of $X$ should be intimately connected with the $K$-theory of the “affine version" of $X$ (see Lam-Shimozono [@LS10] and Lam-Li-Mihalcea-Shimozono [@LLMS]). In view of Givental-Lee [@GL03] and Braverman-Finkelberg [@BF14a; @BF14c], the quantum $K$-theory of $X$ can be defined through the space of quasi-maps, whose union forms a dense subset of the formal power series model ${\mathbf{Q}}_G$ of semi-infinite flag manifolds (cf. Finkelberg-Mirković [@FM99]).
Therefore, it is quite natural to make some rigorous sense of $K_H ( {\mathbf{Q}_{G}})$ and provide the Pieri-Chevalley formula using SMT, which is compatible with the pictures provided by Pittie-Ram and Peterson. This is what we perform in this paper by affording two new theories: [**1)**]{} the Borel-Weil-Bott theory of ${\mathbf{Q}_{G}}$ that enables us to define and calculate a version of $K_H ( {\mathbf{Q}_{G}})$, and [**2)**]{} the SMT of level-zero modules over quantum affine algebras. We remark that the level-zero modules over quantum affine algebras admit an interpretation through the geometry of affine Grassmannian (of Langlands dual type), which is the “affine version” of $X$ (see, e.g., Lenart-Naito-Sagaki-Schilling-Shimozono [@LNSSS2 Introduction]).
In order to explain our results, theories, and ideas more precisely, we need some notation. Let ${\mathfrak{g}}$ denote the Lie algebra of $G$, and let ${\mathfrak{g}}_{\mathrm{af}}$ denote the associated untwisted affine Lie algebra; we fix a Borel subgroup $B$ of $G$ and a maximal torus $H \subset B$, and set $N := [B,B]$. Let $W = N_{G}(H)/H$ be the Weyl group, which is generated by the simple reflections $s_{i}$, $i \in I$; $W$ can be thought of as acting on the dual space ${\mathfrak{h}}^{\ast}$ of the Cartan subalgebra ${\mathfrak{h}}:= {\mathop{\rm Lie}\nolimits}(H)$. We set $W_{{\mathrm{af}}} := W \ltimes Q^{\vee}$, with $Q^{\vee, +} := \sum_{i \in I} {\mathbb{Z}}_{\geq 0} \alpha_{i}^{\vee} \subset
Q^{\vee} := \bigoplus_{i \in I} {\mathbb{Z}}\alpha_{i}^{\vee}$ (the coroot lattice). Let $P = \bigoplus_{i \in I} {\mathbb{Z}}{\varpi}_{i} \subset {\mathfrak{h}}^{\ast}$ be the weight lattice generated by the fundamental weights ${\varpi}_{i}$, $i \in I$, and set $P^+ := \sum_{i \in I} {\mathbb{Z}}_{\ge 0} {\varpi}_{i}$.
For an algebraic group $E$ over ${\mathbb{C}}$, we denote by $E{(\hspace{-1pt}(z)\hspace{-1pt})}$ and $E{[\hspace{-1pt}[z]\hspace{-1pt}]}$ the space of ${\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$-valued points and the space of ${\mathbb{C}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$-valued points of $E$, respectively, viewed as an (ind-)scheme over ${\mathbb{C}}$. Let ${\mathop{\tt ev}\nolimits}_{0} : G{[\hspace{-1pt}[z]\hspace{-1pt}]} \rightarrow G$ be the evaluation map at $z = 0$, and set ${\mathbf{I}}:= {\mathop{\tt ev}\nolimits}_{0}^{-1}(B)$, an Iwahori subgroup of $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$; we also set $\tilde{{\mathbf{I}}} := {\mathbf{I}}\rtimes {\mathbb{C}}^{\ast}$, the semi-direct product group, where the group ${\mathbb{C}}^{\ast}$ (of loop rotations) acts on ${\mathbf{I}}$ as the dilation on $z$. Now, we define ${\mathbf{Q}_{G}^{\mathrm{rat}}}:= G{(\hspace{-1pt}(z)\hspace{-1pt})} / H N{(\hspace{-1pt}(z)\hspace{-1pt})}$, which is a pure ind-scheme of infinite type. Then the set of ${\mathbf{I}}$-orbits is in natural bijection with $W_{{\mathrm{af}}}$; let ${\mathbf{Q}_{G}}( x )$ denote the ${\mathbf{I}}$-orbit closure corresponding to $x \in W_{{\mathrm{af}}}$. We define ${\mathbf{Q}_{G}}:= {\mathbf{Q}_{G}}( e )$.
Our first main result is the following:
\[fnormal\] For each $x \in W_{\mathrm{af}}$, the scheme ${\mathbf{Q}_{G}}( x )$ is normal. In addition, there is an explicit $P^{+}$-graded algebra $R_G$ such that ${\mathbf{Q}_{G}}= {\mathop{\rm Proj}\nolimits}R_G$; here our ${\mathop{\rm Proj}\nolimits}$ is the $P^+$-graded one.
We remark that Theorem \[fnormal\] affirmatively answers [@BF14c Conjecture 2.1] and relevant speculations therein. Also, as we see below, the scheme ${\mathbf{Q}_{G}}$ is far from being “compact” (cf. [@Kat17 Theorem A] and [@FGT (7.1)]). In order to prove Theorem \[fnormal\] naturally, we introduce a “semi-infinite" Bott-Samelson-Demazure-Hansen tower that yields a normal ring $R_G$. From the construction, $R_G$ contains the projective coordinate ring of ${\mathbf{Q}_{G}}$. Moreover, on the basis of the fact that $R_G$ is generated by the primitive degree terms, a detailed comparison with the computation for the dense subset in [@BF14c] implies that the inclusion must be an isomorphism.
For each $x \in W_{\mathrm{af}}$ and $\lambda \in P$, we have an associated $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-equivariant line bundle ${\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}(\lambda)$ over ${\mathbf{Q}_{G}}(x)$. Also, for each $x \in W$ and $\lambda \in P^+$, we have a Demazure submodule $V^{-}_{x} ( \lambda )$, in the sense of [@Kas05], of the level-zero extremal weight module $V ( \lambda )$ (of extremal weight $\lambda$) over the quantum affine algebra $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$ associated to ${\mathfrak{g}}_{{\mathrm{af}}}$.
\[fBWB\] For each $x \in W_{\mathrm{af}}$ and $\lambda \in P$, we have $${\mathop{\rm gch}\nolimits}H^{i} ( {\mathbf{Q}_{G}}( x ), {\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}( \lambda ) ) =
\begin{cases}
{\mathop{\rm gch}\nolimits}V^{-}_{x} ( - w_{\circ} \lambda ) & \text{\rm if $i = 0$ and $\lambda \in P^{+}$}, \\[1.5mm]
0 & \text{\rm otherwise}, \end{cases}$$ where ${\mathop{\rm gch}\nolimits}$ denotes the character taking values in $({\mathbb{Z}}{(\hspace{-1pt}(q^{-1})\hspace{-1pt})})[P]$, and $w_{\circ} \in W$ is the longest element.
The higher cohomology vanishing part of Theorem \[fBWB\] is based on the fact that the ring $R_G$ is free over a polynomial ring with infinitely many variables (Theorem \[R-free\]), which is also an interesting result in its own. We should mention that Theorem \[fBWB\] have an ind-model counterpart in [@BF14c], but there are no implications between these and the two proofs are totally different.
\[fLB\] For each $x \in W_{\mathrm{af}}$, every ${\widetilde{{\mathbf{I}}}}$-equivariant line bundle over the scheme ${\mathbf{Q}_{G}}( x )$ is isomorphic to some ${\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}( \lambda )$ up to character twist.
Although $R_G$ itself is highly infinite-dimensional (it is [*not*]{} even finitely generated), it admits a grading such that it is almost like an Artin algebra in a graded sense. Moreover, Proposition \[fLB\] supplies “graded indecomposable projectives” of $R_G$. These two facts, combined with Theorem \[fBWB\], assert that the category of ${\widetilde{{\mathbf{I}}}}$-equivariant sheaves on ${\mathbf{Q}_{G}}$ (and on ${\mathbf{Q}_{G}^{\mathrm{rat}}}$) behaves almost like the category of coherent sheaves on an affine scheme.
This series of observations enables us to define a reasonable variant of an equivariant $K$-group $K^{\prime}_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}})$ of ${\mathbf{Q}_{G}}$ (and $K_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}^{\mathrm{rat}}})$ of ${\mathbf{Q}_{G}^{\mathrm{rat}}}$) with respect to $\tilde{{\mathbf{I}}}$; see Section \[sec:K-SiFl\] for details. They are rather involved, partly because we need to specify a class of formal power series that is large enough to afford the Pieri-Chevalley rule, and at the same time is small enough so that the Euler character map is injective. Nevertheless, we can prove that $K^{\prime}_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}})$ contains (the classes of) the sheaves $[{\mathcal{O}}_{{\mathbf{Q}_{G}}(y)} ( \lambda )]$ for each $\lambda \in P$ and relevant $y \in W_{{\mathrm{af}}}$. We also prove that our $K_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}}^{\mathrm{rat}})$ is natural enough so that it admits a nil-DAHA action as an analog of Kostant-Kumar [@KK90] for ${\mathbf{Q}_{G}^{\mathrm{rat}}}$ (see Section \[sec:nDAHA\] for details).
Here we recall that in Ishii-Naito-Sagaki [@INS] and Naito-Sagaki [@NS16], the semi-infinite path model of the crystal basis of $V_{x}^{-} ( \lambda )$ is constructed for every $\lambda \in P^+$ and $x \in W_{{\mathrm{af}}}$; it is a specific subset of the set of “semi-infinite” LS paths ${\mathbb{B}^{{\frac{\infty}{2}}}}( \lambda )$ of shape $\lambda$ parametrizing the global crystal basis of $V ( \lambda )$. Note that it is endowed with three functions $$\iota,\,\kappa : {\mathbb{B}^{{\frac{\infty}{2}}}}( \lambda ) \longrightarrow W_{\mathrm{af}}\quad \text{and} \quad
{\mathop{\rm wt}\nolimits}: {\mathbb{B}^{{\frac{\infty}{2}}}}( \lambda ) \longrightarrow P \oplus {\mathbb{Z}}\delta,$$ which are called the initial/final directions and the weight, respectively. We set $${\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} ( \lambda ) :=
\bigl\{ \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}( \lambda ) \mid \kappa ( \eta ) {\succeq}x \bigr\}.$$
In order to make use of the path model above to derive the Pieri-Chevalley formula for ${\mathbf{Q}_{G}^{\mathrm{rat}}}$, we additionally need a combinatorial version of the semi-infinite SMT. This consists of the definition of the initial direction ${\iota(\eta,\,x)} \in W_{{\mathrm{af}}}$ of a semi-infinite LS path $\eta$ with respect to $x$ (based on the existence of the semi-infinite analog of the so-called Deodhar lift), and of the description of tensor product decomposition of crystals in terms of ${\iota(\bullet,\,x)}$ (Theorem \[thm:SMT\] and Theorem \[thm:Dem\]). We remark that our ${\iota(\eta,\,x)}$ is an analogue of the one in [@LiSe03; @LeSh14] in the setting of level-zero extremal weight modules over $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$. Using them, we obtain our Pieri-Chevalley formula:
\[fPCf\] For $\lambda \in P^{+}$ and $x \in W_{{\mathrm{af}}}^{\ge 0}:=W \times Q^{\vee,+}$, we have $$\label{fPC}
[{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )] \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )}] =
\sum_{\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} (- w_{\circ} \lambda )}
e^{{\mathop{\rm fin}\nolimits}(\eta)}q^{{\mathop{\rm nul}\nolimits}(\eta)} \cdot [ {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\iota(\eta,\,x)} )}] \quad \in K^{\prime}_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}}),$$ where ${\mathop{\rm fin}\nolimits}(\eta) \in P$ and ${\mathop{\rm nul}\nolimits}(\eta) \in {\mathbb{Z}}$ for $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(- w_{\circ} \lambda )$ are defined by: $${\mathop{\rm wt}\nolimits}(\eta) = {\mathop{\rm fin}\nolimits}(\eta) + {\mathop{\rm nul}\nolimits}(\eta) \delta.$$
Once generalities on $K^{\prime}_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}})$ and the semi-infinite SMT are given, our strategy for the proof of Theorem \[fPCf\] is along the line of [@LiSe03]. Namely, we compare the functionals $$P \ni \lambda \mapsto \sum_{i \ge 0}(-1)^{i}
{\mathop{\rm gch}\nolimits}H^{i}( {\mathbf{Q}_{G}}, {\mathcal{E}}\otimes_{{\mathcal{O}}_{{\mathbf{Q}_{G}}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) ) \in {\mathbb{C}}[P]{(\hspace{-1pt}(q^{-1})\hspace{-1pt})},$$ where $[{\mathcal{E}}]$ is taken from the both sides of .
This paper is organized as follows. In Section \[sec:notation\], we fix our notation for untwisted affine Lie algebras, and then recall some basic facts about semi-infinite LS paths, extremal weight modules, and their Demazure submodules. In Section \[sec:SMT\], we state a combinatorial version of standard monomial theory for level-zero extremal weight modules, and also its refinement for Demazure submodules; the proofs of these results are given in Sections \[sec:prf-SMT\], \[sec:prf-DC\], and \[sec:prf-Dem\]. In Section \[sec:SiSch\], we first review the formal power series model ${\mathbf{Q}_{G}}$ of semi-infinite flag manifold, and then introduce a semi-infinite version of Bott-Samelson-Demazure-Hansen tower for ${\mathbf{Q}_{G}}$. Then, we study the cohomology spaces of line bundles over ${\mathbf{Q}_{G}}$, and prove the higher cohomology vanishing; also, we describe the spaces of global sections in terms of Demazure submodules of extremal weight modules. As an application, we prove the normality of the semi-infinite Schubert varieties ${\mathbf{Q}_{G}}(x)$, $x \in W_{{\mathrm{af}}}^{\geq 0}$. In Section \[sec:K-SiFl\], after giving a definition of ${\widetilde{{\mathbf{I}}}}$-equivariant $K$-group $K^{\prime}_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}})$ of ${\mathbf{Q}_{G}}$ (and $K_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}^{\mathrm{rat}}})$ of ${\mathbf{Q}_{G}^{\mathrm{rat}}}$), we establish the Pieri-Chevalley formula (Theorem \[thm:PC\]) by combining our geometric results with the semi-infinite SMT. Also, in Section \[sec:nDAHA\], we show that our $K$-group $K_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}^{\mathrm{rat}}})$ admits a natural nil-DAHA action. Appendices mainly contain some technical results concerning the semi-infinite Bruhat order; in particular, we prove the existence of analogs of Deodhar lifts for the semi-infinite Bruhat order.
Acknowledgments. {#acknowledgments. .unnumbered}
----------------
We thank Michael Finkelberg for sending us unpublished manuscripts. S.K. was partially supported by JSPS Grant-in-Aid for Scientific Research (B) 26287004 and Kyoto University Jung-Mung program. S.N. was partially supported by JSPS Grant-in-Aid for Scientific Research (B) 16H03920. D.S. was partially supported by JSPS Grant-in-Aid for Scientific Research (C) 15K04803.
Algebraic setting. {#sec:notation}
==================
Affine Lie algebras. {#subsec:liealg}
--------------------
A graded vector space is a ${\mathbb{Z}}$-graded vector space over ${\mathbb{C}}$ all of whose homogeneous subspaces are finite-dimensional. Let $V = \bigoplus_{m \in {\mathbb{Z}}} V_{m}$ be a graded vector space with $V_{m}$ its subspace of degree $m$. We define $${\mathop{\rm gdim}\nolimits}V := \sum_{m \in {\mathbb{Z}}} \bigl( \dim V_{m} \bigr) q^{m}.$$ Also, we denote by $V^{\vee}$ (resp., $V^{\ast}$) the full (resp., restricted) dual of $V$; note that $V^{\ast} := \bigoplus _{m \in {\mathbb{Z}}} ( V^{\ast} )_{m}$, with $(V^{\ast})_{m}:=(V_{-m})^{\ast}$. In addition, we set ${\widehat{V}}:=\prod_{m \in {\mathbb{Z}}} V_{m}$, which is a completion of $V$.
Let $G$ be a connected, simply-connected simple algebraic group over ${\mathbb{C}}$, and $B$ a Borel subgroup with unipotent radical $N$. We fix a maximal torus $H \subset B$, and take the opposite Borel subgroup $B^{-}$ of $G$ that contains $H$. In the following, for an (arbitrary) algebraic group $E$ over ${\mathbb{C}}$, we denote its Lie algebra ${\mathop{\rm Lie}\nolimits}(E)$ by the corresponding German letter ${\mathfrak{e}}$; in particular, we write ${\mathfrak{g}}={\mathop{\rm Lie}\nolimits}(G)$, ${\mathfrak{b}}={\mathop{\rm Lie}\nolimits}(B)$, ${\mathfrak{n}}={\mathop{\rm Lie}\nolimits}(N)$, and ${\mathfrak{h}}={\mathop{\rm Lie}\nolimits}(H)$. Thus, ${\mathfrak{g}}$ is a finite-dimensional simple Lie algebra over ${\mathbb{C}}$ with Cartan subalgebra ${\mathfrak{h}}$. Denote by $\{ \alpha_{i}^{\vee} \}_{i \in I}$ and $\{ \alpha_{i} \}_{i \in I}$ the set of simple coroots and simple roots of ${\mathfrak{g}}$, respectively, and set $Q := \bigoplus_{i \in I} {\mathbb{Z}}\alpha_i$, $Q^{+} := \sum_{i \in I} {\mathbb{Z}}_{\ge 0} \alpha_i$, and $Q^{\vee} := \bigoplus_{i \in I} {\mathbb{Z}}\alpha_i^{\vee}$, $Q^{\vee,+} := \sum_{i \in I} {\mathbb{Z}}_{\ge 0} \alpha_i^{\vee}$; for $\xi,\,\zeta \in Q^{\vee}$, we write $\xi \ge \zeta$ if $\xi-\zeta \in Q^{\vee,+}$. Let $\Delta$ and $\Delta^{+}$ be the set of roots and positive roots of ${\mathfrak{g}}$, respectively, with $\theta \in \Delta^{+}$ the highest root of ${\mathfrak{g}}$. For a root $\alpha \in \Delta$, we denote by $\alpha^{\vee}$ its dual root. We set $\rho:=(1/2) \sum_{\alpha \in \Delta^{+}} \alpha$ and $\rho^{\vee}:= (1/2) \sum_{\alpha \in \Delta^{+}} \alpha^{\vee}$. Also, let ${\varpi}_{i}$, $i \in I$, denote the fundamental weights for ${\mathfrak{g}}$, and set $$\label{eq:P-fin}
P:=\bigoplus_{i \in I} {\mathbb{Z}}{\varpi}_{i}, \qquad
P^{+} := \sum_{i \in I} {\mathbb{Z}}_{\ge 0} {\varpi}_{i}.$$
Let ${\mathfrak{g}}_{{\mathrm{af}}} = \bigl({\mathfrak{g}}\otimes {\mathbb{C}}[z,z^{-1}]\bigr) \oplus {\mathbb{C}}c \oplus {\mathbb{C}}d$ be the untwisted affine Lie algebra over ${\mathbb{C}}$ associated to ${\mathfrak{g}}$, where $c$ is the canonical central element, and $d$ is the scaling element (or the degree operator), with Cartan subalgebra ${\mathfrak{h}}_{{\mathrm{af}}} = {\mathfrak{h}}\oplus {\mathbb{C}}c \oplus {\mathbb{C}}d$. We regard an element $\mu \in {\mathfrak{h}}^{\ast}:={\mathop{\rm Hom}\nolimits}_{{\mathbb{C}}}({\mathfrak{h}},\,{\mathbb{C}})$ as an element of ${\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$ by setting ${\langle \mu,\,c \rangle}={\langle \mu,\,d \rangle}:=0$, where ${\langle \cdot\,,\,\cdot \rangle}:{\mathfrak{h}}_{{\mathrm{af}}}^{\ast} \times {\mathfrak{h}}_{{\mathrm{af}}} \rightarrow {\mathbb{C}}$ is the canonical pairing of ${\mathfrak{h}}_{{\mathrm{af}}}^{\ast}:={\mathop{\rm Hom}\nolimits}_{{\mathbb{C}}}({\mathfrak{h}}_{{\mathrm{af}}},\,{\mathbb{C}})$ and ${\mathfrak{h}}_{{\mathrm{af}}}$. Let $\{ \alpha_{i}^{\vee} \}_{i \in I_{{\mathrm{af}}}} \subset {\mathfrak{h}}_{{\mathrm{af}}}$ and $\{ \alpha_{i} \}_{i \in I_{{\mathrm{af}}}} \subset {\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$ be the set of simple coroots and simple roots of ${\mathfrak{g}}_{{\mathrm{af}}}$, respectively, where $I_{{\mathrm{af}}}:=I \sqcup \{0\}$; note that ${\langle \alpha_{i},\,c \rangle}=0$ and ${\langle \alpha_{i},\,d \rangle}=\delta_{i0}$ for $i \in I_{{\mathrm{af}}}$. Denote by $\delta \in {\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$ the null root of ${\mathfrak{g}}_{{\mathrm{af}}}$; recall that $\alpha_{0}=\delta-\theta$. Also, let $\Lambda_{i} \in {\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$, $i \in I_{{\mathrm{af}}}$, denote the fundamental weights for ${\mathfrak{g}}_{{\mathrm{af}}}$ such that ${\langle \Lambda_{i},\,d \rangle}=0$, and set $$\label{eq:P}
P_{{\mathrm{af}}} :=
\left(\bigoplus_{i \in I_{{\mathrm{af}}}} {\mathbb{Z}}\Lambda_{i}\right) \oplus
{\mathbb{Z}}\delta \subset {\mathfrak{h}}^{\ast}, \qquad
P_{{\mathrm{af}}}^{0}:=\bigl\{\mu \in P_{{\mathrm{af}}} \mid {\langle \mu,\,c \rangle}=0\bigr\};$$ notice that $P_{{\mathrm{af}}}^{0}=P \oplus {\mathbb{Z}}\delta$, and that $${\langle \mu,\,\alpha_{0}^{\vee} \rangle} = - {\langle \mu,\,\theta^{\vee} \rangle} \quad
\text{for $\mu \in P_{{\mathrm{af}}}^{0}$}.$$
Let $W := \langle s_{i} \mid i \in I \rangle$ and $W_{{\mathrm{af}}} := \langle s_{i} \mid i \in I_{{\mathrm{af}}} \rangle$ be the (finite) Weyl group of ${\mathfrak{g}}$ and the (affine) Weyl group of ${\mathfrak{g}}_{{\mathrm{af}}}$, respectively, where $s_{i}$ is the simple reflection with respect to $\alpha_{i}$ for each $i \in I_{{\mathrm{af}}}$, with length function $\ell:W_{{\mathrm{af}}} \rightarrow {\mathbb{Z}}_{\ge 0}$, which gives the one on $W$ by restriction; we denote by $e \in W_{{\mathrm{af}}}$ the identity element, and by $w_{\circ} \in W$ the longest element. For each $\xi \in Q^{\vee}$, let $t_{\xi} \in W_{{\mathrm{af}}}$ denote the translation in ${\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$ by $\xi$ (see [@Kac Sect. 6.5]); for $\xi \in Q^{\vee}$, we have $$\label{eq:wtmu}
t_{\xi} \mu = \mu - {\langle \mu,\,\xi \rangle}\delta \quad
\text{if $\mu \in {\mathfrak{h}}_{{\mathrm{af}}}^{\ast}$ satisfies ${\langle \mu,\,c \rangle}=0$}.$$ Then, $\bigl\{ t_{\xi} \mid \xi \in Q^{\vee} \bigr\}$ forms an abelian normal subgroup of $W_{{\mathrm{af}}}$, in which $t_{\xi} t_{\zeta} = t_{\xi + \zeta}$ holds for $\xi,\,\zeta \in Q^{\vee}$. Moreover, we know from [@Kac Proposition 6.5] that $$W_{{\mathrm{af}}} \cong
W \ltimes \bigl\{ t_{\xi} \mid \xi \in Q^{\vee} \bigr\} \cong W \ltimes Q^{\vee};$$ we also set $$\label{eq:Wge0}
W_{{\mathrm{af}}}^{\ge 0}:=\bigl\{wt_{\xi} \mid w \in W,\, \xi \in Q^{\vee,+} \bigr\} \subset W_{{\mathrm{af}}}.$$
Denote by ${\Delta_{{\mathrm{af}}}}$ the set of real roots of ${\mathfrak{g}}_{{\mathrm{af}}}$, and by ${\Delta_{{\mathrm{af}}}^{+}}\subset {\Delta_{{\mathrm{af}}}}$ the set of positive real roots; we know from [@Kac Proposition 6.3] that ${\Delta_{{\mathrm{af}}}}=
\bigl\{ \alpha + n \delta \mid \alpha \in \Delta,\, n \in {\mathbb{Z}}\bigr\}$, and ${\Delta_{{\mathrm{af}}}^{+}}=
\Delta^{+} \sqcup
\bigl\{ \alpha + n \delta \mid \alpha \in \Delta,\, n \in {\mathbb{Z}}_{> 0}\bigr\}$. For $\beta \in {\Delta_{{\mathrm{af}}}}$, we denote by $\beta^{\vee} \in {\mathfrak{h}}_{{\mathrm{af}}}$ its dual root, and $s_{\beta} \in W_{{\mathrm{af}}}$ the corresponding reflection; if $\beta \in {\Delta_{{\mathrm{af}}}}$ is of the form $\beta = \alpha + n \delta$ with $\alpha \in \Delta$ and $n \in {\mathbb{Z}}$, then $s_{\beta} =s_{\alpha} t_{n\alpha^{\vee}} \in W \ltimes Q^{\vee}$.
Finally, let $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$ denote the quantized universal enveloping algebra over ${\mathbb{C}}({\mathsf{q}})$ associated to ${\mathfrak{g}}_{{\mathrm{af}}}$, with $E_{i}$ and $F_{i}$, $i \in I_{{\mathrm{af}}}$, the Chevalley generators corresponding to $\alpha_{i}$ and $-\alpha_{i}$, respectively. We denote by $U_{{\mathsf{q}}}^{-}({\mathfrak{g}}_{{\mathrm{af}}})$ the negative part of $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$, that is, the ${\mathbb{C}}({\mathsf{q}})$-subalgebra of $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$ generated by $F_{i}$, $i \in I_{{\mathrm{af}}}$.
Parabolic semi-infinite Bruhat graph. {#subsec:SiBG}
-------------------------------------
In this subsection, we fix a subset $J \subset I$. We set ${Q_{J}}:= \bigoplus_{i \in {J}} {\mathbb{Z}}\alpha_i$, ${Q_{J}^{\vee}}:= \bigoplus_{i \in {J}} {\mathbb{Z}}\alpha_i^{\vee}$, ${Q_{J}^{\vee,+}}:= \sum_{i \in {J}} {\mathbb{Z}}_{\ge 0} \alpha_i^{\vee}$, ${\Delta_{J}}:= \Delta \cap {Q_{J}}$, ${\Delta_{J}}^{+} := \Delta^{+} \cap {Q_{J}}$, and ${W_{J}}:= \langle s_{i} \mid i \in {J}\rangle$. Also, we denote by $$\label{eq:prj}
[\,\cdot\,]_{{J}} :
Q^{\vee} \twoheadrightarrow {Q_{J}^{\vee}}\quad
\text{(resp., $[\,\cdot\,]^{{J}} : Q^{\vee} \twoheadrightarrow Q_{I \setminus {J}}^{\vee}$)}$$ the projection from $Q^{\vee}=Q_{I \setminus {J}}^{\vee} \oplus {Q_{J}^{\vee}}$ onto ${Q_{J}^{\vee}}$ (resp., $Q_{I \setminus {J}}^{\vee}$) with kernel $Q_{I \setminus {J}}^{\vee}$ (resp., ${Q_{J}^{\vee}}$). Let ${W^{J}}$ denote the set of minimal(-length) coset representatives for the cosets in $W/{W_{J}}$; we know from [@BB Sect. 2.4] that $$\label{eq:mcr}
{W^{J}}= \bigl\{ w \in W \mid
\text{$w \alpha \in \Delta^{+}$ for all $\alpha \in {\Delta_{J}}^{+}$}\bigr\}.$$ For $w \in W$, we denote by ${\lfloor w \rfloor}={\lfloor w \rfloor}^{{J}} \in {W^{J}}$ the minimal coset representative for the coset $w {W_{J}}$ in $W/{W_{J}}$. Also, following [@Pet97] (see also [@LS10 Sect. 10]), we set $$\begin{aligned}
({\Delta_{J}})_{{\mathrm{af}}}
& := \bigl\{ \alpha + n \delta \mid
\alpha \in {\Delta_{J}},\,n \in {\mathbb{Z}}\bigr\} \subset \Delta_{{\mathrm{af}}}, \\
({\Delta_{J}})_{{\mathrm{af}}}^{+}
&:= ({\Delta_{J}})_{{\mathrm{af}}} \cap {\Delta_{{\mathrm{af}}}^{+}}=
{\Delta_{J}}^{+} \sqcup \bigl\{ \alpha + n \delta \mid
\alpha \in {\Delta_{J}},\, n \in {\mathbb{Z}}_{> 0} \bigr\}, \\
\label{eq:stabilizer}
({W_{J}})_{{\mathrm{af}}}
& := {W_{J}}\ltimes \bigl\{ t_{\xi} \mid \xi \in {Q_{J}^{\vee}}\bigr\}
= \bigl\langle s_{\beta} \mid \beta \in ({\Delta_{J}})_{{\mathrm{af}}}^{+} \bigr\rangle, \\
\label{eq:Pet}
({W^{J}})_{{\mathrm{af}}}
&:= \bigl\{ x \in W_{{\mathrm{af}}} \mid
\text{$x\beta \in {\Delta_{{\mathrm{af}}}^{+}}$ for all $\beta \in ({\Delta_{J}})_{{\mathrm{af}}}^{+}$} \bigr\};\end{aligned}$$ note that if ${J}= \emptyset$, then $(W^{\emptyset})_{{\mathrm{af}}}=W_{{\mathrm{af}}}$ and $(W_{\emptyset})_{{\mathrm{af}}}=\bigl\{e\bigr\}$. We know from [@Pet97] (see also [@LS10 Lemma 10.6]) that for each $x \in W_{{\mathrm{af}}}$, there exist a unique $x_1 \in ({W^{J}})_{{\mathrm{af}}}$ and a unique $x_2 \in ({W_{J}})_{{\mathrm{af}}}$ such that $x = x_1 x_2$; we define a (surjective) map $$\label{eq:PiJ}
{\Pi^{J}}: W_{{\mathrm{af}}} \twoheadrightarrow ({W^{J}})_{{\mathrm{af}}}, \quad x \mapsto x_{1},$$ where $x= x_1 x_2$ with $x_1 \in ({W^{J}})_{{\mathrm{af}}}$ and $x_2 \in ({W_{J}})_{{\mathrm{af}}}$.
\[dfn:sell\] Let $x \in W_{{\mathrm{af}}}$, and write it as $x = w t_{\xi}$ for $w \in W$ and $\xi \in Q^{\vee}$. We define the semi-infinite length ${\ell^{{\frac{\infty}{2}}}}(x)$ of $x$ by: ${\ell^{{\frac{\infty}{2}}}}(x) = \ell (w) + 2 {\langle \rho,\,\xi \rangle}$.
\[dfn:SiB\]
The (parabolic) semi-infinite Bruhat graph ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ is the ${\Delta_{{\mathrm{af}}}^{+}}$-labeled, directed graph with vertex set $({W^{J}})_{{\mathrm{af}}}$ whose directed edges are of the following form: $x {\xrightarrow{\,\beta\,}} s_{\beta} x$ for $x \in ({W^{J}})_{{\mathrm{af}}}$ and $\beta \in {\Delta_{{\mathrm{af}}}^{+}}$, where $s_{\beta } x \in ({W^{J}})_{{\mathrm{af}}}$ and ${\ell^{{\frac{\infty}{2}}}}(s_{\beta} x) = {\ell^{{\frac{\infty}{2}}}}(x) + 1$. When $J=\emptyset$, we write ${\mathrm{BG}^{{\frac{\infty}{2}}}(W_{{\mathrm{af}}})}$ for $\mathrm{BG}^{{\frac{\infty}{2}}}\bigl((W^{\emptyset})_{{\mathrm{af}}}\bigr)$.
The semi-infinite Bruhat order is a partial order ${\preceq}$ on $({W^{J}})_{{\mathrm{af}}}$ defined as follows: for $x,\,y \in ({W^{J}})_{{\mathrm{af}}}$, we write $x {\preceq}y$ if there exists a directed path from $x$ to $y$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$; we write $x {\prec}y$ if $x {\preceq}y$ and $x \ne y$.
In the case $J = \emptyset$, the semi-infinite Bruhat order on $W_{{\mathrm{af}}}$ is just the generic Bruhat order introduced in [@Lu80]; see [@INS Appendix A.3] for details. Also, for a general $J$, the parabolic semi-infinite Bruhat order on $({W^{J}})_{{\mathrm{af}}}$ is nothing but the partial order on $J$-alcoves introduced in [@Lu97] when we take a special point to be the origin.
In Appendix \[sec:basic\], we recall some of the basic properties of the semi-infinite Bruhat order.
For $x \in ({W^{J}})_{{\mathrm{af}}}$, let ${\mathrm{Lift}}(x)$ denote the set of lifts of $x$ in $W_{{\mathrm{af}}}$ with respect to the map ${\Pi^{J}}:W_{{\mathrm{af}}} \twoheadrightarrow ({W^{J}})_{{\mathrm{af}}}$, that is, $$\label{eq:lift}
{\mathrm{Lift}}(x):=
\bigl\{ x' \in W_{{\mathrm{af}}} \mid {\Pi^{J}}(x') = x \bigr\};$$ for an explicit description of ${\mathrm{Lift}}(x)$, see Lemma \[lem:lift\]. The following proposition will be proved in Appendix \[sec:prf-Deo\].
\[prop:Deo\] If $x \in W_{{\mathrm{af}}}$ and $y \in ({W^{J}})_{{\mathrm{af}}}$ satisfy the condition that $y {\succeq}{\Pi^{J}}(x)$, then the set $${\mathrm{Lift}_{{\succeq}x}(y)}:=
\bigl\{ y' \in {\mathrm{Lift}}(y) \mid y' {\succeq}x \bigr\}$$ has the minimum element with respect to the semi-infinite Bruhat order on $W_{{\mathrm{af}}}$[;]{} we denote this element by $\min {\mathrm{Lift}_{{\succeq}x}(y)}$.
Semi-infinite Lakshmibai-Seshadri paths. {#subsec:SLS}
----------------------------------------
In this subsection, we fix $\lambda \in P^{+} \subset P_{{\mathrm{af}}}^{0}$ (see and ), and set $$\label{eq:J}
J:= \bigl\{ i \in I \mid {\langle \lambda,\,\alpha_i^{\vee} \rangle}=0 \bigr\} \subset I.$$
\[dfn:SBa\] For a rational number $0 < a < 1$, we define ${\mathrm{BG}^{{\frac{\infty}{2}}}_{\lambda,a}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ to be the subgraph of ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ with the same vertex set but having only the edges of the form $x {\xrightarrow{\,\beta\,}} y$ with $a{\langle x\lambda,\,\beta^{\vee} \rangle} \in {\mathbb{Z}}$.
\[dfn:SLS\] A semi-infinite Lakshmibai-Seshadri (LS for short) path of shape $\lambda $ is a pair $$\label{eq:SLS}
\pi = ({\mathbf{x}}\,;\, {\mathbf{a}})
= (x_{1},\,\dots,\,x_{s} \,;\, a_{0},\,a_{1},\,\dots,\,a_{s}), \quad s \ge 1,$$ of a strictly decreasing sequence ${\mathbf{x}}: x_1 {\succ}\cdots {\succ}x_s$ of elements in $({W^{J}})_{{\mathrm{af}}}$ and an increasing sequence ${\mathbf{a}}: 0 = a_0 < a_1 < \cdots < a_s =1$ of rational numbers satisfying the condition that there exists a directed path from $x_{u+1}$ to $x_{u}$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}_{\lambda,a_{u}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ for each $u = 1,\,2,\,\dots,\,s-1$. We denote by ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ the set of all semi-infinite LS paths of shape $\lambda$.
Following [@INS Sect. 3.1] (see also [@NS16 Sect. 2.4]), we endow the set ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ with a crystal structure with weights in $P_{{\mathrm{af}}}$ by the map ${\mathop{\rm wt}\nolimits}:{\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \rightarrow P_{{\mathrm{af}}}$ and the root operators $e_{i}$, $f_{i}$, $i \in I_{{\mathrm{af}}}$; for details, see Appendix \[sec:crystal\]. We denote by ${\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)$ the connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ containing $\pi_{\lambda}:=(e\,;\,0,1) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$.
If $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ is of the form , then we set $$\label{eq:dir}
\iota(\pi):=x_{1} \in ({W^{J}})_{{\mathrm{af}}} \qquad
\text{(resp., $\kappa(\pi):=x_{s} \in ({W^{J}})_{{\mathrm{af}}}$)};$$ we call $\iota(\pi)$ (resp., $\kappa(\pi)$) the initial (resp., final) direction of $\pi$. For $x \in W_{{\mathrm{af}}}$, we set $$\label{eq:SLS-dem}
{\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda) :=
\bigl\{
\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \mid \kappa(\pi) {\succeq}{\Pi^{J}}(x)
\bigr\}.$$
Extremal weight modules and their Demazure submodules. {#subsec:extremal}
------------------------------------------------------
In this subsection, we fix $\lambda \in P^{+} \subset P_{{\mathrm{af}}}^{0}$ (see and ). Let $V(\lambda)$ denote the extremal weight module of extremal weight $\lambda$ over $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$, which is an integrable $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$-module generated by a single element $v_{\lambda}$ with the defining relation that $v_{\lambda}$ is an “extremal weight vector” of weight $\lambda$; recall from [@Kas02 Sect. 3.1] and [@Kas05 Sect. 2.6] that $v_{\lambda}$ is an extremal weight vector of weight $\lambda$ if and only if ($v_{\lambda}$ is a weight vector of weight $\lambda$ and) there exists a family $\{ v_{x} \}_{x \in W_{{\mathrm{af}}}}$ of weight vectors in $V(\lambda)$ such that $v_{e}=v_{\lambda}$, and such that for every $i \in I_{{\mathrm{af}}}$ and $x \in W_{{\mathrm{af}}}$ with $n:={\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \ge 0$ (resp., $\le 0$), the equalities $E_{i}v_{x}=0$ and $F_{i}^{(n)}v_{x}=v_{s_{i}x}$ (resp., $F_{i}v_{x}=0$ and $E_{i}^{(-n)}v_{x}=v_{s_{i}x}$) hold, where for $i \in I_{{\mathrm{af}}}$ and $k \in {\mathbb{Z}}_{\ge 0}$, the $E_{i}^{(k)}$ and $F_{i}^{(k)}$ are the $k$-th divided powers of $E_{i}$ and $F_{i}$, respectively; note that the weight of $v_{x}$ is $x\lambda$. Also, for each $x \in W_{{\mathrm{af}}}$, we define the Demazure submodule $V_{x}^{-}(\lambda)$ of $V(\lambda)$ by $$\label{eq:dem}
V_{x}^{-}(\lambda):=U_{{\mathsf{q}}}^{-}({\mathfrak{g}}_{{\mathrm{af}}})v_{x}.$$
We know from [@Kas94 Proposition 8.2.2] that $V(\lambda)$ has a crystal basis ${\mathcal{B}}(\lambda)$ and the corresponding global basis $\bigl\{G(b) \mid b \in {\mathcal{B}}(\lambda)\bigr\}$; we denote by $u_{\lambda}$ the element of ${\mathcal{B}}(\lambda)$ such that $G(u_{\lambda})=v_{\lambda}$, and by ${\mathcal{B}}_{0}(\lambda)$ the connected component of ${\mathcal{B}}(\lambda)$ containing $u_{\lambda}$. Also, we know from [@Kas05 Sect. 2.8] (see also [@NS16 Sect. 4.1]) that $V_{x}^{-}(\lambda) \subset V(\lambda)$ is compatible with the global basis of $V(\lambda)$, that is, there exists a subset ${\mathcal{B}}_{x}^{-}(\lambda)$ of the crystal basis ${\mathcal{B}}(\lambda)$ such that $$\label{eq:deme}
V_{x}^{-}(\lambda) =
\bigoplus_{b \in {\mathcal{B}}_{x}^{-}(\lambda)} {\mathbb{C}}({\mathsf{q}}) G(b)
\subset
V(\lambda) =
\bigoplus_{b \in {\mathcal{B}}(\lambda)} {\mathbb{C}}({\mathsf{q}}) G(b).$$
For every $x \in W_{{\mathrm{af}}}$, we have $V_{x}^{-}(\lambda) = V_{{\Pi^{J}}(x)}^{-}(\lambda)$ and ${\mathcal{B}}_{x}^{-}(\lambda) = {\mathcal{B}}_{{\Pi^{J}}(x)}^{-}(\lambda)$.
We know the following from [@INS Theorem 3.2.1] and [@NS16 Theorem 4.2.1].
\[thm:isom\] There exists an isomorphism $\Phi_{\lambda}:{\mathcal{B}}(\lambda) \stackrel{\sim}{\rightarrow} {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ of crystals such that $\Phi(u_{\lambda}) = \pi_{\lambda}$ and such that $\Phi_{\lambda}({\mathcal{B}}_{x}^{-}(\lambda)) = {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)$ for all $x \in W_{{\mathrm{af}}}$[;]{} in particular, we have $\Phi_{\lambda}({\mathcal{B}}_{0}(\lambda)) = {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)$.
Let $x \in W_{{\mathrm{af}}}$. If $x$ is of the form $x = wt_{\xi}$ for some $w \in W$ and $\xi \in Q^{\vee}$, then $v_{x} \in V(\lambda)$ is a weight vector of weight $x\lambda =
w\lambda-{\langle \lambda,\,\xi \rangle}\delta$; note that $w\lambda \in \lambda-Q^{+}$. Also, for $i \in I$ (resp., $i=0 \in I_{{\mathrm{af}}}$), the Chevalley generator $F_{i}$ (resp., $F_{0}$) of $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$ acts on $V(\lambda)$ as a (linear) operator of weight $-\alpha_{i} \in Q$ (resp., $-\alpha_{0}=\theta-\delta \in Q + {\mathbb{Z}}_{< 0}\delta$). Therefore, the Demazure submodule $V_{x}^{-}(\lambda) = U_{{\mathsf{q}}}^{-}({\mathfrak{g}}_{{\mathrm{af}}})v_{x}$ has the weight space decomposition of the form: $$V_{x}^{-}(\lambda) =
\bigoplus_{k \in {\mathbb{Z}}}
\Biggl(
\underbrace{
\bigoplus_{\gamma \in Q}
V_{x}^{-}(\lambda)_{\lambda+\gamma+k\delta}}_{=:V_{x}^{-}(\lambda)_{k}}
\Biggr),$$ where $V_{x}^{-}(\lambda)_{k} = \bigl\{ 0 \bigr\}$ for all $k > - {\langle \lambda,\,\xi \rangle}$; in addition, by Theorem \[thm:isom\], together with the definition of the map ${\mathop{\rm wt}\nolimits}:{\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \rightarrow P_{{\mathrm{af}}}$ (see ), we see that if $\gamma \notin -Q^{+}$, then $V_{x}^{-}(\lambda)_{\lambda+\gamma+k\delta} = \bigl\{0\bigr\}$ for all $k \in {\mathbb{Z}}$, since $W_{{\mathrm{af}}}\lambda \subset \lambda - Q^{+} +{\mathbb{Z}}\delta$ by the assumption that $\lambda \in P^{+}$. Here we claim that $V_{x}^{-}(\lambda)_{k}$ is finite-dimensional for all $k \in {\mathbb{Z}}$ with $k \le -{\langle \lambda,\,\xi \rangle}$; we show this assertion by descending induction on $k$. Let $U_{{\mathsf{q}}}^{-}({\mathfrak{g}})$ denote the ${\mathbb{C}}({\mathsf{q}})$-subalgebra of $U_{{\mathsf{q}}}^{-}({\mathfrak{g}}_{{\mathrm{af}}})$ generated by $F_{i}$, $i \in I$. If $k=-{\langle \lambda,\,\xi \rangle}$, then the assertion is obvious since $V_{x}^{-}(\lambda)_{-{\langle \lambda,\,\xi \rangle}} = U_{{\mathsf{q}}}^{-}({\mathfrak{g}})v_{x}$ and $V(\lambda)$ is an integrable $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$-module. Assume that $k < -{\langle \lambda,\,\xi \rangle}$. Observe that $V_{x}^{-}(\lambda)_{k}$ is a $U_{{\mathsf{q}}}^{-}({\mathfrak{g}})$-module generated by $F_{0}V_{x}^{-}(\lambda)_{k+1}$. Because $F_{0}V_{x}^{-}(\lambda)_{k+1}$ is finite-dimensional by our induction hypothesis, and $V(\lambda)$ is an integrable $U_{q}({\mathfrak{g}}_{{\mathrm{af}}})$-module, we deduce that $V_{x}^{-}(\lambda)_{k} = U_{{\mathsf{q}}}^{-}({\mathfrak{g}})(F_{0}V_{x}^{-}(\lambda)_{k+1})$ is also finite-dimensional, as desired.
Now, we define the graded character ${\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda)$ of $V_{x}^{-}(\lambda)$ to be $$\label{eq:gch}
{\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda) : =
\sum_{k \in {\mathbb{Z}}}
\Biggl(
\sum_{\gamma \in Q}
\dim \bigl( V_{x}^{-}(\lambda)_{\lambda+\gamma+k\delta} \bigr)
e^{\lambda+\gamma}
\Biggr) q^{k};$$ observe that $$\label{eq:gch1}
{\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda) \in
\bigl(\underbrace{{\mathbb{Z}}[e^{\nu} \mid \nu \in P]}_{={\mathbb{Z}[P]}}\bigr)
{[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}q^{-{\langle \lambda,\,\xi \rangle}}.$$ For $\gamma \in Q$ and $k \in {\mathbb{Z}}$, we set ${\mathop{\rm fin}\nolimits}(\lambda+\gamma+k\delta):=\lambda+\gamma \in P$ and ${\mathop{\rm nul}\nolimits}(\lambda+\gamma+k\delta):=k \in {\mathbb{Z}}$. Then, by Theorem \[thm:isom\], we have $$\label{eq:gch2}
{\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda) = \sum_{\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))} q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))}.$$
Combinatorial standard monomial theory for semi-infinite LS paths. {#sec:SMT}
==================================================================
In this section, we fix $\lambda,\,\mu \in P^{+} \subset P_{{\mathrm{af}}}^{0}$ (see and ), and set $$J:=\bigl\{i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0 \bigr\}, \qquad
K:=\bigl\{i \in I \mid {\langle \mu,\,\alpha_{i}^{\vee} \rangle}=0 \bigr\}.$$
Standard paths. {#subsec:SP}
---------------
We consider the following condition on $\pi \otimes \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$: $$\label{eq:SM}
\begin{cases}
\text{there exist $x,\,y \in W_{{\mathrm{af}}}$ such that
$x {\succeq}y$ in $W_{{\mathrm{af}}}$}, \\[1mm]
\text{and such that ${\Pi^{J}}(x)=\kappa(\pi)$, ${\Pi^{K}}(y)=\iota(\eta)$};
\end{cases} \tag{\sf SP}$$ we set $${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu):=
\bigl\{\pi \otimes \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu) \mid
\text{$\pi \otimes \eta$ satisfies condition \eqref{eq:SM}}
\bigr\}.$$
\[thm:SMT\] The set ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu) \sqcup \{{\bm{0}}\}$ is stable under the action of the Kashiwara [(]{}or, root[)]{} operators $e_{i}$, $f_{i}$, $i \in I_{{\mathrm{af}}}$, on ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$[;]{} in particular, ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ is a crystal with weights in $P_{{\mathrm{af}}}$. Moreover, ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ is isomorphic as a crystal to ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu)$.
We will give a proof of Theorem \[thm:SMT\] in Section \[sec:prf-SMT\].
Defining chains. {#subsec:DC}
----------------
\[dfn:DC\] Let $\pi=(x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta=(y_{1},\,\dots,\,y_{p}\,;\,{\mathbf{b}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. A defining chain for $\pi \otimes \eta$ is a sequence $x_{1}',\,\dots,\,x_{s}',\,y_{1}',\,\dots,\,y_{p}'$ of elements in $W_{{\mathrm{af}}}$ satisfying the condition: $$\label{eq:DC}
\begin{cases}
x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{p}'
\quad \text{\rm in $W_{{\mathrm{af}}}$}; \\[1.5mm]
{\Pi^{J}}(x_{u}')=x_{u}
\quad \text{\rm for $1 \le u \le s$}; \\[1.5mm]
{\Pi^{K}}(y_{q}')=y_{q}
\quad \text{\rm for $1 \le q \le p$};
\end{cases}
\tag{\sf DC}
$$ we call $x_{1}'$ (resp., $y_{p}'$) the initial element (resp., the final element) of this defining chain.
\[prop:DC\] Let $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. Then, $\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ if and only if there exists a defining chain for $\pi \otimes \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$.
We will give a proof of Proposition \[prop:DC\] in Section \[subsec:prf-DC\].
Now, let $\eta=(y_{1},\,\dots,\,y_{p}\,;\,{\mathbf{b}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. For each $x \in W_{{\mathrm{af}}}$ such that $\kappa(\eta) = y_{p} {\succeq}{\Pi^{K}}(x)$, we define a specific lift ${\iota(\eta,\,x)} \in W_{{\mathrm{af}}}$ of $\iota(\eta) = y_{1} \in ({W^{K}})_{{\mathrm{af}}}$ as follows. Since $y_{p} {\succeq}{\Pi^{K}}(x)$ by the assumption, it follows from Proposition \[prop:Deo\] that ${\mathrm{Lift}_{{\succeq}x}(y_{p})}$ has the minimum element $\min {\mathrm{Lift}_{{\succeq}x}(y_{p})}=:{\widetilde{y}}_{p}$. Similarly, since $y_{p-1} {\succeq}y_{p} = {\Pi^{K}}({\widetilde{y}}_{p})$, it follows again from Proposition \[prop:Deo\] that ${\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{p}}(y_{p-1})}$ has the minimum element $\min {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{p}}(y_{p-1})} = :{\widetilde{y}}_{p-1}$. Continuing in this way, we obtain ${\widetilde{y}}_{p}$, ${\widetilde{y}}_{p-1}$, $\dots$, ${\widetilde{y}}_{1}$. Namely, these elements are defined by the following recursive procedure (from $p$ to $1$): $$\label{eq:lift0}
\begin{cases}
{\widetilde{y}}_{p} := \min {\mathrm{Lift}_{{\succeq}x}(y_{p})}, & \\[1mm]
{\widetilde{y}}_{q} := \min {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{q+1}}(y_{q})}
& \text{for $1 \le q \le p-1$}.
\end{cases}
$$ Finally, we set $$\label{eq:lift0a}
{\iota(\eta,\,x)}:={\widetilde{y}}_{1};$$ this element is sometimes called the initial direction of $\eta$ with respect to $x$.
\[prop:DC2\] Let $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$. Then, $\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ [(]{}or equivalently, there exists a defining chain for $\pi \otimes \eta \in
{\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ by Proposition \[prop:DC\][)]{} if and only if $\kappa(\pi) {\succeq}{\Pi^{J}}({\iota(\eta,\,x)})$ for some $x \in W_{{\mathrm{af}}}$ such that $\kappa(\eta) {\succeq}{\Pi^{K}}(x)$.
We will give a proof of Proposition \[prop:DC2\] in Section \[subsec:prf-DC2\].
Demazure crystals in terms of standard paths. {#subsec:dem-SP}
---------------------------------------------
We set $S:=\bigl\{i \in I \mid {\langle \lambda+\mu,\,\alpha_{i}^{\vee} \rangle}=0 \bigr\}={J}\cap {K}$. For each $x \in W_{{\mathrm{af}}}$, we define ${\mathbb{S}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)
\subset {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ to be the image of ${\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda+\mu)=
\bigl\{ \psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu) \mid \kappa(\psi) {\succeq}{\Pi^{S}}(x) \bigr\}$ under the isomorphism ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda + \mu) \cong {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ in Theorem \[thm:SMT\].
\[thm:Dem\] Let $x \in W_{{\mathrm{af}}}$. For $\pi \otimes \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$, the following conditions , , and are equivalent[:]{} $$\label{eq:Da}
\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu); \tag{\sf D1}$$ $$\label{eq:Db}
\begin{cases}
\text{\rm there exists a defining chain for $\pi \otimes \eta$ whose final element, say $y$, } \\[1mm]
\text{\rm satisfies the condition that ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$};
\end{cases} \tag{\sf D2}
$$ $$\label{eq:Dc}
\kappa(\eta) {\succeq}{\Pi^{K}}(x) \quad \text{\rm and} \quad
\kappa(\pi) {\succeq}{\Pi^{J}}({\iota(\eta,\,x)}). \tag{\sf D3}$$ Therefore, we have $${\mathbb{S}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) =
\bigl\{ \pi \otimes \eta \mid
\text{\rm $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\mu)$ and
$\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}{\iota(\eta,\,x)}}(\lambda)$}
\bigr\},$$ and hence [(]{}see $${\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda+\mu) =
\sum_{ \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\mu) } e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))} q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))}
\underbrace{\sum_{ \pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}{\iota(\eta,\,x)}}(\lambda) } e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))} q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))}
}_{ \text{\rm $={\mathop{\rm gch}\nolimits}V_{{\iota(\eta,\,x)}}^{-}(\lambda)$ by \eqref{eq:gch2}} }.$$
We will give a proof of Theorem \[thm:Dem\] in Section \[sec:prf-Dem\].
Semi-infinite Schubert varieties and their resolutions. {#sec:SiSch}
=======================================================
Geometric setting. {#subsec:geo}
------------------
An (algebraic) variety is an integral separated scheme of finite type over ${\mathbb{C}}$. Also, a pro-affine space is a product of finitely many copies of ${\mathop{\rm Spec}\nolimits}{\mathbb{C}}[ x_{m} \mid m \ge 0 ]$, equipped with a truncation morphism ${\mathop{\rm Spec}\nolimits}{\mathbb{C}}[ x_{m} \mid m \ge 0 ] \rightarrow {\mathop{\rm Spec}\nolimits}{\mathbb{C}}[ x_{m} \mid 0 \le m \le n ]$ for $n \gg 0$; by a morphism of pro-affine spaces, we mean a morphism of schemes that is also continuous with respect to the topology induced by the truncation morphisms (this topology itself is irrelevant to the Zariski topology). For a ${\mathbb{C}}$-vector space $V$, we set ${\mathbb{P}}(V) := (V \setminus \{ 0 \}) / {\mathbb{C}}^{\times}$. We usually regard ${\mathbb{P}}(V)$ as an algebraic variety over ${\mathbb{C}}$ when $\dim V < \infty$, or as an ind/pro-scheme when $\dim V = \infty$ in accordance with the topology of $V$.
For an algebraic group $E$, let $E{[\hspace{-1pt}[z]\hspace{-1pt}]}$, $E{(\hspace{-1pt}(z)\hspace{-1pt})}$, and $E [z]$ denote the set of ${\mathbb{C}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$-valued points, ${\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$-valued points, and ${\mathbb{C}}[z]$-valued points of $E$, respectively; the corresponding Lie algebras are denoted by ${\mathfrak{e}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$, ${\mathfrak{e}}{(\hspace{-1pt}(z)\hspace{-1pt})}$, and ${\mathfrak{e}}[z]$, respectively, with $E$ replaced by its German letter ${\mathfrak{e}}={\mathop{\rm Lie}\nolimits}(E)$. Also, we denote by $R(E)$ the representation ring of $E$.
Recall that $G$ is a connected, simply-connected simple algebraic group over ${\mathbb{C}}$; concerning the Lie algebra ${\mathfrak{g}}={\mathop{\rm Lie}\nolimits}(G)$ and its untwisted affinization ${\mathfrak{g}}_{{\mathrm{af}}}$, we use the notation of Section \[sec:notation\]. We have an evaluation map ${\mathop{\tt ev}\nolimits}_{0} : G{[\hspace{-1pt}[z]\hspace{-1pt}]} \longrightarrow G$ at $z=0$. Let ${\mathbf{I}}:= {\mathop{\tt ev}\nolimits}_{0}^{-1} ( B )$ be an Iwahori subgroup of $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$. Also, for each $i \in I_{{\mathrm{af}}}$, we have a minimal parahoric subgroup ${\mathbf{I}}\subset {\mathbf{I}}(i) \subset G{[\hspace{-1pt}[z]\hspace{-1pt}]}$ corresponding to $\alpha_{i}$, so that ${\mathbf{I}}(i) / {\mathbf{I}}\cong {\mathbb{P}}^{1}$. Note that both $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$ and ${\mathbf{I}}$ admit an action of ${\mathbb{G}_{m}}$ obtained by the scalar dilation on $z$; we denote the resulting semi-direct product groups by ${\widetilde{G}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$ and ${\widetilde{{\mathbf{I}}}}$, respectively. The (finite) Weyl group $W$ of ${\mathfrak{g}}$ is isomorphic to $N_{G}(H)/H$, and $Q^{\vee}$ is isomorphic to $H{(\hspace{-1pt}(z)\hspace{-1pt})}/H{[\hspace{-1pt}[z]\hspace{-1pt}]}$, both of which fit in the following commutative diagram involving the (affine) Weyl group $W_{{\mathrm{af}}}$ of ${\mathfrak{g}}_{{\mathrm{af}}}$: $$\begin{CD}
0 @>>> Q^{\vee} @>>> W_{{\mathrm{af}}} @>>> W @>>> e \\
@. @A{\cong}AA @A{\cong}AA @A{\cong}AA \\
@. H{(\hspace{-1pt}(z)\hspace{-1pt})}/H{[\hspace{-1pt}[z]\hspace{-1pt}]} @>>> N_{G{(\hspace{-1pt}(z)\hspace{-1pt})}}(H)/H{[\hspace{-1pt}[z]\hspace{-1pt}]} @>>>
\underbrace{N_{G{(\hspace{-1pt}(z)\hspace{-1pt})}}(H)/H{(\hspace{-1pt}(z)\hspace{-1pt})}}_{= N_{G}(H)/H}, \\
\end{CD}
$$ where the first row is exact, and the rightmost isomorphism in the second row holds since $N_{G{(\hspace{-1pt}(z)\hspace{-1pt})}}(H) \cong
N_{G{(\hspace{-1pt}(z)\hspace{-1pt})}}(H{(\hspace{-1pt}(z)\hspace{-1pt})}) \cong (N_{G}(H)){(\hspace{-1pt}(z)\hspace{-1pt})}$. In particular, we have a lift $\dot{w} \in N_{G{(\hspace{-1pt}(z)\hspace{-1pt})}}(H)$ for each $w \in W_{{\mathrm{af}}}$. Now, for $x \in W_{{\mathrm{af}}}$ and $i \in I_{{\mathrm{af}}}$, we set $$s_{i} \ast x :=
\begin{cases}
x & \text{if $s_{i} x < x$}, \\
s_{i} x & \text{if $s_{i} x > x$},
\end{cases}
$$ where we denote by $>$ the ordinary Bruhat order on $W_{{\mathrm{af}}}$. Then, the set $W_{{\mathrm{af}}}$ becomes a monoid, which we denote by ${\mathbf{W}_{{\mathrm{af}}}}$, under the product $\ast$; this monoid is also obtained as a subset of the generic Hecke algebra associated to $(W_{{\mathrm{af}}},\,I_{{\mathrm{af}}})$ by setting $a_{i} = 1$ and $b_{i} = 0$ for $i \in I_{{\mathrm{af}}}$ in [@Hum90 Sect. 7.1, Theorem].
Semi-infinite flag manifolds. {#subsec:SiFl}
-----------------------------
Here we review two models of semi-infinite flag manifold associated to $G$, for which the basic references are [@FM99] and [@FFKM99].
Let $L(\lambda)$ denote the (finite-dimensional) irreducible highest weight ${\mathfrak{g}}$-module of highest weight $\lambda \in P^{+}$. Recall that for each $\lambda,\,\mu \in P^{+}$, we have a canonical embedding of irreducible highest weight ${\mathfrak{g}}$-modules (and hence of $G$-modules) up to scalars: $$\label{HWtensor}
L(\lambda + \mu) \hookrightarrow L(\lambda) \otimes_{{\mathbb{C}}} L(\mu).$$ The embedding induces an embedding $$\label{eq:emb}
L(\lambda+\mu) \otimes_{{\mathbb{C}}} R \hookrightarrow
\bigl( L(\lambda) \otimes_{{\mathbb{C}}} L(\mu) \bigr) \otimes_{{\mathbb{C}}} R \cong
(L(\lambda) \otimes_{{\mathbb{C}}} R) \otimes_{R} (L(\mu) \otimes_{{\mathbb{C}}} R)$$ for every commutative, associative ${\mathbb{C}}$-algebra $R$.
\[thm:BG\] Let ${\mathbb{K}}$ be a field containing ${\mathbb{C}}$. The set of collections $\{ \ell^{\lambda} \}_{\lambda \in P^{+}}$ of one-dimensional ${\mathbb{K}}$-vector subspaces $\ell^{\lambda}$ in $L(\lambda) \otimes_{{\mathbb{C}}} {\mathbb{K}}$ such that $\ell^{\lambda} \otimes_{{\mathbb{K}}} \ell^{\mu} = \ell^{\lambda+\mu}$ for every $\lambda,\,\mu \in P^{+}$ [(]{}under the embedding is in bijection with the set of closed ${\mathbb{K}}$-points of $G/B$.
For a ${\mathfrak{g}}$-module $V$, we set $V{[\hspace{-1pt}[z]\hspace{-1pt}]} := V \otimes_{{\mathbb{C}}} {\mathbb{C}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$ and $V{(\hspace{-1pt}(z)\hspace{-1pt})} := V \otimes_{{\mathbb{C}}} {\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$.
\[dfn:DP\] Consider a collection $L = \{ L^{\lambda} \}_{\lambda \in P^{+}}$ of one-dimensional ${\mathbb{C}}$-vector subspaces $L^{\lambda}$ in $L(\lambda){[\hspace{-1pt}[z]\hspace{-1pt}]} = L(\lambda) \otimes_{{\mathbb{C}}} {\mathbb{C}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$ (resp., $L(\lambda){(\hspace{-1pt}(z)\hspace{-1pt})} = L(\lambda) \otimes_{{\mathbb{C}}} {\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$). The datum $L$ is called a formal (resp., rational) Drinfeld-Plücker (DP for short) datum if for every $\lambda,\,\mu \in P^{+}$, the equality $$\label{eq:DP}
L^{\lambda + \mu} = L^{\lambda} \otimes_{{\mathbb{C}}} L^{\mu}$$ holds under the embedding , where $L^{\lambda} \otimes_{{\mathbb{C}}} L^{\mu}$ is considered to be its image under the map $L(\lambda){[\hspace{-1pt}[z]\hspace{-1pt}]} \otimes_{{\mathbb{C}}} L(\mu){[\hspace{-1pt}[z]\hspace{-1pt}]}
\rightarrow L(\lambda){[\hspace{-1pt}[z]\hspace{-1pt}]} \otimes_{{\mathbb{C}}{[\hspace{-1pt}[z]\hspace{-1pt}]}} L(\mu){[\hspace{-1pt}[z]\hspace{-1pt}]}$ (resp., $L(\lambda){(\hspace{-1pt}(z)\hspace{-1pt})} \otimes_{{\mathbb{C}}} L(\mu){(\hspace{-1pt}(z)\hspace{-1pt})}
\rightarrow L(\lambda){(\hspace{-1pt}(z)\hspace{-1pt})} \otimes_{{\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}} L(\mu){(\hspace{-1pt}(z)\hspace{-1pt})}$); we sometimes refer to a collection $\{u^{\lambda}\}_{\lambda \in P^{+}}$ of nonzero elements $u^{\lambda} \in L^{\lambda}$, $\lambda \in P^{+}$, as a formal (resp., rational) DP datum. Let ${\mathbf{Q}_{G}}$ (resp., ${\mathbf{Q}_{G}^{\mathrm{rat}}}$) denote the set of formal (resp., rational) DP data.
\[rem:DP\] By the compatibility condition , a DP datum $\{ L^{\lambda} \}_{\lambda \in P^{+}}$ is determined completely by a collection $\{u^{i}\}_{i \in I}$ of nonzero elements $u^{i} \in L^{{\varpi}_{i}}$ for $i \in I$. We call this collection $\{u^{i}\}_{i \in I}$ DP coordinates.
Let $L = \{ L^{\lambda} \}_{\lambda \in P^{+}} \in {\mathbf{Q}_{G}}$. We define $\deg L^{\lambda}$ to be the degree of a nonzero element in $L^{\lambda}$, viewed as an $L(\lambda)$-valued formal power series (if it is bounded). For each $\xi \in Q^{\vee,+}$, a DP datum of degree $\xi$ is a formal DP datum $L = \{ L^{\lambda} \}_{\lambda \in P^{+}}$ such that $\deg L^{\lambda} \le {\langle \lambda,\,\xi \rangle}$ for all $\lambda \in P^{+}$. For each $\xi \in Q^{\vee,+}$, let ${\mathfrak{Q}}_{G} (\xi)$ denote the set of formal DP data of degree $\xi$. Here we note that if $\xi,\,\zeta \in Q^{\vee,+}$ satisfy $\xi \le \zeta$, i.e., $\zeta - \xi \in Q^{\vee,+}$, then ${\mathfrak{Q}}_{G}(\xi) \subset {\mathfrak{Q}}_{G}(\zeta)$. We set ${\mathfrak{Q}}_{G} := \bigcup_{\xi \in Q^{\vee,+}} {\mathfrak{Q}}_{G}(\xi)$; observe that $${\mathfrak{Q}}_{G}(\xi) \subset {\mathfrak{Q}}_{G} \subset {\mathbf{Q}_{G}}\quad
\text{for each $\xi \in Q^{\vee,+}$}.$$ Also, we remark that ${\mathbf{Q}_{G}}$ has a natural $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-action, and that its subsets ${\mathfrak{Q}}_{G}(\xi)$, $\xi \in Q^{\vee,+}$, and ${\mathfrak{Q}}_{G}$ are stable under the action of $G$ on ${\mathbf{Q}_{G}}$.
\[defbQ\] We have an embedding $$\label{Pemb}
{\mathbf{Q}_{G}}\ni
\{ L^{\lambda} \}_{\lambda \in P^{+}} \mapsto
\{ [ L^{{\varpi}_{i}} ] \} _{i \in I} \in
\prod_{i \in I} {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]}),$$ which gives the set ${\mathbf{Q}_{G}}$ a [(]{}reduced[)]{} structure of an infinite type closed subscheme of $\prod_{i \in I} {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]})$. In particular, ${\mathbf{Q}_{G}}$ is separated.
Because a DP datum $\{ L^{\lambda} \}_{\lambda \in P^{+}}$ is determined uniquely by $\{ L^{{\varpi}_{i}} \} _{i \in I}$ (see Remark \[rem:DP\]), the map above is injective. Also, condition is equivalent to saying that the image of $L^{\lambda} \otimes_{{\mathbb{C}}} L^{\mu}$ lies in the ${\mathbb{C}}$-vector subspace $L(\lambda+\mu){[\hspace{-1pt}[z]\hspace{-1pt}]} \subset
\bigl(L(\lambda) \otimes_{{\mathbb{C}}} L(\mu) \bigr){[\hspace{-1pt}[z]\hspace{-1pt}]}$ for all $\lambda,\,\mu \in P^{+}$. This condition imposes infinitely many equations on $\prod _{i \in I} {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]})$ that define ${\mathbf{Q}_{G}}$ as its closed subscheme. Since each ${\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]})$ is separated, so is ${\mathbf{Q}_{G}}$. This proves the lemma.
\[orbbQ\] The set of $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbits in ${\mathbf{Q}_{G}}$ is labeled by $Q^{\vee,+}$. The codimension of the $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbit corresponding to $\xi \in Q^{\vee,+}$ is equal to $2 {\langle \rho,\,\xi \rangle}$.
\[orbits\] The set of ${\mathbf{I}}$-orbits in ${\mathbf{Q}_{G}}$ is in bijection with the set $W_{{\mathrm{af}}}^{\ge 0} =
\bigl\{w t_{\xi} \mid w \in W,\,\xi \in Q^{\vee,+}\bigr\}$.
Apply (the consequence of) the Bruhat decomposition $G{[\hspace{-1pt}[z]\hspace{-1pt}]} = \bigsqcup_{w \in W} {\mathbf{I}}\dot{w} {\mathbf{I}}$ to each $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbit in ${\mathbf{Q}_{G}}$ described in Theorem \[orbbQ\].
For each $x = w t_{\xi} \in W_{{\mathrm{af}}}^{\ge 0}$, we denote by ${\mathbf{O}}(x)$ the ${\mathbf{I}}$-orbit of ${\mathbf{Q}_{G}}$ that contains a unique $(H \times {\mathbb{G}_{m}})$-fixed point corresponding to $\{z^{ {\langle {\varpi}_{i},\,\xi \rangle} } v_{w w_{\circ} {\varpi}_{i}} \} _{i \in I}$ (see Lemma \[defbQ\]), where for each $\lambda \in P^{+}$ and $w \in W$, we take and fix a nonzero vector $v_{w\lambda}$ of weight $w\lambda$ in $L(\lambda)$; note that the codimension of ${\mathbf{O}}(x) \subset {\overline{{\mathbf{O}}(e)}}$ is given by ${\ell^{{\frac{\infty}{2}}}}(x)$. We set ${\mathbf{Q}_{G}}( x ) := \overline{{\mathbf{O}}( x )} \subset {\mathbf{Q}_{G}}$. For $x,\,y \in W_{{\mathrm{af}}}^{\ge 0}$, we have $$\begin{aligned}
{\mathbf{Q}_{G}}(x) \subset {\mathbf{Q}_{G}}( y )
& \iff x {\succeq}y \quad \text{(\cite[Sect.~5.1]{FFKM99})} \\
& \iff x w_{\circ} {\preceq}yw_{\circ} \quad
\text{(see \cite[Lecture 13]{Pet97})}.\end{aligned}$$ Also, we have ${\mathbf{Q}_{G}}(e) = {\mathbf{Q}_{G}}$ by inspection; in fact, $e \in W_{{\mathrm{af}}}^{\ge 0}$ is the minimum element in the semi-infinite Bruhat order restricted to $W_{{\mathrm{af}}}^{\ge 0}$.
For $\xi \in Q^{\vee,+}$ and $x \in W_{{\mathrm{af}}}^{\ge 0}$, we set ${\mathfrak{Q}}_G (x,\,\xi) := {\mathfrak{Q}}_G (\xi) \cap {\mathbf{Q}_{G}}(x)$, and for $x \in W_{{\mathrm{af}}}^{\ge 0}$, we set ${\mathfrak{Q}}_G (x) := \bigcup_{\xi \in Q^{\vee,+}} {\mathfrak{Q}}_G (x,\,\xi)$.
For each $\xi \in Q^{\vee,+}$, we have an embedding $$\imath_{\xi} :
{\mathbf{Q}_{G}}\ni
\{ L^{\lambda} \}_{\lambda \in P^{+}}
\bigl(= \{ u^{\lambda} \}_{\lambda \in P^{+}} \bigr) \mapsto
\{ z ^{ {\langle \lambda,\,\xi \rangle} } u^{\lambda} \}_{\lambda \in P^{+}} \in {\mathbf{Q}_{G}}.$$ Thus we have its direct limit ${\mathbf{Q}_{G}^{\mathrm{rat}}}\cong \varinjlim_{\xi} {\mathbf{Q}_{G}}$, on which we have an action of $G{(\hspace{-1pt}(z)\hspace{-1pt})}$. By its construction, the embedding $\imath_{\xi}$ is $G {[\hspace{-1pt}[z]\hspace{-1pt}]}$-equivariant, and sends the ${\mathbf{I}}$-orbit ${\mathbf{O}}(x)$ to ${\mathbf{O}}(xt_{\xi})$.
Now, by Lemma \[defbQ\], we have a $G {[\hspace{-1pt}[z]\hspace{-1pt}]}$-equivariant line bundle ${\mathcal{O}}_{{\mathbf{Q}_{G}}}({\varpi}_{i})$ obtained by the pullback of the $i$-th ${\mathcal{O}}(1)$ through . For $\lambda = \sum_{i \in I} m_{i} {\varpi}_{i} \in P$, we set ${\mathcal{O}}_{{\mathbf{Q}_{G}}}(\lambda) :=
\bigotimes_{i \in I} {\mathcal{O}}_{{\mathbf{Q}_{G}}}({\varpi}_{i})^{\otimes m_{i}}$ (as a tensor product of line bundles). Also, for each $x \in W_{{\mathrm{af}}}^{\ge 0}$, we have the restriction ${\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}(\lambda)$ obtained through , which is ${\widetilde{{\mathbf{I}}}}$-equivariant. Similarly, we have $( B \times {\mathbb{G}_{m}})$-equivariant line bundles ${\mathcal{O}}_{{\mathfrak{Q}}_G(x,\xi)} ( \lambda )$ and ${\mathcal{O}}_{{\mathfrak{Q}}_G(x)} ( \lambda )$ by further pullbacks (the latter is $G [z]$-equivariant whenever $x = t_{\zeta}$ for some $\zeta \in Q^{\vee,+}$); we set $$\label{eq:Hn}
H^{n} ( {\mathfrak{Q}}_G(x),\, {\mathcal{O}}_{{\mathfrak{Q}}_G(x)}(\lambda) ) :=
\varprojlim _{\xi} H ^{n} ( {\mathfrak{Q}}_G(x,\,\xi),\,{\mathcal{O}}_{{\mathfrak{Q}}_G(x,\,\xi)}(\lambda) )
\quad \text{for $n \in {\mathbb{Z}}_{\ge 0}$}.$$
Let $\lambda \in P^{+}$. As explained in [@Kat16 Theorem 1.6], the restricted dual of the Demazure submodule $V_{e}^{-} (- w_{\circ} \lambda )$ (see ) of the extremal weight $U_{{\mathsf{q}}}({\mathfrak{g}}_{{\mathrm{af}}})$-module $V(- w_{\circ} \lambda )$ of extremal weight $- w_{\circ} \lambda$ gives rise to an integrable ${\mathfrak{g}}[z]$-module (by taking the classical limit ${\mathsf{q}}\to 1$), called the global Weyl module; we denote it by $W(\lambda)$. Here we note that global Weyl modules carry natural gradings arising from the dilation of the $z$-variable.
\[BFmain\] For $\lambda \in P$, we have $$H^{0} ( {\mathring{\mathbf{Q}}_{G}}, \, {\mathcal{O}}_{{\mathbf{Q}_{G}}}(\lambda) ) \cong
\begin{cases}
W(\lambda)^{\ast} & \text{\rm if $\lambda \in P^{+}$}, \\[1mm]
\{ 0 \} & \text{\rm otherwise},
\end{cases}
$$ where ${\mathring{\mathbf{Q}}_{G}}$ denotes the open dense $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbit in ${\mathbf{Q}_{G}}$.
\[Kmain\] For each $\lambda \in P$ and $x \in W_{{\mathrm{af}}}^{\ge 0}$, we have $${\mathop{\rm gch}\nolimits}\dot{H}^{n}( {\mathfrak{Q}}_G ( x ),\,{\mathcal{O}}_{{\mathfrak{Q}}_G(x)} ( \lambda ) ) =
\begin{cases}
{\mathop{\rm gch}\nolimits}V_{x}^{-}( - w_{\circ} \lambda )
& \text{\rm if $\lambda \in P^{+}$ and $n=0$}, \\
0 & \text{\rm otherwise},
\end{cases}
$$ where $\dot{H}^{n}( {\mathfrak{Q}}_G ( x ),\,{\mathcal{O}}_{{\mathfrak{Q}}_G(x)} ( \lambda ) )$ denotes the space of ${\mathbb{G}_{m}}$-finite vectors in .
\[Kmain-rem\] The assertion of [@Kat16 Theorem 4.12] is only for $x \in W$. The assertion of the form above easily follows from the isomorphism ${\mathbf{Q}_{G}}( x ) \cong {\mathbf{Q}_{G}}( x t_{\xi} )$ for $x \in W$ and $\xi \in Q^{\vee, +}$, which is obtained by means of $\imath_{\xi}$.
\[vample\] For each $\lambda \in P^{++}:=\sum_{i \in I} {\mathbb{Z}}_{> 0} {\varpi}_{i}$, the sheaf ${\mathcal{O}}_{{\mathbf{Q}_{G}}}(\lambda)$ is very ample.
The proof is exactly the same as that of [@Kat16 Corollary 2.7].
Bott-Samelson-Demazure-Hansen towers. {#subsec:BSDH}
-------------------------------------
In this subsection, we construct two kind of (pro-)schemes of infinite type, which can be thought of as “resolutions" of ${\mathbf{Q}_{G}}(x)$ for $x \in W_{{\mathrm{af}}}^{\ge 0}$, and study their properties.
\[lem:Mac\] If $\xi \in Q^{\vee}$ is a strictly antidominant coweight, i.e., ${\langle \alpha_{i},\,\xi \rangle} < 0$ for all $i \in I$, then $\ell(t_{\xi})=-2{\langle \rho,\,\xi \rangle}$, and $\ell(wt_{\xi})=\ell(t_{\xi}) - \ell(w)$ for all $w \in W$[;]{} hence we have $\ell(wt_{\xi})=-{\ell^{{\frac{\infty}{2}}}}(wt_{\xi})$ for all $w \in W$.
\[tweakw\]
${\ell^{{\frac{\infty}{2}}}}( x t_{- 2 m \rho^{\vee}} ) +
{\ell^{{\frac{\infty}{2}}}}( t _{2 m \rho^{\vee}} ) = {\ell^{{\frac{\infty}{2}}}}( x )$ for all $x \in W_{{\mathrm{af}}}$ and $m \in {\mathbb{Z}}$.
There exists $m_{0} \ge 0$ such that $-{\ell^{{\frac{\infty}{2}}}}( x t_{- 2 m \rho^{\vee}} ) =
\ell ( x t_{- 2 m \rho^{\vee}} )$ for all $m \ge m_{0}$.
Part (1) is obvious from the definition of ${\ell^{{\frac{\infty}{2}}}}(\,\cdot\,)$. For part (2), write $x$ as $x =wt_{\xi} \in W_{{\mathrm{af}}}$ for some $w \in W$ and $\xi \in Q^{\vee}$, and take $m_{0} \ge 0$ such that $\xi-2m_{0}\rho^{\vee}$ is strictly antidominant. Then we see from Lemma \[lem:Mac\] that $-{\ell^{{\frac{\infty}{2}}}}( x t_{- 2 m \rho^{\vee}} ) =
\ell ( x t_{- 2 m \rho^{\vee}} )$ for all $m \ge m_{0}$. This proves the lemma.
\[rem:ell\] Keep the setting of Lemma \[tweakw\](2). We have $$\ell ( x t_{- 2 (m_{0}+m) \rho^{\vee}} ) =
\ell ( x t_{- 2 m_{0} \rho^{\vee}} ) + m\ell(t_{-2\rho^{\vee}}) =
\ell ( x t_{- 2 m_{0} \rho^{\vee}} ) + \ell(t_{-2m\rho^{\vee}})$$ for all $m \ge 0$.
In what follows, we fix $x \in W_{{\mathrm{af}}}^{\ge 0}$ unless stated otherwise. For this $x$, we take $m_{0} \ge 0$ as in Lemma \[tweakw\](2), and fix reduced expressions $$\label{eq:reduced}
x t_{- 2 m_{0} \rho^{\vee}} =
s_{i'_1} s_{i'_2} \cdots s_{i'_{\ell'}} \qquad \text{and} \qquad
t_{- 2 \rho^{\vee}} =
s_{i''_1} s_{i''_2} \cdots s_{i''_{\ell}},
$$ where $i'_1,\,\ldots,\,i'_{\ell'},\,i''_1,\,\ldots,\,i''_{\ell} \in I_{{\mathrm{af}}}$, with $\ell'=\ell(x t_{- 2 m_{0} \rho^{\vee}})$ and $\ell=\ell(t_{-2\rho^{\vee}})$. We concatenate these sequences periodically to obtain an infinite sequence $$\label{eq:bi}
{\mathbf{i}}= (
\underbrace{i'_1,\,i'_2,\,i'_3,\,\ldots,\,i'_{\ell'}}_{ \text{for $x t_{- 2 m_{0} \rho^{\vee}}$} },\,
\underbrace{i''_1,\,i''_2,\,\ldots,\,i''_{\ell}}_{\text{for $t_{-2\rho^{\vee}}$}},\,
\underbrace{i''_1,\,i''_2,\,\ldots,\,i''_{\ell}}_{\text{for $t_{-2\rho^{\vee}}$}},\,i''_1,\,\ldots)
\in I_{{\mathrm{af}}}^{\infty},$$ and write it as: ${\mathbf{i}}= (i_{1},\,i_{2},\,\ldots) \in I_{{\mathrm{af}}}^{\infty}$; remark that $s_{i_{1}}s_{i_{2}} \cdots s_{i_{k}}$ is reduced for all $k \ge 0$. For $k \in {\mathbb{Z}}_{\ge 0}$, we set ${\mathbf{i}}_{k} := (i_1,\,i_2,\,\ldots,\,i_{k})$.
Let $k \in {\mathbb{Z}}_{\ge 0}$, and let ${\mathbf{j}}= (i_{j_1},\,\dots,\,i_{j_t})$ be a subsequence of ${\mathbf{i}}_{k}$, where $1 \le j_{1} < \cdots < j_{t} \le k$. We set $\sigma ({\mathbf{j}}) = \sigma_{k}({\mathbf{j}}) : =
\bigl\{1,\,2,\,\ldots,\,k\bigr\} \setminus
\bigl\{j_{1},\,\dots,\,j_{t}\bigr\}$. We identify a subsequence ${\mathbf{j}}$ of ${\mathbf{i}}_{k}$ with a subsequence ${\mathbf{j}}'$ of ${\mathbf{i}}_{k'}$ if and only if $\sigma_{k}({\mathbf{j}}) = \sigma_{k'}({\mathbf{j}}')$ (as subsets of ${\mathbb{Z}}_{>0}$); namely, if $k' \ge k$, then ${\mathbf{j}}= (i_{j_1},\,\dots,\,i_{j_t}) \subset {\mathbf{i}}_{k}$ and ${\mathbf{j}}' = (i_{j_1},\,\dots,\,i_{j_t},\,i_{k+1},\,\dots,\,i_{k'})
\subset {\mathbf{i}}_{k'}$ are identified. Thus, we identify a subsequence ${\mathbf{j}}$ of ${\mathbf{i}}_{k}$ with a subsequence of ${\mathbf{i}}$ by taking the limit in $\varinjlim_{k} {\mathbf{i}}_{k} = {\mathbf{i}}$.
Let $k \in {\mathbb{Z}}_{\ge 0}$, and let ${\mathbf{j}}= (i_{j_1},\,i_{j_2},\,\ldots,\,i_{j_{t}})$ be a subsequence of ${\mathbf{i}}_{k}$. We set $$x({\mathbf{j}}; k) :=
s_{i_{j_1}} \ast s_{i_{j_2}} \ast \cdots \ast s_{i_{j_t}} \in {\mathbf{W}_{{\mathrm{af}}}}.$$
Let $k' \in {\mathbb{Z}}_{\ge 0}$ be such that $k' \ge k$. Because the sequence ${\mathbf{j}}$ above is identified with the subsequence $(i_{j_1},\,\dots,\,i_{j_t},\,i_{k+1},\,\dots,\,i_{k'})$ of ${\mathbf{i}}_{k'}$, we have $$x({\mathbf{j}}; k') :=
\underbrace{s_{i_{j_1}} \ast \cdots \ast s_{i_{j_t}}}_{=x({\mathbf{j}}; k)} \ast
s_{i_{k+1}} \ast \cdots \ast s_{i_{k'}} \in {\mathbf{W}_{{\mathrm{af}}}}.$$
\[stab\] Let $m \in {\mathbb{Z}}_{\ge 0}$, and let ${\mathbf{j}}$ be a subsequence of ${\mathbf{i}}_{ \ell' + m \ell}$. Then, there exists $m_1 \ge m$ such that $x ( {\mathbf{j}}; \ell' + m' \ell ) =
x ( {\mathbf{j}}; \ell' + m'' \ell) \cdot t_{-2 (m' -m'') \rho^{\vee}}$ for every $m' \ge m'' \ge m_{1}$. In particular, the element $x ( {\mathbf{j}}) := x ( {\mathbf{j}}; \ell' + m' \ell ) \cdot
t_{2(m'+m_{0})\rho^{\vee}} \in W_{{\mathrm{af}}}$ does not depend on the choice of $m' \ge m_{1}$.
We first note that $$x ( {\mathbf{j}}; \ell' + m' \ell ) =
x ( {\mathbf{j}}; \ell' + m'' \ell) \ast
\overbrace{
\underbrace{ ( s_{i''_1} \ast s_{i''_2} \ast \cdots \ast s_{i''_{\ell}} ) }_{\text{corresponding to $t_{-2\rho^{\vee}}$}}
\ast \cdots \ast
\underbrace{ ( s_{i''_1} \ast s_{i''_2} \ast \cdots \ast s_{i''_{\ell}} ) }_{\text{corresponding to $t_{-2\rho^{\vee}}$}}
}^{ \text{$(m'-m'')$ times} }.$$ Since $y \ast s_{i} = y s_{i}$ if and only if $\ell ( y s_{i} ) = \ell ( y ) + 1$ for $y \in W_{{\mathrm{af}}}$ and $i \in I_{{\mathrm{af}}}$, it suffices to show that there exists $m_{1} \ge m$ such that $$\label{eq:ell-0}
\ell ( x ( {\mathbf{j}}; \ell' + m'' \ell) \cdot t_{-2 n \rho^{\vee}} ) =
\ell ( x ( {\mathbf{j}}; \ell' + m'' \ell) ) + \ell ( t_{-2 n \rho^{\vee}} )$$ for all $n > 0$ and $m'' \ge m_{1}$. Let $k \in {\mathbb{Z}}_{\ge 0}$ be such that $k \ge m$. Since $x ( {\mathbf{j}}; \ell' + k \ell) =
x ( {\mathbf{j}}; \ell' + m \ell) \ast t_{-2(k-m)\rho^{\vee}}$ and $\ell(t_{-2(k-m)\rho^{\vee}}) = (k-m)\ell$, we see that $$\label{eq:ell-1}
\ell ( x ( {\mathbf{j}}; \ell' + k \ell) ) \ge (k - m) \ell;$$ note that $\ell (y \ast y') \ge
\max \bigl\{ \ell ( y ), \, \ell ( y' ) \bigr\}$ for $y,\,y' \in W_{{\mathrm{af}}}$, as is verified by induction. Now, for each integer $k \ge m$, we set $$d_{k}:= \underbrace{\ell ( t_{- 2 \rho^{\vee}} )}_{=\ell} -
\bigl\{
\ell ( x ( {\mathbf{j}}; \ell' + ( k + 1 ) \ell) ) -
\ell ( x ( {\mathbf{j}}; \ell' + k \ell) ) \bigr\};$$ observe that $d_{k} \in {\mathbb{Z}}_{\ge 0}$. Also, for $k \ge m$, we have $$(k-m)\ell =
\ell ( x ( {\mathbf{j}}; \ell' + k \ell) ) -
\ell ( x ( {\mathbf{j}}; \ell' + m \ell) ) +
\sum_{k'=m}^{k-1} d_{k'}.$$ If $d_{k} > 0$ for infinitely many $k \ge m$, then $(k-m)\ell > \ell ( x ( {\mathbf{j}}; \ell' + k \ell) )$ for $k \gg m$, which contradicts . Hence we deduce that $d_{k} > 0$ only for finitely many $k \ge m$. Thus, if we set $m_{1}:=\max \bigl\{ k \ge m \mid d_{k} > 0 \bigr\}$, then holds. This proves the lemma.
\[si-refl\] For each $y,\,y' \in W_{{\mathrm{af}}}$ such that $y {\preceq}y'$, there exists $m_{2} \in {\mathbb{Z}}_{\ge 0}$ such that $y t_{-2m\rho^{\vee}} \ge y' t_{-2m\rho^{\vee}}$ in the ordinary Bruhat order on $W_{{\mathrm{af}}}$ for all $m \ge m_{2}$.
It suffices to prove the assertion in the case that $y {\xrightarrow{\,\beta\,}} s_{\beta}y = y'$ for some $\beta \in {\Delta_{{\mathrm{af}}}^{+}}$. Here we see from [@INS Corollary 4.2.2] that $\beta$ is either of the following forms:
\(i) $\beta = \alpha$ with $\alpha \in \Delta^{+}$; (ii) $\beta = \alpha+\delta$ with $-\alpha \in \Delta^{+}$.
Moreover, if $y = w t_{\xi}$ with $w \in W$ and $\xi \in Q^{\vee}$, then $\gamma:=w^{-1}\alpha \in \Delta^{+}$ in both cases above. Also, it follows from [@INS Proposition A.1.2] that $$\label{eq:ell}
\ell(w) =
\begin{cases}
\ell(wr_{\gamma})-1 & \text{in case (i)}, \\[1mm]
\ell(wr_{\gamma})-1+2 {\langle \rho,\,\gamma^{\vee} \rangle} & \text{in case (ii)}.
\end{cases}
$$
If we set $\zeta:=\xi-2m\rho^{\vee}$ for $m \in {\mathbb{Z}}$, then $y t_{- 2 m \rho^{\vee}} = w t _{\xi - 2 m \rho^{\vee}} = w t_{\zeta}$, and $$\begin{aligned}
s_{\beta} (y t_{- 2 m \rho^{\vee}}) & =
s_{\alpha+k\delta} w t_{\zeta},
\quad \text{where $k=0$ in case (i), and $k=1$ in case (ii)}, \nonumber \\
& = wt_{\zeta} s_{t_{-\zeta}w^{-1}\alpha+k\delta}
= (y t_{- 2 m \rho^{\vee}}) s_{\gamma + n \delta},
\quad \text{with $n:=k+ {\langle \gamma,\,\zeta \rangle}$}. \label{eq:n}\end{aligned}$$ Therefore, in case (i) (resp., case (ii)), we deduce from [@LNSSS Proposition 5.1(1) (resp., (2)) with $v=e$], together with equalities and that $y't_{-2m\rho^{\vee}} = s_{\beta}y t_{-2m\rho^{\vee}} =
(y t_{-2m\rho^{\vee}}) s_{\gamma + n \delta} < y t_{-2m \rho^{\vee}}$ for all $m \gg 0$. This proves the lemma.
\[BSDH-cover\] For each $y \in W_{{\mathrm{af}}}$ such that $y {\succeq}x$, there exist $k \in {\mathbb{Z}}_{\ge 0}$ and a subsequence ${\mathbf{j}}$ of ${\mathbf{i}}_{k}$ such that $y = x({\mathbf{j}})$.
By Lemma \[si-refl\], there exists $m \gg 0$ such that the assertion of Lemma \[tweakw\](2) holds for $m_{0}+m$ instead of $m_{0}$ (see also Remark \[rem:ell\]), and such that $$yt_{- 2 (m_{0}+m) \rho^{\vee}} <
x t _{- 2 (m_{0}+m) \rho^{\vee}} =
x t _{- 2 m_{0} \rho^{\vee}} \cdot t _{- 2 m \rho^{\vee}};$$ note that ${\mathbf{i}}_{\ell'+m\ell}$ gives a reduced expression for $x t _{- 2 (m_{0}+m) \rho^{\vee}}$. By the Subword Property (see, e.g., [@BB Theorem 2.2.2]), $y t_{- 2 (m_{0}+m) \rho^{\vee}}$ is obtained as a subexpression of the reduced expression for $x t_{- 2 (m_{0}+m) \rho^{\vee}}$ corresponding to ${\mathbf{i}}_{\ell'+m\ell}$. Namely, there exists a subsequence ${\mathbf{j}}=(i_{j_1},\,\dots,\,i_{j_t})$ of ${\mathbf{i}}_{\ell'+m\ell}$ such that $$y t_{- 2 (m_{0}+m) \rho^{\vee}} =
s_{i_{j_1}} \cdots s_{i_{j_t}} =
s_{i_{j_1}} \ast \cdots \ast s_{i_{j_t}} =
x({\mathbf{j}},\ell'+m\ell).$$ Let us take $m_{1} \gg m$ as in Lemma \[stab\]. Then, we see by Remark \[rem:ell\] that for $m' \ge m_{1}$, $$\begin{aligned}
x({\mathbf{j}},\ell'+m'\ell) & =
x({\mathbf{j}},\ell'+m\ell) \ast t_{-2(m'-m)\rho^{\vee}} \\
& =
x({\mathbf{j}},\ell'+m\ell) \cdot t_{-2(m'-m)\rho^{\vee}} =
y t_{- 2 (m'+m_{0}) \rho^{\vee}}.\end{aligned}$$ Thus, we obtain $y = x({\mathbf{j}})$, as desired. This proves the lemma.
For each $i \in I_{{\mathrm{af}}}$ and $y \in W_{{\mathrm{af}}}$, we define a map $$\label{eq:qiy}
q_{i,y} : {\mathbf{I}}(i) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}(y) \rightarrow {\mathbf{Q}_{G}^{\mathrm{rat}}}, \quad
(p,\,L) \mapsto pL.$$
\[one-step\] Let $y \in W_{{\mathrm{af}}}^{\ge 0}$, and $i \in I_{{\mathrm{af}}}$. If $s_{i} y {\succeq}y$, then the map $q_{i,y}$ induces a ${\mathbb{P}}^{1}$-fibration ${\mathbf{I}}(i) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}(y) \rightarrow {\mathbf{Q}_{G}}(y)$, which we also denote by $q_{i,y}$. If $e \preceq s_i y \preceq y$, then the map $q_{i,y}$ induces a birational map ${\mathbf{I}}(i) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}(y) \rightarrow {\mathbf{Q}_{G}}(s_{i}y)$, which we also denote by $q_{i,y}$. In both cases, the map $q_{i,y}$ is proper.
If $s_{i} y {\succeq}y$, then the action of ${\mathbf{I}}(i)$ stabilizes ${\mathbf{O}}(y) \cup {\mathbf{O}}( s_{i} y ) \subset {\mathbf{Q}_{G}}( y )$. Hence taking the closure in ${\mathbf{Q}_{G}}$ implies that ${\mathbf{Q}_{G}}( y )$ admits an ${\mathbf{I}}( i )$-action. Therefore, the assertions hold in this case since ${\mathbf{I}}( i )/ {\mathbf{I}}\cong {\mathbb{P}}^{1}$.
If $e \preceq s_{i} y \preceq y$, then $q_{i,y}^{-1} ( {\mathbf{O}}( s_{i} y ) )$ contains ${\mathbf{I}}\dot{s}_i {\mathbf{I}}\times ^{{\mathbf{I}}} {\mathbf{O}}(y)$. Here we deduce from Lemmas \[lem:si\] and \[lem:dmd\] that if $y' \in W_{{\mathrm{af}}}$ satisfies $y' {\succ}y$, then $s_{i}y' {\succ}s_{i}y$; in particular, we have ${\ell^{{\frac{\infty}{2}}}}( s_i y ) < {\ell^{{\frac{\infty}{2}}}}( s_i y' )$. Hence we have ${\mathbf{I}}\dot{s}_i {\mathbf{I}}\times^{{\mathbf{I}}} {\mathbf{O}}( y ) =
q_{i,y}^{-1} ( {\mathbf{O}}( s_{i} y ) )$. In addition, the unipotent one-parameter subgroup of ${\mathbf{I}}$ corresponding to $\alpha_{i}$ gives an isomorphism ${\mathbb{A}}^{1} \times \dot{s}_i {\mathbf{O}}( y ) \cong {\mathbf{O}}( s_{i} y )$. Therefore, $q_{i,y}$ is birational. Also, by applying the same observation for $s_{i}y$ instead of $y$, we deduce that $q_{i,y}$ is obtained as the restriction of the ${\mathbb{P}}^{1}$-fibration $q_{i,s_{i}y}$ to a closed subscheme. Hence $q_{i,y}$ defines a proper map. This proves the lemma.
For each $k \in {\mathbb{Z}}_{\ge 0}$, we set $x(k) := s_{i_{k}} s_{i_{k-1}} \cdots s_{i_{1}} x$. We claim that $$\label{eq:xk}
{\ell^{{\frac{\infty}{2}}}}( x(k+1) ) = {\ell^{{\frac{\infty}{2}}}}(x (k) ) + 1 \quad \text{for all $k \ge 0$}.$$ Indeed, since $x(k+1) = s_{i_{k}}x(k)$ for $k \ge 0$, we see by that ${\ell^{{\frac{\infty}{2}}}}( x(k+1) ) = {\ell^{{\frac{\infty}{2}}}}(x (k) ) \pm 1$ for each $k \ge 0$. Therefore, it suffices to show that $$\label{eq:xm}
{\ell^{{\frac{\infty}{2}}}}(x((m-m_{0})\ell + \ell')) =
{\ell^{{\frac{\infty}{2}}}}(x(0)) + (m-m_{0})\ell + \ell' \qquad
\text{for all $m \ge m_{0}$};$$ note that $x(0) = x$. We see by the definition that $x((m-m_{0})\ell + \ell') = (xt_{-2m\rho^{\vee}})^{-1}x =
t_{2m\rho^{\vee}}$. Hence we compute: $$\begin{aligned}
{\ell^{{\frac{\infty}{2}}}}(x((m-m_{0})\ell + \ell')) &
= {\ell^{{\frac{\infty}{2}}}}(t_{2m\rho^{\vee}}) = 2{\langle \rho,\,2m\rho^{\vee} \rangle}
= -m {\ell^{{\frac{\infty}{2}}}}(t_{-2\rho^{\vee}}) \nonumber \\
& = m \ell(t_{-2\rho^{\vee}}) = m\ell \quad \text{by Lemma~\ref{lem:Mac}}.
\label{eq:xk1}\end{aligned}$$ Also, by Lemma \[tweakw\](1), we have ${\ell^{{\frac{\infty}{2}}}}(xt_{-2m_{0}\rho^{\vee}}) + {\ell^{{\frac{\infty}{2}}}}(t_{2m_{0}\rho^{\vee}}) = {\ell^{{\frac{\infty}{2}}}}(x)$. Here we deduce that ${\ell^{{\frac{\infty}{2}}}}(xt_{-2m_{0}\rho^{\vee}}) = - \ell(xt_{-2m_{0}\rho^{\vee}}) = -\ell'$ by Lemma \[tweakw\](2), and that ${\ell^{{\frac{\infty}{2}}}}(t_{2m_{0}\rho^{\vee}}) = m_{0}\ell$. Hence we obtain ${\ell^{{\frac{\infty}{2}}}}(x) = -\ell' + m_{0}\ell$. Combining this equality with shows , as desired.
Now, we set $$\label{defbQi}
{\mathbf{Q}_{G}}( {\mathbf{i}}_k ) :=
{\mathbf{I}}( i_1 ) \times^{{\mathbf{I}}} {\mathbf{I}}( i_2 ) \times^{{\mathbf{I}}} \cdots
\times^{{\mathbf{I}}} {\mathbf{I}}( i_k ) \times^{{\mathbf{I}}} {\mathbf{O}}( x ( k ) ).$$ We also define its ambient space $${\mathbf{Q}}^{\#}_G ( {\mathbf{i}}_k ) :=
{\mathbf{I}}( i_1 ) \times^{{\mathbf{I}}} {\mathbf{I}}( i_2 ) \times^{{\mathbf{I}}} \cdots \times^{{\mathbf{I}}}
{\mathbf{I}}( i_k ) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}( x ( k ) ).$$ Since $s_{i_{k}} x ( k ) = x ( k - 1 )$ and $x ( k ) {\succeq}x ( k - 1 )$ for each $k \ge 1$ (see ), we have an ${\widetilde{{\mathbf{I}}}}$-equivariant embedding ${\mathbf{O}}( x ( k - 1 ) ) \hookrightarrow {\mathbf{I}}( i_k ) \times^{{\mathbf{I}}} {\mathbf{O}}( x ( k ) )$ by the latter case of Lemma \[one-step\], and hence an ${\widetilde{{\mathbf{I}}}}$-equivariant embedding ${\mathbf{Q}_{G}}( {\mathbf{i}}_{k-1} ) \hookrightarrow {\mathbf{Q}_{G}}( {\mathbf{i}}_k )$ for each $k \ge 1$. By infinite repetition of these embeddings, we obtain a scheme of infinite type $$\label{Qunion}
{\mathbf{Q}_{G}}( {\mathbf{i}}) := \varinjlim_k {\mathbf{Q}_{G}}( {\mathbf{i}}_k ) = \bigcup_{k \ge 0} {\mathbf{Q}_{G}}( {\mathbf{i}}_k ),$$ endowed with an ${\widetilde{{\mathbf{I}}}}$-action. Also, the multiplication of components yields an ${\widetilde{{\mathbf{I}}}}$-equivariant morphism $${\mathsf{m}}: {\mathbf{Q}_{G}}( {\mathbf{i}}) \longrightarrow {\mathbf{Q}_{G}}( x ).$$ Similarly, we have ${\mathsf{m}}_{k} : {\mathbf{Q}_{G}}^{\#}( {\mathbf{i}}_k ) \rightarrow {\mathbf{Q}_{G}}( x )$ for each $k \ge 0$. The natural inclusion ${\mathbf{Q}_{G}}( {\mathbf{i}}_k ) \hookrightarrow {\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}}_k )$ yields a map ${\mathbf{Q}_{G}}({\mathbf{i}}) \to {\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}}_k )$ by taking the product of factors at position $> k$ from the left in . Namely, we collect the maps $${\mathbf{Q}_{G}}( {\mathbf{i}}_{k'} ) \ni
( p_1,\,\ldots,\,p_{k'},\,L)
\mapsto
(p_1,\,\ldots,\,p_k,\,p_{k+1} \cdots p_{k'}L) \in {\mathbf{Q}_{G}}^{\#}( {\mathbf{i}}_k )$$ for $k'> k$ through , where $p_{j} \in {\mathbf{I}}( i_j )$, $1 \le j \le k'$, and $L \in {\mathbf{O}}(x(k'))$; here each closed point of ${\mathbf{Q}_{G}}( {\mathbf{i}}_{k'} )$ is an equivalence class with respect to the ${\mathbf{I}}^{k'}$-action, and the map above respects equivalence classes. This yields a factorization of ${\mathsf{m}}$ through arbitrary ${\mathsf{m}}_{k}$ in such a way that $$\label{factorBSDH}
{\mathbf{Q}_{G}}( {\mathbf{i}}) \rightarrow
{\mathbf{Q}_{G}}^{\#}( {\mathbf{i}}_{k'} ) \rightarrow
{\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}}_k ) \rightarrow {\mathbf{Q}_{G}}( x )$$ for each $k' \ge k$. This also yields an inclusion $${\mathbf{Q}_{G}}( {\mathbf{i}}) \hookrightarrow
{\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}}) := \varprojlim_k {\mathbf{Q}}^{\#}_G ( {\mathbf{i}}_k ),$$ which fits into the following commutative diagram of ${\widetilde{{\mathbf{I}}}}$-equivariant morphisms for $k < k'$: $$\xymatrix{
{\mathbf{Q}_{G}}({\mathbf{i}}) \ar@{^{(}->}[rr] \ar[d]_{{\mathsf{m}}} \ar[drr] \ar[drrr] & &
{\mathbf{Q}_{G}}^{\#}({\mathbf{i}}) \ar[dll] \ar[d] \ar[dr] & \\
{\mathbf{Q}_{G}}(x) & & {\mathbf{Q}_{G}}^{\#}({\mathbf{i}}_k) \ar[ll]^{{\mathsf{m}}_k} &
{\mathbf{Q}_{G}}^{\#}({\mathbf{i}}_{k'}). \ar[l]
}$$
\[up-normal\] The scheme ${\mathbf{Q}_{G}}( {\mathbf{i}})$ is separated and normal.
The scheme ${\mathbf{Q}_{G}}( {\mathbf{i}})$ is an inclusive union of countably many open subschemes each of which is isomorphic to a pro-affine space bundle over a finite successive ${\mathbb{P}}^{1}$-fibrations. Since each of such a space is separated, we deduce the desired separatedness.
Also, by its construction, each ${\mathbf{Q}_{G}}( {\mathbf{i}}_{k} )$ is a union of pro-affine spaces labeled by subsequences of ${\mathbf{i}}_k$ (so that ${\mathbf{Q}_{G}}({\mathbf{i}}_{k})$ is covered by a total of $2^{k}$-copies of open cover consisting of pro-affine spaces). Thus, ${\mathbf{Q}_{G}}( {\mathbf{i}})$ is a union of countably many pro-affine spaces. Since all of these pieces are normal, we deduce the desired normality. This proves the lemma.
For each subsequence ${\mathbf{j}}\subset {\mathbf{i}}_{k}$, we obtain an ${\mathbf{I}}$-stable pro-affine space ${\mathbf{O}}( {\mathbf{j}}) \subset {\mathbf{Q}_{G}}( {\mathbf{i}}_{k} )$ by replacing ${\mathbf{I}}( i_{j} )$, $1 \le j \le k$, with ${\mathbf{I}}\dot{s}_{i_j} {\mathbf{I}}$ (resp., ${\mathbf{I}}$) if $i_{j} \in {\mathbf{j}}$ (resp., $i_{j} \not\in {\mathbf{j}}$) in ; we refer to ${\mathbf{O}}( {\mathbf{j}})$ as the stratum corresponding to a subsequence ${\mathbf{j}}$ of ${\mathbf{i}}_k$.
\[k-stratum\]
For each $k \in {\mathbb{Z}}_{\ge 1}$, it holds that ${\mathbf{Q}_{G}}( {\mathbf{i}}_k ) = \bigsqcup _{{\mathbf{j}}\subset {\mathbf{i}}_k} {\mathbf{O}}( {\mathbf{j}})$.
Let $k \in {\mathbb{Z}}_{\ge 1}$, and let ${\mathbf{j}}$ and ${\mathbf{j}}'$ be subsequences of ${\mathbf{i}}_k$. Then, ${\mathbf{O}}( {\mathbf{j}}) \subset {\overline{{\mathbf{O}}({\mathbf{j}}')}}$ in ${\mathbf{Q}_{G}}( {\mathbf{i}}_{k} )$ if and only if ${\mathbf{j}}\subset {\mathbf{j}}' (\subset {\mathbf{i}}_k)$.
The proofs of the assertions are straightforward by the definitions.
If we regard ${\mathbf{O}}( {\mathbf{j}})$ as a locally closed subscheme of ${\mathbf{Q}_{G}}( {\mathbf{i}})$ via for ${\mathbf{j}}\subset {\mathbf{i}}_k$, then its images in ${\mathbf{Q}_{G}}^{\#}( {\mathbf{i}}_k )$ and ${\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}}_{k'})$ for $k < k'$ are isomorphic through . Therefore, we can freely replace ${\mathsf{m}}$ with ${\mathsf{m}}_{k}$ when we analyze a single stratum in ${\mathbf{Q}_{G}}({\mathbf{i}})$. We set ${\mathbf{Q}_{G}}({\mathbf{j}}) := \overline{{\mathbf{O}}({\mathbf{j}})}$, where the closure is taken in ${\mathbf{Q}_{G}}( {\mathbf{i}})$.
We have ${\mathbf{Q}_{G}}( {\mathbf{i}}) = \bigsqcup_{{\mathbf{j}}} {\mathbf{O}}( {\mathbf{j}})$, where ${\mathbf{j}}$ runs over all subsequences of ${\mathbf{i}}$ such that $|\sigma({\mathbf{j}})| < \infty$.
Since ${\mathbf{Q}_{G}}({\mathbf{i}}) = \bigcup_{k \ge 1} {\mathbf{Q}_{G}}( {\mathbf{i}}_k )$, the assertion is a consequence of Lemma \[k-stratum\] and the consideration just above.
The map ${\mathsf{m}}$ is surjective.
By Lemma \[BSDH-cover\], each fiber of ${\mathsf{m}}$ along an $(H \times {\mathbb{G}_{m}})$-fixed point is nonempty. Because applying the $\tilde{{\mathbf{I}}}$-action to the set of $(H \times {\mathbb{G}_{m}})$-fixed points exhausts ${\mathbf{Q}_{G}}( x )$ by Corollary \[orbits\], we conclude that ${\mathsf{m}}$ is surjective, as required. This proves the lemma.
The map ${\mathsf{m}}_k$ is ${\widetilde{{\mathbf{I}}}}$-equivariant, birational, and proper for each $k \ge 1$.
Since ${\ell^{{\frac{\infty}{2}}}}( x ( k ) ) - {\ell^{{\frac{\infty}{2}}}}( x ) = k$ for each $k \ge 1$, repeated application of Lemma \[one-step\] shows that ${\mathsf{m}}_k$ is birational and proper for each $k \ge 1$; this is because ${\mathsf{m}}_k$ is obtained as the base change of the composite of morphisms of type $q_{i_j,\,x(j)}$ for $1 \le j \le k$. This proves the lemma.
The map ${\mathsf{m}}$ is also ${\widetilde{{\mathbf{I}}}}$-equivariant and birational since the embedding ${\mathbf{Q}_{G}}({\mathbf{i}}) \subset {\mathbf{Q}_{G}}^{\#} ( {\mathbf{i}})$ is open.
Cohomology of line bundles over ${\mathbf{Q}_{G}}$. {#subsec:coh}
---------------------------------------------------
In this subsection, keeping the setting of the previous subsection, we also assume that $x=e$, the identity element; in this case, we can (and do) take the $m_{0}$ (in Lemma \[tweakw\](2)) to be $0$, so that we have $\ell'=0$ (see ). In particular, we have ${\mathbf{Q}_{G}}( x ) = {\mathbf{Q}_{G}}( e ) = {\mathbf{Q}_{G}}$.
For each $\lambda \in P$, the ${\widetilde{{\mathbf{I}}}}$-equivariant line bundle ${\mathcal{O}}_{{\mathbf{Q}_{G}}( x( k ) )} ( \lambda )$ induces a line bundle over ${\mathbf{Q}_{G}}({\mathbf{i}}_{k})$ for each $k \in {\mathbb{Z}}_{\ge 0}$.
\[H-range\] For each $\lambda \in P$, we have an isomorphism $$H^{0}( {\mathbf{Q}_{G}}({\mathbf{i}}), \,{\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})}(\lambda) ) \cong W(\lambda)^{\ast}$$ of ${\widetilde{{\mathbf{I}}}}$-modules. In particular, the left-hand side carries a natural structure of graded ${\mathfrak{g}}[z]$-module.
We adopt the notation of the previous subsection, with $x=e$, $m_{0}=0$, and $\ell' = 0$. Also, since $\ell ( t_{- 2 \rho^{\vee}} ) =
\ell ( w_{\circ} ) + \ell ( w_{\circ} t_{- 2 \rho^{\vee}} )$, we can rearrange the reduced expression $(i''_1,\,\ldots,\,i''_{\ell}) = (i_1,\,\ldots,\,i_{\ell})$ for $t_{- 2 \rho^{\vee}}$, if necessary, in such a way that the first $\ell(w_{\circ})$-entries $(i_1,\,\ldots,\,i_{\ell(w_{\circ})})$ give rise to a reduced expression for $w_{\circ}$.
The map ${\mathsf{m}}$ factors as ${\mathsf{m}}' \circ {\mathsf{m}}''$, where ${\mathsf{m}}''$ is the map obtained by collapsing the first $\ell ( w_{\circ} )$-factors in ${\mathbf{Q}_{G}}( {\mathbf{i}})$ as: $$\label{eq:mapm}
{\mathbf{Q}_{G}}( {\mathbf{i}}) = {\mathbf{I}}( i_1 ) \times^{{\mathbf{I}}} \cdots \times^{{\mathbf{I}}}
{\mathbf{I}}(i_{\ell ( w_{\circ} )}) \times ^{{\mathbf{I}}} X
\rightarrow G {[\hspace{-1pt}[z]\hspace{-1pt}]} \times ^{{\mathbf{I}}} X,$$ with $X$ a certain scheme admitting an ${\widetilde{{\mathbf{I}}}}$-action, and ${\mathsf{m}}'$ is the natural map $G{[\hspace{-1pt}[z]\hspace{-1pt}]} \times^{{\mathbf{I}}} X \rightarrow {\mathbf{Q}_{G}}$ induced by the action. In particular, we deduce that ${\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} ( \lambda )$ admits a ${\widetilde{G}}{[\hspace{-1pt}[z]\hspace{-1pt}]}$-action, and hence that the space of global sections of ${\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} ( \lambda )$ is a ${\mathfrak{g}}[z]$-module.
Now, for each $m \in {\mathbb{Z}}_{\ge 0}$, we have ${\mathbf{Q}_{G}}( {\mathbf{i}}_{\ell(w_{\circ}) + \ell m} ) \subset {\mathbf{Q}_{G}}({\mathbf{i}})$. Also, for each $k \in {\mathbb{Z}}_{\ge 0}$, we have an inclusion ${\mathbf{Q}_{G}}( {\mathbf{i}}_k ) \hookrightarrow {\mathbf{Q}_{G}}( {\mathbf{i}}_{k + \ell} )$ by sending the stratum ${\mathbf{O}}({\mathbf{j}})$ corresponding to ${\mathbf{j}}\subset {\mathbf{i}}_k$ to the one corresponding to ${\mathbf{j}}_{+} \subset {\mathbf{i}}_{k + \ell}$, which is obtained by shifting all the entries by $\ell$ (so that $\{1,\,2,\,\ldots,\,\ell\} \cap {\mathbf{j}}_{+} = \emptyset$ and $|\sigma_{k+\ell} ({\mathbf{j}}_{+})| = |\sigma_{k} ({\mathbf{j}})| + \ell$); this corresponds to twisting ${\mathop{\rm Image}\nolimits}{\mathsf{m}}= {\mathbf{Q}_{G}}$ to ${\mathbf{Q}_{G}}( t_{2 \rho^{\vee}} )$. It follows that the ${\mathbb{G}_{m}}$-action on the line bundle ${\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}}_k )} ( \lambda )$ is twisted by $t_{2 \rho^{\vee}}$, whose actual effect is of degree $- 2 {\langle \lambda,\,\rho^{\vee} \rangle}$. In particular, the structure map $\pi_{m} : {\mathbf{Q}_{G}}( {\mathbf{i}}_{\ell(w_{\circ})+\ell m} )
\rightarrow \bigl\{ \mathrm{pt} \bigr\}$ for $m \in {\mathbb{Z}}_{\ge 0}$ factors as: $$\label{factor-pi}
\pi_{m} \cong
\bigl( ( \pi_{1})^{\ast} \bigr)^{m} ( \pi_{0} ),$$ where the maps $\pi_0$ and $\pi_1$ are projections $$\begin{aligned}
\pi_{0} & :
{\mathbf{I}}( i_{1} ) \times^{{\mathbf{I}}} {\mathbf{I}}( i_{2} ) \times^{{\mathbf{I}}} \cdots \times^{{\mathbf{I}}}
{\mathbf{I}}( i_{\ell(w_{\circ})}) \times^{{\mathbf{I}}}
{\mathbf{O}}( w_{\circ} ) \rightarrow \bigl\{ \mathrm{pt} \bigr\}, \nonumber \\
\pi_{1} & :
{\mathbf{I}}( i_{1} ) \times^{{\mathbf{I}}} {\mathbf{I}}( i_{2} ) \times^{{\mathbf{I}}} \cdots \times^{{\mathbf{I}}}
{\mathbf{I}}( i_{\ell} ) / {\mathbf{I}}\rightarrow \bigl\{ \mathrm{pt} \bigr\}, \label{pi1}\end{aligned}$$ and $(\pi_{1})^{\ast}$ means the pullback obtained by identifying $\bigl\{\mathrm{pt}\bigr\}$ with ${\mathbf{I}}/ {\mathbf{I}}\in {\mathbf{I}}( i_{\ell} ) / {\mathbf{I}}$ in .
For each $i \in I_{{\mathrm{af}}}$ and an ${\mathbf{I}}$-module $M$, we define ${\mathbf{D}}_{i}(M)$ to be the space of $(H \times {\mathbb{G}_{m}})$-finite vectors in $H^{0} ( {\mathbf{I}}(i) / {\mathbf{I}}, \, {\mathbf{I}}( i ) \times^{{\mathbf{I}}} M^{\ast} )$; this can be thought as a left exact endo-functor in the category of $(H \times {\mathbb{G}_{m}})$-semi-simple ${\mathfrak{I}}$-modules. We set $${\mathbf{D}}_{- 2 \rho^{\vee}} :=
{\mathbf{D}}_{s_{i_1}} \circ {\mathbf{D}}_{s_{i_2}} \circ \cdots \circ {\mathbf{D}}_{s_{i_{\ell}}}
\quad \text{and} \quad
{\mathbf{D}}_{w_{\circ}} :=
{\mathbf{D}}_{s_{i_1}} \circ {\mathbf{D}}_{s_{i_2}} \circ \cdots \circ {\mathbf{D}}_{s_{i_{\ell ( w_{\circ} )}}}.$$ Then, successive application of the Leray spectral sequence to , together with the fact that the ${\mathbb{G}_{m}}$-twist commutes with the whole construction, yields $$H^0 ( {\mathbf{Q}_{G}}({\mathbf{i}}) ,\, {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} ( \lambda ))
\stackrel{\cong}{\longrightarrow}
\varprojlim_{m} {\mathbf{D}}_{- 2 \rho^{\vee}}^{m}
\left( {\mathbf{D}}_{w_{\circ}}
\left( {\mathbb{C}}[ {\mathbf{O}}( w_{\circ} t_{2 m \rho^{\vee}} ) ] \otimes
{\mathbb{C}}_{- t_{- 2 m \rho^{\vee}} \lambda} \right) \right),$$ where we have used the equality $w_{\circ} t_{2 m \rho^{\vee}} w_{\circ}
= t_{- 2 m \rho^{\vee}}$.
In view of with $X$ replaced by ${\mathbf{O}}( w_{\circ} t_{2 m \rho^{\vee}})$, Theorem \[BFmain\] implies $${\mathbf{D}}_{w_{\circ}} \left(
{\mathbb{C}}[ {\mathbf{O}}( w_{\circ} t_{2 m \rho^{\vee}} ) ] \otimes
{\mathbb{C}}_{- t_{- 2 m \rho^{\vee}} \lambda} \right) \cong W ( \lambda )^{\ast},$$ where the grading of the right hand side is shifted by $- 2m {\langle \lambda,\,\rho^{\vee} \rangle}$. Hence, [@Kat16 Theorem 4.13] (cf. [@Kas05 Lemma 2.8] and [@Kat16 Corollary 4.8]) implies $${\mathbf{D}}_{- 2 \rho^{\vee}}^m \left( {\mathbf{D}}_{w_{\circ}} \left(
{\mathbb{C}}[ {\mathbf{O}}( w_{\circ} t_{2 m \rho^{\vee}} ) ] \otimes
{\mathbb{C}}_{- t_{- 2 m \rho^{\vee}} \lambda} \right) \right) \cong W(\lambda)^{\ast}$$ without grading shift. Therefore, we conclude $$H^{0} ( {\mathbf{Q}_{G}}( {\mathbf{i}}), {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})}(\lambda)) \cong W(\lambda)^{\ast}$$ for the choice of ${\mathbf{i}}$ fixed at the beginning of the proof. The assertion for a general ${\mathbf{i}}$ follows from the fact that $${\mathbf{D}}_{- 2 \rho^{\vee}}^m \cong {\mathbf{D}}_{s_{j_1}} \circ \cdots \circ {\mathbf{D}}_{s_{j_{m\ell}}}$$ holds for an arbitrary reduced expression $s_{j_1} \cdots s_{j_{m\ell}}$ for $t_{- 2 m \rho^{\vee}}$. This completes the proof of the proposition.
\[w-approx\] The subset $G[z] \subset G{[\hspace{-1pt}[z]\hspace{-1pt}]}$ is dense.
The original claim is that $G{(\hspace{-1pt}(z)\hspace{-1pt})}$ is generated by the set of ${\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$-valued points of the unipotent groups of $G$. Because the set of ${\mathbb{C}}(z)$-valued points in a one-dimensional unipotent group is dense in the set of ${\mathbb{C}}{(\hspace{-1pt}(z)\hspace{-1pt})}$-valued points in the sense that we can approximate the latter by the former up to an arbitrary order of $z$, it follows that we can approximate $G{(\hspace{-1pt}(z)\hspace{-1pt})}$ by elements of $G(z)$ up to an arbitrary order of $z$. Since such an approximation of an element of $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$ is achieved by elements that are regular at $z = 0$, we conclude the assertion.
\[normal\] We have ${\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})} \cong {\mathcal{O}}_{{\mathbf{Q}_{G}}}$. In particular, the scheme ${\mathbf{Q}_{G}}$ is normal.
We (can) employ the same reduced expression for $t_{-2\rho^{\vee}}$ as in the proof of Proposition \[H-range\]; recall the last sentence of the proof. The pullback defines a map ${\mathsf{m}}^{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}} \rightarrow {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})}$, whose adjunction in turn yields ${\mathcal{O}}_{{\mathbf{Q}_{G}}} \to {\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})}$. From this, by twisting by ${\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )$ for some $\lambda \in P^{+}$, we obtain the following short exact sequence: $$0 \rightarrow {\mathcal{O}}_{{\mathbf{Q}_{G}}} (\lambda) \rightarrow
{\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})} (\lambda) \rightarrow
{\mathop{\rm Coker}\nolimits}\rightarrow 0,$$ from which we deduce a ${\mathfrak{g}}[z]$-module inclusion $$H^0 ( {\mathbf{Q}_{G}}, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) ) \hookrightarrow
H^0 ( {\mathbf{Q}_{G}}, {\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} ( \lambda ) ) \cong
H^0 ( {\mathbf{Q}_{G}}( {\mathbf{i}}), {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} ( \lambda ) ),$$ by taking their global sections. The rightmost one is isomorphic to $W(\lambda)^{\ast}$ by Proposition \[H-range\]. In particular, we have algebra homomorphisms: $$\begin{aligned}
\bigoplus _{\lambda \in P^{+}}
\Gamma ( {\mathbf{Q}_{G}}, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) )
& \subset
\bigoplus _{\lambda \in P^{+}}
\Gamma ( {\mathbf{Q}_{G}}, {\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})} ( \lambda ) ) \\
& \cong \bigoplus _{\lambda \in P^{+}} \Gamma ( {\mathbf{Q}_{G}}({\mathbf{i}}), {\mathcal{O}}_{{\mathbf{Q}_{G}}({\mathbf{i}})} ( \lambda ) ) \\
& \cong \bigoplus _{\lambda \in P^{+}} W ( \lambda )^{\ast};\end{aligned}$$ let us denote the leftmost one by $R'_{G}$ and the rightmost one by $R_{G}$. Since ${\mathbf{Q}_{G}}({\mathbf{i}})$ is normal, we deduce that $R_{G}$ is normal when localized with respect to homogeneous elements in $W({\varpi}_{i})^{\ast}$ for each $i \in I$. For the same reason, $R_{G}$ is an integral domain. The ring structure of $R_{G}$ is induced by the unique (up to scalar) ${\mathfrak{g}}[z]$-module map $$\label{d-mult}
W(\lambda + \mu) \longrightarrow W (\lambda) \otimes_{{\mathbb{C}}} W ( \mu ), \qquad
\lambda,\,\mu \in P^{+},$$ of degree zero. In view of [@Kat16 Proof of Theorem 3.3], we deduce that the multiplication map $W(\lambda)^{\ast} \otimes W (\mu)^{\ast} \rightarrow
W (\lambda + \mu)^{\ast}$ is surjective since is injective. Therefore, $R_{G}$ is a normal ring generated by terms of primitive degree (see, e.g., [@Har77 Chap. II, Exerc. 5.14]; cf. [@Kat16 Proof of Theorem 3.3]).
From the above, it suffices to prove $R_{G}' = R_{G}$. For this purpose, it is enough to prove that the associated graded ring of the projective coordinate ring $R''_{G}$ of ${\mathbf{Q}_{G}}$, which is arising from its structure of a closed subscheme of $\prod_{i \in I} {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]})$, contains $R_{G}$ (see, e.g., [@EGA1 Sect. 2.6] for convention). Recall that the projective coordinate ring of ${\mathbb{P}}( L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]} )$ is $\bigoplus_{n \ge 0} S^{n} \bigl( L({\varpi}_{i})[z]^{\ast}\bigr)$, where $S^{n}V$ denotes the $n$-th symmetric power of a vector space $V$. Thanks to the surjectivity of multiplication map of $R_G$, it is further reduced to seeing that for each $i \in I$, every element of the part $W({\varpi}_{i})^{\ast}$ of degree ${\varpi}_{i}$ in $R_{G}$ is written as the quotient of an element of $\prod_{i \in I} S^{ {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} } \bigl( L ( {\varpi}_{i} ) [z] ^{\ast} \bigr)$ by some power of monomials in elements of $L ( {\varpi}_{j} )[z]^{\ast} \subset W ( {\varpi}_{j} )^{\ast}$, $j \in I$ (note that this condition is particularly apparent in types $A$ and $C$ since $W ( {\varpi}_{i} ) = L ( {\varpi}_{i} ) [z]$ for each $i \in I$).
By [@BF14c Proof of Theorem 3.1], each ${\mathfrak{Q}}_{G}(\xi)$, $\xi \in Q^{\vee, +}$, is a projective variety with rational singularities. By the Serre vanishing theorem [@EGA3-1 Théorème 2.2.1] applied to the ideal sheaf that defines ${\mathfrak{Q}}_{G}(\xi)$ (inside the product of finite-dimensional projective spaces in ${\mathbb{P}}( L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]} )$ obtained by bounding the degree; cf. [@Kat16 (2.1)]), the restriction map $$\label{surj-rest}
\bigotimes_{i \in I} H^{0} \bigl( {\mathbb{P}}( L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]}) ,
{\mathcal{O}}({\langle \lambda,\,\alpha_{i}^{\vee} \rangle}) \bigr) \rightarrow
H^{0} ( {\mathfrak{Q}}_{G}(\xi), {\mathcal{O}}_{{\mathfrak{Q}}_{G}(\xi)} (\lambda) )$$ is surjective for sufficiently large $\lambda \in P^{+}$, where we have used the fact that $$\bigotimes _{i \in I}
H^{0}( {\mathbb{P}}( L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]} ), \mathcal{F}_{i}) \cong
H^{0}\left( \prod_{i \in I} {\mathbb{P}}( L ({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]} ), \boxtimes_{i \in I} \mathcal{F}_{i} \right)$$ holds for vector bundles $\mathcal{F}_{i}$ of finite rank on ${\mathbb{P}}( L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]} )$.
\[H-inj\] For a given degree $n \in {\mathbb{Z}}_{\le 0}$, we can choose $\lambda \in P^{+}$ and $\xi \in Q^{\vee,+}$ sufficiently large in such a way that for every $m \in {\mathbb{Z}}_{>0}$, the restriction map $$W ( m \lambda )^{\ast} \subset W ( m \lambda )^{\vee} =
H^{0} ( {\mathfrak{Q}}_{G}, {\mathcal{O}}_{{\mathfrak{Q}}_{G}} (m \lambda ) ) \longrightarrow
H^{0} ( {\mathfrak{Q}}_{G} ( \xi ), {\mathcal{O}}_{{\mathfrak{Q}}_{G} (\xi)} ( m \lambda ) )$$ is injective at degree greater than or equal to $n$.
[*Proof of Claim \[H-inj\].*]{} Let us denote by $W ({\varpi}_{i} )^{\ast}_{\ge n} \subset W({\varpi}_{i})^{\ast}$ and $W ({\varpi}_{i})_{\le -n} \subset W ({\varpi}_{i})$ the direct sum of the homogeneous components of $W({\varpi}_{i})^{\ast}$ of degree greater or equal to $n$, and the the direct sum of the homogeneous components of $W ( {\varpi}_{i} )$ of degree less than or equal to $-n$, respectively. Also, let $R_{G}^{n}$ be the subring of $R_{G}$ generated by the $W ({\varpi}_{i} )^{\ast}_{\ge n}$, $i \in I$; every homogeneous component of $R_{G}$ of degree greater than or equal to $n$ is contained in $R_{G}^{n}$ by the surjectivity of multiplication map and the fact that $W(\lambda)^{\ast}$ is concentrated in nonpositive degrees. The value of a section of a line bundle over ${\mathop{\rm Proj}\nolimits}R_{G}$ (our ${\mathop{\rm Proj}\nolimits}$ here is the $P^{+}$-graded proj, by which we mean that the $H$-quotient of the subset of the affine spectrum in $\prod_{i \in I} \bigl( W({\varpi}_{i} ) \setminus \{0\} \bigr)$) arising from $R_{G}^{n}$ at a point is determined completely by its image under the projection $${\mathop{\rm pr}\nolimits}:
\prod_{i \in I} {\mathbb{P}}\left( {\widehat{W({\varpi}_{i})}} \right) \setminus Z
\longrightarrow
\prod_{i \in I} {\mathbb{P}}\bigl( W ({\varpi}_{i} )_{\le - n} \bigr)$$ induced by the ${\mathfrak{g}}[z]$-module surjection $W ({\varpi}_{i}) \to W ({\varpi}_{i})_{\le - n}$, where $Z$ denotes the loci in which ${\mathop{\rm pr}\nolimits}$ is not well-defined; for the notation ${\widehat{W({\varpi}_{i})}}$, see Section \[subsec:liealg\]. Thanks to Theorem \[w-approx\], we deduce that $${\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G} \setminus Z ) = {\mathop{\rm pr}\nolimits}( {\mathbf{Q}_{G}}\setminus Z )$$ as the set of closed points. Since the restriction of ${\mathop{\rm pr}\nolimits}$ to ${\mathfrak{Q}}_{G}(\xi) \setminus Z$ for each $\xi \in Q^{\vee,+}$ is a morphism of Noetherian schemes, it follows that ${\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G}(\xi) \setminus Z )$ is a constructible subset of $\prod_{i \in I} {\mathbb{P}}\bigl( W ({\varpi}_{i} )_{\le -n} \bigr)$. Moreover, the irreducibility of ${\mathfrak{Q}}_{G}( \xi )$ forces $\overline{{\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G} ( \xi ) \setminus Z )}$ to be irreducible. Therefore, the equality $${\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G} \setminus Z ) = \bigcup_{\xi \in Q^{\vee,+}} {\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G}(\xi) \setminus Z )$$ implies that there exists some $\xi \in Q^{\vee,+}$ such that $$\label{Z-dense}
{\mathop{\rm pr}\nolimits}( {\mathfrak{Q}}_{G}(\xi) \setminus Z ) \subset {\mathop{\rm pr}\nolimits}( {\mathbf{Q}_{G}}\setminus Z )$$ is Zariski dense, since $\prod_{i \in I} {\mathbb{P}}\left( W ({\varpi}_{i} )_{\le - n} \right)$ is a Noetherian scheme.
Thanks to [@BF14c Proposition 5.1] (cf. Theorem \[Kmain\]), we can find $\lambda$ (by replacing $\xi$ with a larger one if necessary) such that the assertion holds for $m = 1$. Now we assume the contrary to deduce the assertion for $m > 1$. Then, we have an additional equation on ${\mathop{\rm pr}\nolimits}( {\mathbf{Q}_{G}}\setminus Z ) \subset
\prod_{i \in I} {\mathbb{P}}\bigl( W ({\varpi}_{i} )_{\le - n} \bigr)$ vanishing along ${\mathfrak{Q}}_{G}(\xi)$ by (taking sum of) the multiplication of $W(\lambda)^{\ast}_{\ge n}$. However, this is impossible in view of since ${\mathop{\rm Proj}\nolimits}R_{G}$ is reduced (and hence $R_{G}^{n}$ is integral). Thus we have proved Claim \[H-inj\].
We return to the proof of Theorem \[normal\]. We fix $n \in {\mathbb{Z}}_{\le 0}$ and $\beta \in Q^{\vee,+}$ such that Claim \[H-inj\] holds. By replacing $\lambda$ if necessary to guarantee the surjectivity of the restriction map with keeping the situation of Claim \[H-inj\], we deduce that all the maps $$\label{diag}
\begin{split}
\xymatrix{
\bigotimes _{i \in I} H^{0} \bigl( {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]}), {\mathcal{O}}({\langle \lambda,\,\alpha_{i}^{\vee} \rangle}) \bigr)
\ar@{=}[d] \ar[r]
& H^{0}({\mathbf{Q}_{G}}, {\mathcal{O}}_{{\mathbf{Q}_{G}}}(\lambda))
\ar@{^{(}->}[d] \\
\bigotimes _{i \in I} S^{ {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} }
\bigl( L({\varpi}_{i})[z]^{\ast} \bigr)
\ar[r] \ar@{=}[d]
& W (\lambda)^{\ast}
\ar[d] \\
\bigotimes_{i \in I} H^{0} \bigl( {\mathbb{P}}(L({\varpi}_{i}){[\hspace{-1pt}[z]\hspace{-1pt}]}), {\mathcal{O}}( {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}) \bigr)
\ar[r] &
H^{0} ( {\mathfrak{Q}}_{G}(\xi), {\mathcal{O}}_{{\mathfrak{Q}}_{G}(\xi)} ( \lambda ) )
}
\end{split}
$$ are surjective at degree $n$ from the commutativity of the diagram and the surjectivity of the bottom horizontal map. For a degree $n$ element $f \in W ({\varpi}_{i})^{\ast}$ and degree zero element $h_{j} \in L({\varpi}_{j})^{\ast} \subset W({\varpi}_{j})^{\ast}$ for each $j \in I$, we can choose sufficiently large integers $N_{i}$, $i \in I$, such that $$f \cdot \prod_{j \in I} h_{j}^{N_{j}} \in
S^{1+N_{i}} \bigl( L({\varpi}_{i}) [z]^{\ast} \bigr) \cdot
\prod_{j \in I,\,j \ne i} S^{N_{j}} \bigl( L ( {\varpi}_{j} ) [z]^{\ast} \bigr) \subset R_{G}$$ as the corresponding claim is true after sending to the bottom line of . This forces $W ( {\varpi}_{i})^{\ast}$ to be contained in the part of degree ${\varpi}_{i}$ of the associated graded ring of $R_{G}''$, as required. This completes the proof of the theorem.
\[pcr\] The projective coordinate ring $R_{G}$ of ${\mathbf{Q}_{G}}$ arising from the embedding by means of the DP-coordinates is isomorphic to $\bigoplus_{\lambda \in P^{+}} W(\lambda)^{\ast}$.
\[R-free\] The projective coordinate ring $R_{G}$ of ${\mathbf{Q}_{G}}$ in Corollary \[pcr\] is free over the polynomial algebra $A_{G}$ given by the lowest weight components with respect to the $H$-action.
During this proof, we denote by $v_{\lambda} \in W ( \lambda )$ the unique ${\mathfrak{h}}_{{\mathrm{af}}}$-eigenvector of weight $\lambda$ (which is determined up to a scalar, and is the specialization of the corresponding vector of $V ( \lambda )$ through $\mathsf{q} \to 1$).
For each $\lambda \in P^{+}$, we set $${\mathbb{C}}[{\mathbb{A}}^{(\lambda)}] := \bigotimes_{i \in I}
S^{ {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} } {\mathbb{C}}[z] \cong
\bigotimes_{i \in I}
{\mathbb{C}}[ z_{1}^{(i)},\,\ldots,\,
z_{ {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} }^{(i)} ]^{ {\mathfrak{S}}_{{\langle \lambda,\,\alpha_{i}^{\vee} \rangle}} }.$$ By the results [@FL07; @Naoi12], due to Fourier-Littelmann and Naoi, we know that $W (\lambda)$ is a free module over ${\mathbb{C}}[{\mathbb{A}}^{(\lambda)}]$, and the $\lambda$-isotypical component of $W ( \lambda )$ is free of rank one. Here we define the polynomial algebra $A_{G}$ by collecting the $(-\lambda)$-isotypical component of $W ( \lambda )^{\ast}$ for all $\lambda \in P^{+}$. It follows that the ring $A_{G}$ is of the form $$\bigotimes_{i \in I} S^{\bullet} {\mathbb{C}}[z]^{\ast} \cong
\bigoplus_{\lambda \in P^{+}}
\bigotimes_{i \in I} S^{ {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} } {\mathbb{C}}[z]^{\ast} \cong
\bigoplus_{\lambda \in P^{+}} {\mathbb{C}}[{\mathbb{A}}^{(\lambda)}]^{\ast}.$$
Let $\psi \in W ( \mu )^{\ast}$ and $\xi \in {\mathbb{C}}[{\mathbb{A}}^{(\lambda - \mu)}]^{\ast} \subset A_G$, where $\lambda, \mu, \lambda - \mu \in P^+$, and assume that both of them are homogeneous with respect to $P$-weights and degrees. Then, we find a product of ${\mathfrak{h}}_{{\mathrm{af}}}$-eigen PBW basis element $F_1 \in U ( {\mathfrak{n}}^- [z] )$ and monomials $f_1, f_2 \in U ( {\mathfrak{h}}[z] z ) \cong
S^{\bullet} ( {\mathfrak{h}}[z] z )$ such that $\psi ( F _1 f_1 v_{\mu} ) \neq 0$ and $\xi ( f_2 v_{(\lambda - \mu)}) \neq 0$ by the Poincaré-Birkhoff-Witt theorem. It follows that $( \psi \cdot \xi) ( F_1 f_1 f_2 v_{\lambda} ) \neq 0$, since we need to collect the terms $F_1 m_1 v_{\mu} \otimes m_2 v_{\lambda - \mu}$, with $m_1 m_2 = f_1 f_2$, in the tensor product through the embedding $W ( \lambda ) \subset W ( \mu ) \otimes W ( \lambda - \mu )$. This means that $0 \neq \psi \cdot \xi \in W ( \lambda )^{\ast}$; in particular, the ring $R_G$ is torsion-free as an $A_G$-module.
Now, let us fix $i_0 \in I$ and $\lambda \in P^+$ so that ${\langle \lambda,\,\alpha_{i_0}^{\vee} \rangle} = 0$, and set $\lambda_m := \lambda + m {\varpi}_{i_0}$ for each $m \in {\mathbb{Z}}_{\ge 0}$. We also set $A_G^{i_0} := S^{\bullet} {\mathbb{C}}[z]^{\ast} \subset A_G$ for the fixed $i_0$. For each $m, l \in {\mathbb{Z}}_{\ge 0}$ such that $m \ge l$, we denote by $W ( \lambda; m,l )$ the space $({\mathbb{C}}[{\mathbb{A}}^{(l{\varpi}_{i_0})}]^{\ast} \cdot W ( \lambda_{m-l} )^{\ast} )^{\ast}$, which is a ${\mathfrak{b}}[z]$-submodule of $W ( \lambda_m )$. From this description, we have an inclusion $$W ( \lambda; m,l ) \subset W ( \lambda; m,l-1 )$$ when $l > 0$. In particular, $W ( \lambda; m, l )^{\ast}$ is a quotient of $W ( \lambda; m, l - 1 )^{\ast}$. By repeated use of (the dual of) the surjectivity of the multiplication map of $R_G$, we have an embedding: $$\label{W-emb-fw}
\Phi : W ( \lambda_m ) \hookrightarrow W ( \varpi_{i_0} )^{\otimes m} \otimes W ( \lambda ).$$ For $0 \le l \le m$, let $\mathbb W ( \lambda; m, l )$ denote the linear span of pure tensors $$\label{pure-tensor}
\left( \bigotimes_{j = 1}^{m} v_{i_0,j} \right) \otimes v \in
W ( \varpi_{i_0} )^{\otimes m} \otimes W ( \lambda )$$ of ${\mathfrak{h}}_{{\mathrm{af}}}$-eigenvectors in which at most $l$-elements of $\{v_{i_0,j}\}_{j = 1}^{m}$ is of the form $z^e v_{{\varpi}_{i_0}}$ for some $e \in {\mathbb{Z}}_{\ge 0}$. If we denote by $W ( \lambda; m, l )'$ the preimage of ${\mathbb{W}}( \lambda; m, l )$ through $\Phi$, then we have $$W ( \lambda; m, l-1 )' \subset W ( \lambda; m, l )'$$ whenever $l > 0$. By construction, $\{ W ( \lambda; m, l )' \}_{0 \le l \le m}$ forms a ${\mathbb{Z}}$-graded increasing filtration whose associated graded modules $${\mathop{\rm gr}\nolimits}_{l} W( \lambda_m ) :=
W ( \lambda; m, l )' / W ( \lambda; m, l - 1 )', \qquad l \in {\mathbb{Z}}_{\ge 0},$$ stratify $W ( \lambda_m )$.
Here, $W ( \varpi_{i_0} )^{\otimes m} \otimes W ( \lambda )$ admits a graded decomposition coming from the number of elements in $\{v_{i_0,j}\}_{j = 1}^{m}$ of the form $z^e v_{\varpi_{i_0}}$ for $e \in {\mathbb{Z}}_{\ge 0}$ through and . It follows that the space ${\mathbb{W}}( \lambda; m, l )$ is the annihilator of the subspace $$\label{expand-prod}
\sum_{w \in {\mathfrak{S}}_m} \sum_{a=l}^m w ( {\mathbb{C}}[{\mathbb{A}}^{(\varpi_{i_0})}]^{\otimes a} \otimes
W ( \varpi_{i_0} )^{\otimes m-a} )^{\ast} \otimes
W ( \lambda )^\ast \subset ( W ( \varpi_{i_0} ) ^{\otimes m} )^\ast \otimes
W ( \lambda )^{\ast},$$ where ${\mathfrak{S}}_m$ permutes the tensor factors of $W ( \varpi_{i_0} )^{\otimes m}$. Pulling back by $\Phi$, we deduce that $W ( \lambda; m, l-1 )'$ is the annihilator of the space in $W ( \lambda_m )$ through the embedding . Therefore, $W ( \lambda; m, l-1 )' \subset W ( \lambda_m )$ is exactly the annihilator of ${\mathbb{C}}[{\mathbb{A}}^{(l{\varpi}_{i_0})}]^{\ast} \cdot
W ( \lambda_{m-l} )^{\ast} \subset W ( \lambda _m )^{*}$. It follows that we have a canonical isomorphism $${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m) \cong W ( \lambda; m, l ) / W ( \lambda; m, l+1).$$ If we define a subquotient $M(\lambda; n)$ of $R_{G}$ for each $n \in {\mathbb{Z}}_{\ge 0}$ by $$M( \lambda; n ) := \bigoplus _{l \ge 0} M ( \lambda; n )_l, \qquad
M( \lambda; n )_l = \left( {\mathop{\rm gr}\nolimits}_{l} W(\lambda_{n+l}) \right)^{\ast},$$ then $M(\lambda; n)$ admits an $A_G^{i_0}$-action.
From the construction of $M ( \lambda; n )_l$ through $\{W ( \lambda; m, l)\}_{m, l \ge 0}$, we deduce that $M ( \lambda; n )$ is generated by $M ( \lambda; n )_{0}$ as an $A_G^{i_0}$-module. Also, from the construction of $M ( \lambda; n )_l$ through $\{W ( \lambda; m, l)'\}_{m, l \ge 0}$, we deduce that the dual of the multiplication map is the natural map $${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0})}] \otimes {\mathop{\rm gr}\nolimits}_{0} W(\lambda_{n}) \rightarrow {\mathop{\rm gr}\nolimits}_{0} \, W(\lambda_{n+l})$$ of ${\mathbb{C}}[{\mathbb{A}}^{(\lambda_{n+l})}]$-modules. The $({\mathbb{C}}[{\mathbb{A}}^{(m {\varpi}_{i_{0}})}] , {\mathfrak{b}}[z] )$-modules ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$, $0 \le l \le m$, stratify $W ( \lambda_{m} )$. In addition, the maximality of $W ( \lambda; m, l )'$ guarantees that each ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$ is torsion-free as a ${\mathbb{C}}[{\mathbb{A}}^{(\lambda_m)}]$-module.
For each $\lambda \in P^{+}$, we can regard $W ( \lambda )$ as a module corresponding to a vector bundle (or a free sheaf) ${\mathcal{W}}(\lambda)$ over ${\mathbb{A}}^{(\lambda)}$, where ${\mathbb{A}}^{(\lambda)}$ denotes ${\mathop{\rm Spec}\nolimits}{\mathbb{C}}[{\mathbb{A}}^{(\lambda)}]$; its fiber is known to be the tensor product of local Weyl modules $W (\mu,\,x)$, where $\mu \in P^{+}$ and $x \in {\mathbb{C}}$ runs over the configurations of points determined by a point of ${\mathbb{A}}^{(\lambda)}$ (see, e.g., [@Kat16 Theorem 1.4]).
The spaces ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$, $0 \le l \le m$, give torsion-free sheaves ${\mathcal{W}}_l(\lambda_m)$ on ${\mathbb{A}}^{(\lambda_{m})}$ that stratify ${\mathcal{W}}( \lambda_m )$. Hence a section of ${\mathcal{W}}_l (\lambda_m)$ is an equivalence class of the set of sections ${\mathbb{A}}^{(\lambda_m)} \rightarrow {\mathcal{W}}( \lambda_m )$ whose specialization to a general point gives an element of the tensor product of local Weyl modules $$\label{point-division}
v = \sum_{k} \bigotimes_{i,j} v_{i,j}^{(k)} \in
\bigotimes_{i \in I} \bigotimes_{j=1}^{{\langle \lambda,\,\alpha_{i}^{\vee} \rangle}}
W ( {\varpi}_{i}, x_{i,j} )$$ such that exactly $l$-elements in $\{v_{i_{0},j}^{(k)} \}_{j=1}^{m}$ are highest weight vectors, and the other vectors do not have a highest weight component for each $k$.
Since every two points in $\{x_{i,j}\}_{i,j}$ are generically distinct, each pure tensor in divides $\{1,2,\ldots, m\}$ into two subsets $S_{1}$ and $S_{2}$, with $\# S_{1} = l$, so that $\{x_{i_{0},j}\}_{j\in S_{1}}$ carries a highest weight vector (i.e., $v_{i_{0},j}^{(k)} = z^{e}v_{{\varpi}_{i_{0}}}$ for some $e \in {\mathbb{Z}}_{\ge 0}$) and $\{x_{i_{0},j}\}_{j\in S_{2}}$ carries a vector lying in non-highest-weight components (as a ${\mathfrak{b}}$-module). The coordinates $\{x_{i_{0},j}\}_{j\in S_{1}}$ gives rise to the action of ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}]$ on ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$, while the coordinates $\{x_{i_{0},j}\}_{j\in S_{2}}$ gives rise to a ${\mathbb{C}}[{\mathbb{A}}^{(\lambda_{m-l})}]$-module structure on ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$, and these two module structures are (mutually commutative and) distinct. It follows that every pair of elements of ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}]$ and ${\mathop{\rm gr}\nolimits}_{0} W(\lambda_{m-l})$ appears as a section in ${\mathcal{W}}_l ( \lambda_{m} )$ after a generic localization. This particularly gives us the ${\mathfrak{S}}_{m}$-action on ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}] \boxtimes {\mathbb{C}}[{\mathbb{A}}^{(\lambda_{m-l})}]$ and ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$ that changes the order of the highest weight vectors and the one of non-highest weight vectors (or mixes up $S_{1}$ and $S_{2}$). Therefore, ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$ itself is torsion-free as a ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}] \boxtimes {\mathbb{C}}[{\mathbb{A}}^{(\lambda_{m-l})}]$-module.
Because $M ( \lambda; n )$ is generated by $M ( \lambda; n )_{0}$ as an $A_{G}^{i_{0}}$-module, we have an injective map $$\eta : \mathrm{gr}_{l} \, W(\lambda_{m}) \hookrightarrow
{\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}] \otimes \mathrm{gr}_{0} \, W(\lambda_{m-l})$$ of ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0} )}] \boxtimes {\mathbb{C}}[{\mathbb{A}}^{(\lambda_{m-l})}]$-modules, which is an isomorphism after a localization to some Zariski open subset of ${\mathbb{A}}^{(l {\varpi}_{i_0} )} \times {\mathbb{A}}^{(\lambda_{m-l})}$. Since ${\mathop{\rm gr}\nolimits}_{l} W(\lambda_m)$, $0 \le l \le m$, stratifies $W ( \lambda_m )$, we deduce that the $M( \lambda; n )$’s, with $\lambda$ varying, give a stratification of the $A_{G}$-module $R_{G}$. Hence, in order to prove that $R_G$ is free over $A_{G}^{i_0}$, it suffices to verify that each $M ( \lambda; n )$ is a free $A_{G}^{i_0}$-module. By construction, the image of $\eta$ contains ${\mathbb{C}}\otimes {\mathop{\rm gr}\nolimits}_{0} W(\lambda_{m-l})$, which gives a ${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0})}]$-module generator of ${\mathop{\rm gr}\nolimits}_{l} \, W(\lambda_{m})$. Therefore, the map $\eta$ must be an isomorphism. As a consequence, we conclude that $${\mathbb{C}}[{\mathbb{A}}^{(l {\varpi}_{i_0})}]^{\ast} \otimes M ( \lambda; n )_0 \cong M ( \lambda; n )_l$$ through the multiplication map. In other words, $M ( \lambda; n )$ is a free $A_{G}^{i_0}$-module.
Because the above argument is consistent with the filtrations and their associated graded modules arising from a different choice of $i_0 \in I$, we can vary $i_0 \in I$ and construct the associated graded modules inductively on a fixed total order on $I$. This gives a stratification of $R_{G}$ that is free over $A_{G}$. Hence the ring $R_G$ itself is free over $A_G$. This completes the proof of the theorem.
\[coh-e\] For each $\lambda \in P$, we have $${\mathop{\rm gch}\nolimits}H^{n} ( {\mathbf{Q}_{G}}, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) ) =
\begin{cases}
{\mathop{\rm gch}\nolimits}V^{-}_{e}( - w_{\circ} \lambda ) &
\text{\rm if $\lambda \in P^{+}$ and $n = 0$}, \\[1.5mm]
0 &
\text{\rm otherwise}.
\end{cases}$$
We know that ${\mathbf{Q}_{G}}$ is a closed subscheme of $\prod_{i \in I} {\mathbb{P}}( L ( {\varpi}_{i} ){[\hspace{-1pt}[z]\hspace{-1pt}]} )$ by . Therefore, we have a countable set ${\Omega}$ of $I$-tuples of $( H \times {\mathbb{G}_{m}})$-eigenvectors of $\bigsqcup_{i \in I} L( {\varpi}_{i} ){[\hspace{-1pt}[z]\hspace{-1pt}]}$, one for each $i \in I$, so that it induces an affine open cover ${\mathcal{U}}:= \{ {\mathcal{U}}_{S} \}_{S \subset {\Omega}}$ of ${\mathbf{Q}_{G}}$ (where ${\mathcal{U}}_{S} := \bigl\{ f \neq 0 \mid f \in S \bigr\}$) that is closed under intersection. Now, the maps $L({\varpi}_i){[\hspace{-1pt}[z]\hspace{-1pt}]} \setminus \{ 0 \} \rightarrow
{\mathbb{P}}(L({\varpi}_i){[\hspace{-1pt}[z]\hspace{-1pt}]} )$, $i \in I$, induce a (right) $H$-fibration ${\widetilde{\mathbf{Q}}_{G}}$ that defines an open scheme of $\prod_{i \in I} L( {\varpi}_i ){[\hspace{-1pt}[z]\hspace{-1pt}]}$, which corresponds to specifying a nonzero vector $u^{{\varpi}_{i}}$ instead of a one-dimensional ${\mathbb{C}}$-vector subspace $L^{{\varpi}_{i}} \ni u^{{\varpi}_{i}}$ in the definition of DP data; its closure ${\widehat{\mathbf{Q}}_{G}}$, which corresponds to allowing $u^{{\varpi}_{i}} = 0$ in a DP datum, is an affine subscheme of $\prod_{i \in I} L( {\varpi}_i ){[\hspace{-1pt}[z]\hspace{-1pt}]}$ of infinite type. We set $Z := {\widehat{\mathbf{Q}}_{G}}\setminus {\widetilde{\mathbf{Q}}_{G}}$, which is a closed subscheme of ${\widehat{\mathbf{Q}}_{G}}$. Also, the pullback ${\widetilde{{\mathcal{U}}}}_S$ of ${\mathcal{U}}_S$ to ${\widetilde{\mathbf{Q}}_{G}}$ defines an affine open subset of ${\widehat{\mathbf{Q}}_{G}}$. By the finiteness of the defining functions, ${\widetilde{{\mathcal{U}}}}_S \hookrightarrow {\widehat{\mathbf{Q}}_{G}}$ is quasi-compact by [@EGA1 Proposition 1.1.10]. For each finite subset $S \subset {\Omega}$, we set ${\widetilde{{\mathcal{U}}}}^S := \{ {\widetilde{{\mathcal{U}}}}_{T} \}_{T \subset S}$, which is again a collection of affine subschemes that is closed under intersections, and ${\widetilde{U}}^S := \bigcup_{T \subset S}{\widetilde{{\mathcal{U}}}}_T$. In addition, we set ${\widetilde{Z}}_S := {\widehat{\mathbf{Q}}_{G}}\setminus {\widetilde{U}}^S$.
Let us denote the natural projection ${\widetilde{\mathbf{Q}}_{G}}\rightarrow {\mathbf{Q}_{G}}$ by $\pi$. Since ${\mathbf{Q}_{G}}$ is a (right) free quotient of ${\widetilde{\mathbf{Q}}_{G}}$ by $H$, we deduce that ${\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) = ( \pi_{*} {\mathcal{O}}_{{\widetilde{\mathbf{Q}}_{G}}} )^{(H, \lambda)}$, where $\bullet^{(H, \lambda)}$ denotes the $\lambda$-isotypical component with respect to the right $H$-action. Because discarding open sets (in such a way that the remaining ones are closed under intersection) in the Cech complex defines a projective system of complexes satisfying the Mittag-Leffler condition, [@EGA3-1 Proposition 13.2.3] yields an isomorphism $$\label{cech-ref}
H^{n} ( {\widetilde{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widetilde{\mathbf{Q}}_{G}}} ) \cong
\varprojlim_{S} H^{n} ( {\widetilde{U}}^S, {\mathcal{O}}_{{\widetilde{\mathbf{Q}}_{G}}} )
\quad \text{for each $n \in {\mathbb{Z}}_{\ge 0}$}.$$ We have $$\label{des-H}
H^{n} ( {\mathbf{Q}_{G}}, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) ) \cong
H^{n} ( {\widetilde{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widetilde{\mathbf{Q}}_{G}}} )^{(H, \lambda)}
\quad \text{for $n \in {\mathbb{Z}}_{\ge 0}$},$$ since the right $H$-action on ${\widetilde{\mathbf{Q}}_{G}}$ is free, and it induces a semi-simple action on the level of Cech complex.
The long exact sequence of local cohomologies (see [@SGA2 Exposé I, Corollaire 2.9]) yields: $$\label{l-les}
\cdots \rightarrow H^{n}_{{\widetilde{Z}}_S} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} )
\rightarrow H^{n} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} )
\rightarrow H^{n} ( {\widetilde{U}}^S, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) \rightarrow
H^{n+1}_{{\widetilde{Z}}_S} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) \rightarrow \cdots.
$$ Since ${\widehat{\mathbf{Q}}_{G}}$ is affine, this induces $$\label{isom-local}
H^{n} ( {\widetilde{U}}^S, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} )
\stackrel{\cong}{\longrightarrow}
H^{n+1}_{{\widetilde{Z}}_S} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} )
\quad \text{for $n \ge 1$}$$ by [@EGA3-1 Théorème 1.3.1]. Here the quasi-compactness of the embedding ${\widetilde{U}}^S \hookrightarrow {\widehat{\mathbf{Q}}_{G}}$, together with [@SGA2 Exposé II, Proposition 5], implies that $$\label{Koszul}
H^{n}_{{\widetilde{Z}}_S} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) \cong
H^{n} ( K_S ( {\mathbb{C}}[ {\widehat{\mathbf{Q}}_{G}}] ) ),$$ where $K_S ( {\mathbb{C}}[ {\widehat{\mathbf{Q}}_{G}}] )$ denotes the (cohomological) ${\mathbb{C}}[ {\widehat{\mathbf{Q}}_{G}}]$-Koszul complex defined through $S \subset {\Omega}$ (see [@EGA3-1 (1.1.2)]).
In view of , the comparison of and via yields an isomorphism $$\label{genuine}
H_{Z}^{n} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) \cong
\varprojlim_S H^{n} ( K_S ( {\mathbb{C}}[ {\widehat{\mathbf{Q}}_{G}}] ) ) \cong
\varprojlim_S H^{n}_{{\widetilde{Z}}_S} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ).$$ We know from [@EGA3-1 Proposition 1.1.4] that the Koszul complex $K_S ( {\mathbb{C}}[ {\widehat{\mathbf{Q}}_{G}}] )$ has trivial cohomology at degree $< n$ if $S$ contains a regular sequence of length $n$. Here we see from Corollary \[pcr\] that ${\mathbb{C}}[{\widehat{\mathbf{Q}}_{G}}] =
\bigoplus_{\lambda \in P^{+}} W(\lambda)^{\ast}$. Also, by Theorem \[R-free\], we can rearrange ${\Omega}$ if necessary in such a way that for each $i \in I$, the set of the $i$-th components of the elements in ${\Omega}$ contain a regular sequence of arbitrary length. Then, we deduce from that $$H_{Z}^{n} ( {\widehat{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) = \{ 0 \} \quad
\text{for all $n \in {\mathbb{Z}}_{\ge 0}$}.$$ Therefore, and the affinity of ${\widehat{\mathbf{Q}}_{G}}$ imply that $$H^{n} ( {\widetilde{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widehat{\mathbf{Q}}_{G}}} ) =
H^{n} ( {\widetilde{\mathbf{Q}}_{G}}, {\mathcal{O}}_{{\widetilde{\mathbf{Q}}_{G}}} ) = \{ 0 \} \quad
\text{for all $n > 0$}.$$ In view of and Theorem \[BFmain\], we conclude the assertion of the theorem.
\[coh\] For each $x \in W_{{\mathrm{af}}}^{\ge 0}$, the scheme ${\mathbf{Q}_{G}}(x)$ is normal. Moreover, for each $\lambda \in P$ and $x \in W_{{\mathrm{af}}}^{\ge 0}$, we have $${\mathop{\rm gch}\nolimits}H^{n} ( {\mathbf{Q}_{G}}(x), {\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}(\lambda) ) =
\begin{cases}
{\mathop{\rm gch}\nolimits}V^{-}_{x} ( - w_{\circ} \lambda ) &
\text{\rm if $\lambda \in P^{+}$ and $n = 0$}, \\[1.5mm]
0 & \text{\rm otherwise}.
\end{cases}.$$ In particular, we have $${\mathop{\rm gch}\nolimits}H^{0}( {\mathbf{Q}_{G}}(x), {\mathcal{O}}_{{\mathbf{Q}_{G}}(x)}(\lambda) ) \in
({\mathbb{Z}}{(\hspace{-1pt}(q^{-1})\hspace{-1pt})})[P] \subsetneq ({\mathbb{Z}}[P]){(\hspace{-1pt}(q^{-1})\hspace{-1pt})}.$$
Once we know the normality of ${\mathbf{Q}_{G}}$ and the cohomology vanishing result in Theorem \[coh-e\], the same argument as in [@Kat16 Theorem 4.7] (see Theorem \[Kmain\]) yields all the assertions except for the last one.
The last assertion on the character estimate follows from a result about extremal weight modules ([@Kas02 Corollary 5.2]) and the fact that $U_{{\mathsf{q}}}^{-}({\mathfrak{g}})$ is concentrated on subspaces of $q$-degree $\le 0$.
\[w-push\] For an arbitrary $x \in W_{{\mathrm{af}}}^{\ge 0}$, we take ${\mathbf{i}}$ as in . Then we have ${\mathsf{m}}_{\ast} {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\mathbf{i}})} \cong {\mathcal{O}}_{{\mathbf{Q}_{G}}( x )}$.
We adopt the notation of Section \[subsec:BSDH\]. The map ${\mathsf{m}}$ factors as the composite of the map ${\mathsf{m}}$ for $t_{2 m_0 \rho^{\vee}}$ and a successive composite of the $q_{i_{k}, x ( k )}$ for $1 \le k \le \ell'$. The case $x = t_{\xi}$ for $\xi \in Q^{\vee,+}$ is clear since ${\mathbf{Q}_{G}}( t_{\xi} ) \cong {\mathbf{Q}_{G}}$ (through $\imath_{\xi}$). Therefore, it suffices to show that $( q_{i, x} )_{\ast} {\mathcal{O}}_{{\mathbf{I}}(i) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}( x )}
\cong {\mathcal{O}}_{{\mathbf{Q}_{G}}( s_{i} x )}$ for each $x \in W_{{\mathrm{af}}}^{\ge 0}$ and $i \in I_{{\mathrm{af}}}$ such that $s_{i} x \prec x$. This assertion itself follows from Corollary \[coh\] and [@Kat16 Theorem 4.7] (in view of Lemma \[vample\]). Hence we obtain the assertion of the corollary.
$K$-theory of semi-infinite flag manifolds. {#sec:K-SiFl}
===========================================
We keep the notation and setting of Section \[subsec:BSDH\].
\[line-bundles\] Every ${\widetilde{{\mathbf{I}}}}$-equivariant locally free sheaf of rank one [(]{}i.e., line bundle[)]{} on ${\mathbf{Q}_{G}}( x )$ is of the form $\chi \otimes _{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda )$ for some $\lambda \in P$ and an ${\widetilde{{\mathbf{I}}}}$-character $\chi$.
For $x = e$, the boundary of the open $G{[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbit ${\mathbf{O}}$ in ${\mathbf{Q}_{G}}$ is of codimension at least two, and the open $G {[\hspace{-1pt}[z]\hspace{-1pt}]}$-orbit ${\mathbf{O}}$ has a structure of pro-affine bundle over $G / B$. In particular, an ${\widetilde{{\mathbf{I}}}}$-equivariant line bundle over ${\mathbf{O}}$ is the pullback of a $B \times {\mathbb{G}_{m}}$-equivariant line bundle over $G / B$. Because every line bundle over $G/B$ carries a unique $G$-equivariant structure by [@KKV Sect. 3.3], and $B$-equivariant structures of the trivial line bundle ${\mathcal{O}}_{G/B}$ are in bijection with $P$ (since $H^{0}(G/B,\,{\mathcal{O}}_{G/B})={\mathbb{C}}$), we deduce that every $B$-equivariant line bundle over $G / B$ is an $H$-character twist of a $G$-equivariant line bundle, which is obtained as the restriction of some ${\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )$. Consequently, the assertion follows for $x = e$.
Now, for $y_{1},\,y_{2} \in W_{{\mathrm{af}}}^{\ge 0}$ such that $y_{1} \preceq y_{2}$, the restriction map transfers an ${\widetilde{{\mathbf{I}}}}$-equivariant line bundle over ${\mathbf{Q}_{G}}(y_{1})$ to an $\tilde{{\mathbf{I}}}$-equivariant line bundle over ${\mathbf{Q}_{G}}(y_{2})$. Also, for an arbitrary $x \in W_{{\mathrm{af}}}^{\ge 0}$, we can find $\xi \in Q^{\vee,+}$ such that $$\label{eq:DPs}
{\mathbf{Q}_{G}}( xt_{\xi} ) \subset {\mathbf{Q}_{G}}( t_{\xi} ) \subset {\mathbf{Q}_{G}}( x ) \subset {\mathbf{Q}_{G}}(e)={\mathbf{Q}_{G}}$$ by (the proof of) Lemma \[BSDH-cover\], since $x = x ( {\mathbf{j}}) \preceq t_{2m\rho^{\vee}}$ for $m \gg 0$. Because we have ${\mathbf{Q}_{G}}( t_{\xi} ) \cong {\mathbf{Q}_{G}}$ as schemes with an ${\widetilde{{\mathbf{I}}}}$-action, we conclude that a nonisomorphic pair of (${\widetilde{{\mathbf{I}}}}$-equivariant) line bundles over ${\mathbf{Q}_{G}}$ restricts to a nonisomorphic pair of (${\widetilde{{\mathbf{I}}}}$-equivariant) line bundles over ${\mathbf{Q}_{G}}(t_{\xi})$. Since ${\mathbf{Q}_{G}}( x t_{\xi} ) \cong {\mathbf{Q}_{G}}( x )$, the same is true for line bundles over ${\mathbf{Q}_{G}}( x )$. Therefore, by means of , we deduce the assertion of the proposition for an arbitrary $x \in W_{{\mathrm{af}}}^{\ge 0}$ from the case $x = e$. This proves the proposition.
The following is an immediate consequence of Corollary \[coh\] and Proposition \[line-bundles\].
\[cvs\] For each ${\widetilde{{\mathbf{I}}}}$-equivariant line bundle ${\mathcal{L}}$ over ${\mathbf{Q}_{G}}(x)$, we have $$H ^{n} ({\mathbf{Q}_{G}}(x),\,{\mathcal{L}}) = \{ 0 \} \quad \text{\rm for all $n > 0$}.$$
\[faith\] For each ${\widetilde{{\mathbf{I}}}}$-equivariant quasi-coherent sheaf ${\mathcal{E}}$ on ${\mathbf{Q}_{G}}$ such that $$\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{E}}( \lambda ) ) = \{ 0 \} \quad
\text{\rm for all $\lambda \in P$, where
${\mathcal{E}}( \lambda ) = {\mathcal{E}}\otimes_{{\mathcal{O}}_{{\mathbf{Q}_{G}}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}}(\lambda)$},$$ we have ${\mathcal{E}}= \{ 0 \}$.
By the quasi-coherence and ${\widetilde{{\mathbf{I}}}}$-equivariance, every nonzero section of ${\mathcal{E}}$ has an ${\widetilde{{\mathbf{I}}}}$-stable support, which must be a union of ${\mathbf{I}}$-orbits. In addition, it defines a regular section on a complement of finitely many hyperplanes having poles of finite order around the boundary points. Therefore, Lemma \[vample\] implies the desired result.
\[cvb\] Let ${\mathcal{E}}$ be an ${\widetilde{{\mathbf{I}}}}$-equivariant quasi-coherent ${\mathcal{O}}_{{\mathbf{Q}_{G}}}$-module satisfying the following conditions[:]{} $$\label{cvb1}
{\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}},\, {\mathcal{E}}( \lambda ) )^{\ast} \in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]}
\quad \text{\rm for every $\lambda \in P$};$$ $$\label{cvb2}
\begin{array}{l}
\text{\rm there exists $\lambda_{0} \in P$ such that
$\Gamma ( {\mathbf{Q}_{G}},\, {\mathcal{E}}( \lambda ) ) = \{ 0 \}$} \\[2mm]
\text{\rm for all $\lambda \in P$ with
${\langle \lambda,\,\alpha_{i}^{\vee} \rangle} < {\langle \lambda_{0},\,\alpha_{i}^{\vee} \rangle}$
for some $i \in I$}.
\end{array}
$$ Then, we have a resolution $\cdots \rightarrow {\mathcal{P}}^{2}_{{\mathcal{E}}}
\rightarrow {\mathcal{P}}^{1}_{{\mathcal{E}}}
\rightarrow {\mathcal{P}}^{0}_{{\mathcal{E}}}
\rightarrow {\mathcal{E}}\rightarrow 0$ of ${\widetilde{{\mathbf{I}}}}$-equivariant ${\mathcal{O}}_{{\mathbf{Q}_{G}}}$-modules such that
${\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}},\,{\mathcal{P}}^{k}_{{\mathcal{E}}}(\lambda) )^{\ast} \in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]}$ for every $k \ge 0$ and $\lambda \in P$[;]{}
for each $k \ge 0$, the ${\widetilde{{\mathbf{I}}}}$-equivariant ${\mathcal{O}}_{{\mathbf{Q}_{G}}}$-module ${\mathcal{P}}^{k}_{{\mathcal{E}}}$ is a direct sum of line bundles [(]{}if we forget the ${\mathbf{I}}$-module structure[);]{}
for each $m \in {\mathbb{Z}}$ and $\lambda \in P$, the number of direct summands ${\mathcal{P}}^{k}_{{\mathcal{E}}}(\lambda)$ of $\bigoplus_{k \ge 0} {\mathcal{P}}^{k}_{{\mathcal{E}}}(\lambda)$ contributing to the homogeneous subspace of degree $m$ of $\Gamma ( {\mathbf{Q}_{G}},\,\bigoplus _{k \ge 0} {\mathcal{P}}^{k}_{{\mathcal{E}}}(\lambda) )$ is finite.
Moreover, we have $H^{n} ( {\mathbf{Q}_{G}},\,{\mathcal{E}}) = \{ 0 \}$ for all $n > 0$.
Let $$R_{G} := \bigoplus_{\lambda \in P^{+}} W ( \lambda )^{\ast}
= \bigoplus_{\lambda \in P^{+}} H^{0} ( {\mathbf{Q}_{G}},\,{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) )$$ be the projective coordinate ring. Thanks to Lemma \[faith\], the sheaf ${\mathcal{E}}$ is determined by the $R_G$-module $$M ( {\mathcal{E}}) := \bigoplus_{\lambda \in P} H^{0} ( {\mathbf{Q}_{G}}, \, {\mathcal{E}}( \lambda ) ).$$ Because $M({\mathcal{E}})$ is nonpositively graded and each homogeneous subspace with respect to the $(P \times {\mathbb{Z}})$-grading is finite-dimensional, we obtain a surjection $P^{0}_{{\mathcal{E}}} \rightarrow M ({\mathcal{E}})$, where $P^{0}_{{\mathcal{E}}}$ is a direct sum of $(P \times {\mathbb{Z}})$-graded projective $R_G$-modules tensored with ${\widetilde{{\mathbf{I}}}}$-modules; indeed, we can construct the desired maps inductively by starting with $\lambda=\lambda_{0} \in P$, and then by adding the ${\varpi}_{i}$, $i \in I$, repeatedly, by means of the projectivity of $R_{G}$. Since a (graded) projective $R_G$-module is obtained from $R_{G}$ by a grading shift and an ${\widetilde{{\mathbf{I}}}}$-module twist, we deduce that $P^{0}_{{\mathcal{E}}} \cong M ( {\mathcal{P}}^{0}_{{\mathcal{E}}} )$ as ($P \times {\mathbb{Z}}$)-graded $R_G$-modules for a certain direct sum ${\mathcal{P}}^{0}_{{\mathcal{E}}}$ of ${\widetilde{{\mathbf{I}}}}$-equivariant line bundles (with some twist of the ${\widetilde{{\mathbf{I}}}}$-equivariant structure). Here the surjectivity of $P^{0}_{{\mathcal{E}}} \rightarrow M ( {\mathcal{E}})$ of $R_{G}$-modules implies that ${\mathcal{P}}^{0}_{{\mathcal{E}}} \rightarrow {\mathcal{E}}$ is also surjective. Also, by our character estimate, we deduce that ${\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}},\,{\mathcal{P}}^{0}_{{\mathcal{E}}} ( \lambda ) )^{\ast} \in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]}$. Now, let $\Xi ( {\mathcal{E}})$ be the set of those pairs $(\lambda,\,m) \in P \times {\mathbb{Z}}$ for which $$\bigoplus_{ \mu \in \lambda-P^{+} }
\left(\bigoplus _{n > m} \Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{E}}( \mu ) )_{n}\right) \oplus
\bigoplus _{\mu \in \lambda-P^{+}, \, \mu \ne \lambda}
\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{E}}( \mu ) )_{m}= \{0\}.$$ Then, we can rearrange ${\mathcal{P}}^{0}_{{\mathcal{E}}}$, if necessary, to assume that $$\label{eq:sp1}
{\mathop{\rm gdim}\nolimits}\ker \bigl(
\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{P}}^{0}_{{\mathcal{E}}} ( \lambda ) ) \rightarrow
\Gamma ( {\mathbf{Q}_{G}}, {\mathcal{E}}( \lambda ) )
\bigr)^{\ast} \in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]} \quad
\text{for all $\lambda \in P$};$$ $$\label{eq:sp2}
\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{P}}^{0}_{{\mathcal{E}}} ( \lambda ) )^{\ast} = \{ 0 \}
\quad \text{\rm for all $\lambda \in P$ with
${\langle \lambda,\,\alpha_{i}^{\vee} \rangle} < {\langle \lambda_{0},\,\alpha_{i}^{\vee} \rangle}$
for some $i \in I$};
$$ $$\label{eq:sp3}
\ker \bigl(
\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{P}}^0_{{\mathcal{E}}} ( \lambda ) ) \rightarrow
\Gamma ( {\mathbf{Q}_{G}}, {\mathcal{E}}( \lambda ) )
\bigr)_{m} = \{ 0 \}
\quad \text{for every $(\lambda, m) \in \Xi ( {\mathcal{E}})$}.$$ Thanks to and , we can replace ${\mathcal{E}}$ with $\ker \bigl( {\mathcal{P}}^{k}_{{\mathcal{E}}} \rightarrow {\mathcal{P}}^{k-1} _{{\mathcal{E}}} \bigr)$ repeatedly (with the convention ${\mathcal{P}}^{-1}_{{\mathcal{E}}} = {\mathcal{E}}$) to apply the procedure above in order to obtain ${\mathcal{P}}^{k+1} _{{\mathcal{E}}}$ for each $k \ge 0$. This yields an ${\widetilde{{\mathbf{I}}}}$-equivariant resolution $$\label{eq:res}
\cdots
\rightarrow {\mathcal{P}}^{2}_{{\mathcal{E}}}
\rightarrow {\mathcal{P}}^{1}_{{\mathcal{E}}}
\rightarrow {\mathcal{P}}^{0}_{{\mathcal{E}}}
\rightarrow {\mathcal{E}}\rightarrow 0,$$ in which each ${\mathcal{P}}^{k}_{{\mathcal{E}}}$ is a direct sum of ${\widetilde{{\mathbf{I}}}}$-equivariant line bundles (with some twist by ${\widetilde{{\mathbf{I}}}}$-modules). By the construction, we have $$\emptyset =
\bigcap_{k \ge 0} \Xi ( \ker d_{k} ) \subset P \times {\mathbb{Z}},$$ and hence the resolution satisfies the first two of the requirements. Also, taking into account and , we see that the resolution satisfies the third one of the requirements.
Finally, by applying Corollary \[cvs\], we conclude the desired cohomology vanishing. This completes the proof of the theorem.
\[conv\] Keep the setting of Theorem \[cvb\]. We have $$\sum_{k \ge 0}
{\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{P}}^{k}_{{\mathcal{E}}} ( \lambda ) )^{\ast}
\in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]} \quad
\text{\rm for all $\lambda \in P$},$$ and an [(]{}unambiguously defined[)]{} equality $${\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}},\,{\mathcal{E}}( \lambda ) )^{\ast}
= \sum_{k \ge 0}
(-1)^{k} {\mathop{\rm gdim}\nolimits}\Gamma ( {\mathbf{Q}_{G}}, \, {\mathcal{P}}^{k}_{{\mathcal{E}}} ( \lambda ) )^{\ast}
\in {\mathbb{Z}}{[\hspace{-1pt}[q]\hspace{-1pt}]} \quad
\text{\rm for all $\lambda \in P$}.$$
By Theorem \[cvb\](3), there are only finitely many terms ${\mathcal{P}}^{k}_{{\mathcal{E}}}(\lambda)$ contributing to each homogeneous subspace of a fixed $q$-degree of $\Gamma({\mathbf{Q}_{G}},\,{\mathcal{E}}(\lambda))^{\ast}$. Therefore, the projective resolution afforded in Theorem \[cvb\] implies the desired result. This proves the corollary.
We say that a condition depending on $\lambda \in P$ holds for $\lambda \gg 0$ if there exists $\gamma \in P$ and the condition holds for every $\lambda \in \gamma + P^+$.
For an element $f \in ({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$, we define $|f| \in ({\mathbb{Z}}_{\ge 0}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$ as follows: $$f=\sum_{n \in {\mathbb{Z}}_{\le 0}}
\underbrace{\left(\sum_{\nu \in P} c_{\nu,n} e^{\nu} \right)}_{\in {\mathbb{Z}}[P]} q^{n}
\quad \Rightarrow \quad
|f|:=\sum_{n \in {\mathbb{Z}}_{\le 0}}
\underbrace{\left(\sum_{\nu \in P} |c_{\nu,n}| e^{\nu} \right)}_{\in {\mathbb{Z}}_{\ge 0}[P]} q^{n}.$$ We now define $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ to be the following set of formal infinite sums, modulo equivalence relation $\sim$: $$\left\{ f=\sum_{ \lambda \in P }
f_{\lambda} \cdot [ {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )]
\ \Biggm| \
\text{$f_{\lambda} \in ({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$, $\lambda \in P$,
satisfy condition \eqref{eq:K}} \right\}\Biggm/\sim$$ where condition is given by: $$\label{eq:K}
\sum_{ \lambda \in P }
| f_{\lambda} | \cdot
{\mathop{\rm gch}\nolimits}H^{0} ( {\mathbf{Q}_{G}}, \, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda + \mu ) ) \in ({\mathbb{Z}}_{\ge 0}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}
\quad \text{for every $\mu \in P$}, \tag{\#}
$$ and the equivalence relation $\sim$ is: $$\label{eq:finite-support}
f \sim 0 \iff \sum_{ \lambda \in P } f_{\lambda} \cdot
{\mathop{\rm gch}\nolimits}H^{0} ( {\mathbf{Q}_{G}}, \, {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda + \mu ) ) = 0
\quad \text{if $\mu \gg 0$}. \tag{$\sim$}
$$ By construction, $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ is topologically spanned by classes of ${\mathbf{I}}$-equivalent line bundles. Hence we deduce that the following map is well-defined: $$\begin{split}
& {\mathop{\rm Pic}\nolimits}^{{\widetilde{{\mathbf{I}}}}} {\mathbf{Q}_{G}}\times
K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}) \rightarrow K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}), \\
& \left( {\mathcal{L}}, f = \sum f_{\lambda} [{\mathcal{O}}( \lambda)] \right) \mapsto
[{\mathcal{L}}] \cdot f = \sum f_{\lambda} [{\mathcal{L}}\otimes {\mathcal{O}}( \lambda)].
\end{split}
$$
By Corollary \[conv\], each ${\mathcal{E}}$ from Theorem \[cvb\] satisfies $$[{\mathcal{E}}] := \sum_{k \ge 0} (-1)^{k} [ {\mathcal{P}}^{k}_{{\mathcal{E}}}] \in K' _{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}).$$ In particular, thanks to Corollary \[coh\], we have $[{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda )] \in K' _{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ for every $x \in W_{{\mathrm{af}}}^{\ge 0}$ and $\lambda \in P$.
Let ${\mathop{\rm Fun}\nolimits}_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P$ denote the space of $({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$-valued functions on $P$, and let $${\mathop{\rm Fun}\nolimits}^f_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P \subset {\mathop{\rm Fun}\nolimits}_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P$$ be the subset consisting of those functions that are zero on $\gamma + P^{+}$ for some $\gamma \in P$. Then we form a $({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$-module quotient $${\mathop{\rm Fun}\nolimits}^{\mathrm{ess}}_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P
:= {\mathop{\rm Fun}\nolimits}_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P / {\mathop{\rm Fun}\nolimits}^f_{ ({\mathbb{C}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]} } P.$$ For each $\mu \in P$, we regard the assignment $$P \ni \lambda \mapsto
\begin{cases}
{\mathop{\rm gch}\nolimits}V^{-}_{e} ( - w_{\circ} ( \lambda + \mu ) )
& \text{if $\lambda+\mu \in P^{+}$}, \\[1.5mm]
0 & \text{otherwise},
\end{cases}
$$ as an element of ${\mathop{\rm Fun}\nolimits}_{ ({\mathbb{C}}[P]){[\hspace{-0.5pt}[q^{-1}]\hspace{-0.5pt}]} } P$, which we denote by $\Psi ( [{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )] )$. Passing to the quotient, we obtain a map $\Psi : \{ [{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )] \}_{\mu \in P} \ni [{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )]
\mapsto \Psi ( [{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )] ) \in {\mathop{\rm Fun}\nolimits}^{\mathrm{ess}}_{ ({\mathbb{C}}[P]){[\hspace{-0.5pt}[q^{-1}]\hspace{-0.5pt}]} } P$.
\[Windep\] The map $\Psi$ extends to an injective $({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$-linear map $\Psi : K' _{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}) \rightarrow
{\mathop{\rm Fun}\nolimits}^{\mathrm{ess}}_{ ({\mathbb{C}}[P]){[\hspace{-0.5pt}[q^{-1}]\hspace{-0.5pt}]} } P$.
We assume the contrary to deduce a contradiction. Let $C \in K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$, and expand the $C$ as: $$C = \sum_{
n \in {\mathbb{Z}}_{\le 0},\,\nu,\,\mu \in P}
c _{n,\nu,\mu} \, q^{n} e^{\nu} \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}}(\mu)],
\quad \text{with $c_{n,\nu,\mu} \in {\mathbb{Z}}$},$$ inside $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$. We have $\Psi ( C ) = 0$ if and only if there exists $\gamma \in P$ such that $$\sum_{
n \in {\mathbb{Z}}_{\le 0},\,\nu,\,\mu \in P}
c _{n,\nu,\mu} \, q^{n} e^{\nu} \cdot
{\mathop{\rm gch}\nolimits}V_e^- ( - w_{\circ} ( \lambda + \mu ) ) = 0 \qquad \text{for $\mu \in \gamma + P ^+$}.$$ This is exactly the condition $C \sim 0$. Hence the map $\Psi$ defines an injective map. It is $({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$-linear by construction.
For countably many elements ${C_{p}}$, $p \ge 0$, in $K_{{\widetilde{{\mathbf{I}}}}}' ( {\mathbf{Q}_{G}})$ that represent the classes of ${\widetilde{{\mathbf{I}}}}$-equivariant quasi-coherent sheaves, we expand them as: $${C_{p}}= \sum_{ \lambda \in P } a_{\lambda} ( {C_{p}}) [{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )],
\quad \text{with $a_{\lambda} ( {C_{p}}) \in ({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$}$$ by using the procedure of Theorem \[cvb\]; we say that the sum $\sum _{p \ge 0} {C_{p}}$ converges absolutely to an element of $K'_{{\mathbf{I}}} ( {\mathbf{Q}_{G}})$ if there exists some $\lambda_0 \in P$ (uniformly for all $p \ge 0$) such that $a_{\lambda} ( {C_{p}}) = 0$ for all $\lambda \in P$ with ${\langle \lambda,\,\alpha_{i}^{\vee} \rangle} <
{\langle \lambda_{0},\,\alpha_{i}^{\vee} \rangle}$ for some $i \in I$, and if the number of those $( \lambda, p ) \in
P \times {\mathbb{Z}}_{\ge 0}$ for which $a_{\lambda} ( {C_{p}})$ has a nonzero term of $q$-degree $m$ is finite for each $m \in {\mathbb{Z}}$. It is straightforward to see that $\sum_{p \ge 0} {C_{p}}$ defines an element of $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$, which does not depend on the order of the ${C_{p}}$’s.
\[rem:conv\] Since the coefficients for $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ are in ${\mathbb{Z}}$, the sum $\sum_{p \ge 0} {C_{p}}$ must “diverge” or “oscillate” when it does not converge absolutely.
\[indep-of-QGy\] Let $f_{y}\in ({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$, $y \in W_{{\mathrm{af}}} ^{\ge 0}$. Then the formal sum $$\label{fsum}
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}]$$ converges absolutely to an element of $K'_{{\mathbf{I}}} ( {\mathbf{Q}_{G}})$ if and only if $\sum_{y \in W_{{\mathrm{af}}} ^{\ge 0}} | f_{y} |
\in ({\mathbb{Z}}_{\ge 0}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$. Moreover, in this case, the equation $$\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}] = 0$$ implies $f_{y} = 0$ for all $y \in W_{{\mathrm{af}}}^{\ge 0}$.
First, we remark that $[{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}] \in K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ for each $y \in W_{{\mathrm{af}}}^{\ge 0}$ by Corollary \[coh\] and Theorem \[cvb\]. More precisely, by means of the cohomology vanishing: $$H^{\ast}({\mathbf{Q}_{G}},\,{\mathcal{O}}_{{\mathbf{Q}_{G}}(y)}(\mu))=\{0\} \quad \text{if $\mu \notin P^{+}$},$$ we can take $\lambda_{0}=0$ in Theorem \[cvb\] by setting ${\mathcal{E}}={\mathcal{O}}_{{\mathbf{Q}_{G}}(y)}$. In addition, we have $H^{0} ( {\mathbf{Q}_{G}},\,{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}) = {\mathbb{C}}$. Hence the construction in Theorem \[cvb\] implies that $$\label{v-exp}
[{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}] =
[{\mathcal{O}}_{{\mathbf{Q}_{G}}}] + \sum_{ \lambda \in - P^{+} }
a_{y} ( \lambda ) [ {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda ) ] \in K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$$ for some $a_{y}(\lambda) \in ({\mathbb{Z}[P]}){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$. Therefore, the coefficient of $[{\mathcal{O}}_{{\mathbf{Q}_{G}}}]$ in must be the sum $\sum_{y \in W_{{\mathrm{af}}} ^{\ge 0}} f_{y}$, which unambiguously defines an element of $({\mathbb{Z}[P]}){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$ if and only if the coefficient ($\in {\mathbb{Z}}$) of each $q^{n}$, $n \in {\mathbb{Z}}_{\le 0}$, in the sum $\sum_{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y}$ converges absolutely (see Remark \[rem:conv\]). This proves the first assertion.
We prove the second assertion. Let us assume the contrary to deduce a contradiction. Let $S$ be the set of those $y \in W_{{\mathrm{af}}}^{\ge 0}$ for which $f_{y} \neq 0$; denote by $n_{0}$ the maximal $q$-degree of all $f_{y}$, $y \in S$. Also, let $S_{1}$ be the set of those $y \in S$ for which ${\tau}( y ) \not> {\tau}( y')$ for any $y' \in S$, where for $y \in W_{{\mathrm{af}}}^{\ge 0}$ of the form $y = w t_{\xi}$ with $w \in W$ and $\xi \in Q^{\vee,+}$, we set ${\tau}( y ):=\xi$; since a polynomial ring (of finite variables) is Noetherian, we deduce that $|S_{1}| < \infty$. We choose and fix $y_{0} \in S_{1}$ such that $$P^{\#} := \bigl\{
\lambda \in P^{+} \mid
\text{${\langle \lambda,\,{\tau}(y_{0}) \rangle} < {\langle \lambda,\,{\tau}(y) \rangle}$
for all $y \in S$ with $y \ne y_{0}$} \bigr\}$$ is Zariski dense in ${\mathfrak{h}}^{\ast}$. Let $n_{1}$ denote the maximal $q$-degree of those $f_{y}$, $y \in S_{1}$, for which ${\tau}( y ) = {\tau}( y_{0} )$. Then, the subset $$P^{\#\#} := \bigl\{
\lambda \in P^{+} \mid
\text{${\langle \lambda,\,{\tau}(y_{0}) \rangle} < {\langle \lambda,\,{\tau}(y) \rangle} + n_{0} - n_{1}$
for all $y \in S$ with $y \neq y_{0}$}\bigr\}$$ of $P^{\#}$ is still Zariski dense in ${\mathfrak{h}}^{\ast}$.
For each $\lambda \in P^{\#\#}$, the coefficient of the part of degree $\bigl( n_{1} - {\langle \lambda,\,{\tau}( y_{0} ) \rangle} \bigr)$ of $$\Psi \left(
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}] \right)
( \lambda )$$ is equal to $$\sum _{w \in W} f_{ w t_{{\tau}(y_0)} }^{(n_{1})} \cdot
{\mathop{\rm ch}\nolimits}L^{-}_{w} ( - w_{\circ} \lambda ),$$ where $f_{y}^{(n_{1})} \in {\mathbb{Z}}[P]$ is the part of degree $n_{1}$ of $f_{y}$; here, for $w \in W$ and $\mu \in P^{+}$, $L^{-}_{w} ( \mu ):=U({\mathfrak{b}}^{-})L(\mu)_{w\mu}$ denotes the (opposite) Demazure submodule of $L(\mu)$. This defines a ${\mathbb{Z}[P]}$-valued function of $\lambda \in P^{\#\#}$; note that the above is a finite sum. Here we have the equality ${\mathop{\rm ch}\nolimits}L^{-}_{w} ( - w_{\circ} \lambda )^{\ast} =
D_{w w_{\circ}} ( e ^{\lambda} )$ in terms of the Demazure operator $D_{ww_{\circ}}$ for each $w \in W$ (see [@Kum02 Theorem 8.2.9]); recall that the Demazure operator $D_{i}=D_{s_{i}}$, $i \in I$, is defined by $D_{i}(e^{\mu}):=(e^{\mu}-e^{s_{i}\mu-\alpha_{i}})/(1-e^{-\alpha_{i}})$ for $\mu \in P$. Also, we know by [@Mac91 pp.28–29] that the operators $D_{w}$, $w \in W$, form a set of ${\mathbb{Z}[P]}$-linearly independent ${\mathbb{Z}}$-linear operators acting on ${\mathbb{Z}[P]}$. Therefore, we obtain $f_{w t_{{\tau}( y_{0} )}}^{(n_{1})} = 0$ for all $w \in W$. This is a contradiction, and hence we cannot take the $S_{1}$ above from the beginning. Thus, we conclude the desired result. This completes the proof of the proposition.
\[critK\] Let $x \in W_{{\mathrm{af}}}^{\ge 0}$ and $\lambda \in P^{+}$. Consider a collection $f_{y} ( \lambda ) \in ({\mathbb{Z}[P]}){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$, $y \in W_{{\mathrm{af}}} ^{\ge 0}$, such that $\sum_{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} ( \lambda ) \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}]$ converges absolutely in $K'_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}})$. Then, $$\label{class-eq}
[{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda )] =
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} ( \lambda ) \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}]$$ if and only if $$\label{char-eq}
{\mathop{\rm gch}\nolimits}V^{-}_{x} ( - w_{\circ} ( \lambda + \mu ) ) =
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{v} ( \lambda ) \cdot
{\mathop{\rm gch}\nolimits}V^{-}_{y} ( - w_{\circ} \mu ) \quad
\text{\rm for $\mu \gg 0$}.$$
We have an expansion $$[{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda )] =
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} ( \lambda ) \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}]$$ inside $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$. From this, by twisting by the line bundle ${\mathcal{O}}( \mu )$ for $\mu \in P^{+}$, we obtain $$[{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda + \mu )] =
\sum_{y \in W_{{\mathrm{af}}}^{\ge 0}}
f_{y} ( \lambda ) \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )} ( \mu )].$$ By Corollary \[coh\], this equation in turn implies , which proves the “only if” part of the assertion.
We now assume . Then we have $$\Psi ( [{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )} ( \lambda )] ) =
\sum _{y \in W_{{\mathrm{af}}}^{\ge 0}} f_{y} ( \lambda ) \cdot \Psi ( [{\mathcal{O}}_{{\mathbf{Q}_{G}}( y )}] )$$ by Corollary \[coh\]. Therefore, by Theorem \[Windep\], we deduce that both sides of represent the same class in $K'_{ {\widetilde{{\mathbf{I}}}} } ( {\mathbf{Q}_{G}})$. Thus, we have proved the “if” part of the assertion. This proves the corollary.
\[thm:PC\] For each $\lambda \in P^{+}$ and $x \in W_{{\mathrm{af}}}^{\ge 0}$, there holds the equality $$[{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )] \cdot [{\mathcal{O}}_{{\mathbf{Q}_{G}}( x )}] =
\sum_{\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} (- w_{\circ} \lambda )}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))}q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))} \cdot [ {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\iota(\eta,\,x)} )}]$$ in $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$.
By Theorem \[thm:Dem\], we have $${\mathop{\rm gch}\nolimits}V^{-}_{x}( - w_{\circ}(\lambda + \mu) ) =
\sum_{\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} ( - w_{\circ}\lambda )}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))}q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))} \cdot
{\mathop{\rm gch}\nolimits}V^{-}_{{\iota(\eta,\,x)}}( -w_{\circ} \mu )$$ for each $\mu \in P^{+}$. Taking into account the fact that the LHS is zero if $\lambda + \mu \not\in P^{+}$, and the RHS is zero if $\mu \not \in P^{+}$, we conclude the above equation for $\mu \gg 0$. Here we see from Section \[subsec:extremal\] that ${\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta)) \in {\mathbb{Z}}_{\le 0}$ for each $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(-w_{\circ}\lambda)$. Also, we deduce from that for each $m \in {\mathbb{Z}}_{\le 0}$, there exist only finitely many $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} ( - w_{\circ}\lambda )$ such that ${\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta)) \ge m$. Because ${\mathop{\rm gdim}\nolimits}H^{0} ( {\mathbf{Q}_{G}}, \, {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\iota(\eta,\,x)} )} ( \mu ) )
\in {\mathbb{Z}}{[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$ by Corollary \[coh\], we deduce that $$\sum_{\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x} ( - w_{\circ} \lambda )}
e ^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))}q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\eta))} \cdot
[ {\mathcal{O}}_{{\mathbf{Q}_{G}}( {\iota(\eta,\,x)} )}] \in K_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}).$$ From this, by applying Corollary \[critK\], we conclude the desired result. This proves the theorem.
nil-DAHA action on $K_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}^{\mathrm{rat}}})$. {#sec:nDAHA}
=======================================================================================
The nil-DAHA ${{\mathcal{H}\hspace{-5.6pt}\mathcal{H}}}$ (of adjoint type) is the unital ${\mathbb{Z}}[{\mathbf{q}}^{\pm 1}]$-algebra generated by $T_i$, $i \in I_{{\mathrm{af}}}$, and ${\mathbf{e}}(\nu)$, $\nu \in P$, subject to the following relations: $$\label{eq:nDAHA}
\begin{cases}
T_i ( T_i + 1 ) = 0 & \text{for each $i \in I_{{\mathrm{af}}}$}; \\[3mm]
\text{if
$\overbrace{s_i s_j \cdots}^{\text{$m_{ij}$ times}} =
\overbrace{s_j s_i \cdots}^{\text{$m_{ij}$ times}}$, then
$\overbrace{T_i T_j \cdots}^{\text{$m_{ij}$ times}} =
\overbrace{T_j T_i \cdots}^{\text{$m_{ij}$ times}}$} &
\text{for each $i,\,j \in I_{{\mathrm{af}}}$}; \\[3mm]
\text{${\mathbf{e}}( \nu_{1} ) {\mathbf{e}}( \nu_{2} ) =
{\mathbf{e}}( \nu_{1} + \nu_{2} )$ and ${\mathbf{e}}( 0 ) = 1$} &
\text{for each $\nu_{1},\,\nu_{2} \in P$}; \\[3mm]
T_i {\mathbf{e}}( \nu ) - {\mathbf{e}}( s_i \nu ) T_i =
\dfrac{{\mathbf{e}}( s_i \nu ) - {\mathbf{e}}( \nu )}{1 - {\mathbf{e}}( \alpha_i )}
& \text{for each $\nu \in P$ and $i \in I_{{\mathrm{af}}}$}.
\end{cases}
$$ We define ${\mathcal{H}}$ to be the ${\mathbb{Z}}[{\mathbf{q}}^{-1}]$-subalgebra of ${{\mathcal{H}\hspace{-5.6pt}\mathcal{H}}}$ generated by $T_i$, $i \in I$, and ${\mathbf{e}}( \nu )$, $\nu \in P$.
\[KK-reint\] The assignment $$\begin{aligned}
{\mathtt{e}}( {\varpi}_{i} ) & :
[{\mathbb{C}}_{\lambda} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )] \mapsto
[{\mathbb{C}}_{- {\varpi}_{i} + \lambda} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )], \\
{\mathtt{T}}_{i} & : [{\mathbb{C}}_{\lambda} \otimes _{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )] \mapsto
\frac{{\mathtt{e}}(-\lambda) - {\mathtt{e}}(-s_{i}\lambda+\alpha_{i})}
{1-{\mathtt{e}}(\alpha_{i})}
[{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu )],\end{aligned}$$ for each $i \in I$ and $\lambda,\,\mu \in P$, equips $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ with an action of the subalgebra ${\mathcal{H}}$ of ${{\mathcal{H}\hspace{-5.6pt}\mathcal{H}}}$ through the identifications[:]{} $${\mathbf{q}}\mapsto q^{-1}, \qquad
T_{i} \mapsto {\mathtt{T}}_{i} - 1 \quad \text{\rm for $i \in I$}, \qquad
{\mathbf{e}}( \nu ) \mapsto {\mathtt{e}}( \nu ) \quad \text{\rm for $\nu \in P$}.
$$
By the construction, $K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ contains a dense subset isomorphic to $({\mathbb{Z}}[P]){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}
\otimes_{{\mathbb{Z}}} K _{G} ( G / B )$ (see Proposition \[line-bundles\]). Also, we have a surjection $({\mathbb{Z}}[P])[q^{-1}] \otimes_{{\mathbb{Z}}} K _{G} ( G / B )
\twoheadrightarrow
{\mathbb{Z}}[q^{-1}] \otimes_{{\mathbb{Z}}} K_{B}(G/B)$; see, e.g., [@KK90 (3.17)]. Here, for each $i \in I$, the action of ${\mathtt{T}}_i$ is identical to the action of the Demazure operator $D_{i}=D_{s_{i}}$, and the action of ${\mathtt{e}}({\varpi}_{i})$ corresponds to the twist by the $B$-character $-{\varpi}_{i}$; these define an ${\mathcal{H}}$-action on $K _{B} ( G / B )$ by [@KK90 Sect. 3]. Notice that both of the actions of ${\mathtt{e}}(\,\cdot\,)$ and $T_i$, $i \in I$, are neutral with respect to tensoring with ${\mathcal{O}}_{G/B} ( \lambda )$ for each $\lambda \in P$, and that they also commute with the ${\mathbb{G}_{m}}$-twist corresponding to $q^{-1}$. Therefore, the ${\mathcal{H}}$-action on $K_B ( G / B ) \cong {\mathbb{Z}}[P] [{\mathcal{O}}_{G/B}]$ induces an ${\mathcal{H}}$-action on $({\mathbb{Z}[P]})[q^{-1}] \otimes_{{\mathbb{Z}}} K _{G} ( G / B )$ through $$({\mathbb{Z}[P]})[q^{-1}] \otimes_{{\mathbb{Z}}} K _{G} ( G / B ) \cong
({\mathbb{Z}[P]})[q^{-1}] \otimes_{{\mathbb{Z}}} {\mathbb{Z}}[P] [{\mathcal{O}}_{G/B}] =
\bigoplus _{\lambda \in P} ({\mathbb{Z}[P]})[ q^{-1} ] [{\mathcal{O}}_{G/B} ( \lambda )],$$ where the second factor of the leftmost one is responsible for the factors $\{ [{\mathcal{O}}_{G/B} ( \lambda )] \} _{\lambda \in P}$. Finally, we complete $({\mathbb{Z}}[P])[ q^{-1} ] \otimes_{{\mathbb{Z}}} K_{B}(G/B)$ to obtain the desired assertion. This proves the proposition.
The ${\mathcal{H}}$-action in Proposition [\[KK-reint\]]{} is induced by the ${\widetilde{{\mathbf{I}}}}$-character twists and the convolution action of the structure sheaves through $q_{i,e}$ for $i \in I$ [(]{}see .
The assertion holds for the actions of ${\mathcal{H}}$ on $K_B ( G / B )$ and ${\mathbb{Z}}[q^{-1}] \otimes_{{\mathbb{Z}}} K _{B} ( G / B )$ by [@KK90 Sect. 3]. Also, by Lemma \[one-step\], for each $i \in I$, the map $q_{i,e}$ is a ${\mathbb{P}}^{1}$-fibration, and hence $${\mathbb{R}}^{k} (q_{i,e})_{\ast} {\mathcal{O}}_{{\mathbf{I}}( i ) \times ^{{\mathbf{I}}} {\mathbf{Q}_{G}}} \cong
\begin{cases}
{\mathcal{O}}_{{\mathbf{Q}_{G}}} & \text{if $k=0$}, \\[1mm]
\{ 0 \} & \text{if $k \neq 0$}.
\end{cases}
$$ This implies that the convolution action of ${\mathbf{I}}( i ) / {\mathbf{I}}$, $i \in I$, on ${\mathbf{Q}_{G}}$ fixes the classes of $[{\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \lambda )]$ for each $\lambda \in P$ by the projection formula. Taking into account the fact that the twist of $R ( {\widetilde{{\mathbf{I}}}} ) \cong R(B \times {\mathbb{G}_{m}})$ has an effect through the fiber of $q_{i,e}$, we conclude that $T_{i}$, $i \in I$, is identical to the convolution action induced by $q_{i,e}$ through the inclusion ${\mathbb{Z}}[q^{-1}] \otimes_{{\mathbb{Z}}} K _{B} ( G / B )
\subset K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$. This proves the corollary.
For each $\xi \in Q^{\vee,+}$, the natural inclusion map $\imath_{\xi} : {\mathbf{Q}_{G}}\hookrightarrow {\mathbf{Q}_{G}}$ induces an inclusion $(\imath_{\xi})_{\ast} :
K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}}) \hookrightarrow K'_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}})$ of $({\mathbb{Z}[P]}){[\hspace{-1pt}[q^{-1}]\hspace{-1pt}]}$-modules such that $( \imath_{\xi} )_{\ast} [{\mathcal{O}}_{{\mathbf{Q}_{G}}(x)} ( \lambda )] =
[{\mathcal{O}}_{{\mathbf{Q}_{G}}(xt_{\xi})} ( \lambda )]$ for each $x \in W_{{\mathrm{af}}}$. We define $$K_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}^{\mathrm{rat}}}) :=
{\mathbb{Z}}{(\hspace{-1pt}(q^{-1})\hspace{-1pt})} \otimes_{{\mathbb{Z}}{[\hspace{-0.5pt}[q^{-1}]\hspace{-0.5pt}]}}
\varinjlim K'_{{\widetilde{{\mathbf{I}}}}}({\mathbf{Q}_{G}}).$$
\[thm:BFU\] The assignment $$\begin{aligned}
{\mathtt{e}}({\varpi}_{i}) & :
[{\mathbb{C}}_{\lambda} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \mu )] \mapsto
[{\mathbb{C}}_{-{\varpi}_{i} + \lambda} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \mu )]
\quad \text{\rm for $i \in I$}, \\[1mm]
{\mathtt{T}}_i & : [{\mathbb{C}}_{\lambda} \otimes _{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \mu ) ] \\[3mm]
& \hspace*{5mm} \mapsto
\begin{cases}
\dfrac{{\mathtt{e}}( - \lambda ) - {\mathtt{e}}( - s_0 \lambda )}
{1 - {\mathtt{e}}( \alpha_0 )} [{\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \mu )] +
{\mathtt{e}}( - s_0 \lambda ) [ {\mathcal{O}}_{{\mathbf{Q}_{G}}( s_0 t_{\xi} )} ( \mu )]
& \text{\rm for $i = 0$}, \\[7mm]
\dfrac{{\mathtt{e}}( - \lambda ) - {\mathtt{e}}( - s_i \lambda + \alpha_i )}
{1 - {\mathtt{e}}( \alpha_i )} [{\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \mu )]
& \text{\rm for $i \neq 0$},
\end{cases}
$$ where $\xi \in Q^{\vee,+}$, and $\lambda, \mu \in P$, equips $K_{{\widetilde{{\mathbf{I}}}}} ({\mathbf{Q}_{G}^{\mathrm{rat}}})$ with an action of ${{\mathcal{H}\hspace{-5.6pt}\mathcal{H}}}$ through the identifications[:]{} $${\mathbf{q}}\mapsto q^{-1}, \qquad
T_{i} \mapsto {\mathtt{T}}_{i} - 1 \quad \text{\rm for $i \in I_{{\mathrm{af}}}$}, \qquad
{\mathbf{e}}( \nu ) \mapsto {\mathtt{e}}( \nu ) \quad \text{\rm for $\nu \in P$}.$$
Thanks to [@KK90 Sect. 3] (and Lemma \[one-step\]), for each $i \in I$, the action of ${\mathtt{T}}_i$ is induced by the pushforward of an ${\widetilde{{\mathbf{I}}}}$-equivariant inflated sheaf through $q_{i,e}$ (see Section \[subsec:BSDH\]), and the action of ${\mathtt{e}}({\varpi}_{i})$ is induced by an ${\widetilde{{\mathbf{I}}}}$-character twist. Because these geometric counterparts commute with the pullback through $\imath_{\xi}$ for each $\xi \in Q^{\vee, +}$, our formulas define an action of ${\mathcal{H}}$ on $K_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}^{\mathrm{rat}}})$ induced by Proposition \[KK-reint\].
Now, we have $$\label{D-factor}
\frac{f g - e^{- \alpha_0} s_0 (f g )}{1 - e^{- \alpha_{0}}} =
\frac{f - s_{0}(f)}{1 - e^{- \alpha_{0}}} g + s_{0} ( f )
\frac{g - e^{- \alpha_{0}} s_{0} ( g )}{1 - e^{- \alpha_{0}}}
\quad \text{for $f,\,g \in ({\mathbb{C}}[P]){(\hspace{-1pt}(q^{-1})\hspace{-1pt})}$}.
$$ Let $p_{0,t_{\xi}}:{\mathbf{I}}(0) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}(t_{\xi}) \rightarrow {\mathbb{P}}^{1}$ be the inflation of the structure map of ${\mathbf{Q}_{G}}(t_{\xi})$, and let ${\mathcal{E}}(W)$ denote the vector bundle over ${\mathbb{P}}^{1} \cong \tilde{{\mathbf{I}}}(0)/\tilde{{\mathbf{I}}}$ associated to an $\tilde{{\mathbf{I}}}$-module $W$. Then, by taking into account equation , Corollary \[coh\] and Theorem \[cvb\], and [@Kat16 Corollary 4.8], we deduce that for each $\lambda,\,\mu \in P$, $\nu \in P$, and $\xi \in Q^{\vee,+}$ such that $s_{0} t_{\xi} \in W_{{\mathrm{af}}}^{\ge 0}$, $$\begin{aligned}
\sum_{m,\,n \ge 0} (-1)^{m+n} &
{\mathop{\rm gch}\nolimits}H^{m} \Bigl( {\mathbf{Q}_{G}}, \, {\mathbb{R}}^{n} ( q_{0,t_{\xi}} ) _{\ast}
\bigl( {\mathbb{C}}_{- \nu} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi})} ( \lambda ) \bigr)
\otimes_{{\mathcal{O}}_{{\mathbf{Q}_{G}}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}} ( \mu ) \Bigr) \nonumber \\[2mm]
= & \sum_{m \ge 0} (-1)^{m}
{\mathop{\rm gch}\nolimits}H^{m} \bigl( {\mathbf{I}}( 0 ) \times^{{\mathbf{I}}} {\mathbf{Q}_{G}}, \,
{\mathbb{C}}_{- \nu} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi})} ( \lambda + \mu ) \bigr) \nonumber \\[2mm]
= & \sum_{m \ge 0} (-1)^{m}
{\mathop{\rm gch}\nolimits}H^{0} \Bigl( {\mathbb{P}}^{1}, \, {\mathbb{R}}^{m} ( p_{0, t_{\xi}} )_{\ast}
\bigl( {\mathbb{C}}_{- \nu} \otimes_{{\mathbb{C}}}
{\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi})} ( \lambda + \mu ) \bigr) \Bigr) \nonumber \\[2mm]
= & \sum_{m \ge 0} (-1)^{m}
{\mathop{\rm gch}\nolimits}H^{0} \Bigl( {\mathbb{P}}^{1}, \, {\mathbb{C}}_{- \nu} \otimes_{{\mathbb{C}}}
{\mathcal{E}}\bigl( H^{0} ( {\mathbf{Q}_{G}}( t_{\xi} ), \,
{\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \lambda + \mu ) )^{\ast} \bigr) \Bigr) \nonumber \\[2mm]
= & \dfrac{e^{-\nu} \cdot {\mathop{\rm gch}\nolimits}V^{-}_{t_\xi} (-w_{\circ}(\lambda + \mu)) -
e^{- \alpha_0} s_0 ( e^{-\nu} \cdot {\mathop{\rm gch}\nolimits}V^{-}_{t_\xi} (-w_{\circ}(\lambda + \mu)) )}
{1 - e^{- \alpha_0}} \nonumber \\[2mm]
= & \dfrac{ e^{-\nu} - s_0 ( e^{-\nu} )}
{1 - e^{- \alpha_0}}
{\mathop{\rm gch}\nolimits}V^{-}_{t_\xi} (-w_{\circ}(\lambda + \mu)) +
e^{- s_0 \nu} {\mathop{\rm gch}\nolimits}V^{-}_{s_0 t_\xi} (-w_{\circ}(\lambda + \mu)) \nonumber \\[2mm]
= & \dfrac{ e^{-\nu} - s_0 ( e^{-\nu} )}{1 - e^{- \alpha_0}}
\Psi ( [ {\mathcal{O}}_{{\mathbf{Q}_{G}}( t_{\xi} )} ( \lambda ) ] ) ( \mu ) +
e^{- s_0 \nu} \Psi ( [ {\mathcal{O}}_{{\mathbf{Q}_{G}}( s_0 t_{\xi} )} ( \lambda ) ] ) ( \mu ), \label{conv-eqn}\end{aligned}$$ where the first and fourth equalities follow by the Leray spectral sequence. In particular, the term represents the image under $\Psi$ of the convolution of $[{\mathbb{C}}_{-\mu} \otimes_{{\mathbb{C}}} {\mathcal{O}}_{{\mathbf{Q}_{G}}(t_{\xi})} ( \lambda )]$ with respect to $q_{0,t_{\xi}}$. Therefore, from the injectivity of $\Psi$, we conclude that ${\mathtt{T}}_{0}$ is induced by the pushforward of an ${\widetilde{{\mathbf{I}}}}$-equivariant inflated sheaf through $q_{0, t_{\xi}}$ for some $\xi \in Q^{\vee, +}$.
From the above, we deduce that the actions ${\mathtt{T}}_i$, $i \in I_{{\mathrm{af}}}$, and ${\mathbf{e}}(\nu)$, $\nu \in P$, generate the convolution action of Schubert cells and the ${\widetilde{{\mathbf{I}}}}$-character twists of the (thin) affine flag manifold $G{(\hspace{-1pt}(z)\hspace{-1pt})} / {\mathbf{I}}$ on ${\mathbf{Q}_{G}^{\mathrm{rat}}}$ (or rather, on ${\mathbf{Q}_{G}}$). In particular, the $T_{i}$, $i \in I_{{\mathrm{af}}}$, generate the nil-Hecke algebra of affine type by [@KK90 Sect. 3]. Therefore, their commutation relations with ${\mathtt{e}}( {\varpi}_i )$, $i \in I$, imply that the $T_{i}$, $i \in I_{{\mathrm{af}}}$, and the ${\mathtt{e}}( {\varpi}_{i} )$, $i \in I$, satisfy the relations for ${{\mathcal{H}\hspace{-5.6pt}\mathcal{H}}}$ (see also [@BFU Sect. 3.4]); we remark that their convention differs from ours by the twist by the Serre duality and line bundle twist [@BFU Sects. 3.1 and 3.21].
Finally, we complete the proof by observing that ${\mathtt{T}}_0$ preserves $K_{{\widetilde{{\mathbf{I}}}}} ( {\mathbf{Q}_{G}^{\mathrm{rat}}})$ by inspection.
Proof of Theorem \[thm:SMT\]. {#sec:prf-SMT}
=============================
Affine Weyl group action. {#subsec:Weyl}
-------------------------
Let ${\mathcal{B}}$ be a regular crystal in the sense of [@Kas02 Sect. 2.2] (or, a normal crystal in the sense of [@HK p.389]); for example, ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ for $\lambda \in P^{+}$ is a regular crystal by Theorem \[thm:isom\], and hence so is ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ for $\lambda,\,\mu \in P^{+}$. Then we know from [@Kas94 Sect. 7] that the affine Weyl group $W_{{\mathrm{af}}}$ acts on ${\mathcal{B}}$ as follows: for $b \in {\mathcal{B}}$ and $i \in I_{{\mathrm{af}}}$, $$\label{eq:W-act}
s_{i} \cdot b :=
\begin{cases}
f_{i}^{n}b & \text{if $n:={\langle {\mathop{\rm wt}\nolimits}(b),\,\alpha_{i}^{\vee} \rangle} \ge 0$}, \\[1.5mm]
e_{i}^{-n}b & \text{if $n:={\langle {\mathop{\rm wt}\nolimits}(b),\,\alpha_{i}^{\vee} \rangle} \le 0$}.
\end{cases}
$$
Also, for $b \in {\mathcal{B}}$ and $i \in I_{{\mathrm{af}}}$, we define $e_{i}^{\max}b=e_{i}^{{\varepsilon}_{i}(b)}b$ and $f_{i}^{\max}b=f_{i}^{{\varphi}_{i}(b)}b$, where ${\varepsilon}_{i}(b):=\max\bigl\{n \ge 0 \mid e_{i}^{n}b \ne {\bm{0}}\bigr\}$ and ${\varphi}_{i}(b):=\max\bigl\{n \ge 0 \mid f_{i}^{n}b \ne {\bm{0}}\bigr\}$; note that if $b \in {\mathcal{B}}$ satisfies $e_{i}b = {\bm{0}}$ (resp., $f_{i}b = {\bm{0}}$), i.e., ${\varepsilon}_{i}(b)=0$ (resp., ${\varphi}_{i}(b)=0$), then $f_{i}^{\max}b = s_{i} \cdot b$ (resp., $e_{i}^{\max}b = s_{i} \cdot b$).
Connected components of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$. {#subsec:conn}
---------------------------------------------------------------------
Let $\lambda \in P^{+}$, and write it as $\lambda = \sum_{i \in I} m_{i} {\varpi}_{i}$, with $m_{i} \in {\mathbb{Z}}_{\ge 0}$; note that $J=\bigl\{i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}=
\bigl\{i \in I \mid m_{i} = 0\bigr\}$. We define ${\mathop{\rm Par}\nolimits}(\lambda)$ to be the set of $I$-tuples of partitions ${\bm{\rho}}= (\rho^{(i)})_{i \in I}$ such that $\rho^{(i)}$ is a partition of length (strictly) less than $m_{i}$ for each $i \in I$; a partition of length less than $0$ (or $1$) is understood to be the empty partition $\emptyset$. Also, for ${\bm{\rho}}= (\rho^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda)$, we set $|{\bm{\rho}}|:=\sum_{i \in I} |\rho^{(i)}|$, where for a partition $\rho = (\rho_{1} \ge \rho_{2} \ge \cdots \ge \rho_{m})$, we set $|\rho| := \rho_{1}+\cdots+\rho_{m}$. We endow the set ${\mathop{\rm Par}\nolimits}(\lambda)$ with a crystal structure as follows: for ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$ and $i \in I_{{\mathrm{af}}}$, $$e_{i} {\bm{\rho}}= f_{i} {\bm{\rho}}:= {\bm{0}}, \quad
{\varepsilon}_{i} ({\bm{\rho}}) = {\varphi}_{i} ({\bm{\rho}}) := -\infty, \quad
{\mathop{\rm wt}\nolimits}({\bm{\rho}}) := - |{\bm{\rho}}| \delta.$$
We recall from [@INS Sect. 7] the relation between ${\mathop{\rm Par}\nolimits}(\lambda)$ and the set ${\mathop{\rm Conn}\nolimits}({\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda))$ of connected components of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$. We set ${\mathop{\rm Turn}\nolimits}(\lambda):=
\bigl\{k/m_{i} \mid i \in I \setminus {J}\text{ and }
0 \le k \le m_{i}\bigr\}$. By [@INS Proposition 7.1.2], each connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ contains a unique element of the form: $$\label{eq:ext}
\bigl( {\Pi^{J}}(t_{\xi_{1}}),\,\dots,\,{\Pi^{J}}(t_{\xi_{s-1}}),\,e \,;\,
a_{0},\,a_{1},\,\dots,\,a_{s} \bigr),$$ where $s \ge 1$, $\xi_{1},\,\dots,\,\xi_{s-1} \in Q^{\vee}_{I \setminus {J}}$ such that $\xi_{1} > \cdots > \xi_{s-1} > 0=:\xi_{s}$, and $a_{u} \in {\mathop{\rm Turn}\nolimits}(\lambda)$ for all $0 \le u \le s$. For each element of the form (or equivalently, each connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$), we define an element ${\bm{\rho}}= (\rho^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda)$ as follows. First, let $i \in I \setminus {J}$; note that $m_{i} \ge 1$. For each $1 \le k \le m_{i}$, take $0 \le u \le s$ such that $a_{u}$ is contained in the interval $\bigl( (k-1)/m_{i},\,k/m_{i} \bigr]$. Then we define $\rho^{(i)}_{k}$ to be ${\langle {\varpi}_{i},\,\xi_{u} \rangle}$, the coefficient of $\alpha_{i}^{\vee}$ in $\xi_{u}$; we know from (the proof of) [@INS Proposition 7.2.1] that $\rho^{(i)}_{k}$ does not depend on the choice of $u$ above. Since $\xi_{1} > \cdots > \xi_{s-1} > 0=\xi_{s}$, we see that $\rho^{(i)}_{1} \ge \cdots \ge
\rho^{(i)}_{m_{i}-1} \ge \rho^{(i)}_{m_{i}}=0$. Thus, for each $i \in I \setminus {J}$, we obtain a partition of length less than $m_{i}$. For $i \in {J}$, we set $\rho^{(i)}:=\emptyset$. Thus we obtain an element ${\bm{\rho}}= (\rho^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda)$, and hence a map from ${\mathop{\rm Conn}\nolimits}({\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda))$ to ${\mathop{\rm Par}\nolimits}(\lambda)$. Moreover, we know from [@INS Proposition 7.2.1] that this map is bijective; we denote by $\pi_{{\bm{\rho}}} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ the element of the form corresponding to ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$ under this bijection.
\[rem:ext\] Let ${\bm{\rho}}= (\rho^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda)$, with $\rho^{(i)}=(\rho^{(i)}_{1} \ge \cdots )$ for $i \in I$; note that $\rho^{(i)}_{1}=0$ if $\rho^{(i)}=\emptyset$. It follows from the definition that $$\label{eq:par0}
\iota(\pi_{{\bm{\rho}}}) = {\Pi^{J}}(t_{\xi_{1}}), \quad
\text{where} \quad \xi_{1} = \sum_{i \in I}
\rho^{(i)}_{1} \alpha_{i}^{\vee} \in Q^{\vee}_{I \setminus {J}}.$$
For ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, we denote by ${\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda)$ the connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ containing $\pi_{{\bm{\rho}}}$. Also, we denote by ${\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)$ the connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ containing $\pi_{\lambda}=(e\,;\,0,1)$; note that $\pi_{\lambda}=\pi_{{\bm{\rho}}}$ for ${\bm{\rho}}=(\emptyset)_{i \in I}$. We know from [@INS Proposition 3.2.4] (and its proof) that for each ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, there exists an isomorphism ${\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda) \stackrel{\sim}{\rightarrow}
\bigl\{{\bm{\rho}}\bigr\} \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)$ of crystals, which maps $\pi_{{\bm{\rho}}}$ to ${\bm{\rho}}\otimes \pi_{\lambda}$. Hence we have $$\label{eq:isom}
{\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) =
\bigsqcup_{{\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)} {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda) \cong
\bigsqcup_{{\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)}
\bigl\{{\bm{\rho}}\bigr\} \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda) \quad \text{as crystals}.$$
The following lemma is shown by induction on the (ordinary) length $\ell(x)$ of $x$; for part (1), see also [@NS16 Remark 3.5.2]
\[lem:Weyl\]
Let $\lambda \in P^{+}$. If $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ is of the form , then for $x \in W_{{\mathrm{af}}}$, $$\label{eq:tx0}
x \cdot \pi =
\bigl( {\Pi^{J}}(xt_{\xi_{1}}),\,\dots,\,{\Pi^{J}}(xt_{\xi_{s-1}}),\,{\Pi^{J}}(x) \,;\,
a_{0},\,a_{1},\,\dots,\,a_{s} \bigr).$$
Let $\lambda,\,\mu \in P^{+}$. Let ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$, and $\xi,\zeta \in Q^{\vee}$. Then, for $x \in W_{{\mathrm{af}}}$, $$\label{eq:Weyl}
x \cdot \bigl( (t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes (t_{\zeta} \cdot \pi_{{\bm{\chi}}}) \bigr)
= (xt_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes (xt_{\zeta} \cdot \pi_{{\bm{\chi}}}).$$
Let $\xi \in Q^{\vee}$. It follows from Lemma \[lem:SiB\](3) that if $\pi = (x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$, then $$\label{eq:Txi}
T_{\xi}\pi:=
\bigl( {\Pi^{J}}(x_{1}t_{\xi}),\,\dots,\,{\Pi^{J}}(x_{s}t_{\xi})\,;\,{\mathbf{a}}\bigr) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda);$$ the map $T_{\xi} : {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \rightarrow {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ is clearly bijective, with $T_{\xi}^{-1}=T_{-\xi}$. We can verify by the definitions that $$\label{eq:Tx}
\begin{cases}
T_{\xi}e_{i}\pi= e_{i}T_{\xi}\pi, \quad
T_{\xi}f_{i}\pi= f_{i}T_{\xi}\pi
& \text{for $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $i \in I_{{\mathrm{af}}}$}, \\[1.5mm]
{\varepsilon}_{i}(T_{\xi}\pi) = {\varepsilon}_{i}(\pi), \quad
{\varphi}_{i}(T_{\xi}\pi) = {\varphi}_{i}(\pi)
& \text{for $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $i \in I_{{\mathrm{af}}}$}, \\[1.5mm]
{\mathop{\rm wt}\nolimits}(T_{\xi}\pi) = {\mathop{\rm wt}\nolimits}(\pi) - {\langle \lambda,\,\xi \rangle}\delta,
\end{cases}
$$ where $T_{\xi}{\bm{0}}$ is understood to be ${\bm{0}}$.
\[rem:Tx\] Let ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, and assume that $\pi_{{\bm{\rho}}}$ is of the form . For $\xi \in Q^{\vee}$, we see from and that $$T_{\xi}\pi_{{\bm{\rho}}} =
\bigl( {\Pi^{J}}(t_{\xi_{1}+\xi}),\,\dots,\,{\Pi^{J}}(t_{\xi_{s-1}+\xi}),\,{\Pi^{J}}(t_{\xi}) \,;\,
a_{0},\,a_{1},\,\dots,\,a_{s} \bigr) = t_{\xi} \cdot \pi_{{\bm{\rho}}},$$ which implies that $T_{\xi}\pi_{{\bm{\rho}}} \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda)$. Therefore, it follows from that $T_{\xi}({\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda)) = {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda)$.
Quantum Lakshmibai-Seshadri paths. {#subsec:QLS}
----------------------------------
Let $\lambda \in P^{+}$. Let ${\mathop{\rm cl}\nolimits}: {\mathbb{R}}\otimes_{{\mathbb{Z}}} P_{{\mathrm{af}}} \twoheadrightarrow
({\mathbb{R}}\otimes_{{\mathbb{Z}}} P_{{\mathrm{af}}})/{\mathbb{R}}\delta$ denote the canonical projection. For an element $\pi=(x_{1},\,\dots,\,x_{s}\,;\,a_{0},\,a_{1},\,\dots,\,a_{s}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$, we define a piecewise-linear, continuous map ${\mathop{\rm cl}\nolimits}(\pi) : [0,1] \rightarrow ({\mathbb{R}}\otimes_{{\mathbb{Z}}} P_{{\mathrm{af}}})/{\mathbb{R}}\delta$ by $({\mathop{\rm cl}\nolimits}(\pi))(t):={\mathop{\rm cl}\nolimits}({\overline{\pi}}(t))$ for $t \in [0,1]$ (for ${\overline{\pi}}$, see ). As explained in [@NS16 Sect. 6.2], the set $\bigl\{{\mathop{\rm cl}\nolimits}(\pi) \mid \pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)\bigr\}$ is identical to the set ${\mathbb{B}}(\lambda)_{{\mathop{\rm cl}\nolimits}}$ of all “projected (by ${\mathop{\rm cl}\nolimits}$)” LS paths of shape $\lambda$, introduced in [@NS05 (3.4)] and [@NS06 page 117] (see also [@LNSSS2 Sect. 2.2]). Also, by [@LNSSS2 Theorem 3.3], ${\mathbb{B}}(\lambda)_{{\mathop{\rm cl}\nolimits}}$ is identical to the set ${\mathrm{QLS}}(\lambda)$ of all quantum LS paths of shape $\lambda$, introduced in [@LNSSS2 Sect. 3.2]. We can endow the set ${\mathbb{B}}(\lambda)_{{\mathop{\rm cl}\nolimits}} = {\mathrm{QLS}}(\lambda)$ with a crystal structure with weights in ${\mathop{\rm cl}\nolimits}(P_{{\mathrm{af}}})$ in such a way that $$\label{eq:cl}
\begin{cases}
e_{i}{\mathop{\rm cl}\nolimits}(\pi) = {\mathop{\rm cl}\nolimits}(e_{i}\pi), \quad
f_{i}{\mathop{\rm cl}\nolimits}(\pi) = {\mathop{\rm cl}\nolimits}(f_{i}\pi)
& \text{for $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $i \in I_{{\mathrm{af}}}$}, \\[1.5mm]
{\mathop{\rm wt}\nolimits}({\mathop{\rm cl}\nolimits}(\pi))={\mathop{\rm cl}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))
& \text{for $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$}, \\[1.5mm]
{\varepsilon}_{i}({\mathop{\rm cl}\nolimits}(\pi))={\varepsilon}_{i}(\pi), \quad
{\varphi}_{i}({\mathop{\rm cl}\nolimits}(\pi))={\varphi}_{i}(\pi)
& \text{for $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $i \in I_{{\mathrm{af}}}$},
\end{cases}
$$ where we understand that ${\mathop{\rm cl}\nolimits}({\bm{0}}) = {\bm{0}}$. The next theorem follows from [@NS05 Proposition 3.23 and Theorem 3.2].
\[thm:QLS\]
For every $\lambda \in P^{+}$, the crystal ${\mathrm{QLS}}(\lambda)={\mathbb{B}}(\lambda)_{{\mathop{\rm cl}\nolimits}}$ is connected.
For every $\lambda,\,\mu \in P^{+}$, there exists an isomorphism ${\mathrm{QLS}}(\lambda) \otimes {\mathrm{QLS}}(\mu) \cong {\mathrm{QLS}}(\lambda+\mu)$ of crystals. In particular, ${\mathrm{QLS}}(\lambda) \otimes {\mathrm{QLS}}(\mu)$ is connected.
\[lem:conn\] Let $\lambda,\,\mu \in P^{+}$, and set ${J}:=\bigl\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$, ${K}:=\bigl\{ i \in I \mid {\langle \mu,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$. Each connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ contains an element of the form[:]{} $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ for some $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$, ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, and ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$.
Let $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. By Theorem \[thm:QLS\](2), there exists a monomial $X$ in root operators on ${\mathrm{QLS}}(\lambda) \otimes {\mathrm{QLS}}(\mu)$ such that $X \bigl( {\mathop{\rm cl}\nolimits}(\pi) \otimes {\mathop{\rm cl}\nolimits}(\eta) \bigr) =
{\mathop{\rm cl}\nolimits}(\pi_{\lambda}) \otimes {\mathop{\rm cl}\nolimits}(\pi_{\mu})$; recall that $\pi_{\lambda}=(e\,;\,0,\,1) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\pi_{\mu}=(e\,;\,0,\,1) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. It follows from and the tensor product rule for crystals that $X(\pi \otimes \eta)$ is of the form $\pi_{1} \otimes \eta_{1}$ for some $\pi_{1} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ such that ${\mathop{\rm cl}\nolimits}(\pi_{1})={\mathop{\rm cl}\nolimits}(\pi_{\lambda})$ and $\eta_{1} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ such that ${\mathop{\rm cl}\nolimits}(\eta_{1})={\mathop{\rm cl}\nolimits}(\pi_{\mu})$. Here, we see from [@NS16 Lemma 6.2.2] that $$\begin{split}
& \bigl\{ \pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \mid {\mathop{\rm cl}\nolimits}(\pi)={\mathop{\rm cl}\nolimits}(\pi_{\lambda})\bigr\} =
\bigl\{ t_{\xi} \cdot \pi_{{\bm{\rho}}} \mid {\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda),\,\xi \in Q^{\vee}\bigr\}, \\
& \bigl\{ \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu) \mid {\mathop{\rm cl}\nolimits}(\eta)={\mathop{\rm cl}\nolimits}(\pi_{\mu})\bigr\} =
\bigl\{ t_{\zeta} \cdot \pi_{{\bm{\chi}}} \mid {\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu),\,\zeta \in Q^{\vee}\bigr\}.
\end{split}
$$ Therefore, $X(\pi \otimes \eta) =
\bigl(t_{\xi_1} \cdot \pi_{{\bm{\rho}}}\bigr) \otimes \bigl(t_{\zeta_1} \cdot \pi_{{\bm{\chi}}}\bigr)$ for some ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, $\xi_1 \in Q^{\vee}$ and ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$, $\zeta_1 \in Q^{\vee}$. Also, by , we have $t_{-\zeta_1} \cdot \bigl( (t_{\xi_1} \cdot \pi_{{\bm{\rho}}}) \otimes
(t_{\zeta_1} \cdot \pi_{{\bm{\chi}}}) \bigr) =
(t_{\xi_1-\zeta_1} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$; we deduce from and that $t_{\xi_1-\zeta_1} \cdot \pi_{{\bm{\rho}}} = t_{\xi_2} \cdot \pi_{{\bm{\rho}}}$, with $\xi_2=[\xi_1-\zeta_1]^{{J}}$, where $[\,\cdot\,]^{{J}}:Q^{\vee} \twoheadrightarrow Q^{\vee}_{I \setminus {J}}$ is the projection in . We set $\gamma:=[\xi_{2}]_{{K}} \in Q^{\vee}_{K}$, where $[\,\cdot\,]_{{K}}:Q^{\vee} \twoheadrightarrow Q^{\vee}_{{K}}$ is the projection defined as in ; we deduce from and that $t_{-\gamma} \cdot \pi_{{\bm{\chi}}} = \pi_{{\bm{\chi}}}$. In addition, we set $\xi:=\xi_{2}-\gamma$; notice that $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$. Summarizing the above, we have $$\begin{aligned}
(t_{-\gamma}t_{-\zeta_1}) \cdot X(\pi \otimes \eta) & =
(t_{-\gamma}t_{-\zeta_1}) \cdot
\bigl( (t_{\xi_1} \cdot \pi_{{\bm{\rho}}}) \otimes
(t_{\zeta_1} \cdot \pi_{{\bm{\chi}}}) \bigr) =
t_{-\gamma} \cdot
\bigl( (t_{\xi_1-\zeta_1} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} \bigr) \\
& = t_{-\gamma} \cdot
\bigl( (t_{\xi_2} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} \bigr)
= (t_{\xi_2-\gamma} \cdot \pi_{{\bm{\rho}}}) \otimes (t_{-\gamma} \cdot \pi_{{\bm{\chi}}}) \\
& = (t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}. \end{aligned}$$ Because the action of the affine Weyl group $W_{{\mathrm{af}}}$ on ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ is defined by means of root operators (see ), we conclude that $\pi \otimes \eta$ and $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ above are in the same connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. This proves the lemma.
Proof of Theorem \[thm:SMT\]. {#subsec:prf-SMT}
-----------------------------
Recall that $\lambda,\,\mu \in P^{+}$, and ${J}=\bigl\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$, ${K}=\bigl\{ i \in I \mid {\langle \mu,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$.
\[prop:SMT1\] The set ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu) \sqcup \{{\bm{0}}\}$ is stable under the action of the root operators $e_{i}$, $f_{i}$, $i \in I_{{\mathrm{af}}}$, on ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$.
We give a proof of the assertion only for $e_{i}$, $i \in I_{{\mathrm{af}}}$; the proof for $f_{i}$, $i \in I_{{\mathrm{af}}}$, is similar. Let $\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$, and $i \in I_{{\mathrm{af}}}$. We may assume that $e_{i}(\pi \otimes \eta) \ne {\bm{0}}$. Then it follows from the tensor product rule for crystals that $$e_{i}(\pi \otimes \eta) =
\begin{cases}
(e_{i}\pi) \otimes \eta & \text{if ${\varphi}_{i}(\pi) \ge {\varepsilon}_{i}(\eta)$}, \\[1.5mm]
\pi \otimes (e_{i}\eta) & \text{if ${\varphi}_{i}(\pi) < {\varepsilon}_{i}(\eta)$};
\end{cases}
$$ recall from Remark \[rem:SLS\] that $$\label{eq:vevp}
{\varepsilon}_{i}(\eta) = - m^{\eta}_{i} \quad \text{and} \quad
{\varphi}_{i}(\pi) = H^{\pi}_{i}(1)-m^{\pi}_{i}.$$ Let $x,y \in W_{{\mathrm{af}}}$ be such that $x {\succeq}y$ and ${\Pi^{J}}(x) = \kappa(\pi)$, ${\Pi^{K}}(y) = \iota(\eta)$ (see ); we write $x$ and $y$ as: $$\label{eq:xy}
\begin{cases}
x=\kappa(\pi)x_1 & \text{with $x_1 \in ({W_{J}})_{{\mathrm{af}}}$}, \\[1mm]
y=\iota(\eta)y_1 & \text{with $y_1 \in ({W_{K}})_{{\mathrm{af}}}$}.
\end{cases}
$$
#### Case 1.
Assume that ${\varphi}_{i}(\pi) \ge {\varepsilon}_{i}(\eta)$, i.e., $e_{i}(\pi \otimes \eta) = (e_{i}\pi) \otimes \eta$. Note that $\kappa(e_{i}\pi)$ is equal either to $\kappa(\pi)$ or to $s_{i}\kappa(\pi)$ by the definition of the root operator $e_{i}$. If $\kappa(e_{i}\pi)=\kappa(\pi)$, then there is nothing to prove. Hence we may assume that $\kappa(e_{i}\pi) = s_{i} \kappa(\pi)$. Then we deduce from the definition of the root operator $e_{i}$ that the point $t_{1} = \min \bigl\{ t \in [0,1] \mid H^{\pi}_{i}(t)=m^{\pi}_{i} \bigr\}$ is equal to $1$, and hence $$\label{eq:case1a}
H^{\pi}_{i}(1) = m^{\pi}_{i} \quad \text{and} \quad
{\langle \kappa(\pi)\lambda,\,\alpha_{i}^{\vee} \rangle} < 0.$$ By , the equality in , and our assumption that ${\varphi}_{i}(\pi) \ge {\varepsilon}_{i}(\eta)$, we see that $m^{\eta}_{i} \ge 0$, and hence $m^{\eta}_{i}=0$; in particular, we obtain ${\langle \iota(\eta)\mu,\,\alpha_{i}^{\vee} \rangle} \ge 0$. In addition, it follows from that $\kappa(\pi)^{-1}\alpha_{i} \in
- \bigl(\Delta^{+} \setminus {\Delta_{J}}^{+}\bigr) + {\mathbb{Z}}\delta$. Since $x_{1} \in ({W_{J}})_{{\mathrm{af}}}$ by , we have $$\label{eq:x-}
x^{-1}\alpha_{i}=
x_1^{-1}\kappa(\pi)^{-1}\alpha_{i} \in
- \bigl(\Delta^{+} \setminus {\Delta_{J}}^{+}\bigr) + {\mathbb{Z}}\delta \subset -\Delta^{+} + {\mathbb{Z}}\delta.$$ Also, since $s_{i}\kappa(\pi) = \kappa(e_{i}\pi) \in ({W^{J}})_{{\mathrm{af}}}$ and $x_{1} \in ({W_{J}})_{{\mathrm{af}}}$, we have $${\Pi^{J}}(s_{i}x) =
{\Pi^{J}}(s_{i}\kappa(\pi)x_{1}) =
s_{i}\kappa(\pi) = \kappa(e_{i}\pi).$$ If $y^{-1}\alpha_{i} \in \Delta^{+} + {\mathbb{Z}}\delta$, then we see from Lemma \[lem:dmd\](2) (applied to the case $J=\emptyset$) and that $s_{i}x {\succeq}y$ in $W_{{\mathrm{af}}}$. Therefore, $s_{i}x,\,y \in W_{{\mathrm{af}}}$ satisfy condition for $e_{i}(\pi \otimes \eta) = (e_{i}\pi) \otimes \eta$. If $y^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$, then we see from Lemma \[lem:dmd\](3) (applied to the case $J=\emptyset$) and that $s_{i}x {\succeq}s_{i}y$ in $W_{{\mathrm{af}}}$. We now claim that ${\Pi^{K}}(s_{i}y)=\iota(\eta)$. Indeed, since ${\langle \iota(\eta)\mu,\,\alpha_{i}^{\vee} \rangle} \ge 0$ as seen above, we have $\iota(\eta)^{-1}\alpha_{i} \in
\bigl(\Delta^{+} \sqcup (-{\Delta_{K}}^{+})\bigr)+{\mathbb{Z}}\delta$. In addition, since $y_{1} \in ({W_{K}})_{{\mathrm{af}}}$, we deduce that $y^{-1}\alpha_{i} = y_{1}^{-1}\iota(\eta)^{-1}\alpha_{i}$ is contained in $\bigl(\Delta^{+} \sqcup (-{\Delta_{K}}^{+})\bigr)+{\mathbb{Z}}\delta$. However, since $y^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$ by our assumption, we have $y^{-1}\alpha_{i} \in -\Delta_{K}^{+}+{\mathbb{Z}}\delta$, which implies that $s_{y^{-1}\alpha_{i}} \in ({W_{K}})_{{\mathrm{af}}}$. Therefore, we obtain ${\Pi^{K}}(s_{i}y) =
{\Pi^{K}}(ys_{y^{-1}\alpha_{i}}) =
{\Pi^{K}}(y) = \iota(\eta)$, as desired. Thus, $s_{i}x,\,s_{i}y \in W_{{\mathrm{af}}}$ satisfy condition for $e_{i}(\pi \otimes \eta) = (e_{i}\pi) \otimes \eta$.
#### Case 2.
Assume that ${\varphi}_{i}(\pi) < {\varepsilon}_{i}(\eta)$, i.e., $e_{i}(\pi \otimes \eta) = \pi \otimes (e_{i}\eta)$. If $\iota(e_{i}\eta)=\iota(\eta)$, then there is nothing to prove. Hence we may assume that $\iota(e_{i}\eta) = s_{i} \iota(\eta)$. Then we deduce from the definition of the root operator $e_{i}$ that ${\langle \iota(\eta)\mu,\,\alpha_{i}^{\vee} \rangle} < 0$; observe that with notation in and , $t_{0}=0$, and $H^{\eta}_{i}(t)$ is strictly decreasing on $[t_{0},\,t_{1}] = [0,\,t_{1}]$. Thus we obtain $\iota(\eta)^{-1}\alpha_{i} \in
- \bigl(\Delta^{+} \setminus {\Delta_{K}}^{+}\bigr) + {\mathbb{Z}}\delta$. In addition, since $y_{1} \in ({W_{K}})_{{\mathrm{af}}}$ (see ), we deduce that $$y^{-1}\alpha_{i} = y_1^{-1}\iota(\eta)^{-1}\alpha_{i} \in
- \bigl(\Delta^{+} \setminus {\Delta_{K}}^{+}\bigr) + {\mathbb{Z}}\delta \subset
-\Delta^{+} + {\mathbb{Z}}\delta,$$ which implies that $y {\succ}s_{i}y$ by Lemma \[lem:si\]. Since $x {\succeq}y$, we get $x {\succ}s_{i}y$. Also, since $s_{i}\iota(\eta) = \iota(e_{i}\eta) \in ({W^{K}})_{{\mathrm{af}}}$ and $y_{1} \in ({W_{K}})_{{\mathrm{af}}}$, we have $${\Pi^{K}}(s_{i}y) =
{\Pi^{K}}(s_{i}\iota(\eta)y_{1}) =
s_{i}\iota(\eta) = \iota(e_{i}\eta).$$ Thus, $x,\,s_{i}y \in W_{{\mathrm{af}}}$ satisfy condition for $e_{i}(\pi \otimes \eta) = \pi \otimes (e_{i}\eta)$.
This completes the proof of the proposition.
By this proposition, ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ is a subcrystal of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. Hence our remaining task is to prove that ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu) \cong {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu)$ as crystals. Recall from Lemma \[lem:conn\] that each connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ contains an element of the form: $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ for some $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$, ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, and ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$.
\[prop:tx\] Let ${\bm{\rho}}=(\rho^{(i)}) \in {\mathop{\rm Par}\nolimits}(\lambda)$, ${\bm{\chi}}=(\chi^{(i)}) \in {\mathop{\rm Par}\nolimits}(\mu)$, and $\xi=\sum c_{i}\alpha_{i}^{\vee}
\in Q^{\vee}_{I \setminus ({J}\cup {K})}$. Then, $$\label{eq:tx1}
(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$$ if and only if $$\label{eq:tx2}
c_{i} \ge \chi^{(i)}_{1} \quad \text{\rm for all
$i \in I \setminus ({J}\cup {K})$}.$$
We first show the “only if” part; assume that holds. We see that $$\begin{aligned}
& \kappa(t_{\xi} \cdot \pi_{{\bm{\rho}}}) = {\Pi^{J}}(t_{\xi}) \quad \text{by \eqref{eq:tx0}}; \\[1mm]
& \iota(\pi_{{\bm{\chi}}}) = {\Pi^{K}}(t_{\zeta_{1}}), \quad
\text{with} \quad \zeta_{1} = \sum_{i \in I}
\chi^{(i)}_{1} \alpha_{i}^{\vee} \in Q^{\vee}_{I \setminus {K}} \quad \text{by \eqref{eq:par0}}. \end{aligned}$$ Since holds, there exist $x,\,y \in W_{{\mathrm{af}}}$ such that $x {\succeq}y$ in $W_{{\mathrm{af}}}$, and such that $$\begin{cases}
{\Pi^{J}}(x) =
\kappa(t_{\xi} \cdot \pi_{{\bm{\rho}}}) ={\Pi^{J}}(t_{\xi}), & \\[1mm]
{\Pi^{K}}(y) =
\iota(\pi_{{\bm{\chi}}}) = {\Pi^{K}}(t_{\zeta_{1}});
\end{cases}
$$ we write $x$ and $y$ as: $$\begin{cases}
x={\Pi^{J}}(t_{\xi})x_{1}
& \text{with $x_{1} \in ({W_{J}})_{{\mathrm{af}}}$}, \\[1mm]
y={\Pi^{K}}(t_{\zeta_{1}})y_{1}
& \text{with $y_{1} \in ({W_{K}})_{{\mathrm{af}}}$}.
\end{cases}
$$ Because $x \in W_{{\mathrm{af}}}$ is a lift of ${\Pi^{J}}(t_{\xi}) \in ({W^{J}})_{{\mathrm{af}}}$, it follows from Lemma \[lem:lift\] that $x = v t_{\xi+\gamma}$ for some $v \in {W_{J}}$ and $\gamma \in {Q_{J}^{\vee}}$. Similarly, we have $y = v't_{\zeta_1+\gamma'}$ for some $v' \in {W_{K}}$ and $\gamma' \in {Q_{K}^{\vee}}$. Because $x {\succeq}y$ in $W_{{\mathrm{af}}}$, we deduce from Lemma \[lem:SiB\](1) (applied to the case $J=\emptyset$) that $\xi+\gamma \ge \zeta_1+\gamma'$, which implies since $\gamma,\,\gamma' \in Q^{\vee}_{{J}\cup {K}}$.
We next show the “if” part; assume that holds. Recall that $$\begin{aligned}
& \kappa(t_{\xi} \cdot \pi_{{\bm{\rho}}}) = {\Pi^{J}}(t_{\xi}) \quad \text{by \eqref{eq:tx0}}; \\[1mm]
& \iota(\pi_{{\bm{\chi}}}) = {\Pi^{K}}(t_{\zeta_{1}}), \quad
\text{with} \quad \zeta_{1} = \sum_{i \in I}
\chi^{(i)}_{1} \alpha_{i}^{\vee} \in Q^{\vee}_{I \setminus {K}} \quad \text{by \eqref{eq:par0}}. \end{aligned}$$ We set $\gamma:=\sum_{i \in {J}\setminus {K}}
\chi^{(i)}_{1} \alpha_{i}^{\vee} \in {Q_{J}^{\vee}}$. Then it follows from that $\xi+\gamma \ge \zeta_{1}$ since $I \setminus K = (I \setminus (J \cup K)) \sqcup (J \setminus K)$. Hence we deduce from Lemma \[lem:SiB\](2) that $x:=t_{\xi+\gamma} {\succeq}t_{\zeta_1}=:y$ in $W_{{\mathrm{af}}}$. It is obvious that ${\Pi^{K}}(y) = \iota(\pi_{{\bm{\chi}}})$. Also, since $\gamma \in {Q_{J}^{\vee}}$, we see from that ${\Pi^{J}}(x) = {\Pi^{J}}(t_{\xi}) = \kappa(t_{\xi} \cdot \pi_{{\bm{\rho}}})$. Thus, $x$ and $y$ satisfy condition for $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$, which implies . This proves the proposition.
\[prop:conn\] Each connected component of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ contains a unique element of the form[:]{} $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ for some ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$, and $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying condition in Proposition \[prop:tx\]. Therefore, there exists a one-to-one correspondence between the set ${\mathop{\rm Conn}\nolimits}({\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu))$ of connected components of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ and the set of triples $({\bm{\rho}},\,{\bm{\chi}},\,\xi) \in {\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times
Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying condition in Proposition \[prop:tx\].
The “existence” part follows from Lemma \[lem:conn\] and Proposition \[prop:tx\]. Hence it suffices to prove the “uniqueness” part. Let $({\bm{\rho}},\,{\bm{\chi}},\,\xi)$ and $({\bm{\rho}}',\,{\bm{\chi}}',\,\xi')$ be elements in ${\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times
Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying condition in Proposition \[prop:tx\], and suppose that $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ and $(t_{\xi'} \cdot \pi_{{\bm{\rho}}'}) \otimes \pi_{{\bm{\chi}}'}$ are contained in the same connected component of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$. Then there exists a monomial $X$ in root operators such that $X ( (t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} ) =
(t_{\xi'} \cdot \pi_{{\bm{\rho}}'}) \otimes \pi_{{\bm{\chi}}'}$. By the tensor product rule for crystals, we see that $X ( (t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} )
= X_{1} (t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes X_{2}\pi_{{\bm{\chi}}}$ for some monomials $X_{1}$, $X_{2}$ in root operators. Then we have $X_{1} (t_{\xi} \cdot \pi_{{\bm{\rho}}}) = t_{\xi'} \cdot \pi_{{\bm{\rho}}'}$, which implies that $t_{\xi} \cdot \pi_{{\bm{\rho}}}$ and $t_{\xi'} \cdot \pi_{{\bm{\rho}}'}$ are contained in the same connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$, and hence so are $\pi_{{\bm{\rho}}}$ and $\pi_{{\bm{\rho}}'}$. Therefore, by the uniqueness of an element of the form in a connected component of ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ (see Section \[subsec:conn\]), we deduce that ${\bm{\rho}}= {\bm{\rho}}'$. Similarly, we obtain ${\bm{\chi}}={\bm{\chi}}'$. Suppose, for a contradiction, that $\xi \ne \xi'$; we may assume that for some $k \in I \setminus ({J}\cup {K})$, the coefficient of $\alpha_{k}^{\vee}$ in $\xi$ is greater than that in $\xi'$, i.e., the coefficient of $\alpha_{k}^{\vee}$ in $\xi'-\xi$ is a negative integer. Because $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ and $(t_{\xi'} \cdot \pi_{{\bm{\rho}}'}) \otimes \pi_{{\bm{\chi}}'} =
(t_{\xi'} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ are contained in the same connected component, there exists a monomial $Y$ in root operators such that $Y \bigl((t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}\bigr)
= (t_{\xi'} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}
= (t_{\xi+(\xi'-\xi)} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$. Here, the same argument as in the proof of [@INS Lemma 7.1.4] (or, as in the proof of [@INS Proposition 7.1.2]) shows that $$Y^{N} \Bigl((t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}\Bigr)
= (t_{\xi+N(\xi'-\xi)} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}
\quad \text{for all $N \in {\mathbb{Z}}_{\ge 1}$}.$$ Since this element is contained in ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ for all $N \in {\mathbb{Z}}_{\ge 1}$ by Proposition \[prop:SMT1\], it follows from Proposition \[prop:tx\] that the coefficient of $\alpha_{k}^{\vee}$ in $\xi+N(\xi'-\xi)$ is greater than or equal to $\chi^{(k)}_{1}$ for all $N \ge 1$. This contradicts the fact that the coefficient of $\alpha_{k}^{\vee}$ in $\xi'-\xi$ is a negative integer. This proves the proposition.
Now, we write $\lambda$ and $\mu$ as: $\lambda=\sum_{i \in I} m_{i} {\varpi}_{i}$ and $\mu=\sum_{i \in I} n_{i} {\varpi}_{i}$. For $({\bm{\rho}},\,{\bm{\chi}},\,\xi) \in {\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times
Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying , define ${\bm{\omega}}= ({\omega}^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda+\mu)$ as follows. Write ${\bm{\rho}}$, ${\bm{\chi}}$, and $\xi$ as: $$\begin{aligned}
& {\bm{\rho}}= (\rho^{(i)})_{i \in I}, \quad \text{with} \quad
\rho^{(i)}=(\rho^{(i)}_{1} \ge \cdots \ge \rho^{(i)}_{m_i-1} \ge 0) \quad
\text{for $i \in I$}, \\
& {\bm{\chi}}= (\chi^{(i)})_{i \in I}, \quad \text{with} \quad
\chi^{(i)}=(\chi^{(i)}_{1} \ge \cdots \ge \chi^{(i)}_{n_i-1} \ge 0) \quad
\text{for $i \in I$}, \\
& \xi=\sum_{i \in I \setminus ({J}\cup {K})}
c_{i}\alpha_{i}^{\vee}; \quad \text{recall that $c_{i} \ge \chi^{(i)}_{1}$
for all $i \in I \setminus ({J}\cup {K})$}.\end{aligned}$$ Let $i \in I$.
- If $i \in {J}\cap {K}$ (note that $m_{i}=n_{i}=0$), we set ${\omega}^{(i)} := \emptyset$;
- if $i \in {J}\setminus {K}$ (note that $m_{i}=0$), we set ${\omega}^{(i)}:=\chi^{(i)}$, which is a partition of length less than $n_{i} = 0+n_{i}= m_{i}+n_{i}$;
- if $i \in {K}\setminus {J}$ (note that $n_{i}=0$), we set ${\omega}^{(i)}:=\rho^{(i)}$, which is a partition of length less than $m_{i} =m_{i}+0= m_{i}+n_{i}$;
- if $i \in I \setminus ({J}\cup {K})$, we set $${\omega}^{(i)}=
(\rho^{(i)}_{1}+c_{i} \ge \cdots \ge \rho^{(i)}_{m_i-1}+c_{i} \ge c_{i} \ge
\chi^{(i)}_{1} \ge \cdots \ge \chi^{(i)}_{n_i-1}),$$ which is a partition of length less than $(m_{i}-1)+1+(n_{i}-1) + 1=m_{i}+n_{i}$; note that $|{\omega}^{(i)}| = | \rho^{(i)} | + |\chi^{(i)} | + m_{i}c_{i}$.
It follows that ${\bm{\omega}}=({\omega}^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\lambda+\mu)$. Thus we obtain a map $\Theta$ from the set of those $({\bm{\rho}},\,{\bm{\chi}},\,\xi) \in {\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times
Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying to the set ${\mathop{\rm Par}\nolimits}(\lambda+\mu)$; we can easily deduce that the map $\Theta$ is bijective. Also, by direct calculation, we have $$\begin{aligned}
{\mathop{\rm wt}\nolimits}\Bigl( \underbrace{(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}}_{\text{satisfying \eqref{eq:tx2}}} \Bigr)
& = {\mathop{\rm wt}\nolimits}(t_{\xi} \cdot \pi_{{\bm{\rho}}}) + {\mathop{\rm wt}\nolimits}(\pi_{{\bm{\chi}}})
= t_{\xi}\bigl(\lambda-|{\bm{\rho}}|\delta\bigr) + \bigl(\mu-|{\bm{\chi}}|\delta\bigr) \\
& = (\lambda+\mu) - \bigl( |{\bm{\rho}}| + {\langle \lambda,\,\xi \rangle} + |{\bm{\chi}}| \bigr)\delta \\[2mm]
& = (\lambda+\mu) -
\left( |{\bm{\rho}}| + \sum_{i \in I \setminus (J \cup K)} m_{i}c_{i} + |{\bm{\chi}}| \right)\delta \\[2mm]
& = (\lambda+\mu) - \bigl| \Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi) \bigr|\delta.\end{aligned}$$
We claim that the connected component of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ containing $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ is isomorphic, as a crystal, to $\bigl\{\Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi)\bigr\} \otimes {\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu)$, where ${\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu)$ denotes the connected component of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ containing $\pi_{\lambda} \otimes \pi_{\mu} =
(e\,;\,0,1) \otimes (e\,;\,0,1) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. Indeed, let us consider the composite of the following bijections: $$\begin{aligned}
& {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\chi}}}(\mu)
\stackrel{T_{-\xi} \otimes {\mathop{\rm id}\nolimits}}{\longrightarrow}
{\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\rho}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\bm{\chi}}}(\mu)
\quad \text{(see \eqref{eq:Tx} and Remark~\ref{rem:Tx})} \\
& \quad \stackrel{\sim}{\rightarrow}
\bigl(\bigl\{{\bm{\rho}}\bigr\} \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)\bigr) \otimes
\bigl(\bigl\{{\bm{\chi}}\bigr\} \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\mu)\bigr) \quad
\text{(see the comment preceding \eqref{eq:isom})} \\
& \quad \stackrel{\sim}{\rightarrow}
\bigl(\bigl\{{\bm{\rho}}\bigr\} \otimes \bigl\{{\bm{\chi}}\bigr\}\bigr) \otimes
\bigl({\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\mu)\bigr) \quad
\text{(by the tensor product rule for crystals)} \\
& \quad \stackrel{\sim}{\rightarrow}
\bigl\{\Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi)\bigr\} \otimes
\bigl({\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\mu)\bigr),\end{aligned}$$ where the last map sends $({\bm{\rho}}\otimes {\bm{\chi}}) \otimes (\pi \otimes \eta)$ to $\Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi) \otimes (\pi \otimes \eta)$ for each $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda)$ and $\eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\mu)$. We deduce by and the tensor product rule for crystals that the composite of these bijections is an isomorphism of crystals, which sends $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ to $\Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi) \otimes (\pi_{\lambda} \otimes \pi_{\mu})$. Therefore, the connected component of ${\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ containing $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}}$ is mapped to $\bigl\{\Theta({\bm{\rho}},\,{\bm{\chi}},\,\xi)\bigr\} \otimes {\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu)$ under this isomorphism of crystals. It follows from Proposition \[prop:conn\] and the bijectivity of $\Theta$ that $$\label{eq:conn}
{\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu) \cong \bigsqcup_{{\bm{\omega}}\in {\mathop{\rm Par}\nolimits}(\lambda+\mu)}
\bigl\{{\bm{\omega}}\bigr\} \otimes {\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu).$$
\[prop:BN\] As crystals, ${\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu) \cong {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda+\mu)$.
Write $\lambda$ and $\mu$ as: $\lambda = \sum_{i \in I} m_{i}{\varpi}_{i}$ with $m_{i} \in {\mathbb{Z}}_{\ge 0}$, and $\mu = \sum_{i \in I} n_{i}{\varpi}_{i}$ with $n_{i} \in {\mathbb{Z}}_{\ge 0}$, respectively. We know from [@Kas02 Conjecture 13.1(iii)], which is proved in [@BN Remark 4.17], that there exists an isomorphism ${\mathcal{B}}(\lambda+\mu) \stackrel{\sim}{\rightarrow}
\bigotimes_{i \in I} {\mathcal{B}}((m_{i}+n_{i}){\varpi}_{i})$ of crystals, which maps $u_{\lambda+\mu}$ to $\bigotimes_{i \in I} u_{(m_{i}+n_{i}){\varpi}_{i}}$; the restriction of this isomorphism to ${\mathcal{B}}_{0}(\lambda+\mu) \subset {\mathcal{B}}(\lambda+\mu)$ gives an embedding ${\mathcal{B}}_{0}(\lambda+\mu) \hookrightarrow
\bigotimes_{i \in I} {\mathcal{B}}_{0}((m_{i}+n_{i}){\varpi}_{i})$ of crystals. Also, we know from [@Kas02 Conjecture 13.2(iii)], which is proved in [@BN Remark 4.17], that for each $i \in I$, there exists an embedding ${\mathcal{B}}_{0}((m_{i}+n_{i}){\varpi}_{i}) \hookrightarrow
{\mathcal{B}}({\varpi}_{i})^{\otimes (m_{i}+n_{i})}$ of crystals, which maps $u_{(m_{i}+n_{i}){\varpi}_{i}}$ to $u_{{\varpi}_{i}}^{\otimes (m_{i}+n_{i})}$; recall from [@Kas02 Proposition 5.4] that ${\mathcal{B}}({\varpi}_{i})$ is connected. Thus we obtain an embedding $${\mathcal{B}}_{0}(\lambda+\mu) \hookrightarrow
\bigotimes_{i \in I} {\mathcal{B}}_{0}((m_{i}+n_{i}){\varpi}_{i})
\hookrightarrow
\bigotimes_{i \in I} {\mathcal{B}}({\varpi}_{i})^{\otimes (m_{i}+n_{i})}$$ of crystals, which maps $u_{\lambda+\mu}$ to $\bigotimes_{i \in I} u_{{\varpi}_{i}}^{\otimes (m_{i}+n_{i})}$. Here, we recall from [@Kas02 Sect. 10] that for each $j,\,k \in I$, there exists an isomorphism ${\mathcal{B}}({\varpi}_{j}) \otimes {\mathcal{B}}({\varpi}_{k})
\stackrel{\sim}{\rightarrow}
{\mathcal{B}}({\varpi}_{k}) \otimes {\mathcal{B}}({\varpi}_{j})$ of crystals, which maps $u_{{\varpi}_{j}} \otimes u_{{\varpi}_{k}}$ to $u_{{\varpi}_{k}} \otimes u_{{\varpi}_{j}}$. Hence we obtain an isomorphism of crystals $$\bigotimes_{i \in I} {\mathcal{B}}({\varpi}_{i})^{\otimes (m_{i}+n_{i})}
\stackrel{\sim}{\rightarrow}
\left(\bigotimes_{i \in I} {\mathcal{B}}({\varpi}_{i})^{\otimes m_{i}}\right)
\otimes
\left(\bigotimes_{i \in I} {\mathcal{B}}({\varpi}_{i})^{\otimes n_{i}}\right) = : {\mathcal{B}},$$ which maps $\bigotimes_{i \in I} u_{{\varpi}_{i}}^{\otimes (m_{i}+n_{i})}$ to $\left(\bigotimes_{i \in I} u_{{\varpi}_{i}}^{\otimes m_{i}}\right) \otimes
\left(\bigotimes_{i \in I} u_{{\varpi}_{i}}^{\otimes n_{i}}\right)=:b$. From these, we obtain an embedding ${\mathcal{B}}_{0}(\lambda+\mu) \hookrightarrow {\mathcal{B}}$ of crystals, which maps $u_{\lambda+\mu}$ to $b$. Similarly, we obtain an embedding ${\mathcal{B}}_{0}(\lambda) \otimes {\mathcal{B}}_{0}(\mu) \hookrightarrow {\mathcal{B}}$ of crystals, which maps $u_{\lambda} \otimes u_{\mu}$ to $b$. Consequently, there exists an isomorphism of crystals from ${\mathcal{B}}_{0}(\lambda+\mu)$ to the connected component (denoted by ${\mathcal{S}}_{0}(\lambda+\mu)$) of ${\mathcal{B}}_{0}(\lambda) \otimes {\mathcal{B}}_{0}(\mu)$ containing $u_{\lambda} \otimes u_{\mu}$, which maps $u_{\lambda+\mu}$ to $u_{\lambda} \otimes u_{\mu}$. Now, by Theorem \[thm:isom\], we have an isomorphism ${\mathcal{B}}_{0}(\lambda+\mu) \stackrel{\sim}{\rightarrow} {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda+\mu)$ of crystals, which maps $u_{\lambda+\mu}$ to $\pi_{\lambda+\mu}$. In addition, we have an isomorphism ${\mathcal{B}}_{0}(\lambda) \otimes {\mathcal{B}}_{0}(\mu) \stackrel{\sim}{\rightarrow}
{\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\mu)$ of crystals, which maps $u_{\lambda} \otimes u_{\mu}$ to $\pi_{\lambda} \otimes \pi_{\mu}$; by restriction, we obtain an isomorphism of crystals from ${\mathcal{S}}_{0}(\lambda+\mu)$ to ${\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu)$. Summarizing, we obtain the following isomorphism of crystals: $$\begin{array}{ccccccc}
{\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda+\mu) & \stackrel{\sim}{\rightarrow} &
{\mathcal{B}}_{0}(\lambda+\mu) & \stackrel{\sim}{\rightarrow} &
{\mathcal{S}}_{0}(\lambda+\mu) & \stackrel{\sim}{\rightarrow} &
{\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu), \\[3mm]
\pi_{\lambda+\mu} & \mapsto &
u_{\lambda+\mu} & \mapsto &
u_{\lambda} \otimes u_{\mu} & \mapsto &
\pi_{\lambda} \otimes \pi_{\mu}.
\end{array}
$$ This proves the proposition.
By using , Proposition \[prop:BN\], and (with $\lambda$ replaced by $\lambda+\mu$), we conclude that $$\begin{aligned}
{\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu) & \cong
\bigsqcup_{{\bm{\omega}}\in {\mathop{\rm Par}\nolimits}(\lambda+\mu)}
\bigl\{{\bm{\omega}}\bigr\} \otimes {\mathbb{S}^{\frac{\infty}{2}}}_{0}(\lambda+\mu) \\[3mm]
& \cong
\bigsqcup_{{\bm{\omega}}\in {\mathop{\rm Par}\nolimits}(\lambda+\mu)}
\bigl\{{\bm{\omega}}\bigr\} \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}_{0}(\lambda+\mu)
\cong {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu)\end{aligned}$$ as crystals. This completes the proof of Theorem \[thm:SMT\].
\[cor:Sext\] For each ${\bm{\omega}}\in {\mathop{\rm Par}\nolimits}(\lambda+\mu)$, the element $\pi_{{\bm{\omega}}} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu)$ is mapped to $(t_{\xi} \cdot \pi_{{\bm{\rho}}}) \otimes \pi_{{\bm{\chi}}} \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ for some $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$ and ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$ satisfying under the isomorphism ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu) \cong {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ of crystals in Theorem \[thm:SMT\].
Proof of Propositions \[prop:DC\] and \[prop:DC2\]. {#sec:prf-DC}
===================================================
Recall that $\lambda,\,\mu \in P^{+}$, and that ${J}= \bigl\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$, ${K}= \bigl\{ i \in I \mid {\langle \mu,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$.
Proof of Proposition \[prop:DC\]. {#subsec:prf-DC}
---------------------------------
The “if” part is obvious from the definition of defining chains and condition . Let us prove the “only if” part. Assume that $\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$, and write $\pi$ and $\eta$ as: $\pi=(x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta=(y_{1},\,\dots,\,y_{p}\,;\,{\mathbf{b}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$, respectively. It follows from that there exist $x_{s}',\,y_{1}' \in W_{{\mathrm{af}}}$ such that $x_{s}' {\succeq}y_{1}'$ in $W_{{\mathrm{af}}}$, and such that ${\Pi^{J}}(x_{s}')=x_{s}$, ${\Pi^{K}}(y_{1}')=y_{1}$; we write $x_{s}'=x_{s}z_{1}$ for some $z_{1} \in ({W_{J}})_{{\mathrm{af}}}$, and $y_{1}'=y_{1}z_{2}$ for some $z_{2} \in ({W_{K}})_{{\mathrm{af}}}$. Now we set $$\begin{cases}
x_{u}':=x_{u}z_{1} & \text{for $1 \le u \le s$}, \\[1.5mm]
y_{q}':=y_{q}z_{2} & \text{for $1 \le q \le p$}.
\end{cases}
$$ Because $x_{1} {\succeq}x_{2} {\succeq}\cdots {\succeq}x_{s}$ in $({W^{J}})_{{\mathrm{af}}}$ by the definition of semi-infinite LS paths, it follows from Lemma \[lem:SiL\] that $x_{1}' {\succeq}x_{2}' {\succeq}\cdots {\succeq}x_{s}'$ in $W_{{\mathrm{af}}}$. Similarly, we see that $y_{1}' {\succeq}y_{2}' {\succeq}\cdots {\succeq}y_{p}'$ in $W_{{\mathrm{af}}}$. Combining these inequalities with the inequality $x_{s}' {\succeq}y_{1}'$, we obtain $x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{p}'$ in $W_{{\mathrm{af}}}$. Since ${\Pi^{J}}(x_{u}')=x_{u}$ for all $1 \le u \le s$, and ${\Pi^{J}}(y_{q}')=y_{q}$ for all $1 \le q \le p$, the sequence $x_{1}',\,\dots,\,x_{s}',\,y_{1}',\,\dots,\,y_{p}'$ is a defining chain for $\pi \otimes \eta$. This completes the proof of Proposition \[prop:DC\].
Proof of Proposition \[prop:DC2\]. {#subsec:prf-DC2}
----------------------------------
We write $\pi$ and $\eta$ as: $\pi=(x_{1},\,\dots,\,x_{s}\,;\,\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta=(y_{1},\,\dots,\,y_{p}\,;\,\,{\mathbf{b}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$, respectively. First, we prove the “only if” part. Take an (arbitrary) defining chain $x_{1}',\,\dots,\,x_{s}',\,y_{1}',\,\dots,\,y_{p}' \in W_{{\mathrm{af}}}$ for $\pi \otimes \eta$. It suffices to show the following claim.
\[c:DC2-1\] Let $x \in W_{{\mathrm{af}}}$ be such that $y_{p}' {\succeq}x$[;]{} note that $\kappa(\eta)=y_{p}={\Pi^{K}}(y_{p}') {\succeq}{\Pi^{K}}(x)$ by Lemma \[lem:611\]. Then, $\kappa(\pi) {\succeq}{\Pi^{J}}({\iota(\eta,\,x)})$.
[*Proof of Claim \[c:DC2-1\].*]{} For the given $x \in W_{{\mathrm{af}}}$, we define ${\widetilde{y}}_{p},\,{\widetilde{y}}_{p-1},\,\dots,\,{\widetilde{y}}_{1} = {\iota(\eta,\,x)}$ as in . We show by descending induction on $q$ that $$\label{eq:DC2-1}
y_{q}' {\succeq}{\widetilde{y}}_{q} \quad \text{for all $1 \le q \le p$}.$$ Because $y_{p}' {\succeq}x$ and ${\Pi^{K}}(y_{p}')=y_{p}$, it follows that $y_{p}' \in {\mathrm{Lift}_{{\succeq}x}(y_{p})}$, and hence $y_{p}' {\succeq}\min {\mathrm{Lift}_{{\succeq}x}(y_{p})} = {\widetilde{y}}_{p}$. Assume that $q < p$. Since $y_{q}' {\succeq}y_{q+1}'$ by the definition of defining chains, and since $y_{q+1}' {\succeq}{\widetilde{y}}_{q+1}$ by our induction hypothesis, we obtain $y_{q}' {\succeq}{\widetilde{y}}_{q+1}$. In addition, we have ${\Pi^{K}}(y_{q}')=y_{q}$. From these, we deduce that $y_{q}' \in {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{q+1}}(y_{q})}$, and hence $y_{q}' {\succeq}\min {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{q+1}}(y_{q})}
= {\widetilde{y}}_{q}$. Thus we have shown . Hence we have $x_{s}' {\succeq}y_{1}' {\succeq}{\widetilde{y}}_{1} = {\iota(\eta,\,x)}$ by the assumption. Therefore, it follows from Lemma \[lem:611\] that $$\label{eq:kappa-Deo}
\kappa(\pi) = x_{s} = {\Pi^{J}}(x_{s}') {\succeq}{\Pi^{J}}({\iota(\eta,\,x)}).$$ This proves Claim \[c:DC2-1\].
Next, we prove the “if” part. We define ${\widetilde{y}}_{p}$, ${\widetilde{y}}_{p-1}$, $\dots$, ${\widetilde{y}}_{1} = {\iota(\eta,\,x)}$ as in . By the definitions, we have $$\label{eq:DC2-2}
\begin{cases}
{\iota(\eta,\,x)}
= {\widetilde{y}}_{1} {\succeq}{\widetilde{y}}_{2} {\succeq}\cdots {\succeq}{\widetilde{y}}_{p} {\succeq}x, & \\[1mm]
{\Pi^{K}}({\widetilde{y}}_{q}) = y_{q}
\quad \text{for $1 \le q \le p$}.
\end{cases}
$$ Write ${\iota(\eta,\,x)} \in W_{{\mathrm{af}}}$ as: ${\iota(\eta,\,x)} = {\Pi^{J}}({\iota(\eta,\,x)})z$ with $z \in ({W_{J}})_{{\mathrm{af}}}$. Since $\kappa(\pi) {\succeq}{\Pi^{J}}({\iota(\eta,\,x)})$ by the assumption, we deduce from Lemma \[lem:SiL\] that $x_{s}':=\kappa(\pi)z {\succeq}{\Pi^{J}}({\iota(\eta,\,x)})z = {\iota(\eta,\,x)} = {\widetilde{y}}_{1}$. Similarly, if we set $x_{u}':=x_{u}z$ for $1 \le u \le s$, then we have $$\label{eq:DC2-3}
\begin{cases}
x_{1}' {\succeq}x_{2}' {\succeq}\cdots {\succeq}x_{s}' \
({\succeq}{\iota(\eta,\,x)} = {\widetilde{y}}_{1}), & \\[1mm]
{\Pi^{J}}(x_{u}') = x_{u}
\quad \text{for $1 \le u \le s$}.
\end{cases}
$$ Concatenating the sequences in and , we obtain a defining chain $$\label{eq:minDC}
x_{1}' {\succeq}x_{2}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}{\widetilde{y}}_{1} {\succeq}{\widetilde{y}}_{2} {\succeq}\cdots {\succeq}{\widetilde{y}}_{p}$$ for $\pi \otimes \eta \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$. This completes the proof of Proposition \[prop:DC2\].
Proof of Theorem \[thm:Dem\]. {#sec:prf-Dem}
=============================
Recall that $\lambda,\,\mu \in P^{+}$, and that ${J}= \bigl\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$, ${K}= \bigl\{ i \in I \mid {\langle \mu,\,\alpha_{i}^{\vee} \rangle}=0\bigr\}$, and $S = \bigl\{ i \in I \mid {\langle \lambda+\mu,\,\alpha_{i}^{\vee} \rangle}=0\bigr\} = {J}\cap {K}$.
Proof of $\Leftrightarrow$ . {#subsec:23}
----------------------------
We prove the implication $\Rightarrow$ . Let $x_{1}',\,\dots,\,x_{s}',\,y_{1}',\,\dots,\,y_{p}'=:y$ be a defining chain for $\pi \otimes \eta$ such that ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$. Write $x$ as: $x = {\Pi^{S}}(x)z$ for some $z \in ({W_{S}})_{{\mathrm{af}}}$; note that $({W_{S}})_{{\mathrm{af}}} \subset ({W_{J}})_{{\mathrm{af}}} \cap ({W_{K}})_{{\mathrm{af}}}$. We deduce from Lemma \[lem:SiL\] that $${\Pi^{S}}(x_{1}')z,\,\dots,\,{\Pi^{S}}(x_{s}')z,\,
{\Pi^{S}}(y_{1}')z,\,\dots,\,{\Pi^{S}}(y_{p}')={\Pi^{S}}(y)z$$ is also a defining chain for $\pi \otimes \eta$ such that ${\Pi^{S}}({\Pi^{S}}(y)z) = {\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$. Hence we may assume from the beginning that $y {\succeq}x$. We deduce from Lemma \[lem:611\] that $\kappa(\eta) = {\Pi^{K}}(y) {\succeq}{\Pi^{K}}(x)$. Also, the inequality $\kappa(\pi) {\succeq}{\Pi^{J}}({\iota(\eta,\,x)})$ was shown in Claim \[c:DC2-1\] in the proof of Proposition \[prop:DC2\].
The implication $\Rightarrow$ follows from the fact that the defining chain for $\pi \otimes \eta$ satisfies the desired condition in .
Proof of $\Leftrightarrow$ . {#subsec:12}
----------------------------
Let ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$ denote the set of elements in ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \otimes {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ satisfying condition (or equivalently, condition ); by Proposition \[prop:DC\], we see that ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) \subset {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$.
\[lem:Dem1\]
The set ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) \cup \{{\bm{0}}\}$ is stable under the action of the root operator $f_{i}$ for all $i \in I_{{\mathrm{af}}}$.
The set ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) \cup \{{\bm{0}}\}$ is stable under the action of the root operator $e_{i}$ for those $i \in I_{{\mathrm{af}}}$ such that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$.
Let $i \in I_{{\mathrm{af}}}$ be such that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$. Then, $$\label{eq:Dem1-3a}
{\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) = \bigl\{e_{i}^{n}(\pi \otimes \eta) \mid
\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu),\,n \ge 0\bigr\} \setminus \{{\bm{0}}\}.$$
\(1) Let $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$, and let $i \in I_{{\mathrm{af}}}$; we may assume that $f_{i}(\pi \otimes \eta) \ne {\bm{0}}$. We give a proof only for the case that $f_{i}(\pi \otimes \eta) = \pi \otimes f_{i}\eta$; the proof for the case that $f_{i}(\pi \otimes \eta) = f_{i}\pi \otimes \eta$ is similar. We write $\pi$ and $\eta$ as: $\pi=(x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\eta=(y_{1},\,\dots,\,y_{p}\,;\,{\mathbf{b}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$, respectively. Let $x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{p}'$ be a defining chain for $\pi \otimes \eta$ such that ${\Pi^{S}}(y_{p}') {\succeq}{\Pi^{S}}(x)$. Take $0 \le t_{0} < t_{1} \le 1$ as in (with $\pi$ replaced by $\eta$); note that $H^{\eta}_{i}(t)$ is strictly increasing on the interval $[t_{0},\,t_{1}]$. We see from that $f_{i}\eta$ is of the form: $$f_{i} \eta := ( y_{1},\,\ldots,\,y_{k},\,s_{i}y_{k+1},\,\dots,\,
s_{i} y_{m},\,s_{i} y_{m+1},\,y_{m+1},\,\ldots,\,y_{p}\,;\,{\mathbf{b}}')$$ for some $0 \le k \le m \le p-1$ and some increasing sequence ${\mathbf{b}}'$ of rational numbers in $[0,1]$. Here, since $H^{\eta}_{i}(t)$ is strictly increasing on the interval $[t_{0},\,t_{1}]$, it follows that ${\langle y_{n}\mu,\,\alpha_{i}^{\vee} \rangle} > 0$, and hence that $y_{n}^{-1}\alpha_{i} \in (\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta$ for all $k+1 \le n \le m+1$. Hence we deduce that $$\label{eq:yn}
(y_{n}')^{-1}\alpha_{i}
\in (\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta \subset \Delta^{+} + {\mathbb{Z}}\delta
\quad \text{for all $k+1 \le n \le m+1$}
$$ since $y_{n}' = y_{n}z_{n}$ for some $z_{n} \in ({W_{K}})_{{\mathrm{af}}}$. Therefore, it follows from Lemma \[lem:dmd\](3) (applied to the case $J=\emptyset$) that $s_{i}y_{k+1}' {\succeq}\cdots {\succeq}s_{i}y_{m+1}'$. Also, we see from that $s_{i}y_{m+1}' {\succeq}y_{m+1}'$. Thus we obtain $$\label{eq:Dem1-0}
s_{i}y_{k+1}' {\succeq}\cdots {\succeq}s_{i}y_{m+1}' {\succeq}y_{m+1}' {\succeq}\cdots {\succeq}y_{p}';$$ note that ${\Pi^{K}}(s_{i}y_{n}') = s_{i}y_{n}$ for all $k+1 \le n \le m+1$ by Lemma \[lem:si\] since $s_{i}y_{n} \in ({W^{K}})_{{\mathrm{af}}}$. If $t_{1} \ne 1$, then $\kappa(f_{i}\eta) = \kappa(\eta) = y_{p}$, and the final element of the sequence is $y_{p}'$, which satisfies ${\Pi^{S}}(y_{p}') {\succeq}{\Pi^{S}}(x)$ by our assumption. If $t_{1} = 1$, then $m+1=p$, $\kappa(f_{i}\eta) = s_{i}\kappa(\eta) = s_{i}y_{p}$, and the final element of the sequence is $s_{i}y_{m+1}' = s_{i}y_{p}'$. Since $s_{i}y_{m+1}' {\succeq}y_{m+1}'$ as shown above, we deduce from Lemma \[lem:611\] that ${\Pi^{S}}(s_{i}y_{p}') = {\Pi^{S}}(s_{i}y_{m+1}')
{\succeq}{\Pi^{S}}(y_{m+1}') = {\Pi^{S}}(y_{p}') {\succeq}{\Pi^{S}}(x)$. In what follows, we will give a defining chain for $f_{i}(\pi \otimes \eta) = \pi \otimes (f_{i}\eta)$ in which the sequence lies at the tail.
#### Case 1.
Assume that the set $\bigl\{1 \le n \le k \mid {\langle y_{n}\mu,\,\alpha_{i}^{\vee} \rangle} \ne 0\bigr\}$ is nonempty, and let $k_{0}$ be the maximum element of this set. Because the function $H^{\eta}_{i}(t)$ attains its minimum value $m^{\eta}_{i}$ at $t=t_{0}$, it follows that ${\langle y_{k}\mu,\,\alpha_{i}^{\vee} \rangle} =
{\langle y_{k-1}\mu,\,\alpha_{i}^{\vee} \rangle} = \cdots =
{\langle y_{k_{0}+1}\mu,\,\alpha_{i}^{\vee} \rangle} = 0$, and ${\langle y_{k_{0}}\mu,\,\alpha_{i}^{\vee} \rangle} < 0$, which implies that $y_{n}^{-1}\alpha_{i} \in {\Delta_{K}}+ {\mathbb{Z}}\delta$ for all $k_{0}+1 \le n \le k$, and $y_{k_{0}}^{-1}\alpha_{i} \in -(\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta$. Hence we deduce that $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}+ {\mathbb{Z}}\delta$ for all $k_{0}+1 \le n \le k$, and $(y_{k_{0}}')^{-1}\alpha_{i} \in -(\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta$. Therefore, there exists $k_{0} \le k_{1} \le k$ such that $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}^{+} + {\mathbb{Z}}\delta$ for all $k_{1}+1 \le n \le k$, and such that $(y_{k_{1}}')^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$; recall from that $(y_{k+1}')^{-1}\alpha_{i} \in \Delta^{+} + {\mathbb{Z}}\delta$. Hence, in this case, we deduce from Lemma \[lem:dmd\](1) and (3) that $y_{k_{1}}' {\succeq}s_{i}y_{k_{1}+1}' {\succeq}\cdots {\succeq}s_{i}y_{k}' {\succeq}s_{i}y_{k+1}'$; since $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}+ {\mathbb{Z}}\delta$ for all $k_{1}+1 \le n \le k$, we see by Remark \[rem:si\] that ${\Pi^{K}}(s_{i}y_{n}') = {\Pi^{K}}(y_{n}') = y_{n}$ for all $k_{1}+1 \le n \le k$. Thus, we obtain a defining chain $$\label{eq:Dem1-1}
x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{k_{1}}' {\succeq}s_{i}y_{k_{1}+1}' {\succeq}\cdots {\succeq}s_{i}y_{m+1}'
{\succeq}y_{m+1}' {\succeq}\cdots {\succeq}y_{p}'$$ for $f_{i}(\pi \otimes \eta) = \pi \otimes (f_{i}\eta)$.
#### Case 2.
Assume that the set $\bigl\{1 \le n \le k \mid {\langle y_{n}\mu,\,\alpha_{i}^{\vee} \rangle} \ne 0\bigr\}$ is empty, i.e., ${\langle y_{n}\mu,\,\alpha_{i}^{\vee} \rangle} = 0$ for all $1 \le n \le k$; note that $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}+ {\mathbb{Z}}\delta$ for all $1 \le n \le k$. If there exists $1 \le k_{1} \le k$ such that $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}^{+} + {\mathbb{Z}}\delta$ for all $k_{1}+1 \le n \le k$, and $(y_{k_{1}}')^{-1}\alpha_{i} \in -{\Delta_{K}}^{+} + {\mathbb{Z}}\delta$, then we obtain a defining chain of the form for $f_{i}(\pi \otimes \eta) = \pi \otimes (f_{i}\eta)$ in exactly the same way as in Case 1. Hence we may assume that $(y_{n}')^{-1}\alpha_{i} \in {\Delta_{K}}^{+} + {\mathbb{Z}}\delta$ for all $1 \le n \le k$. It follows from Lemma \[lem:dmd\](3) and that $$\label{eq:Dem1-2}
s_{i}y_{1}' {\succeq}\cdots {\succeq}s_{i}y_{k}' {\succeq}s_{i}y_{k+1}' {\succeq}\cdots {\succeq}s_{i}y_{m+1}'
{\succeq}y_{m+1}' {\succeq}\cdots {\succeq}y_{p}';$$ note that by Remark \[rem:si\], ${\Pi^{K}}(s_{i}y_{n}') = {\Pi^{K}}(y_{n}') = y_{n}$ for all $1 \le n \le k$. Now, we define $u_{0}$ to be the maximum element of the set $\bigl\{1 \le u \le s \mid {\langle x_{u}\lambda,\,\alpha_{i}^{\vee} \rangle} \ne 0\bigr\} \cup \{0\}$. We claim that if $u_{0} \ge 1$, then ${\langle x_{s}\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x_{s-1}\lambda,\,\alpha_{i}^{\vee} \rangle} = \cdots =
{\langle x_{u_{0}+1}\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$, and that if $u_{0} \ge 1$, then ${\langle x_{u_{0}}\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$; this would imply that $x_{u}^{-1}\alpha_{i} \in {\Delta_{J}}+ {\mathbb{Z}}\delta$ for all $u_{0}+1 \le u \le s$, and that if $u_{0} \ge 1$, then $x_{u_{0}}^{-1}\alpha_{i} \in -(\Delta^{+} \setminus {\Delta_{J}}^{+}) + {\mathbb{Z}}\delta$. Indeed, since ${\langle y_{n}\mu,\,\alpha_{i}^{\vee} \rangle} = 0$ for all $1 \le n \le k$ by our assumption, we see that $H^{\eta}_{i}(t)$ is identically zero on the interval $[0,\,t_{0}]$, and hence $m^{\eta}_{i} = 0$, from which it follows that ${\varepsilon}_{i}(\eta) = - m^{\eta}_{i} = 0$ by Remark \[rem:SLS\]. Here we recall that $f_{i}(\pi \otimes \eta) = \pi \otimes f_{i}\eta$ (if and) only if ${\varphi}_{i}(\pi) \le {\varepsilon}_{i}(\eta)$ by the tensor product rule for crystals. Hence we see that ${\varphi}_{i}(\pi) = H^{\pi}_{i}(1) - m^{\pi}_{i} = 0$ by Remark \[rem:SLS\]. Since ${\langle x_{s}\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x_{s-1}\lambda,\,\alpha_{i}^{\vee} \rangle} = \cdots =
{\langle x_{u_{0}+1}\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$ by our assumption, we obtain ${\langle x_{u_{0}}\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ if $u_{0} \ge 1$, as desired. Therefore, by the same argument as in Case 1, we get $0 \le u_{0} \le u_{1} \le s$ such that $(x_{u}')^{-1}\alpha_{i} \in {\Delta_{J}}^{+} + {\mathbb{Z}}\delta$ for all $u_{1}+1 \le u \le s$, and such that $(x_{u_{1}}')^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$ if $u_{1} \ge 1$; recall that $(y_{1}')^{-1}\alpha_{i} \in \Delta_{K}^{+} + {\mathbb{Z}}\delta$. Also we note that by Remark \[rem:si\], ${\Pi^{J}}(s_{i}x_{u}') = {\Pi^{J}}(x_{u}') = x_{u}$ for all $u_{1}+1 \le i \le s$. In this case, by Lemma \[lem:dmd\](1) and (3), together with , we obtain a defining chain $$x_{1}' {\succeq}\cdots {\succeq}x_{u_{1}}' {\succeq}s_{i}x_{u_{1}+1}' {\succeq}\cdots {\succeq}s_{i}x_{s}' {\succeq}s_{i}y_{1}' {\succeq}\cdots {\succeq}s_{i}y_{m+1}' {\succeq}y_{m+1}' {\succeq}\cdots {\succeq}y_{p}'$$ for $f_{i}(\pi \otimes \eta) = \pi \otimes (f_{i}\eta)$. This proves part (1).
\(2) Let $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$, and let $i \in I_{{\mathrm{af}}}$ be such that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$; we may assume that $e_{i}(\pi \otimes \eta) \ne {\bm{0}}$. Since $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$, there exists a defining chain for $\pi \otimes \eta$ whose final element, say $y \in W_{{\mathrm{af}}}$, satisfies ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$. We can show the following claims by arguments similar to those in part (1).
\[c:Dem1-1\]
If $e_{i}(\pi \otimes \eta) = e_{i}\pi \otimes \eta$, or if $e_{i}(\pi \otimes \eta) = \pi \otimes (e_{i}\eta)$ and $\kappa(e_{i}\eta)=\kappa(\eta)$, then there exists a defining chain for $e_{i}(\pi \otimes \eta)$ whose final element is $y$.
If $e_{i}(\pi \otimes \eta) = \pi \otimes (e_{i}\eta)$ and $\kappa(e_{i}\eta)=s_{i}\kappa(\eta)$, then there exists a defining chain for $e_{i}(\pi \otimes \eta)$ whose final element is $s_{i}y$.
In case (i) of Claim \[c:Dem1-1\], it is obvious that $e_{i}(\pi \otimes \eta) \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. In case (ii) of Claim \[c:Dem1-1\], we see by the definition of the root operator $e_{i}$ that with notation in and , $t_{1} = 1$, and the function $H^{\eta}_{i}(t)$ is strictly decreasing on $[t_{0},t_{1}] = [t_{0},1]$. Hence we have ${\langle \kappa(\eta)\mu,\,\alpha_{i}^{\vee} \rangle} < 0$, which implies that $\kappa(\eta)^{-1}\alpha_{i} \in
-(\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta$. Since ${\Pi^{K}}({\Pi^{S}}(y)) = {\Pi^{K}}(y) = \kappa(\eta)$, we see that $({\Pi^{S}}(y))^{-1}\alpha_{i} \in - (\Delta^{+} \setminus {\Delta_{K}}^{+}) + {\mathbb{Z}}\delta
\subset - (\Delta^{+} \setminus {\Delta_{S}}^{+}) + {\mathbb{Z}}\delta$, which implies that ${\langle {\Pi^{S}}(y)(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} < 0$. Also, it follows from Lemma \[lem:si\] that $s_{i}{\Pi^{S}}(y) \in ({W^{S}})_{{\mathrm{af}}}$ and ${\Pi^{S}}(s_{i}y) = s_{i}{\Pi^{S}}(y)$. Here, by the assumption, we have ${\langle {\Pi^{S}}(x)(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} =
{\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$. Therefore, we deduce from Lemma \[lem:dmd\](2), together with ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$, that ${\Pi^{S}}(s_{i}y) = s_{i}{\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$. Thus, we conclude that $e_{i}(\pi \otimes \eta) \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. This proves part (2).
\(3) If ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} = 0$, then ${\Pi^{S}}(s_{i}x)={\Pi^{S}}(x)$ by Remark \[rem:si\], and hence ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu) = {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. Hence the assertion is obvious from part (2).
Assume that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$. Then we see from Lemma \[lem:si\] that ${\Pi^{S}}(s_{i}x) = s_{i}{\Pi^{S}}(x) \in ({W^{S}})_{{\mathrm{af}}}$ and ${\Pi^{S}}(s_{i}x) {\succeq}{\Pi^{S}}(x)$, which implies that ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu) \supset {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$. Therefore, by part (2), we obtain the inclusion $\supset$ in . In order to show the opposite inclusion $\subset$ in , it suffices to show that $f_{i}^{\max}(\pi \otimes \eta) \in
{\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$ for all $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. In view of part (1), this assertion itself follows from the following claim.
\[c:Dem1-2\] Let $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. If $f_{i}(\pi \otimes \eta) = {\bm{0}}$, i.e., ${\varphi}_{i}(\pi \otimes \eta) = 0$, then $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$.
[*Proof of Claim \[c:Dem1-2\].*]{} We write $\pi$ and $\eta$ as: $\pi=(x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}})$ and $\eta=(y_{1},\,\dots,\,y_{p}\,;\,{\mathbf{b}})$, respectively. Let $x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{p}'$ be a defining chain for $\pi \otimes \eta$ such that ${\Pi^{S}}(y_{p}') {\succeq}{\Pi^{S}}(x)$. We see from Lemma \[lem:611\] that ${\Pi^{S}}(x_{1}') {\succeq}\cdots {\succeq}{\Pi^{S}}(x_{s}') {\succeq}{\Pi^{S}}(y_{1}') {\succeq}\cdots {\succeq}{\Pi^{S}}(y_{p}')$ is also a defining chain for $\pi \otimes \eta$ satisfying ${\Pi^{S}}({\Pi^{S}}(y_{p}')) = {\Pi^{S}}(y_{p}') {\succeq}{\Pi^{S}}(x)$. Hence we may assume from the beginning that $x_{1}',\,\dots,\,x_{s}',\,y_{1}',\,\dots,\,y_{p}' \in ({W^{S}})_{{\mathrm{af}}}$.
Assume first that the set $$\label{eq:set1}
\begin{split}
& \bigl\{1 \le q \le p \mid
(y_{q}')^{-1}\alpha_{i} \not\in ({\Delta_{K}}^{+} \setminus {\Delta_{S}}^{+}) + {\mathbb{Z}}\delta\bigr\} \\
& \hspace*{20mm}
= \bigl\{1 \le q \le p \mid
(y_{q}')^{-1}\alpha_{i} \in
\bigl((\Delta \setminus {\Delta_{K}}^{+}) \sqcup {\Delta_{S}}^{+}\bigr) + {\mathbb{Z}}\delta\bigr\}
\end{split}
$$ is nonempty. Let $q_{1}$ be the maximum element of this set; notice that ${\langle y_{q}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle y_{q}\mu,\,\alpha_{i}^{\vee} \rangle} =
{\langle y_{q}'\mu,\,\alpha_{i}^{\vee} \rangle} = 0$ for all $q_{1} < q \le p$. Note that $f_{i}(\pi \otimes \eta)={\bm{0}}$ implies $f_{i}\eta = {\bm{0}}$ by the tensor product rule for crystals, and hence $H^{\eta}_{i}(1) - m^{\eta}_{i}=0$. From this it follows that ${\langle y_{q_{1}}'\mu,\,\alpha_{i}^{\vee} \rangle} =
{\langle y_{q_{1}}\mu,\,\alpha_{i}^{\vee} \rangle} \le 0$, and hence $(y_{q_{1}}')^{-1}\alpha_{i} \in ((-\Delta^{+}) \sqcup {\Delta_{S}}^{+}) + {\mathbb{Z}}\delta$ by the definition of $q_{1}$, which implies that ${\langle y_{q_{1}}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \le 0$. Therefore, we see from Lemma \[lem:dmd\](1) and (3) that $$\label{eq:Dem1-2-1}
x_{1}' {\succeq}\cdots {\succeq}x_{s}' {\succeq}y_{1}' {\succeq}\cdots {\succeq}y_{q_{1}}' {\succeq}s_{i}y_{q_{1}+1}' {\succeq}\cdots {\succeq}s_{i}y_{p}';$$ note that ${\Pi^{K}}(s_{i}y_{q}')= {\Pi^{K}}(y_{q}')$ for all $q_{1} < q \le p$ since ${\langle y_{q}'\mu,\,\alpha_{i}^{\vee} \rangle} = 0$. Thus the sequence is also a defining chain for $\pi \otimes \eta$. If $q_{1} = p$, then the final element of is $y_{p}'$, and ${\langle y_{p}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \le 0$. Hence it follows from Lemma \[lem:dmd\](1) that $y_{p}' {\succeq}s_{i}{\Pi^{S}}(x) = {\Pi^{S}}(s_{i}x)$. If $q_{1} < p$, then the final element of is $s_{i}y_{p}'$, and ${\langle y_{p}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$. This implies that $s_{i}y_{p}' \in ({W^{S}})_{{\mathrm{af}}}$ by Lemma \[lem:si\], and that $s_{i}y_{p}' {\succeq}s_{i}{\Pi^{S}}(x) = {\Pi^{S}}(s_{i}x)$ by Lemma \[lem:dmd\](3). Hence we conclude that $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$.
Assume next that the set in is empty, that is, $(y_{q}')^{-1}\alpha_{i} \in ({\Delta_{K}}^{+} \setminus {\Delta_{S}}^{+}) + {\mathbb{Z}}\delta$ for all $1 \le q \le p$; notice that ${\langle y_{q}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$ for all $1 \le q \le p$. Also, since ${\langle y_{q}\mu,\,\alpha_{i}^{\vee} \rangle} =
{\langle y_{q}'\mu,\,\alpha_{i}^{\vee} \rangle} = 0$ for all $1 \le q \le p$, we have $H^{\eta}_{i}(t) = 0$ for all $t \in [0,1]$, and hence ${\varepsilon}_{i}(\eta)=0$. Since $f_{i}(\pi \otimes \eta)={\bm{0}}$ by the assumption, we obtain $f_{i}\pi = {\bm{0}}$ by the tensor product rule for crystals. Let $u_{1}$ be the maximum element of the set $\bigl\{ 1 \le u \le s \mid
(x_{u}')^{-1}\alpha_{i} \not\in ({\Delta_{J}}^{+} \setminus {\Delta_{S}}^{+}) + {\mathbb{Z}}\delta\bigr\} \cup \{0\}$. Then we have ${\langle x_{u}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle x_{u}'\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x_{u}\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$ for all $u_{1} < u \le s$. In addition, we can show by the same argument as above that if $u_{1} \ge 1$, then ${\langle x_{u_{1}}'(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \le 0$. Therefore, it follows from Lemma \[lem:dmd\](1) and (3) that $$\label{eq:Dem1-2-2}
x_{1}' {\succeq}\cdots {\succeq}x_{u_{1}}' {\succeq}s_{i}x_{u_{1}+1} {\succeq}\cdots {\succeq}s_{i}x_{s}' {\succeq}s_{i}y_{1}' {\succeq}\cdots {\succeq}s_{i}y_{p}'.$$ In the same way as for , we can verify that the sequence is a defining chain for $\pi \otimes \eta$ satisfying the condition in . This proves Claim \[c:Dem1-2\].
This completes the proof of Lemma \[lem:Dem1\].
\[cor:Dem2\] Let $x \in W_{{\mathrm{af}}}$, and $i \in I_{{\mathrm{af}}}$. For every $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$, we have $f_{i}^{\max}(\pi \otimes \eta) \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$.
If ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$, then the assertion follows from the proof of Lemma \[lem:Dem1\](3). If ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} < 0$, then we have ${\Pi^{S}}(s_{i}x) = s_{i}{\Pi^{S}}(x) {\preceq}{\Pi^{S}}(x)$ by Lemma \[lem:si\], and hence ${\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu) \supset {\mathbb{D}^{\frac{\infty}{2}}}_{x}(\lambda+\mu)$. Therefore, the assertion follows from Lemma \[lem:Dem1\](1). This proves the corollary.
\[lem:Dem3\] Let $x,\,y \in W_{{\mathrm{af}}}$.
If ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$ in $({W^{S}})_{{\mathrm{af}}}$, then $y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$ for all $({\bm{\rho}},\,{\bm{\chi}},\,\xi) \in {\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times
Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying .
If $y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$ for some $({\bm{\rho}},\,{\bm{\chi}},\,\xi) \in
{\mathop{\rm Par}\nolimits}(\lambda) \times {\mathop{\rm Par}\nolimits}(\mu) \times Q^{\vee}_{I \setminus ({J}\cup {K})}$ satisfying , then ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$.
\(1) By the definitions (see ), $\pi_{{\bm{\rho}}} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $\pi_{{\bm{\chi}}} \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\mu)$ are of the form: $$\label{eq:Dem3b}
\begin{cases}
\pi_{{\bm{\rho}}}=({\Pi^{J}}(t_{\xi_1}),\,\dots,\,{\Pi^{J}}(t_{\xi_{s-1}}),\,e\,;\,{\mathbf{a}}), & \\[1mm]
\pi_{{\bm{\chi}}}=({\Pi^{K}}(t_{\zeta_1}),\,\dots,\,{\Pi^{K}}(t_{\zeta_{p-1}}),\,e\,;\,{\mathbf{b}})
\end{cases}
$$ for some $\xi_{1},\,\dots,\,\xi_{s-1} \in Q^{\vee}_{I \setminus {J}}$ such that $\xi_{1} > \cdots > \xi_{s-1} > 0$, and $\zeta_{1},\,\dots,\,\zeta_{p-1} \in Q^{\vee}_{I \setminus {K}}$ such that $\zeta_{1} > \cdots > \zeta_{p-1} > 0$, respectively. Also, recall from that $$t_{\xi} \cdot \pi_{{\bm{\rho}}}=
({\Pi^{J}}(t_{\xi_1+\xi}),\,\dots,\,
{\Pi^{J}}(t_{\xi_{s-1}+\xi}),\,{\Pi^{J}}(t_{\xi})\,;\,{\mathbf{a}}),$$ and from Lemma \[lem:Weyl\] that $$\label{eq:Dem3c}
\begin{split}
& y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) =
(y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}})) \otimes (y \cdot \pi_{{\bm{\chi}}}), \quad \text{with} \\
&
\begin{cases}
y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}}) =
({\Pi^{J}}(yt_{\xi_1+\xi}),\,\dots,\,{\Pi^{J}}(yt_{\xi_{s-1}+\xi}),\,{\Pi^{J}}(yt_{\xi})\,;\,{\mathbf{a}}), & \\[1mm]
y \cdot \pi_{{\bm{\chi}}}=({\Pi^{K}}(yt_{\zeta_1}),\,\dots,\,{\Pi^{K}}(yt_{\zeta_{p-1}}),\,{\Pi^{K}}(y)\,;\,{\mathbf{b}}).
\end{cases}
\end{split}
$$
Now, if ${\bm{\chi}}=(\chi^{(i)})_{i \in I} \in {\mathop{\rm Par}\nolimits}(\mu)$, with $\chi^{(i)}=(\chi^{(i)}_{1} \ge \chi^{(i)}_{2} \ge \cdots)$ for $i \in I$, then we have $\zeta_{1} = \sum_{i \in I} \chi_{1}^{(i)} \alpha_{i}^{\vee}$ by ; we set $\gamma:=\sum_{i \in {J}\setminus K}
\chi_{1}^{(i)}\alpha_{i}^{\vee} \in {Q_{J}^{\vee}}$. Since $({\bm{\rho}},\,{\bm{\chi}},\,\xi)$ satisfies , and $\chi_{1}^{(i)}=0$ for all $i \in K$, we deduce that $\xi + \gamma \ge \zeta_{1}$, and hence that $$\xi_{1}+\xi+\gamma > \cdots > \xi_{s-1}+\xi+\gamma > \xi+\gamma \ge
\zeta_{1} > \cdots > \zeta_{p-1} > 0.$$ Therefore, it follows from Lemma \[lem:SiB\](2) (applied to the case $J=\emptyset$) that $$\label{eq:dc63}
yt_{\xi_{1}+\xi+\gamma} {\succ}\cdots {\succ}yt_{\xi_{s-1}+\xi+\gamma} {\succ}yt_{\xi+\gamma} {\succeq}yt_{\zeta_{1}} {\succ}\cdots {\succ}yt_{\zeta_{p-1}} {\succ}y = yt_{0} \quad \text{in $W_{{\mathrm{af}}}$};$$ note that by Lemma \[lem:PiJ\], ${\Pi^{J}}(yt_{\xi_{u}+\xi+\gamma}) = {\Pi^{J}}(yt_{\xi_{u}+\xi})$ for all $1 \le u \le s$ since $\gamma \in {Q_{J}^{\vee}}$. Hence the sequence is a defining chain for $y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})$. Since ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$ in $({W^{S}})_{{\mathrm{af}}}$ by the assumption, we conclude that $y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. This proves part (1).
\(2) We divide the proof into several steps.
#### Step 1.
Assume that $x = t_{\zeta'}$ for some $\zeta' \in Q^{\vee}$, and $y = t_{\xi'}$ for some $\xi' \in Q^{\vee}$; in this case, in order to prove that ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$ in $({W^{S}})_{{\mathrm{af}}}$, it suffices to show that $[\xi']^{S} \ge [\zeta']^{S}$ (see Lemma \[lem:SiB\](2)). In the same way as for , we obtain $$\label{eq:Dem3e}
\begin{split}
& y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) =
(y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}})) \otimes (y \cdot \pi_{{\bm{\chi}}}), \quad \text{with} \\
&
\begin{cases}
y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}}) =
({\Pi^{J}}(t_{\xi_1+\xi+\xi'}),\,\dots,\,{\Pi^{J}}(t_{\xi_{s-1}+\xi+\xi'}),\,{\Pi^{J}}(t_{\xi+\xi'})\,;\,{\mathbf{a}}), & \\[1mm]
y \cdot \pi_{{\bm{\chi}}}=({\Pi^{K}}(t_{\zeta_1+\xi'}),\,\dots,\,{\Pi^{K}}(t_{\zeta_{p-1}+\xi'}),\,{\Pi^{K}}(t_{\xi'})\,;\,{\mathbf{b}}).
\end{cases}
\end{split}
$$ Here, since $y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$ satisfies condition (or equivalently, condition ; see Section \[subsec:12\]), we have $$\label{eq:Dem3f}
\kappa(y \cdot \pi_{{\bm{\chi}}}) {\succeq}{\Pi^{K}}(x) \quad \text{and} \quad
\kappa \bigl(y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}}))\bigr) {\succeq}{\Pi^{J}}({\iota(y \cdot \pi_{{\bm{\chi}}},\,x)}).$$ We deduce from the first inequality in that ${\Pi^{K}}(t_{\xi'}) = \kappa(y \cdot \pi_{{\bm{\chi}}}) {\succeq}{\Pi^{K}}(x) = {\Pi^{K}}(t_{\zeta'})$, which implies that $[\xi']^{{K}} \ge [\zeta']^{{K}}$ by Lemma \[lem:SiB\](2). Since $I \setminus S = (I \setminus {K}) \sqcup ({K}\setminus S)$, it remains to show that $[\xi']_{{K}\setminus S} \ge [\zeta']_{{K}\setminus S}$. We define ${\widetilde{y}}_{p}$, ${\widetilde{y}}_{p-1}$, $\dots$, ${\widetilde{y}}_{1}$ by the recursive procedure , that is, $$\begin{split}
& {\widetilde{y}}_{p}:=\min {\mathrm{Lift}_{{\succeq}x}({\Pi^{K}}(t_{\xi'}))}, \quad
\text{with} \quad x = t_{\zeta'}, \\
& {\widetilde{y}}_{p-1}:=\min {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{p}}({\Pi^{K}}(t_{\zeta_{p-1}+\xi'}))}, \\
& \qquad \vdots \\
& {\widetilde{y}}_{1}:=\min {\mathrm{Lift}_{{\succeq}{\widetilde{y}}_{2}}({\Pi^{K}}(t_{\zeta_{1}+\xi'}))}
= {\iota(y \cdot \pi_{{\bm{\chi}}},\,x)}.
\end{split}
$$
\[c:Dem3-1\] The elements ${\widetilde{y}}_{q}$, $1 \le q \le p$, are of the form[:]{} ${\widetilde{y}}_{q} = t_{\zeta_{q}+\xi'+\gamma_{q}}$ for some $\gamma_{q} \in {Q_{K}^{\vee}}$, where we set $\zeta_{p}:=0$.
[*Proof of Claim \[c:Dem3-1\].*]{} We show the assertion by descending induction on $1 \le q \le p$. Assume that $q = p$. We see from Lemma \[lem:lift\] that ${\widetilde{y}}_{p} = z_{p} t_{\xi'+\gamma_{p}}$ for some $z_{p} \in {W_{K}}$ and $\gamma_{p} \in {Q_{K}^{\vee}}$. Since $z_{p} t_{\xi'+\gamma_{p}} = {\widetilde{y}}_{p} {\succeq}x = t_{\zeta'}$, it follows from Lemma \[lem:SiB\](1) and(2) that $t_{\xi'+\gamma_{p}} {\succeq}t_{\zeta'} = x$ in $W_{{\mathrm{af}}}$. Also, we have $t_{\xi'+\gamma_{p}} \in {\mathrm{Lift}}({\Pi^{K}}(t_{\xi'}))$ by Lemma \[lem:lift\]. Combining these, we obtain $t_{\xi'+\gamma_{p}} \in
{\mathrm{Lift}_{{\succeq}x}({\Pi^{K}}(t_{\xi'}))}$. Since ${\widetilde{y}}_{p} = z_{p} t_{\xi'+\gamma_{p}} {\succeq}t_{\xi'+\gamma_{p}}$ in $W_{{\mathrm{af}}}$ by Remark \[rem:SiB\], we deduce that ${\widetilde{y}}_{p} = t_{\xi'+\gamma_{p}}$ by the minimality of ${\widetilde{y}}_{p}$.
Assume that $q < p$; by our induction hypothesis, we have ${\widetilde{y}}_{q+1} = t_{\zeta_{q+1}+\xi'+\gamma_{q+1}}$ for some $\gamma_{q+1} \in {Q_{K}^{\vee}}$. Also, we see from Lemma \[lem:lift\] that ${\widetilde{y}}_{q} = z_{q} t_{\zeta_{q}+\xi'+\gamma_{q}}$ for some $z_{q} \in {W_{K}}$ and $\gamma_{q} \in {Q_{K}^{\vee}}$. Now, the same argument as above shows that $${\widetilde{y}}_{q} = z_{q} t_{\zeta_{q}+\xi'+\gamma_{q}} {\succeq}t_{\zeta_{q}+\xi'+\gamma_{q}} {\succeq}t_{\zeta_{q+1}+\xi'+\gamma_{q+1}} = {\widetilde{y}}_{q+1},$$ and that $t_{\zeta_{q}+\xi'+\gamma_{q}} \in {\mathrm{Lift}}({\Pi^{K}}(t_{\zeta_{q}+\xi'}))$. Hence we obtain ${\widetilde{y}}_{q} = t_{\zeta_{q}+\xi'+\gamma_{q}}$ by the minimality of ${\widetilde{y}}_{q}$. This proves Claim \[c:Dem3-1\].
Because ${\widetilde{y}}_{1} {\succeq}\cdots {\succeq}{\widetilde{y}}_{p} {\succeq}x$ in $W_{{\mathrm{af}}}$, it follows from Lemma \[lem:SiB\](2) and Claim \[c:Dem3-1\] that $$\zeta_{1}+\xi'+\gamma_{1} \ge \zeta_{2}+\xi'+\gamma_{2} \ge \cdots \ge
\zeta_{p}+\xi'+\gamma_{p} \ge \zeta';$$ in particular, we have $$\label{eq:step1-4}
[\zeta_{1}+\xi'+\gamma_{1}]_{{K}\setminus S} \ge
[\zeta']_{{K}\setminus S}.$$ Also, we see by the second inequality in and Claim \[c:Dem3-1\] that $${\Pi^{J}}(t_{\xi+\xi'}) =
\kappa(y \cdot (t_{\xi} \cdot \pi_{{\bm{\rho}}})) {\succeq}{\Pi^{J}}({\iota(y \cdot \pi_{{\bm{\chi}}},\,x)}) =
{\Pi^{J}}(t_{\zeta_{1}+\xi'+\gamma_{1}}),$$ which implies that $[\xi+\xi']^{{J}} \ge
[\zeta_{1}+\xi'+\gamma_{1}]^{{J}}$ by Lemma \[lem:SiB\](2); in particular, we have $[\xi+\xi']_{{K}\setminus S} \ge
[\zeta_{1}+\xi'+\gamma_{1}]_{{K}\setminus S}$. Here, since $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$, we have $[\xi+\xi']_{{K}\setminus S} =
[\xi']_{{K}\setminus S}$. Therefore, we deduce that $$\label{eq:step1-5}
[\xi']_{{K}\setminus S} \ge
[\zeta_{1}+\xi'+\gamma_{1}]_{{K}\setminus S}.$$ Combining and , we obtain $[\xi']_{{K}\setminus S} \ge
[\zeta']_{{K}\setminus S}$, as desired.
#### Step 2.
Assume that $x = t_{\zeta'}$ for some $\zeta' \in Q^{\vee}$, and write $y \in W_{{\mathrm{af}}}$ as $y = v t_{\xi'}$ for some $v \in W$ and $\xi' \in Q^{\vee}$. Let us show the assertion by induction on $\ell(v)$. If $\ell(v) = 0$, i.e., $v = e$, then the assertion follows from Step 1. Assume that $\ell(v) > 0$. We take $i \in I$ such that $\ell(s_{i}v) = \ell(v) -1$; note that $y^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$. Since ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} \le 0$ and ${\langle y\mu,\,\alpha_{i}^{\vee} \rangle} \le 0$, we see by the definition of the root operator $f_{i}$ and that $f_{i}\bigl(y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})\bigr) = {\bm{0}}$, and hence that $$\label{eq:step2-1}
e_{i}^{\max} \bigl( y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \bigr)
= (s_{i}y) \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}).$$ Since $x = t_{\zeta'}$, we have ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} =
{\langle \lambda+\mu,\,\alpha_{i}^{\vee} \rangle} \ge 0$. Therefore, by Lemma \[lem:Dem1\](2), together with , we obtain $(s_{i}y) \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. Hence, by our induction hypothesis, we have ${\Pi^{S}}(s_{i}y) {\succeq}{\Pi^{S}}(x)$. Here we recall that ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \le 0$ since $y^{-1}\alpha_{i} \in -\Delta^{+} + {\mathbb{Z}}\delta$. If ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} < 0$, then ${\Pi^{S}}(y) {\succeq}s_{i}{\Pi^{S}}(y) = {\Pi^{S}}(s_{i}y) {\succeq}{\Pi^{S}}(x)$ by Lemma \[lem:si\] and Remark \[rem:si\]. If ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} = 0$, then ${\Pi^{S}}(y) = {\Pi^{S}}(s_{i}y) {\succeq}{\Pi^{S}}(x)$ by Remark \[rem:si\]. In both cases, we obtain ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$, as desired.
#### Step 3.
Let $x,\,y \in W_{{\mathrm{af}}}$. We see from [@AK] that there exist $i_{1},\,\dots,\,i_{N} \in I_{{\mathrm{af}}}$ such that ${\langle s_{i_{n-1}} \cdots s_{i_{1}}x(\lambda+\mu),\,\alpha_{i_n}^{\vee} \rangle} \ge 0$ for all $1 \le n \le N$, and such that $s_{i_{N}} \cdots s_{i_{1}}x = t_{\zeta'}$ for some $\zeta' \in Q^{\vee}$. Let us show the assertion by induction on $N$. If $N=0$, i.e., $x = t_{\zeta'}$, then the assertion follows from Step 2. Assume that $N > 0$; for simplicity of notation, we set $i:=i_{1}$. It follows from Corollary \[cor:Dem2\] that $$\label{eq:step3-1}
f_{i}^{\max}
\bigl( y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \bigr)
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu).$$
#### Case 3.1.
Assume that ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \le 0$; note that ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} \le 0$ and ${\langle y\mu,\,\alpha_{i}^{\vee} \rangle} \le 0$. We see by the definition of the root operator $f_{i}$ and that $f_{i}^{\max}
\bigl( y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \bigr)=
y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}})$. Hence, by our induction hypothesis, we have ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(s_{i}x)$. Here we recall that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$. If ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} = 0$, then we have ${\Pi^{S}}(s_{i}x) = {\Pi^{S}}(x)$ by Remark \[rem:si\]. If ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$, then it follows from Lemma \[lem:si\] that ${\Pi^{S}}(s_{i}x) = s_{i}{\Pi^{S}}(x) {\succeq}{\Pi^{S}}(x)$. In both cases, we obtain ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$, as desired.
#### Case 3.2.
Assume that ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$; note that ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} \ge 0$ and ${\langle y\mu,\,\alpha_{i}^{\vee} \rangle} \ge 0$. We see by the definition of the root operator $e_{i}$ and that $e_{i} \bigl( y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \bigr) = {\bm{0}}$, and hence $$f_{i}^{\max}
\bigl( y \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}) \bigr)=
(s_{i}y) \cdot \bigl( t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}).$$ Hence, by our induction hypothesis, we have ${\Pi^{S}}(s_{i}y) {\succeq}{\Pi^{S}}(s_{i}x)$. As in Case 3.1, we see that ${\Pi^{S}}(s_{i}x) {\succeq}{\Pi^{S}}(x)$, and hence ${\Pi^{S}}(s_{i}y) {\succeq}{\Pi^{S}}(x)$. Also, since ${\langle y(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} > 0$, it follows from Lemma \[lem:si\] that ${\Pi^{S}}(s_{i}y) = s_{i}{\Pi^{S}}(y)$, and hence from Lemma \[lem:dmd\](2) that ${\Pi^{S}}(y) {\succeq}{\Pi^{S}}(x)$.
This proves part (2), and completes the proof of Lemma \[lem:Dem3\].
Now, the equivalence $\Leftrightarrow$ follows from the next lemma.
\[lem:DemX\] Let $\psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu)$, and assume that $\psi$ is mapped to $\pi \otimes \eta \in {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ under the isomorphism ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu) \cong {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ in Theorem \[thm:SMT\]. Let $x \in W_{{\mathrm{af}}}$.
If $\psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda+\mu)$, then $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$.
If $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$, then $\psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda+\mu)$.
By [@NS16 Lemma 5.4.1], there exist $i_{1},\,i_{2},\,\dots,\,i_{N} \in I_{{\mathrm{af}}}$ satisfying the conditions that $$\begin{cases}
{\langle s_{i_{n-1}}s_{i_{n-2}} \cdots
s_{i_{2}}s_{i_{1}}x(\lambda+\mu),\,\alpha_{i_n}^{\vee} \rangle} \ge 0
& \text{for all $1 \le n \le N$, and} \\[1.5mm]
f_{i_{N}}^{\max}f_{i_{N-1}}^{\max} \cdots f_{i_{2}}^{\max}f_{i_{1}}^{\max}\psi =
t_{\xi'} \cdot \pi_{{\bm{\omega}}}
& \text{for some $\xi' \in Q^{\vee}$ and ${\bm{\omega}}\in {\mathop{\rm Par}\nolimits}(\lambda+\mu)$}.
\end{cases}
$$ We prove part (1) by induction on $N$. Assume that $N=0$, i.e., $\psi = t_{\xi'} \cdot \pi_{{\bm{\omega}}}$; recall from that $\kappa(\psi) = {\Pi^{S}}(t_{\xi'})$. Since $\psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda+\mu)$ by the assumption, we have $$\label{eq:DemX-1}
{\Pi^{S}}(t_{\xi'}) = \kappa(\psi) {\succeq}{\Pi^{S}}(x).$$ By Corollary \[cor:Sext\], $\psi = t_{\xi'} \cdot \pi_{{\bm{\omega}}}$ is mapped to $t_{\xi'} \cdot
\bigl(t_{\xi} \cdot \pi_{{\bm{\rho}}} \otimes \pi_{{\bm{\chi}}}\bigr)$, which is $\pi \otimes \eta$, for some $\xi \in Q^{\vee}_{I \setminus ({J}\cup {K})}$ and ${\bm{\rho}}\in {\mathop{\rm Par}\nolimits}(\lambda)$, ${\bm{\chi}}\in {\mathop{\rm Par}\nolimits}(\mu)$ satisfying under the isomorphism ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda+\mu) \cong {\mathbb{S}^{\frac{\infty}{2}}}(\lambda+\mu)$ of crystals in Theorem \[thm:SMT\]. Therefore, we deduce from Lemma \[lem:Dem3\](1), together with , that $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$.
Assume that $N > 0$. For simplicity of notation, we set $i_{1}:=i$; note that ${\langle x(\lambda+\mu),\,\alpha_{i}^{\vee} \rangle} \ge 0$. We see from [@NS16 Corollary 5.3.3] that $f_{i}^{\max}\psi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}s_{i}x}(\lambda+\mu)$. By our induction hypothesis, we have $f_{i}^{\max}(\pi \otimes \eta)
\in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}s_{i}x}(\lambda+\mu)$. Since $\pi \otimes \eta = e_{i}^{k} f_{i}^{\max}(\pi \otimes \eta)$ for some $k \ge 0$, we deduce from Lemma \[lem:Dem1\](3) that $\pi \otimes \eta \in {\mathbb{D}^{\frac{\infty}{2}}}_{{\succeq}x}(\lambda+\mu)$. This proves part (1).
We can prove part (2) similarly, using Lemma \[lem:Dem3\](2) instead of Lemma \[lem:Dem3\](1).
Appendices. {#appendices. .unnumbered}
===========
Basic properties of the semi-infinite Bruhat order. {#sec:basic}
===================================================
We fix $J \subset I$ and $\lambda \in P^{+} \subset P_{{\mathrm{af}}}^{0}$ (see and ) such that $\big\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle}= 0 \bigr\} = {J}$.
\[lem:PiJ\]
It holds that $$\label{eq:PiJ2}
\begin{cases}
{\Pi^{J}}(w)={\lfloor w \rfloor}
& \text{\rm for all $w \in W$}; \\[1mm]
{\Pi^{J}}(xt_{\xi})={\Pi^{J}}(x){\Pi^{J}}(t_{\xi})
& \text{\rm for all $x \in W_{{\mathrm{af}}}$ and $\xi \in Q^{\vee}$};
\end{cases}
$$ in particular, $$\label{eq:PiJ3}
({W^{J}})_{{\mathrm{af}}}
= \bigl\{ w {\Pi^{J}}(t_{\xi}) \mid w \in W^J,\,\xi \in Q^{\vee} \bigr\}.$$
For each $\xi \in Q^{\vee}$, the element ${\Pi^{J}}(t_{\xi})$ is of the form[:]{} ${\Pi^{J}}(t_{\xi})=z_{\xi}t_{\xi+\phi_{{J}}(\xi)}$ for [(]{}a unique[)]{} $z_{\xi} \in {W_{J}}$ and $\phi_{{J}}(\xi) \in {Q_{J}^{\vee}}$.
For $\xi,\,\zeta \in Q^{\vee}$, $$\label{eq:PiJ1}
{\Pi^{J}}(t_{\xi}) = {\Pi^{J}}(t_{\zeta}) \iff \xi-\zeta \in {Q_{J}^{\vee}}.$$
\[lem:si\] Let $x \in ({W^{J}})_{{\mathrm{af}}}$, and $i \in I_{{\mathrm{af}}}$. Then, $$\label{eq:si1}
s_{i}x \in ({W^{J}})_{{\mathrm{af}}} \iff
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \ne 0 \iff
x^{-1}\alpha_{i} \in (\Delta \setminus {\Delta_{J}})+{\mathbb{Z}}\delta.$$ Moreover, in this case, $$\label{eq:simple}
\begin{cases}
x {\xrightarrow{\,\alpha_{i}\,}} s_{i}x \iff
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0 \iff
x^{-1}\alpha_{i} \in (\Delta^{+} \setminus {\Delta_{J}}^{+})+{\mathbb{Z}}\delta, & \\[1.5mm]
s_{i}x {\xrightarrow{\,\alpha_{i}\,}} x \iff
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} < 0 \iff
x^{-1}\alpha_{i} \in -(\Delta^{+} \setminus {\Delta_{J}}^{+})+{\mathbb{Z}}\delta. &
\end{cases}
$$
\[rem:si\] Keep the setting of Lemma \[lem:si\]. If $x^{-1}\alpha_{i} \in {\Delta_{J}}+{\mathbb{Z}}\delta$, i.e., ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$, then ${\Pi^{J}}(s_{i}x) = x$.
\[lem:dmd\] Let $x,\,y \in ({W^{J}})_{{\mathrm{af}}}$ be such that $x {\preceq}y$, and let $i \in I_{{\mathrm{af}}}$.
If ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} \le 0$, then $s_{i}x {\preceq}y$.
If ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \ge 0$ and ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, then $x {\preceq}s_{i}y$.
If ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, or if ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ and ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, then $s_{i}x {\preceq}s_{i}y$.
\[lem:SiB\]
Let $w,\,v \in {W^{J}}$, and $\xi,\,\zeta \in Q^{\vee}$. If $w{\Pi^{J}}(t_{\xi}) {\succeq}v{\Pi^{J}}(t_{\zeta})$, then $[\xi]^{{J}} \ge [\zeta]^{{J}}$, where $[\,\cdot\,]^{{J}}:Q^{\vee} \twoheadrightarrow
Q^{\vee}_{I \setminus {J}}$ is the projection in .
Let $w \in {W^{J}}$, and $\xi,\,\zeta \in Q^{\vee}$. Then, $w{\Pi^{J}}(t_{\xi}) {\succeq}w{\Pi^{J}}(t_{\zeta})$ if and only if $[\xi]^{{J}} \ge [\zeta]^{{J}}$.
Let $x,\,y \in ({W^{J}})_{{\mathrm{af}}}$ and $\beta \in {\Delta_{{\mathrm{af}}}^{+}}$ be such that $x {\xrightarrow{\,\beta\,}} y$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$. Then, ${\Pi^{J}}(xt_{\xi}) {\xrightarrow{\,\beta\,}} {\Pi^{J}}(yt_{\xi})$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ for all $\xi \in Q^{\vee}$. Therefore, if $x {\succeq}y$, then ${\Pi^{J}}(xt_{\xi}) {\succeq}{\Pi^{J}}(yt_{\xi})$ for all $\xi \in Q^{\vee}$.
\[rem:SiB\] Let $w \in W$. Since $w \ge e$ in the ordinary Bruhat order on $W$, we see that $w {\succeq}e$ in the semi-infinite Bruhat order on $W_{{\mathrm{af}}}$. Hence it follows from Lemma \[lem:SiB\](3) that $wt_{\xi} {\succeq}t_{\xi}$ for all $\xi \in Q^{\vee}$.
\[lem:SiL\] Let $x,\,y \in ({W^{J}})_{{\mathrm{af}}}$ be such that $x {\preceq}y$ in $({W^{J}})_{{\mathrm{af}}}$. Then, $xz {\preceq}yz$ in $W_{{\mathrm{af}}}$ for all $z \in ({W_{J}})_{{\mathrm{af}}}$.
Let $z \in ({W_{J}})_{{\mathrm{af}}}$. We know from [@Pet97] (see also [@A Theorem 3.3]) that ${\ell^{{\frac{\infty}{2}}}}(xz) = {\ell^{{\frac{\infty}{2}}}}(x) + {\ell^{{\frac{\infty}{2}}}}(z)$ and ${\ell^{{\frac{\infty}{2}}}}(yz) = {\ell^{{\frac{\infty}{2}}}}(y) + {\ell^{{\frac{\infty}{2}}}}(z)$. Also, we may assume that $x {\xrightarrow{\,\beta\,}} y$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$ for some $\beta \in {\Delta_{{\mathrm{af}}}^{+}}$; by the definition of ${\mathrm{BG}^{{\frac{\infty}{2}}}\bigl(({W^{J}})_{{\mathrm{af}}}\bigr)}$, we have $y = s_{\beta}x$, with ${\ell^{{\frac{\infty}{2}}}}(y)={\ell^{{\frac{\infty}{2}}}}(x)+1$. Therefore, $yz = s_{\beta}xz$, and $${\ell^{{\frac{\infty}{2}}}}(yz)
= {\ell^{{\frac{\infty}{2}}}}(y)+{\ell^{{\frac{\infty}{2}}}}(z)
= {\ell^{{\frac{\infty}{2}}}}(x)+1+{\ell^{{\frac{\infty}{2}}}}(z) = {\ell^{{\frac{\infty}{2}}}}(xz)+1.$$ Thus, we obtain $xz {\xrightarrow{\,\beta\,}} yz$ in ${\mathrm{BG}^{{\frac{\infty}{2}}}(W_{{\mathrm{af}}})}$, as desired.
\[lem:611\] If $x,\,y \in W_{{\mathrm{af}}}$ satisfy $x {\succeq}y$, then ${\Pi^{J}}(x) {\succeq}{\Pi^{J}}(y)$.
Proof of Proposition \[prop:Deo\]. {#sec:prf-Deo}
==================================
\[lem:lift\] Let $x \in ({W^{J}})_{{\mathrm{af}}}$, and write it as[:]{} $x = w {\Pi^{J}}(t_{\xi}) \in ({W^{J}})_{{\mathrm{af}}}$ for some $w \in {W^{J}}$ and $\xi \in Q^{\vee}$ [(]{}see . Then, ${\mathrm{Lift}}(x) = \bigl\{ w' t_{\xi+\gamma} \mid
w' \in w{W_{J}},\, \gamma \in {Q_{J}^{\vee}}\bigr\}$.
We set $L := \bigl\{ w' t_{\xi+\gamma} \mid
w' \in w{W_{J}},\, \gamma \in {Q_{J}^{\vee}}\bigr\}$. We first prove that $L \subset {\mathrm{Lift}}(x)$. Let $w' t_{\xi+\gamma} \in L$. Then we see from and that ${\Pi^{J}}(w' t_{\xi+\gamma}) = {\Pi^{J}}(w'){\Pi^{J}}(t_{\xi+\gamma})
= {\lfloor w' \rfloor}\Pi(t_{\xi}) = w \Pi(t_{\xi}) = x$. This proves the inclusion $L \subset {\mathrm{Lift}}(x)$.
We next show that $L \supset {\mathrm{Lift}}(x)$. Each element of ${\mathrm{Lift}}(x)$ is of the form $x z$ for some $z \in ({W_{J}})_{{\mathrm{af}}}={W_{J}}\ltimes {Q_{J}^{\vee}}$; we write $z$ as $z = v_{1}t_{\gamma_{1}}$ for some $v_{1} \in {W_{J}}$ and $\gamma_{1} \in {Q_{J}^{\vee}}$. Since ${\Pi^{J}}(t_{\xi}) = v_{2}t_{\xi + \gamma_{2}}$ for some $v_{2} \in {W_{J}}$ and $\gamma_{2} \in {Q_{J}^{\vee}}$, we have $xz = (w {\Pi^{J}}(t_{\xi})) (v_{1}t_{\gamma_{1}}) =
w (v_{2}t_{\xi + \gamma_{2}})(v_{1}t_{\gamma_{1}}) =
wv_{2}v_{1}t_{ v_{1}^{-1}(\xi + \gamma_{2}) + \gamma_{1} }$, which is of the form $w v t_{\xi+\gamma}$ for some $v \in {W_{J}}$ and $\gamma \in {Q_{J}^{\vee}}$. Thus, this element is contained in $L$. This proves the opposite inclusion, and hence the lemma.
Now, we give a proof of Proposition \[prop:Deo\]. If $J=I$, then the assertion is obvious. Hence we may assume that $J \subsetneqq I$.
#### Step 1.
Assume that $x = t_{\xi}$ for some $\xi \in Q^{\vee}$, and $y = {\Pi^{J}}(t_{\zeta})$ for some $\zeta \in Q^{\vee}$; since ${\Pi^{J}}(t_{\xi})=y {\succeq}{\Pi^{J}}(x)={\Pi^{J}}(t_{\xi})$ by the assumption, it follows from Lemma \[lem:SiB\](1) that $[\zeta]^{{J}} \ge [\xi]^{{J}}$, where $[\,\cdot\,]^{{J}}:Q^{\vee} \twoheadrightarrow Q_{I \setminus {J}}^{\vee}$ is the projection in . We set $\gamma:=[\zeta-\xi]_{{J}} \in {Q_{J}^{\vee}}$, where $[\,\cdot\,]_{{J}}:Q^{\vee} \twoheadrightarrow {Q_{J}^{\vee}}$ is the projection in ; note that $[\zeta-\gamma]_{{J}} = [\xi]_{{J}}$. We claim that $t_{\zeta-\gamma}$ is the minimum element of ${\mathrm{Lift}_{{\succeq}x}(y)}$. It is clear by Lemma \[lem:lift\] that $t_{\zeta-\gamma} \in {\mathrm{Lift}}(y)$. In addition, since $[\zeta-\gamma]^{{J}} = [\zeta]^{{J}} \ge [\xi]^{{J}}$ and $[\zeta-\gamma]_{{J}} = [\xi]_{{J}}$, we have $\zeta - \gamma \ge \xi$, and hence $t_{\zeta - \gamma} {\succeq}t_{\xi} = x$ by Lemma \[lem:SiB\](2). Thus, $t_{\zeta - \gamma} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Now, by Lemma \[lem:lift\], each element $y' \in {\mathrm{Lift}_{{\succeq}x}(y)} \subset {\mathrm{Lift}}(y)$ is of the form $y'=v't_{\zeta-\gamma'}$ for some $v' \in {W_{J}}$ and $\gamma' \in {Q_{J}^{\vee}}$. Since $v' t_{\zeta-\gamma'} = y' {\succeq}x = t_{\xi}$ in $W_{{\mathrm{af}}}$ by the assumption, we deduce from Lemma \[lem:SiB\](1) that $\zeta-\gamma' \ge \xi$; in particular, $[\zeta-\gamma']_{{J}} \ge [\xi]_{{J}}$. Here, since $\gamma=[\zeta-\xi]_{{J}}$ by the definition, we have $[\zeta-\gamma']_{{J}} \ge [\xi]_{{J}}=[\zeta-\gamma]_{{J}}$. Also, since $\gamma,\,\gamma' \in {Q_{J}^{\vee}}$, we have $[\zeta-\gamma']^{{J}} = [\zeta-\gamma]^{{J}}$. Combining these, we obtain $\zeta-\gamma' \ge \zeta-\gamma$. Therefore, by Remark \[rem:SiB\] and Lemma \[lem:SiB\](2), $y'=v' t_{\zeta-\gamma'} {\succeq}t_{\zeta-\gamma'} {\succeq}t_{\zeta-\gamma}$. Thus, $t_{\zeta-\gamma}$ is the minimum element of ${\mathrm{Lift}_{{\succeq}x}(y)}$.
In the following, we fix $\Lambda \in P^{+}$ and $\lambda \in P^{+}$ such that $\bigl\{ i \in I \mid {\langle \Lambda,\,\alpha_{i}^{\vee} \rangle} = 0\bigr\} = \emptyset$ and $\bigl\{ i \in I \mid {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} = 0\bigr\} = {J}$. Note that ${\langle \Lambda,\,\beta^{\vee} \rangle} \ne 0$ for all $\beta \in {\Delta_{{\mathrm{af}}}}$.
#### Step 2.
Let $x \in W_{{\mathrm{af}}}$, and assume that $y = {\Pi^{J}}(t_{\zeta})$ for some $\zeta \in Q^{\vee}$. We deduce from [@AK] that there exist $i_{1},\,\dots,\,i_{N} \in I_{{\mathrm{af}}}$ such that ${\langle s_{i_{n-1}} \cdots s_{i_{1}}x\Lambda,\,\alpha_{i_n}^{\vee} \rangle} > 0$ for all $1 \le n \le N$, and such that $s_{i_{N}} \cdots s_{i_{1}}x = t_{\xi}$ for some $\xi \in Q^{\vee}$. We show the assertion of the proposition by induction on $N$. If $N=0$, then the assertion follows from Step 1. Assume that $N \ge 1$; for simplicity of notation, we set $i:=i_{1} \in I_{{\mathrm{af}}}$. Since $$\label{eq:s2-1}
{\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$$ by the assumption, it follows that $x^{-1}\alpha_{i} \in \Delta^{+}$, and hence ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \ge 0$. Also, by Lemma \[lem:si\] and Remark \[rem:si\], $$\label{eq:Pix}
{\Pi^{J}}(s_{i}x) =
\begin{cases}
s_{i}{\Pi^{J}}(x) & \text{if ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$}, \\[1mm]
{\Pi^{J}}(x) & \text{if ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$}.
\end{cases}
$$
#### Case 2.1.
Assume that $i \in I \setminus {J}$, and hence ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle}=
{\langle \lambda,\,\alpha_{i}^{\vee} \rangle} > 0$; note that $s_{i}y \in ({W^{J}})_{{\mathrm{af}}}$ by Lemma \[lem:si\]. We first claim that $s_{i}y {\succeq}{\Pi^{J}}(s_{i}x)$. Indeed, if ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, then it follows from Lemma \[lem:dmd\](3), , and the assumption $y {\succeq}{\Pi^{J}}(x)$ that $s_{i}y {\succeq}s_{i}{\Pi^{J}}(x) = {\Pi^{J}}(s_{i}x)$. If ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle} = {\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$, then it follows from Lemma \[lem:si\], , and the assumption $y {\succeq}{\Pi^{J}}(x)$ that $s_{i}y {\succ}y {\succeq}{\Pi^{J}}(x) = {\Pi^{J}}(s_{i}x)$. In both cases, we obtain $s_{i}y {\succeq}{\Pi^{J}}(s_{i}x)$, as desired. Hence, by our induction hypothesis, $${\mathrm{Lift}_{{\succeq}s_{i}x}(s_{i}y)}=
\bigl\{ y'' \in W_{{\mathrm{af}}} \mid
\text{\rm ${\Pi^{J}}(y'') = s_{i}y$ and $y'' {\succeq}s_{i}x$}
\bigr\}$$ has the minimum element $y''_{\min}$. We next claim that $s_{i}y''_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Indeed, since ${\langle y''_{\min}\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle {\Pi^{J}}(y''_{\min})\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle s_{i}y\lambda,\,\alpha_{i}^{\vee} \rangle} = - {\langle y\lambda,\,\alpha_{i} \rangle} < 0$, it follows from Lemma \[lem:si\] that ${\Pi^{J}}(s_{i}y''_{\min}) = s_{i} {\Pi^{J}}(y''_{\min}) = s_{i}(s_{i}y) = y$, which implies that $s_{i}y''_{\min} \in {\mathrm{Lift}}(y)$. In addition, the inequality ${\langle y''_{\min}\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ above implies that ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$. Since ${\langle s_{i}x\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ by , we deduce from Lemma \[lem:dmd\](3), together with the assumption $y''_{\min} {\succeq}s_{i}x$, that $s_{i}y''_{\min} {\succeq}x$. Thus we get $s_{i}y''_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$, as desired. Finally, we claim that $$\label{eq:min2-1}
\text{$s_{i}y''_{\min}$ is the minimum element of ${\mathrm{Lift}_{{\succeq}x}(y)}$}.$$ Let $y' \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Since ${\langle y'\lambda,\,\alpha_{i}^{\vee} \rangle}=
{\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} = {\langle \lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ by our assumption, we see that $(y')^{-1}\alpha_{i} \in \Delta^{+} + {\mathbb{Z}}\delta$, and hence ${\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$. Also, since ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ by , it follows from Lemma \[lem:dmd\](3) that $s_{i}y' {\succeq}s_{i}x$, which implies that $s_{i}y' \in {\mathrm{Lift}_{{\succeq}s_{i}x}(s_{i}y)}$. Therefore, we obtain $s_{i}y' {\succeq}y''_{\min}$. Since ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ as seen above, and ${\langle s_{i}y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, we deduce from Lemma \[lem:dmd\](3) that $y' {\succeq}s_{i}y''_{\min}$. This shows .
#### **Case 2.2.**
Assume that $i \in {J}$, and hence ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle}=
{\langle \lambda,\,\alpha_{i}^{\vee} \rangle} = 0$. We first claim that $y {\succeq}{\Pi^{J}}(s_{i}x)$. Indeed, if ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle}
= {\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, then it follows from Lemma \[lem:dmd\](1), , and the assumption $y {\succeq}{\Pi^{J}}(x)$ that $y {\succeq}s_{i}{\Pi^{J}}(x) = {\Pi^{J}}(s_{i}x)$. If ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle} = {\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} = 0$, then $y {\succeq}{\Pi^{J}}(x) = {\Pi^{J}}(s_{i}x)$ by . In both cases, we obtain $y {\succeq}{\Pi^{J}}(s_{i}x)$, as desired. Hence, by our induction hypothesis, $${\mathrm{Lift}_{{\succeq}s_{i}x}(y)}=
\bigl\{ y'' \in W_{{\mathrm{af}}} \mid
\text{\rm ${\Pi^{J}}(y'') = y$ and $y'' {\succeq}s_{i}x$}
\bigr\}$$ has the minimum element $y''_{\min}$. We set $$y'_{\min}:=
\begin{cases}
y''_{\min} &
\text{if ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} > 0$}, \\[1mm]
s_{i}y''_{\min} &
\text{if ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} < 0$};
\end{cases}$$ remark that $y'_{\min} {\preceq}y''_{\min}$ by Lemma \[lem:si\]. First, we show that $y'_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Since $y''_{\min} \in {\mathrm{Lift}}(y)$ and $i \in {J}$, it follows from Remark \[rem:si\] that $y'_{\min} \in {\mathrm{Lift}}(y)$. Also, since ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, we see by Lemma \[lem:si\] that $s_{i}x {\succeq}x$. Hence we have $y_{\min}' = y_{\min}'' {\succeq}s_{i}x {\succeq}x$ if ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} > 0$. If ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} < 0$, then we deduce from Lemma \[lem:dmd\](3) that $y_{\min}' = s_{i}y_{\min}'' {\succeq}s_{i}(s_{i}x) = x$ since $y_{\min}'' {\succeq}s_{i}x$. Thus, in both cases, we obtain $y'_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$, as desired. Next, we show that $$\label{eq:min2-2}
\text{$y'_{\min}$ is the minimal element of ${\mathrm{Lift}_{{\succeq}x}(y)}$}.$$ Let $y' \in {\mathrm{Lift}_{{\succeq}x}(y)}$. If ${\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, then it follows from Lemma \[lem:dmd\](1) that $y' {\succeq}s_{i}x$, and hence $y' \in {\mathrm{Lift}_{{\succeq}s_{i}x}(y)}$. This implies that $y' {\succeq}y_{\min}''$. If ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} > 0$, then we have $y' {\succeq}y_{\min}'' = y_{\min}'$ by the definition. If ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} < 0$, then we see from Lemma \[lem:si\] that $y' {\succeq}y_{\min}'' {\succeq}s_{i}y_{\min}'' = y_{\min}'$. Assume now that ${\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$. Since ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, it follows from Lemma \[lem:dmd\](3) that $s_{i}y' {\succeq}s_{i}x$. In addition, since $i \in {J}$ and $y' \in {\mathrm{Lift}}(y)$, we see from Remark \[rem:si\] that ${\Pi^{J}}(s_{i}y')=y$. Hence we obtain $s_{i}y' \in {\mathrm{Lift}_{{\succeq}s_{i}x}(y)}$, so that $s_{i}y' {\succeq}y''_{\min}$; note that ${\langle s_{i}y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ by our assumption. If ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} > 0$, then we deduce from Lemma \[lem:dmd\](2) that $y' {\succeq}y''_{\min} = y'_{\min}$. If ${\langle y''_{\min}\Lambda ,\, \alpha_{i}^{\vee} \rangle} < 0$, then we deduce from Lemma \[lem:dmd\](3) that $y' {\succeq}s_{i}y''_{\min} = y'_{\min}$. Thus, in all cases, we have shown that $y' {\succeq}y'_{\min}$, as desired.
#### **Case 2.3.**
Assume that $i = 0$. In this case, we have ${\langle y\lambda,\,\alpha_{0}^{\vee} \rangle}=
{\langle \lambda,\,\alpha_{0}^{\vee} \rangle} < 0$ since $\alpha_{0}=-\theta+\delta$, where $\theta \in \Delta^{+}$ is the highest root. By the same argument as that at the beginning of Case 2.2, we see that $y {\succeq}{\Pi^{J}}(s_{0}x)$. Hence, by the induction hypothesis, $${\mathrm{Lift}_{{\succeq}s_{0}x}(y)}=
\bigl\{ y'' \in W_{{\mathrm{af}}} \mid
\text{\rm ${\Pi^{J}}(y'') = y$ and $y'' {\succeq}s_{0}x$}
\bigr\}$$ has the minimum element $y''_{\min}$. Since ${\langle x\Lambda ,\, \alpha_{0}^{\vee} \rangle} > 0$ by , it follows from Lemma \[lem:si\] that $s_{0}x {\succeq}x$, which implies that $y''_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Here we claim that $$\text{$y''_{\min}$ is the minimal element of ${\mathrm{Lift}_{{\succeq}x}(y)}$.}$$ Let $y' \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Then we have ${\langle y'\Lambda ,\, \alpha_{0}^{\vee} \rangle} < 0$. Indeed, we deduce from Lemma \[lem:lift\] that $y'=z t_{\zeta+\gamma}$ for some $z \in {W_{J}}$ and $\gamma \in {Q_{J}^{\vee}}$. Since $z \in {W_{J}}$ and $\theta \in \Delta^{+} \setminus {\Delta_{J}}^{+}$ (recall that $J \subsetneqq I$), we see that $z^{-1}\theta \in \Delta^{+} \setminus {\Delta_{J}}^{+}$, and hence that ${\langle y'\Lambda,\,\alpha_{0}^{\vee} \rangle} =
{\langle \Lambda,\,-z^{-1}\theta^{\vee} \rangle} < 0$. Since ${\langle x\Lambda ,\, \alpha_{0}^{\vee} \rangle} > 0$ by , it follows from Lemma \[lem:dmd\](1) that $y' {\succeq}s_{0}x$, and hence $y' \in {\mathrm{Lift}_{{\succeq}s_{0}x}(y)}$. This shows that $y' {\succeq}y''_{\min}$.
#### Step 3.
Let $x \in W_{{\mathrm{af}}}$, $y \in ({W^{J}})_{{\mathrm{af}}}$, and write $y$ as $y = v {\Pi^{J}}(t_{\zeta})$ for some $v \in {W^{J}}$ and $\zeta \in Q^{\vee}$. We show the assertion by induction on $\ell(v)$. If $\ell(v)=0$, then the assertion follows from Step 2. Assume that $\ell(v) \ge 1$, and take $i \in I$ such that ${\langle v\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$; note that in this case, $v^{-1}\alpha_{i} \in - (\Delta^{+} \setminus {\Delta_{J}}^{+})$, and $s_{i}v \in W^{J}$ (see, for example, [@LNSSS Lemmas 5.8 and 5.9]), and hence that $s_{i}y \in ({W^{J}})_{{\mathrm{af}}}$. Also, for all $y' \in {\mathrm{Lift}}(y)$, we have ${\langle y'\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle v\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, which implies that $$\label{eq:S3-1}
{\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0.$$
#### Case 3.1.
Assume that ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$; note that ${\langle {\Pi^{J}}(x)\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \ge 0$. Since ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} = {\langle v\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, it follows from Lemma \[lem:dmd\](2) that $s_{i}y {\succeq}{\Pi^{J}}(x)$. Hence, by our induction hypothesis, $${\mathrm{Lift}_{{\succeq}x}(s_{i}y)} =
\bigl\{ y'' \in W_{{\mathrm{af}}} \mid
\text{\rm ${\Pi^{J}}(y'') = s_{i}y$ and $y'' {\succeq}x$}
\bigr\}$$ has the minimum element $y''_{\min}$. Since ${\langle y''_{\min}\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle s_{i}y\lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, it follows that $(y''_{\min})^{-1}\alpha_{i} \in \Delta^{+}$, and hence ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$. This implies that $s_{i}y''_{\min} {\succeq}y''_{\min} {\succeq}x$ by Lemma \[lem:si\]. In addition, we see by Lemma \[lem:si\] that ${\Pi^{J}}(s_{i}y''_{\min}) = s_{i}{\Pi^{J}}(y''_{\min}) = s_{i}(s_{i}y) = y$. Therefore, we conclude that $s_{i}y''_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Here we claim that $$\label{eq:min3-1}
\text{$s_{i}y''_{\min}$ is the minimum element of ${\mathrm{Lift}_{{\succeq}x}(y)}$}.$$ Let $y' \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Since ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ by the assumption, and ${\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ by , we deduce from Lemma \[lem:dmd\](2) that $s_{i}y' {\succeq}x$. In addition, we see by Lemma \[lem:si\] that ${\Pi^{J}}(s_{i}y') = s_{i} {\Pi^{J}}(y') = s_{i}y$, which implies that $s_{i}y' \in {\mathrm{Lift}_{{\succeq}x}(s_{i}y)}$, and hence $s_{i}y' {\succeq}y''_{\min}$. Because ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle s_{i}y'\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, it follows from Lemma \[lem:dmd\](3) that $y' {\succeq}s_{i}y''_{\min}$. This shows .
#### Case 3.2.
Assume that ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$; note that ${\langle x\lambda,\,\alpha_{i}^{\vee} \rangle} \le 0$. Since ${\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, it follows from Lemma \[lem:dmd\](2) and (3), together with Lemma \[lem:si\] and Remark \[rem:si\], that $s_{i}y {\succeq}{\Pi^{J}}(s_{i}x)$. Hence, by our induction hypothesis, $${\mathrm{Lift}_{{\succeq}s_{i}x}(s_{i}y)} =
\bigl\{ y'' \in W_{{\mathrm{af}}} \mid
\text{\rm ${\Pi^{J}}(y'') = s_{i}y$ and $y'' {\succeq}s_{i}x$}
\bigr\}$$ has the minimum element $y''_{\min}$; as in Case 3.1, we obtain ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$. Since ${\langle s_{i}x\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, we deduce from Lemma \[lem:dmd\](3) that $s_{i}y''_{\min} {\succeq}x$. In addition, we see by Lemma \[lem:si\] that ${\Pi^{J}}(s_{i}y_{\min}'') = s_{i} {\Pi^{J}}(y_{\min}'') = s_{i} (s_{i}y) = y$, which implies that $s_{i}y''_{\min} \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Here we claim that $$\label{eq:min3-2}
\text{$s_{i}y''_{\min}$ is the minimum element of ${\mathrm{Lift}_{{\succeq}x}(y)}$}.$$ Let $y' \in {\mathrm{Lift}_{{\succeq}x}(y)}$. Since ${\langle y'\lambda,\,\alpha_{i}^{\vee} \rangle} =
{\langle y\lambda,\,\alpha_{i}^{\vee} \rangle} < 0$, it follows that ${\langle y'\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$. Also, since ${\langle x\Lambda,\,\alpha_{i}^{\vee} \rangle} < 0$ by the assumption, we deduce from Lemma \[lem:dmd\](3) that $s_{i}y' {\succeq}s_{i}x$. In addition, we see by Lemma \[lem:si\] that ${\Pi^{J}}(s_{i}y') = s_{i}{\Pi^{J}}(y') = s_{i}y$, which implies that $s_{i}y' \in {\mathrm{Lift}_{{\succeq}s_{i}x}(s_{i}y)}$, and hence $s_{i}y' {\succeq}y''_{\min}$. Because ${\langle y''_{\min}\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$ and ${\langle s_{i}y'\Lambda,\,\alpha_{i}^{\vee} \rangle} > 0$, it follows from Lemma \[lem:dmd\](3) that $y' {\succeq}s_{i}y''_{\min}$. This shows .
This completes the proof of Proposition \[prop:Deo\].
Crystal structure on ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$. {#sec:crystal}
==================================================================
We fix $\lambda \in P^{+} \subset P_{{\mathrm{af}}}^{0}$ (see and ). Let $$\label{eq:pi2}
\pi = ({\mathbf{x}}\,;\, {\mathbf{a}})
= (x_{1},\,\dots,\,x_{s} \,;\, a_{0},\,a_{1},\,\dots,\,a_{s}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda).$$ Define ${\overline{\pi}}:[0,1] \rightarrow {\mathbb{R}}\otimes_{{\mathbb{Z}}} P_{{\mathrm{af}}}$ to be the piecewise-linear, continuous map whose “direction vector” on the interval $[a_{u-1},\,a_{u}]$ is $x_{u}\lambda \in P_{{\mathrm{af}}}$ for each $1 \le u \le s$, that is, $$\label{eq:olpi}
{\overline{\pi}} (t) :=
\sum_{k = 1}^{u-1}(a_{k} - a_{k-1}) x_{k}\lambda + (t - a_{u-1}) x_{u}\lambda
\quad
\text{for $t \in [a_{u-1},\,a_u]$, $1 \le u \le s$}.$$ We know from [@INS Proposition 3.1.3] that ${\overline{\pi}}$ is an (ordinary) LS path of shape $\lambda$, introduced in [@Lit95 Sect. 4]. We set $$\label{eq:wt}
{\mathop{\rm wt}\nolimits}(\pi):= {\overline{\pi}}(1) = \sum_{u = 1}^{s} (a_{u}-a_{u-1})x_{u}\lambda \in P_{{\mathrm{af}}}.$$
Now, we define root operators $e_{i}$, $f_{i}$, $i \in I_{{\mathrm{af}}}$. Set $$\label{eq:H}
\begin{cases}
H^{\pi}_{i}(t) := {\langle {\overline{\pi}}(t),\,\alpha_{i}^{\vee} \rangle} \quad
\text{for $t \in [0,1]$}, \\[1.5mm]
m^{\pi}_{i} :=
\min \bigl\{ H^{\pi}_{i} (t) \mid t \in [0,1] \bigr\}.
\end{cases}
$$ As explained in [@NS16 Remark 2.4.3], all local minima of the function $H^{\pi}_{i}(t)$, $t \in [0,1]$, are integers; in particular, the minimum value $m^{\pi}_{i}$ is a nonpositive integer (recall that ${\overline{\pi}}(0)=0$, and hence $H^{\pi}_{i}(0)=0$).
We define $e_{i}\pi$ as follows. If $m^{\pi}_{i}=0$, then we set $e_{i} \pi := {\bm{0}}$, where ${\bm{0}}$ is an additional element not contained in any crystal. If $m^{\pi}_{i} \le -1$, then we set $$\label{eq:t-e}
\begin{cases}
t_{1} :=
\min \bigl\{ t \in [0,\,1] \mid
H^{\pi}_{i}(t) = m^{\pi}_{i} \bigr\}, \\[1.5mm]
t_{0} :=
\max \bigl\{ t \in [0,\,t_{1}] \mid
H^{\pi}_{i}(t) = m^{\pi}_{i} + 1 \bigr\};
\end{cases}
$$ notice that $H^{\pi}_{i}(t)$ is strictly decreasing on the interval $[t_{0},\,t_{1}]$. Let $1 \le p \le q \le s$ be such that $a_{p-1} \le t_{0} < a_p$ and $t_{1} = a_{q}$. Then we define $e_{i}\pi$ to be $$\label{eq:epi}
\begin{split}
& e_{i} \pi := (
x_{1},\,\ldots,\,x_{p},\,s_{i}x_{p},\,s_{i}x_{p+1},\,\ldots,\,
s_{i}x_{q},\,x_{q+1},\,\ldots,\,x_{s} ; \\
& \hspace*{40mm}
a_{0},\,\ldots,\,a_{p-1},\,t_{0},\,a_{p},\,\ldots,\,a_{q}=t_{1},\,
\ldots,\,a_{s});
\end{split}
$$ if $t_{0} = a_{p-1}$, then we drop $x_{p}$ and $a_{p-1}$, and if $s_{i} x_{q} = x_{q+1}$, then we drop $x_{q+1}$ and $a_{q}=t_{1}$.
Similarly, we define $f_{i}\pi$ as follows. Note that $H^{\pi}_{i}(1) - m^{\pi}_{i}$ is a nonnegative integer. If $H^{\pi}_{i}(1) - m^{\pi}_{i} = 0$, then we set $f_{i} \pi := {\bm{0}}$. If $H^{\pi}_{i}(1) - m^{\pi}_{i} \ge 1$, then we set $$\label{eq:t-f}
\begin{cases}
t_{0} :=
\max \bigl\{ t \in [0,1] \mid H^{\pi}_{i}(t) = m^{\pi}_{i} \bigr\}, \\[1.5mm]
t_{1} :=
\min \bigl\{ t \in [t_{0},\,1] \mid H^{\pi}_{i}(t) = m^{\pi}_{i} + 1 \bigr\};
\end{cases}
$$ notice that $H^{\pi}_{i}(t)$ is strictly increasing on the interval $[t_{0},\,t_{1}]$. Let $0 \le p \le q \le s-1$ be such that $t_{0} = a_{p}$ and $a_{q} < t_{1} \le a_{q+1}$. Then we define $f_{i}\pi$ to be $$\label{eq:fpi}
\begin{split}
& f_{i} \pi := ( x_{1},\,\ldots,\,x_{p},\,s_{i}x_{p+1},\,\dots,\,
s_{i} x_{q},\,s_{i} x_{q+1},\,x_{q+1},\,\ldots,\,x_{s} ; \\
& \hspace{40mm}
a_{0},\,\ldots,\,a_{p}=t_{0},\,\ldots,\,a_{q},\,t_{1},\,
a_{q+1},\,\ldots,\,a_{s});
\end{split}
$$ if $t_{1} = a_{q+1}$, then we drop $x_{q+1}$ and $a_{q+1}$, and if $x_{p} = s_{i} x_{p+1}$, then we drop $x_{p}$ and $a_{p}=t_{0}$. In addition, we set $e_{i} {\bm{0}}= f_{i} {\bm{0}}:= {\bm{0}}$ for all $i \in I_{{\mathrm{af}}}$.
\[thm:SLS\]
The set ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda) \sqcup \{ {\bm{0}}\}$ is stable under the action of the root operators $e_{i}$ and $f_{i}$, $i \in I_{{\mathrm{af}}}$.
For each $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$ and $i \in I_{{\mathrm{af}}}$, we set $$\begin{cases}
{\varepsilon}_{i} (\pi) :=
\max \bigl\{ n \ge 0 \mid e_{i}^{n} \pi \neq {\bm{0}}\bigr\}, \\[1.5mm]
{\varphi}_{i} (\pi) :=
\max \bigl\{ n \ge 0 \mid f_{i}^{n} \pi \neq {\bm{0}}\bigr\}.
\end{cases}
$$ Then, the set ${\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$, equipped with the maps ${\mathop{\rm wt}\nolimits}$, $e_{i}$, $f_{i}$, $i \in I_{{\mathrm{af}}}$, and ${\varepsilon}_{i}$, ${\varphi}_{i}$, $i \in I_{{\mathrm{af}}}$, defined above, is a crystal with weights in $P_{{\mathrm{af}}}$.
\[rem:SLS\] Let $\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$, and $i \in I_{{\mathrm{af}}}$. If $e_{i}\pi \ne {\bm{0}}$, then we deduce from the definition of the root operator $e_{i}$ that $m^{e_{i}\pi}_{i} = m^{\pi}_{i}+1$. Hence it follows that ${\varepsilon}_{i}(\pi)=-m^{\pi}_{i}$. Similarly, we have ${\varphi}_{i}(\pi)=H^{\pi}_{i}(1)-m^{\pi}_{i}$.
A formula for graded characters of Demazure submodules. {#sec:tx-gch}
=======================================================
\[prop:gch-tx\] For each $x \in W_{{\mathrm{af}}}$ and $\xi \in Q^{\vee}$, there holds the equality $${\mathop{\rm gch}\nolimits}V_{xt_{\xi}}^{-}(\lambda) = q^{-{\langle \lambda,\,\xi \rangle}}{\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda).$$
Let $\pi = (x_{1},\,\dots,\,x_{s}\,;\,{\mathbf{a}}) \in {\mathbb{B}^{{\frac{\infty}{2}}}}(\lambda)$. We see that $$\begin{aligned}
\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)
& \Rightarrow x_{s} {\succeq}{\Pi^{J}}(x)
\Rightarrow {\Pi^{J}}(x_{s}t_{\xi}) {\succeq}{\Pi^{J}}({\Pi^{J}}(x)t_{\xi})
\quad \text{by Lemma~\ref{lem:SiB}\,(3)} \\
& \Rightarrow {\Pi^{J}}(x_{s}t_{\xi}) {\succeq}{\Pi^{J}}(xt_{\xi})
\quad \text{by \eqref{eq:PiJ2}} \\
& \Rightarrow T_{\xi}(\pi) \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda);\end{aligned}$$ for the definition of $T_{\xi}$, see . From this, we conclude that $T_{\xi}({\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)) \subset
{\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda)$. Replacing $x$ by $xt_{\xi}$, and $T_{\xi}$ by $T_{-\xi}$, we obtain $T_{-\xi}({\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda)) \subset
{\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)$, and hence ${\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda) \subset T_{\xi}({\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda))$. Combining these, we conclude that $T_{\xi}({\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)) = {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda)$. Therefore, using , we compute: $$\begin{aligned}
{\mathop{\rm gch}\nolimits}V_{xt_{\xi}}^{-}(\lambda)
& = \sum_{\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}xt_{\xi}}(\lambda)} e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))} q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))}
= \sum_{\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(T_{\xi}(\pi)))}
q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(T_{\xi}(\pi)))} \\[3mm]
& = \sum_{\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\pi)-{\langle \lambda,\,\xi \rangle}\delta)}
q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\pi)-{\langle \lambda,\,\xi \rangle}\delta)}
\quad \text{by \eqref{eq:Tx}} \\[3mm]
& = \sum_{\pi \in {\mathbb{B}^{{\frac{\infty}{2}}}}_{{\succeq}x}(\lambda)}
e^{{\mathop{\rm fin}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))}
q^{{\mathop{\rm nul}\nolimits}({\mathop{\rm wt}\nolimits}(\pi))-{\langle \lambda,\,\xi \rangle}}
= q^{-{\langle \lambda,\,\xi \rangle}} {\mathop{\rm gch}\nolimits}V_{x}^{-}(\lambda). \end{aligned}$$ This proves the proposition.
[LN3SXX]{}
T. Akasaka and M. Kashiwara, Finite-dimensional representations of quantum affine algebras, [*Publ. Res. Inst. Math. Sci.*]{} [**33**]{} (1997), 839–867.
D. Anderson, S. Griffeth, and E. Miller, Positivity and Kleiman transversality in equivariant $K$-theory of homogeneous spaces, [*J. Eur. Math. Soc.*]{} [**13**]{} (2011), 57–84.
T. Arakawa, Two-sided BGG resolutions of admissible representations, [*Represent. Theory*]{} [**18**]{} (2014), 183–222.
S. Baldwin and S. Kumar, Positivity in $T$-equivariant $K$-theory of flag varieties associated to Kac-Moody groups II, [*Represent. Theory*]{} [**21**]{} (2017), 35–60.
J. Beck and H. Nakajima, Crystal bases and two-sided cells of quantum affine algebras, [*Duke Math. J.*]{} [**123**]{} (2004), 335–402.
A. Björner and F. Brenti, “Combinatorics of Coxeter Groups”, Graduate Texts in Mathematics Vol. 231, Springer, New York, 2005.
A. Braverman and M. Finkelberg, Semi-infinite Schubert varieties and quantum $K$-theory of flag manifolds, [*J. Amer. Math. Soc.*]{} [**27**]{} (2014), 1147–1168.
A. Braverman and M. Finkelberg, Weyl modules and $q$-Whittaker functions, [*Math. Ann.*]{} [**359**]{} (2014), 45–59.
A. Braverman and M. Finkelberg, Twisted zastava and $q$-Whittaker functions, [*J. London Math. Soc.*]{} [**96**]{} (2017), 309–325.
A. Braverman and M. Finkelberg, unpublished notes.
A. Braverman and D. Gaitsgory, Geometric Eisenstein series, [*Invent. Math.*]{} [**150**]{} (2002), 287–384.
M. Brion, Positivity in the Grothendieck group of complex flag varieties, [*J. Algebra*]{} [**258**]{} (2002), 137–159.
I. Cherednik and B. Feigin, Rogers-Ramanujan type identities and nil-DAHA, [*Adv. Math.*]{} [**248**]{} (2013), 1050–1088.
B. Feigin, M. Finkelberg, A. Kuznetsov, and I. Mirković, Semi-infinite flags. II. Local and global intersection cohomology of quasimaps’ spaces, [*in*]{} “Differential Topology, Infinite-dimensional Lie Algebras, and Applications”, pp. 113–148, Amer. Math. Soc. Transl. Ser. 2, Vol. 194, Amer. Math. Soc., Providence, RI, 1999.
M. Finkelberg and I. Mirković, Semi-infinite flags. I. Case of global curve $\mathbf{P}^{1}$, [*in*]{} “Differential Topology, Infinite-dimensional Lie Algebras, and Applications”, pp. 81–112, Amer. Math. Soc. Transl. Ser. 2, Vol. 194, Amer. Math. Soc., Providence, RI, 1999.
S. Fishel, I. Grojnowski, and C. Teleman, The strong Macdonald conjecture and Hodge theory on the loop Grassmannian, [*Ann. Math.*]{} [**168**]{} (2008), 175–220.
G. Fourier and P. Littelmann, Weyl modules, Demazure modules, KR-modules, crystals, fusion products and limit constructions, [*Adv. Math.*]{} [**211**]{} (2007), 566–593.
A. Givental and Y.-P. Lee, Quantum $K$-theory on flag manifolds, finite-difference Toda lattices and quantum groups, [*Invent. Math.*]{} [**151**]{} (2003), 193–219.
P. Gille, Le problème de Kneser-Tits, Séminaire Bourbaki Vol. 2007/2008, Astérisque No. 326 (2009), Exp. No. 983, vii, 39–81 (2010).
A. Grothendieck and J. Dieudonné, “Eléments de Géométrie Algébrique. I: Le langage des schémas”, Inst. Hautes Études Sci. Publ. Math. No. 4, 1960.
A. Grothendieck and J. Dieudonné, “Eléments de Géométrie Algébrique. III: Étude cohomologique des faisceaux cohérents. I.”, Inst. Hautes Études Sci. Publ. Math. No. 11, 1961.
A. Grothendieck, “Cohomologie Locale des Faisceaux Cohérents et Théorèmes de Lefschetz Locaux et Globaux. Fasc. I: Exposés 1–8; Fasc. II: Exposés 9–13”, Séminaire de Géométrie Algébrique 1962, Troisième édition, corrigée, Rédigé par un groupe d’auditeurs Institut des Hautes Études Scientifiques, Paris 1965.
R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics Vol. 52, Springer-Verlag, New York-Heidelberg, 1977.
J. Hong and S.-J. Kang, “Introduction to Quantum Groups and Crystal Bases”, Graduate Studies in Mathematics Vol. 42, Amer. Math. Soc., Providence, RI, 2002.
J. E. Humphreys, “Reflection Groups and Coxeter Groups”, Cambridge Studies in Advanced Mathematics Vol. 29, Cambridge University Press, Cambridge, 1990.
M. Ishii, S. Naito, and D. Sagaki, Semi-infinite Lakshmibai-Seshadri path model for level-zero extremal weight modules over quantum affine algebras, [*Adv. Math.*]{} [**290**]{} (2016), 967–1009.
V. G. Kac, “Infinite Dimensional Lie Algebras”, 3rd Edition, Cambridge University Press, Cambridge, UK, 1990.
M. Kashiwara, Crystal bases of modified quantized enveloping algebra, [*Duke Math. J.*]{} [**73**]{} (1994), 383–413.
M. Kashiwara, On level-zero representations of quantized affine algebras, [*Duke Math. J.*]{} [**112**]{} (2002), 117–175.
M. Kashiwara, Level zero fundamental representations over quantized affine algebras and Demazure modules, [*Publ. Res. Inst. Math. Sci.*]{} [**41**]{} (2005), 223–250.
S. Kato, Demazure character formula for semi-infinite flag manifolds, preprint 2016, arXiv:1605.04953.
S. Kato, Frobenius splitting of thick flag manifolds of Kac-Moody algebras, preprint 2017, arXiv:arXiv:1707.03773.
A. Klyachko, Stable bundles, representation theory and Hermitian operators, [*Selecta Math.*]{} (N.S.) [**4**]{} (1998), 419–445.
F. Knop, H. Kraft, and T. Vust, The Picard group of a $G$-variety, [*in*]{} “Algebraische Transformationsgruppen und Invariantentheorie”, pp. 77–87, DMV Sem. Vol. 13, Birkhäuser, Basel, 1989.
B. Kostant and S. Kumar, $T$-equivariant $K$-theory of generalized flag varieties, [*J. Differential Geom.*]{} [**32**]{} (1990), 549–603.
S. Kumar, “Kac-Moody Groups, their Flag Varieties and Representation Theory”, Progress in Mathematics Vol. 204, Birkhäuser Boston, Boston, MA, 2002.
T. Lam and M. Shimozono, Quantum cohomology of $G/P$ and homology of affine Grassmannian, [*Acta Math.*]{} [**204**]{} (2010), 49–90.
T. Lam, C. Li, L. Mihalcea, and M. Shimozono, A conjectural Peterson isomorphism in $K$-theory, arXiv:1705.03435.
C. Lenart, S. Naito, D. Sagaki, A. Schilling, and M. Shimozono, A uniform model for Kirillov-Reshetikhin crystals I: Lifting the parabolic quantum Bruhat graph, [*Int. Math. Res. Not.*]{} [**2015**]{} (2015), 1848–1901.
C. Lenart, S. Naito, D. Sagaki, A. Schilling, and M. Shimozono, A uniform model for Kirillov-Reshetikhin crystals II: Alcove model, path model, and $P=X$, [*Int. Math. Res. Not.*]{} [**2017**]{} (2017), 4259–4319.
C. Lenart and A. Postnikov, Affine Weyl groups in $K$-theory and representation theory, [*Int. Math. Res. Not.*]{} [**2007**]{} (2007), no. 12, Art. ID: rnm038, 65 pp.
C. Lenart and M. Shimozono, Equivariant $K$-Chevalley rules for Kac-Moody flag manifolds, [*Amer. J. Math.*]{} [**136**]{} (2014), 1175–1213.
P. Littelmann, Paths and root operators in representation theory, [*Ann. of Math.*]{} (2) [**142**]{} (1995), 499–525.
P. Littelmann and Seshadri, A Pieri-Chevalley type formula for $K(G/B)$ and standard monomial theory, [*in*]{} “Studies in memory of Issai Schur”, pp. 155–176, Progr. Math. Vol. 210, Birkhäuser Boston, Boston, MA, 2003.
G. Lusztig, Hecke algebras and Jantzen’s generic decomposition patterns, [*Adv. Math.*]{} [**37**]{} (1980), 121–164.
G. Lusztig, Periodic $W$-graphs, [*Represent. Theory*]{} [**1**]{} (1997), 207–279.
I.G. Macdonald, “Notes on Schubert Polynomials”, Publications du Laboratoire de combinatoire et d’informatique mathématique Vol. 6, Dép. de mathématiques et d’informatique, Université du Québec à Montréal, 1991.
I.G. Macdonald, “Affine Hecke Algebras and Orthogonal Polynomials”, Cambridge Tracts in Mathematics Vol. 157, Cambridge University Press, Cambridge, 2003.
O. Mathieu, Positivity of some intersections in $K_{0}(G/B)$, [*J. Pure Appl. Algebra*]{} [**152**]{} (2000), 231–243.
E. Mukhin, V. Tarasov, and A. Varchenko, Schubert calculus and representations of the general linear group, [*J. Amer. Math. Soc.*]{} [**22**]{} (2009), 909–940.
S. Naito, F. Nomoto, and D. Sagaki, Specialization of nonsymmetric Macdonald polynomials at $t=\infty$ and Demazure submodules of level-zero extremal weight modules, arXiv:1511.07005, to appear in [*Trans. Amer. Math. Soc.*]{}, DOI:10.1090/tran/7114.
S. Naito and D. Sagaki, Crystal of Lakshmibai-Seshadri paths associated to an integral weight of level zero for an affine Lie algebra, [*Int. Math. Res. Not.*]{} [**2005**]{} (2005), no. 14, 815–840.
S. Naito and D. Sagaki, Path model for a level-zero extremal weight module over a quantum affine algebra. II, [*Adv. Math.*]{} [**200**]{} (2006), 102–124.
S. Naito and D. Sagaki, Demazure submodules of level-zero extremal weight modules and specializations of Macdonald polynomials, [*Math. Z.*]{} [**283**]{} (2016), 937–978.
K. Naoi, Weyl modules, Demazure modules and finite crystals for non-simply laced type, [*Adv. Math.*]{} [**229**]{} (2012), 875–934.
D. Peterson, Quantum Cohomology of $G/P$, Lecture notes, Massachusetts Institute of Technology, Spring 1997.
H. Pittie and A. Ram, A Pieri-Chevalley formula in the $K$-theory of a $G/B$-bundle, [*Electron. Res. Announc. Amer. Math. Soc.*]{} [**5**]{} (1999), 102–107.
[^1]: Key words and phrases: semi-infinite flag manifold, normality, $K$-theory, Pieri-Chevalley formula, standard monomial theory, semi-infinite Lakshmibai-Seshadri path. Mathematics Subject Classification 2010: Primary 17B37; Secondary 14N15, 14M15, 33D52, 81R10.
| {
"pile_set_name": "ArXiv"
} |
Electron acceleration driven by high-power lasers has attracted much effort, since it could lead to the development of compact electron accelerators. Such devices are expected to be useful in many applications, ranging from coherent X-ray generation [@nakajima08_naturephys; @schlenvoigt08_naturephys] to electron diffraction imaging [@baum07_pnas]. In laser-driven particle acceleration, the challenge is to find an efficient way to apply the electromagnetic field. Among the many proposed methods, the use of radially polarized laser beams (RPLBs) for electron acceleration in vacuum has been the object of many investigations [@salamin06_pra; @varin05_pre; @varin06_pre; @salamin07_optlett; @fortin10_jpb; @wong10_optexpress; @singh11_prstab; @marceau12_optlett]. In this scheme, which was recently demonstrated experimentally [@payeur12_apl], electrons experience sub-cycle acceleration from the longitudinal electric field component at the center of an ultra-intense TM$_{01}$ beam. This mechanism induces a strong longitudinal compression effect that could theoretically lead to the production of attosecond electron bunches [@varin06_pre; @karmakar07_lpb].
Many studies that have investigated on-axis [@varin05_pre; @salamin06_pra; @fortin10_jpb; @wong10_optexpress] and off-axis [@varin06_pre; @singh11_prstab] electron acceleration in RPLBs are based on the usual *paraxial approximation*. In these analyses, it is generally assumed that the paraxial approximation is valid whenever $k_0 z_R \gtrsim 125$ (or $w_0/\lambda_0 \gtrsim 2.5$), where $z_R$ and $w_0$ are respectively the Rayleigh range and beam waist size, and $\lambda_0 = 2\pi/k_0$ is the dominant wavelength of the laser.
In this Letter, we question the validity of the paraxial approximation in the above-mentioned parameter regime. Using a rigorous solution to Helmholtz equation for a TM$_{01}$ beam as well as its paraxial counterpart, we highlight significant differences in the electron acceleration dynamics predicted by the exact and the paraxial fields. The exact fields are used to investigate the origin and importance of these differences.
A nonparaxial TM$_{01}$ beam propagating along the positive $z$ axis with its beam waist at $z=0$ is described in complex notation (i.e., $E={\textrm}{Re}\{\tilde{E}e^{j(\omega_0 t-\phi_0)}\}$, where $\omega_0=ck_0$ and $\phi_0$ are the beam frequency and phase constant) by the following field components in cylindrical coordinates $(r,\phi,z)$ [@sheppard99_ol; @april08b_optlett]: $$\begin{aligned}
&\tilde{E}_r = -j E_0 e^{-k_0 a} j_2 (k_0 \tilde{R}) \sin \tilde{\theta} \cos \tilde{\theta} \label{eq:exactEr} \\
&\tilde{E}_z = -\tfrac{2}{3}jE_0 e^{-k_0 a}\left[j_0 (k_0 \tilde{R}) + j_2(k_0 \tilde{R}) P_2(\cos \tilde{\theta}) \right] \label{eq:exactEz} \\
&\tilde{B}_\phi = B_0 e^{-k_0 a} j_1 (k_0 \tilde{R}) \sin \tilde{\theta} \label{eq:exactHphi} \, .\end{aligned}$$ Here $E_0=c B_0$ are amplitude parameters, $k_0$ is the beam wavenumber, $\tilde{R}=[r^2+(z+ja)^2]^{1/2}$ is the complex radius, $\cos \tilde{\theta}=(z+ja)/\tilde{R}$ defines the complex angle $\tilde{\theta}$, $j_n(\cdot)$ is the $n$-th order spherical Bessel function of the first kind, and $P_2(\cdot)$ is the Legendre polynomial of degree 2. The parameter $a$ is a real and positive constant called the confocal parameter. The latter may be used to characterize the degree of paraxiality of the beam since it is monotonically related to the Rayleigh range and beam waist size by the relation $z_R = k_0 w_0^2/2 = [\sqrt{1+(k_0a)^2}-1]/k_0$. The power carried along the $z$ axis by a nonparaxial TM$_{01}$ beam is [@april10_josaa] $$\begin{aligned}
\!\!\! P \!=\! \frac{P_0 e^{-2k_0 a}}{ k_0^3 a^3} \left[2k_0 a \sinh(2k_0 a)\!-\!\cosh(2k_0 a) \!+\! 1 \!-\! 2k_0^2 a^2 \right],\end{aligned}$$ where $P_0 = \pi |E_0|^2/8 \eta_0 k_0^2$. Note that the fields described in Eqs. – represent a rigorous solution to Helmholtz equation. We will thus refer to them as the *exact* TM$_{01}$ *fields*.
In the paraxial limit, namely when $k_0 a \gg 1$, the fields – can be expanded as power series of the parameter $\delta = 1/k_0 a$. Using the normalized coordinates $\rho = r(2a/k_0)^{-1/2}$ and $\zeta = z/a$ (since $z_R=k_0w_0^2/2 \approx a$), we find, up to terms of order $\delta^3$: $$\begin{aligned}
&\begin{aligned}
\tilde{E}_r = - \tfrac{1}{\sqrt{2}}E_0 \Big[ \rho f^2\delta^{3/2} - \big( 3j\rho f^3\!+\!3&\rho^3 f^4 \!-\!\tfrac{j}{2}\rho^5 f^5 \big)\delta^{5/2}\\
& \ \ + \mathcal{O} (\delta^{7/2}) \Big] \exp(j\Psi)
\end{aligned}\label{eq:corrEr}\\
&\begin{aligned}
\tilde{E}_z &= jE_0 \Big[\big( f^2 \!-\! j \rho^2 f^3 \big)\delta^2 \\
&- \big(jf^3 \!+\! 5\rho^2 f^4 \!-\! \tfrac{7j}{2} \rho^4 f^5 \!-\! \tfrac{1}{2}\rho^6 f^6 \big)\delta^3+ \mathcal{O} (\delta^{4}) \Big] \exp(j\Psi)
\end{aligned}\label{eq:corrEz} \\
&\begin{aligned}
\tilde{B}_\phi = -\tfrac{1}{\sqrt{2}}B_0 \Big[ \rho f^2\delta^{3/2} - \big( j\rho f^3\!+\!2&\rho^3 f^4 \!-\!\tfrac{j}{2}\rho^5 f^5 \big)\delta^{5/2}\\
& \ \ + \mathcal{O} (\delta^{7/2}) \Big] \exp(j\Psi) \ ,
\end{aligned}\label{eq:corrHphi}\end{aligned}$$ where $f=1/(\zeta+j)$ and $\Psi = - (\zeta/\delta + \rho^2f )$. The lowest-order terms of Eqs. – correspond to the well-known *paraxial* TM$_{01}$ *fields*, which are commonly used to analyze electron acceleration in RPLBs [@varin05_pre; @salamin06_pra; @varin06_pre; @fortin10_jpb; @wong10_optexpress; @singh11_prstab], $$\begin{aligned}
&\tilde{E}_r^{(0)} = -\tfrac{1}{\sqrt{2}}\, E_0 \rho f^2\delta^{3/2} \exp(j\Psi) = c B_\phi^{(0)}\label{eq:paraEr}\\
&\tilde{E}_z^{(0)} = jE_0 \big( f^2 - j \rho^2 f^3 \big)\,\delta^2 \exp(j\Psi) \label{eq:paraEz}\ .\end{aligned}$$
To simulate electron acceleration in RPLBs, we numerically integrate the Newton-Lorentz equations: $$\begin{aligned}
\frac{d{\mathbf}{r}}{dt} = {\mathbf}{v} \ ,\ \ \ \frac{d{\mathbf}{v}}{dt} = -\frac{e}{\gamma m_e}\left[{\mathbf}{E}+ {\mathbf}{v}\times{\mathbf}{B} - \frac{{\mathbf}{v}}{c^2}\left({\mathbf}{v}\cdot{\mathbf}{E}\right)\right] ,\end{aligned}$$ where $e$, $m_e$, ${\mathbf}{r}$, ${\mathbf}{v}$ are the electron’s charge, mass, position, velocity, respectively, and $\gamma=(1-|{\mathbf}{v}|^2/c^2)^{-1/2}$. We also suppose that the laser beam is pulsed, which we model by multiplying ${\mathbf}{E}$ and ${\mathbf}{B}$ by ${\textrm}{sech}(\xi/\xi_0)$, where $\xi = \omega_0 t - k_0 z$. This is to ensure that the fields satisfy Maxwell’s equation in the limit $\xi_0 \gg 1$ for any value of the phase $\xi$ [@wong10_optexpress]. Furthermore, we use $\lambda_0 = 800$ nm, although it can be readily shown that the results are scalable to any value of $\lambda_0$ [@wong10_optexpress; @marceau12_optlett].
We consider a cloud of electrons initially at rest in the $(r,z)$ plane outside the laser pulse. The initial position of the electrons are drawn randomly from a two-dimensional Gaussian distribution centered at the origin with standard deviation $\sigma_r = \sigma_z = \lambda_0/10$. The electrons are accelerated by a pulsed TM$_{01}$ beam with $k_0 a = 500$ (which corresponds to $w_0 \approx 5\lambda_0$), a value generally considered well inside the paraxial regime. Space-charge effects are neglected; each trajectory is computed independently. The relativistic electron bunch produced after performing the numerical simulation with the exact and the paraxial TM$_{01}$ fields from the same initial conditions are shown in Fig. \[fig:bunches\]. A comparison of the datasets in Fig. \[fig:bunches\](a)–(c) immediately shows an enormous difference in the bunch transverse extent; the angular divergence of the bunch accelerated by the exact fields is about 100 times larger than the bunch accelerated by the paraxial fields. Moreover, the electron bunch energy gain distribution is very different from one case to the other, as shown in Fig. \[fig:bunches\](d). The average energy gain obtained with the paraxial fields is however close to the energy gain near the optical axis with the exact fields, as shown in Fig. \[fig:bunches\](e). This agrees with previous results reported in [@marceau12_optlett] for on-axis acceleration. Since the near-axis electrodynamics is similar in both cases, the strong longitudinal compression predicted by the paraxial fields is also observed with the exact fields.
![(Color online) Electron bunch accelerated by a TM$_{01}$ pulsed beam with $P=10^{15}$ W, $k_0 a = 500$, $\xi_0=14.21$, $\phi_0=\pi$. The simulation was performed with: (i) the exact fields \[Eqs. –\], (ii) the paraxial fields \[Eqs. –\], and (iii) the corrected paraxial fields \[Eqs. – up to $\mathcal{O}(\delta^3)$\]. (a) Bunch snapshot and (b)–(c) close-up view of the front end; (d) energy gain distribution; (e) average energy gain versus final radial coordinate. The electrons are initially at rest outside the laser pulse in the $(r,z)$ plane and distributed randomly according to a Gaussian distribution centered at the origin with $\sigma_r = \sigma_z = \lambda_0/10$. The results are computed 15 ps after the passage of the beam at $z=0$. Only $N=200$ electrons are shown in (a)–(c), while $N=50000$ different initial conditions were used to obtain the results in (d) and (e). \[fig:bunches\]](./Marceau_Fig1abc "fig:"){width="\columnwidth"}\
![(Color online) Electron bunch accelerated by a TM$_{01}$ pulsed beam with $P=10^{15}$ W, $k_0 a = 500$, $\xi_0=14.21$, $\phi_0=\pi$. The simulation was performed with: (i) the exact fields \[Eqs. –\], (ii) the paraxial fields \[Eqs. –\], and (iii) the corrected paraxial fields \[Eqs. – up to $\mathcal{O}(\delta^3)$\]. (a) Bunch snapshot and (b)–(c) close-up view of the front end; (d) energy gain distribution; (e) average energy gain versus final radial coordinate. The electrons are initially at rest outside the laser pulse in the $(r,z)$ plane and distributed randomly according to a Gaussian distribution centered at the origin with $\sigma_r = \sigma_z = \lambda_0/10$. The results are computed 15 ps after the passage of the beam at $z=0$. Only $N=200$ electrons are shown in (a)–(c), while $N=50000$ different initial conditions were used to obtain the results in (d) and (e). \[fig:bunches\]](./Marceau_Fig1d "fig:"){width="\columnwidth"}\
![(Color online) Electron bunch accelerated by a TM$_{01}$ pulsed beam with $P=10^{15}$ W, $k_0 a = 500$, $\xi_0=14.21$, $\phi_0=\pi$. The simulation was performed with: (i) the exact fields \[Eqs. –\], (ii) the paraxial fields \[Eqs. –\], and (iii) the corrected paraxial fields \[Eqs. – up to $\mathcal{O}(\delta^3)$\]. (a) Bunch snapshot and (b)–(c) close-up view of the front end; (d) energy gain distribution; (e) average energy gain versus final radial coordinate. The electrons are initially at rest outside the laser pulse in the $(r,z)$ plane and distributed randomly according to a Gaussian distribution centered at the origin with $\sigma_r = \sigma_z = \lambda_0/10$. The results are computed 15 ps after the passage of the beam at $z=0$. Only $N=200$ electrons are shown in (a)–(c), while $N=50000$ different initial conditions were used to obtain the results in (d) and (e). \[fig:bunches\]](./Marceau_Fig1e "fig:"){width="\columnwidth"}\
To understand why such important discrepancies between the exact and the paraxial cases arise for off-axis electrons, it is instructive to look at the equations of motion. When electrons close to the optical axis interact with the laser beam, they are primarily accelerated in the positive $z$ direction by the longitudinal electric field. For an electron with $v_r \ll v_z$, the equation governing the radial velocity is approximately $$\begin{aligned}
\frac{dv_r}{dt} \approx \frac{-e}{\gamma m_e} \left(E_r - v_z B_\phi \right) \ . \label{eq:Lorentz}\end{aligned}$$ From Eq. , we see that the paraxial $\tilde{E}_r$ and $\tilde{B}_\phi$ field components are perfectly symmetric, i.e., $\tilde{E}_r^{(0)} = c\tilde{B}_\phi^{(0)}$. Therefore, according to Eq. , an electron travelling primarily along the optical axis at a relativistic velocity will feel a quasi-null force in the radial direction. This explains the observation of electron bunches with very narrow transverse extent when performing the simulations with the paraxial fields.
![(Color online) Normalized amplitude of (a) $-eE_z$ and (b) $-e(E_r - cB_\phi)$ at $t=0$ (exact fields). During sub-cycle acceleration, electrons will spend most of their time between a maximum of $-eE_z$ and the minimum located behind \[an example is shown by the rectangle in (a) and (b)\]. In this phase configuration, electrons travel at $v_z \approx c$ and feel, according to (b), a radial force directed outward. Beam parameters are the same as in Fig. \[fig:bunches\]. \[fig:fields\]](./Marceau_Fig2 "fig:"){width="1\columnwidth"}\
However, this argument does not hold in the exact case. According to Eqs. and , the exact $\tilde{E}_r$ and $\tilde{B}_\phi$ field components are not perfectly symmetric. The perturbative series and show that the symmetry between $\tilde{E}_r$ and $\tilde{B}_\phi$ is broken as soon as the first nonparaxial correction is introduced. Since $E_r$ and $B_\phi$ reach their maximum value at $\rho \approx [(1+\zeta^2)/2]^{1/2}$, the ratio of the maximum value of the radial field term in Eq. , $|E_r - cB_\phi|$, and the accelerating field, $|E_z|$, scales as $ \delta^{1/2}(1+\zeta^2)^{1/2}$. The radial force component may therefore not be neglected when $k_0 a \gtrsim 100$, and its relative importance becomes greater as $\zeta$ increases. Moreover, Fig. \[fig:fields\] demonstrates that the phase configuration in which electrons spend most of their time during sub-cycle acceleration is associated with a radial force directed outward. This explains why the angular divergence of the bunch accelerated by the exact solution is larger instead of smaller.
To verify explicitly if the discrepancy between the electron bunches in Fig. \[fig:bunches\] originates from the articifial symmetry between the paraxial $\tilde{E}^{(0)}_r$ and $\tilde{B}^{(0)}_\phi$ field components, we have performed the same numerical simulation with the *corrected paraxial fields*, i.e., the paraxial fields with the first nonparaxial correction to each component \[Eqs. – up to $\mathcal{O}(\delta^3)$\]. The results obtained are shown in Fig. \[fig:bunches\]. We immediately see a much better agreement with the exact case. While the final positions of the electrons do not match perfectly, the spatial configuration of the bunches and their energy gain distribution are much more similar. Note that the small differences in the energy gain distribution can be understood from Fig. \[fig:bunches\](e). Indeed, we see that the corrected paraxial results get less accurate as $\rho$ increases. This is a consequence of the fact that the magnitude of the nonparaxial corrections increases as we move further from the optical axis. Taking into account the first nonparaxial correction to $\tilde{E}_r$ and $\tilde{B}_\phi$ thus only offers a limited solution for electrons far from the optical axis.
Despite the fact that the validity of the paraxial approximation was never fully adressed in the literature for $k_0z_R$ values well above 100, an order of magnitude comparison may be performed with existing results. In [@karmakar07_lpb], an RPLB with $w_0 = 3\lambda_0$ and second-order field accuracy was used to accelerate electrons from a target of size comparable to the initial conditions used here. The resulting electron bunch angular divergence was reported to be approximately $\Delta \theta \approx 3^\circ$. With the parameters used in this Letter, we obtain $\Delta \theta \approx 5^\circ$ with the exact fields, which is of the same order of magnitude as in [@karmakar07_lpb], compared to $\Delta \theta < 0.1^\circ$ for the paraxial fields.
In conclusion, in a parameter regime where the paraxial approximation was previously considered valid, we have highlighted significant differences between the properties of electron bunches accelerated by paraxial and exact TM$_{01}$ beams. These differences originate from the symmetry between the paraxial $\tilde{E}_r^{(0)}$ and $\tilde{B}_\phi^{(0)}$ fields. This artificial symmetry is broken as soon as the first nonparaxial corrections to the electromagnetic field are taken into account, which allows to obtain more accurate results. The considerations presented in this Letter are also believed to apply to ultrashort pulses, since the relation $\tilde{E}_r = c\tilde{B}_\phi$ always holds for the paraxial TM$_{01}$ fields, regardless of the pulse duration [@varin05_pre]. Our study thus advocates that special care has to be taken when working under the paraxial approximation in the context of electron acceleration in RPLBs. It should be reminded that under relativistic conditions, nonparaxial field corrections may always yield major differences in the trajectories of off-axis electrons, even for very large values of $k_0 a$. Solutions as accurate as possible, ideally exact, should thus be used.
This research was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC), the Canadian Institute for Photonic Innovations (CIPI), and Calcul Québec Université Laval (computational resources).
[10]{}
K. Nakajima, Nature Phys. **4**, 92 (2008).
H.-P. Schlenvoigt, K. Haupt, A. Debus, F. Budde, O. Jäckel, S. Pfotenhauer, H. Schwoerer, E. Rohwer, J. G. Gallacher, E. Brunetti, R. P. Shanks, S. M. Wiggins, and D. A. Jaroszynski, Nature Phys. **4**, 130 (2008).
P. Baum and A. H. Zewail, Proc. Natl. Acad. Sci. USA **104**, 18409 (2007).
Y. I. Salamin, Phys. Rev. A **73**, 043402 (2006).
C. Varin, M. Pich[é]{}, and M. A. Porras, Phys. Rev. E **71**, 026603 (2005).
C. Varin and M. Pich[é]{}, Phys. Rev. E **74**, 045602(R) (2006).
Y. I. Salamin, Opt. Lett. **32**, 90 (2007).
P.-L. Fortin, M. Pich[é]{}, and C. Varin, J. Phys. B **43**, 025401 (2010).
L. J. Wong and F. X. K[ä]{}rtner, Opt. Express **18**, 25035 (2010).
K. P. Singh and M. Kumar, Phys. Rev. ST – Accel. Beams **14**, 030401 (2011).
V. Marceau, A. April, and M. Pich[é]{}, Opt. Lett. **37**, 2442 (2012).
S. Payeur, S. Fourmaux, B. E. Schmidt, J.-P. MacLean, C. Tchervenkov, F. L[é]{}gar[é]{}, M. Pich[é]{}, and J.-C. Kieffer, Appl. Phys. Lett. **101**, 041105 (2012).
A. Karmakar and A. Pukhov, Laser Part. Beams **25**, 371 (2007).
C. J. R. Sheppard and S. Saghafi, Opt. Lett. **24**, 1543 (1999).
A. April, Opt. Lett. **33**, 1563 (2008).
A. April, J. Opt. Soc. Am. A **27**, 76 (2010).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give a procedure that can be used to automatically satisfy invariants of a certain shape. These invariants may be written with the operations intersection, composition and converse over binary relations, and equality over these operations. We call these invariants [sentence]{}s that we interpret over graphs. For questions stated through sets of these sentences, this paper gives a semi-decision procedure we call graph saturation. It decides entailment over these [sentence]{}s, inspired on graph rewriting. We prove correctness of the procedure. Moreover, we show the corresponding decision problem to be undecidable. This confirms a conjecture previously stated by the author [@amperspiegelRAMICS].'
author:
- 'Sebastiaan J. C. Joosten'
bibliography:
- 'Amperspiegel.bib'
title: Finding models through graph saturation
---
This is an accepted preprint, which will be published in the Journal of Logical and Algebraic Methods in Programming (JLAMP).
Introduction
============
The question ‘what models does a set of formulas $\mathcal{T}$ have’ has practical relevance, as it is an abstraction of an information system: We interpret the data set stored in an information system at a certain point in time as a model, and each invariant of the system corresponds to a formula in $\mathcal{T}$. This correspondence is the core idea behind languages such as Ampersand [@AmpersandRAMICS], that define an information system this way. Users of an information system try to change the data set continually. These changes might violate the constraints. While Ampersand responds to such violations by rejecting the change, it would be convenient to automatically add data items such that all constraints are satisfied. The question then becomes: what data items should be added? We solve this question partially by means of a graph saturation procedure.
The question ‘does a set of formulas $\mathcal{T}$ have a model satisfying all formulas’ essentially asks whether $\mathcal{T}$ is free of contradictions. So far, we did not discuss the language in which we can write the formulas in $\mathcal{T}$. Several interesting problems arise when restricting the language in which we can write formulas: the satisfiability problem is obtained by restricting to disjunctions of positive and negative literals. Restricting to linear integer equalities, we obtain the linear programming problem. In this paper, we restrict those formulas to equalities over terms, in which terms are expressions of relations combined through the allegorical operations[^1]. We define *[sentence]{}* to be a formula over the restricted language considered in this paper (Definition \[def:tr\]).
Our interest in this language stems from experience in describing systems in Ampersand. All operations from relation-algebra are part of the Ampersand language. The operations considered here include only the most frequently used subset of those operations. Therefore, many of the formulas used in Ampersand will be sentences as considered in this work. We therefore consider this work a step towards an Ampersand system that helps the user find models.
Approach
--------
We give a short summary of the basic algorithm presented here, so we can better relate our approach to other literature, describe our contributions, and give the outline of this paper. Italicised words in the next paragraph are defined later.
The algorithm aims to determine whether there is a particular *model* for a set of *[sentence]{}s*, say $\mathcal{T}$, and is guaranteed to terminate if no such *model* exists. It proceeds to construct a (possibly infinite) *model* otherwise. The procedure has two phases: first, we translate the *[sentence]{}s* in $\mathcal{T}$ into a set of *[graph rule]{}s*. We then apply a saturation procedure on the *[graph rule]{}s*. This procedure creates a *chain* of *graphs*, whose limit is a *least consequence graph*. A *graph* contains a *conflict* if it has an edge with the label $\bot{}$. If a *least consequence graph* contains a *conflict*, then there is no *model* for $\mathcal{T}$. Otherwise, the *least consequence graph* corresponds to a *model* of $\mathcal{T}$, if the *[graph rule]{}s* correspond to $\mathcal{T}$ according to a straightforward *translation*. We abort the procedure as soon as a *conflict* arises, because we can be sure that no *models* for $\mathcal{T}$ exist in this case. A second question we can answer through the same algorithm is that of *entailment*: *entailment* is the question whether a *[sentence]{}* $\phi$ follows from a set of *[sentence]{}s* $\mathcal{T}$.
In an information system, a least consequence graph is a well suited to determine which data items to add: If conflict free, it corresponds to a graph that maintains the invariants. At the same, it only contains necessary consequences: it will not cause data items to be added that have nothing to do with the change the user made.
Related Work
------------
We compare the work in this paper to existing work in two ways: work it is similar to in motivation, and work it is similar to in implementation from an abstract perspective. In motivation, our research is closely related to the Alcoa tool, which we’ll discuss first. In approach, our methods are related to description logics and to graph rewriting, which we’ll discuss second.
#### The Alcoa Tool.
Our search for a reasoner for Ampersand is related to Alcoa [@Jackson2000AlcoaTA], which is the analyzer for Alloy [@jackson2002alloy], a language based on Z [@Z]. Like Ampersand, the languages Z and Alloy are based on relations. Alloy is a simplification of Z: it reduces the supported operations to a set that is small yet powerful. This paper differs from Alloy in the expressivity of its operations, however: Alloy allows writing full first order formula’s plus the Kleene-star, making it a language that is even more expressive than Ampersand. We compare to Alcoa because this work is similar in purpose.
In Alloy, a user may write assertions, which are formulas that the user believes follow from the specification. Alcoa tries to find counterexamples to those assertions, as well as a finite model for the entire specification. Unfortunately, several properties of the Alcoa tool hinder our purposes in Ampersand: Alcoa requires an upper bound on the size of (or number of elements in) the model. It does not perform well if this bound is too large. In a typical information system, the amount of data is well above what can be considered ‘too large’. As an additional complication, we cannot adequately predict the size of the model we might require. This is why we look at other methods for achieving similar goals.
#### Description Logics.
We can regard our procedure as a way to derive facts from previously stated facts: this is what happens in terms of [sentence]{}s between subsequent graphs in the chain we create. So called description logics are languages used in conjunction with an engine, that gives a procedure to learn new facts from previously learned facts, using declarative statements (or rules) in the corresponding description logic. For a good overview of description logics, see the book on that topic by Baader [@baader2003description].
A set of derivation rules is consistent if it has a model. For a highly expressive description logic such as OWL DL, determining consistency is undecidable. Still, a rule engine for OWL DL will happily try to learn new facts until a model is found. Users of OWL DL typically need to ensure that the stated derivation rules together with the rule engine give a terminating procedure. For many description logics, termination of its rule engine is syntactically guaranteed, and these logics are consequentially decidable. The description logic for which the language and implementation is closest to our language is the logic $\mathcal{EL}$ and its extensions proposed by Baader et al [@Baader:2005aa; @EL++]. Instead of using tableau-based procedures, as most description logics, it uses a saturation-based reasoner. Syntax of the derivation rules is limited to ensure termination of any saturation procedure: $\mathcal{EL}$ allows statements about unary relations using top, bottom, individual elements called ‘nominal’, and conjunction. Statements about binary relations use a different syntax, that can be translated into [sentence]{}s using composition, converse and the identity relation (but not necessarily vice-versa). By modeling $\mathcal{EL}$’s unary relations as binary relations that are a subset of the identity relation, all of $\mathcal{EL}$ and its extensions can be expressed through the [sentence]{}s described in this paper. In particular, the syntax of $\mathcal{EL}$ does not have disjunctions, thus eliminating the need for backtracking. In fact, $\mathcal{EL}$ is designed such that its consistency can be decided deterministically in polynomial time. Its extensions have different complexity bounds, but preserve polynomial runtime for the fragment that falls within $\mathcal{EL}$.
In our work, we do not work under the assumption of termination: neither the user or the syntax guarantees it. This allows us to use a richer language than one that is syntactically guaranteed to terminate. Despite this lack of termination, we do ensure termination in case of conflicts: a conflict will be found if our [sentence]{}s imply it. This allows the user to approach certain problems through any set of rules within the grammar, rather than just those sets for which the implementation is guaranteed to terminate. The implementation presented in our work applies [graph rule]{}s ‘fairly’ to ensure this. Fair application of rules is typically not required in the implementation of description logic engines.
#### Graph Rewriting.
A central concept in graph rewriting is that a pushout can be used to apply a *[graph rule]{}* on a graph, as described by Wolfram Kahl [@kahl2001relation]. The usual idea of such a pushout is that it models execution by removing a portion of the graph, and replacing it with the result of the execution step. Graph rewriting might then terminate when no rules can be applied anymore. Our approach diverges on this point: rather than execution, a step models learning a deducible conclusion. Rather than terminating when no step is possible, we are interested in the limit of the sequence of graphs. For this reason, the notions of weak pushout step and weak pushout don’t coincide exactly: we ensure that the sequence of graphs form a chain in order for the limit to exist.
The term saturation is borrowed from the saturation procedure in resolution procedures, introduced by Robinson in 1965 [@Robinson:1965aa]. His procedure solves an entailment problem over a certain language. As in his procedure, our procedure adds derivable facts iteratively.
Contributions and Paper Outline
-------------------------------
We mentioned how this paper contributes by comparing it to related work: Compared to the work on $\mathcal{EL}$, our approach allows [sentence]{}s in a richer language, and we present a translation to [graph rule]{}s to separate the semantics from the core of the implementation. Compared to the work on graph-rewriting, we present a new graph-based manipulation algorithm, and give an interpretation of those graphs as models for sets of [sentence]{}s.
We also relate the contribution of this paper to a paper presenting Amperspiegel [@amperspiegelRAMICS]. This earlier paper by the author conjectured that the problem whether no least consequence graph exists is undecidable. It also contains a procedure for finding such graphs, which it conjectures to be correct. We will show that the procedure in the paper is an instance of the variations of the procedure described here. To simplify the presentation of our results in this paper, the definition of a least consequence graph is slightly different here: A least consequence graphs always exist according to the definitions used in this paper. In the terminology of this paper, the conjecture just mentioned would be: the problem whether a least consequence graph contains a conflict is undecidable. This paper proves the stated result.
The procedure presented in this work is simpler than the one presented earlier. However, the latter can be obtained by applying optimizations to the former. We show correctness of the procedure, and show that the existence of a conflict free least consequence graph implies the existence of models for a set of sentences. Semi-decidability of consistency is not surprising in this setting: the logic we consider is less expressive than several logics for which semi-decidability is established. Our contribution lies in presenting an intuitive, flexible, graph-based algorithm that does not use backtracking.
The outline of this paper is as follows: we define the syntax and semantics of [sentence]{}s in Section \[sec:background\], and define the problems our procedure aims to solve: deciding consistency and entailment. Section \[sec:graphrules\] then introduces the heart of the procedure by defining least consequence graphs and indicating how to obtain them through [graph rule]{}s. Section \[sec:Rules\] connects these two, by giving a translation of [sentence]{}s to [graph rule]{}s. The procedure is given as an algorithm in Section \[sec:procedure\], and we indicate how to use the procedure to decide consistency and entailment. Before going to the conclusion, we indicate why we cannot hope to do better than giving a possibly non-terminating procedure, by proving undecidability in Section \[sec:undecidable\]. Conclusion and acknowledgements are in Section \[sec:conclusion\].
Background and Problem Statement {#sec:background}
================================
As this paper primarily deals with directed labeled graphs, we choose to use these graphs for the semantics of [sentence]{}s as well. There is no fundamental difference between this presentation and the usual binary relation based semantics usually presented as the canonical allegory (or as the canonical model for relation algebra). However, using graphs now simplifies our proofs later on, and makes it that we do not have to define them later. Graphs are defined as follows:
A directed labeled graph $G = (\mathcal{L},V,E)$ is given by a set of labels $\mathcal{L}$, a set of vertices $V$, and a set of edges $E \subseteq \mathcal{L} \times V \times V$. The set of all graphs with labels $\mathcal{L}$ is written as $\mathbb{G}_\mathcal{L}$. We write *graph* when we mean a directed labeled graph. We say that a graph is *finite* if both its set of vertices $V$ and its set of edges $E$ are finite. The cardinality of $V$ is written $|G|$. A graph with no vertices (and therefore no edges) is called *empty*, written $\mathbb{0}_\mathcal{L}$.
Terms are built inductively from relation symbols $\mathcal{L}$, combined with the operations $\_\sqcap\_$, $\_\fcmp\_$, and $\_{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}$. The operations stand for intersection, relational composition, and relational converse, respectively. The set of all terms over $\mathcal{L}$ is denoted as ${{\mathbb{E}_{\mathcal{L}}}}$. We use the same letter $\mathcal{L}$ to indicate labels in graphs, as well as relation symbols in terms. This notation is deliberately chosen because of the semantics given in Definition \[def:semantics\] below.
\[def:semantics\] For a graph $G=(\mathcal{L},V,E)$, the *semantics* of a term $\mathbb{e} \in {{\mathbb{E}_{\mathcal{L}}}}$, written as ${\left\lBrack\mathbb{e}\right\rBrack_{G}} \subseteq V \times V$, is as in representable relation algebra: $$\begin{aligned}
{\left\lBrackl\right\rBrack_{G}} &= \left\{(x,y) \mid (l,x,y) \in E \right\} \\
{\left\lBrack\mathbb{e}_1 \sqcap \mathbb{e}_2\right\rBrack_{G}} &= {\left\lBrack\mathbb{e}_1\right\rBrack_{G}} \cap {\left\lBrack\mathbb{e}_2\right\rBrack_{G}} \\
{\left\lBrack\mathbb{e} {^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\,\right\rBrack_{G}} &= \left\{ (y,x) \mid (x,y)\in{\left\lBrack\mathbb{e}\right\rBrack_{G}}\right\} \\
{\left\lBrack\mathbb{e}_1 \fcmp \mathbb{e}_2\right\rBrack_{G}} &= {\left\{ (x,y) \mid \exists z.\ (x,z)\in{\left\lBrack\mathbb{e}_1\right\rBrack_{G}} \ \wedge \ (z,y)\in{\left\lBrack\mathbb{e}_2\right\rBrack_{G}}\right\}}\end{aligned}$$
A [sentence]{} is the proposition stating that two terms are equal:
\[def:tr\] Given the terms $\mathbb{e}_1,\mathbb{e}_2 \in {{\mathbb{E}_{\mathcal{L}}}}$, the pair $(\mathbb{e}_1,\mathbb{e}_2)$ is a *[sentence]{}*, written $\mathbb{e}_1 = \mathbb{e}_2$. We write $\mathbb{e}_L \sqsubseteq \mathbb{e}_R$ for a [sentence]{} of the shape $\mathbb{e}_L = \mathbb{e}_L \sqcap \mathbb{e}_R$. We say that a [sentence]{} *holds* in graph $G$ if ${\left\lBrack\mathbb{e}_1\right\rBrack_{G}} = {\left\lBrack\mathbb{e}_2\right\rBrack_{G}}$, in which case we write: $G \vDash \mathbb{e}_1 = \mathbb{e}_2$. If $\mathcal{T}$ is a set of sentences, we say that it holds in $G$ if each of the sentences holds in $G$, written $G \vDash \mathcal{T}$.
\[lem:subseteq\] Let $\mathbb{e}_1,\mathbb{e}_2 \in {{\mathbb{E}_{\mathcal{L}}}}$, and $G\in\mathbb{G}_\mathcal{L}$. $$\begin{aligned}
G \vDash \mathbb{e}_1 \sqsubseteq \mathbb{e}_2 \quad &\Leftrightarrow \quad {\left\lBrack\mathbb{e}_1\right\rBrack_{G}} \subseteq {\left\lBrack\mathbb{e}_2\right\rBrack_{G}}\\
G \vDash \mathbb{e}_1 = \mathbb{e}_2 \quad &\Leftrightarrow \quad G \vDash \mathbb{e}_1 \sqsubseteq \mathbb{e}_2 \ \wedge \ G \vDash \mathbb{e}_2 \sqsubseteq \mathbb{e}_1\end{aligned}$$
We deviate slightly from allegories: First, we are working in an untyped setting, or put differently: in an allegory with only a single object. In ‘typed allegories’, allegories with more than one object, relational composition is a partial operation. This deviation is not fundamental: we are simply adding more terms to our language than would be there in the typed setting. A second deviation is that we have not introduced identity morphisms. We introduce the identity symbol ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$ by treating it as a symbol in $\mathcal{L}$. Our approach generalizes to multiple identity symbols, as one would expect in allegories with multiple objects, but this is out of scope in favor of a simplified presentation.
Apart from the identity symbol ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$, we also introduce bottom and top ($\bot$ and $\top$) as symbols in $\mathcal{L}$. In Definition \[def:standard\] we give the interpretation of these designated relation symbols, defining a graph as standard if it adheres to this interpretation.
\[def:standard\] We say that a set of labels $\mathcal{L}$ is *standard* with the (possibly empty) set of constant elements $\mathcal{C}$ if $\bot,\top,{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\in \mathcal{L}$ and $\mathcal{C}\subseteq\mathcal{L}$. We refer to elements in $\mathcal{C}$ simply as constants. Let $\mathcal{L}$ be a standard set of labels with the constants $\mathcal{C}$. A graph $G = (\mathcal{L},V,E)$ is called *standard* if $V\neq \{\}$, and: $$\begin{aligned}
{\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}} &= \{(x,x) \mid x\in V\} \\
{\left\lBrack\top\right\rBrack_{G}} &= \{(x,y) \mid x,y\in V\} \\
{\left\lBrack\bot\right\rBrack_{G}} &= \{\} \\
\forall c\in\mathcal{C}.\ {\left\lBrackc\right\rBrack_{G}} &= \{(c,c)\}\end{aligned}$$
This work looks at models for $\mathcal{T}$, and investigates whether $\mathcal{T}$ entails $\phi$. We can now give the definitions that necessary to make this precise.
\[def:model\] Let $\mathcal{T}$ be a set of [sentence]{}s over a standard set of labels $\mathcal{L}$ (with constants $\mathcal{C}$). We say that the graph $G\in \mathbb{G}_\mathcal{L}$ is a *model* for $\mathcal{T}$ if every [sentence]{} in $\mathcal{T}$ holds in $G$ and $G$ is standard. We say that $\mathcal{T}$ is *consistent* if such a graph exists. We may refer to any set of [sentence]{}s $\mathcal{T}$ as an instance of the consistency problem.
\[def:entails\] Let $\mathcal{T}$ be a set of [sentence]{}s over a standard set of labels $\mathcal{L}$, and let $\phi$ be a [sentence]{} over $\mathcal{L}$. We say that $(\mathcal{T},\phi)$ is an instance of the *entailment problem*. We say that $\mathcal{T}$ *entails* $\phi$ if for all standard graphs G, $G \vDash \mathcal{T}$ implies $G \vDash \phi$.
Our use of ‘standard’ in these definitions is not a restriction: given a graph $G$ over a language $\mathcal{L}$ with $\bot,\top,{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\not\in \mathcal{L}$, we can make it into a standard graph $G'$ over $\mathcal{L} \cup \{\bot,\top,{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\}$, choosing the constants $\mathcal{C}=\{\}$, and adding the edges according to Definition \[def:standard\]. Then $G \vDash \phi$ if and only if $G' \vDash \phi$ for $\phi \in {{\mathbb{E}_{\mathcal{L}}}}$, as $\phi$ cannot talk about $\bot,\top$ or ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$.
We prove a straightforward correspondence between the consistency problem and the entailment problem:
There is a standard graph $G$ such that $G \vDash \mathcal{T}$ if and only if $\mathcal{T}$ does not entail $\bot = \top$.
We first prove that if $\mathcal{T}$ entails $\bot = \top$, then there is no standard graph with $G \vDash \mathcal{T}$: A standard graph must have at least one vertex, say $v$. Then $(v,v)\in{\left\lBrack\top\right\rBrack_{G}}$, and $(v,v)\not\in{\left\lBrack\bot\right\rBrack_{G}}$, so ${\left\lBrack\bot\right\rBrack_{G}} \neq {\left\lBrack\top\right\rBrack_{G}}$. For the other direction: Suppose there is no standard graph with $G \vDash \mathcal{T}$, then entailment of any formula follows by definition.
We proceed with a small example of sentences, an entailment and a consistency problem. As an example, we make an administration of people and rooms. We use the label ${\tt i}$ to denote which room a person Inhabits, and ${\tt r}$ to denote which people are Roommates. We think of the labels in terms of their semantics: as binary relations. We show how these relations are connected by the sentence expressing: Two people are roommates if and only if they share a room: ${\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}$. This gives a one-sentence theory $\mathcal{T} = \{{\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\}$ on a standard set of labels that contains $\tt i$ and $\tt r$.
We ask ourselves if being a roommate is a transitive relation. That is, does $\mathcal{T}$ entail ${\tt r}\fcmp{\tt r} \sqsubseteq {\tt r}$ or not? The answer is negative. A possible counter-example our procedure may produce is a graph $G$ with: $$\begin{aligned}
{\left\lBrack\tt i\right\rBrack_{G}} &= \{(0,3),(1,4),(2,3),(2,4)\} \\
{\left\lBrack\tt r\right\rBrack_{G}} &= \{(0,0),(0,2),(1,1),(1,2),(2,0),(2,1),(2,2)\}\end{aligned}$$ In this example, $0,1,2$ are people, and $3,4$ are their rooms. While $0$ and $1$ are roommates of $2$, $0$ is not a roommate of $1$. Note that person $2$ has two rooms in this example. We may wish to forbid this: the sentence ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$ expresses that $\tt i$ is univalent (if two rooms are inhabited by the same person, those two must be the same room). Now $\mathcal{T} = \{{\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\ ,\ {\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\}$ entails ${\tt r}\fcmp{\tt r} \sqsubseteq {\tt r}$, and our procedure shows this, as we will demonstrate in Section \[sec:procedure\].
We elaborate on the same example for checking consistency, and add some constants to $\mathcal{L}$. Let $\mathcal{C} = \{\texttt{`Liz'},\texttt{`Jon'},\texttt{`Batcave'},\texttt{`Room 11'}\}$. Let’s say we want Liz and Jon to be roommates, and ask ourselves if that’s possible. That is, we wish to solve the consistency problem for: $$\begin{aligned}
\mathcal{T} = \{\ & {\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\\
,\ &{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\\
,\ &\texttt{`Liz'} \fcmp \top \fcmp \texttt{`Jon'} \sqsubseteq {\tt r} \}\end{aligned}$$ Our procedure then produces a graph like $G$ with: $$\begin{aligned}
{\left\lBrack\tt i\right\rBrack_{G}} &= \{(\texttt{`Liz'},0),(\texttt{`Jon'},0)\} \\
{\left\lBrack\tt r\right\rBrack_{G}} &= \{(\texttt{`Liz'},\texttt{`Jon'})\}\end{aligned}$$ Without going into details on why, we remark that our procedure comes up with a new room, here called $0$, even with the Batcave and Room 11 available. Finally, if we require Liz and Jon to be in their rooms of their choice, the Batcave and Room 11 respectively, our procedure detects that the requirements are no longer consistent. That is, the following theory is not consistent: $$\begin{aligned}
\mathcal{T} = \{\ & {\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\ ,\ {\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\\
,\ &\texttt{`Liz'} \fcmp \top \fcmp \texttt{`Jon'} \sqsubseteq {\tt r}\\
,\ &\texttt{`Liz'} \fcmp \top \fcmp \texttt{`Batcave'} \sqsubseteq {\tt i}\\
,\ &\texttt{`Jon'} \fcmp \top \fcmp \texttt{`Room 11'} \sqsubseteq {\tt i}
\}\end{aligned}$$
[Graph Rule]{}s and Consequence Graphs {#sec:graphrules}
======================================
This section defines a least consequence graph, and gives conditions on a chain of graphs that ensure that its limit is a least consequence graph. When a graph is a least consequence graph, we can use it to answer both the entailment problem and the consistency problem. The conditions on a chain of graphs tell us how [graph rule]{}s should be applied by possible implementations. Basically ‘least consequence graph’ characterises that all [graph rule]{}s are applied correctly and sufficiently. We define [graph rule]{}s in Definition \[def:gr\], least consequence graphs in Definition \[def:lcg\], and conclude the section with the conditions that give us a least consequence graph, proven in Lemma \[lem:fairchain\] and \[lem:pushoutchain\].
We introduce special notation for two basic operations on graphs: relabeling of vertices, and taking the union of two graphs. Suppose we have a function $f : V_1 \to V_2$, where $V_1$ is the set of vertices of some graph. We can apply the function on the corresponding graph, written $\hat{f}$: $$\hat{f}((\mathcal{L},V_1,E)) = \left(\mathcal{L},\{f(v) \mid v\in V_1\},\{(l,f(x),f(y)) \mid (l,x,y) \in E\}\right)$$ For taking the union of two graphs, we simply write $\cup$, defined as follows: $$(\mathcal{L}_1,V_1,E_1) \cup (\mathcal{L}_2,V_2,E_2) = (\mathcal{L}_1 \cup \mathcal{L}_2,V_1 \cup V_2,E_1 \cup E_2)$$ This leads to a natural definition of subgraph:
We say that $G_1$ *is a subgraph of* $G_2$ if $G_1 \cup G_2 = G_2$. It follows that a subgraph of a finite graph is again finite. If $G_1$ is a subgraph of $G_2$ and $G_1,G_2\in\mathbb{G}_\mathcal{L}$ for some $\mathcal{L}$, we write $G_1 \xrightarrow{\subseteq} G_2$.
In this article, we consider the set of labels $\mathcal{L}$ to be arbitrary but fixed. The relation ‘subgraph’ forms a complete lattice over $\mathbb{G}_\mathcal{L}$, which justifies the following definition:
Given a set of labels $\mathcal{L}$. We say that $S : \mathbb{N} \rightarrow \mathbb{G}_\mathcal{L}$ is a *chain* if for all $i \in \mathbb{N}$, $S(i)$ is a subgraph of $S(i+1)$. The union of all graphs in a chain, written $S(\infty)$, is called the *supremum*, defined as $S(\infty) = (\mathcal{L},\bigcup_{i} E_i,\bigcup_{i} V_i)$ with $S(i) = (\mathcal{L},E_i,V_i)$.
The way we use graph rewriting is most closely related to the single-pushout rewriting found in the literature (e.g. [@kahl2001relation]). In this approach, [graph rule]{}s are related through a morphism that is, for instance, a partial function. Vertices in the left hand side of the rule not related to the right hand side get removed upon application of the rule. Similarly, vertices on the right hand side get inserted. In our setting, we need the application of a rule to form a chain. To make sure that we can do this, we use ‘subgraph’ as a condition on [graph rule]{}s.
\[def:gr\] A pair of graphs $(L,R)$ is called a *[graph rule]{}* if $L$ is a subgraph of $R$, and $R$ is finite. We say that a set $\mathcal{R}$ is a *set of [graph rule]{}s with labels* $\mathcal{L}$ if each $(L,R)\in\mathcal{R}$ is a [graph rule]{}, and $L,R\in \mathbb{G}_\mathcal{L}$.
We proceed by giving an example of a [graph rule]{}, and do so visually. A graph can be drawn in the usual way. Figure \[fig:graph\] is an example of a graph with $k,l,m\in\mathcal{L}$. A picture does not specify the set of labels $\mathcal{L}$, only the set of edges and the set of vertices. An example of a [graph rule]{} is given in Figure \[fig:rulpair\] and \[fig:rulcompact\]. Using the subgraph condition allows us to draw a [graph rule]{} $(L,R)$ in a single figure, using small dots for nodes in $R$ but not in $L$, and big dots and solid edges for what is in $L$, and therefore in $R$.
[0.2]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$k$\] (2); (3) edge\[bend left\] node\[label=below:$m$\] (1); (2) edge node\[label=right:$l$\] (3); (1) edge \[loop left,distance=1cm,in=135,out=-135\] node\[label=$l$\] (1);
[0.45]{}$\left(\begin{array}{c}
\begin{tikzpicture}[-latex,semithick]
\node[](c) {};
\node[vertex,label=above:$0$](0) [left = 10pt of c] {};
\node[vertex,label=above:$1$](1) [right = 10pt of c] {};
\path (0) edge node[label=above:$l$] {} (1);
\end{tikzpicture}
\end{array},
\begin{array}{c}
\begin{tikzpicture}[-latex,semithick]
\node[](c) {};
\node[vertex,label=above:$0$](0) [left = 10pt of c] {};
\node[vertex,label=above:$1$](1) [right = 10pt of c] {};
\node[vertex,label=below:$2$](2) [below =of c] {};
\path (0) edge node[label=above:$l$] {} (1);
\path (0) edge node[label=below:$l$] {} (2);
\path (2) edge node[label=below:$l$] {} (1);
\end{tikzpicture}\end{array}
\right)$
[0.3]{}
\(c) ; (0) \[left = 10pt of c\] ; (1) \[right = 10pt of c\] ; (2) \[below =of c\] ; (0) edge\[thick\] node\[label=above:$l$\] (1); (0) edge\[dotted\] node\[label=below:$l$\] (2); (2) edge\[dotted\] node\[label=below:$l$\] (1);
\
[0.2]{}
\(0) ; (1) \[right =of 0\] ; (0) edge node\[label=above:$l$\] (1); (0) edge\[loop below,distance=1cm,in=-135,out=-45\] node\[label=above left:$l$ \] (0);
[0.45]{}
\(0) ; (1) \[right =of 0\] ; (0) edge node\[label=above:$l$\] (1); (1) edge\[loop below,distance=1cm,out=-135,in=-45\] node\[label=above right: $l$\] (1);
[0.3]{}
\(0) ; (1) \[right =of 0\] ; (0) edge node\[label=above:$l$\] (1); (0) edge\[loop below,distance=1cm,in=-135,out=-45\] node\[label=above left:$l$ \] (0); (0) \[below =of 0\]; (1) \[right =of 0\] ; (0) edge node\[label=above:$l$\] (1); (0) edge\[loop below,distance=1cm,in=-135,out=-45\] node\[label=above left:$l$ \] (0);
We present a saturation procedure, so we need to capture when a graph is ‘saturated’. For this purpose, we define ‘maintained’, which indicates that a rule is applied sufficiently in a graph. For defining ‘maintained’, we first define graph embeddings:
Let $G_1,G_2\in\mathbb{G}_\mathcal{L}$. If $\hat{f}(G_1) \xrightarrow{\subseteq} G_2$, then $(f,G_1,G_2)$ is an *embedding* of $G_1$ in $G_2$. In such case, we write $G_1 \xrightarrow{f} G_2$. We say that $G_1$ is *embedded* in $G_2$ if such an $f$ exists, written $G_1 \rightarrow G_2$. It follows immediately that $G \xrightarrow{f} \hat{f}(G)$.
We briefly explain our notations with the observation that embeddings form a category: its objects are graphs with labels $\mathcal{L}$, and its arrows are embeddings. Although $\_ \xrightarrow{\subseteq} \_ = \_ \xrightarrow{\lambda x . x} \_ $, note that $G_1 \xrightarrow{\subseteq} G_2$ is only the identity arrow if $G_1 = G_2$, which is why we avoid writing $\_ \xrightarrow{id}\_ $.
\[def:lcg\] A [graph rule]{} $(L,R)$ with $L = (\mathcal{L},V_L,E_L)$ is *maintained* in $G$ if for every embedding $L\xrightarrow{f}G$, there is an embedding $R\xrightarrow{g}G$ such that $f(v) = g(v)$ for all $v \in V_L$. If for a set of [graph rule]{}s $\mathcal{R}$, each [graph rule]{} in $\mathcal{R}$ is maintained in $G$, we say that $G$ is a *consequence graph* maintaining $\mathcal{R}$. If furthermore $S$ is a subgraph of $G$, and $(S,G)$ is maintained in each consequence graph maintaining $\mathcal{R}$, then $G$ is a *least consequence graph* of $S$ maintaining $\mathcal{R}$.
We use chains to find least consequence graphs. We look at two properties: ‘fairness’ and ‘weak pushout’, that help establish graphs to be a consequence graph and least, respectively. To get some intuition, and hopefully help dispel some overly optimistic conjectures, we look at some examples before defining these two properties.
We begin with an example of an embedding. Let $L = (\{k,l,m\},\{0,1\},\{(l,0,1)\})$ and $R = (\{k,l,m\},\{0,1,2\},\{(l,0,1),(l,0,2),(l,2,1)\})$ be graphs. Note that $(L,R)$ is the [graph rule]{} drawn in Figure \[fig:rulpair\]. We can embed $L$ into the graph $G_\mathrm{\ref{fig:graph}}$ as shown in Figure \[fig:graph\]. A corresponding embedding is $(f,L,G_\mathrm{\ref{fig:graph}})$ with $f(i) = a$ for $i\in\{0,1\}$. There is also an embedding for $R$: $(g,R,G_\mathrm{\ref{fig:graph}})$ with $g(i) = a$ for $i\in\{0,1,2\}$, which satisfies $g(i)=f(i)$ for $i\in\{0,1\}$. However, the [graph rule]{} is not maintained, as for the embedding $(f',L,G_\mathrm{\ref{fig:graph}})$ with $f'(0) = b,\ f'(1) = c$, there is no such $g$.
As an example of a consequence graph, let $\mathcal{R} = \{(L,R)\}$ with $(L,R)$ as defined above, and let $G_\mathrm{\ref{fig:consequence1}} = (\{k,l,m\},\{0,1\},\{(l,0,1),(l,0,0)\})$, as drawn in Figure \[fig:consequence1\]. Then $G_\mathrm{\ref{fig:consequence1}}$ is a consequence graph maintaining $\mathcal{R}$. It is, however, not a least consequence graph of $L$ maintaining $\mathcal{R}$, since Figure \[fig:consequence2\] gives a consequence graph maintaining $\mathcal{R}$ in which $(L,G)$ is not maintained. We believe every least consequence graph of $L$ maintaining $\mathcal{R}$ is infinite and even infinitely branching: loops in such consequence graphs would make that they are no longer ‘least’, and to every edge with label $l$ there need to be two more of such edges in order to maintain $\mathcal{R}$.
The graph $G_\mathrm{\ref{fig:consequence1}}$ defined above is an example of a least consequence graph of $G_\mathrm{\ref{fig:consequence1}}$ maintaining $\mathcal{R}$. Graph $G_\mathrm{\ref{fig:consequence3}}=(\{k,l,m\},\{0,1,2,3\},\{(l,0,1),(l,0,0),(l,2,3),(l,2,2)\})$, consisting of two disjunctive copies of $G_\mathrm{\ref{fig:consequence1}}$, is a least consequence graph too, see Figure \[fig:consequence3\]. If a least consequence graph is unique, it must be the empty graph.
From the definition of maintained it follows that if $L \xrightarrow{\subseteq} M \xrightarrow{\subseteq} R$ and $(L,R)$ is maintained in $G$, then $(L,M)$ is maintained in $G$ too. Consequently, if $(L,R)$ is maintained in a least consequence graph of $L$ maintaining $\mathcal{R}$, then $(L,R)$ is maintained in every consequence graph maintaining $\mathcal{R}$.
The following definition gives a sufficient condition to reach a consequence graph:
Given a set of [graph rule]{}s $\mathcal{R}$ and a chain $S$. We say that $S$ is a *fair chain* for $\mathcal{R}$ if for each [graph rule]{} $(L,R)\in \mathcal{R}$ and for each embedding $L\xrightarrow{f}S(i)$ there exists a $j\in\mathbb{N}$ and an embedding $R\xrightarrow{g}S(j)$ with $f(v) = g(v)$ for all $v$ in the set of vertices of $L$.
\[lem:fairchain\] If $S$ is a fair chain for $\mathcal{R}$, $S(\infty)$ is a consequence graph maintaining $\mathcal{R}$.
By definition, $S(\infty)$ is a consequence graph if we can show that $R$ is embedded in $S(\infty)$ for every $L$ that is embedded in it. Take such an embedding $L\xrightarrow{f}S(\infty)$. Then for each edge $(l,u,v)$ of $L$ there is an $i$ such that $(l,f(u),f(v))$ is an edge in $S(i)$. Take the largest such $i$, then $f$ embeds $L$ in $S(i)$, and therefore $g$ embeds $R$ in $S(j)$ for some $j$ with $f(v) = g(v)$, so $g$ also embeds $R$ in $S(\infty)$.
We define [weak pushout step]{} as an upper limit to each step, to ensure that a consequence graph found as a supremum of a chain built out of these steps is also a least consequence graph.
Let $G_1$ and $G_2$ be graphs in $\mathbb{G}_\mathcal{L}$, and let $(L,R)$ be a [graph rule]{}. We say that $(G_1,G_2)$ is a *[weak pushout step]{}* for $(L,R)$ if the following hold:
- $G_1$ is a subgraph of $G_2$.
- There are embeddings $L\xrightarrow{f}G_1$ and $R\xrightarrow{g}G_2$ such that $f(v)=g(v)$ for all vertices in $L$.
- If there are embeddings $G_1\xrightarrow{f'}G$ and $R\xrightarrow{g'}G$ such that $f'(f(v)) = g'(v)$ for all vertices in $L$, then there is an embedding $G_2\xrightarrow{h}G$ such that $f'(v) = h(v)$ for all vertices in $G_1$.
Like in our variation of [graph rule]{}s, we use a [weak pushout step]{} as a variation of the categorical pushout[^2] that is typically used in graph rewriting, to ensure that chains are formed. In such a (weak) pushout, the requirement of subgraphs is missing, making the entire definition symmetrical ($G_1$ and $R$ can be switched). A pushout, as compared to a weak pushout, additionally requires that the embeddings $f'$ and $g'$ at the end of our definition, is unique. These subtle differences arise out of our need to form chains, which aren’t typical structures in graph rewriting.
\[def:pushoutchain\] Let $S$ be a chain with $S(i) = (\mathcal{L},E_i,V_i)$, and let $\mathcal{R}$ be a set of [graph rule]{}s. If for each $i$, either $S(i) = S(i+1)$ or there exists an $r\in\mathcal{R}$ such that $(S(i),S(i+1))$ is a [weak pushout step]{} for $r$, then $S$ is a *simple [weak pushout chain]{}* under $\mathcal{R}$. *[Weak pushout chain]{}s* are inductively defined:
1. every simple [weak pushout chain]{} under $\mathcal{R}$ is a [weak pushout chain]{} under $\mathcal{R}$,
2. if for each $i$, there exists an $s$, which is a [weak pushout chain]{} under $\mathcal{R}$ with $s(0) = S(i)$ and $s(\infty) = S(i+1)$, then $S$ is a [weak pushout chain]{} under $\mathcal{R}$, \[enum:inductive\]
3. nothing else is a [weak pushout chain]{}.
For most of this paper, it suffices to consider simple [weak pushout chain]{}s.
There is a way to draw [weak pushout step]{}s that is convenient in practice, although it can leave parts implicit. On a [weak pushout step]{} $(G_1,G_2)$ for $(L,R)$, as drawn in Figure \[fig:wps\], large vertices indicate vertices in the image of $f$ for $L\xrightarrow{f}G_1$. The applied [graph rule]{} is that of Figure \[fig:rulcompact\]. Edges in $\hat{f}(L)$ are drawn slightly thicker. The corresponding $\hat{g}$ for $R\xrightarrow{g}G_2$ is drawn with dotted lines. Since the drawing is of a [weak pushout step]{}, small vertices connected to dotted lines are in $G_2$ but not in $G_1$. The graph $G_1$ is the graph of Figure \[fig:graph\], and $G_2$ is the graph in Figure \[fig:graph2\].
[0.4]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$k$\] (2); (3) edge\[bend left\] node\[label=below:$m$\] (1); (2) edge\[thick\] node\[label=right:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=$l$\] (1); (x) \[right=of 2\] ; (2) edge\[dotted\] node\[label=below:$l$\] (x); (x) edge\[dotted\] node\[label=below:$l$\] (3);
[0.4]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$k$\] (2); (3) edge\[bend left\] node\[label=below:$m$\] (1); (2) edge node\[label=right:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=$l$\] (1); (x) \[right=of 2\] ; (2) edge node\[label=below:$l$\] (x); (x) edge node\[label=below:$l$\] (3);
A [weak pushout chain]{} does not necessarily have a consequence graph as its supremum: we can construct a [weak pushout chain]{} with $S(i) = G$ for any graph $G$. However, the following holds:
\[lem:pushoutchain\] If $S$ is a [weak pushout chain]{} under $\mathcal{R}$ and $S(\infty)$ is a consequence graph maintaining $\mathcal{R}$, then $S(\infty)$ is a least consequence graph of $S(0)$ maintaining $\mathcal{R}$.
Let $G$ be a consequence graph. We first consider the case in which $S$ is a simple [weak pushout chain]{}. By induction on $i$, we prove that $(S(0),S(i))$ is maintained in $G$: For $i=0$, $(S(0),S(0))$ is trivially maintained in any graph. For $S(i+1)$, assume $(S(0),S(i))$ is maintained in $G$ by induction. If $S(i+1) = S(i)$, then $S(i+1)$ is trivially embedded in $G$. If $S(i)\neq S(i+1)$, then $(S(i),S(i+1))$ is a [weak pushout step]{} for some $(L,R) \in \mathcal{R}$. Given an embedding $S(0)\rightarrow G$, as $L$ is embedded in $S(i)$, transitively $L$ is embedded in $G$. Since $G$ is a consequence graph, $R$ is embedded in $G$ such that, by definition, there exists an embedding of $S(i+1)$ into $G$. We conclude that for all $i$, $S(i)$ is embedded in $G$. To conclude that $S(\infty)$ is also embedded in $G$, note that the individual embeddings of $S(i)$ in $G$ have a limit (each embedding function is contained in the next by $f'(v) = h(v)$). The case in which the [weak pushout chain]{} $S$ is not simple follows inductively from composing embeddings. Therefore $S(\infty)$ is a least consequence graph.
A chain that is both fair and a [weak pushout chain]{} is called a fair [weak pushout chain]{}. A fair [weak pushout chain]{} has a least consequence graph as its supremum. This gives a way to create least consequence graphs, which we’ll come back to in Section \[sec:procedure\].
Translation between [Sentence]{}s and [Graph Rule]{}s {#sec:Rules}
=====================================================
This section shows how to turn [sentence]{}s into [graph rule]{}s. For every [sentence]{}, there is a corresponding [graph rule]{} that is maintained if and only if the [sentence]{} holds. This allows us to use [graph rule]{}s in order to reason about [sentence]{}s. We introduce a translate function ${\mathcal{E}_\mathcal{L}}: {{\mathbb{E}_{\mathcal{L}}}}\rightarrow \mathbb{G}_\mathcal{L}$ in Definition \[def:translation\] to make precise which graph belongs to a term. Lemma \[lem:translatesound\] states how the two correspond in the case of [sentence]{}s of the shape $\_ \sqsubseteq \_$. Using Lemma \[lem:subseteq\], this means we can encode a set of [sentence]{}s as a set of [graph rule]{}s.
\[def:translation\] Given a term $\mathbb{e}$, we say that ${\mathcal{E}_\mathcal{L}}(\mathbb{e})$ is the *translation* of $\mathbb{e}$. We define ${\mathcal{E}_\mathcal{L}}: {{\mathbb{E}_{\mathcal{L}}}}\rightarrow \mathbb{G}_\mathcal{L}$ as follows: $$\begin{aligned}
{\mathcal{E}_\mathcal{L}}(l)&=\left(\mathcal{L},\{0,1\},\{(l,0,1)\}\right)\\
{\mathcal{E}_\mathcal{L}}(\mathbb{e} {^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{})&=\hat{f}({\mathcal{E}_\mathcal{L}}(\mathbb{e})) \text{\quad with $f(v) = 1 - v$ for $v<2$ and $f(v) = v$ for $v\geq 2$.}\\
{\mathcal{E}_\mathcal{L}}(\mathbb{e}_1 \fcmp \mathbb{e}_2)&=\hat{f_1}({\mathcal{E}_\mathcal{L}}(\mathbb{e}_1)) \cup \hat{f_2}({\mathcal{E}_\mathcal{L}}(\mathbb{e}_2))\\
&\qquad\text{with $f_1(0) = 0$ and $f_1(v) = v + \left|{\mathcal{E}_\mathcal{L}}(\mathbb{e}_2)\right| - 1$ for $v\neq 0$,}\\
&\qquad\text{and $f_2(0) = \left|{\mathcal{E}_\mathcal{L}}(\mathbb{e}_2)\right|$ and $f_2(v) = v$ for $v\neq 0$.}\\
{\mathcal{E}_\mathcal{L}}(\mathbb{e}_1 \sqcap \mathbb{e}_2)&={\mathcal{E}_\mathcal{L}}(\mathbb{e}_1) \cup \hat{f}({\mathcal{E}_\mathcal{L}}(\mathbb{e}_2))\\
&\qquad\text{with $f(v) = v$ for $v<2$ and $f(v) = v + \left|{\mathcal{E}_\mathcal{L}}(\mathbb{e}_1)\right| - 2$ for $v\geq 2$.}\end{aligned}$$ For notational convenience, ${\mathcal{E}_\mathcal{L}}\left(\mathbb{e}_L \sqsubseteq \mathbb{e}_R\right) = \left({\mathcal{E}_\mathcal{L}}(\mathbb{e}_L),{\mathcal{E}_\mathcal{L}}(\mathbb{e}_L \sqcap \mathbb{e}_R)\right)$. It follows that ${\mathcal{E}_\mathcal{L}}\left(\mathbb{e}_L \sqsubseteq \mathbb{e}_R\right)$ is a [graph rule]{}.
As an example of how the translation works, the graphs in Figure \[fig:rulpair\] are ${\mathcal{E}_\mathcal{L}}(l)$ and ${\mathcal{E}_\mathcal{L}}\left(l\sqcap l\fcmp l\right)$ respectively. As a whole, the [graph rule]{} in Figure \[fig:rulpair\] is ${\mathcal{E}_\mathcal{L}}\left(l\sqsubseteq l\fcmp l\right)$.
The vertices $0$ and $1$ of ${\mathcal{E}_\mathcal{L}}(\mathbb{e})$ can intuitively be understood as the variables $x$ and $y$ as in Definition \[def:semantics\]. Lemma \[lem:translate\] makes this precise:
\[lem:translate\] $(v_0,v_1) \in {\left\lBrack\mathbb{e}\right\rBrack_{G}}$ if and only if there is an $f$ such that ${\mathcal{E}_\mathcal{L}}(\mathbb{e})\xrightarrow{f}G$ with $f(i) = v_i$ for $i<2$.
The statement follows by induction on $\mathbb{e}$, using that the vertices in ${\mathcal{E}_\mathcal{L}}(\mathbb{e})$ are $\left\{i \mid i \in \mathbb{N} \wedge i<\left|{\mathcal{E}_\mathcal{L}}(\mathbb{e})\right|\right\}$.
We can use Lemma \[lem:translate\] to show a connection between [graph rule]{}s and [sentence]{}s:
\[lem:translatesound\] A [sentence]{} $\mathbb{e}_L \sqsubseteq \mathbb{e}_R$ holds in $G$ if and only if $\left({\mathcal{E}_\mathcal{L}}(\mathbb{e}_L), {\mathcal{E}_\mathcal{L}}(\mathbb{e}_L \sqcap \mathbb{e}_R)\right)$ is maintained in $G$.
Suppose the [sentence]{} holds in $G$, and ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_L)\xrightarrow{f}G$. It follows from Lemma \[lem:translate\] that $(f(0),f(1)) \in {\left\lBrack\mathbb{e}_L\right\rBrack_{G}}$. As the [sentence]{} holds, $(f(0),f(1)) \in {\left\lBrack\mathbb{e}_R\right\rBrack_{G}}$. Using Lemma \[lem:translate\], take $g$ with ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_R)\xrightarrow{g}G$ and $f(v) = g(v)$ for $v < 2$. Following Definition \[def:translation\], construct $g'$ such that ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_L \sqcap \mathbb{e}_R)\xrightarrow{g'}G$ and $f(v) = g'(v)$ for $v$ in the vertices of ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_L)$.
For the other direction, suppose $\left({\mathcal{E}_\mathcal{L}}(\mathbb{e}_L),{\mathcal{E}_\mathcal{L}}(\mathbb{e}_L \sqcap \mathbb{e}_R)\right)$ is maintained in $G$, and let $(x,y) \in {\left\lBrack\mathbb{e}_L\right\rBrack_{G}}$. By Lemma \[lem:translate\], there is an $f$ such that ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_L)\xrightarrow{f}G$ with $f(0) = x$ and $f(1) = y$. Since the [graph rule]{} is maintained, there is a $g$ such that ${\mathcal{E}_\mathcal{L}}(\mathbb{e}_L \sqcap \mathbb{e}_R)\xrightarrow{g}G$ with $g(0)=f(0)=x$ and $g(1)=f(1)=y$. Again using Lemma \[lem:translate\], $(x,y)\in{\left\lBrack\mathbb{e}_L \sqcap \mathbb{e}_R\right\rBrack_{G}} = {\left\lBrack\mathbb{e}_L\right\rBrack_{G}} \cap {\left\lBrack\mathbb{e}_R\right\rBrack_{G}} \subseteq {\left\lBrack\mathbb{e}_R\right\rBrack_{G}}$, so the [sentence]{} holds in $G$.
We use [graph rule]{}s to deal with the requirements in Definition \[def:standard\] to make a graph standard, in a way similar to Lemma \[lem:translatesound\]. We give a set of [graph rule]{}s that make checking if a standard graph exists easy: A standard graph exists provided that ${\left\lBrack\bot\right\rBrack_{G}} = \{\}$, and that a set of additional [graph rule]{}s, which we will call the standard-rules, is maintained. This motivates the following definitions:
\[def:conflict\] Let $\bot \in \mathcal{L}$. The relation symbol $\bot$ stands for an empty relation. A graph for which ${\left\lBrack\bot\right\rBrack_{G}} = \{\}$ is *conflict free*. If $G = (\mathcal{L},V,E)$ is conflict free, we have $\forall x y. (\bot,x,y) \not\in E$, so we call any edge $(\bot,x,y)$ a *conflict*.
Let $\top \in \mathcal{L}$. The relation symbol $\top$ stands for the full relation. We refer to the [graph rule]{} $((\mathcal{L},\{0,1\},\{\}),(\mathcal{L},\{0,1\},\{(\top,0,1)\}))$ as the *top-rule*, since any graph $G = (\mathcal{L},V,E)$ satisfies ${\left\lBrack\top\right\rBrack_{G}} = \{(x,y) \mid x,y \in V\}$ if and only if $G$ maintains the top-rule.
Let $\bot,\top \in \mathcal{L}$. The [graph rule]{} $((\mathcal{L},\{\},\{\}),(\mathcal{L},\{0\},\{\}))$ is called the *nonempty-rule*. A graph $G = (\mathcal{L},V,E)$ maintains the nonempty-rule if and only if $V \neq \{\}$.
[0.2]{}
\(0) ; (1) \[right =of 0\]; (0) edge\[dotted\] node\[label=below:$\top$\] (1);
[0.25]{}
\(0) ; (0) edge\[dotted,in=135,out=-135,loop,distance=1cm\] node\[label=left :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$\] (0);
[0.25]{}
\(0) ; (1) \[right =of 0\]; (1) edge\[thick,bend left\] node\[label=below :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (0); (0) edge\[dotted,bend left\] node\[label=above :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (1);
[0.25]{}
\(0) ; (2) \[right =of 0\]; (1) \[right =of 2\]; (0) edge\[thick\] node\[label=above :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (2); (2) edge\[thick\] node\[label=above :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (1); (0) edge\[dotted,bend right\] node\[label=below :${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (1);
A conflict-free graph $G$ that maintains the top-rule, satisfies ${\left\lBrack\top\right\rBrack_{G}} \neq {\left\lBrack\bot\right\rBrack_{G}}$ if and only if it maintains the nonempty-rule.
The relation symbol ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$ models the identity relation $\{(x,x) \mid x \in V\}$. However, we do not let ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}}$ represent this relation directly. Instead, we let ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$ stand for an equivalence relation and ensure that we can make a graph based on equivalence classes, in which ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}} = \{(x,x) \mid x \in V\}$ holds.
Given a set of relation symbols $\mathcal{L}$, we say that the following set of [graph rule]{}s are the *identity-rules* for $\mathcal{L}$: $$\begin{aligned}
\big((\mathcal{L},\{0\},\{\})&\ ,\ \ (\mathcal{L},\{0\},\{({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}},0,0)\})\big)\label{idt:reflexive} \\
{\mathcal{E}_\mathcal{L}}\big({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} &\sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\big)\label{idt:symmetric} \\
{\mathcal{E}_\mathcal{L}}\big({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\fcmp {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}&\sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\big)\label{idt:transitive} \\
\forall l\in\mathcal{L}.\ {\mathcal{E}_\mathcal{L}}\big({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\fcmp l \fcmp {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}&\sqsubseteq l \big)\label{idt:eachl}\end{aligned}$$
Identity-rules (\[idt:reflexive\]) to (\[idt:eachl\]) can be understood as ensuring ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$ is reflexive, symmetric, transitive, and a congruence respectively. The identity-rules hold under the standard semantics of ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$, that is: if for some graph $G = (\mathcal{L},E,V)$, we have ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}} = \{(x,x) \mid x \in V\}$ then the identity-rules are maintained in $G$. The following lemma speaks about the other direction:
\[lem:identrules\] Let $G = (\mathcal{L},V,E)$ be a graph in which the identity-rules for $\mathcal{L}$ are maintained. There is an idempotent $f$ such that $\hat{f}(G) \xrightarrow{\subseteq} G$, and ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{\hat{f}(G)}} = \{(f(x),f(x)) \mid x \in V\}$.
Since the first three identity-rules for $\mathcal{L}$ are maintained in $G$, ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}}$ is an equivalence relation on $V$. Let $f$ be some function that takes a canonical element from the equivalence class. It follows that ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{\hat{f}(G)}} = \{(f(x),f(x)) \mid x \in V\}$, and it remains to be shown that $\hat{f}(G) \xrightarrow{\subseteq} G$. For the vertices, this is immediate. For the edges: For all $(l,x,y) \in E$ we show that $(l,f(x),f(y))\in E$. By our choice of $f$, $({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}},f(x),x)\in E$ and $({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}},y,f(y))\in E$. Suppose $(l,x,y)\in E$. Since Identity-rule (\[idt:eachl\]) for $l$ is maintained in $G$, we get $(l,f(x),f(y)) \in E$. Therefore $\hat{f}(G) \xrightarrow{\subseteq} G$.
Lemma \[lem:identrules\] gives us exactly the desired semantics for ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$: for $(\mathcal{L},V',E') = \hat{f}(G)$, we have ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{\hat{f}(G)}} = \{(x,x) \mid x \in V'\}$. Furthermore, it states that $\hat{f}(G)$ and $G$ are mutually embedded ($G \rightarrow \hat{f}(G)$ holds for all $f$).
We now proceed to introduce constants, through a set of [sentence]{}s. This characterisation is similar to how points are characterized in relation algebra, see for instance work by Schmidt and Ströhlein [@Schmidt:1985aa]. If $c$ is a constant, then $p = c\fcmp \top$ is a point (sometimes called a right ideal). The corresponding constant can be retrieved from a point: $c = p \fcmp p{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}$. Our presentation here in terms of constants rather than points is a matter of personal preference. These rules state that the relation ${\left\lBrackc\right\rBrack_{G}}$ should be nonempty, the cross-product of two sets, and a subset of the identity relation. Finally, we state that for two different constants, ${\left\lBrackc_1\right\rBrack_{G}}$ and ${\left\lBrackc_2\right\rBrack_{G}}$ should be non-overlapping.
Let $\mathcal{L}$ be a standard set of labels with constants $\mathcal{C}$, we say that the following set of [graph rule]{}s are the *constant-rules* for $\mathcal{C}$: $$\begin{aligned}
\forall c\in\mathcal{C}.&& {\mathcal{E}_\mathcal{L}}\big(\top &\sqsubseteq \top \fcmp c \fcmp \top \big)\label{pr:exists} \\
\forall c\in\mathcal{C}.&& {\mathcal{E}_\mathcal{L}}\big(c \fcmp \top \fcmp c &\sqsubseteq c \big) \label{pr:crossprod} \\
\forall c\in\mathcal{C}.&& {\mathcal{E}_\mathcal{L}}\big(c &\sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\big)\label{pr:ident} \\
\forall c_1, c_2\in\mathcal{C}.\ c_1 \neq c_2 \Rightarrow && {\mathcal{E}_\mathcal{L}}\big(c_1 \fcmp c_2 &\sqsubseteq \bot \big)\label{pr:neq}\end{aligned}$$ The top-, nonempty-, identity-, and constant-rules together are called the *standard-rules* for $\mathcal{C}$ and $\mathcal{L}$, written $\mathcal{S}_{\mathcal{C},\mathcal{L}}$.
Similar to our treatment of ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$, we would like to find an $f$ such that ${\left\lBrackc\right\rBrack_{\hat{f}(G)}} = \{(c,c)\}$. The $f$ of Lemma \[lem:identrules\] gives us a graph that is isomorphic to one in which $\forall c\in\mathcal{C}.\ {\left\lBrackc\right\rBrack_{\hat{f}(G)}} = \{(c,c)\}$ holds, provided that $G$ is conflict free and maintains the standard-rules. Lemma \[lem:existsgraph\] says that, for finding a model with ‘standard semantics’, it suffices to find a conflict free graph that maintains the standard-rules.
\[lem:existsgraph\] Let $\mathcal{L}$ be a standard set of labels with constants $\mathcal{C}$. Let $\mathcal{T}$ be a set of [sentence]{}s over $\mathcal{L}$ of the shape $\_ \sqsubseteq \_$. We define $\mathcal{R} = \mathcal{S}_{\mathcal{C},\mathcal{L}} \cup \{{\mathcal{E}_\mathcal{L}}(t)\mid t\in \mathcal{T}\}$. Let $G'$ be a conflict free consequence graph maintaining $\mathcal{R}$, then there is a graph $G = (\mathcal{L},E,V)$, and functions $f$ and $g$ such that:
1. $G = \hat{f}(\hat{g}(G)) =\hat{f}(G')$, and $\hat{g}(G) \xrightarrow{\subseteq} G'$.
2. The graph $G$ is standard.
3. Every [sentence]{} in $\mathcal{T}$ holds in $G$.
We begin the proof by constructing $G$ and $f$, based on $G' = (\mathcal{L},E',V')$. By Lemma \[lem:identrules\], there is an idempotent function $h$ with ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{\hat{h}(G')}} = \{(h(x),h(x)) \mid x \in V'\}$. Top- and nonempty-rules are maintained in $G'$, so by constant-rule (\[pr:exists\]), there are vertices $v_1,v_2\in V'$ with $(c,v_1,v_2)\in E'$ for each $c\in\mathcal{C}$. Let $m : \mathcal{C} \to V'$ such that for each $c$, $\exists v\in V'.\ (c,m(c),v)\in E'$, therefore $\exists v\in V'.\ (c,h(m(c)),h(v))\in E'$. Using constant-rule (\[pr:ident\]), it follows that $h(m(c)) = h(v)$, so $(c,h(m(c)),h(m(c))) \in E'$. From constant-rule (\[pr:neq\]) and that $G'$ is conflict free, we get $(c_1,h(m(c_2)),h(m(c_2)))\not\in E'$ iff $c_1\neq c_2$. We conclude that $h \circ m$, the function that maps $c\in\mathcal{C}$ to $h(m(c))$, is injective, so $m$ is injective. Therefore, there is a $V$ and an $m'$ with $\mathcal{C}\subseteq V$ such that $m' : V \to V'$ is bijective and $m(c) = m'(c)$ for $c\in \mathcal{C}$. Let $f = h \circ m'$, defining $G = \hat{f}(G')$. Let $g$ be the inverse of $m'$, giving $\hat{f}(\hat{g}(G)) = \hat{h}(G) = G$ since $h$ is idempotent. We have ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}} = \{(x,x) \mid x\in V\}$ by our choice of $h$. Also $G$ is a consequence graph of $\mathcal{R}$ since $G = \hat{f}(G')$ and $G'$ is a consequence graph of $\mathcal{R}$. From $(c,h(m(c)),h(m(c))) \in E'$ we get $(c,c)\in{\left\lBrackc\right\rBrack_{G}}$. Using constant-rule (\[pr:crossprod\]) and constant-rule (\[pr:ident\]), now with ${\left\lBrack{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}\right\rBrack_{G}} = \{(x,x) \mid x\in V\}$, we get $\{(c,c)\}={\left\lBrackc\right\rBrack_{G}}$. All properties now follow.
A Procedure to Find a Standard Graph {#sec:procedure}
====================================
A set of [sentence]{}s is satisfiable if and only if there is no $i$ such that $S(i)$ contains a conflict in a corresponding fair [weak pushout chain]{}. This follows from the previous sections as follows: Given a set of [sentence]{}s $\mathcal{T}'$ with relation symbols $\mathcal{L}$, Lemma \[lem:subseteq\] shows that we can find an equivalent set of [sentence]{}s $\mathcal{T}$ such that each [sentence]{} is of the shape $\_ \sqsubseteq \_$. We derive a set of [graph rule]{}s $\mathcal{R}$ that includes the standard-rules and the translation of the [sentence]{}s in $\mathcal{T}$. By making a fair [weak pushout chain]{} $S$ starting in the empty graph, we obtain a supremum that is a least consequence graph of $\mathbb{0}_\mathcal{L}$ maintaining $\mathcal{R}$. If this graph contains a conflict, then any graph maintaining $\mathcal{R}$ will, so $\mathcal{T}'$ is unsatisfiable. If not, we can apply Lemma \[lem:existsgraph\] to find a model for $\mathcal{T}'$. In this section, we look at constructing fair [weak pushout chain]{}s, based on a set of [graph rule]{}s $\mathcal{R}$ that include the standard-rules.
An Algorithm for Fair [Weak Pushout Chain]{}s {#sec:algorithm}
---------------------------------------------
Assume that the set of [sentence]{}s $\mathcal{T}$ is finite. Consequently, only finitely many relation symbols $\mathcal{L}$ are used in those [sentence]{}s. We restrict $\mathcal{L}$ to those relation symbols that are actually used in $\mathcal{T}$. This makes the corresponding set of [graph rule]{}s $\mathcal{R}$ (including the standard-rules) finite. Thus, we can construct a fair [weak pushout chain]{} for $\mathcal{R}$. Algorithm \[alg:fairpushoutchain\] gives a procedure for this.
$(n \in \mathbb{N},\mathcal{L},E,\mathcal{R})$ Let $G = (\mathcal{L},\{i\mid i\in\mathbb{N},\ i < n\},E)$, produce $G$ Let $W = \{\}$ be our worklist
\[lem:algSound\] Algorithm \[alg:fairpushoutchain\] constructs a fair [weak pushout chain]{} starting in $G$ under $\mathcal{R}$, the limit of which is a least consequence graph of $G$ under $\mathcal{R}$.
The algorithm constructs a [weak pushout chain]{}, because the graph constructed on Line \[lnr:makeGraph\] is part of a [weak pushout step]{} for a [graph rule]{} in $\mathcal{R}$. Let $S : \mathbb{N} \to \mathbb{G}_\mathcal{L}$ describe the [weak pushout chain]{} generated (with $S(0) = G$). Pick an arbitrary $N$. Since the set of [graph rule]{}s is finite, also the number of functions $f$ with $f(v) \leq N$ that embed left-hand sides of [graph rule]{}s into $S(\infty)$ is finite. For some $i$, all such embeddings are in $S(i)$. If an embedding is picked on Line \[lnr:pickGraph\], there is a $g$ such that $R\xrightarrow{g}S(\infty)$, since such a $g$ is added on Line \[lnr:makeGraph\]. Therefore, for each embedding $f$ with $f(v) \leq N$ such that $L\xrightarrow{f}G$, there is a $g$ such that $R\xrightarrow{g}S(\infty)$ with $f(v)=g(v)$. The domain for every such $f$ is finite, so we can pick an $N$ for every $f$ such that $f(v) \leq N$. Therefore, the [weak pushout chain]{} is fair. Lemma \[lem:fairchain\] and \[lem:pushoutchain\] complete the proof.
The algorithm can be changed into a semi-decision procedure to decide whether the limit contains a conflict: If $G$ contains a conflict, then any limit in which $G$ occurs will contain the conflict. Therefore, if we are only interested in whether the limit has a conflict, we can abort the algorithm as soon as $G \cup g(R)$ in Line \[lnr:makeGraph\] has a conflict. Vice versa, if the limit has a conflict, then there will be a graph $G$ in some iteration of the algorithm that has that conflict. This gives a semi-decision procedure. We can use this to decide consistency, using $\mathbb{0}_\mathcal{L}$ as the initial graph.
The same procedure can be used to prove entailment. Say we wish to determine if $\mathcal{T}$ entails $\phi$ for a problem on a standard set of labels $\mathcal{L}$, for $\phi$ equal to $\mathbb{e}_1 \sqsubseteq \mathbb{e}_2$. Assume without loss of generality that $l\not\in\mathcal{L}$. We introduce a new label $l$: $\mathcal{L}' = \mathcal{L} \cup \{l\}$. Let $\mathcal{T}' = \mathcal{T} \cup \{l \sqsubseteq \mathbb{e}_1, \mathbb{e}_2 \sqcap l \sqsubseteq \bot\}$. Let $\mathcal{R}$ be the standard rules plus the derived rules of $\mathcal{T}'$. This time, run the algorithm with $(\mathcal{L}',\{0,1\},\{(l,0,1)\})$ as the initial graph: we obtain a least consequence graph maintaining $\mathcal{R}$. If this graph does not contain a conflict, there is a standard graph $G$ in which $\mathcal{R}$ is maintained, and therefore $\mathcal{T}$ holds in $G$, but $\phi$ does not hold as ${\left\lBrackl\right\rBrack_{G}} \subseteq {\left\lBrack\mathbb{e}_1\right\rBrack_{G}}$ but ${\left\lBrackl\right\rBrack_{G}} \cap {\left\lBrack\mathbb{e}_1\right\rBrack_{G}} = \{\}$ for ${\left\lBrackl\right\rBrack_{G}}$ nonempty, since $(\mathcal{L}',\{0,1\},\{(l,0,1)\}){}\ \xrightarrow{} G$. If the obtained graph does contain a conflict, then all consequence graphs of $\mathcal{R}$ with nonempty $l$ contain a conflict. Suppose $G$ is standard, each of $\mathcal{T}$ holds, there is a pair (labeled $l$) in ${\left\lBrack\mathbb{e}_1\right\rBrack_{G}}$, and that pair is not in ${\left\lBrack\mathbb{e}_2\right\rBrack_{G}}$, then we get a contradiction to the statement that all consequence graphs of $\mathcal{R}$ with nonempty $l$ contain a conflict. In other words: for each standard $G \vDash \mathcal{T}$, we have ${\left\lBrack\mathbb{e}_1\right\rBrack_{G}} \subseteq {\left\lBrack\mathbb{e}_2\right\rBrack_{G}}$, so $\mathcal{T}$ entails $\phi$. This shows that a least consequence graph can be used to decide entailment. By terminating our procedure when a conflict is found, we can prove entailment if it holds (and do not terminate otherwise). This can be extended to $\phi$ of the shape $\mathbb{e}_1 = \mathbb{e}_2$, by applying this procedure to both $\mathbb{e}_1 \sqsubseteq \mathbb{e}_2$ and $\mathbb{e}_2 \sqsubseteq \mathbb{e}_1$.
There is another case in which we can abort: once the graph maintains all [graph rule]{}s in $\mathcal{R}$, we hit Line \[lnr:stop\], and $G$ is equal to the limit. In such a case, we have found the limit of the chain given by Algorithm \[alg:fairpushoutchain\], and can immediately decide whether or not it is conflict free. Unfortunately, even if conflict free graphs that maintain all [graph rule]{}s exist (so by definition of least consequence graph, the limit is conflict free), we do not necessarily hit this case. Section \[sec:undecidable\] shows that we cannot hope to find an algorithm that decides whether or not a conflict free consequence graph exists.
Optimizations for Implementations
---------------------------------
We discuss some possible optimizations for the purpose of showing correctness of the algorithm described by the author in an earlier paper [@amperspiegelRAMICS]. The earlier algorithm is not Algorithm \[alg:fairpushoutchain\], but an optimized version thereof. We only describe a few optimizations, that suffice to show that the algorithm presented earlier is correct as well.
As optimizations, we allow changing the outcome of the algorithm, but require that the proof of Lemma \[lem:algSound\] remains valid. In particular, instead of the graph $G \cup g(R)$ constructed on Line \[lnr:makeGraph\], we can make a larger graph $S(\infty)$ if $G \cup g(R) \subseteq S(\infty)$ and $S(\infty)$ is the limit of a (not necessarily fair) [weak pushout chain]{}. Through this change, the algorithm no longer constructs simple [weak pushout chain]{}s, but Lemma \[lem:algSound\] still holds.
As an instance of this, observe that we can combine [graph rule]{}s, as this is a form of combining [weak pushout step]{}s: suppose $(L,R)$ and $(L',R')$ are [graph rule]{}s in $\mathcal{R}$, such that $L' \xrightarrow{f} R$. Then we can find an $R''$ such that $(R,R'')$ is a [weak pushout step]{} of $(L',R')$. We can then safely replace the [graph rule]{} $(L,R)$ for $(L,R'')$ in $\mathcal{R}$, as a [weak pushout step]{} of $(L,R'')$ is the limit of a chain that satisfies the aforementioned condition. Apart from changing the set of [graph rule]{}s $\mathcal{R}$, we can change the algorithm such that the standard-rules are always maintained after each step. Let $G'$ be a graph constructed in that way. According to Lemma \[lem:existsgraph\], we represent the graph $G'$ by the graph $\hat{f}(G')$, making it such that we do not need to store the relation-symbols $\top$, $\bot$, ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$, or the constants in $\mathcal{C}$. We do need to keep track of which vertices originally belong to which equivalence classes, in order to be able to produce the underlying $G'$ in each step. Since the function $f$ possibly maps several vertices of $G'$ to one vertex in $\hat{f}(G')$, the original graph $\hat{f}_{i}(G'_{i})$ is not necessarily a subgraph of the newly generated graph $\hat{f}_{i+1}(G'_{i+1})$. On the other hand, if we are only interested in whether or not there is a conflict in the least consequence graph, then we only need to keep track of the least vertex of each class such that the $N$ chosen on Line \[lnr:pickGraph\] corresponds to a minimal embedding of $f$. This is precisely the algorithm proposed in the earlier paper [@amperspiegelRAMICS], showing it is a semi-decision procedure for deciding whether a least consequence graph contains a conflict.
Example Run of the Optimized Algorithm
--------------------------------------
We return to one of the examples given in Section \[sec:background\]: the entailment problem that asks whether $\mathcal{T} = \{{\tt r} = {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\ ,\ {\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\}$ entails ${\tt r}\fcmp{\tt r} \sqsubseteq {\tt r}$. We construct a $\mathcal{T}'$ for the entailment problem as described in Section \[sec:algorithm\]: $$\begin{aligned}
\mathcal{T}' = \{\ & {\tt r} \sqsubseteq {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\ ,\ {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \sqsubseteq {\tt r} \ ,\ {\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \fcmp {\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{} \ ,\ l \sqsubseteq {\tt r}\fcmp{\tt r} \ ,\ l \sqcap {\tt r} \sqsubseteq \bot\ \}\end{aligned}$$ Using the translation of Section \[sec:Rules\], Figure \[fig:exgr\] gives the [graph rule]{}s we work with. We use the optimizations just described, and do not restate the standard-rules.
[0.19]{}
\(c) ; (0) \[left = 10pt of c\] ; (1) \[right = 10pt of c\] ; (2) \[below =of c\] ; (0) edge\[thick\] node\[label=above:$\tt r$\] (1); (0) edge\[dotted\] node\[label=below:$\tt i$\] (2); (1) edge\[dotted\] node\[label=below:$\tt i$\] (2);
[0.19]{}
\(c) ; (0) \[left = 10pt of c\] ; (1) \[right = 10pt of c\] ; (2) \[below =of c\] ; (0) edge\[dotted\] node\[label=above:$\tt r$\] (1); (0) edge\[thick\] node\[label=below:$\tt i$\] (2); (1) edge\[thick\] node\[label=below:$\tt i$\] (2);
[0.19]{}
\(c) ; (0) \[left = 10pt of c\] ; (1) \[right = 10pt of c\] ; (2) \[below =of c\] ; (0) edge\[dotted\] node\[label=above:${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$\] (1); (2) edge\[thick\] node\[label=below:$\tt i$\] (0); (2) edge\[thick\] node\[label=below:$\tt i$\] (1);
[0.19]{}
\(c) ; (0) \[left = 10pt of c\] ; (1) \[right = 10pt of c\] ; (2) \[below =of c\] ; (0) edge\[thick\] node\[label=above:$l$\] (1); (0) edge\[dotted\] node\[label=below:$\tt r$\] (2); (2) edge\[dotted\] node\[label=below:$\tt r$\] (1);
[0.19]{}
\(0) ; (1) \[right =of 0\]; (0) edge\[thick,bend left=90\] node\[label=[\[label distance=-3\]$l$]{}\] (1); (0) edge\[thick\] node\[label=[\[label distance=-3\]$\tt r$]{}\] (1); (0) edge\[dotted,bend right=90\] node\[label=[\[label distance=-3\]below:$\bot$]{}\] (1);
We start the procedure with $n = 2$ and $E = \{(l,0,1)\}$. Note that per our optimizations, the self loops $({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{},0,0)$ and $({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{},1,1)$ are implicitly there, as well as all $\top$ edges. Only one rule does not hold: $l \sqsubseteq {\tt r}\fcmp{\tt r}$, and consequently only one [graph rule]{} is not maintained. A pushout step for it gives $n = 3$ and $E = \{(l,0,1), ({\tt r},0,2), ({\tt r},2,1)\}$ as the next call to ProduceChain. Again only one rule is not maintained, the one for ${\tt r} \sqsubseteq {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}$. This time, our work-list contains two elements: one for each edge labeled $\tt r$. Both have a maximum node number of $2$, so we can choose either. We pick $f$ that maps to $({\tt r},0,2)$, and $n = 4$ and $E = \{(l,0,1), ({\tt r},0,2), ({\tt r},2,1), ({\tt i},0,3), ({\tt i},2,3)\}$. This time, ${\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \sqsubseteq {\tt r}$ is also not maintained: $({\tt i},0,3)$ but $({\tt r},0,0)$ is missing. The highest node number assigned to $N$ is $3$ however, so we need to finish treating ${\tt r} \sqsubseteq {\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}$. Next iteration: $n = 5$ and $E = \{(l,0,1), ({\tt r},0,2), ({\tt r},2,1), ({\tt i},0,3), ({\tt i},1,4), ({\tt i},2,3), ({\tt i},2,4)\}$. Subsequently: $n = 5$ and $E = \{(l,0,1), ({\tt r},0,2), ({\tt r},2,1), ({\tt i},0,3), ({\tt i},1,4), ({\tt i},2,3), ({\tt i},2,4), ({\tt r},0,0)\}$, then $({\tt r},2,2)$ is added. At this point, we have a choice again, between ${\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \sqsubseteq {\tt r}$ and ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$. We apply the former first: after several iterations it gives us the graph that satisfies all rules except ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$: $$\begin{aligned}
{\left\lBrackl\right\rBrack_{G}} &= \{(0,1)\} \\
{\left\lBrack\tt i\right\rBrack_{G}} &= \{(0,3),(1,4),(2,3),(2,4)\} \\
{\left\lBrack\tt r\right\rBrack_{G}} &= \{(0,0),(0,2),(1,1),(1,2),(2,0),(2,1),(2,2)\}\end{aligned}$$ Since we did not use ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$ yet, and all other rules are satisfied up to this point, we are exactly in the place we would have been if ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$ wasn’t present. This is (minus the $l$) the graph given in Section \[sec:background\] as a possible graph our algorithm could give. If we would have handled $({\tt r},2,1)$ before $({\tt r},0,2)$ instead, we would have gotten a graph with a different numbering.
We now proceed by applying ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$. The pushout step adds $({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{},3,4)$. We have not described precisely how our optimizations proceed at this point, but we need to renumber the nodes such that $3$ and $4$ are identified. For preserving fairness, we renumber high to low: the node $4$ is relabeled to $3$. This can cause some pushout steps to get assigned a lower $N$, but never a higher one. We proceed with the graph $G'$: $$\begin{aligned}
{\left\lBrackl\right\rBrack_{G'}} &= \{(0,1)\} \\
{\left\lBrack\tt i\right\rBrack_{G'}} &= \{(0,3),(1,3),(2,3)\} \\
{\left\lBrack\tt r\right\rBrack_{G'}} &= \{(0,0),(0,2),(1,1),(1,2),(2,0),(2,1),(2,2)\}\end{aligned}$$ At this point, ${\tt i}\fcmp{\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \sqsubseteq {\tt r}$ does not hold, and the resulting action is to insert $({\tt r},0,1)$. Subsequently, $l \sqcap {\tt r} \sqsubseteq \bot$ does not hold and we insert a conflict. We abort concluding that the entailment holds.
While we needed several iterations to conclude entailment, we saved many iterations by treating the standard rules separately. If we had applied ${\tt i}{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}\fcmp{\tt i} \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$ earlier, we would have derived the contradiction sooner.
Presentation of the Algorithm
-----------------------------
We conclude this section with a note on the presentation in this paper. In the earlier paper, we presented the efficient implementation [@amperspiegelRAMICS] as discussed in the previous paragraph. This does not allow us to talk about the limit of the procedure. Using the same presentation would have alleviated the need for Lemma \[lem:identrules\]. However, the simpeler presentation used in this paper allows us to argue that the limit of a chain always exists. This simplifies many of the other proofs in this paper.
We give an example that shows why it is problematic to describe limits in the more involved presentation: Given the [graph rule]{}s $\mathcal{E}\left(l \sqsubseteq l\fcmp l\right)$, $\mathcal{E}\left(l \sqsubseteq {{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\right)$ and the identity-rules, Figure \[fig:wpcloop\] shows a part of a [weak pushout chain]{}. Following the procedure for the given rules, we obtain the graphs in Figure \[fig:wpcstep1\] and \[fig:wpcstep2\]. After every step, we could decide to apply the identity-rules until they are maintained. If we construct a chain like this, and proceed in a similar manner as illustrated in Figure \[fig:wpcloop\], we indeed construct a fair [weak pushout chain]{}. The limit of this chain is an infinite graph in which every two vertices are connected by an edge labeled $l$, as well as an edge labeled ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}$, which indeed maintains the [graph rule]{}s. If we apply the mentioned optimizations and choose a representation of the graph as intended in Lemma \[lem:identrules\] after each graph, defining the ‘limit’ becomes problematic: We do not need to draw edges with the label ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$, as they are given by the drawn vertices, and the graph representation after the step in Figure \[fig:wpcstep2\] is drawn in Figure \[fig:wpcend\]. This graph is isomorphic to the one we started with, showing we end up in a sequence that alternates between two graphs. None of these graphs maintains any of the given [graph rule]{}s, despite the ‘underlying’ chain being fair. Since a well defined limit is an important concept in many lemmas, we chose to use chains as described in this paper.
[0.19]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$l$\] (2); (2) edge\[bend left\] node\[below=12pt,label=:$l$\] (1); (3) edge\[bend left\] node\[label=below:$l$\] (1); (2) edge node\[label=right:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=below:$l$\] (1);
[0.3]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$l$\] (2); (2) edge\[bend left\] node\[below=12pt,label=:$l$\] (1); (3) edge\[bend left\] node\[label=below:$l$\] (1); (2) edge\[thick\] node\[label=right:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=below:$l$\] (1); (x) \[right=of 2\] ; (2) edge\[dotted\] node\[label=above:$l$\] (x); (x) edge\[dotted\] node\[label=below:$l$\] (3);
[0.3]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$l$\] (2); (2) edge node\[below=10pt,label=:$l$\] (1); (3) edge\[bend left,thick\] node\[label=below:$l$\] (1); (3) edge\[dotted\] node\[below right=6pt,label=${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$\] (1); (2) edge node\[below right=2pt,label=:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=below:$l$\] (1); (x) \[right=of 2\] ; (2) edge node\[label=above:$l$\] (x); (x) edge node\[label=below:$l$\] (3);
[0.19]{}
\(1) ; (2) \[right =of 1\] ; (3) \[below =of 2\] ; (1) edge\[bend left\] node\[label=above:$l$\] (2); (2) edge\[bend left\] node\[below=12pt,label=:$l$\] (1); (3) edge\[bend left\] node\[label=below:$l$\] (1); (2) edge node\[label=right:$l$\] (3); (1) edge \[loop left,in=135,out=-135,distance=1cm\] node\[label=below:$l$\] (1);
A Proof of Undecidability {#sec:undecidable}
=========================
The following decision problem is undecidable: given a set of [sentence]{}s $\mathcal{T}$, is there a standard graph $G$ in which every [sentence]{} in $\mathcal{T}$ holds?
This proof closely follows a proof by Krisnadhi and Lutz [@Krisnadhi:2007aa] on ‘conjunctive query answering’. We use a reduction from the undecidable problem whether two context free grammars have an empty intersection. This problem is given by two grammars with non-terminals $N_1$ and $N_2$, a common set of terminals $T$, and production rules $P_i \subseteq N_i\times(N_i \cup T)^*$ with $i\in\{1,2\}$. The sets $N_1, N_2$ and $T$ are mutually disjoint. The question to be answered is whether there exists a sequence of terminals that is generated by the two starting nodes $s_i \in N_i$.
We make an encoding by choosing $\mathcal{C},\mathcal{L}$ and $\mathcal{T}$ such that there is a standard graph in which the [sentence]{}s in $\mathcal{T}$ hold if and only if the context free grammars have an empty intersection. We encode every terminal and nonterminal with corresponding relation symbols, and use the constant symbol $\epsilon$ for the empty word: $\mathcal{C} = \{\epsilon\}$ and $\mathcal{L} = \{\epsilon,\top,\bot,{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{}\} \cup N_1 \cup N_2 \cup T$. For $\mathcal{T}$ we use a [sentence]{} for each production rule, one for each terminal, and a final [sentence]{} that requires the two grammars to have an empty intersection: $$\begin{gathered}
\mathcal{T} = \left\{\sigma_0 \fcmp \cdots \fcmp \sigma_k \sqsubseteq n_j \mid (n_j,\sigma_0\cdots\sigma_k) \in P_1 \cup P_2 \right\}
\cup
\left\{{{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{} \sqsubseteq t \fcmp t{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{} \mid t \in T \right\} \\
\cup
\left\{\epsilon \fcmp s_1 \fcmp s_2{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}\fcmp \epsilon \sqsubseteq \bot \right\}\end{gathered}$$
We show that there is a standard graph in which $\mathcal{T}$ holds if the grammars have an empty intersection, and there is no such graph if the grammars share a word. First suppose the grammars have an empty intersection. We construct a graph as follows: The vertices are words over $T$ where $\epsilon$ is the empty word. There are edges $(t,u,u t)$ for each non-terminal $t\in T$, edges $(n,u,u p)$ if the word $p$ is a valid parse of $n \in N_1 \cup N_2$, and edges to make the graph standard: $$\begin{gathered}
E = \left\{(t,u,u t) \mid u \in T^*,\ t \in T\right\} \cup \{(\epsilon,\epsilon,\epsilon)\} \cup \{({{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}{},u,u) \mid u\in T^*\} \\ \cup \{(\top,u,v) \mid u,v\in T^*\}
\cup \left\{(n,u,u p) \mid u,p \in T^*,\ n \in N_1 \cup N_2,\ n\text{ parses }p\right\}\end{gathered}$$ It can be checked that $G=(\mathcal{L},T^*,E)$ is standard and all [sentence]{}s in $\mathcal{T}$ hold. In particular, ${\left\lBracks_1 \fcmp s_2\right\rBrack_{G}} = \{\}$ as there is no word $p$ such that $s_1$ parses $p$ and $s_2$ parses $p$.
Now suppose for a proof by contradiction that $G=(\mathcal{L},V,E)$ is a standard graph in which all [sentence]{}s in $\mathcal{T}$ hold, and that $w = t_0 \cdots t_{n-1}$ is a word that is parsed by $s_1$ and $s_2$. Since $\mathcal{T}$ holds, there is a path in $G$ with vertices $\epsilon=v_0,\ldots,v_n$ and edges $(t_0,v_0,v_1), \ldots, (t_{n-1},v_{n-1},v_n)$. By induction on the parse-tree of how $s_1$ parses $w$, there is an edge $(s_1,v_0,v_n)\in E$. Similarly, $(s_2,v_0,v_n)\in E$. Since $\epsilon = v_0$ and $G$ is standard, $(\epsilon,\epsilon) \in {\left\lBrack\epsilon\fcmp s_1\fcmp s_2{^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}\fcmp\epsilon\right\rBrack_{G}}$. Since all [sentence]{}s in $\mathcal{T}$ hold, $(\epsilon,\epsilon) \in {\left\lBrack\bot\right\rBrack_{G}}$, contradicting that $G$ is standard.
Undecidability of entailment follows as a corollary: there are no standard graphs in which every sentence in $\mathcal{T}$ holds iff the sentence $\top = \bot$ is entailed.
We end with a final remark about the proof we presented. The relation symbol $\top$ and the operation $\sqcap$ were not used in the proof. Thus, this proof of undecidability holds if $\mathcal{T}$ is restricted to [sentence]{}s of a simpler shape. By application of Lemma \[lem:existsgraph\], we conclude that deciding whether a conflict free consequence graph under $\mathcal{R}$ exists is undecidable. Equivalently, given $\mathcal{R}$, it is undecidable to determine whether any least consequence graph under $\mathcal{R}$ contains a conflict.
Conclusion {#sec:conclusion}
==========
In this paper, we have given a translation of [sentence]{}s into [graph rule]{}s, and have proven that for a graph $G$, a [sentence]{} is maintained in $G$ if and only if the translated [graph rule]{} holds in $G$. Furthermore, when allowing the [sentence]{}s to use extra relation-symbols with a dedicated meaning, we can add corresponding [graph rule]{}s to ensure that this dedicated meaning is preserved. In addition, we showed that there exists a least consequence graph of a set of [graph rule]{}s, which one may be able to find through a semi-decision procedure. This procedure also allows us to determine whether a set of [sentence]{}s is consistent. Finally, we have shown that in a sense, we cannot do better than to give a semi-decision procedure: The problem of whether a set of [sentence]{}s has a standard graph in which all [graph rule]{}s hold is undecidable.
Our procedure can partially automate preserving invariants in information systems. Its implementation and evaluation is foreseen in Ampersand, but considered outside the scope of this paper.
#### Acknowledgements {#acknowledgements .unnumbered}
We thank Wolfram Kahl and Stef Joosten for their thoughts and comments on this paper and earlier versions. Part of the research presented in this paper was performed at the University of Innsbruck, Austria, supported by the Austrian Science Fund (FWF) project Y757. Supported by The Netherlands Organisation for Scientific Research (NWO) project 639.023.710.
[^1]: These are $\fcmp$, $\sqcap$, ${^{\mkern-1mu{}{\raise0.3ex\hbox{\tiny$\smallsmile$}}}\kern-0.1em{}}{}$ and ${{{\text{\usefont{U}{dsrom}{m}{n}1}}_{}}}$. See the book by Freyd and Scedrov for details on allegories [@FS90].
[^2]: Pushouts in a category with embeddings as arrows
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We developed a strategic of optimal portfolio based on information theory and Tsallis statistics. The growth rate of a stock market is defined by using $q$-deformed functions and we find that the wealth after n days with the optimal portfolio is given by a $q$-exponential function. In this context, the asymptotic optimality is investigated on causal portfolios, showing advantages of the optimal portfolio over an arbitrary choice of causal portfolios. Finally, we apply the formulation in a small number of stocks in brazilian stock market $[B]^{3}$ and analyzed the results.'
address:
- 'Colegiado de Física, Departamento de Ciências Exatas e da Terra, Universidade do Estado da Bahia, Brasil'
- 'Colegiado de Engenharia Civil, Universidade Federal do Vale do São Francisco, Brasil'
author:
- 'Marco A. S. Trindade'
- Sergio Floquet
- 'Lourival M. Silva Filho'
title: 'Portfolio Theory, Information Theory and Tsallis Statistics'
---
Econophysics ,Stock Market and Portfolio Theory ,Nonextensive Theory and q-Gaussians
Introduction
============
The Modern Portfolio Theory was introduced by Harry Markovitz in 1952 [@Mark]. The basic principle is the mean-variance approach [@Sharpe] so that the expected return is maximized for a given level of risk, i.e., a constraint on the variance. This reflects the idea of diversification in investment and risk aversion. Posteriorly, Cover explored these concepts in the context of information theory [@Cover; @Cover1; @Cover2; @Cover3]. Particularly, Kelly introduced the concept of log-optimal portfolio [@Kelly]. An asymptotic equipartition property for the stock market as well as the asymptotic optimality of log-optimal investment was derived by Algoet and Cover [@Cover4]. Still in this context of quantitative finance, the Cover’s portfolio was defined in reference [@Cover5].
It is important to highlight two seminal references in the applications of information theory in finance [@theil65; @fama65b]. In [@theil65] Theil and Leenders analyzed data on the Amsterdam Stock Exchange in the period November 2, 1959, through October 31, 1963, concluding that the Amsterdam Stock Exchange has a memory of one day declining in price, advancing or remaining unchanged. The relative entropy was used as a measure of inaccuracy for the forecasts. Fama [@fama65b], in turn, applied the Theil-Leenders tests in data from the New York Exchange for the period June 2, 1952, to October 29, 1962, conducting that the proportions of securities declining or advancing today on the New York Stock Exchange are not auspicious in predicting proportions declining or advancing tomorrow.
A relevant risk measure in portfolio optimization is the entropic value-at-risk (EVarR), introduced by Ahmadi-Javid [@ahmadi11; @ahmadi12]. In addition to being coherent (i.e., it satisfies the properties translation invariance, subadditivity, monotonicity, positive homogeneity), it is strongly monotone and strictly monotone. Recently, a promising sample-based portfolio optimization has been proposed [@ahmadi19]. An interesting approach is that EVaR approach outperforms CVaR (Conditional Value-at-Risk) approach as the sample size increases. This is because the number of variables and constraints of the EVaR is independent of the sample size. In addition, under a analysis of real have better best mean and worst return rates and Sharpe ratios compared with the previous one. An investigation about performance hypothesis testing through the Sharpe ratio was conducted by Memmel [@memmel03]. It has been shown that we can use as test statistic the Sharpe ratio difference divided by its asymptotic standard deviation. Aifan Ling et al. [@ling19] developed a robust multi-period mean-LPM (lower partial moment) portfolio selection model considering transaction cost under an asymmetric uncertainty set. This model provides better returns an Shape ratios when real market data are analyzed.
A recent approach for the field of quantitative finance and economics, called econophysics [@St1], lies in the use tools derived from statistical physics [@Bou]. Relations between physics and economics are long standing. The gravity model of international trade, for example, mimics the law of gravitation and it was proposed in 1954, by Walter Isard [@Isard]. In the econophysics framework associated to information theory, some works [@Zu1; @Zu2] were developed exploring the complexity-entropy causality plane. In the reference [@Zu1], applications to distinguish the stage of the development of the stock market are conducted and in [@Zu2] the efficiency of sovereign bond markets is investigated through complexity-entropy causality plane revealing correlations and hidden structures in the daily values of bond indices. Zunino et al. [@zunino16] analyzed the time-varying informational efficiency of European corporate bond markets as well as the impact of the 2008 financial crises on setorial indices related to the aforementioned titles. An interesting result is that before the crisis, all sectors present similar efficient behaviors. In the post-crisis each sector follow its own dynamic.
Another techniques emerge from non-extensive statistical mechanics [@Tsallis2; @Tsallis3] proposed by Tsallis in 1988 [@Tsallis1], as a generalization of Boltzmann-Gibbs statistics. The Tsallis entropy carries a non-extensive parameter q such that in the limit $q \rightarrow 1$ the Boltzmann-Gibbs entropy is recovered. The maximization of the Tsallis entropy under appropriate constraints leads to the distribution of Tsallis $q$-Gaussian [@Tsallis2; @Tsallis3; @Tsallis4]. It has been widely used in the study of complex systems [@Tsallis2; @Tsallis3], including the stock market. The $q$-Gaussians allow a consistent way of describing high-frequency financial observations due to finite variance and temporal autocorrelations [@TsBor]. Option pricing formulas based on Tsallis statistics were derived by Borland [@Borland]. As highlighted in [@TsBor], one advantage is the existence of explicit closed-form solutions. The generalized Black-Sholes (B-S) equation was obtained with entropic index $q$. This model fits, for $q=1.5$, stock returns more realistically than B-S standard $(q=1)$. The probability distribution of stock returns is non-Gaussian [@mandelbrot63; @fama65]. On the other hand, the modern portfolio theory assumes that returns follow a Gaussian distribution, which results in a less realistic scenario. In order to overcome these limitations, in this paper, we merge Cover’s approach from information theory to the mathematical tools of Tsallis statistics. In section \[sec1\] we present the Nonextensive Statistics and some of yours results that will be using in section \[sec2\] to formulate the nonextensive version of the Cover’s portfolio using the $q$-deformed functions and the $q$-product as key elements. In addition, an asymptotic study is carried out to substantiate the work. In section \[sec3\] we will apply the above for-mentioned formalism for the analysis of a small number of stocks in brazilian stock market and comparing the results with Cover’s portfolio describe by a Gaussian distribution, besides calculating the Sharpe Ratio and Sortino Ratio. Finally, we present the final considerations and perspectives in section \[sec4\].
Nonextensive Statistics \[sec1\]
================================
Tsallis entropy [@Tsallis1] is a generalization of the Boltzmann-Gibbs entropy. It is given by$$\begin{aligned}
S_{q} & = & k_{B} \frac{1- \displaystyle \sum_{i=1}^{n} p_{i}^q }{1-q} , \end{aligned}$$ with $\displaystyle \sum_{i=1}^{n} p_{i} = 1$, where for two systems independent $A$ and $B$ we have the propriety $$\begin{aligned}
S_{q}(A+B) = S_{q}(A) + S_{q}(B) +(1-q)S_{q}(A)S_{q}(B).\end{aligned}$$ The parameter $q$ represents the nonextensive of the systems, when for $q=1$ we recover the Boltzmann-Gibbs entropy, for $q<1$ we obtain the superextensivity (additivity of the entropy) and for $q>1$ we obtain the subextensivity (decrease of the entropy).
Tsallis introduced the function $q$-logarithm and $q$-exponential $$\begin{aligned}
\ln_{q}(x) &=& \displaystyle \frac{ x^{1-q}-1}{1-q}, \ \ \exp_{q}(x) = [1+(1-q)x]_{+}^{\frac{1}{1-q}} \label{eq-lnq}\end{aligned}$$ with $[A]_{+}:=max(0,A)$, that for $q=1$ we recover the logarithm and exponential functions, and allow write Tsallis entropy as $S_{q} = -\sum_{i=1}^{n} p_{i}^{q} \ln_{q}p_{i}$. Another important operation between the functions is the $q$-product, $\otimes_{q}$, that is defined by [@Borges] $$x \otimes_{q} y := [x^{1-q}+y^{1-q}-1]_{+}^{\frac{1}{1-q}}. \label{eq-q-produto}$$
The $q$-Gaussian function can be introduced as $$\begin{aligned}
f_{q}(x) & = & \frac{1}{C_{q}|\sigma|} exp_{q}\left[ -\frac{(x-\mu)^{2} }{\sigma^{2}} \right]\end{aligned}$$ where $C_{q}$ is the normalization constant given by $$C_q=\left\{ \begin{array}{ll}
\frac{2\sqrt{\pi}\Gamma \left(\frac{1}{1-q}\right)}{(3-q)\sqrt{1-q}\Gamma \left(\frac{3-q}{2(1-q)}\right)}&
-\infty<q<1 \\
\vspace{0.1cm}
\sqrt{\pi},& \ \ q=1 \\
\vspace{0.1cm}
\frac{\sqrt{\pi}\Gamma \left(\frac{3-q}{2(q-1)}\right)}{\sqrt{q-1}\Gamma \left(\frac{1}{1-q}\right)} ,& 1<q<3, \\
\end{array} \right.$$ which has an asymptotic ($x\gg 1$) power law behavior given by $ \exp_{q}(x^{2}) \sim x^{ \frac{2}{1-q}}$. In a similar way we have the multivariate $q$-Gaussian distribution $$\begin{aligned}
\label{qGau}
f_{q}(\textbf{x})&=&C_{d,q}\exp_{q}\left[ \frac{ (x_{1}-\mu_{1}) ^{2} }{\sigma_{1}^{2}} + \cdots + \frac{ ( x_{d} -\mu_{d} ^{2}) }{\sigma_{d}^{2}} \right],\end{aligned}$$ where $\sigma_{i}$ and $\mu_{i}$ are parameters to be determined and the normalizing constant is given by $ C_{d,q} = \displaystyle \frac{ 1 }{ |\sigma_{1}\cdots \sigma_{d}| w_{d} I_{q,d} }$ where $w_{d}$ is the surface area of the unit sphere in $\mathbb{R}^{d}$ dimensional space, with $ \displaystyle w_{d} = \frac{ 2\pi^{ \frac{d}{2} } }{ \Gamma(\frac{d}{2}) } $, and $ \displaystyle I_{q,d} = \int_{0}^{+\infty} r^{d-1} e_{q}^{-r^{2}} $.
Umarov and Tsallis [@Umarov1] obtained a formula for compute the normalizing constant $$\begin{aligned}
C_{d,q}= \frac{\left(\frac{3-q_{1}}{2}\right)^{\frac{d-1}{2}}\left(\frac{3-q_{2}}{2}\right)^{\frac{d-2}{2}}...
\left(\frac{3-q_{d-1}}{2}\right)^{\frac{1}{2}}}{(C_{q}^{q})^{-1}(C_{q_{1}}^{q_{1}})^{-1}...(C_{q_{d-1}}^{q_{d-1}})^{-1}},\end{aligned}$$ with $$\begin{aligned}
q_{n}=\frac{2q+n(1-q)}{2+n(1-q)}, \ \ n=0 \pm 1, \pm 2,...\end{aligned}$$ where $C_{q}$ is the normalizing constant of one-dimensional $q$-Gaussian.
As pointed out in many papers [@TsBor; @borland2004a; @queiros2005a; @vellekoop2007] the stock market data express a fat tail behavior and a power law for the cumulative return distribution in the asymptotic case ($x\gg 1$) [@drozdz2007b; @pan2008; @eryigit2009] what makes the $q$-Gaussian distribution attractive to describe stock market data. In cases [@borland2004a; @gopikrishnana1998; @drozdz2007a; @ruiz2018] an inverse cubic power law for cumulative distribution was obtained and it is related to $q$-Gaussian distribution with $q=1.5$.
Cover’s q-Portfolio \[sec2\]
============================
The stock market is given by a vector $\mathbf{X}=(X_{1}, X_{2},..., X_{n})$, where $X_{i}$ is the relative price, i.e. the ratio of the price at the end of the day to the price at the beginning of the day. We have that $X_{i}\geq 0$, $i=1,...,m$ where $m$ is the number of stocks. In this work we apply the Tsallis statistic formalism in the Cover’s portfolio theory, where the parameter $q$ is responsible by take account the nonextensivity of stock market. Following Cover [@Cover5] we introduced the growth $q$-rate of a stock market portfolio $\textbf{b}$: $$\begin{aligned}
W_{q}(\textbf{b}, f)&=&E(\ln_{q}\textbf{b}^{t}\textbf{x}) \nonumber \\
&=&\int \ln_{q}(\textbf{b}^{t}\textbf{x})f_{q} (\textbf{x})d\textbf{x}.\end{aligned}$$ We define the optimal growth $q$-rate $W_{q}^{*}(f_{q})$ as $$\begin{aligned}
W^{*}_{q}(f)=\max_{\textbf{b}} W_{q}(\textbf{b}, f_{q}).\end{aligned}$$ The growth optimal portfolio $b^{*}$ is one that achieves the maximum of $W_{q}(\textbf{b}, f_{q})$. Let $\textbf{X}_{1},\textbf{X}_{2},...,\textbf{X}_{n}$ be random vectors *i.i.d* with density probability function $f_{q}$. The $q$-wealth after n days using the portfolio $b^{*}$ is given by $$\begin{aligned}
S_{n}^{*(q)}=\bigotimes_{(q);i=1}^{n} \textbf{b}^{*t} \textbf{X}_{i},\end{aligned}$$ where the $q$-product $\otimes_{q}$ is defined in (\[eq-q-produto\]).
Using the strong law of large numbers [@Durrett], we have $$\begin{aligned}
\frac{1}{n}\ln_{q}S_{n}^{*(q)}&=&\frac{1}{n}\ln_{q}\bigotimes_{(q);i=1}^{n}\textbf{b}^{*t} \textbf{X}_{i} \nonumber \\
&=& \frac{1}{n}\sum_{i=1}^{n}\ln_{q}\textbf{b}^{*t} \textbf{X}_{i} \rightarrow W^{*}_{q} \ \ a.s.,\end{aligned}$$ so that $S_{n}^{*(q)}=\exp_{q}(nW^{*}_{q})$. This justifies our definition of $q$-wealth analogously to reference [@Cover]. Now, we will consider the following assumption: $$\label{des1}
E \left(\frac{S_{n}^{(q)}}{S_{n}^{*(q)}}\right)\leq 1,$$ in order to show the asymptotic optimality of the $ln_{q}$-optimal portfolio for the causal portfolios. The causal portfolio strategy is a sequence of mappings $b_{i}: T^{m(i-1)}\rightarrow B$, where the portfolio $\textbf{b}_{i}(\textbf{X}_{1},\textbf{X}_{2},..., \textbf{X}_{i-1})$ is used on day $i$. We define $B=\{\textbf{b} \in \mathbb{R}^{m}: \textbf{b}_{i} \geq 0, \sum_{i=1}^{m}\textbf{b}_{i}=1 \}$ as the set of allowed portfolios. Using the Markov inequality [@Durrett], we have from (\[des1\]): $$\begin{aligned}
P(S_{n}^{(q)} > \lambda_{n}S_{n}^{*(q)}) \leq \lambda_{n}^{-1}.\end{aligned}$$ Therefore $$\begin{aligned}
P\left(\frac{1}{n}\ln_{q}\frac{S_{n}^{(q)}}{S_{n}^{*(q)}} > \frac{\ln_{q}\lambda_{n}}{n} \right) \leq \lambda_{n}^{-1}.\end{aligned}$$ Consequently $$\begin{aligned}
\sum_{n=1}^{\infty} P\left(\frac{1}{n}\ln_{q}\frac{S_{n}^{(q)}}{S_{n}^{*(q)}} > \frac{n^{2-2q}-1}{n(1-q)} \right) \leq \frac{\pi^{2}}{6},\end{aligned}$$ where we consider $\lambda_{n}=n^{2}$. Then $$\begin{aligned}
P\left(\frac{1}{n}\ln_{q}\frac{S_{n}^{(q)}}{S_{n}^{*(q)}} > \frac{n^{2-2q}-1}{n(1-q)} ,\textrm{infinitely often} \right)=0\end{aligned}$$ using the Borel-Cantelli lemma [@Durrett]. Thus, there exists an $N$ such that for all $n>N$: $$\label{alt}
\frac{1}{n}\ln_{q}\frac{S_{n}^{(q)}}{S_{n}^{*(q)}} \leq \frac{n^{2-2q}-1}{n(1-q)},$$ for almost every sequence from the stock market. This implies (for $q>0.5$) with probability 1: $$\limsup_{n \rightarrow \infty} \frac{1}{n} \ln_{q}\frac{S_{n}^{(q)}}{S_{n}^{*(q)}} \leq 0.$$
We have that for almost every sequence from stock market, $S_{n}^{*(q)}$ is greater than the wealth of any investor, i. e., the $ln_{q}$-optimal portfolio is better than any other portfolio under the assumptions aforementioned. We can also show that $$\begin{aligned}
\max_{\textbf{b}_{1},...,\textbf{b}_{n}}E \left( \ln_{q}S_{n}^{(q)} \right) &=&\max_{\textbf{b}_{1},...,\textbf{b}_{n}}E \left[ \ln_{q}\left( \bigotimes_{(q);i=1}^{n} \textbf{b}_{i}^{t}\textbf{X}_{i} \right) \right] \nonumber \\
&=&\sum_{i=1}^{n}\max_{\textbf{b}_{1},...,\textbf{b}_{n}}E \left( \ln_{q}\textbf{b}_{i}^{t}
(\textbf{X}_{1},...,\textbf{X}_{i}) \right) \nonumber \\
&=&\sum_{i=1}^{n}E \left( \ln_{q}\textbf{b}^{*t}\textbf{X}_{i} \right) \nonumber \\
&=& nW^{*}_{q}.\end{aligned}$$ Consequently, $$\begin{aligned}
E \left( \ln_{q}S_{n}^{*(q)} \right) & \geq & E \left( \ln_{q}S_{n} \right) ,\end{aligned}$$ i.e., the $\ln_{q}$-optimal portfolio maximizes the expected $\ln_{q}$ of the final wealth. So we show theoretically that the q-Portfolio will provide higher relative wealth results that the Cover’s portfolio and, for $q \rightarrow 1$, we recover the Cover’s portfolio theory from the q-Portfolio.
Computational results \[sec3\]
==============================
In this section, we apply our formulation in the Brazilian stock market $[B]^{3}$ - Brazil, Stock Exchange and Over-the-Counter Market located at São Paulo - Brazil, using *R* language and environment [@r-cran]. We use the package GetHFData [@gethfdata] that download and aggregate high frequency data from Brazilian stock market using File Transfer Protocol (FTP) and the package DEoptim [@deoptim] that implements the Differential Evolution Algorithm [@diffevol]. This is a useful method for solution of global optimization problems. Also, we use the packages Pracma, Cubature and R2Cuba [@pracma; @cubature; @becs] that compute numerical multi-dimensional integration from Gauss-Kronrod, hypercubes and Monte Carlo methods, respectively.
Importing the Brazilian stock market data from the period 01/01/2018 to 04/30/2018, which represent 80 days of movement in $[B]^{3}$, we choose analyze the stocks with higher trading in $[B]^{3}$ as: Brazilian Petroleum Corporation - Petrobas (PETR4), Vale S.A. (VALE3), Bank of Brazil (BBAS3) and Bradesco Bank (BBDC4).
The Gaussian distribution was used in the Cover’s portfolio theory [@muller88; @inuiguchi2000; @li2013; @tunc2013; @matesanz08; @markowitz12], as well as $q$-Gaussian distribution [@zhao2018]. To apply our formalism describe in section \[sec2\] we choose a multivariate $q$-Gaussian as joint probability density distribution of the vector of price relatives, as defined in (\[qGau\]), and compare this results with the multivariate Gaussian distribution for the Cover’s portfolio theory related to relative prices [@Cover5].
The parameters $q$, $\mu_{i}$ and $\sigma_{i}$ are estimated through maximum likelihood method for each set of stocks, and then the growth rate is maximized in order to construct the Cover’s portfolio for the Gaussian distribution and $q$-Portfolio.
After maximization, we apply the acquired portfolio for the next month 05/01/2018 to 05/31/2018, that represent 21 days of movement in $[B]^{3}$, and calculate the wealth relative for two stocks as present in figure \[fig-2\], and for three and four stocks as shows in figure \[fig-3\]. Analyzing this figures we can observe that the q-portfolio brings a better results that the Gaussian Cover’s portfolio. Indeed the q-portfolio presents a higher wealth relative in $67,10 \%$ of all days, achieving $75,00 \%$ for three stocks set.
\(a) (b)
\(c) (d)
\(e) (f)
\(a) (b)
\(c) (d)
(e)
We present the results of wealth relative to two, three and four stocks for the period in table \[tab-2\], which reinforce that wealth relative for $q$-portfolio presents higher value at the end of the period in almost all situations, except for BBAS3 and PETR4 case.
[cccccccccccc]{} & & & & & & & & & & & BBAS3\
& & & & & & & & & & & BBDC4\
& & & & & & & & & & & PETR4\
& & & & & & & & & & & VALE3\
Wealth relative & & & & & & & & & & &\
Gaussian CP & & & & & & & & & & &\
Wealth relative & & & & & & & & & & &\
$q$-Portfolio & & & & & & & & & & &\
$q$-Wealth relative & & & & & & & & & & &\
$q$-Portfolio & & & & & & & & & & &\
& & & & & & & & & & &\
& & & & & & & & & & &\
At end of the period of 21 days of movement, we can see that wealth relative for $q$-Portfolio brings better results that the Gaussian Cover’s portfolio in ten of the eleven cases analyzed. In eight of the eleven cases the q-portfolio begin with higher values that the Gaussian Cover’s portfolio, brings better results at the end of the period, showing that our method deals better with Brazilian stock market $[B]^{3}$ data compared to the Gaussian Cover’s portfolio theory, being able to predict a higher wealth relative value.
The parameter $q$ in all cases is between $1.39$ to $1.65$ closer to value obtained in [@TsBor; @Borland], indicating that behavior followed by the stock market is well represented by this range of parameters in Tsallis statistics. In the same way we can apply our formalism to the case with several stocks and analyze the portfolio of many sizes.
A measure of the portfolio’s risk-adjusted return can be realized by calculating the Sharpe ratio $$\begin{aligned}
S_{a} & = & \displaystyle \frac{E \left[ R_{t} - R_{f} \right] }{ \sigma \left[ R_{t} \right] } \end{aligned}$$ and the Sortino ratio $$\begin{aligned}
S_{o} & = & \displaystyle \frac{ E \left[ R_{t} - T \right] }{TDD}\end{aligned}$$ where $R_{t}$ is portfolio’s return, $R_{f}$ denotes the risk-free rate of return, $\sigma \left[ R_{t} \right]$ is the standard deviation of the portfolio return, $T$ denotes the target or require of return for an investment and $TDD = \sqrt{\frac{1}{N} \sum_{i=1}^{N} min(0,R_{i} - T)^{2} }$ is the target downside deviation.
We use a risk free rate and a require rate of returns $R_{f}= T = 0$ and calculate the Sharpe and Sortino ratio for the q-Portfolio and the Gaussian Cover’s Portfolio. The results are present in table \[tab-s\] and for $63,6\%$ of the cases the q-Portfolio performs better that the Gaussian Cover’s Portfolio for both, Sharpe Ratio and Sortino Ratio.
[cccccccccccc]{} & & & & & & & & & & & BBAS3\
& & & & & & & & & & & BBDC4\
& & & & & & & & & & & PETR4\
& & & & & & & & & & & VALE3\
Sharpe Ratio & & & & & & & & & & &\
Gaussian CP & & & & & & & & & & &\
Shape Ratio & & & & & & & & & & &\
q-Portfolio & & & & & & & & & & &\
Sortino Ratio & & & & & & & & & & &\
Gaussian CP & & & & & & & & & & &\
Sortino Ratio & & & & & & & & & & &\
$q$-Portfolio & & & & & & & & & & &\
The values from table \[tab-s\] indicating that q-Portfolio present a better risk-adjusted performance considering the total volatility and the downside risk, being able to generate a higher return that the Gaussian Cover’s Portfolio.
Conclusions \[sec4\]
=====================
The Tsallis statistic allows us to generalize usual concepts through a non-extensive parameter $q$. At the limit $q \rightarrow 1$, the usual expressions are recovered. Examples of interest include deformations of functions and algebraic operations. We explore these tools in Cover’s approach from information theory to portfolio theory. In this way we define the growth $q$-rate $W_{q}$ of a stock market portfolio, the optimal growth $q$-rate $W_{q}^{*}$, $q$-wealth after n days using the portfolio $b^{*}$ and we show that the $q$-wealth after $n$ days with the optimal portfolio is given by the $q$-exponential function. In the context of causal portfolio we studied the asymptotic optimality and we derive that for, almost every sequence from stock market the optimal $q$-wealth after n days is greater than the $q$-wealth of any investor. It is important to note that the parameter $q$ establishes a correlation between stock prices on each day. In fact, by equation (\[qGau\]), we can note that the proposed $q$-Gaussian is not a product of $q$-Gaussians, but it is a $q$-product. Based on this approach a Brazilian stock market analysis was performed in a small number of stocks. We show that $q$-wealth is better suited to empirical data than standard wealth and the optimal value for the non-extensive parameter $q$ between $1.39$ to $1.65$. An analysis of the Sharpe ratio and Sortino ratio also ratifies advantages in using q-Portfolio. As perspectives we intend to investigate i.i.d. markets to time-dependent market processes in this scenario. Furthermore, we pretend apply this formulation to analyze the Post-Modern Portfolio Theory exploring the nonextensive version of Downside Risk and of log-normal distribution.
[00]{}
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Vincent De Comarmond$^1$, Robert de Mello Koch$^{1,2}$ and Katherine Jefferies$^{1}$\
$^{1}$ National Institute for Theoretical Physics,\
Department of Physics and Centre for Theoretical Physics,\
University of the Witwatersrand,\
Wits, 2050, South Africa\
\
$^{2}$Stellenbosch Institute for Advanced Studies,\
Stellenbosch, South Africa\
\
E-mail:
title: Surprisingly Simple Spectra
---
Introduction
============
In the last few years interesting new progress has been made in the study of the dynamics of multimatrix models. Starting from the remarkable observation[@Corley:2001zk] that the Schur polynomials are a complete basis of gauge invariant operators which diagonalize the two point function of the free ($g_{YM}^2=0$) super Yang-Mills theory, similar bases have been found for multimatrix models[@Balasubramanian:2004nb; @de; @Mello; @Koch:2007uu; @de; @Mello; @Koch:2007uv; @Kimura:2007wy; @Bekker:2007ea; @Brown:2007xh; @Bhattacharyya:2008rb; @Brown:2008rr; @Kimura:2008wy; @Kimura:2009wy; @Ramgoolam:2008yr]. For these bases, the two point function is diagonal and known exactly as a function of $N$ (but of course, at $g_{YM}^2=0$). The fact that the $N$ dependence is known exactly suggests that these results will be useful for going beyond the planar approximation.
When would this be needed? Often “the planar limit” and “the large $N$ limit” are taken as synonyms. This is not, in general, accurate. For example, if we imagine computing the two point correlator of an operator with a bare dimension $\Delta$ of most $\Delta\sim J$ with ${J^2\over N}\ll 1$, then summing the planar diagrams will capture the large $N$ limit. For operators with a dimension larger than this, exploding combinatoric factors overpower the nonplanar (${1\over N^2}$) suppression and the planar approximation is completely ineffective[@Balasubramanian:2001nh]. In this scenario, to get the correct large $N$ limit, it is necessary to sum a lot more than just the planar diagrams. On general grounds we expect the large $N$ limit to be simpler than the full theory[@'tHooft:1973jz]. The planar diagrams are a small subset of all possible diagrams, so that it is quite natural to expect that summing only the planar diagrams will give a much simpler problem. Why should the large $N$ limit be simple when one needs to sum much more than just the planar diagrams? The answer to this question will probably not be general, but rather will depend on the specific dynamical problem considered and must be answered case by case. A very pedestrian approach is simply to compute the large $N$ limit and then to look for simplifications. This has been accomplished[@Koch:2008ah; @Koch:2008cm; @Koch:2009jc] in a number of interesting examples including LLM geometries [@Lin:2004nb; @Balasubramanian:2005mg] and the near horizon geometry of a bound state of giant gravitons[@shahin]. [*The results are remarkably simple*]{}. Indeed, as an example, for ${1\over 2}$ BPS-correlators in the presence of $M$ giant gravitons with $M$ of order $N$, [@Koch:2008ah; @Koch:2008cm; @Koch:2009jc] showed that the usual ${1\over N}$ expansion is replaced by a ${1\over M+N}$ expansion. Further, if one expands the exact correlators (which because they are ${1\over 2}$ BPS do not depend on $g_{YM}^2$ but only on $N$ or $N+M$) the expansion coefficients for correlators in the background of $M$ giants are exactly the same as the expansion coefficients for correlators with no giants present! This remarkably simple result was confirmed holographically[@Maldacena:1997re] by matching to graviton dynamics in the LLM geometries using the formalism of [@Skenderis:2007yb]. For near-BPS operators corresponding to BMN loops[@Berenstein:2002jq] it was argued in [@Chen:2007gh; @us; @hai1] that the usual ’t Hooft coupling $g_{YM}^2N$ is replaced by the effective ’t Hooft coupling $g_{YM}^2 (N+M)$. For additional interesting related studies see[@shahin; @hai2].
In this article we will consider the problem of computing the anomalous dimension of an operator with a bare dimension of order $N$. To answer this question, we need to go well beyond the planar limit; we find the methods and approach of[@Balasubramanian:2004nb; @de; @Mello; @Koch:2007uu; @de; @Mello; @Koch:2007uv; @Kimura:2007wy; @Bekker:2007ea; @Brown:2007xh; @Bhattacharyya:2008rb; @Brown:2008rr; @Kimura:2008wy; @Kimura:2009wy] once again surprisingly effective. The results again exhibit a remarkable simplicity - the spectrum of anomalous dimension can be matched to the spectrum of a set of oscillators! We see once again that the large $N$ limit is indeed a simple limit.
The operators we consider, restricted Schur polynomials, will be built using $O(N)$ $Z$s and 3 or 4 “impurities” ($Y$s), where $Z$ and $Y$ are complex adjoint scalars of ${\cal N}=4$ super Yang Mills theory. The case of operators with two impurities was studied in[@Koch:2010gp]. The dilatation operator when acting on a restricted Schur polynomial, produces terms that have a combination $ZY-YZ$ appearing. In [@Koch:2010gp], the techniques of [@de; @Mello; @Koch:2007uv; @Bekker:2007ea] were used to separate the $Z$ and the $Y$ and then write the results as a linear combination of restricted Schur polynomials. This method is very cumbersome as it involves the inversion of a matrix. This must be done analytically so its tedious (for the case of two impurities one must invert a $6\times 6$ matrix). More than two impurities was effectively out of reach. In this article we develop a new formula (in section 2) which avoids this matrix inversion. This allows us to handle the cases of three and four impurities without much trouble. The resulting formulas for the dilatation operator are quite lengthy (see Appendix A), but their spectrum is surprisingly simple.
[*Our results suggest that for the class of operators considered, the Hamiltonian defined by the dilatation operator is integrable - it is just a set of oscillators. This is an example of an integrable dilatation operator, obtained by summing both planar and non-planar diagrams.* ]{}
The operators we consider can be mapped to giant gravitons[@McGreevy:2000cw] in spacetime[@Balasubramanian:2001nh; @Corley:2001zk; @Berenstein:2004kk]. There is in fact already a known connection between the geometry of giant gravitons and harmonic oscillators[@Mikhailov:2000ya; @Beasley:2002xv; @Kinney:2005ej; @Biswas:2006tj; @Mandal:2006tk; @jurgis]. Our work differs from these results in at least two important ways. Firstly, we claim that the complete spectrum (not just the BPS spectrum!) has a connection to harmonic oscillators. Secondly, we have very good control over the set of operators we consider. Our operators are dual to a two giant system. Previous studies captured the full set of BPS states and consequently were not able to distinguish (for example) giant graviton plus graviton from excited giant graviton. Our study captures only states of the two giant system. Thus, it is rather natural to associate our oscillators with excitations modes of a giant graviton. [*In the same way that there are oscillators in the worldsheet theory of a string describing the oscillation modes of a string, we are describing the oscillators that would be present in a world volume description of giant gravitons describing the oscillations modes of a giant graviton D3 brane.*]{}
The problem of computing anomalous dimensions for operators with a large ${\cal R}$-charge has been considered before by a number of authors. The restricted Schur polynomials we consider in this article are built by distributing “impurities” (given by $Y$s), in an operator built mainly from $Z$s. If we replace the $Y$s by words containing $O(\sqrt{N})$ letters (which may be $Z$, $Y$ or other field or derivative of a field), these words are naturally identified with open strings[@Balasubramanian:2002sa; @Sadri:2003mx; @Berenstein:2003ah]. In this case the dilatation operator reproduces the dynamics of open strings ending on a giant graviton[@Berenstein:2006qk; @de; @Mello; @Koch:2007uu; @de; @Mello; @Koch:2007uv; @Bekker:2007ea]. The mixing of operators is highly constrained. Indeed, in [@de; @Mello; @Koch:2007uv; @Bekker:2007ea] it was shown that operators which mix can differ at most by moving one box around on the Young diagram labeling the operator. Another interesting basis to consider is the Brauer basis[@Kimura:2007wy; @Kimura:2009wy]. This basis is built using Brauer algebra projectors. The structure constants of the Brauer algebra are $N$ dependent. There is an elegant construction of a class of BPS operators [@yusuke] in which the natural $N$ dependence appearing in the definition of the operator[@heslop] is reproduced by the Brauer algebra projectors[@yusuke]. Finally, another very natural approach to the problem, is to adopt a basis that has sharp quantum numbers for the global symmetries of the theory[@Brown:2007xh; @Brown:2008rr]. The action of the anomalous dimension operator in this sharp quantum number basis is very similar to the action in the restricted Schur basis: again operators which mix can differ at most by moving one box around on the Young diagram labeling the operator[@Brown:2008rs]. For further related interesting work see [@tomyusuke; @Huang:2010ne]. Finally, for a rather general approach which correctly counts and constructs the weak coupling BPS operators see[@jurgis].
We now conclude this introduction with a description of what is to follow. In section 2 we will describe the action of the dilatation operator on restricted Schur polynomials. The main result of this section is formula (\[final\]) which gives a very explicit description of the action of the dilatation operator. In section 3 we will describe, in broad terms, the set of operators we consider and explain how the dynamics of the large $N$ limit simplifies. In section 4 we explain how to construct the projectors needed to evaluate the action of the dilatation operator. A very detailed description of the specific operators we consider is given in section 5; we also describe a limit in which our system of two giants should be well described as a system of $D3$ brane giants plus open strings. In this limit we see that the dilatation operator reduces to a lattice version of the second derivative, with the Young diagram labels of the restricted Schur polynomials defining the lattice. In section 6 we present our numerical results and draw some general conclusions from them in section 7. The explicit result for the dilatation operator is given in Appendix A; the intertwiners which enter into the expression for the dilatation operator are described in Appendix B.
Action of the Dilatation Operator
=================================
[*In this section we will study the action of the one loop dilatation operator on restricted Schur polynomials built using two complex adjoint scalars. The main result of this section is the surprisingly simple result (\[final\]) for the action of the dilatation operator.*]{}
[0.2cm]{}
We will consider the action of the one loop dilatation operator in the $SU(2)$ sector[@Beisert:2003tq] of ${\cal N}=4$ super Yang Mills theory $$D = - g_{\rm YM}^2 {\rm Tr}\,\big[ Y,Z\big]\big[ \partial_Y ,\partial_Z\big]$$ on the restricted Schur polynomial $$\chi_{R,(r,s)}(Z^{\otimes \, n},Y^{\otimes \, m})
=
{1\over n!m!}\sum_{\sigma\in S_{n+m}}{\rm Tr}_{(r,s)}(\Gamma_R(\sigma))
Z^{i_1}_{i_{\sigma(1)}}\cdots Z^{i_n}_{i_{\sigma(n)}}Y^{i_{n+1}}_{i_{\sigma(n+1)}}\cdots Y^{i_{n+m}}_{i_{\sigma(n+m)}}\, .$$ The labels of our restricted Schur polynomial $\chi_{(R,(r,s))}$ are (i) $R$, which is a Young diagram with $n+m$ boxes or equivalently an irreducible representation of $S_{n+m}$, (ii) $r$, which is a Young diagram with $n$ boxes or equivalently an irreducible representation of $S_n$ and (iii) $s$ which is a Young diagram with $m$ boxes or equivalently an irreducible representation of $S_m$. The notation ${\rm Tr}_{(r,s)}$ implies that one should only trace over the subspace carrying the irreducible representation[^1] $(r,s)$ of $S_n\times S_m$ inside the carrier space for irreducible representation $R$ of $S_{n+m}$. This trace is most concretely realized by including a projector $P_{R\to (r,s)}$ (from the carrier space of $R$ to the carrier space of $(r,s)$) and tracing over all of $R$. A simple calculation yields[^2] [$$\begin{aligned}
\label{basicreslt}
D \, \chi_{R,(r,s)}(Z^{\otimes \, n},Y^{\otimes \, m})=&&
{g_{\rm YM}^2\over (n-1)!(m-1)!}\sum_{\psi\in S_{n+m}}
\Tr_{(r,s)}\left(\Gamma_R ((n,n+1) \psi -\psi (n,n+1))\right)\times\cr
\times Z^{i_1}_{i_{\psi (1)}}&&
\cdots
Z^{i_{n-1}}_{i_{\psi (n-1)}}
(YZ-ZY)_{i_{\psi (n)}}^{i_n}\delta^{i_{n+1}}_{i_{\psi (n+1)}}
Y^{i_{n+2}}_{i_{\psi (n+2)}}\cdots Y^{i_{n+m}}_{i_{\psi (n+m)}}
\, .\end{aligned}$$ ]{} The sum over $\psi$ runs only over permutations for which $\psi (n+1)=n+1$. To perform this sum over $\psi$, write the sum over $S_{n+m}$ as a sum over cosets of the $S_{n+m-1}$ subgroup obtained by keeping precisely those permutations that satisfy $\psi (n+1)=n+1$. The result follows immediately from the reduction rule for Schur polynomials (see [@de; @Mello; @Koch:2004ws] and appendix C of [@de; @Mello; @Koch:2007uu]) [$$\begin{aligned}
\nonumber
D \, \chi_{R,(r,s)}=&&
{g_{\rm YM}^2\over (n-1)!(m-1)!}\sum_{\psi\in S_{n+m-1}}\sum_{R'}c_{RR'}\,\Tr_{(r,s)}\Big(\Gamma_R ((n,n+1))\Gamma_{R'}(\psi) \cr
&&-\Gamma_{R'}(\psi) \Gamma_R((n,n+1))\Big)\, Z^{i_1}_{i_{\psi (1)}}
\cdots
Z^{i_{n-1}}_{i_{\psi (n-1)}}
(YZ-ZY)_{i_{\psi (n)}}^{i_n}
Y^{i_{n+2}}_{i_{\psi (n+2)}}\cdots Y^{i_{n+m}}_{i_{\psi (n+m)}}
\, .\end{aligned}$$ ]{} The sum over $R'$ runs over all representations that can be subduced from $R$. Concretely, $R'$ runs over all Young diagrams that can be obtained from $R$ by dropping a single box; $c_{RR'}$ is the weight[^3] of the box that must be removed from $R$ to obtain $R'$. We will make use of the following notation for restricted characters $$\chi_{R,(r,s)}(\sigma )={\rm Tr}_{(r,s)}\Big(\Gamma_R(\sigma)\Big)={\rm Tr}\Big(P_{R\to (r,s)}\Gamma_R(\sigma)\Big)\, .$$ Now, using the identity (bear in mind that $\psi (n+1)=n+1$) $$Z^{i_1}_{i_{\psi (1)}} \cdots Z^{i_{n-1}}_{i_{\psi (n-1)}} (YZ-ZY)_{i_{\psi (n)}}^{i_n}
Y^{i_{n+2}}_{i_{\psi (n+2)}} \cdots Y^{i_{n+m}}_{i_{\psi (n+m)}}
={\rm Tr}\left(\Big((n,n+1)\, \psi -\psi\,(n,n+1)\Big)Z^{\otimes n}Y^{\otimes m} \right)$$ where $${\rm Tr} (\sigma Z^{\otimes n}Y^{\otimes m})=
Z^{i_1}_{i_{\sigma(1)}}\cdots Z^{i_n}_{i_{\sigma(n)}}Y^{i_{n+1}}_{i_{\sigma(n+1)}}\cdots Y^{i_{n+m}}_{i_{\sigma(n+m)}}\, ,$$ and (this identity is proved in [@Bhattacharyya:2008rc]) $${\rm Tr} (\sigma Z^{\otimes n}Y^{\otimes m})=\sum_{T,(t,u)}{d_T n! m!\over d_t d_u (n+m)!}\chi_{T,(t,u)}(\sigma^{-1})\chi_{T,(t,u)}(Z,Y)$$ we obtain $$D\chi_{R,(r,s)}(Z,Y)=\sum_{T,(t,u)} M_{R,(r,s);T,(t,u)}\chi_{T,(t,u)}(Z,Y)\, ,$$ [$$M_{R,(r,s);T,(t,u)}=g_{YM}^2\sum_{\psi\in S_{n+m-1}}\sum_{R'}
{c_{RR'} d_T n m\over d_t d_u (n+m)!}
\Tr_{(r,s)}\Big(\Gamma_R ((n,n+1))\Gamma_{R'}(\psi)-\Gamma_{R'}(\psi) \Gamma_R((n,n+1))\Big)\times$$ $$\times \chi_{T,(t,u)}(\psi^{-1}(n,n+1) - (n,n+1)\psi^{-1})\, .$$ ]{} The sum over $\psi$ can be done by using the fundamental orthogonality relation $$\begin{aligned}
\label{final}
M_{R,(r,s);T,(t,u)} &&= - g_{YM}^2\sum_{R'}{c_{RR'} d_T n m\over d_{R'} d_t d_u (n+m)}
\Tr\Big( \Big[ \Gamma_R((n,n+1)),P_{R\to (r,s)}\Big]I_{R'\, T'}\times \cr
&&\times \Big[\Gamma_T((n,n+1)),P_{T\to (t,u)}\Big] I_{T'\, R'}\Big) \, .\end{aligned}$$ The reader should consult Appendix B for a definition of the intertwiners $I_{R'\, T'}$. This expression for the one loop dilatation operator is exact in $N$. It is one of the key results of this article.
To obtain the spectrum of anomalous dimensions, we need to consider the action of the dilatation operator on normalized operators. The two point function for restricted Schur polynomials has been computed in [@Bhattacharyya:2008rb] $$\langle\chi_{R,(r,s)}(Z,Y)\chi_{T,(t,u)}(Z,Y)^\dagger\rangle =
\delta_{R,(r,s)\,T,(t,u)}f_R {{\rm hooks}_R\over {\rm hooks}_{r}\, {\rm hooks}_s}\, .$$ In this expression $f_R$ is the product of the weights in Young diagram $R$ and ${\rm hooks}_R$ is the product of the hook lengths of Young diagram $R$. The normalized operators can be obtained from $$\chi_{R,(r,s)}(Z,Y)=\sqrt{f_R \, {\rm hooks}_R\over {\rm hooks}_r\, {\rm hooks}_s}O_{R,(r,s)}(Z,Y)\, .$$ In terms of these normalized operators $$DO_{R,(r,s)}(Z,Y)=\sum_{T,(t,u)} N_{R,(r,s);T,(t,u)}O_{T,(t,u)}(Z,Y)$$ [$$N_{R,(r,s);T,(t,u)}= - g_{YM}^2\sum_{R'}{c_{RR'} d_T n m\over d_{R'} d_t d_u (n+m)}
\sqrt{f_T \, {\rm hooks}_T\, {\rm hooks}_r \, {\rm hooks}_s \over f_R \, {\rm hooks}_R\, {\rm hooks}_t\, {\rm hooks}_u}\times$$ $$\times\Tr\Big(\Big[ \Gamma_R((n,n+1)),P_{R\to (r,s)}\Big]I_{R'\, T'}\Big[\Gamma_T((n,n+1)),P_{T\to (t,u)}\Big]I_{T'\, R'}\Big) \, .$$ ]{} This last expression will be used later when we numerically study the spectrum of the dilatation operator.
Excited Giant Graviton Bound States
===================================
[*The goal of this section is to clearly define the class of operators being considered and to outline the approximations that can be made in the large $N$ limit.*]{}
[0.2cm]{}
In this article we will study restricted Schur polynomials labeled by Young diagrams with at most two columns. The number of $Z$s appearing is $\alpha N$ where $2-\alpha\equiv\zeta\ll 1$. The number of $Y$s appearing is fixed to be $O(1)$. These operators are dual to giant gravitons that wrap an $S^3$ in the $S^5$ of the AdS$_5\times$S$^5$ background. Since the restricted Schur polynomials furnish a suitable basis for the two giant system, we know that these operators capture all excitations (BPS and non supersymmetric) of the two giant system. For a study of excitations of the single giant system using restricted Schur polynomials see [@Koch:2010gp]. For a spacetime study of excitations of the single giant system using the Born-Infeld action see [@Das:2000st]. The most important result from [@Das:2000st; @Koch:2010gp] for us, is that all the deformations of the single threebrane giant graviton that we consider, are supersymmetric.
The mixing of these operators with restricted Schur polynomials that have three columns (or more) is suppressed by a factor of order ${1\over\sqrt{N}}$. This factor arises from the normalization of the restricted Schur polynomials: the three column restricted Schur polynomials (with one short column) have a two point function which is smaller than the two column restricted Schur polynomials by a factor of order ${1\over N}$[@Koch:2010gp]. Thus, at large $N$ we can focus on the two column restricted Schur polynomials, which is a huge simplification. The analog of the statement that for operators with a dimension of $O(1)$, different trace structures do not mix is: [*at large $N$ restricted Schur polynomials $\chi_{R,(r,s)}$ with $R$ a Young diagram with $n$ columns, each of which has length of $O(N)$, do not mix with operators $\chi_{R',(r',s')}$ that have $n'\ne n$ columns.*]{} The fact that the two column restricted Schur polynomials are a decoupled sector at large $N$ is to be expected. Indeed, at large $N$ these operators correspond to a well defined stable semi-classical object in spacetime (the two giant system). We expect that $n$ column restricted Schur polynomials are also a decoupled sector at large $N$ for the same reason.
Simple Projectors
=================
[*When all Young diagrams labeling the restricted Schur polynomial have at most 2 columns, the projector $P_{R\to (r,s)}$ simplifies dramatically. The goal of this section is to explain the simplification and exploit it to efficiently build the relevant projectors.*]{}
[0.2cm]{}
The projector $P_{R\to (r,s)}$ projects from representation $R$ of $S_{n+m}$ to representation $(r,s)$ of $S_n\times S_m$. One issue, which complicates things considerably, is that representation $(r,s)$ can be subduced more than once when irreducible representation $R$ is decomposed into irreducible representations of the $S_n\times S_m$ subgroup. Giving a general rule to specify precisely how this multiplicity is resolved is nontrivial. For the operators we consider in this article, this problem does not arise. As soon as a system of three or more gaint gravitons are considered, it will be necessary to deal with this issue. Fortunately a well defined approach to resolving these multiplicities has been outlined in [@Kimura:2008wy]. Basically, [@Kimura:2008wy] considers elements in the group algebra $CS_{n+m}$ which are invariant under conjugation by $CS_n\times CS_m$. The Cartan subalgebra of these elements are the natural generalization of the Jucys-Murphy elements which define a Cartan subalgebra for $S_n$[@Okunkov]. The multiplicities will be labelled by the eigenvalues of this Cartan subalgebra[@Kimura:2008wy]. It would be very interesting to work out the details of this proposal in the context of multigiant systems.
Given $R$ we imagine removing some boxes; after removing these boxes one is left with $r$. The removed boxes are assembled to produce $s$. If we specify both $R$ and the boxes that are to be removed to obtain $r$ we obtain every representation exactly once. For example, consider all possible operators that can be constructed from the following $R$ by removing the three boxes shown $$\young({\,}{\,},{\,}{\,},{\,}{*},{\,}{*},{\,},{\,},{\,},{*})$$ For the boxes that are removed, we must respect edges that are joined, which means the two boxes removed from the short column must remain stacked on top of each other. Thus, there are two possible irreducible representations $s$ that can be produced, implying we can subduce two possible irreducible representations of $S_9 \times S_3$, by removing these three boxes $$\young({\,}{\,},{\,}{\,},{\,},{\,},{\,},{\,},{\,})\, \young({\,},{\,},{\,})\qquad
\young({\,}{\,},{\,}{\,},{\,},{\,},{\,},{\,},{\,})\, \young({\,}{\,},{\,})$$ Thus, in total we’d get the following $S_9 \times S_3$ irreducible representations subduced from the $R$ given above (sum over all possible ways to remove boxes to get this result) $$\young({\,}{\,},{\,}{\,},{\,}{\,},{\,}{\,},{\,})\,\young({\,},{\,},{\,}) \oplus
\young({\,}{\,},{\,}{\,},{\,}{\,},{\,},{\,},{\,})\,\young({\,},{\,},{\,}) \oplus
\young({\,}{\,},{\,}{\,},{\,}{\,},{\,},{\,},{\,})\,\young({\,}{\,},{\,}) \oplus
\young({\,}{\,},{\,}{\,},{\,},{\,},{\,},{\,},{\,})\,\young({\,},{\,},{\,}) \oplus
\young({\,}{\,},{\,}{\,},{\,},{\,},{\,},{\,},{\,})\,\young({\,}{\,},{\,}) \oplus
\young({\,}{\,},{\,},{\,},{\,},{\,},{\,},{\,},{\,})\,\young({\,},{\,},{\,})$$ The dimensions of these representations are 42, 48, 96, 27, 54 and 8 respectively. This sums to give 275 which is indeed the dimension of $R$.
To construct the actual projector, we need only build an operator which will assemble the removed boxes in the correct way to produce $s$. We will give an example of how to construct this operator; the general case should be clear. Lets start with the representation $R$ shown, removing the boxes indicated $$\young({\,}{\,},{\,}{*},{\,},{*})$$ Our projector will act in the subspace spanned by the two sets of states $$|1\rangle =|\young({\,}{\,},{\,}{2},{\,},{1})\rangle\qquad
|2\rangle =|\young({\,}{\,},{\,}{1},{\,},{2})\rangle$$ We are using a Young-Yamonouchi basis. Thus, each state given above could be any one of $d_{\tiny\yng(2,1,1)}=3$ states, corresponding to the number of ways to complete the labels. When acting in the subspace, the operator which organizes the boxes into representation $s$ is $$P_s={d_s\over m!}\sum_{\sigma\in S_m}\chi_s(\sigma)\Gamma_R(\sigma)\, .$$ All that remains is to supply a formula for the action of $\Gamma_R(\sigma)$ (for $\sigma =1,(12)$) when acting on the subspace spanned by $|1\rangle$ and $|2\rangle$. The action of $\Gamma_R(\sigma)$ on any Young-Yamonouchi state is well known. For the states above $$\Gamma_R\left((12)\right)|1\rangle = -{1\over 3}|1\rangle +{\sqrt{8}\over 3}|2\rangle$$ $$\Gamma_R\left((12)\right)|2\rangle = {1\over 3}|2\rangle +{\sqrt{8}\over 3}|1\rangle$$ so that $$\Gamma_R\left( (12)\right)=-{1\over 3}|1\rangle\langle 1| +{\sqrt{8}\over 3}|2\rangle\langle 1|
+{1\over 3}|2\rangle\langle 2| +{\sqrt{8}\over 3}|1\rangle\langle 2|$$ and $$\Gamma_R\left( 1\right)= |1\rangle\langle 1| + |2\rangle\langle 2| \, .$$
The Radial Direction
====================
[*In this section we describe a limit in which the dilatation operator simplifies significantly. There are two columns in the Young diagrams labeling the restricted Schur polynomials. When the first column contains $O(\sqrt{N})$ boxes more than the second, the dilatation operator simplifies to a lattice realization of the second derivative. The Young diagram label itself defines the lattice.*]{}
[0.2cm]{}
Three Impurities
----------------
The three impurity operators are built using many $Z$s and three $Y$s. To specify these operators, we need to give the three Young diagrams labeling the restricted Schur polynomial. The second Young diagram, $r$, (which specifies an irreducible representation of $S_n$) is specified by stating the number of rows with two boxes ($=b_0$) and the number of rows with a single box ($=b_1$). The third Young diagram label, $s$, (which specifies an irreducible representation of $S_3$) and the first Young diagram label, $R$, (which specifies an irreducible representation of $S_{n+3}$) can now be built from $r$ by specifying which boxes in $R$ are to be removed to obtain $r$ and how these boxes are to be organized into an $S_3$ irreducible representation. There are 6 possibilities $$\chi_A(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,2,2,1,1)\,\yng(1,1,1)}(Z,Y)
\qquad
\chi_B(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,2,1,1,1,1)\,\yng(1,1,1)}(Z,Y)$$ $$\chi_C(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,2,1,1,1,1)\,\yng(2,1)}(Z,Y)
\qquad
\chi_D(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,1,1,1,1,1,1)\,\yng(1,1,1)}(Z,Y)$$ $$\chi_E(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,1,1,1,1,1,1)\,\yng(2,1)}(Z,Y)
\qquad
\chi_F(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,1,1,1,1,1,1,1,1)\,\yng(1,1,1)}(Z,Y)$$ The corresponding normalized operators are denoted using the capital letter $O$. In view of the discussion of section 3, we know that $b_0$ is $O(N)$ and $b_1$ ranges from 0 or 1 to $O(N)$. The action of the dilatation operator is given in Appendix A.
The ${\cal R}$-charge of an operator in the field theory maps into the angular momentum of the dual string theory state. Thanks to the Myers effect[@Myers:1999ps] the angular momentum of the string theory state determines its size. Identifying the two columns of the Young diagrams with the two threebranes, the number of boxes in each column determines the angular momentum and hence the size of each threebrane. In the limit that $N-b_0=O(N)$, $b_0=O(N)$ and $b_1=O(\sqrt{N})$ we have non-maximal giants which are separated by a distance of $O(1)$ in string units. In this limit, we expect the dynamics to simplify. Indeed, the system should be described by two $D3$ brane giant gravitons with open strings stretching between then. The action of the dilatation operator becomes $$DO_A(b_0,b_1) =g_{YM}^2\left(N-b_0\right)\times O\left({1\over b_1}\right)$$ $$\begin{aligned}
\nonumber
DO_B(b_0,b_1) =&&-{4\over 3}g_{YM}^2\left(N-b_0\right)\left[
O_B(b_0+1,b_1-2)-2\,O_B(b_0,b_1) + O_B(b_0-1,b_1+2) \right]\cr
&&+{2\sqrt {2}\over 3}g_{YM}^2\left(N-b_0\right)\left[O_C(b_0+1,b_1-2)
-2O_C(b_0+1,b_1)+O_C(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_C(b_0,b_1) =&&{2\sqrt {2}\over 3}g_{YM}^2(N-b_0)\left[
O_B(b_0+1,b_1-2)-2O_B(b_0,b_1)+O_B(b_0-1,b_1+2)\right]\cr
&&-{2\over 3}g_{YM}^2(N-b_0)\left[ O_C(b_0+1,b_1-2)-2O_C(b_0,b_1)+O_C(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_D(b_0,b_1) =&&-{4\over 3}g_{YM}^2(N-b_0)\left[
O_D(b_0+1,b_1-2)-2O_D(b_0,b_1)+O_D(b_0-1,b_1+2)\right]\cr
&&+{2\sqrt {2}\over 3}g_{YM}^2(N-b_0)\left[
O_E(b_0+1,b_1-2)-2O_E(b_0,b_1)+O_E(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_E(b_0,b_1) = &&{2\sqrt {2}\over 3}g_{YM}^2(N-b_0)\left[
O_D(b_0+1,b_1-2)-2O_D(b_0,b_1)+O_D(b_0-1,b_1+2)\right]\cr
&&-{2\over 3}g_{YM}^2(N-b_0)\left[
O_E(b_0+1,b_1-2)-2O_E(b_0,b_1)+O_E(b_0-1,b_1+2)\right]\end{aligned}$$ $$DO_F(b_0,b_1) =g_{YM}^2\left(N-b_0\right)\times O\left({1\over b_1}\right)$$ These results have a natural interpretation. Notice that there are four operators for which the $S_m$ representation is the totally antisymmetric representation. We will see that there are also four operators for which the corresponding states remain supersymmetric; this agreement between the number of operators for which the $S_m$ representation is the totally antisymmetric representation and the number of supersymmetric states, holds in general for the two giant system. Looking at the labels, it is natural to interpret $O_A(b_0,b_1)$ as a state in which we deform only the larger threebrane. Recall from section 3, that deforming a single threebrane gives us a supersymmetric state so it seems natural for $O_A(b_0,b_1)$ to remain supersymmetric. Similarly, $O_F(b_0,b_1)$ can be interpreted as a state in which we deform only the smaller threebrane and a similar comment can be made. The fact that the combinations $O_B(b_0,b_1)+\sqrt{2}O_C(b_0,b_1)$ and $O_D(b_0,b_1)+\sqrt{2}O_E(b_0,b_1)$ are annihilated by $D$ implies that there are another two supersymmetric ways to deform the pair of threebranes. Finally, notice that if we set $O_B(b_0,b_1)-O_C(b_0,b_1)/\sqrt{2}\equiv O_{B-C}(b_0,b_1)$ and $O_D(b_0,b_1)-O_E(b_0,b_1)/\sqrt{2}\equiv O_{D-E}(b_0,b_1)$ we have $$\begin{aligned}
\nonumber
DO_{B-C}(b_0,b_1) = &&-2g_{YM}^2(N-b_0)\left[
O_{B-C}(b_0+1,b_1-2)-2O_{B-C}(b_0,b_1)+O_{B-C}(b_0-1,b_1+2)\right]\cr
DO_{D-E}(b_0,b_1) = &&-2g_{YM}^2(N-b_0)\left[
O_{D-E}(b_0+1,b_1-2)-2O_{D-E}(b_0,b_1)+O_{D-E}(b_0-1,b_1+2)\right]\end{aligned}$$ The right hand side again is a discretization of the second derivative. [*It is the Young diagram itself that is defining the lattice.*]{} After recalling that the number of boxes in each column sets the angular momentum and hence the radius[^4] of the corresponding threebrane, its clear that the radius of the giant graviton together with local physics in this radial direction has emerged.
Four Impurities
---------------
For the case of four impurities there are nine possible operators that we can define $$\chi_A(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,2,2,2,1)\, \yng(1,1,1,1)}(Z,Y)
\qquad
\chi_B(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,2,2,1,1,1)\, \yng(1,1,1,1)}(Z,Y)$$ $$\chi_C(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\,\yng(2,2,2,2,2,1,1,1)\, \yng(2,1,1)}(Z,Y)
\qquad
\chi_D(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,2,1,1,1,1,1)\, \yng(1,1,1,1)}(Z,Y)$$ $$\chi_E(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,2,1,1,1,1,1)\, \yng(2,1,1)}(Z,Y)
\qquad
\chi_F(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,2,1,1,1,1,1)\, \yng(2,2)}(Z,Y)$$ $$\chi_G(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,1,1,1,1,1,1,1)\, \yng(1,1,1,1)}(Z,Y)
\qquad
\chi_H(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,2,1,1,1,1,1,1,1)\, \yng(2,1,1)}(Z,Y)$$ $$\chi_I(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,2,1,1,1,1,1)\, \yng(2,2,1,1,1,1,1,1,1,1,1)\, \yng(1,1,1,1)}(Z,Y)$$ Again, the corresponding normalized operators are denoted using the capital letter $O$. The action of the dilatation operator is given in Appendix A.
In the limit that $N-b_0=O(N)$, $b_0=O(N)$ and $b_1=O(\sqrt{N})$ the dynamics again simplifies. The action of the dilatation operator becomes $$DO_A(b_0,b_1)=(N-b_0)g_{YM}^2\times O\left({1\over b_1}\right)$$ $$\begin{aligned}
\nonumber
DO_B(b_0,b_1) =&&-{3\over 2}g_{YM}^2\left(N-b_0\right)\left[
O_B(b_0+1,b_1-2)-2\,O_B(b_0,b_1) + O_B(b_0-1,b_1+2) \right]\cr
&&+{\sqrt {3}\over 2}g_{YM}^2\left(N-b_0\right)\left[O_C(b_0+1,b_1-2)
-2O_C(b_0+1,b_1)+O_C(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_C(b_0,b_1) =&&{\sqrt {3}\over 2}g_{YM}^2(N-b_0)\left[
O_B(b_0+1,b_1-2)-2O_B(b_0,b_1)+O_B(b_0-1,b_1+2)\right]\cr
&&-{1\over 2}g_{YM}^2(N-b_0)\left[ O_C(b_0+1,b_1-2)-2O_C(b_0,b_1)+O_C(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_D(b_0,b_1) =&&-2g_{YM}^2\left(N-b_0\right)\left[
O_D(b_0+1,b_1-2)-2\,O_D(b_0,b_1) + O_D(b_0-1,b_1+2) \right]\cr
&&+{2\over \sqrt{3}}g_{YM}^2\left(N-b_0\right)\left[O_E(b_0+1,b_1-2)
-2O_E(b_0+1,b_1)+O_E(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_E(b_0,b_1) =&&-2g_{YM}^2(N-b_0)\left[
O_E(b_0+1,b_1-2)-2O_E(b_0,b_1)+O_E(b_0-1,b_1+2)\right]\cr
&&+{2\over \sqrt{3}}g_{YM}^2(N-b_0)\left[ O_D(b_0+1,b_1-2)-2O_D(b_0,b_1)+O_D(b_0-1,b_1+2)\right]\cr
&&+{2\sqrt{6}\over 3}g_{YM}^2(N-b_0)\left[ O_F(b_0+1,b_1-2)-2O_F(b_0,b_1)+O_F(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_F(b_0,b_1) =&&-2g_{YM}^2\left(N-b_0\right)\left[
O_F(b_0+1,b_1-2)-2\,O_F(b_0,b_1) + O_F(b_0-1,b_1+2) \right]\cr
&&+{2\sqrt{6}\over 3}g_{YM}^2\left(N-b_0\right)\left[O_E(b_0+1,b_1-2)
-2O_E(b_0+1,b_1)+O_E(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_G(b_0,b_1) =&&-{3\over 2}g_{YM}^2\left(N-b_0\right)\left[
O_G(b_0+1,b_1-2)-2\,O_G(b_0,b_1) + O_G(b_0-1,b_1+2) \right]\cr
&&+{\sqrt{3}\over 2}g_{YM}^2\left(N-b_0\right)\left[O_H(b_0+1,b_1-2)
-2O_H(b_0+1,b_1)+O_H(b_0-1,b_1+2)\right]\end{aligned}$$ $$\begin{aligned}
\nonumber
DO_H(b_0,b_1) =&&-{1\over 2}g_{YM}^2\left(N-b_0\right)\left[
O_H(b_0+1,b_1-2)-2\,O_H(b_0,b_1) + O_H(b_0-1,b_1+2) \right]\cr
&&+{\sqrt{3}\over 2}g_{YM}^2\left(N-b_0\right)\left[O_G(b_0+1,b_1-2)
-2O_G(b_0+1,b_1)+O_G(b_0-1,b_1+2)\right]\end{aligned}$$ $$DO_I(b_0,b_1)=(N-b_0)g_{YM}^2\times O\left({1\over b_1}\right)$$ We can again identify combinations of operators that are annihilated by $D$, that is, that are BPS. Apart from $O_A(b_0,b_1)$ and $O_I(b_0,b_1)$ we have $O_B(b_0,b_1)+\sqrt{3}O_C(b_0,b_1)$, $O_D(b_0,b_1)+\sqrt{3}O_E(b_0,b_1)+\sqrt{2}O_F(b_0,b_1)$ and $O_G(b_0,b_1)+\sqrt{3}O_H(b_0,b_1)$. Notice that all of the BPS operators from this section and the last can be written as $$O_{\rm BPS}(R,r)=\sum_s \sqrt{d_s} O_{R,(r,s)}(b_0,b_1)$$ where $d_s$ is the dimension of the irreducible representation $s$ of the symmetric group. Finally, notice that if we set $\sqrt{3}O_B(b_0,b_1)-O_C(b_0,b_1)\equiv O_{B-C}(b_0,b_1)$, $\sqrt{2}O_D(b_0,b_1)-O_F(b_0,b_1)\equiv O_{D-F}(b_0,b_1)$, $O_D(b_0,b_1)-\sqrt{3}O_E(b_0,b_1)+\sqrt{2}O_F(b_0,b_1)\equiv O_{DF-E}(b_0,b_1)$ and $\sqrt{3}O_G(b_0,b_1)-O_H(b_0,b_1)\equiv O_{G-H}(b_0,b_1)$, we have [$$\begin{aligned}
\nonumber
DO_{B-C}(b_0,b_1) = &&-2g_{YM}^2(N-b_0)\left[
O_{B-C}(b_0+1,b_1-2)-2O_{B-C}(b_0,b_1)+O_{B-C}(b_0-1,b_1+2)\right]\cr
DO_{D-F}(b_0,b_1) = &&-2g_{YM}^2(N-b_0)\left[
O_{D-F}(b_0+1,b_1-2)-2O_{D-F}(b_0,b_1)+O_{D-F}(b_0-1,b_1+2)\right]\cr
DO_{DF-E}(b_0,b_1) = &&-4g_{YM}^2(N-b_0)\left[
O_{DF-E}(b_0+1,b_1-2)-2O_{DF-E}(b_0,b_1)+O_{DF-E}(b_0-1,b_1+2)\right]\cr
DO_{G-H}(b_0,b_1) = &&-2g_{YM}^2(N-b_0)\left[
O_{G-H}(b_0+1,b_1-2)-2O_{G-H}(b_0,b_1)+O_{G-H}(b_0-1,b_1+2)\right]\end{aligned}$$ ]{} The right hand side is again a discretization of the second derivative.
Numerical Results
=================
[*In this section we describe the result of numerically diagonalizing the dilatation operator.*]{}
[0.2cm]{}
When setting up a numerical computation of the spectrum of the anomalous dimension operator, we need to specify the maximum value for the difference between the number of boxes in the long column and the number of boxes in the short column. Given this value, denoted $a_{\rm max}$, we are able to determine how many operators participate in our problem and we are able to describe the resulting spectrum rather explicitly. We will focus on the case of even $a_{\rm max}$. In this case the difference between the number of boxes in the long column and the number of boxes in the short column is always an even number.
Two Impurities
--------------
For a given value of $a_{\rm max}$ there are $2+2a_{\rm max}$ states in total. There are ${3\over 2}a_{\rm max}+1$ zero eigenvalues (corresponding to supersymmetric states). The remaining eigenvalues are $$\lambda_i =8g_{YM}^2 i\qquad i=1,2,\cdots,{a_{\rm max}\over 2}+1\, .$$
Three Impurities
----------------
For a given value of $a_{\rm max}$ there are $1+3a_{\rm max}$ states in total. There are $2 a_{\rm max}$ zero eigenvalues (corresponding to supersymmetric states). The remaining eigenvalues are $$\lambda_i =8g_{YM}^2 i\qquad i=1,2,\cdots,{a_{\rm max}\over 2}\, ,$$ each with a degeneracy of two and a single maximum eigenvalue $\lambda=4a_{\rm max}g_{YM}^2+8g_{YM}^2$. This degeneracy almost certainly indicates a symmetry enhancement in the large $N$ limit.
Four Impurities
---------------
For a given value of $a_{\rm max}$ there are $1 +{9\over 2}a_{\rm max}$ states in total. There are ${5\over 2}a_{\rm max}-1$ zero eigenvalues. The eigenvalues are again evenly spaced with a level spacing of $8g_{YM}^2$ and they are again degenerate. The low lying eigenvalues $$\lambda_i = 8 g_{YM}^2 i\qquad i=1,2,\cdots,{a_{\rm max}\over 2}\, ,$$ have a degeneracy which alternates between 3 and 4. Thus, we find three eigenvalues $\lambda =8g_{YM}^2$, followed by four eigenvalues $\lambda=16g_{YM}^2$, followed by three eigenvalues $\lambda=24g_{YM}^2$, followed by four eigenvalues $\lambda = 32g_{YM}^2$ and so on.
If $a_{\rm max}$ is a multiple of 4, the larger eigenvalues are given by $\lambda=4a_{\rm max}g_{YM}^2 +8g_{YM}^2$, $\lambda=4a_{\rm max}g_{YM}^2 + 16g_{YM}^2$ and $$\lambda_i=4a_{\rm max}g_{YM}^2 + 16g_{YM}^2 +16ig_{YM}^2\qquad i=1,2,\cdots,{a_{\rm max}\over 4}\, ;$$ all of these eigenvalues on the last line above are non-degenerate.
If $a_{\rm max}$ (which by assumption is even) is not a multiple of 4, the larger eigenvalues are given by $\lambda=4a_{\rm max}g_{YM}^2 +8g_{YM}^2$ with a degeneracy of 2 and $$\lambda_i=4a_{\rm max}g_{YM}^2 + 16g_{YM}^2 +16ig_{YM}^2\qquad i=1,2,\cdots,{a_{\rm max}+2\over 4}\, ;$$ all of these eigenvalues on the last line above are non-degenerate.
Once again, the degeneracies observed almost certainly indicate a symmetry enhancement in the large $N$ limit.
Discussion
==========
In this article we have computed the one loop anomalous dimension of an operator built from $O(N)$ $Z$s and 3 or 4 $Y$ “impurities”. What lessons can be learnt from these results, together with the results of [@Koch:2010gp], which dealt with the case of 2 impurities? Before we start the discussion, it is useful to recall the structure of the operators which participate in the case of two impurities $$\chi_A(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1)\,\yng(2,2,2,2,2,1,1)\,\yng(1,1)}(Z,Y)
\qquad
\chi_B(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1)\,\yng(2,2,2,1,1,1,1,1,1)\,\yng(1,1)}(Z,Y)$$ $$\chi_D(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1)\,\yng(2,2,2,2,1,1,1,1)\,\yng(2)}(Z,Y)
\qquad
\chi_E(b_0,b_1)=\chi_{\tiny \yng(2,2,2,2,2,1,1,1,1)\,\yng(2,2,2,2,1,1,1,1)\,\yng(1,1)}(Z,Y)$$ Firstly, the actual result for the one loop dilation operator (see Appendix A) is rather complicated. This is to be expected - it was obtained by summing a huge class of Feynman diagrams - much more than just the planar diagrams. On the other hand, the spectra of anomalous dimensions obtained are rather simple. To obtain a numerically tractable problem, we have been forced to keep the value of $a_{\rm max}$ finite (recall that this parameter measures the maximum value for the difference between the number of boxes in the long column and the number of boxes in the short column). In the large $N$ limit[^5] $a_{\rm max}=\zeta N$ goes to infinity. Our discussion assumes we are working in this $a_{\rm max}\to\infty$ limit.
We have restricted ourselves to a study of less than five impurities. This restriction is not imposed because our methods break down for five or more impurities, but only because the details of writing down the projection operators and evaluating the dilatation operator becomes increasingly complicated as the number of impurities is increased. In particular, there is no problem in principle with taking $O(N)$ impurities. The case of no impurities or one impurity are simple to handle analytically - all of these operators are annihilated by the one loop dilatation operator[@Koch:2010gp].
For the case of two impurities there are three times as many zero eigenvalue states as there positive eigenvalue states. There are ${a_{\rm max}\over 2}$ positive eigenvalue states with a constant energy level spacing of $8g_{YM}^2$. Thus, it is natural to associate an oscillator with a set of $\sim{a_{\rm max}\over 2}$ states. With this assumption, the dilatation operator acting on the two impurity operators gives the spectrum of three harmonic oscillators with a level spacing of zero and a single harmonic oscillator with a level spacing of $8g_{YM}^2$. Looking at the two impurity operators given above, we see that there are three operators with impurities in the antisymmetric representation (${\tiny \yng(1,1)}$) and one operator with the impurities in the symmetric representation (${\tiny \yng(2)}$).
For the case of three impurities, the dilatation operator gives the spectrum of four harmonic oscillators with a level spacing of zero and two harmonic oscillators with a level spacing of $8g_{YM}^2$; each oscillator again has $\sim{a_{\rm max}\over 2}$ states. Looking at the three impurity operators given in section 5.1, we see that there are four operators (A,B,D and F) with impurities in the antisymmetric representation (${\tiny \yng(1,1,1)}$) and two operators (C and E) with the impurities in the ${\tiny \yng(2,1)}$ representation.
For the case of four impurities, recall that we had an interesting degeneracy structure - the degeneracy alternates between three degenerate states and four degenerate states. This is naturally explained as three oscillators with a level spacing of $8g_{YM}^2$ and a fourth with a level spacing of $16g_{YM}^2$; with this interpretation each oscillator again has $\sim{a_{\rm max}\over 2}$ states. Thus, for the case of four impurities, the dilatation operator gives the spectrum of five harmonic oscillators with a level spacing of zero, three harmonic oscillator with a level spacing of $8g_{YM}^2$ and one with a spacing of $16g_{YM}^2$. Looking at the four impurity operators given in section 5.2, we see that there are five operators (A,B,D,G and I) with impurities in the antisymmetric representation (${\tiny \yng(1,1,1,1)}$), three operators (C,E and H) with the impurities in the ${\tiny \yng(2,1,1)}$ representation and one (F) with impurities in the ${\tiny \yng(2,2)}$ representation[^6].
It is rather easy to guess the result for a general number of impurities. If the number of impurities is even $=2n$ we expect to obtain a set of oscillators with frequency $\omega_i$ and degeneracy $d_i$ given by $$\omega_i =8ig_{YM}^2,\qquad d_i=2(n-i)+1,\qquad i=0,1,...,n\, .$$ If the number of impurities is odd $=2n+1$ we expect to obtain a set of oscillators with frequency $\omega_i$ and degeneracy $d_i$ given by $$\omega_i =8ig_{YM}^2,\qquad d_i=2(n-i+1),\qquad i=0,1,...,n\, .$$ This conjecture passes a simple counting test: $\sum_i d_i$ is equal to the number of restricted Schur polynomials that can be defined. Further, the degeneracies $d_i$ match the number of each type of oscillator that can be defined: $d_i$ is equal to the number of operators which have the impurities organized into a Young diagram with $i$ boxes in the short column.
A beautiful simple picture is emerging from the rather complicated formulas obtained for the dilatation operator: the dilatation operator is equivalent to a set of harmonic oscillators. For each type of operator there is a single oscillator and the frequency of the oscillator is determined by the representation which organizes the impurities. Since a set of harmonic oscillators is an integrable system, this system we have studied here is an example of an integrable dilatation operator, obtained by summing planar and non-planar diagrams.
${\cal N}=4$ super Yang-Mills theory has an $SU(4)$ ${\cal R}$-symmetry. A $U(2)$ subgroup of the ${\cal R}$-symmetry can be used to rotate the $Y$s and $Z$s into each other. Thus, although the operators we have studied do not treat the $Y$s and $Z$s symmetrically, a more complete study working with the complete set of two column restricted Schur polynomials would yield a description in which $Z$ and $Y$ appear on an equal footing. If we are to recover this symmetric description its clear that the oscillators which emerge for each particular impurity configuration must themselves have frequencies which are multiples of $8g_{YM}^2$ - exactly as we have found. Exploring this further we should be able to show that the spectrum is filling out $U(2)$ multiplets. By including more species of impurities we should also be able to see more of the expected global symmetry group.
What is the AdS/CFT dual interpretation of our results? The operators we have considered are dual to giant gravitons. A connection between the geometry of giant gravitons and harmonic oscillators was already uncovered in [@Biswas:2006tj; @Mandal:2006tk; @jurgis]. This work quantizes the moduli space of Mikhailov’s giant gravitons. Consequently one is capturing a huge space of states. It is this huge space of states that connects to harmonic oscillators. Our study has focused on a two giant system. Consequently, we know that the oscillators that we have captured are associated to this two giant system and excitations of it. We have thus arrived at a slightly more refined statement of how the harmonic oscillator enters. The critical reader might question whether our set of operators includes excitations corresponding (for example) to the two giant system plus a graviton. This would seem to be a small perturbation of the two giant system that is not an excitation of it - the graviton is an excitation of spacetime. We do not have such excitations among our states: these states correspond to operators with a small third column, which have decoupled at large $N$. In much the same way that by quantizing the possible excitation modes of a string one obtains a set of oscillators, its natural to think that our oscillators arise from the quantization of the possible excitation modes of a giant graviton.
We have described a limit (the first column of the Young diagram contains $O(\sqrt{N})$ boxes more than the second) in which the dilatation operator simplifies to a lattice realization of the second derivative. [*It is the Young diagram itself that is defining the lattice.*]{} After recalling that the number of boxes in each column sets the angular momentum and hence the radius of the corresponding threebrane, its clear that the radius of the giant graviton together with local physics in this radial direction has emerged. Notice that for BMN loops the number of lattice sites is $O(\sqrt{N})$; for the operators we have studied here the number of lattice sites is $O(N)$.
One can contemplate a number of tests for our proposal. With a thorough understanding of the vibrational modes of the giant graviton system, one could imagine realizing definite classical membrane geometries by considering coherent states of the oscillators. One might then compare the energy of these states with the volume of the membrane times the membrane tension. Although naively the field theory and gravity computations are never simultaneously valid, one might hope that for operators which are close to BPS, the perturbative result might agree with the strong coupling answer (see [@Berenstein:2003ah] for a relevant discussion).
In addition to questions we pointed out above, there are a number of clear directions for further study. Given the simplicity of our results, it should be possible to construct an analytic solution. This is under investigation[@bhw]. It would also be interesting to understand how our results are modified at higher loops. One could also consider the case of $n>2$ column restricted Schur polynomials and more species of impurities. A much more general question would be to ask when (and how) simple systems are expected to emerge from multimatrix models. For a single matrix model, it is well known that the planar limit is nicely captured by the dynamics of $N$ non-interacting non-relativistic fermions in an external potential. In this paper we have argued that the large $N$ limit of a class of operators dual to giant gravitons is captured by a collection of harmonic oscillators. Presumably every semiclassical object in spacetime (string, giant graviton, black hole, etc) is associated with the emergence of a simple system in the large $N$ limit of the corresponding class of operators in the field theory. Can we make this connection sharper and more useful?
[1.0cm]{}
[*Acknowledgements:*]{} We would like to thank Tom Brown, Warren Carlson, Norman Ives, Yusuke Kimura, Hai Lin, Dimitri Polyakov, Sanjaye Ramgoolam, Peter Roenne, Stephanie Smith and Michael Stephanou for pleasant discussions and/or helpful correspondence. This work is based upon research supported by the South African Research Chairs Initiative of the Department of Science and Technology and National Research Foundation. Any opinion, findings and conclusions or recommendations expressed in this material are those of the authors and therefore the NRF and DST do not accept any liability with regard thereto.
Dilatation Operator for Three or Four impurities
================================================
The dilatation operator for the case of two impurities has been given in [@Koch:2010gp]. In what follows $$DO=g_{YM}^2\hat{D}O\, .$$
Three Impurities
----------------
$$\hat{D}O_A(b_0,b_1) =\sqrt{\left(N-b_0-b_1-2\right)\left(N-b_0+1\right)}\left[
4\,b_1\,\sqrt{{\frac{b_1+4}{b_1+2}}}{1\over \left(b_1+2\right) \left( b_1+3 \right)}
O_B(b_0,b_1)\right.$$ $$-2\,\sqrt {{\frac {b_1+4}{b_1+2}}}\sqrt {2}{1\over \left( b_1+2 \right)}
O_C(b_0,b_1)
+8\,\sqrt{\left(b_1+4\right)\left(b_1+1\right)\over\left(b_1+2\right)\left(b_1+3\right)}
{\frac{1}{\left(b_1+3 \right)\left( b_1+2 \right)}} O_D(b_0-1,b_1+2)$$ $$\left. +2\,\sqrt{\left(b_1+4\right)\left(b_1+1\right)\over \left(b_1+3\right)\left(b_1+2\right)}{\frac {\sqrt {2}}
{ \left( b_1+3 \right)\left( b_1+2 \right)}}O_E(b_0-1,b_1+2)\right]
+(N-b_0-b_1-2)\left[{\frac {12}{ \left( b_1+2 \right) \left( b_1+3 \right) }}O_A(b_0,b_1)\right.$$ $$\left.
-4\,{\frac {\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) } \left( b_1+5 \right) }{ \left( b_1+3 \right) ^{2} \left( b_1+2 \right) }}O_B(b_0-1,b_1+2)
+2\,{\frac {\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) }\sqrt {2}}{ \left( b_1+3 \right) ^{2}}}O_C(b_0-1,b_1+2)\right]$$ $$\hat{D}O_B(b_0,b_1) =\sqrt{\left(N-b_0-b_1-1\right)\left(N-b_0\right)}\left[
-{4\over 3}\, \sqrt {{\frac { \left( b_1+2 \right) \left( b_1-1 \right) }{b_1\, \left( b_1+1 \right) }}}
{\left( b_1-2 \right) \left( b_1+3 \right) \over b_1\left( b_1+1 \right)}O_B(b_0+1,b_1-2)\right.$$ $$+{2\over 3}\,{b_1+3\over b1}\sqrt{{\frac{\left(b_1+2\right)\left(b_1-1\right)}{\left(b_1+1\right)b_1}}}\sqrt {2}O_C(b_0+1,b_1-2)
-{32\over 3}\,{{b_1}^{2}+2\,b_1-3 \over b_1(b_1+1)(b_1+2)^2} \sqrt {{\frac {b_1+2}{b_1}}}O_D(b_0,b_1)$$ $$\left.
-{2\sqrt{2}\over 3}\sqrt {b_1+2\over b_1}{\frac {(b_1+3) (3b_1-2)}{b_1(b_1+2)(b_1+1)}}O_E(b_0,b_1)
+8\,\sqrt {\left( b_1+3 \right)b_1\over \left( b_1+2 \right)\left( b_1+1 \right)}{\frac {1}{\left(b_1+1\right)\left(b_1+2\right)}}O_F(b_0-1,b_1+2)
\right]$$ $$+\sqrt{\left(N-b_0-b_1-2\right)\left(N-b_0+1\right)}\left[
{2\over 3}\sqrt{\left(b_1+4\right)\left(b_1+1\right)\over \left(b_1+2\right)\left(b_1+3\right)}
{\frac {\sqrt{2}b_1}{\left(b_1+3\right)}}O_C(b_0-1,b_1+2)\right.$$ $$-{4\over 3}\,\sqrt { \left( b_1+4 \right)\left( b_1+1 \right)\over\left( b_1+3 \right)\left( b_1+2 \right)}
{\frac { \left( b_1+5 \right) b_1}{ \left( b_1+3 \right)\left( b_1+2 \right)}}O_B(b_0-1,b_1+2)$$ $$\left.
+4\,\sqrt {b_1+4\over b_1+2}{\frac {b_1}{\left( b_1+3 \right)\left( b_1+2 \right)}}O_A(b_0,b_1)\right]
+\left( N-b_0-b_1-1 \right)\left[
-4\sqrt {{\frac {b_1-1}{b_1+1}}}{\left( b_1+3 \right)\over\left(b_1+1\right) b_1}O_A(b_0+1,b_1-2)\right.$$ $$+{4\over 3}\,{\frac{(b_1+3)(b_1^3+5b_1^2+8b_1-12)}{(b_1+1)b_1(b_1+2)^2}}O_B(b_0,b_1)
-{2\sqrt {2}\over 3}\,{\frac {(b_1^2+2b_1-4)(b_1+3)}{(b_1+1)(b_1+2)^2}}O_C(b_0,b_1)$$ $$\left.
-{8\over 3}\, \sqrt {{\frac {b_1+3}{b_1+1}}}
{\left( b_1+4 \right) b_1 \over \left( b_1+2 \right)^2 \left( b_1+1 \right)}O_D(b_0-1,b_1+2)
+{4\over 3}\, \sqrt {2}\sqrt {{\frac {b_1+3}{b_1+1}}}{b_1\over \left( b_1+2 \right)^2}
O_E(b_0-1,b_1+2)\right]$$ $$+\left( N-b_0+1 \right) \left[
{4\over 3}\,{\frac { \left( b_1+4 \right) {b_1}^{2} }{ \left( b_1+3 \right) \left( b_1+2 \right) ^{2}}}O_B(b_0,b_1)
+{8\over 3}\,{\frac{\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) }b_1\, \left( b_1+4 \right) }
{ \left( b_1+3 \right) ^{2} \left( b_1+2 \right) ^{2}}}O_D(b_0-1,b_1+2)\right.$$ $$\left.
-{2\over 3}\,{\frac{\sqrt {2} \left( b_1+4 \right) b_1}{ \left( b_1+2 \right) ^{2}}}
O_C(b_0,b_1)
+{2\over 3}\,{\frac{\sqrt {2}\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) }b_1\, \left( b_1+4 \right) }
{ \left( b_1+3 \right) ^{2} \left( b_1+2 \right) ^{2}}}O_E(b_0-1,b_1+2)\right]$$ $$\hat{D}O_C(b_0,b_1) =\sqrt {\left(N-b_0-b_1-1\right)\left(N-b_0\right)}\left[
{2\sqrt {2}\over 3}\, \sqrt {{\frac { \left( b_1+2 \right) \left( b_1-1 \right) }{ \left( b_1+1 \right) b_1}}}
{\left( b_1-2 \right)\over b_1+1 }O_B(b_0+1,b_1-2)\right.$$ $$-{2\over 3}\,\sqrt {{\frac { \left( b_1+2 \right) \left( b_1-1 \right) }{ \left( b_1+1 \right) b_1}}}O_C(b_0+1,b_1-2)
+{2\sqrt{2}\over 3}\sqrt{{\frac {b_1+2}{b_1}}}{(b_1-1)(3b_1+8) \over (b_1+1)(b_1+2)^2}O_D(b_0,b_1)$$ $$\left.
-{4\over 3}\sqrt{b_1+2\over b_1}{1\over (b_1+1)(b_1+2)}O_E(b_0,b_1)
+2\,{\frac{\sqrt{\left(b_1+2\right)\left(b_1+3\right)\left(b_1+1\right)b_1}\sqrt{2}}{\left( b_1+1 \right)^2\left(b_1+2\right)^2}}O_F(b_0-1,b_1+2)
\right]$$ $$+\sqrt {\left(N-b_0-b_1-2\right)\left(N-b_0+1\right)}\left[
-2\,{\frac {\sqrt { \left( b_1+4 \right) \left( b_1+2 \right) }\sqrt {2}}{ \left( b_1+2 \right) ^{2}}}O_A(b_0,b_1)\right.$$ $$\left.
+{2\sqrt{2}\over 3}\,\sqrt{\left(b_1+4\right)\left(b_1+1\right)\over \left(b_1+3\right)\left(b_1+2\right)}
{\frac{\left(b_1+5\right)}{\left(b_1+2\right)}}
O_B(b_0-1,b_1+2)
-{2\over 3}\,\sqrt{\left(b_1+4\right)\left(b_1+1\right)\over \left(b_1+2\right)\left(b_1+3\right)}
O_C(b_0-1,b_1+2)\right]$$ $$+\left( N-b_0-b_1-1 \right)
\left[2\,\sqrt {{\frac {b_1-1}{b_1+1}}}\sqrt {2} {1\over b_1+1}O_A(b_0+1,b_1-2)\right.$$ $$-{2\sqrt{2}\over 3}\,{\frac {(b_1^2+2b_1-4)(b_1+3)}{(b_1+1)(b_1+2)^2}}O_B(b_0,b_1)
+{2\over 3}{\frac{b_1(b_1^2+2b_1-1)}{(b_1+1)(b_1+2)^2}}O_C(b_0,b_1)$$ $$\left.
-{2\over 3}\, \sqrt {2}\sqrt {{\frac {b_1+3}{b_1+1}}}
{ \left( b_1+4 \right) b_1 \over \left( b_1+2 \right)^2 \left( b_1+1 \right)}O_D(b_0-1,b_1+2)
+{2\over 3}\, \sqrt {{\frac {b_1+3}{b_1+1}}}{b_1\over \left( b_1+2 \right)^2}
O_E(b_0-1,b_1+2)\right]$$ $$+\left( N-b_0+1 \right)\left[
-{2\over 3}\,{\frac { \sqrt {2} \left( b_1+4 \right) b_1}{ \left( b_1+2 \right) ^{2}}}O_B(b_0,b_1)
+{2\over 3}\,{\frac { \left( b_1+4 \right) \left( b_1+3 \right) }{ \left( b_1+2 \right) ^{2}}}O_C(b_0,b_1)\right.$$ $$\left.
-{4\over 3}\,{\frac {\sqrt {2}\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) } \left( b_1+4 \right) }{ \left( b_1+3 \right) \left( b_1+2 \right) ^{2}}}O_D(b_0-1,b_1+2)
-{2\over 3}\,{\frac {\sqrt { \left( b_1+1 \right) \left( b_1+3 \right) } \left( b_1+4 \right) }
{ \left( b_1+3 \right) \left( b_1+2 \right) ^{2}}}O_E(b_0-1,b_1+2)\right]$$ $$\hat{D}O_D(b_0,b_1) =\sqrt{\left(N-b_0-b_1\right)\left(N-b_0-1\right)}\left[
-{4\over 3}\sqrt {\left(b_1+1\right)\left( b_1-2 \right)\over b_1 \left( b_1-1 \right)}
\,{\frac { \left( b_1-3 \right) \left( b_1+2 \right) }{{b_1} \left( b_1-1 \right)}}O_D(b_0+1,b_1-2)\right.$$ $$\left.
+{2\over 3}\,{\frac { \left( b_1+2 \right) \sqrt {b_1\, \left( b_1-1 \right) \left( b_1+1 \right) \left( b_1-2 \right) }\sqrt {2}}
{b_1\, \left( b_1-1 \right) ^{2}}}O_E(b_0+1,b_1-2)
-4\,{\frac { \left( b_1+2 \right) \sqrt {b_1\, \left( b_1-2 \right)}}{{b_1}^{2}
\left( b_1-1 \right) }}O_F(b_0,b_1)\right]$$ $$+\sqrt { \left( N-b_0-b_1-1 \right) \left( N-b_0 \right) }
\left[{\frac {8}{b_1\left( b_1+1 \right)}}\sqrt { \left( b_1+2 \right) \left( b_1-1 \right)\over \left( b_1+1 \right) b_1}O_A(b_0+1,b_1-2)
\right.$$ $$+{2\sqrt{2}\over 3}\,{\frac {\sqrt{b_1( b_1+2)}(b_1-1)(3b_1+8)}{(b_1+1)(b_1+2)^2 b_1}}O_C(b_0,b_1)
-{32\over 3}{(b_1^2+2b_1-3)\sqrt{b_1(b_1+2)}\over (b_1+2)^2 b_1^2 (b_1+1)}O_B(b_0,b_1)$$ $$-{4\over 3}\, \sqrt {{\frac {b_1\, \left( b_1+3 \right) }{ \left( b_1+2 \right) \left( b_1+1 \right) }}}
{\left( b_1-1 \right) \left( b_1+4 \right) \over\left(b_1+2\right)\left(b_1+1\right)}O_D(b_0-1,b_1+2)$$ $$\left.
+{2\over 3}\, \sqrt {{\frac {b_1\, \left( b_1+3 \right) }{ \left( b_1+2 \right) \left( b_1+1 \right) }}}\sqrt {2}
{\left( b_1-1 \right)\over b_1+2 }O_E(b_0-1,b_1+2)\right]
+ \left( N-b_0 \right) \left[
{4\over 3}{\frac {(b_1-1)(b_1^3+b_1^2+16)}{(b_1+1)b_1^2(b_1+2)}}
O_D(b_0,b_1)\right.$$ $$+{8\over 3}\, \sqrt {{\frac {b_1-1}{b_1+1}}}{ {b_1}^{2}-4 \over b_1^2 \left( b_1+1 \right)}O_B(b_0+1,b_1-2)
-{4\over 3}\, \sqrt {2}\sqrt {{\frac {b_1-1}{b_1+1}}} {b_1+2 \over b_1^2}O_C(b_0+1,b_1-2)$$ $$\left.
-{2\sqrt {2}\over 3}\,{\frac {(b_1^2+2b_1-4)(b_1-1)}{(b_1+1)b_1^2}}O_E(b_0,b_1)
+4\, \left( b_1-1 \right) \sqrt {{\frac {b_1+3}{b_1+1}}}{1\over \left( b_1+2 \right)\left( b_1+1 \right)}
O_F(b_0-1,b_1+2)\right]$$ $$+\left( N-b_0-b_1 \right) \left[
{4\over 3}\,{\frac { \left( b_1-2 \right) \left( b_1+2 \right) ^{2} }{ \left( b_1-1 \right) {b_1}^{2}}}O_D(b_0,b_1)
-{2\sqrt {2}\over 3}\,{\frac { \left( {b_1}^{2}-4 \right) }{{b_1}^{2}}}O_E(b_0,b_1)\right.$$ $$\left.
-{8\over 3}\,\sqrt{\left(b_1+1\right)\over \left(b_1-1\right)}
{\frac {\left(b_1^2-4\right)}{{b_1}^{2} \left( b_1-1 \right)}}O_B(b_0+1,b_1-2)
-{2\sqrt {2}\over 3}\,\sqrt{\left(b_1+1\right)\over\left(b_1-1\right)}{\frac{\left(b_1^2-4\right)}{{b_1}^{2} \left( b_1-1 \right)}}
O_C(b_0+1,b_1-2)\right]$$ $$\hat{D}O_E(b_0,b_1) =\sqrt { \left( N-b_0-b_1 \right) \left( N-b_0-1 \right) }\left[
{2\sqrt{2}\over 3}\sqrt {\left( b_1+1 \right) \left( b_1-2 \right) \over b_1\, \left( b_1-1 \right)}
\,{\frac { \left( b_1-3 \right)}
{b_1}}O_D(b_0+1,b_1-2)\right.$$ $$\left.
-{2\over 3}\,{\frac {\sqrt {b_1\, \left( b_1-1 \right) \left( b_1+1 \right) \left( b_1-2 \right)}}{b_1\, \left( b_1-1 \right) }}
O_E(b_0+1,b_1-2)
+2\,{\frac {\sqrt {2}\sqrt {b_1\, \left( b_1-2 \right) }}{{b_1}^{2}}}
O_F(b_0,b_1)\right]$$ $$+\sqrt { \left( N-b_0-b_1-1 \right) \left( N-b_0 \right) }\left[
2\,{\frac {\sqrt {2}\sqrt { \left( b_1+2 \right) \left( b_1-1 \right)
\left( b_1+1 \right) b_1}}{{b_1}^{2} \left( b_1+1 \right) ^{2}}}
O_A(b_0+1,b_1-2)\right.$$ $$-{4\over 3}{\sqrt{b_1(b_1+2)}\over (b_1+2)(b_1+1)b_1}O_C(b_0,b_1)
-{2\sqrt {2}\over 3}\sqrt {b_1(b_1+2)}{(b_1+3)(3b_1-2) \over (b_1+1)b_1^2(b_1+2)}O_B(b_0,b_1)$$ $$+{2\over 3}\sqrt{{\frac {b_1(b_1+3)}{(b_1+2)(b_1+1)}}}\sqrt {2}
{b_1+4\over b_1+1}O_D(b_0-1,b_1+2)
\left.
-{2\over 3}\,\sqrt {{\frac {b_1\, \left( b_1+3 \right) }{ \left( b_1+1 \right) \left( b_1+2 \right) }}}
O_E(b_0-1,b_1+2)\right]$$ $$+\left( N-b_0 \right)\left[
-{2\sqrt {2}\over 3}{\frac {(b_1^2+2b_1-4)(b_1-1)}{(b_1+1)b_1^2}}O_D(b_0,b_1)
\right.$$ $$+{2\over 3}\,{\frac {(b_1+2)(b_1^2+2b_1-1)}{(b_1+1)b_1^2}}O_E(b_0,b_1)
-2\,\sqrt {{\frac {b_1+3}{b_1+1}}}\sqrt {2} {1\over b_1+1}O_F(b_0-1,b_1+2)$$ $$\left. +{2\over 3}\sqrt {2}\sqrt {{\frac {b_1-1}{b_1+1}}} {b_1^2-4 \over (b_1+1)b_1^2}O_B(b_0+1,b_1-2)
-{2\over 3}\sqrt{{\frac {b_1-1}{b_1+1}}} {b_1+2 \over b_1^2}O_C(b_0+1,b_1-2)\right]$$ $$+\left( N-b_0-b_1 \right)\left[{2\over 3}\,{\frac { \left( {b_1}^{2}-3\,b_1+2 \right) }{{b_1}^{2}}}O_E(b_0,b_1)
-{2\sqrt {2}\over 3}\,{\frac{\left( {b_1}^{2}-4 \right) }{{b_1}^{2}}}O_D(b_0,b_1)\right.$$ $$\left.
+{4\sqrt {2}\over 3}\,{\frac { \left( b_1-2 \right) \sqrt { \left( b_1+1 \right) \left( b_1-1 \right) }}{{b_1}^{2}
\left( b_1-1 \right) }}O_B(b_0+1,b_1-2)
+{2\over 3}\,{\frac {\left(b_1-2\right)\sqrt{\left(b_1+1\right)\left(b_1-1\right)}}{{b_1}^{2}\left(b_1-1\right)}}
O_C(b_0+1,b_1-2)\right]$$ $$\hat{D}O_F(b_0,b_1) =\sqrt { \left( N-b_0-b_1 \right) \left( N-b_0-1 \right) }\left[
8\,{\frac {\sqrt {b_1\, \left( b_1-1 \right) \left( b_1+1 \right) \left( b_1-2 \right) }
}{{b_1}^{2} \left( b_1-1 \right) ^{2}}}O_B(b_0+1,b_1-2)\right.$$ $$+2\,{\frac {\sqrt {2}
\sqrt {b_1\, \left( b_1-1 \right) \left( b_1+1 \right) \left( b_1-2 \right) }}{{b_1}^{2} \left( b_1-1 \right) ^{2}}}O_C(b_0+1,b_1-2)
-4\, \left( b_1+2 \right) \sqrt {{\frac {b_1-2}{b_1}}}
{1\over b_1 \left( b_1-1 \right)}O_D(b_0,b_1)$$ $$\left.
+2\,\sqrt {{\frac {b_1-2}{b_1}}}\sqrt {2}{1\over b_1} O_E(b_0,b_1)
\right]
+4\,{\frac { \left( b_1-3 \right) \sqrt { \left( b_1+1 \right) \left( b_1-1 \right) } \left( N-b_0-1 \right) }{b_1\, \left( b_1-1 \right) ^{2}}}
O_D(b_0+1,b_1-2)$$ $$-2\,{\frac {\sqrt { \left( b_1+1 \right) \left( b_1-1 \right) }\sqrt {2} \left( N-b_0-1 \right) }{ \left( b_1-1 \right) ^{2}}}
O_E(b_0+1,b_1-2)
+12\,{\frac {N-b_0-1}{b_1\, \left( b_1-1 \right) }}O_F(b_0,b_1)$$
Four Impurities
---------------
$$\hat{D}O_A(b_0,b_1)=\sqrt{ (N-b_0-b_1-3)(N-b_0+1)}\left[
6\sqrt{b_1+5\over b_1+3}{b_1\over (b_1+4)(b_1+2)}
O_B(b_0,b_1)\right.$$ $$-2\sqrt{3}\sqrt{b_1+5\over b_1+3} {1\over b_1+2}O_C(b_0,b_1)
+12{\sqrt{(b_1+1)(b_1+5)}\over (b_1+3)(b_1+2)(b_1+4)} O_D(b_0-1,b_1+2)$$ $$\left. + 4\sqrt {3}{\sqrt{(b_1+5)(b_1+1)}\over (b_1+3)(b_1+4)(b_1+2)}
O_E(b_0-1,b_1+2)\right]
+(N-b_0-b_1-3)\left[
{\frac {24}{(b_1+2)(b_1+4)}}O_A(b_0,b_1)\right.$$ $$\left.
-6\sqrt{b_1+1\over b_1+3}{\frac{b_1+6}{(b_1+2)(b_1+4)}}O_B(b_0-1,b_1+2)
+ 2\sqrt {3}\sqrt{b_1+1\over b_1+3}{\frac {1}{b_1+4}}O_C(b_0-1,b_1+2)\right]$$ $$\hat{D}O_B(b_0,b_1)=\sqrt { \left( N-b_0-b_1-2 \right)\left( N-b_0 \right) }\left[
{\sqrt{3}\over 2}\sqrt{\left(b_1+3\right)(b_1-1)}{b_1+4\over b_1(b_1+1)}O_C(b_0+1,b_1-2)
\right.$$ $$-{3\over 2} {\sqrt{(b_1+3)(b_1-1)}\over (b_1+1)(b_1+2)}{(b_1+4)(b_1-2)\over b_1}O_B(b_0+1,b_1-2)
+3 {(b_1+4)(b_1-1)(b_1-6)\over b_1(b_1+3)(b_1+2)^2}\sqrt{b_1+3\over b_1+1}O_D(b_0,b_1)$$ $$-\sqrt {3} {(b_1+4)(3b_1-2)\over b_1(b_1+2)^2} \sqrt{b_1+3\over b_1+1}O_E(b_0,b_1)
$$ . + [2b\_1(b\_1+4)(b\_1+1)(b\_1+3)(b\_1+2)\^2]{}(9 O\_G(b\_0-1,b\_1+2)+O\_H(b\_0-1,b\_1+2))\] $$$$ +$$$$ +(N-b\_0-b\_1-2)+(N-b\_0+1)$${\vskip 1cm}
$$ O\_C(b\_0,b\_1)=$$$$ ++(N-b\_0-b\_1-2)$$$$ +(N-b\_0+1)$${\vskip 1cm}
$$ O\_D(b\_0,b\_1)= $$$$ +3 O\_G(b\_0,b\_1) - O\_H(b\_0,b\_1) $$$$ + . 12O\_I(b\_0-1,b\_1+2)\] +$$$$ +(N-b\_0-b\_1-1)+(N-b\_0)+(N-b\_0-b\_1-1)+(N-b\_0)$${\vskip 1cm}
$$ O\_E(b\_0,b\_1)=$$$$ + +(N-b\_0-b\_1-1)$$$$ +(N-b\_0)+(N-b\_0-b\_1-1)+(N-b\_0)$${\vskip 1cm}
$$ O\_F(b\_0,b\_1)=++(N-b\_0-b\_1-1)$$$$ $$$$ +(N-b\_0)$${\vskip 1cm}
$$ O\_G(b\_0,b\_1)=$$$$ ++(N-b\_0-b\_1)+(N-b\_0-1)$${\vskip 1cm}
$$ O\_H(b\_0,b\_1)=$$$$ +$$$$ +(N-b\_0-b\_1)+(N-b\_0-1)$$$$ $${\vskip 1cm}
$$ O\_I [(b\_0,b\_1)]{}=+(N-b\_0-2)$$
A Discussion on Intertwiners
============================
When $S^n$ acts on $V^{\otimes n}$ $n>1$ it furnishes a reducible representation. Imagine that this includes the irreducible representations $R$ and $T$. Representing the action of $\sigma$ as a matrix $\Gamma(\sigma )$, in a suitable basis we can write $$\Gamma(\sigma)=\left[
\matrix{\Gamma_R(\sigma) &0 &\cdots\cr 0 &\Gamma_S(\sigma) &\cdots\cr \cdots &\cdots &\cdots}\right]\, .$$ If we restrict ourselves to an $S_{n-1}$ subgroup of $S_n$, then in general, both $R$ and $S$ will subduce a number of representations. Assume for the sake of this discussion that $R$ subduces $R_1'$ and $R_2'$ and that $S$ subduces $S_1'$ and $S_2'$. Then, for $\sigma\in S_{n-1}$ we have $$\Gamma(\sigma)=
\left[
\matrix{
\Gamma_{R_1'}(\sigma) &0 &0 &0 &\cdots\cr
0 &\Gamma_{R_2'}(\sigma) &0 &0 &\cdots\cr
0 &0 &\Gamma_{S_1'}(\sigma) &0 &\cdots\cr
0 &0 &0 &\Gamma_{S_2'}(\sigma) &\cdots\cr
\cdots &\cdots &\cdots &\cdots &\cdots}\right]\, .$$ Imagine that $S_1'=R_1'$, that is, one of the irreducible representations subduced by $R$ is also subduced by $S$. Then, a simple application of the fundamental orthogonality relation gives $$\sum_{\sigma\in S_{n-1}}
\left[
\matrix{
\Gamma_{R_1'}(\sigma) &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
\cdots &\cdots &\cdots &\cdots &\cdots}\right]_{ij}
\left[
\matrix{
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
0 &0 &\Gamma_{S_1'}(\sigma) &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
\cdots &\cdots &\cdots &\cdots &\cdots}\right]_{ab}$$ $$={(n-1)!\over d_{R_1'}}\delta_{R_1'S_1'}
\left[
\matrix{
0 &0 &{\bf 1} &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
\cdots &\cdots &\cdots &\cdots &\cdots}\right]_{ib}
\left[
\matrix{
0 &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
{\bf 1} &0 &0 &0 &\cdots\cr
0 &0 &0 &0 &\cdots\cr
\cdots &\cdots &\cdots &\cdots &\cdots}\right]_{aj}$$ $$\equiv {(n-1)!\over d_{R_1'}}\delta_{R_1'S_1'} (I_{R_1'S_1'})_{ib}(I_{S_1'R_1'})_{aj}$$ where the form of the intertwiners has been spelled out.
[30]{} -2pt
S. Corley, A. Jevicki and S. Ramgoolam, “Exact correlators of giant gravitons from dual N = 4 SYM theory,” Adv. Theor. Math. Phys. [**5**]{}, 809 (2002) \[arXiv:hep-th/0111222\]. V. Balasubramanian, D. Berenstein, B. Feng and M. x. Huang, “D-branes in Yang-Mills theory and emergent gauge symmetry,” JHEP [**0503**]{}, 006 (2005) \[arXiv:hep-th/0411205\]. R. de Mello Koch, J. Smolic and M. Smolic, “Giant Gravitons - with Strings Attached (I),” JHEP [**0706**]{}, 074 (2007), arXiv:hep-th/0701066. R. de Mello Koch, J. Smolic and M. Smolic, “Giant Gravitons - with Strings Attached (II),” JHEP [**0709**]{} 049 (2007), arXiv:hep-th/0701067. Y. Kimura and S. Ramgoolam, “Branes, Anti-Branes and Brauer Algebras in Gauge-Gravity duality,” arXiv:0709.2158 \[hep-th\]. D. Bekker, R. de Mello Koch and M. Stephanou, “Giant Gravitons - with Strings Attached (III),” arXiv:0710.5372 \[hep-th\]. T. W. Brown, P. J. Heslop and S. Ramgoolam, “Diagonal multi-matrix correlators and BPS operators in N=4 SYM,” arXiv:0711.0176 \[hep-th\]. R. Bhattacharyya, S. Collins and R. d. M. Koch, “Exact Multi-Matrix Correlators,” JHEP [**0803**]{}, 044 (2008) \[arXiv:0801.2061 \[hep-th\]\]. T. W. Brown, P. J. Heslop and S. Ramgoolam, “Diagonal free field matrix correlators, global symmetries and giant gravitons,” arXiv:0806.1911 \[hep-th\]. Y. Kimura and S. Ramgoolam, “Enhanced symmetries of gauge theory and resolving the spectrum of local operators,” Phys. Rev. D [**78**]{}, 126003 (2008) \[arXiv:0807.3696 \[hep-th\]\]. Y. Kimura, “Non-holomorphic multi-matrix gauge invariant operators based on Brauer algebra,” arXiv:0910.2170 \[hep-th\]. S. Ramgoolam, “Schur-Weyl duality as an instrument of Gauge-String duality,” AIP Conf. Proc. [**1031**]{}, 255 (2008) \[arXiv:0804.2764 \[hep-th\]\]. V. Balasubramanian, M. Berkooz, A. Naqvi and M. J. Strassler, “Giant gravitons in conformal field theory,” JHEP [**0204**]{}, 034 (2002) \[arXiv:hep-th/0107119\]. G. ’t Hooft, “A Planar Diagram Theory for Strong Interactions,” Nucl. Phys. B [**72**]{}, 461 (1974);\
G. ’t Hooft, “A Two-Dimensional Model For Mesons,” Nucl. Phys. B [**75**]{}, 461 (1974). R. de Mello Koch, “Geometries from Young Diagrams,” JHEP [**0811**]{}, 061 (2008) \[arXiv:0806.0685 \[hep-th\]\]. R. de Mello Koch, N. Ives and M. Stephanou, “Correlators in Nontrivial Backgrounds,” Phys. Rev. D [**79**]{}, 026004 (2009) \[arXiv:0810.4041 \[hep-th\]\]. R. de Mello Koch, T. K. Dey, N. Ives and M. Stephanou, “Correlators Of Operators with a Large R-charge,” arXiv:0905.2273 \[hep-th\]. H. Lin, O. Lunin and J. M. Maldacena, “Bubbling AdS space and 1/2 BPS geometries,” JHEP [**0410**]{}, 025 (2004) \[arXiv:hep-th/0409174\]. V. Balasubramanian, J. de Boer, V. Jejjala and J. Simon, “The library of Babel: On the origin of gravitational thermodynamics,” JHEP [**0512**]{}, 006 (2005) \[arXiv:hep-th/0508023\],\
V. Balasubramanian, V. Jejjala and J. Simon, “The library of Babel,” Int. J. Mod. Phys. D [**14**]{}, 2181 (2005) \[arXiv:hep-th/0505123\]. V. Balasubramanian, J. de Boer, V. Jejjala and J. Simon, “Entropy of near-extremal black holes in AdS$_5$,” JHEP [**0805**]{}, 067 (2008) \[arXiv:0707.3601 \[hep-th\]\],\
R. Fareghbal, C. N. Gowdigere, A. E. Mosaffa and M. M. Sheikh-Jabbari, “Nearing Extremal Intersecting Giants and New Decoupled Sectors in N = 4 SYM,” JHEP [**0808**]{}, 070 (2008) \[arXiv:0801.4457 \[hep-th\]\]. J. M. Maldacena, “The large N limit of superconformal field theories and supergravity,” Adv. Theor. Math. Phys. [**2**]{}, 231 (1998) \[Int. J. Theor. Phys. [**38**]{}, 1113 (1999)\] \[arXiv:hep-th/9711200\];\
S. S. Gubser, I. R. Klebanov and A. M. Polyakov, “Gauge theory correlators from non-critical string theory,” Phys. Lett. B [**428**]{}, 105 (1998) \[arXiv:hep-th/9802109\];\
E. Witten, “Anti-de Sitter space and holography,” Adv. Theor. Math. Phys. [**2**]{}, 253 (1998) \[arXiv:hep-th/9802150\]. K. Skenderis and M. Taylor, “Anatomy of bubbling solutions,” JHEP [**0709**]{}, 019 (2007) \[arXiv:0706.0216 \[hep-th\]\]. D. E. Berenstein, J. M. Maldacena and H. S. Nastase, “Strings in flat space and pp waves from N = 4 super Yang Mills,” JHEP [**0204**]{}, 013 (2002) \[arXiv:hep-th/0202021\]. H. Y. Chen, D. H. Correa and G. A. Silva, “Geometry and topology of bubble solutions from gauge theory,” Phys. Rev. D [**76**]{}, 026003 (2007) \[arXiv:hep-th/0703068\]. R. de Mello Koch, T. K. Dey, N. Ives and M. Stephanou, “Hints of Integrability Beyond the Planar Limit,” JHEP [**1001**]{}, 014 (2010) \[arXiv:0911.0967 \[hep-th\]\]. H. Lin, A. Morisse and J. P. Shock, “Strings on Bubbling Geometries,” JHEP [**1006**]{}, 055 (2010) \[arXiv:1003.4190 \[hep-th\]\]. H. Lin, “Studies on 1/4 BPS and 1/8 BPS geometries,” arXiv:1008.5307 \[hep-th\]. J. McGreevy, L. Susskind and N. Toumbas, “Invasion of the giant gravitons from anti-de Sitter space,” JHEP [**0006**]{}, 008 (2000) \[arXiv:hep-th/0003075\];\
M. T. Grisaru, R. C. Myers and O. Tafjord, “SUSY and Goliath,” JHEP [**0008**]{}, 040 (2000) \[arXiv:hep-th/0008015\];\
A. Hashimoto, S. Hirano and N. Itzhaki, “Large branes in AdS and their field theory dual,” JHEP [**0008**]{}, 051 (2000) \[arXiv:hep-th/0008016\]. D. Berenstein, “A toy model for the AdS/CFT correspondence,” JHEP [**0407**]{}, 018 (2004) \[arXiv:hep-th/0403110\]. R. d. M. Koch, G. Mashile and N. Park, “Emergent Threebrane Lattices,” Phys. Rev. D [**81**]{}, 106009 (2010) \[arXiv:1004.1108 \[hep-th\]\]. A. Mikhailov, “Giant gravitons from holomorphic surfaces,” JHEP [**0011**]{}, 027 (2000) \[arXiv:hep-th/0010206\]. C. E. Beasley, “BPS branes from baryons,” JHEP [**0211**]{}, 015 (2002) \[arXiv:hep-th/0207125\]. J. Kinney, J. M. Maldacena, S. Minwalla and S. Raju, “An index for 4 dimensional super conformal theories,” Commun. Math. Phys. [**275**]{}, 209 (2007) \[arXiv:hep-th/0510251\]. I. Biswas, D. Gaiotto, S. Lahiri and S. Minwalla, “Supersymmetric states of N = 4 Yang-Mills from giant gravitons,” JHEP [**0712**]{}, 006 (2007) \[arXiv:hep-th/0606087\]. G. Mandal and N. V. Suryanarayana, “Counting 1/8-BPS dual-giants,” JHEP [**0703**]{}, 031 (2007) \[arXiv:hep-th/0606088\]. J. Pasukonis and S. Ramgoolam, “From counting to construction of BPS states in N=4 SYM,” arXiv:1010.1683 \[hep-th\]. V. Balasubramanian, M. x. Huang, T. S. Levi and A. Naqvi, “Open strings from N = 4 super Yang-Mills,” JHEP [**0208**]{}, 037 (2002) \[arXiv:hep-th/0204196\],\
O. Aharony, Y.E. Antebi, M. Berkooz and R. Fishman, “Holey sheets: Pfaffians and subdeterminants as D-brane operators in large $N$ gauge theories,” JHEP [**0212**]{}, 096 (2002) \[arXiv:hep-th/0211152\]. D. Sadri and M. M. Sheikh-Jabbari, “Giant hedge-hogs: Spikes on giant gravitons,” Nucl. Phys. B [**687**]{}, 161 (2004) \[arXiv:hep-th/0312155\]. D. Berenstein, “Shape and holography: Studies of dual operators to giant gravitons,” Nucl. Phys. B [**675**]{}, 179 (2003) \[arXiv:hep-th/0306090\]. D. Berenstein, D. H. Correa and S. E. Vazquez, “A study of open strings ending on giant gravitons, spin chains and integrability,” \[arXiv:hep-th/0604123\],\
D. Berenstein and S. E. Vazquez, “Integrable open spin chains from giant gravitons,” JHEP [**0506**]{}, 059 (2005) \[arXiv:hep-th/0501078\],\
D. Berenstein, D. H. Correa and S. E. Vazquez, “Quantizing open spin chains with variable length: An example from giant gravitons,” Phys. Rev. Lett. [**95**]{}, 191601 (2005) \[arXiv:hep-th/0502172\],\
D. H. Correa and G. A. Silva, “Dilatation operator and the Super Yang-Mills duals of open strings on AdS Giant Gravitons,” \[arXiv:hep-th/0608128\].
Y. Kimura, “Quarter BPS classified by Brauer algebra,” JHEP [**1005**]{}, 103 (2010) \[arXiv:1002.2424 \[hep-th\]\]. E. D’Hoker and A. V. Ryzhov, “Three-point functions of quarter BPS operators in N = 4 SYM,” JHEP [**0202**]{}, 047 (2002) \[arXiv:hep-th/0109065\],\
E. D’Hoker, P. Heslop, P. Howe and A. V. Ryzhov, “Systematics of quarter BPS operators in N = 4 SYM,” JHEP [**0304**]{}, 038 (2003) \[arXiv:hep-th/0301104\],\
P. J. Heslop and P. S. Howe, “OPEs and 3-point correlators of protected operators in N = 4 SYM,” Nucl. Phys. B [**626**]{}, 265 (2002) \[arXiv:hep-th/0107212\]. T. W. Brown, “Permutations and the Loop,” arXiv:0801.2094 \[hep-th\]. T. W. Brown, “Cut-and-join operators and N=4 super Yang-Mills,” arXiv:1002.2099 \[hep-th\],\
M. x. Huang, “Higher Genus BMN Correlators: Factorization and Recursion Relations,” arXiv:1009.5447 \[hep-th\]. N. Beisert, C. Kristjansen and M. Staudacher, “The dilatation operator of N = 4 super Yang-Mills theory,” Nucl. Phys. B [**664**]{}, 131 (2003) \[arXiv:hep-th/0303060\]. R. de Mello Koch and R. Gwyn, “Giant graviton correlators from dual SU(N) super Yang-Mills theory,” JHEP [**0411**]{}, 081 (2004) \[arXiv:hep-th/0410236\]. R. Bhattacharyya, R. de Mello Koch and M. Stephanou, “Exact Multi-Restricted Schur Polynomial Correlators,” arXiv:0805.3025 \[hep-th\]. S. R. Das, A. Jevicki and S. D. Mathur, “Vibration modes of giant gravitons,” Phys. Rev. D [**63**]{}, 024013 (2001) \[arXiv:hep-th/0009019\]. Andrei Okounkov and Anatoly Vershik, “A new approach to representation theory of symmetric groups,” Selecta Mathematica, [**2**]{}, 581-605,\
Andrei Okounkov and Anatoly Vershik, “A New Approach to the Representation Theory of the Symmetric Groups II,” Journal of Mathematical Sciences [**131**]{}, 5471-5494.
R. C. Myers, “Dielectric-branes,” JHEP [**9912**]{}, 022 (1999) \[arXiv:hep-th/9910053\].
W. Carlson, R. de Mello Koch and H. Lin, work in progress.
[^1]: In general, because $(r,s)$ can be subduced more than once, we should include a multiplicity index. We will not write this index explicitly in this article.
[^2]: Our index conventions are $(YZ)^i_k = Y^i_j Z^j_k$.
[^3]: Recall that the weight of a box in row $i$ and column $j$ is $N - i + j$.
[^4]: The giant graviton threebrane wraps an S$^3$ of a given radius. It is the radius of this S$^3$ that we call the “radius of the threebrane”.
[^5]: $\zeta$ was defined in section 3.
[^6]: Note that by looking at the representation that organizes the impurities we have been able to read off the frequencies of the harmonic oscillators appearing. This is where it ends; in particular we are [*not*]{} claiming that $O_F$ are the operators corresponding to the frequency $16g_{YM}^2$ operators! Operators with a good scaling dimension are a complicated linear combination of the various possible $O$s.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The Tunka Radio Extension (Tunka-Rex) is an antenna array spread over an area of about 1 km$^2$. The array is placed at the Tunka Advanced Instrument for cosmic rays and Gamma Astronomy (TAIGA) and detects the radio emission of air showers in the band of 30 to 80 MHz. During the last years it was shown that a sparse array such as Tunka-Rex is capable of reconstructing the parameters of the primary particle as accurate as the modern instruments. Based on these results we continue developing our data analysis. Our next goal is the reconstruction of cosmic-ray energy spectrum observed only by a radio instrument. Taking a step towards it, we develop a model of aperture of our instrument and test it against hybrid TAIGA observations and Monte-Carlo simulations. In the present work we give an overview of the current status and results for the last five years of operation of Tunka-Rex and discuss prospects of the cosmic-ray energy estimation with sparse radio arrays.'
address:
- '$^1$ Institut für Kernphysik, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany'
- '$^2$ Institute of Applied Physics ISU, Irkutsk, 664020, Russia'
- '$^3$ Institut für Prozessdatenverarbeitung und Elektronik, Karlsruhe Institute of Technology (KIT)'
- '$^4$ Skobeltsyn Institut of Nuclear Physics MSU, Moscow, 119991, Russia'
- '$^5$ Department of Physics and Astronomy, University of Delawere, Newark, DE, USA'
author:
- 'V. Lenok$^1$, P.A. Bezyazeekov$^2$, N.M. Budnev$^2$, D. Chernykh$^2$, O. Fedorov$^2$, O.A. Gress$^2$, A. Haungs$^1$, R. Hiller$^1$[^1], T. Huege$^1$[^2], Y. Kazarina$^2$, M. Kleifges$^3$, D. Kostunin$^1$, E.E. Korosteleva$^4$, L.A. Kuzmichev$^4$, N. Lubsandorzhiev$^4$, T. Marshalkina$^2$, R. Monkhoev$^2$, E. Osipova$^4$, A. Pakhorukov$^2$, L. Pankov$^2$, V.V. Prosin$^4$, F.G. Schröder$^1,^5$, D. Shipilov$^2$ and A. Zagorodnikov$^2$ (Tunka-Rex Collaboration)'
title: Current Status and New Challenges of The Tunka Radio Extension
---
Introduction
============
Digital radio antenna arrays are a rapidly developing instrument for observation and study of ultra-high energy messengers in PeV-EeV range [@schroeder2017; @huege2016]. Radio detection has evolved from understanding signal formation and early hardware development to the stage of real physical measurements.
![\[cluster\] [*Left*]{}: one of the clusters of Tunka-Rex. Three generations of the antennas are marked in the picture along with the Tunka-Grande and Tunka-133 detectors. The black lines schematically represent the cabling. [*Right*]{}: map of the present footprint of the detectors of TAIGA. ](cluster+map.pdf){width="1.\textwidth"}
Tunka-Rex is a radio antenna array designed for detecting cosmic rays. It is located at the site of the TAIGA detector [@taiga] at the Tunka valley in Russia. The Tunka-Rex detector has been operating since 2012. It has been extended several times and currently consists of 57 antennas in the dense core of the detector (about 1 km$^2$ area) and 6 antennas in the satellite clusters used as a high-energy extension. In total the Tunka-Rex radio array covers about 3 km$^2$. As antenna type we use SALLA [@salla] operating in the band of 30–80 MHz. The present configuration of Tunka-Rex detector is shown in the Fig. \[cluster\].
Highlights of Recent Results
============================
Some important updates of the analysis have been done in Tunka-Rex. Recently we developed a method of template analysis of the data. Essentials of this method are the following:
- Pre-reconstruction using the standard procedure of Tunka-Rex [@jcap]. This stage provides the shower incoming direction, the core position, and the energy estimation.
- Creating a library of CoREAS simulations [@coreas] for each event obtained in the previous step with the goal to cover all possible depths of the shower maxima.
- Least-square fit of the simulated signal envelops to the reconstructed ones. The shower maximum and primary-particle energy are reconstructed on this stage.
Results of applying this method to the data are shown in the Fig. \[xmax\], left [@trex_template].
To better understand atmospheric impact we compared the predictions of the refractivity given by the standard CORSIKA atmosphere and the values given by GDAS [@gdas]. The results of this comparison are displayed in the Fig. \[xmax\], right.
Also, mainly due to the worsening of the radio noise environment on the site some important updates have been introduced in the signal processing chain [@kostunin_arena18]:
- Matched filtering and background suppression with convolutional neural networks [@shipilov_arena18; @marshalkina_arena18].
- A sliding window for noise estimation was introduced to make the noise estimation procedure more robust against RFI.
Tunka-Rex has resolution of the X$_{max}$ reconstruction about 30 g/cm$^2$. The resolution of the energy reconstruction is approximately 10% with systematic uncertainties of the instrument absolute calibration of about 20%.
![\[xmax\][*Left*]{}: Mean shower maximum as function of primary energy reconstructed by different experiments measuring the electromagnetic component of air showers [@trex_template]. [*Right*]{}: difference between the refractivity at the CORSIKA [@corsika] standard atmosphere and the one calculated with GDAS profiles for the Tunka location.](xmaxOverview_dEdX.pdf){height="13pc"}
![\[xmax\][*Left*]{}: Mean shower maximum as function of primary energy reconstructed by different experiments measuring the electromagnetic component of air showers [@trex_template]. [*Right*]{}: difference between the refractivity at the CORSIKA [@corsika] standard atmosphere and the one calculated with GDAS profiles for the Tunka location.](atmo.pdf){height="13pc"}
Aperture and Exposure Estimation
================================
The estimation of the instrument exposure is a key component for the correct reconstruction of the cosmic-ray flux measured by the observatory.
The general way to estimate the flux $J$ as a function of energy of the primary particle $E$ is the following [@auger_exposure]: $$J(E) = \frac{\mathrm{d}^4 N}{\mathrm{d} E \, \mathrm{d} s \, \mathrm{d} \Omega \, \mathrm{d} t} \approx \frac{\Delta N_{sel}(E)}{\Delta E} \frac{1}{\varepsilon(E)},$$ where $N$ is the number of primaries with energy between $E$ and $E+\mathrm{d} E$ passing through a surface element $\mathrm{d} s$ within a solid angle $\mathrm{d}\Omega$ and within time $\mathrm{d}t$. $\Delta N_{sel}(E)$ is the number of events passing the selection criteria in a given analysis in the energy bin centered in $E$ and having a width $\Delta E$. $\varepsilon(E)$ is the energy-dependent exposure.
The exposure can be expressed in the form of the integral of the detection efficiency $\xi$: $$\varepsilon(E) = \int \limits _{T_{sel}} \int \limits _{\Omega_{sel}} \int \limits _{S_{sel}} \xi \left( E,t,\vartheta,\varphi,x,y \right) \cos \vartheta \, \mathrm{d} s \, \mathrm{d} \Omega \, \mathrm{d} t =
\int \limits _T A (E,t) \, \mathrm{d} t,
\label{exposure}$$ with $\mathrm{d} \Omega = \sin \vartheta \, \mathrm{d} \vartheta \, \mathrm{d} \varphi$ and $\mathrm{d} s = \mathrm{d} x \,\mathrm{d} y$. After partial integration one obtains the value $A$, which is called instant aperture. The detection efficiency depends strongly on energy, direction, and shower core location as well as on observation time due to instrumental instability over time.
The idea is to use of the full efficiency regime and fiducial selection of area and angular coverage of the sky. Using this regime of full efficiency is the most reliable way to estimate the cosmic-ray flux since instrumental effects of changing the observed mass composition near the detection threshold do not influence the final results. In this case the aperture in the equation (\[exposure\]) can be factorized in the following way $$\varepsilon(E) = \int \limits _T \, S (E,t) A_{\Omega} (E,t) \, \mathrm{d} t,
\label{full-efficiency}$$ where $S (E,t)$, $A_{\Omega} (E,t)$ are the fiducial area of the instrument and the angular part of the aperture with fiducial selection of the viewing angle. Full efficiency means that the detection efficiency is equal to one, and is omitted hereafter. In the simplest case of stable operation the equation (\[full-efficiency\]) can be simplified $$\varepsilon(E) = S(E) A_{\Omega}(E) T,$$ where $T$ is the observation time at full efficiency.
In comparison to particle and non-imaging air-Cherenkov detectors, the full efficiency regime of radio measurements in the bandwidth we use has a distinguishing feature: the detection efficiency for arrival directions near the geomagnetic field is suppressed. This makes straightforward calculations of the instrumental full-efficiency angular coverage impossible.
Recently we developed a model taking into account this feature. This model uses elliptical shower footprints for calculation of detection efficiency for a given area and incoming direction. Parametrization of the ellipse parameters in this model is obtained from our experimental data.
This simple model, however, does not describe the observed detection efficiency in required detail. It works sufficiently well only in the intermediate range of zenith angles. Comparisons of the model against the hybrid observations with TAIGA detectors as well as detailed description of the model can be found in [@trex_exposure].
Calculations of aperture can be done with high precision by utilizing an analytical approach.
Since the main problem in the analytical calculation of the angular part of the aperture is the suppression around the magnetic field direction we can introduce a simple cut around this region. An additional cut for low zenith angles is introduced due to the low number of triggered antennas in this region.
Utilizing the considered cuts one obtains: $$A_{\Omega} (E) = 2 \int \limits _{\vartheta_{min}\left(E\right)} ^{\vartheta_{max}}
\left[ \pi
- \arccos \left( \frac{\cos \alpha_{min}\left(E\right) - \cos \vartheta \cos \vartheta_m}{\sin \vartheta \sin \vartheta_m} \right)
\right] \cos \vartheta \sin \vartheta \, \mathrm{d} \vartheta.
\label{a_omega}$$ In this equation $\vartheta_{max}$ is the maximal zenith angle, $\vartheta_{min}$ — minimal zenith angle, $\alpha_{min}$ — minimal geomagnetic angle required (recall that geomagnetic angle is the angle between the arrival direction and the geomagnetic field), $\vartheta_{M}$ — zenith angle of geomagnetic field. These cuts are applied in the fiducial selection of the events. The integral is solved numerically.
Details of the derivation this formula can be found in the appendix.
The presented method allows calculation of the experimental aperture semi-analytically taking advantage of energy dependent cuts. The choice of cut parametrization for the particular mode is under investigation as well as overall performance of the method.
Conclusion
==========
Tunka-Rex has been operating and successfully measuring cosmic rays since 2012. Reliable reconstruction methods and instrument calibration were developed during the last years. The main results of the Tunka-Rex can be summarized as following:
- Development of a robust reconstruction method for data containing only a few antennas with signals [@trex_template; @trex_reco].
- Experimental determination of the energy (10%) and X$_{max}$ (30 g/cm$^2$) precision by comparing radio and air-Cherenkov measurements of the same events [@cross-check].
- Comparing the energy scales of the Tunka-133 and KASCADE-Grande instruments with its radio extensions [@energy_scale].
- Development of an aperture and exposure model [@trex_exposure] and utilizing this model for the reconstruction of the mean shower maximum as function of energy [@trex_template].
The construction of Tunka-Rex was funded by the German Helmholtz association and the Russian Foundation for Basic Research (Grant No. HRJRG-303). This work has been supported by the Helmholtz Alliance for Astroparticle Physics (HAP), the Russian Federation Ministry of Education and Science (Tunka shared core facilities, unique identifier RFMEFI59317X0005, agreements: 3.9678.2017/8.9, 3.904.2017/4.6, 3.6787.2017/7.8, 3.6790.2017/7.8), the Russian Foundation for Basic Research (grants 16-02-00738, 17-02-00905, 18-32-00460) In preparation of this work we used calculations performed on the HPC-cluster “Academician V.M. Matrosov” \[36\] and on the computational resource ForHLR II funded by the Ministry of Science, Research and the Arts Baden-Württemberg and DFG (“Deutsche Forschungsgemeinschaft”). A part of the data analysis was performed using the radio extension of the Offline framework developed by the Pierre Auger Collaboration [@offline].
Details on Equation \[a\_omega\]
================================
The idea of the derivation of equation \[a\_omega\] can be illustrated with Fig. \[limits\]. First, we perform the integration of the cosine over area I, and then exclude the integral over area II. The coordinate system is chosen for simplification in a way that the azimuthal angle of the center of area II is zero.
![\[limits\]Cuts and areas used in the calculations (azimuthal equidistant projection of the sky). The integral is over the gray area.](integral_limits.pdf){height="15pc"}
Let us begin with the integral for area I. It is useful to recall that cosine in the initial integral holds information that the detector is flat. $$(I) = \int \limits _{\Omega'} \cos \vartheta \, \mathrm{d} \Omega =
\int \limits _{0} ^{2\pi} \int \limits _{\vartheta_{min}} ^{\vartheta_{max}} \cos \vartheta \sin \vartheta \, \mathrm{d} \vartheta \, \mathrm{d} \varphi =
2\pi \int \limits _{\vartheta_{min}} ^{\vartheta_{max}} \cos \vartheta \sin \vartheta \, \mathrm{d} \vartheta.$$ This equation can be integrated further $$(I) = \pi \left[ \cos ^2 \vartheta_{min} - \cos ^2 \vartheta_{max} \right].$$ This is the standard equation for the calculation of the aperture in the case of full efficiency and absence of any azimuth-dependent cuts. If $\vartheta_{min}=0$ then $(I) = \pi \left[ 1 - \cos ^2 \vartheta_{max} \right].$
For the area II we begin with the same equation and exploiting the symmetry for the integration: $$\label{ii}
(II) = \int \limits _{\Omega'} \cos \vartheta \, \mathrm{d} \Omega =
2 \int \limits _{\tilde{\Omega}} \cos \vartheta \, \mathrm{d} \Omega =
2 \int \limits _{\vartheta_{min}} ^{\vartheta_{max}} \int \limits _{0} ^{f(\vartheta)} \cos \vartheta \sin \vartheta \, \mathrm{d} \varphi \, \mathrm{d} \vartheta.$$ As upper limit of the integral over $\varphi$ we use dependency of $\varphi$ on $\vartheta$ on the circle of radius $\alpha_{min}$ around the direction to the local magnetic field ($\vartheta_M$, $\varphi$). We use equation for circle on spherical surface adopted from [@bronshtein] $$\cos \alpha_{min} = \cos \vartheta \cos \vartheta_M + \sin \vartheta \sin \vartheta_M \cos \varphi.$$ Hence it is easy to express the required dependency $$\varphi = \arccos \left( \frac{\cos \alpha_{min} - \cos \vartheta \cos \vartheta_M}{\sin \vartheta \sin \vartheta_M} \right).$$ Using this equation as the upper limit in \[ii\] the integral is $$(II) = 2 \int \limits _{\vartheta_{min}} ^{\vartheta'_{max}}
\arccos
\left( \frac{\cos \alpha_{min} - \cos \vartheta \cos \vartheta_M}{\sin \vartheta \sin \vartheta_M}
\right)
\cos \vartheta \sin \vartheta
\mathrm{d} \vartheta.$$ Here $\vartheta'_{max}$ is the maximal zenith angle of area II.
The final step is to perform the subtraction $$(I) - (II) =
\pi \left[ \cos ^2 \vartheta_{min} - \cos ^2 \vartheta_{max} \right]
- 2 \int \limits _{\vartheta_{min}} ^{\vartheta'_{max}}
\arccos
\left( \frac{\cos \alpha_{min} - \cos \vartheta \cos \vartheta_M}{\sin \vartheta \sin \vartheta_M}
\right)
\cos \vartheta \sin \vartheta
\mathrm{d} \vartheta.$$ This form is convenient for practical calculations. However, it can be can be expressed in the following shorter form $$(I) - (II) = 2 \int \limits _{\vartheta_{min}} ^{\vartheta_{max}}
\left[ \pi
- \arccos \left( \frac{\cos \alpha_{min} - \cos \vartheta \cos \vartheta_M}{\sin \vartheta \sin \vartheta_M} \right)
\right] \cos \vartheta \sin \vartheta \, \mathrm{d} \vartheta.$$
References {#references .unnumbered}
==========
[9]{} F.G. Schröder, Prog. Part. Nucl. Phys. 93, 1 (2017), 1607.08781 T. Huege, Phys. Rept. 620, 1 (2016), 1601.07426 L.A. Kuzmichev et al. — TAIGA Coll., Phys.Atom.Nucl. 81 (2018) no.4, 497-507 P. Abreu et al. — Pierre Auger Coll., JINST 7 (2012) P10011, 1209.3840 P.A. Bezyazeekov et al. — Tunka-Rex Coll., JCAP 01 (2016) 052, doi:10.1088/1475-7516/2016/01/052 T. Huege et at., AIP Conf.Proc. 1535 (2013) no.1, 128, 1301.2132 P.A. Bezyazeekov et al. — Tunka-Rex Coll., Phys.Rev. D97 (2018) no.12, 122004, 1803.06862 D. Heck et al., Forschungszentrum Karlsruhe Report FZKA 6019 (1998) Global Data Assimilation System (GDAS). https://www.ncdc.noaa.gov/data-access/model-data/model-datasets/global-data-assimilation-system-gdas D. Kostunin et al. — Tunka-Rex Coll., proceeding of ARENA 2018 (submitted to publication). D. Shipilov et al. — Tunka-Rex Coll., proceeding of ARENA 2018 (submitted to publication). T. Marshalkina et al. — Tunka-Rex Coll.. proceeding of ARENA 2018 (submitted to publication). P. Abreu et al. — Pierre Auger Coll., Astropart.Phys. 34 (2011) 368-381, 1010.6162 O. Fedorov et al. — Tunka-Rex Coll., PoS ICRC2017 (2018) 387, 1712.00974 D. Kostunin et al. — Tunka-Rex Coll., Astropart.Phys. 74 (2016) 79-86, 1504.05083 P.A. Bezyazeekov et al. — Tunka-Rex Coll., JCAP 1601, 052 (2016), 1509.05652 W.D. Apel et al. — Tunka-Rex and LOPES Coll., Phys. Lett. B763, 179 (2016) S. Argiro et al. — Pierre Auger Coll., Nucl.Instrum.Meth. A580 (2007) 1485-1496, 0707.1652 I.N. Bronshtein, K.A. Semendyayev, G. Musiol, H. Muehlig 2007 [*Handbook of Mathematics*]{} (Berlin: Springer) chapter 3 p 175
[^1]: now at the University of Zürich
[^2]: also at Vrije Universiteit Brussel, Belgium
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A combination of ground–based (NTT and VLT) and HST (HDF–N and HDF–S) public imaging surveys have been used to collect a sample of 1712 I–selected and 319 $K\leq 21$ galaxies observed with an extended spectral coverage from the U to the K band. Photometric redshifts have been obtained for all these galaxies, using a spectral library computed from Bruzual and Charlot models. The results have been compared with the prediction of an analytic rendition of the current CDM hierarchical models for galaxy formation that explicitly accounts for magnitude limits and dust extinction. We focus in particular on two observed quantities: the galaxy redshift distribution at $K\leq
21$ and the evolution of the UV luminosity density. The former has been proposed by Kauffmann and Charlot (1998) as a very robust prediction of any CDM hierarchical model, and we show that it is remarkably constant among different cosmological models. The derived photometric redshift distribution is in agreement with the hierarchical CDM prediction, with a fraction of only $5\%$ of galaxies detected at $z\geq 2$. This result strongly supports hierarchical scenarios where present–day massive galaxies are the result of merging processes. The observed UV luminosity density in our I-selected sample is confined within a factor of 4 over the whole range $0 < z < 4.5$, in agreement with previous spectroscopic and photometric surveys. CDM models in a critical ($\Omega =
1$,$\Lambda =0$) Universe are not able to produce the density of UV photons that is observed at $z\geq 3$. CDM models in $\Lambda$–dominated universe are in better agreement at $3 \leq z
\leq 4.5$, but predict a pronounced peak at $z\simeq 1.5$ and a drop by a factor of 8 from $z=1.5$ to $z=4$ that is not observed in the data. We conclude that improvements are required in the treatment of the physical processes directly related to the SFR, e.g. the starbust activity in merger processes and/or different recipes for linking the supernovae feedback to the star formation activity.
author:
- |
Adriano Fontana$^1$, Nicola Menci$^1$, Sandro D’Odorico$^2$, Emanuele Giallongo$^1$,Francesco Poli$^1$, Stefano Cristiani$^{3,4}$ , Alan Moorwood$^2$, Paolo Saracco$^5$\
$^{1}$ Osservatorio Astronomico di Roma, 00040 Monte Porzio, Italy\
$^{2}$ European Southern Observatory, 85748 Garching bei München, Germany\
$^{3}$ Dipartimento di Astronomia dell’ Università, 35122 Padova, Italy\
$^{4}$ Space Telescope European Coordinating Facility, European Southern Observatory, 85748 Garching bei München, Germany\
$^{5}$ Osservatorio Astronomico di Brera, 22055 Merate (LC) Italy
title: 'High redshift evolution of optically and IR–selected galaxies: a comparison with CDM scenarios '
---
-1truecm 2[cm$^{-2}$]{} Ø[$\Omega 4632$]{} HSD
Introduction
============
Understanding how massive galaxies formed and evolved is one of the major goals of present–day cosmology. Currently favoured theoretical scenarios attempt to describe “ab initio” the global formation and evolution of galaxies from primordial fluctuations including the main physical processes involved (e.g. Kauffmann, White & Guiderdoni 1993, Cole et al 1994, Baugh et al 1998). These “hierarchical” models naturally predict galaxies to form from smaller units that accrete gas and merge to build up present–day massive objects. These models are challenged by several observations suggesting that bulges and ellipticals were formed at a very early stage of the Universe and slowly evolved thereafter (Bernardi et al 1999, Schade et al 1999 and references therein).
Kauffmann and Charlot (1998; KC98 hereafter) proposed the use of the redshift distribution of K–band limited samples to address this issue. The main advantage here is that the K band traces the IR radiation produced by ordinary stars at any $z\leq4$ (see Fig. 1 of KC98) and is little affected by dust extinction , and it is therefore a reliable tracer of the mass in stars already assembled in galaxies at any redshift. As shown by KC98, hierarchical models prevent massive galaxies to be already assembled at $z\geq1$, and the expected number of galaxies at $z\geq 1.5$ is about 4 times lower than the predictions of Pure Luminosity Evolutionary (PLE) models. Unfortunately, the existing spectroscopic surveys (Cowie et al. 1996, Cohen et al. 1999) still lack the required completeness at faint IR magnitudes in this critical redshift regime.
The evolution of the global star–formation rate as a function of $z$ has long been recognized as a powerful tool to trace galaxy evolution. First results from spectroscopic (Lilly et al. 1995) and color–estimated (Madau et al 1996, 1998, Connolly et al 1997) redshift surveys suggested a steep rise and fall of the SFR with a main peak at $z\simeq 2$. Photometric redshift analysis (Giallongo et al 1998, G98 hereafter, Pascarelle et al 1999) and spectroscopic surveys at low and high $z$ (Treyer et al. 1998, Cowie et al. 1999, Steidel et al. 1999), inclusion of dust corrections and far–IR detections (Hughes et al 1998) are now modifying this picture.
In this work we have used deep ground-based and HST multi-band observations from UV to IR to obtain photometric redshifts for galaxies in an optical ($I<27.5$) and IR ($K<21$) sample. The redshift distributions of the K-band limited sample and of the UV luminosity density have been compared with the results of the CDM models to test their fundamentals properties.
The basic ingredients: Data and models
======================================
Multicolor catalogs
-------------------
The observations used in this paper cover the full wavelength range from the UV to the K band and have sub-arcsec image quality. Two of the fields were observed with the ESO 3.5 NTT SUSI imager: the first (hereafter BR1202) is centered on the z=4.7 QSO BR1202-07, the second is a neighbouring field (NTT Deep Field, NTTDF hereafter). The BVRI images and catalogs of these fields are described in G98 and Arnouts et al (1999a) respectively, complemented by NTT observations in J and K (Saracco et al. 1999) and in the U band. The latter observations and the procedures to obtain the final UBVRIJK catalogs are fully described in Fontana et al. (in preparation).
The third dataset results from the VLT-NICMOS observations of the HDF-S (Fontana et al 1999).
Finally, we have used the HDF-N and HDF–S with the IR observations obtained at Kitt Peak and at NTT–SOFI (Da Costa et al, 1998), respectively. For the HDF–N we have used the multicolor catalog published by Fernandez–Soto et al (1999), which uses an optimal technique to match the optical and IR images that have a quite different seeing. A similar catalog for the HDF–S has been provided by the same authors and is available at the WEB address [http://www.ess.sunysb.edu/astro/hfds/home.html]{}. Only WFPC bands have been used in the optical, for consistency with the HDF–N. In both cases we have clipped the outer regions of the frame with lower S/N.
Despite different origins, these data are sufficiently homogeneous for the purpose of the paper. Indeed, all catalogs have been obtained with similar procedures and software, and we have obtained photometric redshifts only for objects that are significantly above the detection threshold, so that small differences in the detection procedures are not expected to be important. All the multicolor catalogs use the optical images as detection frame, as is appropriate for the estimate of the UV luminosity density at high redshift. We have also performed an independent object detection on the K images alone to ensure that all the galaxy at $K<21$ were included in our optically-selected catalogs.
Photometric redshifts
---------------------
The multicolor catalogs have been used to derive photometric redshifts for all the galaxies in the sample, using a code already described elsewhere (G98, Fontana et al in preparation). The code is based on the synthetic models of the Bruzual and Charlot GISSEL library, with the addition of intergalactic absorption (Madau et al 1996) and dust reddening (SMC–like Pei 1992). The accuracy on the HDF–N spectroscopic sample is $\sigma _z\sim 0.06 (0.3)$ in the redshift interval $z=0-1.5 (1.5-3.5)$.
At fainter flux levels the reliability of photometric redshifts has been estimated with Monte Carlo simulations (Arnouts et al 1999b). We have defined a bright sample at $I_{AB}\le26$ that includes a subsample of the two HDFs and the NTTDF, and a fainter HDF(N+S) sample to $I_{AB}\le27.5$.
It is known that Galactic stars are significant sources of false high redshift candidates, especially in the brightest samples (Steidel et al 1999). Obvious stars have been excluded at $I_{AB}\leq 25.5$ in the HDF-S on the basis of the SExtractor morphological classification (Arnouts et al 1999a). The morphological selection removes all the $z\geq5$ candidates in the HDF–S. These objects are typically bright ($I = 20 - 24$) and are always detected in the JHK bands. They would be formally assigned to $z\geq 5$ since they are nearly undetected in V and have a negative $J-K$, typical of M star spectra. Analogously, we have used the detailed morphological classification developed in Poli et al (1999) to identify stars down to $I_{AB} \simeq
25.3$ in the NTTDF sample.
An analytical rendition of hierarchical models
----------------------------------------------
To compare these results with the present understanding of galaxy formation and evolution we have developed an analytical rendition of the hierarchical models (e.g. Cole et al. 1994). The prescription used to treat all the physical processes involved are identical to the “Durham” rendition, and we refer the reader to their list of papers for the details, while the complete formulations of our analytical rendition are given in Poli et al 1999 (see also Menci and Cavaliere 1999). Rather than following the history of each halo within a Monte Carlo scheme, we produce the statistical distributions of the main physical properties of galaxies in the DM halos. Our treatment extends the White and Frenk 1991 approach explicitly including the merging of galaxies in common halos through dynamical friction. This is accomplished by computing for all galaxies in DM halos the probability that the dynamical friction time is smaller than the halo survival time (as given by Lacey and Cole 1993). For the average quantities of interest here, this approach produces the same outputs as the “Durham” approach. We emphasize that we have not attempted to modify or improve the current models, that have several free parameters tuned to match the local properties of galaxies (counts, the I-band Tully-Fisher relation and the B-band luminosity function), since our aim is simply to compare their prediction with the picture emerging from our data.
The only improvement introduced is a self–consistent treatment of the dust absorption, by defining an effective optical depth $
\tau_\lambda$(Guiderdoni & Rocca–Volmerange 1987) that is used to suppress the expected luminosity: $$\tau_\lambda = \tau_{dust}^0 (1-\omega)^{1/2} (A_\lambda/ A_V) (Z_g/ Z_\odot)^s f_g$$
where $ (1-\omega)^{1/2} (A_\lambda/ A_V) (Z_g/ Z_\odot)^s$ is a metal–dependent extinction law, $f_g$ is the gas fraction $f_g = m_g / (m_g+m_*)$ (computed by the code) and $\tau_{dust}^0 $ is a gas–to–dust ratio chosen to match the observed B luminosity function at low redshift (Somerville and Primack 1998).
We adopt in this paper four different models, a Standard CDM (SCDM, $\Omega=1, \Lambda = 0, h=0.5$ ), an Open CDM ($\Omega=0.5,
\Lambda = 0, h=0.7$), a low-density flat model ($\Lambda$CDM, $\Omega=0.3, \Lambda = 0.7, h=0.6$) and a tilted model, ( $\Omega=1,
\Lambda = 0, h=0.5$). The power spectrum normalization and the parameters that describe the IMF, the star formation process and the galaxy merging are taken from Heyl et al. (1995) for SCDM, Open and $\Lambda$CDM models, and from Poli et al (1999) for the tilted model.
The K$<21$ samples
==================
\[nzK\]
\[epsi2700\]
The normalized cumulative redshift distributions of the 319 objects in the K–limited sample is shown in Fig.1. The upper panel shows the different distributions in the five fields considered here as well as the total distribution. The results from the five fields are clearly consistent, within the observed field–to–field scatter. As expected, the most discordant distributions come from the two smallest fields (BR1202 and VLTTC). For the purpose of the KC98 test, what is critical is the number of massive high $z$ galaxies detected in the sample. Only 9% of the galaxies are found at $z\geq 1.6$ (8%, 7% and 12% in the NTTDF, HDF–N and HDF–S respectively), and only 5% at $z\geq\ 2$ (6%, 4% and 2%). A similar conclusion was reached by Saracco et al 1999.
In the lower panel, the total distribution is compared to the predictions of PLE models (as adopted from KC98) and of the CDM models described in the previous section. All the CDM models are reasonably consistent with each other, despite the wide variety of cosmological and physical parameters adopted. As expected, the Open and the $\Lambda$ models predict the slower evolution. Conversely, there is a large difference between the ensemble of hierachical models and the PLE predictions in the KC98 rendition. This fact, that was already stressed by KC98, applies not only to the SCDM used by KC98 but also to other models with different cosmologies, strenghtening the validity of the KC98 test.
A main result of this paper is that [*the observed cumulative distribution is in good agreement with that predicted by the hierarchical models*]{}, strongly supporting hierarchical scenarios where massive objects are the result of merging processes in recent epochs.
To reiterate that the $K \leq 21$ threshold corresponds to a selection with respect to the mass of the galaxies we have labelled the upper x-axis of Fig.1 with the mass in stars contained in an evolved galaxy at the corresponding redshift, normalized to K=21. At $z\geq 1.6$, the minimum mass in stars in objects at $K\leq 21$ is $\simeq
10^{10}M_{\odot}$. However, a significant contribution to the K luminosity may also be due to the AGB population during a starburst phase. For comparison, the same K=21 luminosity may be obtained from a $z=1.6$ galaxy of only 0.1 Gyrs of age with a constant star formation rate of $10 M_{\odot}/$yr (and hence a mass of $10^{9}M_{\odot}$). In conclusion, both massive evolved objects and/or strong starbursts may contribute to the counts at $z\geq 1.6$ and $K\leq 21$. Since both classes are relatively rare in “bottom–up” hierarchical models, the redshift distribution is a sensitive test of these models.
The agreement between the observed and predicted distribution is slightly worse at $z\leq 0.5$. The observed [*differential*]{} distribution (Fig.1[*c*]{}) has indeed a paucity of galaxies at very low redshifts with respect to the theoretical predictions, an effect that produces the steeper cumulative distribution shown in Fig.1[*b*]{}. This is likely to be a combination of selection effects (these small fields have been explicitly chosen to be free of bright local galaxies) and of the slope of the faint end of the luminosty functions in CDM models, that is steeper than locally observed.
The cosmological evolution of the UV luminosity density
=======================================================
We show in Fig.2 the cosmological evolution of the UV luminosity density $\phi_{2700}$ as estimated from the photometric redshifts at $I_{AB}\leq 26$ (HDF–N + HDF–S +NTTDF, left panels) and at $I_{AB}\leq 27.5$ (HDF–N + HDF–S, right panels), for two different cosmologies. The $L_{2700}$ luminosity of each galaxy in the sample is directly obtained from the best–fitting spectrum, and falls in the range of the observed magnitudes at any redshift $z>0.25$. At $z>2.4$, most objects are undetected in the IR bands, and the fitting spectra are constrained by the corresponding upper limits.
The availability of these three different fields allows us for the first time to compare the evolution of $\phi_{2700}$ in different fields. At $I_{AB} \le 26$ the NTTDF is in good agreement with the HDF–N, while the HDF–S shows significant differences at $0.75 < z < 1.5$ and at $z>3.5$, due to the variance in the total counts and in the redshift distributions. At this stage, it is not clear whether the overall discrepancy among the fields, and most notably between HDF–N and HDF–S, is due to a real cosmic variance or to some instrumental effect, and we consider it as an estimate of the global uncertainities in this analysis. At $z\leq 1.5$, the results from the NTTDF and the HDF–N are consistent with those from spectroscopic surveys (Treyer 1998, Lilly 1995) when the corrections for steep luminosity functions are adopted, as seems appropriate for fields dominated by blue star–forming galaxies. At higher $z$ all the fields concur to a scenario where the UV luminosity density is relatively constant from $z=1$ to $z=4.5$.
The overall picture emerging from Fig.2 is that the UV luminosity density does not change by more than a factor of 4 over the redshift range $0 < z < 4.5$, the only exception beeing the $z\simeq
4$ redshift bin in the HDF–S (but see below).
The comparison with CDM models is less straightforward here. We have overplotted in Fig.2 the prediction of two well–studied examples, the SCDM and $\Lambda$–CDM. At variance with previous works, we have not corrected the observed values for incompleteness or extinction, but rather we have explicitly shown the differential effects of the inclusion of a magnitude limit and different dust extinction curves on the theoretical expectations. The comparison shows that the $\phi_{2700}$ overall shape is hardly recovered by the current CDM models. In particular, the SCDM model is not able to produce the density of UV photons that is observed at $z\geq 3$, while $\Lambda$–CDM is in better agreement at $3 \leq z \leq 4.5$, but predicts a pronounced peak at $z\simeq 1.5$ and a drop by a factor of 8 from $z=1.5$ to $z=4$ that is not observed in the data.
A more accurate comparison at high redshift can be carried out by plotting (Fig.3) the redshift evolution of the UV luminosity density at a shorter wavelength (1400 Å), where the best–fitting spectra are tied to the observed R and I bands, and comparing it with spectroscopic surveys and CDM models. Photometric surveys are consistent with the results of spectroscopic surveys on brighter samples with the exception of the HDF–S, especially at $z\geq 3.5$. The large variance between the HDF–N and the HDF–S is due to an intrinsic lack of high redshift galaxies in the latter. In particular 4 objects are identified at $z\geq 5$ in the HDF–N, while no convincing candidate is found in the HDF–S, after removing obvious stars. It should be noted that when a standard color selection as in Madau et al. (1996) is applied, a comparable number of B-dropout galaxies can be found in the HDF-N and -S. However, these represent only a fraction of the high redshift galaxies found by the photometric redshift technique (see Pascarelle et al. 1998, Fontana et al in preparation) that uses the IR bands as additional constraints. These additional high-z candidates are brighter and more numerous in the HDF–N than in the HDF–S, producing the different values shown in Fig. 3.
\[epsi1400\]
Summary
=======
We have collected and analyzed a sample of 1712 I–selected and 319 $K\leq21$ galaxies from public deep imaging surveys, mainly the two HDFs and the NTTDF. We have derived photometric redshifts for the whole sample in an homogeneuos way. The results may be summarized as follows:
- The redshift distribution of the $K\leq 21$ sample (Fig. 1) is dominated by objects at low or intermediate redshift, with a fraction of only $9\%$ of galaxies detected at $z\geq 1.6$ and of $5\%$ at $z\geq 2$;
- The UV luminosity density $\phi_{2700}$ is confined within a factor of 4 from $z=0$ to $z=4.5$ (Fig.2), and is relatively constant from $z=1$ to $z=4.5$, although significant field–to–field variations exist and dominate over statistical uncertainties;
- A comparison between HDF–N and HDF–S shows that the UV luminosity density $\phi_{1400}$ at $z\geq 4.5$ is still poorly determined, probably due to the cosmic variance between these two small fields. $\phi_{1400}$ changes by a factor of $\simeq 6$ between the two fields at $z\geq 4.5$, since no convincing $z\geq
5$ candidate is found in the HDF–S, compared to 4 in the HDF–N (2 of which have spectroscopic confirmation).
We have compared these results with the predictions of our analytical rendition of popular CDM models. We chose not to correct the data for incompleteness or dust extinction, but rather to include both effects in the theoretical model we compare with.
The $K\leq21$ redshift distribution at $z\geq 1$ directly reflects the number of massive galaxies already assembled at $z\geq1$ (KC98). The agreement that we find between the observed distribution and the prediction of an ensemble of CDM models (Fig.1[*b*]{}) strongly supports a key feature of these theoretical scenarios, i.e. that massive objects are the result of merging processes in recent epochs.
On the other hand the overall shape of the UV luminosity density, that is tied to the physical mechanisms driving the star formation processes, is not easily reproduced by current CDM models. The comparison between the observed evolution and the prediction of two different models (SCDM and $\Lambda$–CDM) shows that the SCDM model is not able to produce the density of UV photons that is observed at $z\geq 3$. Given the $I_{AB}\leq27.5$ limit applied to both observations and models, the discrepancy means that the current SCDM model fails to reproduce the bright tail of the luminosity function. $\Lambda$–CDM is in better agreement at $3 \leq z \leq 4.5$, but predicts a pronounced peak at $z\simeq 1.5$ and a drop by a factor of 8 from $z=1.5$ to $z=4$ that is not observed in the data. Such a result holds for all the adopted extinction laws. This implies that further refinements are required in the treatment of the physical processes directly related to the SFR. For instance, adopting a weaker feedback would increase the luminosity of fainter galaxies that dominate the statistics at $z\geq 2$ yielding a less steep decline of the SFR. Another possibility is that merging activity at $z\geq 2 $ is effective in enhancing the luminosity and/or the number density of faint galaxies at such $z$. A first attempt to include these effects has been described by Somerville, Primack and Faber 1998. These - or other - changes will require a global recalibration of the model parameters, in order to fit the increasing number of observables at low and high redshift.
[**Acknowledgments**]{}\
We thank Alvio Renzini for stimulating discussions on this research topics, the referee, Guinevere Kauffmann, for several useful comments which improved the paper and F. Governato, S. Savaglio and V. Testa for comments on earlier versions of this work. The paper is based on observations made with: the ESO VLT Antu telescope at the Paranal Observatory, the ESO New Technology Telescope at the La Silla Observatory (some of which under the EIS programs 59.A-9005(A), 60.A-9005(A)), the NASA/ESA Hubble Space Telescope and the Kitt Peak National Observatory. The ultraviolet observations of the NTTDF were performed in SUSI-2 guaranteed time of the Observatory of Rome in the framework of the ESO-Rome Observatory agreement for this instrument.
Arnouts, D’Odorico, S., Cristiani, S., Zaggia, S., Fontana, A., Giallongo, E., 1999a, A&A 341, 641 Arnouts, S., Cristiani, S., Moscardini, L., Matarrese, S.,Lucchin, F., Fontana, A., Giallongo, E., 1999b, MNRAS in press, [astro-ph/9902290]{} Baugh, C. M., Cole, S., Frenk, C. S., Lacey, C. G., 1998, ApJ, 498, 504
Bernardi, M., Renzini, A., da Costa, L. N., Wegner, G., Alonso, M. V., Pellegrini, P. S., Rite’, C., Willmer, C. N. A., 1999, ApJ, 508, 143
Cohen, J. G., Blandford, R., Hogg, D. W., Pahre, M. A., & Shopbell, P. L. 1999, ApJ, 512, 30
Cole, S., Aragón–Salamanca, Frenk, C. S., Navarro, J. F., Zepf, S. E., 1994, MNRAS, 271, 781
Connolly, A. J., Szalay, A. S., Dickinson, M., SubbaRao, M. U., & Brunner, R. J. 1997, ApJ, 486, L11
Cowie, L. L., Songaila, A., Barger, A. J. 1999, AJ in press, astro-ph/9904345
Cowie, L. L., Songaila, A., Hu, E. M., Cohen, J. G. 1996, AJ, 112, 839
Da Costa, L. et al , 1998, A&A subm. [astro-ph/9812105]{} Fernandez-Soto, A., Lanzetta, K. M., Yahil, A, 1999, ApJ 513, 34
A. Fontana A., D’Odorico S., Fosbury, R., Giallongo, E., Hook, I., Poli, F., Renzini, A., Viezzer R., , 1999, A&A 343, L19
Giallongo, E., D’Odorico, S., Fontana, A., Cristiani, S., Egami, E., Hu, E., McMahon, R. G., 1998, AJ 115, 2169 (G98)
Guiderdoni, B., & Rocca-Volmerange, B. 1987, A&A, 186, 1
Heyl, J. S., Cole, S., Frenk, C. S., Navarro, J. F., 1995, MNRAS, 264, 755 Hughes, D.H., Serejeant, S., Dunlop, J., Rowan–Robinson, M., Blain, A., Mann, R. G., Ivison, R., Peacock, J., Efstathiou, A., gear, W., Oliver, S., Lawrence, A., Longair, M.m Goldscmith, P., Jenness, T. 1998, Nature, 394,241 Kauffmann, G., White, S. D. M., Guiderdoni, B., 1993, MNRAS, 264, 201 Kauffmann, G. and Charlot, S., 1998, MNRAS, 297, L23 (KC98) Lacey, C., Cole, S., 1993, MNRAS 262, 627L Lilly, S. J., Tresse, L., Hammer, F., Crampton, D., & Le Fèvre, O. 1995 ApJ, 455, 108 Madau, P., 1995, ApJ, 441, 18 Madau, P., Ferguson, H. C., L., Dickinson, M., Giavalisco, M., Steidel, C. C., Fruchter, A., 1996 MNRAS, 283,1388
Madau, P., Pozzetti, L., Dickinson, M., 1998, ApJ, 498, 106 Menci, N., Cavaliere, A., 1999, MNRAS subm.
Pascarelle, S. M., Lanzetta, K. M., Fernandez–Soto, A., 1998, ApJ 508, L1 Pei, Y. C. 1992, ApJ, 395, 130 Poli, F., Giallongo, E., Menci, N., D’Odorico, S., Fontana, A., 1999, ApJ in press. Saracco, P., D’Odorico, S., Moorwood, A., Buzzoni, A., Cuby, J.-G., Lidman, C. 1999, A&A, in press Schade D., Lilly, S. J., Crampton, D., Ellis, R. S., Le Févre, O., Hammer, F., Brinchmann, J., Abraham, R., Colless, M., Glazebrook, K., tresse, L., Broadhurst, T., 1999 ApJ in press, astro-ph/9906171
Somerville, R. S., & Primack, J. R. 1998, MNRAS submitted, astro-ph/9802268 Somerville, R. S., & Primack, J. R. , Faber, S., M. 1998, MNRAS submitted, astro-ph/9806228
Steidel, C. C., M., Adelberger, K. L., Giavalisco, M., Dickinson, Pettini, M., 1999, ApJ 519, 1
Treyer, M. AQ., Ellis, R. S., Milliard, B., Donas J., Bridges, T. J. 1998, MNRAS, 300, 303
White, S., D. M., & Frenk, C. S. 1991, ApJ, 379, 52
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We demonstrate high-fidelity reversible transfer of quantum information from the polarisation of photons into the spin-state of an electron-hole pair in a semiconductor quantum dot. Moreover, spins are electrically manipulated on a sub-nanosecond timescale, allowing us to coherently control their evolution. By varying the area of the electrical pulse, we demonstrate phase-shift and spin-flip gate operations with near-unity fidelities. Our system constitutes a controllable quantum interface between flying and stationary qubits, an enabling technology for quantum logic in the solid-state.'
author:
- 'A. Boyer de la Giroday'
- 'A.J. Bennett'
- 'M.A. Pooley'
- 'R.M. Stevenson'
- 'N. Sköld'
- 'R.B. Patel'
- 'I. Farrer'
- 'D.A. Ritchie'
- 'A.J. Shields'
title: 'All-electrical coherent control of the exciton states in a single quantum dot'
---
Emerging quantum technologies such as quantum computing and quantum cryptography promise to revolutionise the way information is processed by controlling quantum bits or “qubits” in two-level quantum systems [@nielsen; @bouwmeester]. Photonic “flying” qubits are used for quantum communication [@gisin; @zeilinger2; @zeilinger1] as they easily travel in space or through standard optical fibres while solid-state “stationary” qubits are usually more convenient to perform quantum computations [@ladd; @koppens; @press; @greilich]. These two embodiments have thus far been developed independently, but coherent control of both types of qubits and coherent transfer of quantum information between them are necessary to extend the potential of quantum technologies [@kosaka].
Optical absorption of photons in a semiconductor to create an electron-hole pair (or exciton) provides a natural mechanism to initialise a stationary qubit. Selection rules determine the mapping of the photon polarisation onto the spin-state of the exciton. Once initialised manipulation of solid-state spins has been achieved with pulsed magnetic fields at milli-kelvin temperatures [@koppens], coherent optical beams [@press; @greilich] or couplings to localised optical modes [@imamoglu]. However, these techniques require bulky setups or synchronised lasers, making them somewhat impractical and incompatible for large-scale applications. New methods to control these qubits that take full advantage of well-established semiconductor technology will be advantageous when it comes to controlling large numbers of qubits.
Here, we advocate using the spin-state of an exciton trapped in a self-assembled semiconductor quantum dot (QD) as an electrically-controllable qubit in which quantum information from flying qubits can be efficiently transferred and easily restored. An exciton has two optically-active spin eigenstates which couple to linearly polarised photons and are energetically separated by a fine-structure splitting ($s$) resulting from the exchange interaction [@bayer; @gammon]. We control the splitting and the orientation of the eigenstates via an applied vertical electric field [@bennett], which allows for sub-nanosecond electrical coherent spin manipulation. We demonstrate high-fidelity phase-shift and spin-flip operations using dynamical modulation of the electric field.
The device used was similar in design to those of previous experiments [@patel] and consists in a $p-i-n$ heterostructure grown by molecular beam epitaxy. A single layer of InAs QDs with dot density $<$1$\mu$m$^{-2}$ was grown at the center of the intrinsic region made of a 10nm GaAs quantum well clad with a short period superlattice equivalent to Al$_{0.75}$Ga$_{0.25}$As on each side, which prevents tunneling of the carriers out of the dot region when the structure is biased. Doping extends into the superlattice and allows application of an electric field along the growth direction. The electric field $F$ is calculated using $F=\left(\frac{V-V_{bi}}{d}\right)$, where $V$ is the bias applied to the structure, $V_{bi}$=2.2V is the built-in potential and $d$=140nm is the thickness of the intrinsic region. This $p-i-n$ device is encased in a weak planar microcavity consisting of 14 (4) periods below (above) the dot layer. We used standard photolithography and wet etching techniques to fabricate a diode with an area of 35$\mu$m$\times$60$\mu$m. Excitation and photon collection occurs through an opaque metallic film on the sample surface patterned with micron-diameter apertures. The sample was cooled to $\approx$5K and excited quasi-resonantly using a mode-locked Ti:Sapphire laser.
We start by characterising the physical properties of our QD used as a solid-state photonic interface. Fig. 1(a) shows a schematic energy diagram of the QD. The exciton spin-state is initialised through quasi-resonant optical excitation and phonon relaxation. The phonon energy was measured to be $\approx$34.6meV corresponding to the 1-LO phonon in bulk GaAs. The eigenstates $\left|X_{H}\right\rangle$ and $\left|X_{V}\right\rangle$, coupled to linearly polarised photons, are separated by a tunable energy $s$. A vertical electric field reduces $s$, eventually causing an inversion in its sign. In Fig. 1(b), $\left|s\right|$ is shown to vary linearly with gradient $\pm$0.26$\mu$eVkV$^{-1}$cm. An anticrossing is measured at a field $F_{0}$=-155.4kVcm$^{-1}$ where the splitting is reduced to its minimum value of $s_{0}$=0.4$\mu$eV. This anticrossing originates from a field-dependent coupling between the eigenstates.
This coupling also leads to mixing of the eigenstates polarisation character and thus rotation of their orientation (Fig. 1(c)). We use the notation $\left|X_{H}\right\rangle$ and $\left|X_{V}\right\rangle$ to refer to the spin eigenstates away from the anticrossing, where $H$ and $V$ are the horizontal and vertical orientations in the lab-frame defined by the polarisation of photons to which the eigenstates couple. As the field is swept across $F_{0}$, the polarisation of those photons rotates by 90 degrees. At the anticrossing, the eigenstates $\left|X_{D}\right\rangle$ and $\left|X_{A}\right\rangle$ couple to photons with diagonal ($D$) and anti-diagonal ($A$) polarisations in the lab-frame.
Importantly, quasi-resonant optical excitation allows for initialisation of any superposition of spin-states by mapping the polarisation of an “input” excitation photon into the spin-state of the exciton [@flissikowski; @kowalik]. Considering an input photon of polarisation $\left|\Psi_{in}\right\rangle=\cos\theta\left|H\right\rangle+e^{i\varphi}\sin\theta \left|V\right\rangle$, the idealised coherent time evolution of the exciton spin-state (away from the anticrossing) is given by
$$\left|\Psi_{X}(t)\right\rangle=\cos\theta\left|X_{H}\right\rangle+e^{i\frac{st}{\hbar}}e^{i\varphi}\sin\theta\left|X_{V}\right\rangle.$$
The splitting $s$ therefore introduces a phase difference between the eigenstates accumulated over time at a rate of $st/\hbar$ and leading to coherent oscillations of the spin-state [@hudson]. Those oscillations can be observed by initialising an exciton and measuring its radiative emission in superpositions of eigenstates, leading to an ideal signal of intensity $I(t)\propto\cos\left(st/\hbar\right)e^{-t/\tau_{r}}$ (Fig. 1(d)), where $\tau_{r}$ is the radiative lifetime measured to be 1.28ns$\pm$0.08ns for our QD. Fitting those oscillations allows for temporal measurements of $\left|s\right|$ that agree well with spectral measurements (Fig. 1(b)).
![(colour online). (a) Schematic energy diagram of the QD used in our experiment where the polarisation of the laser is transferred into the exciton’s spin-state through quasi-resonant excitation. (b) and (c) Evolution of $\left|s\right|$ and orientation of the eigenstates relative to the lab-frame as a function of vertical electric field. Spectral and temporal measurements of $\left|s\right|$ are in good agreement. (d) High-visibility coherent oscillations of the spin-state for three different values of $\left|s\right|$. The exciton was initialised in a maximum superposition of eigenstates and emission was time-resolved along the same orientation.\[fig1\]](fig1){width="1.\columnwidth"}
We now characterise the use of an exciton as an interface between optical and solid-state qubits (Fig. 2). Measurements were performed at an electric field of -175kVcm$^{-1}$ where $\left|s\right|\approx$5$\mu$eV and where the eigenstates are oriented along $H$ and $V$. Fig. 2(a) illustrates the operating principle where an input qubit $\left|\Psi_{in}\right\rangle$ is first encoded into the polarisation of a photon defined by the parameters $\theta$ and $\varphi$ that are used as coordinates to represent the optical qubit on the Poincaré sphere. Quasi-resonant excitation of the QD then maps the polarisation of $\left|\Psi_{in}\right\rangle$ into the spin-state of the exciton qubit $\left|\Psi_{X}(t)\right\rangle$, therefore defining a point with the same coordinates ($\theta$,$\varphi$) on the Bloch sphere of the solid-state qubit. The phase accumulation between the eigenstates rotates the qubit around the equator of the Bloch sphere. Eventually, the electron-hole pair in the QD recombines and emits a photon of polarisation $\left|\Psi_{out}\right\rangle$ depending on the time spent in the solid-state $\left|\Psi_{X}(t)\right\rangle$.
The time evolution of $\left|\Psi_{X}(t)\right\rangle$ is characterised in Fig. 2(b), where the exciton was initialised in a diagonal superposition of eigenstates (corresponding to $\left|X_{D}\right\rangle$ on the Bloch sphere) and the emission was measured in linear $\left\{\left|H\right\rangle,\left|V\right\rangle\right\}$, diagonal $\left\{\left|D\right\rangle,\left|A\right\rangle\right\}$ and circular $\left\{\left|R\right\rangle,\left|L\right\rangle\right\}$ bases. According to Eq. (1) and Fig. 2(a), no oscillations are observed when measuring $\left|\Psi_{out}\right\rangle$ along the eigenstates while oscillations of angular frequency $s/\hbar$ are observed when measuring in diagonal and circular basis states, with a dephasing corresponding to the respective positions of those states on the equator.
We measure the fidelity of the interface $f_{in}$=$\left|\left\langle\Psi_{out}|\Psi_{in}\right\rangle\right|^{2}$ for six different input qubits prepared in the linear, diagonal, and circular states (Fig. 2(c)). Exciting an eigenstate leads to $f_{in}$=0.95$\pm$0.03 decaying with a timescale of 78ns$\pm$17ns, which is limited by spin scattering. Exciting a maximum superposition leads to an initial fidelity $f_{in}$=0.81$\pm$0.03 which oscillates due to the finite value of $s$. The envelope of the time-resolved fidelity decays with a timescale of 3.0ns$\pm$0.4ns limited by cross-dephasing, which randomizes the phase relationship between the two superimposed eigenstates [@stevenson2; @hudson]. Cross-dephasing is significantly longer than the radiative lifetime and the time needed to perform electrical manipulation of the qubit. The amplitude of the oscillation in the fidelity is thought to be reduced by the finite response time of the APD as well as the uncertainty in the time at which the superposition is created originating from the finite linewidth of the transition.
![(colour online). (a) Operating principle of our solid-state photonic interface. An optical qubit $\left|\Psi_{in}\right\rangle$ is first encoded in the polarisation of a photon. Quantum information is then coherently transferred from $\left|\Psi_{in}\right\rangle$ into the spin-state $\left|\Psi_{X}(t)\right\rangle$ of the exciton which rotates around the equatorial plan of the Bloch sphere at an angular velocity $\left|s\right|/\hbar$. Finally, the electron-hole pair recombines and emits a photon of polarisation $\left|\Psi_{out}\right\rangle$ depending on the time spent in the solid-state. (b) Complete characterisation of the time evolution of $\left|\Psi_{X}(t)\right\rangle$ prepared in the $\left|X_{D}\right\rangle$ spin-state by time-resolved measurements in linear, diagonal, and circular bases. Each signal has been normalised by the sum of the two measurements in the corresponding basis (vertical scale from 0 to 1). The symbols refer to measurements along different polarisations. (c) Fidelity $f_{in}$=$|\left\langle\Psi_{out}|\Psi_{in}\right\rangle|^{2}$ of the interface. The symbols refer to different input qubits.\[fig2\]](fig2){width="1.\columnwidth"}
Electrical coherent control of the phase accumulated by the qubit when initialised in a superposition of eigenstates can be demonstrated by dynamic modulation of $\left|s\right|$. We operate at values $\left|s\right|>$5$\mu$eV so that the orientation of the eigenstates does not change significantly with electric field. The time evolution is then given by $$\left|\Psi_{X}(t)\right\rangle=\cos\theta\left|X_{H}\right\rangle+ e^{i\frac{\int^{t}_{0}s(\tau)d\tau}{\hbar}}
e^{i\varphi}\sin\theta\left|X_{V}\right\rangle,$$ where $s(\tau)$ is modulated by an electrical pulse inducing faster phase accumulation during the gate operation (Fig. 3(a)). By applying a pulse of constant width but different amplitudes, we show that the phase-shift is proportional to the gate amplitude (Fig. 3(b)). Some deviation from linear behaviour arises from ringing in the electrical signal due to limited speed of our device and limited bandwidth of the pulse generator. Phase-shifts were fitted from time-resolved measurements shown in Fig. 3(c) where the exciton was initialised in a diagonal superposition $\left|X_{D}\right\rangle$ and emission was measured both along the diagonal and anti-diagonal polarisations. A 500ps gaussian electrical pulse was applied 250ps after the laser initialises the qubit. The intensity falls during the gate operation as emission is Stark-shifted out of the detection window. Fig. 3(d) and 3(e) show measurements in the diagonal basis after initialisation in $\left|X_{D}\right\rangle$, respectively without gate and with a $\pi$ phase-shift gate applied, the latter corresponding to a modulation of $\left|s\right|$ from a minimum of $\approx$5$\mu$eV to a maximum of $\approx$10$\mu$eV, estimated from the Stark shift resulting from the electrical pulse. The fidelity $f_{G}$ of a gate operation is given by [@zoller] $f_{G}=\left|\left\langle\Psi_{G}|U_{I}|\Psi_{in}\right\rangle\right|^{2}/f_{in}$, where $\Psi_{G}$ is the state of the photon emitted by the exciton after the gate operation and $U_{I}$ is the ideal gate transformation matrix. Normalisation to $f_{in}$ is used to evaluate the fidelity of the gate independently of the initialisation of the spin-state discussed previously. To take dephasing into account, the definition can be extended to [@li] $f_{G}=Tr\left[\rho_{G}U_{I}\rho_{X}U^{\ast}_{I}\right]/Tr\left[\rho^{2}_{X}\right]$, where $\rho_{G}=\left|\Psi_{G}\right\rangle\left\langle\Psi_{G}\right|$ is the density matrix of the photon emitted by the exciton after the gate operation, $\rho_{X}=\left|\Psi_{X}\right\rangle\left\langle\Psi_{X}\right|$ is the density matrix of the exciton state with no gate applied, and $Tr$ represents the trace. Fidelity as a function of time for the phase-shift gate is shown in Fig. 3(f). Irregularities in the evolution of the fidelity just after the gate are due to ringing in the electrical signal. Away from the gate, the fidelity fluctuates between 0.9 and 1.
![(colour online). (a) The electrical gate modulates $\left|s\right|$ and induces a faster phase accumulation between the spin eigenstates. The gated time-evolution (bottom) is consequently dephased compared to the ungated operation (top). (b) Phase-shift as a function of gate amplitude. (c) Normalised (colour scale from 0 to 1) time-resolved data for an exciton initialised in a diagonal superposition $\left|X_{D}\right\rangle$ and measurements in the diagonal basis. (d) and (e) Radiative emission from the QD initialised in $\left|X_{D}\right\rangle$ respectively without gate and with a $\pi$ phase-shift gate. (f) Fidelity of the $\pi$-shift gate operation as a function of time.\[fig3\]](fig3){width="1.\columnwidth"}
More complex operations are achieved by operating at small values of $\left|s\right|$ where the orientation of the eigenstates are strongly sensitive to the electric field (Fig. 1(c)). This ability to change not only the energetic splitting of the two levels but also to dynamically vary the eigenstates allows complete control of the qubit: any input state can be mapped onto any output state. To demonstrate this we now describe a “spin flip” between the two eigenstates, the operating principle of which is shown in Fig. 4(a). The exciton is initialised in the eigenstate $\left|X_{D}\right\rangle$ at minimum splitting $\left|s\right|$=0.4$\mu$eV and is represented by a point at the north pole of the Bloch sphere with the axis labelled $D$. An electrical pulse similar to that used to obtain the $\pi$ phase-shift is applied to the structure and rotates the eigenstates by 45 degrees without changing the orientation of the spin-state. As a consequence, the exciton is now in a superposition $\left|X_{D}\right\rangle$ represented at the equator of the Bloch sphere and still oriented diagonally in the lab-frame. A $\pi$ phase is then accumulated during the gate operation leading to the superposition $\left|X_{A}\right\rangle$. After the pulse, the eigenstates rotate back to their initial pre-pulse orientation with the exciton being now in its eigenstate $\left|X_{A}\right\rangle$, corresponding to a spin flip from $\left|X_{D}\right\rangle$. Fig. 4(b) and 4(c) show emission from both eigenstates (oriented along $D$ and $A$), after exciting $\left|X_{D}\right\rangle$. The polarisation of the emission after the gate is reversed, corresponding to a spin-flip in the solid-state. Fig. 4(d) shows the fidelity as a function of time obtained for the spin-flip operation shown in Fig. 4(c). The fidelity is 0.97 after the gate and then fluctuates due to ringing in the electrical signal.
![(colour online). (a) Operating principle of the spin-flip gate. The exciton is initialised at minimum splitting in its eigenstate $\left|X_{D}\right\rangle$. Modulating the electric field changes the value of $\left|s\right|$ and rotates the eigenstates by 45 degrees resulting in a superposition $\left|X_{D}\right\rangle$. A $\pi$ phase-shift is accumulated during the gate operation leading to a superposition $\left|X_{A}\right\rangle$. Returning to the initial value of the electric field induces another 45 degrees rotation of the eigenstates resulting in a spin flip. (b) and (c) Radiative emission from the QD respectively without gate and with the spin-flip gate. The trace with empty (plain) circles is obtained when exciting and measuring along the same (opposite) eigenstates. (d) Fidelity of the spin-flip gate as a function of time.\[fig4\]](fig4){width="1.\columnwidth"}
In conclusion, we demonstrated that an exciton in a QD is a high-fidelity solid-state photonic interface, on which arbitrary phase-shifts and spin-flips can be performed electrically. Quantum information encoded in the polarisation of a photon was transfered into and restored from the spin-state of the exciton. Coherent manipulation of the spin qubits was achieved through dynamical modulation of a vertical electric field and high-fidelity phase-shift and spin-flip gate operations were demonstrated. Using small gate areas and on-chip electronics will reduce ringing in the electrical signals and improve the gate fidelities. It will also allow operation times below 10ps [@chau], so that more than 300 operations can be performed within the coherence time. Furthermore, reducing the local optical density of states with cavity QED [@purcell] or using quantum dot molecules to separate the electron-hole pair and prevent recombination [@michler] would increase the exciton’s lifetime by orders of magnitude and allow for triggered emission. Extending our scheme to two qubits could be achieved in the near term through the creation of biexcitons [@li]. Further increase in the number of qubits would be achievable using site-positioned dots with low fine-structure splitting in devices with local gates, which is technologically feasible [@mohan].
We thank the EU FP7 FET programme for partial support through the Q-ESSENCE Integrated Project. Some of the authors (A.B.d.l.G, M.A.P. and R.B.P.) would also like to thank EPSRC and TREL for financial support.
[27]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study the scaling regimes for the Kardar–Parisi–Zhang equation with noise correlator $R(q)\propto (1 + w \, q^{-2 \rho})$ in Fourier space, as a function of $\rho$ and the spatial dimension $d$. By means of a stochastic Cole–Hopf transformation, the critical and correction-to-scaling exponents at the roughening transition are determined to all orders in a $(d - d_c)$ expansion. We also argue that there is a intriguing possibility that the rough phases above and below the lower critical dimension $d_c = 2 (1 + \rho)$ are genuinely different which could lead to a re-interpretation of results in the literature.'
author:
- 'E. Frey, U. C. Täuber H. K. Janssen'
title: |
Scaling regimes and critical dimensions in the\
Kardar–Parisi–Zhang problem
---
euromacr
€
The Kardar–Parisi–Zhang (KPZ) equation, which was originally introduced to describe growth of rough surfaces [@kardar-parisi-zhang:86], displays generic scale invariance, as well as a non-equilibrium roughening transition separating a smooth from a rough phase above the [*lower*]{} critical dimension. As a consequence of its mapping to the noisy Burgers equation [@forster-nelson-stephen:77], to the statistical mechanics of a directed polymer in a random environment [@huse-henley-fisher:85], as well as to other interesting equilibrium and non-equilibrium systems (for recent reviews, see Refs. [@halpin-healy-zhang:95; @krug:97]), the KPZ problem has emerged as one the fundamental theoretical models defining possible universality classes for non-equilibrium scaling phenomena and phase transitions.
In one dimension, the roughness and dynamic exponents, $\chi$ and $z$, have long been determined exactly by means of the dynamic renormalization group (RG), utilizing the symmetries of the problem [@forster-nelson-stephen:77]. Furthermore, it has been demonstrated that the associated scaling functions can be computed to high precision by means of the self-consistent one-loop or mode-coupling approximation [@hwa-frey:91; @frey-taeuber-hwa:96]. For $d > d_c$, a two-loop RG calculation [@frey-taeuber:94] indicated that the critical behavior of the roughening transition might be described by an exact set of exponents as suggested earlier on the basis of scaling arguments [@doty-kosterlitz:92]. Using a directed-polymer representation, Lässig was able to demonstrate the validity of this statement to all orders in a $(2 + \varepsilon)$ expansion [@laessig:95] (see also Refs. [@janssen_unpub:97; @wiese:98]). In addition, these results were evidence for a upper critical dimension $d_{\rm uc} = 4$ of the roughening transition, and suggested that the ensuing strong-coupling rough phase was not accessible within perturbation theory. This is in accord with the divergence of the coupling constant $g$ upon approaching the lower critical dimension from below [@frey-taeuber:94].
The scaling behavior in the strong-coupling rough phase above $d_c$ has been very controversial. Based on very different assumptions and analytic approaches, diverse values for the scaling exponents have been postulated, see, e.g., Refs. [@halpin-healy:90; @stepanow:97; @laessig:98]. In addition, some authors have claimed $d_{{\rm uc}} = 4$ to be the upper critical dimension for the rough phase as well [@moore-etal:95; @laessig-kinzelbach:97; @bhattacharjee:98], where the scaling exponents assume the values known in infinite dimensions [@derrida-spohn:88; @mezard-parisi:91]. In contrast, numerical studies observed merely continuously varying exponents as $d$ was increased [@ala_nissila-hjelt-kosterlitz-venaelaeinen:93]. In addition, the validity of the continuum Langevin description has been questioned in this regime, and a conceivable breakdown of universality has been conjectured (see, e.g., Refs. [@marsili-bray:96; @newman-swift:97]).
In order to shed light on some of these open issues, we shall find it useful to add a long-range power-law contribution to the usual spatially local stochastic noise of the KPZ equation, as first introduced by Medina [*et al.*]{} [@medina-hwa-kardar-zhang:89]. More generally, we investigate the Langevin equation, $$\partial_t s({\bf x},t) = D \, {\bf \nabla}^2 s({\bf x},t)
+ {D \, g \over 2} \, \left[ {\bf \nabla} s({\bf x},t) \right]^2
+ \zeta({\bf x},t) \ ,
\label{kpzeqn}$$ with Gaussian noise characterized by zero mean and variance $\langle
\zeta({\bf x},t) \, \zeta({\bf x}^\prime,t^\prime) \rangle = 2 \,
R_0({\bf x}-{\bf x}^\prime) \, \delta(t-t^\prime)$, which we assume to be [*local*]{} in time, but which may contain spatially long-range power law contributions of the form $R_0({\bf x}-{\bf x}^\prime)
\propto |{\bf x} - {\bf x}^\prime|^{2 \rho - d}$. Thus, typically we have $R_0({\bf q}) = D \left( 1 + w \, |{\bf q}|^{- 2 \rho} \right)$ in Fourier space. Notice that setting ${\bf u} = - {\bf \nabla} s$ in Eq. (\[kpzeqn\]) leads to the noisy Burgers equation, with the usual locally [*conserved*]{} noise for $w = 0$ or $\rho = 0$, but with [ *non-conserved*]{} noise for $\rho = 1$ (termed model B in Ref. [@forster-nelson-stephen:77]). We find as the most important effects of adding the long-range noise term: (i) the lower critical dimension for the roughening transition is shifted [*upwards*]{} to $d_c = 2 \left( 1 + \rho \right)$; (ii) above $d_c$, there are [ *two*]{} subtly distinct regimes for the [*smooth*]{} phase, characterized by different correction-to-scaling exponents; (iii) below $d_c$, there are two distinct [*rough*]{} regimes, governed by the short-range and long-range noise RG fixed points, respectively, and separated by a line $\rho_*(d)$ in the $(\rho,d)$ plane.
In the following, above $d_c$, we derive an [*exact*]{} integral equation for the noise correlation function by means of a stochastic Cole–Hopf transformation. Based on the ensuing minimally renormalized RG flow functions, we compute critical exponents at the roughening transition to [*all*]{} orders in a $(d - d_c)$ expansion, and determine the [*exact*]{} scaling exponents in the smooth phase above $d_c$. We demonstrate that the strong-coupling rough phase above $d_c$ is perturbationally inaccessible, but probably characterized by a spatially local noise correlator. Below the lower critical dimension $d_c$, we determine the scaling exponents at the long-range noise fixed point [*exactly*]{}, provided such a non-trivial finite fixed point exists. We discuss different analytical RG approaches to such a strong-coupling fixed point, and obtain an approximate expression for the separatrix $\rho_*(d)$. Finally, we discuss whether the rough phases above and below $d_c$, are “continuously” connected as a function of space dimension $d$ and the magnitude of the noise correlation exponent $\rho$. We shall argue that there is a intriguing possibility that these two rough phases are genuinely different.
A convenient starting point for a systematic analysis of a Langevin equation like Eq. (\[kpzeqn\]) is its reformulation in terms of a dynamic generating functional [@frey-taeuber:94], $$\begin{aligned}
{\cal J} [{\tilde s},s] &=& \int \! d^dx \int \! dt \; {\tilde s}({\bf x},t)
\biggl[ \partial_t s({\bf x},t) - D \, {\bf \nabla}^2 s({\bf x},t)
\label{dynfun} \\
&&- {D \, g \over 2} \left[ {\bf \nabla} s({\bf x},t) \right]^2
- \int \! d^dx^\prime R_0({\bf x}-{\bf x}^\prime) \,
{\tilde s}({\bf x}^\prime,t) \biggr] \ , \nonumber \end{aligned}$$ which allows the calculation of expectation values by means of path integrals with the statistical weight $\exp(-{\cal J}[{\tilde s},s])$. Upon directly applying a Cole–Hopf transformation to the KPZ equation, Eq. (\[kpzeqn\]), one obtains a diffusion equation subject to multiplicative noise which is often interpreted as a directed polymer in a random potential [@huse-henley-fisher:85]. If we recast the same idea in terms of the above dynamic functional the corresponding stochastic Cole–Hopf transformation [@janssen_unpub:97] — in the appropriately discretized version of Eq. (\[dynfun\]) in the Ito representation — reads $$\begin{aligned}
n({\bf x},t) &=& {2 \over g} \, \exp \left\{ {g \over 2} \,
\bigl[ s({\bf x},t) + D \, R_0({\bf 0}) \, t \bigr] \right\} \ ,
\label{holktr} \\
{\tilde n}({\bf x},t + \tau) &=& {\tilde s}({\bf x},t + \tau)
\exp \left\{ - {g \over 2} \bigl[ s({\bf x},t) + D \, R_0({\bf 0}) \, t
\bigr] \right\} \, . \nonumber\end{aligned}$$ This leads to a dynamic generating functional, $$\begin{aligned}
&&{\cal J} [{\tilde n},n]
= \int \! d^dx \int \! dt \biggl\{ {\tilde n}({\bf x},t)
\Bigl[ \partial_t n({\bf x},t) - D\, {\bf \nabla}^2 n({\bf x},t) \Bigr]
\label{holkdf} \\
&&- {D \, g^2 \over 4} \int \! d^dx^\prime \, {\tilde n}({\bf x},t) \,
n({\bf x},t) \, R_0({\bf x}-{\bf x}^{\prime}) \,
\tilde{n}({\bf x}^{\prime },t) \, n({\bf x}^{\prime },t) \biggr\} \ ,
\nonumber \end{aligned}$$ very reminiscent of the field theory for diffusion-limited pair annihilation [@lee:94]. A remarkable feature of ${\cal J} [{\tilde
n},n]$ is that there exist [*no*]{} loop diagrams contributing to a renormalization of the diffusion propagator. Hence the dynamic exponent is $z = 2$ whenever standard perturbation theory is applicable. This leaves us with the renormalization of the four-point noise vertex, which is formally achieved to [*all*]{} orders in the perturbation expansion via a Bethe-Salpeter equation, as graphically depicted in Fig. \[fig:bethe\_salpeter\]. Analytically, this leads to an integral equation for the renormalized noise correlator $R$, $$\begin{aligned}
R({\bf k},{\bf k}^\prime;\mu^2) = R_0({\bf k}-{\bf k}^\prime)
+ {g^2 \over 4} \int \! {d^dp \over (2\pi)^d} \,
{R_0({\bf k}-{\bf p}) \over p^2 + \mu^2} \, R({\bf p},{\bf k}'; \mu^2)
\ ,
\label{betsal}\end{aligned}$$ where $\mu^2 = i \omega / 2 D + q^2 / 4$, and ${\bf q}$ and $\omega$ are the momentum and frequency transfers from right to left in the vertices of Fig. \[fig:bethe\_salpeter\], while ${\bf k}-{\bf
k}^\prime$ denotes the momentum transfer from bottom to top. The standard perturbation expansion in terms of $g^2$ is equivalent to the Neumann series for this Fredholm integral equation. The [*exact*]{} relation (\[betsal\]) can be used to determine the scaling function for the renormalized noise self-consistently [@janssen-taeuber-frey:98].
0.6
In this letter, we focus on the asymptotic scaling exponents. Upon inserting the noise correlator into the Bethe–Salpeter equation (\[betsal\]), one readily obtains the full renormalization of the short-range noise strength $D$ within the minimal-subtraction procedure, which here leads to a formal [*double*]{} expansion with respect to $\varepsilon = d - 2$ [*and*]{} $\rho$. On the other hand, the non-analytic long-range contribution itself is not renormalized perturbationally. In terms of the renormalized counterparts $u$ and $v$ for the bare couplings $g^2$ and $w \, g^2$, respectively, the ensuing [*exact*]{} RG $\beta$ functions are actually precisely those of the one-loop theory, $$\beta_u = \varepsilon \, u - (u + v)^2 / 2 \ , \quad
\beta_v = (\varepsilon - 2 \rho) \, v \ .
\label{mnbeta}$$ For $d > d_c = 2(1+\rho)$, Eq. (\[mnbeta\]) shows that $v \to 0$ asymptotically. Therefore, the usual [*short-range*]{} noise KPZ scenario applies .
In addition to the trivial fixed point $u_* = 0$ and the strong-coupling fixed point $u_* = \infty$, there appears an [ *unstable*]{} critical fixed point $(u_c,v_c) = (2 \, \varepsilon,0)$, marking the location of a non-equilibrium [*roughening transition*]{} above the [*lower*]{} critical dimension $d_c$. At this $O(\varepsilon)$ fixed point, the general scaling relation $\chi + z =
2$ is valid, and because of $z_c = 2$ we find the marginal roughness exponent $\chi_c = 0$ [@doty-kosterlitz:92; @laessig:95]. Furthermore, the negative eigenvalue of the associated stability matrix defines the crossover or inverse correlation length exponent $\phi_c = \nu_c^{-1} = d - 2$, whereas its positive eigenvalue yields the correction-to-scaling exponent $\omega_c = d - d_c$. Assuming that the Chayes–Fisher bound $\nu > 2/d$ [@chayes-chayes-fisher-spencer:86] applies for the crossover length scale in this problem (equivalent to a directed polymer in a random environment), one finds that this lower bound is reached in $d_{{\rm uc}} = 4$ dimensions, which therefore constitutes the [ *upper*]{} critical dimension for the roughening transition, beyond which $\phi_c = 2$, $\nu_c = 1/2$ [@laessig:95; @taeuber-frey:95].
In the [*smooth*]{} phase, both couplings $u \to 0$ and $v \to 0$, and a more careful analysis of the RG flow of the ratio $w = v / u$ is required (details will be presented elsewhere [@janssen-taeuber-frey:98]). One finds that actually $w \to
\infty$, and thus the smooth phase, as opposed to the roughening transition, is characterized by the [*long-range*]{} algebraic noise, with roughness exponent $\chi_{{\rm sm}} = 1 + \rho - d / 2 \leq 0$, and $z_{{\rm sm}} = 2$. Furthermore, [*two distinct*]{} scaling regimes emerge, distinguished by different correction-to-scaling exponents: For $d_c < d < 2 (1 + 2 \rho)$, one finds $\omega_1 = d - 2
(1 + \rho)$ and $\omega_2 = 2 (1 + 2 \rho) - d$, whereas for $d \geq
2(1+2\rho)$, $\omega_1 = d - 2 (1 + 2 \rho)$ and $\omega_2 = 2 \rho$. Precisely at $d = d_c$, i.e., for $\rho = (d - 2) / 2$, there appears a fixed [*line*]{}, which is unstable for $w < 1$, and stable for $w >
1$.
Finally, the RG analysis also tells us that the rough phase emerging above $u_c$ is a genuine [*strong-coupling*]{} phase in the sense that it remains inaccessible to perturbative methods, even to [*all*]{} orders in $\varepsilon$. Numerical solutions of the flow equations also suggest that $w \to 0$ in the rough phase; hence the algebraic noise correlations appear to be [*irrelevant*]{} in the strong-coupling regime.
Up to now we have restricted our discussion to dimensions larger than the lower critical dimension, where there is a kinetic roughening transition from a smooth to a rough phase. If one tries to extend the above analysis to $d< d_c = 2(1+\rho)$, where there exists [*no*]{} roughening transition, Eq. (\[mnbeta\]) implies that $v \to \infty$ at long length scales. Thus, the [*minimally*]{} renormalized perturbation theory based on the stochastic Cole–Hopf transformation breaks down, and one must resort to other methods. Fortunately, however, we can draw some important [*exact*]{} conclusions already from the general structure of the field theory (\[dynfun\]). As a consequence of (a) the underlying Galilean invariance, which fixes the renormalization of $g$, (b) the fact that the non-analytic noise term proportional to $D w$ cannot renormalize, and (c) the momentum dependence of the non-linear vertex, there are merely [*two*]{} independent renormalization constants to be determined, namely for the renormalized fields, $s_R = Z^{1/2} \, s$ and the renormalized diffusion constant, $D_R = Z_D D$. Upon defining $\gamma = \mu
\partial_\mu \vert_0 \ln Z$ and $\zeta = \mu \partial_\mu \vert_0 \ln
Z_D$, the RG $\beta$ functions can be expressed as $\beta_u = (d - 2 -
\gamma - 2 \, \zeta) u$ and $\beta_v = (d - 2 - 2 \, \rho - 3 \,
\zeta) v$. Then by solving the RG equation for the correlation function near a stable RG fixed point, we furthermore identify $\chi =
(2 - d + \gamma_*) / 2$ and $z = 2 + \zeta_*$. The existence of a [*non-zero, finite*]{} RG fixed point $u_*$ then immediately leads to the scaling relation $\chi + z = 2$ [@frey-taeuber:94]. Similarly, at [*any*]{} long-range fixed point $0 < v_* < \infty$, $\zeta_*$ is fixed, giving the [*exact*]{} values $$z_{{\rm lr}} = (4 + d - 2 \, \rho) / 3 \ , \quad
\chi _{{\rm lr}} = (2 - d + 2 \, \rho) / 3
\label{lrexpo}$$ for the scaling exponents at the long-range fixed point, [*provided*]{} $0 < u_* < \infty$ as well.
In general, these two scaling fixed points compete, and the short-range fixed point must evidently be dominant, if $z_{{\rm sr}} <
z_{{\rm lr}}$ (and vice versa), which indeed implies $\beta_v > 0$ and hence $v \to 0$. In one dimension, and for purely local noise ($w =
0$), the non-linear reversible force term proportional to $g$ can be shown to fulfill the detailed-balance conditions, guaranteeing that the stationary probability distribution becomes ${\cal P}_{{\rm st}}
\propto \exp \left( -\int dx [\nabla s(x)]^2 /2 \right)$, as for the linear equation. In this Gaussian static theory, there can be no field renormalization ($Z = 1$), hence $\zeta = 0$, leading to the familiar one-dimensional KPZ short-range scaling exponents $\chi_{{\rm
sr}} = 1 / 2$ and $z_{{\rm sr}} = 3 / 2$ [@forster-nelson-stephen:77; @kardar-parisi-zhang:86]. Comparing the latter with the long-range dynamic exponent $z_{{\rm lr}} = (5 - 2
\rho) / 3$, we find that the short-range fixed point remains [ *stable*]{}, provided $\rho < 1/4$ [@medina-hwa-kardar-zhang:89]. After some controversy in the literature, this result has been confirmed by simulations for the noisy Burgers equation [@hayot-jayaprakash:96].
Notice that a [*minimal*]{} renormalization scheme can never arrive at the exact one-dimensional exponents $\chi_{{\rm sr}} = 1 / 2$ and $z_{{\rm sr}} = 3 / 2$. Instead, in order to address the rough phase below $d_c$, one may devise a [*non-minimal*]{} renormalization procedure at fixed dimension $d$ and $\rho$ [@frey-taeuber:94; @janssen-taeuber-frey:98]. Alternatively, one may utilize the mapping to the Burgers equation and hence to driven diffusive systems, for which a well-defined $(2 - d)$ expansion exists. Adding long-range correlated noise, this actually leads to the identical stability condition $\rho < 1/4$ for the short-range fixed point in $d = 1$ [@janssen-schmittmann:98]. In the long-range regime, the case $\rho = 1$, corresponding to the Burgers equation with [*non-conserved*]{} noise, is accessible through an $\varepsilon$ expansion below the upper critical dimension $d_{{\rm
uc}} = 4$ of this model [@forster-nelson-stephen:77]. The dynamic exponent here is actually obtained to all orders in $\varepsilon$, and reads $z_{{\rm lr}} = (2 + d) / 3$.
In order to further discuss the implications of the above [*exact*]{} results for the KPZ equation with long-range correlated noise it is instructive to consider the scaling regimes in the $(d,\rho)$ landscape; see Fig. \[fig:phase\_diagram\]. One should notice that there are two [*qualitatively*]{} distinct regions separated by the lower critical dimension $d_c(\rho) = 2 (1 + \rho)$.
0.8
For $d < d_c(\rho)$ there is only a rough phase and the short-range (SR) and long-range (LR) noise fixed points compete. Above the lower critical dimension there appears a phase transition from a smooth to a rough phase and the RG analysis indicates that LR noise constitutes an irrelevant perturbation in the rough phase.
[F]{}rom the above analysis we know the values of the scaling exponents in the domain of attraction of the LR fixed point exactly. Since the scaling exponents at the SR fixed point are independent of $\rho$, the only additional information needed to determine the SR exponents is the separatrix $\rho_* (d)$ between the domains of attraction of the LR and SR fixed points. Of course, for the latter to be true one has to require that the exponents are continuous at $\rho_* (d)$. This has been shown explicitly for $d=1$, and seems very reasonable in general. From the analysis in this letter we cannot determine the exact form of the separatrix. Yet, we actually can locate some points on this curve. In $d = 1$, the short-range fixed point is stable for $\rho < \rho_* = 1/4$; for $\rho = 1$ (the Burgers equation with non-conserved noise), and below $d_{{\rm uc}} = 4$, there is only the long-range regime. If we assume that the separatrix $\rho_*(d)$ between the short-range and long-range regimes extends up to four dimensions, then it definitely contains the points $(1,4)$, $(1,1/4)$, and probably also $(0,0)$ in the $(\rho,d)$ plane. A simple linear interpolation yields the function $\rho_*(d) \approx d / 4$ (dashed line in Fig. \[fig:phase\_diagram\]); recent computer simulations in fact found $\rho_*(2) \approx 1/2$ [@li:97]. Inserting the linear interpolation approximation (which we do not expect to be exact), $\rho_*(d) \approx d/4$, into Eq. (\[lrexpo\]) yields $$z_{{\rm sr}} \approx (8 + d) / 6 \ , \quad
\chi _{{\rm sr}} \approx (4 - d) / 6 \ .
\label{srexpo}$$ These values remarkably coincide with Halpin-Healy’s results obtained in a functional RG [@halpin-healy:90] and also with a more recent perturbative mode-coupling study by Bhattacharjee [@bhattacharjee:98].
There are now two possible scenarios. The first one is that all the scaling exponents for the strong-coupling phase are continuous over the whole $(\rho,d)$ plane and in particular upon crossing the line marking the lower critical dimension. This would necessarily imply (if not prove) that $d_{\rm uc} = 4$ is the [*upper*]{} critical dimension for the rough phase in agreement with some recent speculations [@laessig-kinzelbach:97] but disagreement with computer simulations [@ala_nissila-hjelt-kosterlitz-venaelaeinen:93]. However, one should probably expect to encounter singularities as the lower critical dimension $d_c(\rho)$ is crossed. A second more intriguing scenario is therefore that there are two fundamentally different rough phases, one below and one above the lower critical dimension $d_c (\rho) = 2 (1 + \rho)$, which we call type-I (SR-I) and type-II (SR-II). This would then allow for the scaling exponents to be discontinuous at $d_c (\rho)$. Below $d_c (\rho)$ one would have scaling exponents close to those obtained from the linear interpolation of the separatrix in Eq. (\[srexpo\]). In particular the upper critical dimension of the type-I rough phase is $d_{uc} =4$. Above the lower critical dimension $d_c (\rho)$ one would have a different set of exponents (e.g. those given by the numerical simulations) and the upper critical dimension must not necessarily be equal to $4$ or any other finite value. At present there is proof for neither of the above scenarios but the following observations are quite indicative. First, from the RG analysis exploiting the Cole–Hopf transformation we have learned that the strong-coupling phase above $d_c(\rho)$ is not accessible by perturbation theory even to infinite order. On the other hand the rough phase at $d=1$ is accessible by standard perturbation theory using a mapping of the KPZ equation to a driven diffusion model [@janssen-schmittmann:98]. Second, for $\rho=0$ an explicit two-loop calculation [@frey-taeuber:94] shows that the fixed point value of the coupling constant $g$ approaches infinity as the lower critical dimension approaches $2$ from below. Third, this scenario would provide a coherent picture for most of the available numerical and analytical results for the KPZ equation. Some of the analytical approaches (e.g. functional RG and mode-coupling theory) are self-consistent theories and hence necessarily start out with correlated noise. This in effect shifts the lower critical dimension upwards, and it is well possible that this automatically constrains the results to the SR-I phase as opposed to the SR-II phase these studies were supposedly aiming at. It is difficult to judge on other analytical methods and how they fit into the scheme discussed here. Lastly, there have been suggestions for a breakdown of a continuum description, and perhaps even universality, in the rough phase [@marsili-bray:96; @newman-swift:97]. Physically, in a microscopically rough regime, the underlying lattice as well as details of the dynamic rules may well be important; in addition, one might question the existence of a well-defined short-distance expansion [@laessig:98] in this phase. Clearly, these issues require further clarification through approaches that extend beyond the continuum equation, Eq. (\[kpzeqn\]), and equivalent field theory methods.
We benefited from discussions with J.L. Cardy, T. Halpin-Healy, and K.J. Wiese. E.F. and U.C.T. are grateful for support from the DFG through Heisenberg and habilitation fellowships Fr 850/3-1 and Ta 177/2-1,2. H.K.J. acknowledges support from the SFB 237 (“Unordnung und große Fluktuationen”), funded by the DFG.
[10]{}
, [Parisi, G.]{} and [Zhang, Y.-C.]{}, (1986) 889.
, [Nelson, D. R.]{} and [Stephen, M. J.]{}, (1977) 732.
, [Henley, C. L.]{} and [Fisher, D. S.]{}, (1985) 2924.
and [Zhang, Y.-C.]{}, (1995) 215.
, (1997) 139.
and [Frey, E.]{}, (1991) R7873.
, [T[ä]{}uber, U. C.]{} and [Hwa, T.]{}, (1996) 4424.
and [T[ä]{}uber, U. C.]{}, (1994) 1024.
and [Kosterlitz, J. M.]{}, (1992) 1979.
, (1995) 559.
, [*unpublished*]{} (1997).
, (1998).
, (1990) 711.
, (1997) R4853.
, (1998) 2366.
, (1995) 4257.
and [Kinzelbach, H.]{}, (1997) 903.
, (1998) L93.
and [Spohn, H.]{}, (1988) 817.
and [Parisi, G.]{}, (1991) 809.
, [Hjelt, T.]{}, [Kosterlitz, J. M.]{} and [ Venäläinen, O.]{}, (1993) 207.
and [Bray, A. J.]{}, (1996) 2750.
and [Swift, M. R.]{}, (1997) 2261.
, [Hwa, T.]{}, [Kardar, M.]{} and [Zhang, Y.-C.]{}, (1989) 3053.
, (1994) 2633.
, [Täuber, U. C.]{} and [Frey, E.]{}, (1998).
, [Chayes, L.]{}, [Fisher, D. S.]{} and [Spencer, T.]{}, (1986) 2999.
and [Frey, E.]{}, (1995) 6319.
and [Jayaprakash, C.]{}, (1996) 4681.
and [Schmittmann, B.]{}, (1998).
, (1997) 1178.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, a notion of affine walled Brauer-Clifford superalgebras $BC_{r, t}^{\rm aff} $ is introduced over an arbitrary integral domain $R$ containing $2^{-1}$. These superalgebras can be considered as affinization of walled Brauer superalgebras in [@JK]. By constructing infinite many homomorphisms from $BC_{r, t}^{\rm aff}$ to a class of level two walled Brauer-Clifford superagebras over $\mathbb C$, we prove that $BC_{r, t}^{\rm aff} $ is free over $R$ with infinite rank. We explain that any finite dimensional irreducible $BC_{r, t}^{\rm aff} $-module over an algebraically closed field $F$ of characteristic not $2$ factors through a cyclotomic quotient of $BC_{r, t}^{\rm aff} $, called a cyclotomic (or level $k$) walled Brauer-Clifford superalgebra $ BC_{k, r, t}$. Using a previous method on cyclotomic walled Brauer algebras in [@RSu1], we prove that $BC_{k, r, t}$ is free over $R$ with super rank $(k^{r+t}2^{r+t-1} (r+t)!, k^{r+t}2^{r+t-1} (r+t)!)$ if and only if it is admissible in the sense of Definition \[condi-kcycl\]. Finally, we prove that the degenerate affine (resp., cyclotomic) walled Brauer-Clifford superalgebras defined by Comes-Kujawa in [@CK] are isomorphic to our affine (resp., cyclotomic) walled Brauer-Clifford superalgebras.'
address:
- 'M.G. Department of Mathematics, East China Normal University, Shanghai, 2000240, China'
- 'H.R. School of Mathematical Science, Tongji University, Shanghai, 200092, China'
- 'L.S. School of Mathematical Science, Tongji University, Shanghai, 200092, China'
- 'Y.S. School of Mathematical Science, Tongji University, Shanghai, 200092, China'
author:
- 'Mengmeng Gao, Hebing Rui, Linliang Song, Yucai Su'
title: 'Affine walled Brauer-Clifford superalgebras'
---
[^1]
Introduction
============
In his pioneer’s work, Schur considered $V^{\otimes r}$, the $r$-th tensor product of the natural module $V$ of the general linear group $GL_n(\mathbb C)$. This is a left $GL_n(\mathbb C)$-module such that $GL_n(\mathbb C)$ acts on $V^{\otimes r}$ diagonally. There is a right action of the symmetric group $\Sigma_r$ on $V^{\otimes r}$, and two such actions commute with each other. This enabled Schur to establish a duality between the polynomial representations of $GL_n(\mathbb C)$ and the representations of symmetric groups over $\mathbb C$. Later on, such a result was generalized by Brauer [@B], Sergeev [@Ser] and Lehrer-Zhang[@GZ] and so on. In these cases, the $r$-th tensor product $V^{\otimes r}$ are considered where $V$ is the natural module of a symplectic group or an orthogonal group or a queer Lie superalgebra $\mathfrak{q}(n)$ or an orthosymplectic supergroup and so on. The Brauer algebras and the Hecke-Clifford superalgebras etc naturally appear as endomorphism algebras of $V^{\otimes r}$.
Koike [@Koi] considered the mixed tensor modules $V^{\otimes r}\OTIMES (V^*)^{\otimes t}$ of the $r$-th power of the natural module $V$ and the $t$-th power of the dual natural module $V^*$ of the general linear group $GL_n(\mathbb C)$ for various $r,t\in\Z^{\ge0}$. This led him to introduce the notion of walled Brauer algebras in [@Koi] (see also [@Tur]). Shader and Moon [@SM] set up super Schur-Weyl dualities between walled Brauer algebras and general linear Lie superalgebras, by studying mixed tensor modules of general linear Lie superalgebras ${\frak{gl}}_{m|n}$. Brundan and Stroppel [@BS4] established super Schur-Weyl dualities between level two Hecke algebras $H_r^{p,q}$ and ${\frak{gl}}_{m|n}$, by studying tensor modules $K_\l\OTIMES V^{\otimes r}$ of Kac modules $K_\l$ with the $r$-th power $V^{\otimes r}$ of the natural module $V$ of ${\frak{gl}}_{m|n}$. This led them to obtain various results including the celebrated one on Morita equivalences between blocks of categories of finite dimensional ${\frak{gl}}_{m|n}$-modules and categories of finite dimensional left modules over some generalized Khovanov’s diagram algebras [@BS]. By studying tensor modules $M^{r,t}:=V^{\otimes r}\OTIMES K_\l\OTIMES (V^*)^{\otimes t}$ of Kac modules $K_\l$ with the $r$-th power of the natural module $V$ and the $t$-th power of the dual natural module $V^*$ of $\frak{gl}_{m|n}$, two of authors [@RSu; @RSu1] introduced a new class of associative algebras, referred to affine walled Brauer algebras, over a commutative ring containing $1$.[^2] They established super Schur-Weyl dualities between level two walled Brauer algebras $B_{2,r,t}$ and general linear Lie superalgebras, which enables them to classify highest weight vectors of ${\frak{gl}}_{m|n}$-modules $M^{r,t}$, and to determine decomposition numbers of $B_{2,r,t}$ arising from super Schur-Weyl duality. In order to further study representation theory of queer Lie superalgebras and to establish higher level mixed Schur-Weyl duality between queer Lie superalgebras and some associative algebras, a natural question is, what kind of algebras may come into the play if one replaces general linear Lie superalgebras ${\frak{gl}}_{m|n}$ by queer Lie superalgebras $\mathfrak{q}(n)$. This is one of the motivations of the present paper to introduce the notion of affine walled Brauer-Clifford superalgebras. Another motivation comes from two of authors’ work on the Jucys-Murphy elements of walled Brauer algebras in [@RSu].
In 2014, Jung and Kang [@JK] introduced the notion of walled Brauer superalgebras or walled Brauer-Clifford superalgebras $BC_{r, t}$ so as to establish the mixed Schur-Weyl-Sergeev duality for queer Lie superalgebras $\mathfrak{q}(n)$. The superalgebra $BC_{r, t}$ can be considered as a generalization of a Hecke-Clifford superalgebra and a walled Brauer algebra. In the present paper, we construct Jucys-Murphy elements for $BC_{r, t}$ and study its properties in detail. Through these elements, we can introduce the notion of affine Brauer-Clifford superalgebras $BC_{r, t}^{\rm aff}$ in a ring theoretical way. Using arguments similar to those in [@RSu], we construct infinite many homomorphisms between the affine Brauer-Clifford superalgebra $BC_{r, t}^{\rm aff}$ and walled Brauer-Clifford superalgebras, and thus we are able to prove that the affine walled Brauer-Clifford superalgebra is free over $R$ with infinite rank if the defining parameter $\omega_1$ is zero. However, many affine walled Brauer-Clifford superalgebras which appear in the higher level mixed Schur-Weyl-Sergeev duality have non-zero defining parameter $\omega_1$. In order to overpass this, we consider level two mixed Schur-Weyl-Sergeev duality for $\mathfrak{q}(n)$ and prove that a class of level two walled Brauer-Clifford superalgebras over $\mathbb C$ with non-zero $\omega_1$ have required super-dimensions. Using these level two walled Brauer-Clifford superalgebras instead of walled Brauer-Clifford superalgebras used before, we can establish infinite many superalgebra homomorphisms and hence prove the freeness of the affine walled Brauer-Clifford superalgebra over $R$ no matter whether $\omega_1$ is zero or not. This is one of the points which is different from the work in [@RSu].
It is a natural problem to give a classification of finite dimensional irreducible $BC_{r, t}^{\rm aff} $-modules over an algebraically closed field of characteristic not $2$. By introducing cyclotomic quotients of $BC_{r, t}^{\rm aff}$, called cyclotomic walled Brauer-Clifford superalgebras, we are able to prove that any finite dimensional irreducible $BC_{r, t}^{\rm aff} $-module factors through a cyclotomic walled Brauer-Clifford superalgebra. We define this superalgebra over $R$ and prove that it is free over $R$ with required rank if and only if it is admissible in the sense of Definition \[condi-kcycl\]. In a sequel, we will classify finite dimensional irreducible modules for affine and cyclotomic walled Brauer-Clifford superalgebras over an arbitrary (algebraically closed) field with characteristic not $2$.
We notice that degenerate affine walled Brauer-Clifford algebras and their cyclotomic quotients are also introduced by Comes and Kujawa via affine oriented Brauer-Clifford supercategory [@CK] in the early of July, 2017 (at that time we had obtained our affine and cyclotomic walled Brauer-Clifford algebras), and they prove that their degenerate affine walled Brauer-Clifford superalgebras are free over $R$ with infinite rank. We introduce affine walled Brauer-Clifford algebras or their cyclotomic quotients in terms of generators and defining relations in order to study relationship between finite dimensional modules of affine walled Brauer-Clifford algebras or their cyclotomic quotients and those of queer Lie superalgebras. Motivated by [@BCNR], we prove that a degenerate affine walled Brauer-Clifford superalgebra in [@CK] is isomorphic to one of our affine walled Brauer-Clifford superalgebras. We also prove that their cyclotomic walled Brauer-Clifford superalgebra is isomorphic to one of ours if both of these two superalgebras are admissible in the sense of Definition \[condi-kcycl\]. This also gives a proof of the freeness of Comes-Kujawa’s cyclotomic walled Brauer-Clifford superalgebra. Such a result is not available in [@CK].
We organize our paper as follows. In section 2, we recall the notion of walled Brauer-Clifford superalgebras $BC_{r, t}$ in [@JK]. Several properties on the Jucys-Murphy elements of $BC_{r, t}$ are given. This leads us to introduce the notion of affine walled Brauer-Clifford superalgebras in section 3. We give infinite many homomorphisms between $BC_{r, t}^{\rm aff} $ and walled Brauer-Clifford superalgebras. We also define cyclotomic walled Brauer-Clifford superalgebras. In section 4, we use higher level mixed Schur-Weyl-Sergeev dualities to prove that a class of level two walled Brauer-Clifford algebras with non-zero parameter $\omega_1$ have required super-dimensions over $\mathbb C$. In section 5, we construct infinite many homomorphisms between $BC_{r, t}^{\rm aff} $ and level two walled Brauer-Clifford superalgebras which appear in the higher level mixed Schur-Weyl-Sergeev dualities in section 4. This in turn enables us to mimic arguments in [@RSu] to prove the freeness of $BC_{r, t}^{\rm aff} $ over $R$. In particular, $BC_{r, t}^{\rm aff} $ is of infinite super-rank. In section 6, we prove that a cyclotomic walled Brauer-Clifford superalgebra is free over $R$ with required super rank if and only if it is admissible. Finally, in section 7, we prove that the degenerate affine (resp., cyclotomic) walled Brauer-Clifford superalgebra in [@CK] is isomorphic to ours. In the later case, we need to assume that both superalgebras are admissible.
Walled Brauer-Clifford superalgebras
====================================
Throughout, we assume that $R$ is an integral domain containing $2^{-1}$. Let $\Sigma_r$ be the [*symmetric group*]{} in $r$ letters. Then $\Sigma_r$ is generated by $s_1, \ldots, s_{r-1}$, subject to the relations (for all admissible $i$ and $j$): $$\label{symm} s_i^2=1, \ \ s_is_{i+1}s_i=s_{i+1}s_is_{i+1}, \ \ \text{ $s_is_j=s_js_i$, if $|i-j|>1$.}$$ Each $s_i$ can be identified with the simple reflection $(i, i+1)$, where $(i, j)\in \Sigma_r$, which switches $i , j$ and fixes others. In this paper, we always assume that $\Sigma_r$ acts on the right of the set $\{1, 2, \ldots, r\}$.
The [*Hecke-Clifford algebra*]{} $HC_{r}$ was introduced by Sergeev [@Ser] in order to study the $r$-th tensor product of the natural module for the queer Lie superalgebras $\mathfrak{q}(n)$. It is the associative $R$-superalgebra generated by even elements $s_1, \ldots, s_{r-1}$ and odd elements $c_1, \ldots, c_r$ subject to together with the following defining relations (for all admissible $i, j$): $$\label{sera} c_i^2=-1, \ \ c_ic_j=-c_jc_i, \ \ \text{$w^{-1} c_i w=c_{(i)w}, \forall w\in \Sigma_r$}.$$
In this paper, we denote $\Z_i=\{0, 1, \ldots, i-1\}$. We always use $\alpha_j$ to denote the $j$-th coordinate of $\alpha\in\Z_i^r$ for $1\le j\le r$. Let $|\alpha|=\sum_{j=1}^r \alpha_j$. The Hecke-Clifford algebra $HC_r$ is free over $R$ with basis $\{c^{\alpha} w\mid w\in \Sigma_r, \alpha\in \Z_2^r\}$, where $c^\alpha=c_1^{\alpha_1}\cdots c_r^{\alpha_r}$ (see [@K]). Since $s_1, \ldots, s_{r-1}$ (resp., $c_1, \ldots, c_r$) are even (resp., odd), the even (resp., odd) subspace of $HC_r$ is spanned by $\{c^{\alpha} w\mid w\in \Sigma_r, \alpha\in \mathbb Z_2^r, |\alpha|\in 2\mathbb Z\}$ (resp., $\{c^{\alpha} w\mid w\in \Sigma_r, \alpha\in \mathbb Z_2^r, |\alpha|\not\in 2\mathbb Z\}$). In particular, the super rank of $HC_r$ is $(2^{r-1} r!, 2^{r-1} r!)$.
We need $\overline{HC}_r$ as follows. As the $R$-superalgebra, it is generated by the even elements $\bar s_1, \ldots, \bar s_{r-1}$ and odd elements $\bar c_1, \ldots, \bar c_r$ subject to the relations (for all admissible $i$ and $j$): $$\label{aseradual}\begin{aligned} & \bar s_i^2=1, \ \ \bar s_i\bar s_{i+1}\bar s_i=\bar s_{i+1}\bar s_i\bar s_{i+1}, \ \ \text{and $\bar s_i\bar s_j=\bar s_j\bar s_i$, if $|i-j|>1$, }\\
& \bar c_i^2=1, \ \ \bar c_i\bar c_j=-\bar c_j\bar c_i, \ \ \text{and $w^{-1} \bar c_i w=\bar c_{(i)w}, \forall w\in \Sigma_r$. }\\ \end{aligned}$$ In this case, we identify $\bar s_i$ with $s_i$. If $\sqrt{-1}\in R$£¬ then $HC_r$ is the $\overline{HC}_r$ by setting $\bar c_i=\sqrt{-1}c_i$ and $\bar s_i=s_i$. Let $$\label{mmpy} L_1=0, \text{ and $ L_i=\mathfrak L_i+c_i \mathfrak L_i c_i $, $2\le i\le r$,}$$ where $\mathfrak L_i=\sum_{j=1}^{i-1} (j, i)$. These elements, which are known as Jucys-Murphy elements of $HC_r$, satisfy the following relations for all admissible $i, j, k$: $$\label{jm-wba1}\begin{aligned} & L_i L_j=L_jL_i, \ \ s_i L_k=L_k s_i\ \ \text{ if $k\neq i, i+1$},\\
& s_iL_is_i=L_{i+1}-(1-c_ic_{i+1})s_i,\ \ c_iL_k=(-1)^{\delta_{i, k}} L_k c_i,\\
\end{aligned}$$ where $\delta_{i, k}=1$ if $i=k$, and $0$ otherwise. If we denote by $\bar L_1, \ldots, \bar L_r$ the Jucys-Murphy elements of $\overline {HC}_{r}$, then $\bar L_1=0$ and $\bar L_i= \bar {\mathfrak L}_i -\bar c_i \bar {\mathfrak L}_i \bar c_i$, where $\bar {\mathfrak L}_j=\sum_{k=1}^{j-1} (\bar k, \bar j)$. In this case, we identify $\bar i$ with $i$ for all $1\le i\le r$. So, $\Sigma_r$ can be identified with the symmetric group on the set $\{\bar 1, \ldots, \bar r\}$, and turns out to be $$\label{jm-wba2}\begin{aligned} &\bar L_i \bar L_j=\bar L_j\bar L_i, \ \ \bar s_i \bar L_k=\bar L_k \bar s_i\ \ \text{ if $k\neq i, i+1$},\\
& \bar s_i\bar L_i\bar s_i=\bar L_{i+1}-(1+
\bar c_i\bar c_{i+1})\bar s_i,\ \ {\bar c_i\bar L_k=(-1)^{\delta_{i, k}}} \bar L_k \bar c_i.\\
\end{aligned}$$ Considering $-L_i$ (resp., $-\bar L_i$) as abstract generators $x_i$ (resp., $\bar x_i$) yields the notion of the affine Hecke-Clifford algebra $HC_{r}^{\rm aff}$ (resp., ${\overline{HC}}{\ssc\,}_{r}^{\rm aff}$) defined as follows.
The [*affine Hecke-Clifford*]{} algebra $HC_{r}^{\rm aff}$ is the associative $R$-superalgebra generated by even elements $s_1, \ldots, s_{r-1}, x_1$ and odd elements $c_1, \ldots, c_r$ subject to –, together with the following defining relations (for all admissible $i$ and $j$): $$\label{asera} x_1x_2=x_2x_1, \ \ x_1c_i=(-1)^{\delta_{i, 1}} c_ix_1, \ \ s_jx_1=x_1s_j, \text{if $j\neq 1$, }$$ where $x_2=s_1x_1s_1-(1-c_1c_2)s_1$. Later on, we need $\overline{HC}_r^{\rm aff}$ as follows. As the $R$-superalgebra, it is generated by even elements $\bar s_1, \ldots, \bar s_{r-1}, \bar x_1$ and odd elements $\bar c_1, \ldots, \bar c_r$ subject to together with the following defining relations (for all admissible $i$ and $j$): $$\label{aseradual2}
\bar x_1\bar x_2=\bar x_2\bar x_1, \ \ \bar x_1\bar c_i=(-1)^{\delta_{i, 1}} \bar c_i\bar x_1, \ \ \bar s_j\bar x_1=\bar x_1\bar s_j, \text{if $j\neq 1$, }$$ where $\bar x_2=\bar s_1\bar x_1\bar s_1-(1+\bar c_1\bar c_2)\bar s_1$. Certainly, $\overline {HC}_r^{\rm aff}$ is $HC_r^{\rm aff}$ if $\sqrt{-1}\in R$. For $1\le i\le r$, define $$\label{relsmurp} x_i=x_i'-L_i, \text{ and $\bar x_i=\bar x_i'-\bar L_i$,}$$ where $x_{i}'=s_{i-1}\cdots s_1x_1 s_1\cdots s_{i-1}$, $x_1'=x_1$ and $\bar x_i'=\bar s_{i-1}\cdots \bar s_1 \bar x_1 \bar s_1\cdots \bar s_{i-1}$, $\bar x_1'=\bar x_1$. Then we have the following relations for all admissible $i$ and $j$: $$\label{comm-hc} \begin{aligned} & x_{i+1}=s_{i}x_{i}s_{i}-(1-c_ic_{i+1}) s_i, \text{ and $x_ix_j=x_jx_i$,}\\
& \bar x_{i+1}=\bar s_{i}\bar x_{i}\bar s_{i}-(1+\bar c_i\bar c_{i+1}) \bar s_i, \text{ and $\bar x_i\bar x_j=\bar x_j\bar x_i$.}
\\ \end{aligned}$$ For all $\alpha\in \mathbb N^r$, define $x^\alpha= x_1^{\alpha_1}\cdots x_r^{\alpha_r}$ and $\bar x^\alpha=\bar x_1^{\alpha_1}\cdots \bar x_r^{\alpha_r}$. It is proved in [@K] that $HC_r^{\rm aff}$ has basis $\{x^{\alpha}c^{\beta} w \mid w\in \mathfrak S_r, \alpha\in \mathbb N^r,
\beta\in \Z_2^r\}$. The even (resp., odd ) subspace of $HC_r^{\rm aff}$ is spanned by all $ x^\alpha c^{\beta} w$ such that $|\beta|\in 2\mathbb Z$ (resp., $|\beta|\not\in 2\mathbb Z $). Similar results hold for $\overline{HC}_r^{\rm aff}$.
We are going to recall the definition of the [*walled Brauer-Clifford superalgebra*]{} $BC_{r, t}$. This superalgebra was introduced by Jung and Kang in [@JK] so as to study the mixed tensor product of the natural module and its linear dual for the queer Lie superalgebra $\mathfrak{q}(n)$. The original $BC_{r, t}$ is defined via [*$(r, t)$-superdiagrams*]{} in [@JK]. In this paper, we use its equivalent definition.
[@JK Theorem 5.1] \[wsera\] The walled Brauer-Clifford superalgebra $ BC_{r,t}$ is the associative $R$-superalgebra generated by even generators $e_1$, $s_1, \ldots, s_{r-1}$, $\bar s_{1}, \ldots, \bar s_{t-1}$, and odd generators $c_1, \ldots, c_r, \bar c_1, \ldots, \bar c_{t}$ subject to – together with the following defining relations for all admissible $i, j$:
[2]{}
1. $e_1 c_1=e_1\bar c_1 $, $c_1 e_1=\bar c_1 e_1 $,
2. $\bar s_j c_i=c_i \bar s_j$, $ s_i \bar c_j=\bar c_j s_i$,
3. $c_i \bar c_j=-\bar c_j c_i$, $s_i\bar s_j=\bar s_j s_i$,
4. $e_1^2=0$,
5. $e_1 s_1 e_1=e_1=e_1\bar s_1 e_1$,
6. $s_i e_1=e_1 s_i$, $\bar s_i e_1=e_1 \bar s_i$, if $i\neq 1$,
7. $e_1s_1\bar s_1 e_1 s_1=e_1s_1\bar s_1 e_1 \bar s_1$,
8. $s_1 e_1s_1\bar s_1 e_1 =\bar s_1 e_1s_1\bar s_1 e_1$,
9. $c_i e_1=e_1 c_i$ and $\bar c_i e_1=e_1\bar c_i$, if $i\neq 1$,
10. $e_1c_1e_1=0=e_1\bar c_1 e_1$.
\[tau\] There is a unique $R$-linear anti-involution $\tau: BC_{r, t}\rightarrow BC_{r, t}$, which fixes all of its generators.
It follows from Definition \[wsera\], immediately.
It is known that the subalgebra of $BC_{r, t}$ generated by even generators $s_1, \ldots, s_{r-1}$, $\bar s_1, \ldots, \bar s_{t-1}$ and $e_1$ is isomorphic to the walled Brauer algebra $ B_{r, t}(0)$ in [@Koi; @Tur]. This enables us to use freely the results on $B_{r, t}(0)$ in [@RSu] so as to simplify our presentation. Write $s_{i, j}=s_i s_{i+1, j}$ if $i<j$ and $s_{i, i}=1$ and $s_{i, j}=s_{i, j+1} s_{ j}$ if $i>j$. Similarly, we have $\bar s_{i, j}$’s, etc. Following [@RSong], define $D_{r, t}^f=\{1\}$ if $f=0$ and $$\begin{aligned}
&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\label{drt} {D}_{r,t}^f=\Big\{ s_{f,i_f} \bar s_{f, j_f} \cdots s_{1,i_1}\bar s_{1,{j_1}}\,\Big|&\!\!\! k \le {j_k}\le t, 1\le k\le f,\nonumber\\
&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&\!\!\!{1 \le i_1< i_{2} < \ldots <i_f\le r} \Big\} \text{ if $0<f\le \min\{r, t\}$.}\end{aligned}$$
Define $e^0=1$ and $e^f=e_1e_2\cdots e_f$ if $0<f\le \min\{r, t\}$, where $e_i=e_{i, i}$ and $e_{i, j}=(\bar 1, \bar j) (1, i) e_1 (1, i)(\bar 1, \bar j)$ for all admissible $i, j$.
[@JK Theorem 5.1]\[wbhsa321\] The walled Brauer-Clifford superalgebra $BC_{r, t}$ has $R$-basis $$\label{bcabasis} \left\{ c^{\alpha} d_1^{-1}e^f w d_2 \bar c^{\beta}\mid 0\le f\le \min\{r, t\}, w\in \Sigma_{r-f}\times { {\Sigma}}_{\overline{t-f}}, d_1, d_2\in D_{r, t}^f, (\alpha, \beta)\in \Z_2^r\times \Z_2^t \right \}.$$ In particular, the super rank of $BC_{r, t}$ is $(2^{r+t-1} (r+t)!, 2^{r+t-1} (r+t)!)$.
The basis of $BC_{r, t}$ given in is a refinement of $X$ given in the proof of [@JK Theorem 5.1]. We remark that each $d_1^{-1}e^f w d_2$ corresponds to a unique walled Brauer diagram in [@RSu].
\[wsba-1\] For any positive integer $k$, the subalgebra of $BC_{k+r, k+t}$ generated by even elements $e_{k+1}, s_{k+1}, \ldots$, $s_{k+r-1}$, $\bar s_{k+1}, \ldots, \bar s_{k+t-1}$ and odd elements $c_{k+1}, \bar c_{k+1}$ is isomorphic to $BC_{r, t}$.
Easy exercise using Theorem \[wbhsa321\] and Definition \[wsera\].
\[spannwbc\] Let $BC_{k-1, k-1}$ be the subalgebra of $BC_{k, k}$ generated by $e_1, s_1, \ldots, s_{k-2}$, $\bar s_1,\ldots, \bar s_{k-2}$ and $c_1, \bar c_1$. Then $e_k BC_{k, k}$ is a left $BC_{k-1, k-1}$-module spanned by all $ e_k c_k^\sigma s_{k, j}\bar s_{k, l}$ such that $\sigma\in \Z_2$ and $1\le j, l\le k$.
It is enough to prove that the left $BC_{k-1, k-1}$-module $V_k$ spanned all $ e_k c_k^\sigma s_{k, j}\bar s_{k, l}$ is a right $BC_{k, k}$-module. If so, then $V_k=e_k BC_{k, k}$ by the fact that $e_k\in V_k$. We have $V_k s_i\subset V_k$ and $V_k c_1\subset V_k$ since $$e_k c_k^\sigma s_{k, j}\bar s_{k, l}s_i=\begin{cases} {s_{i} e_k c_k^\sigma s_{k, j}\bar s_{k, l}} &\text{if $i<j$,}\\
{e_k c_k^\sigma s_{k, j+1}\bar s_{k, l}} &\text{if $i=j$,}\\
{s_{i-1}e_k c_k^\sigma s_{k, j}\bar s_{k, l}} &\text{if $i>j$, }\\ \end{cases} \ \text{ and } \
e_k c_k^\sigma s_{k, j}\bar s_{k, l} c_1=\begin{cases} \epsilon c_1 e_k c_k^\sigma s_{k, j}\bar s_{k, l} &\text{if $j>1$,}\\
e_k c_k^{\sigma+1} s_{k, j}\bar s_{k, l} &\text{if $j=1$, }\\
\end{cases}$$ where $\epsilon=1$ (resp., $-1$) if $\sigma=0$ (resp., $1$). Similarly, $V_k \bar s_i\subset V_k$ and $V_k \bar c_1\subset V_k$. Finally, $V_k e_1\subset V_k$ since $$\label{step1} e_k c_k^\sigma s_{k, j}\bar s_{k, l}e_k =\begin{cases} 0 &\text{if $j=k=l$,}\\
c_{k-1}^\sigma s_{k-1, j} e_k &\text{if $l=k>j$,}\\
\bar c_{k-1}^\sigma \bar s_{k-1, l} e_k &\text{if $j=k>l$,}\\
e_{k-1} c_{k-1}^\sigma s_{k-1, j}\bar s_{k-1, l}e_k &\text{if $j, l<k$.}\\
\end{cases}$$
\[ese1\] We have $e_k BC_{k, k} e_k=e_{k} BC_{k-1, k-1}$ for all $k\ge 2$ and $e_1 BC_{1, 1} e_1=0$.
We have $e_k BC_{k, k} e_k\subseteq e_{k} BC_{k-1, k-1}$ by Lemma \[spannwbc\] and . When $k\ge 2$, the inverse inclusion follows from the equation $e_k=e_ks_{k-1}e_k$ and $e_k x=x e_k$ for any $x\in BC_{k-1,k-1}$.
\[RSu-GRSS\] For all admissible $i, j$, let $ y_i=\sum_{j=1}^{i-1} (e_{i, j}+\bar e_{i, j})-L_i$, and $\bar y_i=\sum_{j=1}^{i-1} (e_{j, i}-\bar e_{j, i})-\bar L_i$, where $\bar e_{i, j}=c_i e_{i, j} c_i$. Then $y_i={\mathfrak y_i}+c_i {\mathfrak y_i} c_i$ and $\bar y_j= {\bar {\mathfrak y}_j}-\bar c_j {\bar {\mathfrak y}_j} \bar c_j$, where $\mathfrak y_i$ (resp., ${\bar {\mathfrak y}}_j$) is $y_i$ (resp., $\bar y_j$) in [@RSu (3.5)] in the case $\delta_1=0$. So, $$\label{fraky} \mathfrak y_i=\mbox{$\sum\limits_{j=1}^{i-1}$} (e_{i, j}-(j, i)), \text{ and } \bar {\mathfrak y}_i=
\mbox{$\sum\limits_{j=1}^{i-1}$} (e_{j, i}-(\bar j, \bar i)).$$
\[usef\] With the notations above, the following results hold in $ BC_{r, t}$ $($for all admissible $i, j)$,
[2]{}
1. $s_j y_i=y_i s_j$, $\bar s_j \bar y_i=\bar y_i\bar s_j$ if $j\neq i-1, i$,
2. $s_j\bar y_i=\bar y_i s_j$, $\bar s_j y_i=y_i\bar s_j$ if $j\neq i-1$,
3. $y_i c_i=-c_i y_i$, $\bar y_i \bar c_i=-\bar c_i \bar y_i$,
4. $y_i c_j=c_j y_i$, $\bar y_i \bar c_j=\bar c_j\bar y_i$ if $i\neq j$,
5. $y_i\bar c_j=\bar c_j y_i$, $\bar y_i c_j=c_j\bar y_i$ if $j\geq i$,
6. $y_{i} y_{i+1}=y_{i+1}y_i$, $\bar y_i \bar y_{i+1}=\bar y_{i+1}\bar y_i$,
7. $y_i(e_i+\bar y_i-\bar e_i)=(e_i+\bar y_i-\bar e_i)y_i $,
8. $e_{i} \bar y_i=e_i( L_i-\bar L_i)$, $e_{i} y_i=e_i(\bar L_i-L_i)$,
9. $e_i s_i y_is_i=s_i y_is_i e_i$, $e_j \bar s_j \bar y_j\bar s_j=\bar s_j \bar y_j\bar s_j e_j$,
10. $y_i\tilde y_i =\tilde y_i y_i$, $\bar y_i {\tilde {\bar y}}_i ={\tilde {\bar y}}_i \bar y_i$,
11. $ e_i y_i^k c_i e_i=0$, $\forall k\in \mathbb N$,
12. $e_i y_i^{2n} e_i=0 $, $e_i \bar y_i^{2n} e_i=0$, $\forall n\in \mathbb N$,
13. $e_i y_i e_i=e_i \bar y_i e_i=0$,
where $\tilde y_i=s_iy_is_i-(1-c_ic_{i+1})s_i$ and ${\tilde {\bar y}}_i=\bar s_i\bar y_i\bar s_i-(1+\bar c_i\bar c_{i+1})\bar s_i$.
We assume $\sqrt{-1}\in R$. Then $BC_{r, t}\cong BC_{t, r}$. It is reasonable since we can embed $R$ into a larger integral domain containing $\sqrt{-1}$. The required isomorphism sends (a): $\sqrt{-1} c_i $ (resp., $\sqrt{-1}\bar c_j$) in $BC_{r, t}$ to $\bar c_i$ (resp., $c_j$) in $BC_{t, r}$; (b): $e_1$ to $e_1$; (c): $s_i$ (resp., $\bar s_j$) in $BC_{r, t}$ to $\bar s_i$ (resp., $s_j$) in $BC_{t, r}$. So, it suffices to verify one of equations in (1)–(6), (8)–(13) except (11).
\(1) If $j\neq i, i-1$, then $s_j c_i=c_is_j$ and $s_j\mathfrak y_i=\mathfrak y_i s_j$ by and [@RSu Lemma 3.3(6)]. So, $$s_j y_i=s_j(\mathfrak y_i+c_i\mathfrak y_i c_i)= (\mathfrak y_i+c_i\mathfrak y_i c_i)s_j=y_i s_j.$$
\(2) If $j\neq i-1$, then $s_j\bar {\mathfrak y}_i=\bar {\mathfrak y}_i s_j$ by [@RSu Lemma 3.3(7)]. By Definition \[wsera\](2), $$s_j\bar y_i=s_j(\bar{\mathfrak y}_i-\bar c_i\bar{\mathfrak y}_i\bar c_i)=s_j\bar y_i.$$
\(3) Since $c_i^2=-1$, we have $c_iy_i=c_i(\mathfrak y_i+c_i\mathfrak y_i c_i)=-\mathfrak y_i c_i+c_i \mathfrak y_i=-y_i c_i$.
\(4) By , and Definition \[wsera\](9), $c_ic_j=-c_jc_i$, $c_j L_i=L_ic_j$ and $e_{i, k}c_j=c_je_{i, k}$ if $i\neq j$. By Definition \[RSu-GRSS\], we have $y_ic_j=c_jy_i$.
\(5) If $j>k$, then $e_{i, k} \bar c_j=\bar c_j e_{i,k}$. By Definition \[wsera\](2)–(3) and Definition \[RSu-GRSS\], we have $y_i \bar c_j=\bar c_j y_i$.
\(6) Since $\mathfrak y_i=\sum_{j=1}^{i-1} (e_{i, j}-(j, i))$ (see ), we have $\mathfrak y_i c_{i+1}=c_{i+1}\mathfrak y_i$. By [@RSu Lemma 3.3(9)], $\mathfrak y_i\mathfrak y_{i+1}=\mathfrak y_{i+1}\mathfrak y_{i}$. Thus, $$\begin{aligned} y_i y_{i+1} &=(\mathfrak y_i+c_i\mathfrak y_ic_i) y_{i+1}\overset{(4)}=\mathfrak y_i y_{i+1}+c_i\mathfrak y_i y_{i+1}c_i\\
&= \mathfrak y_i\mathfrak y_{i+1}+c_{i+1}\mathfrak y_i\mathfrak y_{i+1}c_{i+1}+c_i \mathfrak y_i \mathfrak y_{i+1} c_i+c_ic_{i+1} \mathfrak y_i\mathfrak y_{i+1} c_{i+1} c_i.\\ \end{aligned}$$ Applying the anti-involution $\tau$ on the above equation yields $ y_i y_{i+1}=y_{i+1} y_i$.
\(7) We have $$\begin{aligned} y_i (e_i-\bar e_i+\bar y_i)&=(\mathfrak y_i+c_i\mathfrak y_i c_i) (e_i+{\bar {\mathfrak y}}_i-\bar c_i {\bar {\mathfrak y}}_i \bar c_i-c_i e_ic_i)\\
& =\mathfrak y_i(e_i+{\bar {\mathfrak y}}_i)-\bar c_i \mathfrak y_i (e_i+{\bar {\mathfrak y}}_i)\bar c_i +c_i \mathfrak y_i (e_i+{\bar {\mathfrak y}}_i) c_i-
\bar c_i c_i \mathfrak y_i (e_i+{\bar {\mathfrak y}}_i) c_i\bar c_i. \\
\end{aligned}$$ Applying the anti-involution $\tau$ on the above equation and using $(e_i+{\bar {\mathfrak y}}_i )\mathfrak y_i = \mathfrak y_i (e_i+{\bar {\mathfrak y}}_i )$ (see [@RSu Lemma 3.3(4)]) yields (7).
\(8) By [@RSu Lemma 3.3(1)], $e_i\mathfrak y_i=e_{i} (-\mathfrak L_i+{\bar {\mathfrak L}}_i)$. So, $$\begin{aligned} e_iy_i &=e_i\mathfrak y_i+e_i c_i\mathfrak y_i c_i=e_i\mathfrak y_i(1+\bar c_i c_i)=-e_i(\mathfrak L_i-\bar{\mathfrak L}_i)(1+\bar c_i c_i)=-e_i\mathfrak L_i+e_i{\bar {\mathfrak L}}_i
-e_i(\mathfrak L_i-{\bar {\mathfrak L}}_i)\bar c_i c_i\\ & =-e_i (\mathfrak L_i+ c_i \mathfrak L_i c_i)+e_i(\bar {\mathfrak L}_i-
\bar c_i \bar {\mathfrak L}_i \bar c_i )=-e_i
(L_i-\bar L_i).\\ \end{aligned}$$
\(9) By [@RSu Lamma 3.3(5)], $e_i s_i\mathfrak y_i s_i= s_i\mathfrak y_i s_i e_i$. So, $$e_i s_i y_i s_i=e_i s_i c_i\mathfrak y_i c_i s_i+ e_i s_i \mathfrak y_i s_i
=c_{i+1} e_i s_i\mathfrak y_i s_i c_{i+1}+ s_i \mathfrak y_i s_i e_i=s_i y_i s_i e_i.$$
\(10) We define $\textbf{m}_i=s_i\mathfrak y_i s_i-s_i$. By [@RSu Lemma 3.3(5)]), $\textbf{m}_i \mathfrak y_i=\mathfrak y_i \textbf{m}_i $. So, $$\label{yy123} \begin{aligned}& y_i\tilde y_i =(\mathfrak y_i+c_i\mathfrak y_ic_i) (s_i(\mathfrak y_i+c_i\mathfrak y_ic_i)s_i-(1-c_ic_{i+1})s_i)\\
=&\mathfrak y_i \textbf{m}_i + \mathfrak y_i (s_ic_i \mathfrak y_i c_i s_i+c_ic_{i+1}s_i)+c_i\mathfrak y_ic_i (s_i (\mathfrak y_i+c_i\mathfrak y_i c_i) s_i-(1-c_ic_{i+1})s_i)\\
=&\mathfrak y_i \textbf{m}_i +c_{i+1} \mathfrak y_i \textbf{m}_i c_{i+1}+c_i\mathfrak y_i c_i \textbf{m}_i +
c_i\mathfrak y_i c_i(s_ic_i\mathfrak y_ic_is_i +c_ic_{i+1}s_i )\\
=& \mathfrak y_i \textbf{m}_i +c_{i+1}\mathfrak y_i \textbf{m}_i c_{i+1}+c_i\mathfrak y_ic_is_i\mathfrak y_is_i-
c_i\mathfrak y_i c_i s_i -c_i\mathfrak y_ic_{i+1}s_i+c_i\mathfrak y_i c_i s_i c_i \mathfrak y_i c_i s_i\\
=& \mathfrak y_i \textbf{m}_i +c_{i+1}\mathfrak y_i \textbf{m}_i c_{i+1}+ c_i\mathfrak y_i \textbf{m}_i c_i -c_i\mathfrak y_ic_{i}s_i-c_ic_{i+1}
\mathfrak y_is_i\mathfrak y_i s_i c_ic_{i+1}\\
=&\mathfrak y_i \textbf{m}_i +c_{i+1}\mathfrak y_i \textbf{m}_i c_{i+1}+ c_i\mathfrak y_i \textbf{m}_i c_i+c_{i+1}c_i\mathfrak y_i \textbf{m}_i c_ic_{i+1}.
\end{aligned}$$ Applying the anti-involution $\tau$ on yields $\tilde y_i y_i= y_i\tilde y_i$.
\(11) If $k$ is odd, then $$e_i y_i^k c_i e_i=e_i y_i^k \bar c_i e_i\overset{(5)}=e_i\bar c_i y_i^k e_i\overset{(3)} =-e_i y_i^k c_i e_i,$$ forcing $ e_i y_i^k c_i e_i=0$. If $k$ is even, then $$\label{eybary} e_{i} y_i^{k-1} \bar y_i=e_{i} y_{i}^{k-1}(\bar y_i+e_i-\bar e_i)-e_iy_i^{k-1} e_i+e_iy_i^{k-1}\bar e_i
\overset{(7)}=e_{i}(\bar y_i+e_i-\bar e_i)y_{i}^{k-1}-e_iy_i^{k-1} e_i+e_iy_i^{k-1}\bar e_i.$$ So, $e_{i} y_i^{k-1} \bar y_i c_ie_i=e_i \bar y_i y_i^{k-1} c_i e_i$. On the other hand, $e_{i} y_i^{k-1} \bar y_i c_ie_i=-e_i c_i
y_i^{k-1}\bar y_i e_i=e_{i} c_i y_i^k e_i $ and $e_i \bar y_i y_i^{k-1} c_i e_i=-e_i y_i^k c_ie_i=-e_{i} c_i y_i^k e_i$. So, $ e_i y_i^k c_i e_i=0$ for even $k$.
\(12) We have $e_k y_k^{2n} e_k=0$ since $$e_k y_k^{2n} e_k =e_k y_k^{2n} \bar c_k^2 e_k\overset{(5)}=e_k \bar c_k y_k^{2n} c_k e_k
= e_k c_k y_k^{2n}c_k e_k\overset{(3)}= e_k y_k^{2n} c_k^2 e_k=-e_k y_k^{2n} e_k.$$
\(13) If $j<i$, then $e_i c_i e_{i, j} c_i e_i=e_i \bar c_i e_{i, j} \bar c_i e_i=e_{i} e_{i, j} e_i=e_i$ and $e_ic_i(j, i)c_ie_i=e_i \bar c_i (j, i) \bar c_i e_i=e_i$. So, $$e_i y_i e_i=2 e_i\mathfrak y_i e_i=2\mbox{$\sum\limits_{j=1}^{i-1}$} e_i(e_{i, j}-(i, j))e_i=0.$$
\[equal123\] There is a unique element $\omega_{a, k}\in BC_{k-1, k-1}$ such that $e_k y_k^a e_k=\omega_{a, k} e_k$. Similarly, there is a unique element $\bar \omega_{a, k}\in BC_{k-1, k-1}$ such that $e_k {\bar y_k}^a e_k=\bar \omega_{a, k} e_k$. Moreover, $\omega_{2n,k}=\bar \omega_{2n,k}=0$.
The existence of an $\omega_{a, k}$ follows from Proposition \[ese1\] and the uniqueness of such an element follows from Theorem \[wbhsa321\]. Finally, we have $\omega_{2n,k}=\bar \omega_{2n,k}=0$ by Lemma \[usef\](12).
\[y1\] For $n\in \mathbb N$, $e_i \bar y_i^{2n+1}\!=\!\sum_{j=0}^n a^{(i)}_{2n+1, j} e_i y_i^{2j+1}$ for some $a_{2n+1,j}^{(i)}\in$ $
R[\omega_{3, i}, \ldots,\omega_{2n-1, i} ]$ such that
1. $a_{2n+1, n}^{(i)}=-1$,
2. $a_{2n+1, j}^{(i)}=a_{2n-1, j-1}^{(i)}$, $1\le j\le n$,
3. $a^{(i)}_{2n+1, 0}=\sum_{j=1}^{n-1} a^{(i)}_{2n-1, j} \omega_{2j+1, i}$.
When $n=0$, we have $e_i(y_i+\bar y_i)=0$ by Lemma \[usef\](8). So, $a_{1, 0}^{(i)}=-1$. In general, we have $$\begin{aligned} e_i y_i^{2j-1} \bar y_i^2& = e_i y_i^{2j-1} ( \bar y_i+e_i-\bar e_i) \bar y_i-e_i y_i^{2j-1} e_i \bar y_i\ \ (\text{by Lemma~\ref{usef}(11)})\\
& =e_i (\bar y_i+e_i-\bar e_i) y_i^{2j-1} \bar y_i +\omega_{2j-1,i} e_i y_i \ \ (\text{by Lemma~\ref{usef}(7)})
\\ & =-e_i y_i^{2j} \bar y_i+\omega_{2j-1,i} e_i y_i\ \ (\text{ by Lemma~\ref{usef}(8)}). \end{aligned}$$ Similarly, using $\omega_{2j, i}=0$ yields $e_i y_i^{2j} \bar y_i = -e_i y_i^{2j+1}$. So, $e_i y_i^{2j-1} \bar y_i^2=e_i y_i^{2j+1}+w_{2j-1,i} e_i y_i$. By inductive assumption on $n$ and $\omega_{1, i}=0$, we have the result, immediately.
\[y2\] For positive integers $n$, $e_i \bar y_i^{2n}\!=\!\sum_{j=0}^n a^{(i)}_{2n, j} e_i y_i^{2j}$ for some $a_{2n,j}^{(i)}\!\in\!
R[\omega_{3, i}, \ldots,\omega_{2n-1, i} ]$ such that
1. $a_{2n, n}^{(i)}=1$,
2. $a_{2n, j}^{(i)}=a_{2n-2, j-1}^{(i)}$, $1\le j\le n$,
3. $a^{(i)}_{2n, 0}=\sum_{j=1}^{n-1} a^{(i)}_{2n-2, j} \omega_{2j+1, i}$.
We have $e_i y_i^{2j} \bar y_i^2=-e_iy^{2j+1} \bar y_i=e_iy_i^{2j+2}+\omega_{2j+1, i} e_1$. By inductive assumption on $n$ and $\omega _{1,i}=0$, we immediately have the result.
We can assume $k \ge 2$ (resp., $n\ge 2$) in the Lemma \[free1\] since $y_1=\bar y_1=0$ (resp., $\omega_{1, k}=\bar \omega_{1, k}=0$ by Lemma \[usef\](13)).
\[free1\] We have $\bar \omega_{2n-1, k}\in $ $ R[\omega_{3, k}, \ldots, \omega_{2n-1, k}]$ if $k, n\in\Z^{\ge2}$. Furthermore, both $\omega_{2n-1, k}$ and $\bar\omega_{2n-1, k}$ are central in $BC_{k-1, k-1}$.
By Lemma \[y1\] and inductive assumption on $k$, we have the first statement. To prove the second, note that any $h\in \{e_1, s_1, \ldots, s_{k-2}, c_1\}$ commutes with $e_k, y_k$. So, $e_k (h \omega_{2n-1, k})=e_k(\omega_{2n-1, k} h)$. By Theorem \[wbhsa321\], $h \omega_{2n-1, k}=\omega_{2n-1, k} h$. Finally, we need to check that $e_k (h \omega_{2n-1, k})=e_k(\omega_{2n-1, k} h)$ for any $h\in \{\bar s_1, \ldots, \bar s_{k-2}, \bar c_1\}$. In this case, we use Lemma \[y1\]. More explicitly, we can use $\bar y_k$ instead of $y_k$ in $e_k y_k^{2n-1} e_k$. Thus $h \omega_{2n-1, k}=\omega_{2n-1, k} h$, as required.
In the following, we define $$\label{hhk} h_k =y_k+e_k+\bar e_k, \text{ and } \bar h_k=\bar y_k+e_k-\bar e_k, \text{ for all admissible $k$.}$$
\[syk\] For $k,a\in\Z^{\ge 1}$, we have $$s_k y_{k+1}^a=h_k^a s_k-\mbox{$\sum\limits_{b=0}^{a-1}$} h_k ^{a-1-b}y_{k+1}^b+\mbox{$\sum\limits_{b=0}^{a-1}$} (-1)^{a-b} c_kc_{k+1} h_k^{a-b-1} y_{k+1}^b,$$ where $h_k$ is given in .
It is easy to verify the result by induction on $a$.
\[dzjk\] Suppose $1\le j\le k-1$. Define $ z_{j, k}=s_{j, k-1} h_{k-1} s_{k-1, j}$, and $ \bar z_{j, k}=\bar s_{j, k-1} \bar h_{k-1} \bar s_{k-1, j}$, where $h_{k-1}$ and $\bar h_{k-1}$ are given in . Then
1. $ z_{j, k}=\sum_{\ell=1}^{k-1} e_{j, \ell}-\sum_{1\le s\le k-1, s\neq j} (s, j)+c_j \left( \sum_{\ell=1}^{k-1} e_{j, \ell}-\sum_{1\le s\le k-1, s\neq j} (s, j)\right) c_j$,
2. $\bar z_{j, k}=\sum_{\ell=1}^{k-1} e_{ \ell, j}-\sum_{1\le s\le k-1, s\neq j} (\bar s, \bar j)-
\bar c_j \left(\sum_{\ell=1}^{k-1} e_{ \ell, j}-\sum_{1\le s\le k-1, s\neq j} (\bar s, \bar j)\right) \bar c_j$.
Easy exercise.
Note that $\omega_{0,k}=0$ and $\omega_{1, k}=\bar \omega_{1, k}=0$ (see Lemma \[usef\](13)), and $e_k h=0$ for $h\in BC_{k-1, k-1}$ if and only if $h=0$. We will use these facts frequently in the proof of the following lemma, where we use the terminology that a monomial in $z_{j, k+1}$’s and $\bar z_{j, k+1}$’s is a [*leading term*]{} in an expression if it has the highest degree by defining ${\rm deg\,}z_{i,j}={\rm deg\,}\bar z_{i,j}=1$.
\[omed\] For any positive integer $n$, $\omega_{2n+1, k+1}$ can be written as an $R$-linear combination of monomials in $ z_{j, k+1}$’s and $\bar z_{j, k+1}$’s for $1\le j\le k$ such that the leading terms of $\omega_{2n+1, k+1}$ are $2\sum_{j=1}^{k}(-z_{j, k+1}^{2n}+\bar z_{ j, k+1}^{2n})$.
By Corollary \[equal123\] and Lemma \[usef\](8), we have $$\label{ommmm}
\omega_{2n+1,k+1}e_{k+1}=e_{k+1} y_{k+1}^{2n+1} e_{k+1}=e_{k+1} (\bar L_{k+1}- L_{k+1}) y_{k+1}^{2n}e_{k+1}.$$ Note that $(j,k+1)=s_{j,k}s_ks_{k,j}$ and $s_{j,k}, s_{k,j}$ commute with $y_{k+1},e_{k+1}, c_{k+1}$ (see Lemma \[usef\](1) and ). Considering the right-hand side of and expressing $L_{k+1}$ by , we see that a term of $-e_{k+1} L_{k+1} y_{k+1}^{2n}e_{k+1}$ becomes $$\begin{aligned}{\sc\!}&-s_{j, k} e_{k+1} (s_{k} y_{k+1}^{2n}\!+
c_{k+1} s_{k} c_{k+1}y_{k+1}^{2n}) e_{k+1} s_{ k, j}=-2s_{j, k} e_{k+1} (s_k y_{k+1}^{2n}) e_{k+1}s_{k, j}\\
& =-2s_{j, k} e_{k+1}\left \{h_k^{2n} s_k -\mbox{$\sum\limits_{b=0}^{2n-1}$} h_k ^{2n-b-1} y_{k+1}^b
+ \mbox{$\sum\limits_{b=0}^{2n-1}$}(-1)^{2n-b} c_k c_{k+1} h_k^{2n-b-1} y_{k+1}^b\right\} e_{k+1} s_{k, j} \\
&=-2\!s_{j, k} h_k^{2n}e_{k+1}s_ke_{k+1}s_{k, j}+2s_{j, k}\mbox{$\sum\limits_{b=0}^{2n-1}$} h_k^{2n-b-1} e_{k+1} y_{k+1}^b e_{k+1} s_{k, j} \ \ (\text{by Lemma~\ref{usef}(11)})
\\ &=-2\!s_{j, k}e_{k+1}\Big( h_k^{2n}-\mbox{$\sum\limits_{b=0}^{2n-1}$} h_k^{2n-b-1} \omega_{b, k+1}\Big)s_{k, j}.\\ \end{aligned}$$ By inductive assumption, the right-hand side of the above equation can be written as an $R$-linear combination of monomials with the required form such that the leading term is $-2z_{j, k+1}^{2n}$. Finally, we consider terms in concerning $\bar L_{k+1}$, namely we need to deal with $$e_{k+1} (\bar j, \overline{ k\!+\!1}) y_{k+1}^{2n} e_{k+1}- e_{k+1} \bar c_{k+1} (\bar j, \overline{ k\!+\!1})\bar c_{k+1} y_{k+1}^{2n} e_{k+1}=2 e_{k+1}(\bar j, \overline{ k\!+\!1}) y_{k+1}^{2n} e_{k+1} .$$ Applying $\tau$ on $ e_{k+1}\bar y_{k+1}^{2n}$ and using Lemma \[y2\] and inductive assumption on $n$, we can use $ \bar y_{k+1}^{2n} e_{k+1}$ to replace $y_{k+1}^{2n} e_{k+1}$ in $e_{k+1} (\bar j, \overline{k\!+\!1}) y_{k+1}^{2n} e_{k+1}$ (by forgetting lower terms). This enables us to consider $ e_{k+1} (\overline j,
\overline {k\!+\!1})\bar y_{k+1}^{2n} e_{k+1}$ instead. As above, this term can be written as the required form with leading term $ 2\bar z_{j, k+1}^{2n}$. The proof is completed.
\[commomega\]For $a\!\in\!\Z^{\ge0},{\ssc\,}k\!\in\!\Z^{\ge 1}$, both $\omega_{a, k+1}$ and $\bar \omega_{a, k+1}$ commute with $y_{k+1}$, $\bar y_{k+1}, c_{l}, \bar c_{l}$, $l\ge k+1$.
Since $\omega_{a, 1}=\bar \omega_{a, 1}=\omega_{1, k}=\bar \omega_{1, k}=\omega_{2a, k}=\bar\omega_{2a, k}=0$ for all admissible $a, k$, we can assume $a, k\in \mathbb Z^{\ge 2}$ and $2\nmid a$. In order to verify that $\omega_{a, k+1}$ and $\bar \omega_{a, k+1}$ commute with $y_{k+1}$, by Lemmas \[free1\], \[dzjk\]–\[omed\], it suffices to prove that $y_{k+1}$ commutes with both $z_{j, k+1}$ and $\bar z_{j, k+1}$ for $1\le j\le k$. By Lemma \[dzjk\], $z_{j, k+1}=\mathfrak z_{j, k+1}+c_j \mathfrak z_{j, k+1} c_j$, where $$\mathfrak z_{j, k+1}=\mbox{$\sum\limits_{\ell=1}^k$} e_{j, \ell}-\mbox{$\sum\limits_{1\le s\le k, s\neq j}$} (s, j),$$ which is $z_{j, k+1} $ in [@RSu Lemma 3.9]. Obviously, $c_{k+1} \mathfrak z_{j, k+1}=\mathfrak z_{j, k+1} c_{k+1}$. By Lemma \[usef\](4), we have $y_{k+1}c_j=c_j y_{k+1}$. So, $$\label{zy123} \begin{aligned} z_{j,k+1}y_{k+1}&
=\mathfrak z_{j, k+1} (c_{k+1} \mathfrak y_{k+1}c_{k+1}+\mathfrak y_{k+1})+c_j \mathfrak z_{j, k+1}(c_{k+1} \mathfrak y_{k+1}c_{k+1}+\mathfrak y_{k+1})c_j\\
&=c_{k+1} \mathfrak z_{j, k+1}\mathfrak y_{k+1} c_{k+1}+\mathfrak z_{j, k+1}\mathfrak y_{k+1}+c_j \mathfrak z_{j, k+1} \mathfrak y_{k+1} c_j+c_jc_{k+1} \mathfrak z_{j, k+1} \mathfrak y_{k+1}c_{k+1}c_j. \\
\end{aligned}$$ Recall that $\tau$ is the $R$-linear anti-involution in Lemma \[tau\]. By Definition \[RSu-GRSS\] and Lemma \[dzjk\], $\tau$ fixes both $z_{j, k+1}$ and $y_{k+1}$. So, $y_{k+1} z_{j,k+1}=\tau(z_{j,k+1}y_{k+1})$. Since $\mathfrak y_{k+1} \mathfrak z_{j, k+1} =\mathfrak z_{j, k+1} \mathfrak y_{k+1} $ (see [@RSu Lemma 3.11]), we have $z_{j,k+1}y_{k+1}=y_{k+1} z_{j,k+1}$ by . One can check $\bar z_{j, k+1} y_{k+1}=y_{k+1} \bar z_{j, k+1}$ similarly via Definition \[RSu-GRSS\] and the equation $\bar z_{j,k+1}=\bar {\mathfrak z}_{j, k+1}-\bar c_j \bar {\mathfrak z}_{j, k+1} \bar c_j$. This proves that $y_{k+1}$ commutes with $\omega_{a, k+1}$ and $\bar \omega_{a, k+1}$. We remark that one can check both $\omega_{a, k+1}$ and $\bar \omega_{a, k+1}$ commute with $\bar y_{k+1}$, similarly. By Lemma \[dzjk\], one can easily check that $z_{j, k+1}$ and $\bar z_{j, k+1}$ commute with $c_l$ and $\bar c_l$ for all $l\ge k+1$.
Affine walled Brauer-Clifford superalgebras
===========================================
In this section, we assume that $R$ is an integral domain containing $\omega_1$ and $2^{-1}$. Motivated by Definition \[wsera\] and Lemma \[usef\], we introduce the notion of affine walled Brauer-Clifford superalgebra over $R$ as follows.
\[awbsa\] The [*affine walled Brauer-Clifford superalgebra*]{} $ BC_{r,t}^{\text{\rm aff}}$ is the associative $R$-superalgebra generated by odd elements $c_1, \ldots, c_r$, $\bar c_1, \ldots, \bar c_t$ and even elements $e_1, x_1, \bar x_1$, $s_1, \ldots, s_{r-1}$, $\bar s_1, \ldots, \bar s_{t-1}$, and two families of even central elements $\omega_{2k+1}, \bar \omega_{k}$, $ k\in\Z^{\ge1}$ subject to –, , and Definition \[wsera\](1)–(10) together with the following defining relations for all admissible $i$:
[2]{}
1. $e_1(x_1+\bar x_1)=(x_1+\bar x_1)e_1=0$,
2. $e_1s_1x_1s_1=s_1x_1s_1e_1$,
3. $x_1(e_1+\bar x_1-\bar e_1 )=(e_1-\bar e_1+\bar x_1)x_1$,
4. $e_1\bar s_1\bar x_1\bar s_1=\bar s_1\bar x_1\bar s_1 e_1$,
5. $e_1x_1^{2k+1} e_1=\omega_{2k+1} e_1$, $\forall k\in \mathbb N$,
6. $e_1x_1^{2k}e_1=0$, $\forall k\in \mathbb N$,
7. $e_1\bar x_1^{k}e_1=\bar \omega_{k}e_1$, $\forall k\in \mathbb Z^{>0}$,
8. $x_1 \bar c_{i}=\bar c_{i} x_1$,
9. $\bar x_1 c_i= c_i \bar x_1$,
10. $ x_1 \bar s_i=\bar s_i x_1$,
11. $\bar x_1 s_i=s_i \bar x_1$.
For the simplification of presentation, we set $\omega_{2k}=0$, $\forall k\in \mathbb N$. The following result follows from Definition \[awbsa\], immediately.
\[antiaff\] There is an $R$-linear anti-involution $\sigma: BC_{r,t}^{\rm aff}\rightarrow BC_{r,t}^{\rm aff}$, which fixes all generators of $BC_{r, t}^{\rm aff}$ in Definition $\ref{awbsa}$.
Lemmas \[assum\] and \[assum1\] can be proven by arguments similar to those for Lemmas \[y1\] and \[y2\].
\[assum\] For any $n\in \mathbb N$, $e_1\bar x_1^{2n+1}=\sum_{j=0}^n a_{2n+1, j} e_1 x_1^{2j+1}$ for some $a_{2n+1, j}\in BC_{r, t}^{\rm aff}$ such that
1. $a_{2n+1, n}=-1$,
2. $a_{2n+1, j}=a_{2n-1, j-1}$ for all $1\le j\le n-1$,
3. $a_{2n+1, 0}=\sum_{j=0}^{n-1} a_{2n-1, j}\omega_{2j+1}$.
In particular, $a_{2n+1,j}\in R[ \omega_3, \ldots, \omega_{2n-1}]$, for all $0\le j\le n$.
\[assum1\] For any positive integer $n$, $e_1\bar x_1^{2n}=\sum_{j=0}^n a_{2n, j} e_1 x_1^{2j}$ for some $a_{2n,j}\in BC_{r,t}^{\rm aff }$ such that
1. $a_{2n, n}=1$,
2. $a_{2n, j}=a_{2n-1, j-1}$ for all $1\le j\le n-1$,
3. $a_{2n, 0}=\sum_{j=0}^{n-1} a_{2n-2, j}\omega_{2j+1}$.
In particular, $a_{2n, j}\in R[ \omega_3, \ldots, \omega_{2n-1}]$, for all $0\le j\le n$.
\[assump1\] If $e_1$ is $R[\omega_3, \omega_5, \ldots, \bar \omega_1, \bar \omega_2,\ldots]$-torsion-free, then $\bar \omega_{2n+1} =\sum_{i=0}^{n} a_{2n+1, i} \omega_{2i+1}$ and $\bar \omega_{2n}=0$ for all $n\in \mathbb N $. In particular, $\bar \omega_1=-\omega_1$.
By Definition \[awbsa\](1), $(\omega_1+\bar \omega_1)e_1=0$. If $e_1$ is $R[\omega_3, \omega_5, \ldots, \bar \omega_1, \bar \omega_2,\ldots]$-torsion-free, $\bar \omega_1=- \omega_1$. In general, by Lemma \[assum\], $e_1\bar x_1^{2n+1}e_1=\sum_{j=0}^n a_{2n+1, j} e_1 x_1^{2j+1}e_1$. So, $\bar \omega_{2n+1}=\sum_{j=0}^n a_{2n+1, j} \omega_{2j+1}$. Similarly, by Lemma \[assum1\] and Definition \[awbsa\](6), $\bar \omega_{2n}=\sum_{j=0}^n a_{2n, j} \omega_{2j}=0$.
\[admis-para\]From here onwards, we always assume that $\bar \omega_{2n}=0$ and $\bar \omega_{2n+1}$’s are given in Corollary \[assump1\]. Otherwise, we would have $e_1=0$ provided that $R$ is a field, in which case, $BC_{r, t}^{\rm aff}$ turns out to be $HC_{r}^{\rm aff}\boxtimes {\bar {HC}_t}^{\rm aff}$, the outer tensor product of two affine Hecke-Clifford superalgebras! We remark that the tensor product is that for superalgebra in the sense that $$(x\boxtimes y ) (x_1\boxtimes y_1)=(-1)^{[y][x_1]} xx_1\boxtimes y y_1,$$ for any homogenous elements $x, x_1\in HC_{r}^{\rm aff} $ and $y, y_1\in \overline{HC}_{t}^{\rm aff} $, where $[x]$, called the parity of $x$, is $1$ (resp., $0$) if $x$ is odd (resp., even).
\[level-1\]For any $k\in \mathbb Z^{>0}$, there is a superalgebra homomorphism $\Phi_k: BC_{r, t}^{\rm aff} \rightarrow BC_{r+k, t+k}$ sending $s_i, \bar s_j, e_1, x_1, \bar x_1, c_l, \bar c_m, \omega_a, \bar \omega_a$ to $s_{k+i}, \bar s_{k+j}, e_{k+1}, y_{k+1}, \bar y_{k+1}, c_{k+l}, \bar c_{k+m}, \omega_{a, k+1}, \bar \omega_{a, k+1}$ respectively for all admissible $a, i, j, l, m$’s.
It is enough to verify the images of generators of $BC_{r, t}^{\rm aff}$ satisfy the defining relations for $BC_{r, t}^{\rm aff}$ in Definition \[awbsa\]. We say $\Phi_k$ satisfies the relation if the images of generators satisfy this relation.
By Lemmas \[dzjk\]–\[commomega\], the images of $\omega_a$ and $\bar \omega_a$ commute with the images of other generators. By Corollary \[wsba-1\], $\Phi_k$ satisfies – and Definition \[wsera\](1)–(10). $\Phi_k$ satisfies and by Lemma \[usef\](1), (3), (4), (10). Further, $\Phi_k$ satisfies Definition \[awbsa\](1)–(4) by Lemma \[usef\](7)–(9). In this case, we need $(y_{k+1}+\bar y_{k+1})e_{k+1}=0$, which can be obtained by applying the anti-involution $\tau$ on Lemma \[usef\](8). $\Phi_k$ satisfies Definition \[awbsa\](5)–(7) by Corollary \[equal123\] and Lemma \[usef\](11)–(13). Finally, $\Phi_k$ satisfies Definition \[awbsa\](8)–(11) by Lemma \[usef\](2), (5).
In [@RSu], two of the authors proved the freeness of the affine walled Brauer algebra via bases of infinite many walled Brauer algebras. The key point is the existence of infinite many homomorphisms between the affine walled Brauer algebra and walled Brauer algebras [@RSu Theorem 3.12]. In the current case, Theorem \[level-1\] is the counterpart of [@RSu Theorem 3.12]. Since $\omega_{1, k}=0$ for all $k$, what we can do is to use Theorem \[level-1\] to prove the freeness of affine walled Brauer-Clifford superalgebras with parameters $\omega_1=0$. However, many affine walled Brauer-Clifford algebras which appear in the higher mixed Schur-Weyl-Sergeev dualities have non-zero parameter $\omega_1$. For details, see section 4. For this reason, we use level two walled Brauer-Clifford superalgebras (with special parameters) instead of walled Brauer-Clifford superalgebras later on. This is one of the points which is different from the work in [@RSu].
In $BC_{r, t}^{\rm aff}$, we define $x_i, x_i', \bar x_j$ and $\bar x_j'$ as in for all admissible $i$ and $j$.
\[hecrel\] We have the following results for admissible $i$ and $j$:
[2]{}
1. $x_i c_i=-c_i x_i$ and $\bar x_i \bar c_i=-\bar c_i \bar x_i$,
2. $x_i c_j=c_j x_i$, $\bar x_i \bar c_j=\bar c_j \bar x_i$ if $i\neq j$,
3. $x_i\bar c_j=\bar c_j x_i$ and $\bar x_i c_j=c_j \bar x_i$.
\(1) and (2) follow from , , and . (3) follows from Definition \[awbsa\](8)–(11).
\[hecrel2\] We have the following results for all admissible $i, j$:
[2]{}
1. $x_i'(\bar x_j'+e_{i, j}-\bar e_{i, j})=(\bar x_j'+e_{i, j}-\bar e_{i, j}) x_i'$,
2. $\bar x_j'( x_i'+e_{i, j}+\bar e_{i, j})=( x_i'+e_{i, j}+\bar e_{i, j})\bar x_j'$,
3. $e_{i, j}(x_i'+\bar x_j')=0$,
4. $(x_i'+\bar x_j')e_{i, j}=0$,
5. $e_i{x_i'}^kc_ie_i=0$,$\forall k\in \mathbb N$,
6. $e_i{\bar x_i'}^kc_ie_i=0$,$\forall k\in \mathbb N$.
Multiplying $(1, i) (\bar 1, \bar j)$ on both sides of Definition \[awbsa\](3) yields (1). By Definition \[awbsa\](1),(3), we know that $\bar x_1(x_1+e_1+\bar e_1)=(x_1+e_1+\bar e_1)\bar x_1$. So (2) can be proved similarly. Multiplying $(1, i)(\bar 1, \bar j)$ on both side of Definition \[awbsa\](1) yields (3) and (4). We have $c_i x_i'=-x_i'c_i$ (resp., $\bar c_i x_i'=x_i' \bar c_i$) by and Lemma \[hecrel\](1) (resp., Definition \[awbsa\](8)–(11)). Also, (1) is the counterpart of Lemma \[usef\](7). So, (5) and (6) can be verified by arguments similar to those in the proof of Lemma \[usef\](11). We leave the details to the reader.
\[relxpri\] We have the following results for all admissible $i,j, k, l$:
[2]{}
1. $e_{i, k}x_j'=x_j'e_{i, k}$, if $i\neq j$,
2. $e_{i, k}\bar x_l'=\bar x_l' e_{i, k}$, if $k\neq l$,
3. $e_{i, j}(x_i')^a e_{i, j}=\omega_a e_{i, j}$, $\forall a\in \mathbb N$,
4. $e_{i, j}(\bar x_j')^a e_{i, j}=\bar \omega_a e_{i, j}, \forall a\in \mathbb N$.
We have $e_1x_2'=x_2'e_1$ by Definition \[awbsa\](2). Multiplying $(2, j)$ on both sides of the equation yields $e_1 x_j'=x_j' e_1$. Since $i\neq j$, multiplying $(1,i)(\bar 1, \bar k)$ on both sides of $e_1 x_j'=x_j' e_1$ yields (1). (2) can be verified similarly. (3) and (4) follow from Definition \[awbsa\](5)–(7).
We consider $\ BC_{r,t}^{\text{\rm aff}}$ as a filtrated superalgebra by setting $$\text{deg}{ s_i}=\text{deg}{ \bar s_j}=\text{deg}{ e_1}=\text{deg}{ c_n}=\text{deg}{ \bar c_m}=\text{deg}{ \omega_a}=\text{deg} {\bar \omega_a}=0 \text{ and } \
\text{deg} {x_k}=\text{deg}{ \bar x_\ell}=1,$$ for all admissible $a, i, j, k, \ell, m, n$. Let $(BC_{r,t}^{\rm aff})^{(k)}$ be the super $R$-submodule spanned by monomials with degrees less than or equal to $k$ for $k\in{\mathbb{Z}}^{\ge0}$. Then we have the following filtration $$\label{filtr}
BC_{r,t}^{\text{\rm aff}}\supset\ldots\supset (BC_{r,t}^{\text{\rm aff}})^{(1)}\supset(BC_{r,t}^{\text{\rm aff}})^{(0)}\supset (BC_{r,t}^{\text{\rm aff}})^{(-1)}=0.$$ Let ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})\!=\!\oplus_{i\ge0}( BC_{r,t}^{\text{\rm aff}})^{[i]}$, where $( BC_{r,t}^{\text{\rm aff}})^{[i]}\!=\!( BC_{r,t}^{\text{\rm aff}})^{(i)}/( BC_{r,t}^{\text{\rm aff}})^{(i-1)}$. Then ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$ is a $\mathbb Z$-graded superalgebra associated to $BC_{r,t}^{\text{\rm aff}}$. We use the same symbols to denote elements in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$. In particular, $x_i=x_i'$ and $\bar x_j'=\bar x_j$ in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$.
\[regm\] We say that $\textbf{m}$ is a regular monomial of $BC_{r,t}^{\rm aff}$ if $\textbf{m}= x^{\alpha} d {\bar x}^{\beta} \prod_{n\in \mathbb Z^{>0}} \omega_{2n+1}^{a_{2n+1}}$, for some $d\in S$, $a_{2n+1}\in \mathbb N$ and $ (\alpha, \beta)\in \mathbb N^r\times \mathbb N^t$, where $S$ is given in , $x^\alpha=\prod_{i=1}^r x_i^{\alpha_i}$ and $\bar x^\beta=\prod_{i=1}^t \bar x_i^{\beta_i}$.
\[spanned\] As an $R$-module, $BC_{r,t}^{\rm aff}$ is spanned by all regular monomials in Definition $\ref{regm}$.
Let $M$ be the $R$-submodule of $ BC_{r,t}^{\rm aff}$ spanned by all regular monomials $\textit{\textbf{m}} $ in Definition \[regm\]. We want to prove $$\label{veri}h\textit{\textbf{m}} \in M\mbox{ \ \ for any generator $h$ of $ BC_{r,t}^{\rm aff}$}.$$ If so, we have $M=BC_{r,t}^{\rm aff}$ since $1\in M$.
We prove (\[veri\]) by induction on $|\alpha|$. If $|\alpha|=0$, i.e., $\alpha_i=0$ for all possible $i$’s, then (\[veri\]) follows from Theorem \[wbhsa321\] unless $h= \bar x_1$. In the later case, by , and Lemma \[hecrel\], we need to compute $\bar x_k e^f $ when $1\le k\le t$ and $f>0$. If $k\in \{1,2, \ldots, f\}$, by Lemma \[hecrel2\](4), we use $-x_k$ instead of $\bar x_k$ since we work on the graded superalgebra ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$. So, $h{\textit{\textbf{m}}}\in M$. Otherwise, $k>f$. By Lemma \[relxpri\](2), we can use $e^f \bar x_k $ instead of $\bar x_k e^f $. So, (\[veri\]) follows from Lemma \[hecrel\] and Theorem \[wbhsa321\].
Suppose $|\alpha|>0$. By ,, Lemma \[hecrel\] and Theorem \[wbhsa321\], we see that (\[veri\]) holds unless $h\in \{\bar x_1, e_1\}$. Suppose $h=\bar x_1$. By Lemma \[hecrel2\](1) and (2), $\bar x_1 x_i=x_i\bar x_1$ in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$. So, we need to deal with $\bar x_k e^f $ when $1\le k\le t$. They are the cases that we have dealt with. So, $\bar x_1 \textbf{m}\in M$.
Finally, we assume $h=e_1$. If $\alpha_i\neq 0$ for some $i$ with $ 2\le i\le r$, then $e_1 x_i=x_i e_1$ in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$ (see Lemma \[relxpri\](1)). By inductive assumption on $|\alpha|$, we have (\[veri\]). In order to finish the proof, it remains to consider the case that $x^{\alpha}=x_1^{\alpha_1}$ such that $\alpha_1>0$. In this case, $$\label{veri1}\textbf{m} = x_1^{\alpha_1} {c}^{\gamma } d_1^{-1} e^f w d_2 {\bar c}^{\delta} \bar x^{\beta}\in M,$$ where $d_1, d_2\in \mathscr D_{r, t}^f$ and $\beta\in \mathbb N^t$ and $(\gamma, \delta)\in \Z_2^r \times \Z_2^t $. Write $d_1 e_1 d_1^{-1}=e_{i, j}$ for some $i,j$. By and inductive assumption on $|\alpha|$, we can use $d_1^{-1} x_i^{\alpha_1}$ to replace $x_1^{\alpha_1} d_1^{-1}$ in . So, we need to verify $$\label{v1} e_{i, j} x_i^{\alpha_1} c^{\gamma} e^f w d_2 {\bar c}^{\delta } \bar x^{\beta}\in M.$$ By Lemma \[hecrel2\](3) and inductive assumption, it is enough to verify $$\label{v2} e_{i, j} \bar x_j^{\alpha_1}c^{\gamma} e^f w d_2 {\bar c}^{\delta } \bar x^{\beta}\in M.$$ If $j\ge f+1$, then follows from Lemma \[relxpri\](2) and Theorem \[wbhsa321\]. Otherwise, $j\le f$. If $i=j$, by inductive assumption, we use $( \bar x_i+\bar L_i)^{\alpha_1}$ instead of $\bar x_i^{\alpha_1} $ in $e_{i} \bar x_i^{\alpha_1} c^\gamma e_i$. If $\gamma_i=0$, then $e_{i} \bar x_i^{\alpha_1} e_i=\bar \omega_{\alpha_1} e_1$ in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$ (see Lemma \[relxpri\](4)). If $\gamma_i\neq 0$, then $e_{i} \bar x_i^{\alpha_1}c_i e_i=0$ in ${\rm gr} ( BC_{r,t}^{\text{\rm aff}})$ (see Lemma \[hecrel2\](5)). In any case, follows from inductive assumption on $|\alpha|$. Finally, we assume $i\neq j$. If $i\neq k$, then $e_{i,j}c_k=c_ke_{i,j}$. By inductive assumption, we need to consider $e_{i,j} x_i^{\alpha_1} c_i^{\gamma_i} e_j=e_{i,j} x_i^{\alpha_1} e_j c_i^{\gamma_i}$. Since $$e_{i,j} x_i^{\alpha_1} e_j= e_{i,j} e_j x_i^{\alpha_1} = (i, j)x_i^{\alpha_1}e_j=x_j^{\alpha_1} (i, j)e_j$$ in ${\rm gr}(BC_{r,t}^{\text{\rm aff}})$, by inductive assumption and our previous results on $h\in$ $\{s_1, \ldots, s_{r-1}, c_1,\ldots, c_r, x_1\}$, we have . So is true. This completes the proof.
\[affinewbcsa\] Let $I$ be the two-sided ideal of $BC_{r, t}^{\rm aff}$ generated by $\omega_{2k+1}-{\tilde \omega}_{2k+1}$, where $\tilde {\omega}_{2k+1}\in R$ for all $k\in \mathbb Z^{>0}$. Let $\widetilde {BC}_{r, t}=BC_{r, t}^{\rm aff}/I$.
\[cwbcsa1\] Let $I$ be the two-sided ideal of $\widetilde{BC}_{r, t}$ generated by $f(x_1)$ and $g(\bar x_1)$, where $$\label{fg} f(x_1)= x_1^k\mbox{$\prod\limits _{i=1}^m$} (x_1^2-u_i^2), \text{ and }g(\bar x_1)= \bar x_1^{k_1}\mbox{$\prod\limits _{j=1}^{m_1}$}(\bar x_1^2-\bar u_j^2),$$ for some non-zero $u_1, \ldots, u_m, \bar u_1, \ldots,\bar u_{m_1}\in R$ such that $\ell=k+2m=k_1+2m_1$ and $$\label{efg3211} e_1 f(x_1)=(-1)^k e_1 g(\bar x_1).$$ The [*level $\ell$ or cyclotomic walled Brauer-Clifford superalgebra $BC_{\ell, r, t}$* ]{} is the quotient algebra $\widetilde{BC}_{r, t}/I$.
In section 6, we will explain the reason why $f(x_1)$ and $g(\bar x_1)$ have to satisfy and .
\[cregm\] We say that $\textbf {m}$ is a regular monomial of $\widetilde{BC}_{r, t}$ (resp., $ BC_{\ell, r, t}$) if it is of form $ x^{\alpha} d {\bar x}^{\beta} $, for some $d\in S$, and $ (\alpha, \beta)\in \mathbb N^r\times \mathbb N^t$ (resp. $\mathbb Z_\ell^r\times \mathbb Z_\ell^t$), where $S$ is given in .
\[level-l-span\] As $R$-modules, both $\widetilde{BC}_{r, t}$ and $BC_{\ell, r,t}$ are spanned by their regular monomials.
By Proposition \[spanned\], $\widetilde{BC}_{r, t}$ is spanned by all its regular monomials. Let $\phi_\ell: \widetilde{BC}_{r, t}\twoheadrightarrow BC_{\ell, r,t}$ be the canonical epimorphism. It is enough to verify the image of a regular monomial $\textbf{m}$ of $\widetilde{BC}_{r, t}$ can be expressed as a linear combination of regular monomials of $BC_{\ell, r,t}$. If both $\alpha\in \Z_\ell^r$ and $\beta\in \Z_\ell^t$, then the images of $\textbf{m}$ is a regular monomial of $ BC_{\ell, r,t}$. Otherwise, either $\alpha_i\ge \ell$ or $\beta_j\ge \ell$ for some possible $i$ or $j$. Since $\widetilde {BC}_{r, t}$ inherits the graded structure of $BC_{r, t}^{\rm aff}$, it results in a graded structure on $BC_{\ell, r, t}$. So, either $x_i^{\alpha_i}$ or $\bar x_j^{\beta_j}$ can be expressed as a linear combination of elements in $BC_{\ell, r, t}$ with lower degrees. Using these elements to replace either $x_i^{\alpha_i}$ or $\bar x_j^{\beta_j}$ in the image of $\textbf{m}$ and considering the inverse images of such elements in $\widetilde{BC}_{r, t}$, we see that the image of $\textbf{m}$ can be expressed as a linear combination of regular monomials of $BC_{\ell, r, t}$, as required.
A basis of $BC_{2, r, t}$ with special parameters
=================================================
Let $\mfg=\mathfrak{q}(n)$ be the queer Lie superalgebra of rank $n$ over $\C$, which has a basis $e_{ij}=E_{i,j}+E_{-i,-j}$ (even element), $f_{i,j}=E_{i,-j}+E_{-i,j}$ (odd element) for $i,j\in I^+=\{1,2,...,n\},$ where $E_{i,j}$ is the $2n\times2n$ matrix with entry $1$ at $(i,j)$ position and zero otherwise for $i,j\in I=I^+\cup I^-$, and $I^-=-I^+$. Let $V=\C^{n|n}=V_{\bar0}\oplus V_{\bar1}$ be the natural $\mfg$-module (and the natural ${\mathfrak{gl}}_{n|n}$-module) with basis $\{v_i\,|\,i\in I\}$. Then $v_i$ has the parity $[v_i]=[i]\in\Z_2$, where $[i]=0$ and $[-i]= 1$ for $i\in I^+$. Let $V^*$ be the linear dual space of $V$ with dual basis $\{\bar v_i\,|\,i\in I\}$. Thus $V^*$ is a left $\mfg$-module with action $$\begin{aligned}
\label{action-dual}
E_{a,b}\bar v_i=-(-1)^{[a]([a]+[b])}\d_{i,a}\bar v_b\mbox{ for }a,b,i\in I.\end{aligned}$$ Let $\fh=\fh_{\bar 0}\oplus\fh_{\bar 1}$ be a Cartan subalgebra of $\mfg$ with even part $\fh_{\bar0}={\rm span}\{e_{i,i}\,\,|\,i\in I^+\}$ and odd part $\fh_{\bar 1}={\rm span}\{f_{i,i}\,\,|\,i\in I^+\}$. Let $\fh^*_{\bar0}$ be the dual space of $\fh_{\bar0}$ with $\{\es_i\,|\,i\in I^+\}$ being the dual basis of $\{e_{i,i}\,\,|\,i\in I^+\}$. Then an element $\l\in\fh^*$ (called a [*weight*]{}) can be written as $$\label{weight-}\l=\SUM{i\in I^+}{}\l_i\es_i=(\l_1,...,\l_n)\mbox{ \ with \ }\l_i\in\C.$$ Let $M$ be any $\mfg$-module. For any $r,t\!\in\!\Z^{\ge0}$, set $M^{r,t}=M\OTIMES V^{\otimes r}\OTIMES (V^*)^{\otimes t}$. For convenience we denote the ordered set $$\label{ordered-set}J=\{0\}\cup J_1\cup J_2,\
\mbox{ where $J_1=\{1,...,r\}$, $J_2=\{\bar 1,...,\bar t\}$,}$$ and $0\prec1\prec\ldots\prec r\prec\bar1\prec\ldots\prec\bar t$. We write $M^{r,t}$ as $$\label{M-st==}M^{r,t}=\OT{i\in J}V_i,\mbox{ \ where $V_0=M$, $V_i=V$ if $0\prec i\prec\bar1$, and $V_i=V^*$ if $i\succ r$,}$$ (hereafter all tensor products will be taken according to the order in $J$), which is a left $U(\mfg)^{\otimes(r+t+1)}$-module (where $U(\mfg)$ is the universal enveloping algebra of $\mfg$), with the action given by $$\Big(\OT{i\in J} g_i\Big)\Big(\OT{i\in J} x_i\Big)=(-1)^{\sum\limits_{i\in J}{}[g_i]\sum\limits_{j\prec i}{}[x_j]}\OT{i\in J}(g_ix_i)\mbox{ for }g_i\in U(\mfg),\ x_i\in V_i.$$
For the purpose of proving a basis of level $2$ walled-Brauer Clifford superalgebra, we take $n=2m$ to be even integer. We denote $I_1^+=\{1,...,m\},$ $I_2^+=m+I_1^+$. Thus $I^+=I_1^+\cup I_2^+$. For $i\in I_1^+$, we denote $i_{\bullet}=i+m\in I_2^+$. For $i\in I_2^+$, we denote $i_{\circ}=i-m\in I_1^+$. Let $M=L_\l$ be the finite dimensional simple $\mfg$-module of type $M$ with dominant highest weight Then $End_g(L_\lambda)$ is one dimensional. Denote by $v_\l$ a fixed highest weight vector of $L_\l$ with even parity, and $(L_\l)_\l$ the highest weight space of $L_\l$, which is $2^m$-dimensional with a basis where the products are taken in any fixed order $($changing the order only changes the vectors by a factor $\pm1)$. For $i\in I_2^+$, we have Let $C$ be the PBW monomial basis of $U(\mfg^-\oplus\fh)$. We say a basis element $a\in C$ has [*length*]{} $\ell(a):=k$ if $a$ contains $k$ factors; for instance, $\ell(b^{\th})=|\th|$. For $i\in\Z^{\ge0}$, let $C_i=\{a\in C\,|\,\ell(a)=i\}.$ Set Let $(L_\l)_i$ be the subspace of $L_\l$ spanned by $cv_\l$ for $c\in C$ with $\ell(c)\le i$. Set $(L_\l)_{-1}=0$. Note that elements of $\mfg^+$ acting on $L_\l$ send $(L_\l)_i$ to $(L_\l)_{i-1}$.
\[basi1lemm\] For $i\in\Z^{\ge0}$, the set $D_iv_\l$ is $\C$-linear independent under modulo $(L_\l)_{i-1}$.
Assume $c:=\sum_{\si\in\Z_2^m:|\si|=i}a_{\si} u^{\si} v_\l\in (L_\l)_{i-1}$ for some $a_{\si}\in\C$ with at least one $a_{\si}\ne0$. Take a $\tilde\si\in\Z_2^m$ such that $a_{\tilde\si}\ne0$. Assume $\tilde\si_\ell\ne0$ for some $\ell\in I_1^+$. Applying $f_{\ell,\ell_{\bullet}}\in\mfg^+$ to $c$, by moving $f_{\ell,\ell_{\bullet}}$ to the right until it meets $v_\l$, using the commutation relation $[f_{\ell,\ell_{\bullet}},f_{j_{\bullet},j}]=\d_{\ell{\ssc\,}j}(e_{\ell\ell}+e_{\ell_{\bullet},\ell_{\bullet}})$ (which is a Cartan element commuting with $f_{i_{\bullet},i}$ for $i\ne\ell$), we can easily obtain $f_{\ell,\ell_{\bullet}}c=\sum_{\si\in\Z_2^m:\si_{\ell}\ne0}a'_{\si}(2p-(\ell+\ell_{\bullet})) u^{\si-1_\ell}v_\l\in (L_\l)_{i-2}$, where $a'_{\si}=\pm a_{\si}$, $1_\ell=(\d_{1\ell},...,\d_{m\ell})\in\Z_2^m$. Note that $2p-(\ell+\ell_{\bullet})\ne0$ by . Now induction on $|\th|$ gives that $a_{\tilde\si}=0$, a contradiction with the assumption.
For each $i=1,2,...,$ by Lemma \[basi1lemm\], we can choose a maximal subset $\hat C_i$ of $C_i$ satisfying the following conditions (i.e., we extend $D_i$ to a basis $\hat C_i$ of $(L_\l)_i$ modulo $(L_\l)_{i-1}$, thus $\#\hat C_i={\rm dim}(L_\l)_i/(L_\l)_{i-1}$):
- $\hat C_i\supset D_i$;
- $\{u v_\l\,|\,u\in\hat C_i\}$ is a $\C$-linear independent subset of $L_\l$.
Then we have the following basis of $L_\l$, We say the basis element $wv_\l$ has [*length*]{} $\ell(wv_\l):=\ell(w)$. Then from our choice of $\hat C_i$, we immediately have the following.
\[Basis20\] Let $\alpha\in C$ be a monomial basis element of length $j$. Then $\alpha v_\l$ is a combination of basis elements in $B^e$ with length $\le j$.
Take the following basis of $M^{r,t}$, $$\label{basis-M-0}
B_M=\Big\{b_M= b\otimes\OT{i\in J_1}v_{k_i}\otimes\OT{i\in J_2}\bar v_{k_i}\,\Big|\,b\in B^e,\,
k_i\in I\Big\}.$$ Introduce the following elements, $$\begin{aligned}
\label{def-Omega}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
\bar e_{ij}=E_{ij}-E_{-i,-j},\ \ \ \bar f_{ij}=E_{-i,j}-E_{i,-j}\ \ \in\ \ {\mathfrak {gl}}_{n|n},\nonumber\\
&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&\Omega_0\!=\!\mbox{$\sum\limits_{i,j\in I}$}(-1)^{[j]}E_{ij}\OTIMES E_{ji}\in {\mathfrak {gl}}_{n|n}^{\otimes 2},\ \ \
\Omega_1\!=\!\mbox{$\sum\limits_{i,j\in I^+}$}e_{ij}\OTIMES \bar e_{ji}\!-\!\mbox{$\sum\limits_{i,j\in I^+}$}f_{ij}\OTIMES \bar f_{ji}\in\mfg\OTIMES {\mathfrak {gl}}_{n|n}.\end{aligned}$$ For $a,b\in J$ with $a\prec b$, we define $\pi_{ab}:U(\mfg)\otimes U({\mathfrak {gl}}_{n|n})\to U(\mfg)\otimes U({\mathfrak {gl}}_{n|n})^{\otimes(r+t)}$ by $$\label{pi-ab}
\pi_{ab}(x\OTIMES y)=1\OTIMES\cdots\OTIMES1\OTIMES x\OTIMES 1\OTIMES\cdots\OTIMES1\OTIMES y\OTIMES1\OTIMES\cdots\OTIMES1,$$ where $x$ and $y$ are in the $a$-th and $b$-th tensors respectively. Similarly we have $\pi_a:U(\mfg)\to U(\mfg)^{\otimes(r+t+1)}$ which sends $x$ to the $a$-th tensor.
\[casm\] We can use to define the following elements of the endomorphism algebra ${\rm End}_{U(\mfg)}(M^{r,t})^{\rm op}$, $$\begin{aligned}
\label{operator--1}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
s_i=\pi_{i,i+1}(\Omega_0)|_{M^{r,t}}\ (1\!\le\! i\!<\!r),\ \ \ \
\bar s_j=\pi_{\bar j,\overline{j+1}}(\Omega_0)|_{M^{r,t}}\ (1\!\le\! j\!<\!t),\nonumber\\&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
x'_i=-\pi_{0i}(\Omega_1)|_{M^{r,t}}\ (1\!\le\! i\!\le\! r),\ \ \ \ \bar x'_j=-\pi_{0\bar j}(\Omega_1)|_{M^{r,t}}\ (1\!\le\! j\!\le\! t),
\nonumber\\&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!& e_i\!=\!-\pi_{i\bar i}(\Omega_0)|_{M^{r,t}}\,(1\!\le\! i\!\le\!\min\{r,t\}),\ \
c_i\!=\!\pi_i(c)\, (1\!\le\! i\!\le\! r),\ \ \bar c_i\!=\!\pi_{\bar i}(\bar c)\,(1\!\le\! i\!\le\! t),\end{aligned}$$ where $c:V\to V$ (resp., $\bar c:V^*\to V^*$) is the automorphism such that $c(v_{\pm i})=\pm v_{\mp i}$ (resp., $\bar c(\bar v_{\pm i})=\bar v_{\mp i}$). Set $x_1=x'_1,\,\bar x_1=\bar x'_1$.
Observe that $c^2=-1$ and $\bar c^2=1$, and $c,\bar c$ correspond to maps $c,\bar c:I\to I$ such that
\[SMSM\]
1. The minimal polynomial of $x_1$ with respect to $M^{r,t}$ is $f(x)=x^2-p(p+1)$.
2. The minimal polynomial of $\bar x_1$ with respect to $M^{r,t}$ is $g(\bar x_1)=x^2-(p-n+1)(p-n)$.
3. We have $e_1x_1e_1=-n(2p-n+1)e_1$ with respect to $M^{r,t}$.
\(a) We may assume $r=1,\,t=0$. Note that the only possible highest weight in the finite dimensional $\mfg$-module $L_\l\otimes V$ is $\mu=\l+\epsilon_1$, which is a typical dominant weight. Thus $L_\l\otimes V$ must be complete reducible, and thus a direct sum of finite copies of $L_{\mu}$. Observe that the set $\{u^\th\otimes v_{\pm1}\,|\,u^\th\in B_0\}$, with $2^{m+1}$ elements, is a maximal set of $\C$-linear independent highest weight vectors of weight $\mu$. Since $L_\mu$ occupies $2^m$ $\C$-linear independent highest weight vectors, we see that $L_\l\otimes V=L_\mu^{\oplus2}$, which as a $\mfg$-module is generated by $v^\pm_\mu:=v_\l\otimes v_{\pm1}$. One can easily verify that $v_\mu^\pm x_1=\mp(p+1)v_\mu^\pm\pm f_{11}v_\mu^\mp $. Thus $v_\mu^+,f_{11}v_\mu^{-}$ (resp., $v_\mu^-,f_{11}v_\mu^{+}$) span a 2-dimensional $x_1$-invariant subspace of $L_\l\otimes V$, and the minimal polynomial of $x_1$ in this subspace is $f(x)=x_1^2-p(p+1)$. Since $x_1$ commutes with the $\mfg$-action and $L_\l\otimes V$ is generated by $v_\mu^\pm$, we see $f(x_1)$ is also the minimal polynomial of $x_1$ in $M^{r,t}$.
\(b) We can assume $r=0,\,t=1.$ Similar to the arguments in (a), we have $L_\l\otimes V^*=L_\nu^{\oplus2}$ with highest weight $\nu=\l-\epsilon_n$ (which is again a typical dominant weight) and two highest weight vectors $v_\nu^\pm:=v_\l\otimes\bar v_{\pm n}$. In addition, $v_\nu^\pm\bar x_1=\pm(p-n)v_\nu^\pm+f_{nn} v_\nu^\mp$. Thus the minimal polynomial of $\bar x_1$ is $g(\bar x_1)=\bar x_1^2-(p-n+1)(p-n)$.
\(c) We can assume $r=t=1.$ Then for $a,b\in I$, we have $$\begin{array}{lll}(v_\l\otimes v_a\otimes \bar v_b)e_1x_1e_1=(-1)^{[a]}\d_{ab}\sum\limits_{i\in I}(v_\l\otimes v_i\otimes\bar v_i)x_1e_1\\[4pt]
\phantom{===}=(-1)^{[a]}\d_{ab}\sum\limits_{i\in I}(-1)^{[i]}\big((p\!+\!1\!-\!i)v_\l\otimes v_i\otimes\bar v_i\big)e_1\!=\!-n(2p\!-\!n\!+\!1)(v_\l\otimes v_a\otimes \bar v_b)e_1.\end{array}$$ Since $L_\l\otimes V\otimes V^*$ is generated by $v_\l\otimes v_a\otimes \bar v_b$ for $a,b\in I$, and $e_1,x_1$ commute with the $\mfg$-action, we obtain (c).
\[p4\] For $k\in I^+$, we have $$\begin{aligned}
&\!\!\!\!\!\!\!\!\!&
(v_\lambda\otimes v_{\pm k})x_1=\mp\lambda_k v_\lambda \otimes v_{\pm k}\pm (-1)^{[ v_\lambda]}f_{k,k}v_{\lambda}\otimes v_{\mp k}\mp \mbox{$\sum\limits_{j<k}$}e_{k,j}v_\lambda \otimes v_{\pm j}\mp\(-1)^{[v_\lambda]} \mbox{$\sum\limits_{j<k}$}f_{k,j}v_\lambda \otimes v_{\mp j},\\
&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
(v_\lambda\otimes{\bar v_{\pm k}}) \bar x_1=\pm\lambda_k v_\lambda \otimes {\bar v_{\pm k}}\mp (-1)^{[ v_\lambda]}f_{k,k}v_{\lambda}\otimes \bar v_{\mp k} \pm \mbox{$\sum\limits_{i>k}$}e_{i,k}v_\lambda \otimes \bar v_{\pm i} \mp\(-1)^{[ v_\lambda]}\mbox{$\sum\limits_{i>k}$}f_{i,k}v_\lambda \otimes {\bar v_{\mp i}}.\end{aligned}$$
The result follows from the definitions of $x_1$ and $\bar x_1$.
For any $a\in I$, we set $a^+=|a|\in I^+$. Then gives Now we assume $BC_{2,r, t}$ is the level two walled-Brauer Clifford superalgebra such that $x_1,\bar x_1$ satisfy the degree 2 polynomials in Lemma \[SMSM\], and parameters satisfy where $ p\!\in\!\C\backslash\Z$, or $p\in\Z$ with $p>2m$ or $p<0$, and $m\in\Z^{>0}$ satisfies $m\ge 2(r+t)$. By Lemma 4.4 and Definition 3.1(1)(3), we have $e_1 f(x_1)=e_1g(\bar x_1)$. We take $n=2m$. Take the weight $\l$ as in , then we can define the space $ M^{r,t}$ as in .
\[hom\] There is an algebra homomorphism $\varphi: BC_{2,r, t} \rightarrow End_{\mfg}(M^{r,t})^{op}$ such that $\varphi$ sends the generators $e_1$, $x_1$, $\bar x_1$, $s_i$’s, $\bar s_j$’s, $c_m$’s, $\bar c_n$’s to the same symbols defined in the Definition \[casm\].
By Lemma 4.4(a)-(b), we need to show the images of the generators satisfy the relations in Definition \[awbsa\]. First it is easy to see that (2.1)–(2.3) and Definition 2.1 (1)–(10) are satisfied (cf.[@JK] or [@BGJKW Theorem 1.4]). By [@HKS Thoerem 7.4.1], (2.7)–(2.8) are satisfied. Moreover, (2), (4), (8)–(11) in Definition \[awbsa\] follows from the Definition \[casm\].
Let $\Omega_{0,i}=\pi_{0,i}(\Omega_1) $ for $i\in J_1\cup J_2$, $S_{i,j}=\pi_{i,j}(\Omega_0) $ for $i,j\in J_1\cup J_2$, and $C_i=\pi_i(C)$ for $i\in J_1\cup J_2$, where $C=\sum_{i\in I^{+}} \bar f_{i,i}$. Then $c_i= C_i|_{M^{r,t}}$ for $i\in J_1$ and $\bar c_i= C_{\bar i}|_{M^{r,t}}$ for $\bar i\in J_2$. It follows from the proof of [@HKS Thoerem 7.4.1] that
- $\Omega_1(1\otimes C )=-(1\otimes C )\Omega_1$,
- $S_{i,i+1}\Omega_{0,i}S_{i,i+1}=\Omega_{0,i+1}$,
- $(1^{\otimes {i-1}}\otimes C \otimes 1) S_{i,i+1}=S_{i,i+1}(1^{\otimes i }\otimes C)$,
- $\Omega_{0,i}\Omega_{0,j}-\Omega_{0,j}\Omega_{0,i}=(\Omega_{0,j}-\Omega_{0,i})S_{i,j}+(\Omega_{0,j}+\Omega_{0,i})
C_iC_jS_{i,j}$.
Assuming that $i=1$ and $j=\bar 1$ in (d), we have $$\label{sss1} \Omega_{0,1}(\Omega_{0,\bar 1}+S_{1,\bar 1}-C_1C_{\bar 1}S_{1,\bar 1})=\Omega_{0,\bar 1}\Omega_{0,1}+\Omega_{0,\bar 1}S_{1,\bar 1}+\Omega_{0,\bar 1}C_1C_{\bar 1}S_{1,\bar 1}.$$ By (a)-(c), we get $\Omega_{0, \bar 1} C_1C_{\bar 1} S_{1, \bar 1}=-C_1S_{1, \bar 1} C_1\Omega_{0, 1}$, $\Omega_{0, \bar 1}S_{1,\bar 1}=S_{1,\bar 1} \Omega_{0, 1}$, $C_1 C_{\bar 1} S_{1, \bar 1}=C_1S_{1, \bar 1} C_1$, respectively. So, the right hand side of is equal to $$\label{sss2}\Omega_{0,\bar 1}\Omega_{0,1}+S_{1,\bar 1}\Omega_{0,1}-C_1S_{1,\bar 1}C_1\Omega_{0,1}=(\Omega_{0,\bar 1}+S_{1,\bar 1}-C_1S_{1,\bar 1}C_1)\Omega_{0,1}.$$ Now, Definition \[awbsa\](3) is satisfied by -. Consider the following $\mfg$-homomorphism:
- $\alpha:V \otimes V^{*} \rightarrow \mathbb C$ such that$u\otimes \phi\mapsto (-1)^{[{\phi}] [{u}]} \phi(u)$;
- $\beta:\mathbb C\rightarrow V\otimes V^{*} $ such that $1\mapsto\sum_{i\in I}v_i\otimes \bar v_i$.
Then $e_1=-\beta\circ\alpha$. Since $ \alpha\circ {x_1}^a\circ\beta \in End_{\mfg}(L_\lambda)$, and $L(\lambda)$ is of type $M$, we have $\alpha\circ {x_1}^a\circ\beta=-{\omega_a}$, for some $ {\omega_a}\in\mathbb C$. Then $e_1x_1^ae_1=(-\beta\circ\alpha)\circ x_1^a\circ(-\beta\circ\alpha)=-{\omega_a}\beta\circ\alpha=\omega_a e_1$. Similarly, $e_1\bar x_1^a e_1=\bar \omega_a e_1$ for some ${\bar \omega_a}\in\mathbb C$. So, Definition \[awbsa\] (5), (7) are satisfied. Since $c_1^2=-1,\,\bar c_1^2=1,\,c_1x_1=-x_1c_1$, we can get that $e_1c_1^2x_1^{2a}e_1=-e_1x_1^{2a}e_1$. Hence $\omega_{2a}=-\omega_{2a}$ and $\,\omega_{2a}=0$, and Definition \[awbsa\] (6) is satisfied. Finally, Definition \[awbsa\] (1) follows from Lemma \[p4\].
By Proposition \[hom\], $M^{r,t}$ is a right $BC_{2,r, t}$-module. For any $\alpha,\b\in\Z_2^r$, $\bar\alpha,\bar\beta\in\Z_2^t$, we define the following elements of $BC_{2,r, t}$: $$\label{x-al-bx-b}
c^\alpha=\mbox{$\prod\limits_{i=1}^rc_i^{\alpha_i},\ \ x'^\beta=\prod\limits_{j=1}^r x'^{\beta_j}_j,$}\ \
\bar c^{\bar\alpha}=\mbox{$\prod\limits_{i=1}^t\bar c_i^{\bar\alpha_i},\ \ \bar x'^{\bar \beta}=\prod\limits_{j=1}^t\bar x'^{\bar\beta_j}_j,$}$$where the product in $x'^\beta$ is written in the order $x_r'^{\beta_r}\cdots x_1'^{\beta_1}$ (thus $x_1'$ acts first on $M^{r,t}$) and the like for $\bar x'^{\bar\beta}$.
\[theo-2222\]The monomials $$\label{monom}
{\textit{\textbf{m}}}:=d_1^{-1}c^\alpha x'^\beta e^f \bar x'^{\bar\beta}\bar c^{\bar\alpha}wd_2,$$ with $\alpha,\beta\in\Z_2^r,\,\bar\alpha,\bar\beta\in\Z_2^t$ and $d_1,e^f,w,d_2$ as in , are $\mathbb C$-linearly independent elements of $ BC_{2,r, t}$.
\#1[[***\#1[$\ssc\,$]{}***]{}]{}
Suppose there is a nonzero $\mathbb C$-combination $\tif c:=\sum_{\tif m}r_{\tif m}\tif m$ of monomials being zero. We fix a monomial $\tilde{\tif m}:=\tilde d^{-1}_1c^{\tilde \alpha} x'^{\tilde \beta} e^{\tilde f} \bar x'^{\bar{\tilde \beta}}\bar c^{\bar{\tilde \alpha}}\tilde w\tilde d_2$ in $\tif c$ with nonzero coefficient $r_{\tilde{\tif m}}\ne0$ which satisfies the following conditions:
- $|\tilde\beta|+|\bar{\tilde\beta}|$ is maximal;
- $\tilde f$ is minimal among all monomials satisfying (i).
We take the basis element $v= v_\l\OTIMES\OTIMES_{i\in J_1}v_{k_i}\OTIMES\OTIMES_{i\in J_2}\bar v_{k_i}\in B_{ M}$ (cf. ) such that (note that here is the place where we require condition that $2(r+t)\le m$)
- for $1\le i\le r$, $k_i=i$ if $\tilde\b_i=0$ and $k_i=-(m+i)$ if $\tilde\b_i=1$;
- for $1\le i\le \tilde f$, $k_{\bar i}=k_i^+$;
- for $\tilde f<i\le t$, $k_{\bar i}=r+i_{\bullet}$ if $\bar{\tilde \b}_i=0$ and $k_{\bar i}=-(r+i)$ if $\bar{\tilde \b}_i=1$.
Take $$\begin{aligned}
\label{b'===}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
z:=(v) c^{\tilde\alpha}\tilde d_1{\tif c}\tilde d_2^{-1}\tilde w^{-1}\bar c^{\bar{\tilde\alpha}}\in M^{r,t},\\
\label{1b'===}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
\tilde u=\mbox{$\prod\limits_{i=1}^r$}f^{\tilde\b_i}_{i_{\bullet},i}
\mbox{$\prod\limits_{i=1}^t$}f^{\bar{\tilde\b}_i}_{r+i_{\bullet},r+i}v_\l\in B^e,
\\
\label{2b'===}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
\tilde v=\tilde u\OTIMES\OT{i\in J_1}v_{i}\OTIMES\OT{i\in J_2}\bar v_{r+i_{\bullet}}\in B_{ M},
\end{aligned}$$ where the product in is in the same order as in . We want to prove that when written $z$ as a combination of basis $B_{ M}$ in , the coefficient $\chi^{z}_{\tilde v}$ of $\tilde v$ is nonzero. Thus assume a monomial $\tif m$ in appears in the expression of $\tif c$ with $r_{\tif m}\ne0$ and consider the following element, $$\begin{aligned}
\label{u-11111}
z_1&\!\!\!:=\!\!\!&(v)c^{\tilde\alpha}\tilde d_1{\tif m}\tilde d_2^{-1}\tilde w^{-1}\bar c^{\bar{\tilde\alpha}}=
(v)c^{\tilde\alpha}\tilde d_1d_1^{-1}c^\alpha x'^\b e^f\bar x'^{\bar\b}\bar c^{\bar\alpha}wd_2\tilde d_2^{-1}\tilde w^{-1}\bar c^{\bar{\tilde\alpha}}\nonumber\\
&\!\!\!=\!\!\!&
\Big(\,v_\l\otimes\OT{i\in J_1}v_{ c^{\g_i+\tilde\alpha_i}(k_{(i)\tilde d_1d_1^{-1}})}\otimes\OT{i\in J_2}\bar v_{k_{(i)\tilde d_1d_1^{-1}}}\Big)x'^\b e^f\bar x'^{\bar\b}\bar c^{\bar\alpha}wd_2\tilde d_2^{-1}\tilde w^{-1}\bar c^{\bar{\tilde\alpha}},\end{aligned}$$where $\g_i=\alpha_{(i)\tilde d_1d_1^{-1}}$, and where the last equation is understood as “equal up to a sign” (cf. ), which follows by noting that elements in $\Sigma_r\times\bar{\Sigma}_t$ have natural right actions on $J_1\cup J_2$ by permutations and $c$ acts on $I$ by . Write $z_1$ as a $\mathbb C$-combination of basis $B_{ M}$. If $\tilde v$ appears as a term with a nonzero coefficient in the combination, then we say that $z_1$ [produces]{} $\tilde v$.
Note that $\tilde u$ has length $|\b|+|\bar\b|$. By Definition \[casm\] and from our choice of $B^e$ in , we see that factors of $\tilde u$ can be only contributed by the actions of $x'_i$’s and $\bar x'_i$’s, and each $x'_i$ or $\bar x'_i$ can at most contribute one length of $\tilde u$ by observing the following: if the first factor of a term in $\pi_{0i}(\Omega_{1})$ for $i\in J_1\cup J_2$ acting on the first factor of an element in $B_M$ changes the first factor to a basis element in $B^e$ then this $\pi_{0i}(\Omega_{1})$ may contribute one length, otherwise the first factor is changed to a combination of basis elements with length not increasing by Lemma \[Basis20\]. We see that $z_1$ cannot produce a basis element with degree higher than $|\b|+|\bar\b|$. Thus $\tilde v$ cannot be produced if $|\b|+|\bar\b|<|\tilde\b|+|\bar{\tilde\b}|$. So by condition (i), we can assume Then $f{\!}\ge{\!}\tilde f$ by condition (ii).
For any basis element $b_M$ written as in , we say $k_i$ the [*$i$-th label*]{} of $b_M$ for $i\in J_1\cup J_2$. Note from that all factors of $\tilde u$ have the following form Thus when $x'_i=-\pi_{0i}(\Omega_1)$ for $1\le i\le r$ is applied to the element inside the bracket, it can only change its $i$-th label, say $\ell$, to $\pm\ell$, $\pm(\ell-m)$. Since $\tilde d_1d_1^{-1}$ only permutes labels and $c^{\g_i+\tilde\alpha_i}$ only changes labels up to a sign, in order for a term in to contribute to $\chi^{z_1}_{\tilde v}$, we need at least $f$ pairs $(i,\bar j)\in J_1\times J_2$ such that the $i$-th label $k_i$ and $j$-th label $k_{\bar j}$ satisfy the condition $k_i^+=k_{\bar j}^+$ or $k_i^+=k_{\bar j}^++m$. From our choice of the vector $v$, we must have $f\le\tilde f$. Thus we can suppose $\tilde f=f$ by the fact that $f\ge \tilde f$.
Set $J_{f}=\{i,\bar i\,|\,1\le i\le \tilde f=f\}\subset J_1\cup J_2$ (cf. ). If $d_1\ne \tilde d_1$, then by definition , we have $$\label{j---in}\tilde j :=(j)\tilde d_1d_1^{-1}\notin J_{f}\mbox{ \ for some $j\in J_{f}$}.$$ Say $\tilde j \!\in\! J_1$ (the proof is similar if $\tilde j \!\in\! J_2$), then $f\!<\!\tilde j \!\le\! r$. Condition (1) shows that $k^+_{\tilde j }= \tilde j$. Then conditions (2) and (3) show that there is no $\bar\ell\in J_2$ with $k^+_{\tilde j }= k^+_{\bar\ell }$ or $k^+_{\tilde j }= k^+_{\bar\ell }+m$. Since all factors of $\tilde u$ have the form , we see that $z_1$ cannot produce the basis element $\tilde v$. Thus we can suppose $\tilde d_1=d_1$. Then $c^{\g_i+\tilde\alpha_i}(k_{(i)\tilde d_1d_1^{-1}})=c^{\alpha_i-\tilde\alpha_i}(k_i)$ (note that $c^2=1$ acting on $I$). If $\alpha_i\ne\tilde\alpha_i$ for some $1\le i\le f$, then $c^{\alpha_i-\tilde\alpha_i}(k_i)=-k_i$ and after applying $x'^{\beta}$ to the element inside the bracket in , we obtain an element which satisfies the condition that either its $i$-th label is not $i$ (in this case after we apply $e^f$ we obtain the zero element) or else its zero-th factor cannot contain the factor $\prod_{i=1}^rf^{\tilde\b_i}_{i_{\bullet},i}$. In any case we cannot obtain the element $\tilde v$. Thus we can assume $\alpha_i=\tilde\alpha_i$ for $1\le i\le f$. Similarly, we can assume $\alpha_i=\tilde\alpha_i$ for $f<i\le r$, i.e., $\alpha=\tilde\alpha$.
By conditions (1) and (2), we see that if $\beta_i\ne\tilde\b_i$ for some $i$ with $1\le i\le f$, or $\beta_i=1>\tilde\b_i$ for some $i\in J_1$, then again $z_1$ cannot produce the basis element $\tilde v$. Thus we suppose: $\beta_i\!=\!\tilde\b_i$ if $1\!\le\! i\!\le\! f$, and $\beta_i\le\tilde\b_i$ for $i\!\in\! J_1$. If $\tilde\b_i=1$ but $\b_i=0$ for some $i\in J_1$, then by , $z_1$ can only produce some basis elements which have at least a tensor factor, say $v_\ell$, with $\ell\in I^-$, and thus $\tilde v$ cannot be produced. Hence we can suppose $\b=\tilde\b$. Dually, we can suppose $\bar\beta\!=\!\bar{\tilde\beta}$.
Rewrite $wd_2\tilde d_2^{-1}\tilde w^{-1}$ as $wd_2\tilde d_2^{-1}\tilde w^{-1}=d_{20}\tilde d_{20}^{-1}w'$, where $d_{20}=wd_2w^{-1}$, $\tilde d_{20}=w\tilde d_2w^{-1}$ and $w'=w\tilde w^{-1}$. Note that $w'\in \mathfrak S_{r-f} \times \bar{\mathfrak S}_{t-f}$, which only permutes elements of $(J_1\cup J_2)\bs J_{f}$. We see that if $ d_{20}\ne \tilde d_{20}$, then as in , there exists some $j\in J_{f}$ with $\tilde j :=(j)d_{20}\tilde d_{20}^{-1}w'\notin J_{f }$, thus $\tilde u_{ M}$ cannot be produced. So assume $ d_{20}=\tilde d_{20}$. Similarly we can suppose $w'=1$. Then the same arguments after show that we can assume $\bar\alpha=\bar{\tilde\alpha}$ (cf. ).
The above has in fact proved that if the coefficient $\chi^{z_1}_{\tilde v}$ is nonzero then $z_1$ in must satisfy $(d_1,\alpha,\beta,f,{\bar\beta},{\bar\alpha},w,d_2)=
(\tilde d,{\tilde \alpha},{\tilde \beta},{\tilde f},{\bar{\tilde \beta}},{\bar{\tilde \alpha}},\tilde w,\tilde d_2)$, i.e., $z_1=(v)\tilde{\tif m}$. In this case, one can easily verify that $\chi^{z_1}_{\tilde v}\!=\!\pm1$. This proves that $z$ defined in is nonzero, a contradiction. The theorem is proven.
\[level-2\] $BC_{2, r, t}$ has a $\mathbb C$-basis which consists of all regular monomials of it.
We have the result immediately from Corollary \[level-l-span\] and Theorem \[theo-2222\].
Homomorphisms between $\widetilde{BC}_{r, t}$ and $BC_{2, r+k, t+k}$
=====================================================================
In this section, we generalize Theorem \[level-1\] so as to establish infinite many homomorphisms from $ BC^{\rm aff}_{r, t}$ to $BC_{2, r+k, t+k}$ for all positive integers $k$, where $BC_{2, r+k, t+k}$ are level two walled Brauer-Clifford superalgebras which appear in the higher level mixed Schur-Weyl-Sergeev duality in section 4. As an application, we prove that $ BC^{\rm aff}_{r, t}$ has $R$-basis which consists of all regular monomials in the Definition \[regm\]. Recall $x_i', \bar x_j'$ in .
\[usefaff\] For all admissible $i, j$, we have the following results in $BC_{r, t}^{\rm aff}$: $s_j\bar x'_i=\bar x'_i s_j$, $\bar s_j x'_i=x'_i\bar s_j$, $x'_i\bar c_j=\bar c_j x'_i$ and $\bar x'_i c_j=c_j\bar x'_i$.
Easy exercises.
\[usefaffco\] Recall $y_i$ and $\bar y_j$ in Definition $\ref{RSu-GRSS}$. The following results hold in $ BC_{r, t}^{\rm aff} $ for all admissible $i,j$:
[2]{}
1. $x'_i \bar y_i=\bar y_i x'_i$,
2. $\bar x'_i y_i=y_i \bar x'_i$,
3. $x'_{i+1} y_i= y_i x'_{i+1}$,
4. $\bar x'_{i+1} \bar y_i= \bar y_i \bar x'_{i+1}$,
5. $(\bar x'_{i+1}+\bar y_{i+1}) x_j'= x_j'(\bar x'_{i+1}+\bar y_{i+1})$ if $j\le i$,
6. $(x'_{i+1}+y_{i+1}) x_j'=x_j'(x'_{i+1}+y_{i+1})$, if $j\le i$,
7. $(\bar x'_{i+1}+\bar y_{i+1}) \bar x_j'= \bar x_j'(\bar x'_{i+1}+\bar y_{i+1})$ if $j\le i$,
8. $(x'_{i+1}+y_{i+1})\bar x_j'=\bar x_j'(x'_{i+1}+y_{i+1})$, if $j\le i$.
By symmetry, it is enough to prove (1), (3), (5), (6).
\(1) If $j\le i-1$, then $x_i' e_{j, i}=e_{j, i}x_i'$, $x'_i\bar c_j=\bar c_j x'_i$, and $\bar s_j x'_i=x'_i\bar s_j$ by Lemmas \[relxpri\](1) and \[usefaff\]. So, $$x'_i \bar y_i=x_i'\Big(\mbox{$\sum\limits_{j=1}^{i-1}$} (e_{j, i}-\bar e_{j, i})-\bar L_i\Big)=\bar y_i x'_i.$$ One can check (3) via Definition \[RSu-GRSS\] similarly.
\(5) By Lemmas \[hecrel2\](1), \[relxpri\](1)–(2), $x'_j(\bar x'_{i+1}+e_{j, i+1}-\bar e_{j, i+1})=(\bar x'_{i+1}+e_{j, i+1}-\bar e_{j, i+1})x'_j$ and $x'_j e_{s, i+1 }=e_{s, i+1 }x_j'$ and $x'_j \bar e_{s, i+1 }=\bar e_{s, i+1 }x_j'$ whenever $j\neq s$. Since $x'_j \bar L_{i+1}=\bar L_{i+1} x_j'$, we have $$\begin{aligned} (\bar x'_{i+1}+\bar y_{i+1})x_j' &=\Big(\bar x'_{i+1}+e_{j, i+1}-\bar e_{j, i+1}+\mbox{$\sum\limits_{1\le s\le i, s\neq j}$}(e_{s, i+1 }-\bar e_{s, i+1})-\bar L_{i+1}\Big)x_j'\\
&= x_j'\Big(\bar x'_{i+1}+e_{j, i+1}-\bar e_{j, i+1}+\mbox{$\sum\limits_{1\le s\le i, s\neq j}$}(e_{s, i+1 }-\bar e_{s, i+1})-\bar L_{i+1}\Big)\\ & =x_j' (\bar x'_{i+1}+\bar y_{i+1}).\\
\end{aligned}$$
\(6) By , and Lemma \[relxpri\](1), we have $$\begin{aligned} x_1(x_{i+1}'+y_{i+1})& =x_1\Big(x_{i+1}+\mbox{$\sum\limits_{j=1}^{i}$} (e_{i+1, j}+\bar e_{i+1, j})\Big)= \Big(x_{i+1}+\mbox{$\sum\limits_{j=1}^{i}$} (e_{i+1, j}+\bar e_{i+1, j})\Big)x_1\\ &= (x_{i+1}'+y_{i+1})x_1.\\ \end{aligned}$$ Applying $(1, j)$ on both sides of the above equation yields (6).
For the simplification of notation, we define $$\label{zi} z_i=x_i'+y_i \text{ and $\bar z_j=\bar x_j'+\bar y_j$}$$ for all admissible $i$ and $j$.
\[usefcycl\]The following results hold in $ BC_{r, t}^{\rm aff} $ for all admissible $i, j$:
[2]{}
- $s_j z_i =z_i s_j$, $\bar s_j \bar z_i=\bar z_i \bar s_j$, if $j\neq i-1, i$,
- $s_j \bar z_i=\bar z_i s_j$, $\bar s_j z_i =z_i \bar s_j$, if $j\neq i-1$,
- $z_i c_i=-c_iz_i $, $\bar z_i \bar c_i=-\bar c_i \bar z_i$,
- $z_i c_j=c_j z_i $, $\bar z_i \bar c_j=\bar c_j \bar z_i$, if $i\neq j$,
- $z_i\bar c_j=\bar c_j z_i$, $\bar z_i c_j=c_j\bar z_i$, if $i\leq j$,
- $z_i(e_i+\bar z_i-\bar e_i)=(e_i+\bar z_i-\bar e_i)z_i $,
- $e_{i} \bar z_i=-e_i( x_i+\bar L_i)$, $e_{i} z_i=e_i(x_i+\bar L_i)$,
- $e_i s_i z_is_i=s_i z_i s_i e_i$, $e_j \bar s_j \bar z_j\bar s_j=\bar s_j \bar z_j\bar s_j e_j$,
- $z_i\tilde z_i =\tilde z_iz_i$,
- $\bar z_i\tilde{\bar z_i} =\tilde{\bar z_i}\bar z_i$,
- $e_i z_i e_i=e_i x'_i e_i=\omega_1 e_i $,
- $ e_i z_i^k c_i e_i=0$, $\forall k\in \mathbb N$,
- $e_i z_i^{2n} e_i=0 $, $e_i (\bar z_i)^{2n} e_i=0$ $\forall n\in \mathbb N$,
where $\tilde z_i=(s_i z_i s_i -(1-c_ic_{i+1})s_i)$ and $\tilde{\bar z_i}=(\bar s_i \bar z_i \bar s_i -(1+\bar c_i\bar c_{i+1})\bar s_i)$.
(1)–(5) follows from Lemma \[usef\](1)–(5), Lemma \[usefaff\] and and . (6) follows from Lemmas \[usef\](7), \[hecrel2\](1), \[usefaffco\](1)–(2). (7) follows from Lemmas \[usef\](8), \[hecrel2\](3)–(4). (8) follows from Lemmas \[usef\](9), \[relxpri\](1). Multiplying $(2, i+1)(1, i)$ on both sides of $x_1x_2=x_2x_1$ (see ) yields $$\label{bii1} x_i'(x_{i+1}'-(1-c_ic_{i+1})s_i)=(x_{i+1}'-(1-c_ic_{i+1})s_i)x_i'.$$ Now, (9) follows from , Lemmas \[usef\](10), \[usefaffco\](3)–(4). We leave (10) to the reader since it can be verified, similarly. (11) follows from Lemmas \[usef\](13), \[relxpri\](3). Via (6), one can prove (12) by arguments similar to those for Lemma \[usef\](11). Finally, one can verify (13) by arguments similar to those for Lemma \[usef\](12).
From here to the end of Theorem \[alghomcyco\], we assume the ground ring $R$ is $\mathbb C$. Also, $BC_{2, r, t}$ is one of those which appear in the higher level mixed Schur-Weyl-Sergeev duality in section 4.
\[step2\] The $e_k BC_{2, k, k}$ is the left $BC_{2, k-1, k-1}$-module spanned by $e_k c_k^{\sigma_1}x_k'^{\sigma_2} s_{k, j} \bar s_{k, l}$ for all $ \sigma_1, \sigma_2\in \Z_2$ and $ 1\le j, l\le k$, where $BC_{2, k-1, k-1}$ is the subalgebra of $BC_{2, k, k}$ generated by $c_1, \bar c_1$,$ x_1, \bar x_1$, $ s_1, \ldots, s_{k-2}$ and $\bar s_1, \ldots, \bar s_{k-2}$.
This result, which is a counterpart of Lemma \[spannwbc\], can be proved similarly.
\[esecycl\]
1. $e_k BC_{2, k, k} e_k=e_{k} BC_{2, k-1, k-1}$.
2. Recall $z_k$ and $\bar z_k$ in . There is a unique $\xi_{a, k}$ $($resp., $\bar \xi_{a, k}$$)$ in $BC_{2,k-1, k-1}$ such that $e_k z_k^a e_k=\xi_{a, k} e_k$ (resp., $e_k \bar z_k^a e_k=\bar \xi_{a, k} e_k)$. Moreover, $\xi_{2n,k}=\bar \xi_{2n,k}=0$, and $\xi_{1, k}=\omega_1$.
\(1) follows from Lemma \[step2\] (see the proof of Proposition \[ese1\]) and (2) follows from (1) and Corollary \[level-2\] for $BC_{2, k, k}$ and Lemma \[usefcycl\](11), (13).
Lemmas \[assumcycl\] and \[assum1cycl\] can be proven by arguments similar to those for Lemma \[y1\]–\[y2\].
\[assumcycl\] For any $n\in \mathbb N$, $e_i\bar z_i^{2n+1}=\sum_{j=0}^n a^{(i)}_{2n+1, j} e_iz_i^{2j+1}$ for some $a^{(i)}_{2n+1,j}\in R[ \xi_{3,i}, \ldots, \xi_{2n-1,i}]$ such that
1. $a^{(i)}_{2n+1, n}=-1$,
2. $a^{(i)}_{2n+1, j}=a^{(i)}_{2n-1, j-1}$ for all $1\le j\le n-1$,
3. $a^{(i)}_{2n+1, 0}=\sum_{j=0}^{n-1} a^{(i)}_{2n-1, j}\xi_{2j+1,i}$.
\[assum1cycl\] For any positive integer $n$, $e_i\bar z_i^{2n}=\sum_{j=0}^n a^{(i)}_{2n, j} e_i z_i^{2j}$ for some $a^{(i)}_{2n,j}\in R[ \xi_{3,i}, \ldots, \xi_{2n-1,i}]$ such that
1. $a^{(i)}_{2n, n}=1$,
2. $a^{(i)}_{2n, j}=a^{(i)}_{2n-1, j-1}$ for all $1\le j\le n-1$,
3. $a^{(i)}_{2n, 0}=\sum_{j=0}^{n-1} a^{(i)}_{2n-2, j}\xi_{2j+1,i}$.
We can assume $k \ge 2$ (resp., $n\ge 1$) in the Lemma \[free1cycl\] since $z_k=x_k'+y_k$ and $\bar z_k=\bar x_k'+\bar y_k$ (resp., $\xi_{1, k}=\omega_1$ and $\bar \xi_{1, k}=-\omega_1$ by Lemma \[usefcycl\](11)).
\[free1cycl\] We have $\bar \xi_{2n+1, k}\in R[\xi_{3, k}, \ldots, \xi_{2n+1, k}]$ if $k \in\Z^{\ge2}$ and $n\in \mathbb Z^{\ge 1}$. Furthermore, both $\xi_{2n+1, k}$ and $\bar\xi_{2n+1, k}$ are central in $BC_{2,k-1, k-1}$.
The first statement follows from Lemma \[assumcycl\]. We have $s_jz_k=z_ks_j$ and $c_jz_k=z_kc_j$ for all $j\le k-1$ by Lemma \[usefcycl\](1)–(4). Since $y_k=\sum_{i=1}^{k-1} (e_{k, i}+\bar e_{k, i})-L_k$£¬ $z_k=x_k'+y_k=x_k+\sum_{i=1}^{k-1} (e_{k, i}+\bar e_{k, i})$. By and Lemma \[relxpri\](1) and (2), $x_1z_k=z_kx_1$ for $k\ge 2$. Obviously, $e_1$ commutes with $x_k'$ and $e_{k, i}, \bar e_{k, i}, c_k, (k, i)$ whenever $i\neq 1$ and $k\ge 2$. Since $e_1 e_{k, 1}=e_1 (k, 1)$, we have $e_1z_k=z_ke_1$.
We have proved that $h$ commutes with $e_k, z_k$ for any $h\in \{e_1, s_i,c_j,x_1 \mid 1\le i\le k-2\ ,1\le j\le k-1\} $. So $e_k (h \xi_{a, k})=e_k(\xi_{a, k} h)$. By Corollary \[level-2\] and Proposition \[esecycl\], $h \xi_{2n+1, k}=\xi_{2n+1, k} h$. Finally, we need to check $e_k (h \xi_{a, k})=e_k(\xi_{a, k} h)$ for any $h\in \{\bar s_i,\bar c_j,\bar x_1 \mid 1\le i\le k-2\ ,1\le j\le k-1\}$. In this case, we use Lemma \[assumcycl\] so as to use $\bar z_k $ instead of $z_k$ in $e_k z_k^{2n+1} e_k$. Therefore, $h \xi_{2n+1, k}=\xi_{2n+1, k} h$, as required.
For $k,a\in\Z^{\ge 1}$, we have $$\label{sykcycl}
s_k z_{k+1}^a =h_k^a s_k-\mbox{$\sum\limits_{b=0}^{a-1}$} h_k^{a-1-b}z_{k+1}^b +\mbox{$\sum\limits_{b=0}^{a-1}$} (-1)^{a-b} c_kc_{k+1} h_k^{a-b-1}
z_{k+1}^b,$$ where $h_k=z_k+e_k+\bar e_k$.
The result can be easily checked by induction on $a$.
Recall $z_i, \bar z_j$ in for all admissible $i$ and $j$. For all $1\le j\le k-1$, define $$\label{z-bar-zcycl} p_{j, k}=s_{j, k-1} (z_{k-1} +e_{k-1}+\bar e_{k-1})s_{k-1, j}, \ \ \text{and $\bar p_{j, k}=\bar s_{j, k-1} (\bar z_{k-1}+e_{k-1}-\bar e_{k-1})\bar s_{k-1, j}$.}$$ Note that $\xi_{0,k}=0$, and $e_k h=0$ for $h\in BC_{2,k-1, k-1}$ if and only if $h=0$. We will use this fact freely in the proof of the following lemma, where we use the terminology that a monomial in $p_{j, k+1}$’s $\bar p_{j, k+1}$’s, and $ x_j', \bar x_j'$ is a [*leading term*]{} in an expression if it has the highest degree by defining ${\rm deg\,}p_{i,j}={\rm deg\,}\bar p_{i,j}= {\rm deg\,}x'_j={\rm deg\,}\bar x'_j=1$.
\[omedcycl\] For any positive integer $n$, $\xi_{2n+1, k+1}$ can be written as an $R$-linear combination of monomials in $ p_{j, k+1}$, $\bar p_{j, k+1}$, $x'_j$ and $\bar x'_j$ for $1\le j\le k$ such that the leading terms of $\xi_{2n+1, k+1}$ are $-2\sum_{j=1}^{k}(p_{j, k+1}^{2n}+p_{j, k+1}^{2n-1} x_j'-\bar p_{ j, k+1}^{2n}-\bar p_{j, k+1}^{2n-1}\bar x_j')$.
We have $x_1^2=p(p+1)$ and $\bar x_1^2=(p-n+1)(p-n)$ in $BC_{2, r, t}$ by Lemma \[SMSM\]. So, $$\label{ommmmcycl}{\sc\!}
\begin{aligned}
& \xi_{2n+1,k+1} e_{k+1}\overset{\rm Prop.~5.5} = e_{k+1}z_{k+1}^{2n+1} e_{k+1}=e_{k+1}y_{k+1} z_{k+1}^{2n}e_{k+1}+e_{k+1}x'_{k+1}z_{k+1}^{2n}e_{k+1}\\ = & e_{k+1}y_{k+1}
z_{k+1}^{2n}e_{k+1}+e_{k+1}(x'_{k+1})^2 z_{k+1}^{2n-1}e_{k+1}+e_{k+1}x'_{k+1}y_{k+1}z_{k+1}^{2n-1}e_{k+1}\\ = & e_{k+1}y_{k+1}z_{k+1}^{2n}e_{k+1}+
p(p+1)e_{k+1} z_{k+1}^{2n-1}e_{k+1}-e_{k+1}\bar x'_{k+1}y_{k+1}z_{k+1}^{2n-1}e_{k+1} \text{ (by Lemma~\ref{hecrel2}(3))}\\= &e_{k+1}y_{k+1}z_{k+1}^{2n}e_{k+1}+p(p+1)e_{k+1} z_{k+1}^{2n-1}e_{k+1}+e_{k+1}\bar y_{k+1}\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1} \text{ (Lemmas~\ref{usef}(8),~\ref{usefaffco}(2))} \\=& e_{k+1}(-L_{k+1}+\bar L_{k+1})z_{k+1}^{2n}e_{k+1}+p(p+1)\xi_{2n-1, k+1} e_{k+1} +e_{k+1}(L_{k+1}-\bar L_{k+1} )\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}.
\end{aligned}\!\!$$ Recall $h_k=z_k+e_k+\bar e_k$ in . Considering the right-hand side of and expressing $L_{k+1}$ by , using $(j,k+1)=$ $s_{j,k}s_ks_{k,j}$ and the fact that $s_{j,k},s_{k,j}$ commute with $x'_{k+1},y_{k+1},e_{k+1}$, a term in the linear combination of $e_{k+1} L_{k+1}z_{k+1}^{2n}e_{k+1}$ becomes $$\label{step44} s_{j, k} e_{k+1} s_{k} z_{k+1}^{2n} e_{k+1}
s_{ k, j}\! +
s_{j, k} e_{k+1}c_{k+1} s_{k} c_{k+1} z_{k+1}^{2n} e_{k+1} s_{ k, j}.$$ Since $c_{k+1} x_{k+1}'=-x_{k+1}'c_{k+1}$ and $c_{k+1} y_{k+1}=-y_{k+1} c_{k+1}$, we have $c_{k+1} z_{k+1}^{2n}=z_{k+1}^{2n} c_{k+1}$. Note that $e_{k+1}c_{k+1}=e_{k+1}\bar c_{k+1}$, and $\bar c_{k+1}$ commutes with $s_k, x_{k+1}'$ and $y_{k+1}$. So $$\label{step45} \begin{aligned} & s_{j, k} e_{k+1}c_{k+1} s_{k} c_{k+1} z_{k+1}^{2n} e_{k+1} s_{ k, j}=s_{j, k} e_{k+1} \left(s_{k} z_{k+1}^{2n}\right) e_{k+1}
s_{ k, j}\! \\
\overset { \eqref{sykcycl}}=& s_{j, k} e_{k+1}\left( h_k^{2n} s_k -\mbox{$\sum\limits_{b=0}^{2n-1}$} h_k^{2n-b-1} z_{k+1}^b
+(-1)^{b} \mbox{$\sum\limits_{b=0}^{2n-1}$} c_k c_{k+1} h_k^{2n-b-1} z_{k+1}^b \right)e_{k+1}s_{k, j} \\=&s_{j, k} h_k^{2n}s_{k,j}e_{k+1}-s_{j, k}\mbox{$\sum\limits_{b=0}^{2n-1}$} h_k^{2n-b-1} s_{k, j} \xi_{b,k+1}e_{k+1}\text{ (by Lemma~\ref{usefcycl}(12)).}\end{aligned}$$ By induction assumption, the leading terms of $\xi_{b, k+1}$ are of degree $b-1$. So, the leading term of $s_{j, k} e_{k+1} \left(s_{k} z_{k+1}^{2n}\right) e_{k+1}
s_{ k, j}$ is $p_{j, k+1}^{2n}$ and hence $-e_{k+1} L_{k+1}z_{k+1}^{2n}e_{k+1}$ contributes to the leading terms $-2\sum_{j=1}^k p_{j, k+1}^{2n}$.
We compute $e_{k+1} L_{k+1}\bar x_{k+1}'z_{k+1}^{2n-1}e_{k+1}$. A term of it becomes $2s_{j, k} e_{k+1}s_k\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}s_{k,j}$ since $$e_{k+1}c_{k+1}s_k c_{k+1}\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}= e_{k+1}s_k\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}.$$ Thus, it is enough to compute the leading terms of $s_{j, k} e_{k+1}s_k\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}s_{k,j}$. Since $x_k'$ commutes with $z_{k+1}$ and $e_{k+1}$, and $e_{k+1} s_k\bar x_{k+1}'=e_{k+1} \bar x_{k+1}'s_k=-e_{k+1} x_{k+1}'s_k=-e_{k+1} s_k x_{k}'$, we have $$s_{j, k} e_{k+1}s_k\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}s_{k,j}=-s_{j, k} e_{k+1}s_k z_{k+1}^{2n-1}e_{k+1}s_{k,j} x'_{j}$$ By , $s_{j, k} e_{k+1}s_k\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}s_{k,j}$ contributes to the leading term $\-p_{j, k+1}^{2n-1}x_j'$ whose degree is $2n$. Finally, we need to compute the leading terms of $e_{k+1} \bar L_{k+1}z_{k+1}^{2n}e_{k+1}$ and $-e_{k+1}\bar L_{k+1} \bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}$, By Lemma \[assum1cycl\], one can use $\bar z_{k+1}^{2n}$ to replace $z_{k+1}^{2n}$ in $e_{k+1} \bar L_{k+1}z_{k+1}^{2n}e_{k+1}$. Thus, $e_{k+1} \bar L_{k+1}z_{k+1}^{2n}e_{k+1}$ contributes to the leading terms $\bar p_{j, k+1}^{2n}$. Similarly, a term of $e_{k+1}\bar L_{k+1}\bar x'_{k+1}z_{k+1}^{2n-1}e_{k+1}$ is of form $\bar x_j' e_{k+1} (\bar j, \overline {k+1}) z_{k+1}^{2n-1} e_{k+1}$ whose leading term $\bar p_{j, k+1}^{2n-1}\bar x_j'$ is of degree $2n$. The proof is completed.
\[commomegacycl\]For $a\!\in\!\Z^{\ge0},{\ssc\,}k\!\in\!\Z^{\ge 1}$, both $\xi_{a, k+1}$ and $\bar \xi_{a, k+1}$ commute with $x'_{k+1}+y_{k+1}$ and $\bar x'_{k+1}+\bar y_{k+1}$.
By Proposition \[esecycl\], we can assume that $a=2n+1$ and $n\geq 1$. By Lemma \[omedcycl\], $\xi_{2n+1, k}$ can be written as a linear combinations of monomials in $p_{j, k+1}$, $\bar p_{j, k+1}$, $x'_j$, and $\bar x'_j$ for $1\le j\le k$. From , $$\label{pze} p_{j, k+1}=x_j'+z_{j, k+1}, \text{ and $\bar p_{j, k+1}=\bar x_j'+\bar z_{j, k+1}$,}$$ where $z_{j,k+1} ,\bar z_{j,k+1}$ are defined in Lemma \[dzjk\]. So, it is enough to prove that $z_{j, k+1}$, $\bar z_{j, k+1}$, $x'_j$ and $\bar x'_j$ commute with $\bar x'_{k+1}+\bar y_{k+1}$ if $1\le j\le k$,. By Lemma \[usefaffco\](5)–(6), both $x'_j$ and $\bar x'_j$ commute with $\bar x'_{k+1}+\bar y_{k+1}$. Finally, $z_{j, k+1}$, $\bar z_{j, k+1}$ commute with $\bar x'_{k+1}+\bar y_{k+1}$ since (1): both $z_{j, k+1}$ and $\bar z_{j,k+1}$ commute with $y_{k+1}$ and $\bar y_{k+1}$ (see the proof of Lemma \[commomega\]) and (2): $z_{j,k+1}$ and $\bar z_{j,k+1}$ are linear combinations of elements which commutes with both $\bar x'_{k+1}$ and $ { x'_{k+1}}$, by Lemmas \[dzjk\], \[relxpri\](1)–(2), and \[usefaff\] and and . This proves that $\xi_{a, k+1}$ commutes with $x'_{k+1}+y_{k+1}$ and $\bar x'_{k+1}+\bar y_{k+1}$ and so is $\bar \xi_{a, k+1}$ by Lemma \[assumcycl\].
\[commom1egacycl\]For $a\!\in\!\Z^{\ge0},{\ssc\,}k\!\in\!\Z^{\ge 1}$, both $\xi_{2a+1, k+1}$ and $\bar \xi_{2a+1, k+1}$ commute with $c_j$ and $\bar c_j$ if $j \geq k+1$.
We have proved that both $z_{j, k+1}$ and $\bar z_{j, k+1}$ commute with $c_j$ and $\bar c_j$ if $j \geq k+1$ in the proof of Lemma \[commomega\]. If $j\le k$, then $ x_j'$ and $\bar x_j $ commute with $c_l$ and $\bar c_l$ for $l\ge k+1$, by Lemma \[usefaff\] and and . Now, the result follows from and Lemma \[omedcycl\], immediately.
In the following result, we assume the ground field is $\mathbb C$ since we use level two walled Brauer-Clifford superalgebras in section 4. After we have proved the freeness of cyclotomic walled Brauer-Clifford sueralgebras in section 6, we know that the following result is available over an integral domain $R$.
\[alghomcyco\] For any $k\in \mathbb Z^{>0}$, there is a superalgebra homomorphism $\phi_k: BC_{r,t}^{\rm aff}\rightarrow BC_{2, r+k,t+k}$ sending $\omega_{2n+1}, \bar \omega_{2n+1}, e_1, s_i, \bar s_j, c_l, \bar c_m, x_1, \bar x_1$ to $\xi_{2n+1, k+1}, \bar \xi_{2n+1, k+1}, e_{k+1}, s_{k+i}, \bar s_{k+j}, c_{k+l}, \bar c_{k+m}, z_{k+1}$, $ \bar z_{k+1}$ respectively, for all admissible $i, j, l, m, n$.
It is enough to verify the images of generators of $BC_{r, t}^{\rm aff}$ satisfy the defining relations for $BC_{r, t}^{\rm aff}$ in Definition \[awbsa\]. If so, $\phi_k$ is an algebra homomorphism. Since $\phi_k$ sends even (resp., odd) generators to even (resp., odd) elements in $BC_{2, r+k,t+k}$, $\phi_k$ is a superalgebra homomorphism.
By Corollary \[wsba-1\], $\phi_k$ satisfies – and Definition \[wsera\](1)–(10). By Lemma \[usefcycl\](1),(3),(9),(10), $\phi_k$ satisfies and . Applying anti-involution $\sigma$ on Lemma \[usef\](7), we see that $\phi_k$ satisfies the Definition \[awbsa\](1). By Lemma \[usefcycl\](8), (6) (resp., (11)–(13)), $\phi_k$ satisfies Definition \[awbsa\](2)–(4) (resp., (5)–(7)). Finally, $\phi_k$ satisfies Definition \[awbsa\](8)–(11) by Lemma \[usefcycl\](2), (5).
The following result is a counterpart of [@RSu Theorem 4.14].
\[main1cyco\] Suppose $R$ is a domain which contains $2^{-1}$ and $ \omega_1$. Then $ BC_{r,t}^{\rm aff}$ is free over $R$ spanned by all regular monomials in Definition $\ref{regm}$. In particular, $BC_{r,t}^{\rm aff}$ is of infinite rank.
By Proposition \[spanned\], it is enough to prove that $M$, the set of all regular monomials of $BC_{r, t}^{\rm aff},$ is linear independent over $\mathbb Z[\omega_1, 2^{-1}]$, where $\omega_1$ is an indeterminate. By fundamental theorem on algebras, it suffices to prove it for sufficient many $\omega_1$’s. This can be done by choosing $\omega_1$ as in . So, it is enough to prove that $M$ is linear independent over $\mathbb C$ for infinite many $\omega_1$’s in .
By Lemma \[hecrel\](1)–(3) and Definition \[regm\], we assume that a regular monomial $\mathbf{m}$ of $BC_{r, t}^{\rm aff} $ is of form $$\label{regmm} \mathbf{m}= c^{\alpha} x^{\beta} d_1^{-1} e^f wd_2 {\bar x}^{\gamma} {\bar c}^{\delta}
\mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}},$$ where $(\alpha, \delta)\in \mathbb Z_2^r\times \mathbb Z_2^t$ and $(\beta, \gamma)\in \mathbb N^r\times \mathbb N^t$ and $d_1, d_2\in D_{r, t}^f$ and $w\in \Sigma_{r-f}\times \Sigma_{\overline{t-f}}$ and $0\le f\le \min\{r, t\}$. So, it is equivalent to prove that the above regular monomials are linear independent. If it were false, then there is a finite subset $S\subset M$ such that $\sum_{\mathbf{ m}\in S} r_{\textbf{m}} {\textbf {m}}=0$ and $r_{\mathbf {m}}\neq 0$ for all $\textbf {m}\in S$. For each $\mathcal S$, we set $$\label{k-kmcyco}\begin{aligned} & \tilde k =\max\big\{|\beta|+\SUM{n}{} (2n+1) a_{2n+1}\mid
c^{\alpha} x^{\beta} d_1^{-1} e^f wd_2 {\bar x}^{\gamma} {\bar c}^{\delta}
\mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in \mathcal S \big\},\\
&
\hat k=\max\big\{|\gamma|+\SUM{n}{} (2n+1) a_{2n+1}\mid c^{\alpha} x^{\beta} d_1^{-1} e^f wd_2 {\bar x}^{\gamma} {\bar c}^{\delta}
\mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in \mathcal S \big\}.
\\ \end{aligned}$$ If $\tilde k\geq \hat k$, we define $k=\tilde k$ and $$\label{k11} \begin{aligned}
& f_0=\min\big\{f\mid c^{\alpha} x^{\beta} d_1^{-1} e^f wd_2 {\bar x}^{\gamma} {\bar c}^{\delta} \mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in S, |\beta|+\SUM{n}{} (2n+1) a_{2n+1}=k\,\big\},\\
& k_1= \max\big\{|\gamma|\mid c^{\alpha} x^{\beta} d_1^{-1} e^{f_0} wd_2 {\bar x}^{\gamma} {\bar c}^{\delta} \mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in S, |\beta|+\SUM{n}{} (2n+1) a_{2n+1}= k\,\big\}.\\
\end{aligned}$$ If $\tilde k<\hat k$, we define $k=\hat k$ and $$\label{k12}\begin{aligned}& f_0=\min\big\{f\mid c^{\alpha} x^{\beta} d_1^{-1} e^f wd_2 {\bar x}^{\gamma} {\bar c}^{\delta} \mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in S, |\gamma|+\SUM{n}{} (2n+1) a_{2n+1}=k\,\big\}, \\ & k_1= \max\big\{|\beta|\mid c^{\alpha} x^{\beta} d_1^{-1} e^{f_0} wd_2 {\bar x}^{\gamma} {\bar c}^{\delta} \mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}}\in S, |\gamma|+\SUM{n}{} (2n+1) a_{2n+1}= k\big\}.\\
\end{aligned}$$
Let $\phi_k{\ssc\!}:\! BC_{r,t}^{\text{\rm aff}}{\ssc\!}\!\rightarrow {\ssc\!}\! BC_{2,r+k,t+k}(\omega_1)$ be the superalgebra homomorphism in Theorem \[alghomcyco\]. Since $x_i =x_i'-L_i $ and $\bar x_i=\bar x_i'-\bar L_i$, $$\label{ximage} \phi_k(x_i)=\mbox{$\sum\limits_{j=1}^k$} (e_{k+i, j}+\bar e_{k+i, j})+x'_{k+i}-L_{k+i},\ \text{and}\ \phi_k(\bar x_i)=\mbox{$\sum\limits_{j=1}^k$}( e_{j, k+i}-\bar e_{j, k+i})+\bar x'_{k+i} -\bar L_{k+i}.$$ Using Lemma \[omedcycl\] to express $\xi_{2n+1,k+1}$ for $n\in \mathbb Z^{\ge 1}$, we see that some terms of $\phi_k(\mathbf m)$ are of forms (we will see in the next paragraph that other terms of $\phi_k(\mathbf m)$ will not contribute to our computations) $$\label{prod1cycl} \mbox{$\prod_{i=1}^r { c}_{i+k}^{\alpha_i}\prod\limits_{i=1}^r$} (k\! +\!i, i_{1}) \cdots (k\!+\!i, i_{\beta_i}) \phi_k (d_1^{-1} e^f w d_2)
\mbox{$\prod\limits_{j=1}^t$} (\overline{k\!+\!j}, \bar j_1) \cdots (\overline{k\!+\!j}, \bar j_{\gamma_j})\mbox{$\prod\limits_{j=1}^t$} { \bar c}_{j+k}^{\delta_j} \mbox{$\prod\limits_{n\ge 1}$} \mathbf c_{2n+1},$$ where $\mathbf c_{2n+1}$, which comes from $\xi_{2n+1, k+1}$, ranges over products of $a_{2n+1}$ disjoint cycles in $\Sigma_k$ (or $\bar {\Sigma}_k)$ such that each cycle is of length $2n+1$.
By Theorem \[theo-2222\], $BC_{r, t}$ is a subalgebra of $BC_{2, r, t}$ and so is the walled Brauer algebra, say $B_{r, t}(0)$ which is isomorphic to the subalgebra generated by $e_1, s_1, \ldots, s_{r-1}$ and $\bar s_1, \ldots, \bar s_{t-1}$. Similarly, we have the subalgebra $B_{r+k, t+k}(0)$ of $BC_{2, r+k, t+k}$. It is known that $B_{r, t}(0)$ can be defined by so called $(r, t)$-walled Brauer diagrams. Each of them is a diagram with $(r\!+\!t)$ vertices on the top and bottom rows, and vertices on both rows are labeled from left to right by $r, \ldots,2, 1, \bar 1, \bar 2, \ldots, \bar t$. Every vertex $i\!\in\!\{1, 2, \ldots, r\}$ (resp., $\bar i\!\in $ $\{\bar 1, \bar 2, \ldots, \bar t\}$) on each row must be connected to a unique vertex $\bar j$ (resp., $j$) on the same row or a unique vertex $j$ (resp., $\bar j$) on the other row. The pairs $[i, j]$ and $[\bar i, \bar j]$ are called [*vertical edges*]{}, and the pairs $[\bar i, j]$ and $[i, \bar j]$ are called [*horizontal edges*]{}. By definition, a $\phi_k(d_1^{-1} e^f w d_2)$ in corresponds to a unique $(r+k, t+k)$-walled Brauer diagram such that $[i, i]$ and $[\bar j, \bar j]$ are its vertical edges for all $1\le i, j\le k$ (see e.g. [@RSu]). We call the terms of the form (\[prod1cycl\]) the [*leading terms*]{} if
- $k=|\beta|+\sum_n (2n+1) a_{2n+1}$ if $\tilde k\geq \hat k$ and $k=|\gamma|+\sum_n(2n+1) a_{2n+1}$ if $\tilde k<\hat k$. (cf. ),
- the corresponding $f$ in (\[prod1cycl\]) is $f_0$ in ,
- $|\gamma|=k_1$ if $\tilde k\geq \hat k$ and $|\beta|=k_1$ if $\tilde k<\hat k$,
- in the first case of (i), the juxtapositions of the sequences $i_1, i_2, \ldots, i_{\beta_i}$ for $1\!\le \!i\!\le\! r$ and $\mathbf c_{2n+1}$, $n\ge 1$ run through all permutations of the sequences in $1, 2, \ldots, k$ and the sequences $\bar j_1, \bar j_2, \ldots,\bar j_{\gamma_j}$, $1\le j\le t$ run through all permutations of the sequence $\bar 1, \bar 2, \ldots, \bar k_1$; while in the second case of (i), the juxtapositions of the sequences $\bar j_1, \bar j_2, \ldots, \bar j_{\gamma_j}$ for $1\le j\le t$ and $\mathbf c_{2n+1}$, $n\ge 1$ run through all permutations of the sequences in $\bar 1, \bar 2, \ldots, \bar k$ and the sequence of $i_1, i_2, \ldots, i_{\beta_i}$, $1\le i\le r$ run through all permutations of sequence $1, 2, \ldots, k_1$.
By Theorem \[monom\], all $\tilde {\mathbf{m}}=c^{\tilde\alpha} x'^{\tilde \beta} \tilde d_1^{-1} e^{\tilde f} \tilde w\tilde d_2 {\bar x'}^{\tilde \gamma} {\bar c}^{\tilde \delta} \in BC_{2, r+k, t+k}$ consist of a basis of $BC_{2, r+k, t+k}$ over $\mathbb C$, where $\tilde \alpha, \tilde \beta\in \mathbb Z_2^{r+k}$, $\tilde \gamma, \tilde \delta\in \mathbb Z_2^{t+k}$, and $\tilde d_1, \tilde d_2\in D_{r+k, t+k}^{\tilde f}$, $\tilde w\in \Sigma_{r+k-\tilde f}\times \Sigma_{\overline{t+k-\tilde f}}$ and $0\le \tilde f\le \min\{r+k, t+k\}$. Such monomials will be called normal monomials. Moreover, $\tilde{\mathbf{m}}$ is called an admissible monomial if
1. $\tilde \alpha_i=\tilde \delta_j=0$ for all $1\le i\le k$, $1\le j\le k_1$ if $\tilde k\ge \hat k$; or $\tilde \alpha_i=\tilde \delta_j=0$ for all $1\le i\le k_1$ and $1\le j\le k$ if $\tilde k<\hat k$,
2. the corresponding walled Brauer diagram of $\tilde d_1^{-1} e^{\tilde f } \tilde w \tilde d_2$ satisfies (1)–(5) as follows:
1. $\tilde f= f_0$,
2. no vertical edge of form $[i, i]$ and $[\bar j, \bar j]$, $1\le i\le k$, $1\le j\le k_1$ if $\tilde k \geq \hat k$,
3. no vertical edge of form $[i, i]$ and $[\bar j, \bar j]$, $1\le i\le k_1$, $1\le j\le k$ if $\tilde k<\hat k$,
4. no horizontal edge of form $[i, \bar j]$, $1\!\le\! i\! \le\! k$, at the bottom row if $\tilde k \geq \hat k$,
5. no horizontal edge of form $[i, \bar j]$, $1\!\le\! j\! \le\! k$, at the top row if $\tilde k<\hat k$,
3. $\tilde \beta_i=\tilde \gamma_j=0$, for all $1\le i\le r+k$ and $1\le j\le t+k$.
In the following, we assume that $\tilde k<\hat k$ (the case $\tilde k\ge \hat k$ can be dealt with in a similar way). A $\phi_k(\mathbf{m})$ contributes admissible monomials of $BC_{2, k+r, k+t}$ only when $\mathbf{m}\in \mathcal S$ is given in such that $k=|\gamma|+\sum_{n} (2n+1) a_{2n+1}$, $ f=f_0$ and $k_1=|\beta|$. More explicitly, the leading terms exactly appear in $\phi_k(\mathbf m)$ which are admissible monomials of $BC_{2, k+r, k+t} $. We claim that other terms in $\phi_k( \sum_{\mathbf m\in \mathcal S} r_{\mathbf {m}} {\mathbf m})$ are obtained from by
- using the terms $e_{k+i, i_j}, \bar e_{k+i,i_j}$ of $ \phi_k(x_i)$ to replace some $(k+i, i_{j})$,
- using the terms $e_{i, j}, \bar e_{i, j}$ $1\le i, j\le k$ (resp., $\bar e_{j_i, k+j}, e_{j_i, k+j}$ of $\phi_k(\bar x_j)$) to replace ($\bar i, \bar j)$ in $\mathbf c_{2n+1}$ (resp., $(\overline{k+j}, \bar j_i)$) if (1) does not occur,
- using the term $x_{k+i}'$ of $\phi_k(x_i)$ to replace $(k+i, i_j)$ or using the term $\bar x_{k+i}'$ of $\phi_k(\bar x_i)$ to replace $(\overline{k+i}, \bar i_j)$, or using either $x_j'$ or $\bar x_j'$ to replace some $(\bar i, \bar j)$ in $\mathbf c_{2n+1}$, provided that neither (1) nor (2) occurs,
- using some $(\overline{k+j}, \bar s)$ or $\bar c_s (\overline{k+j}, \bar s)\bar c_s$, $s>k$ to replace $(\overline{k+j}, \bar j_i)$; or using $(i, j), c_j(i, j)c_j$ to replace $(\bar i, \bar j)$ in $\mathbf c_{2n+1}$; or using $(k+i, s)$, $c_s(k+i, s) c_s$, $s>k_1$, to replace $(k+i, i_j)$, provided that (1)–(3) do not occur,
- using $\bar c_{j_i} (\overline{k+j}, \bar j_i) \bar c_{j_i}$ to replace $(\overline{k+j}, \bar j_i)$, or using $\bar c_{j} (\bar i, \bar j)\bar c_j$ to replace $(\bar i, \bar j)$ in $\mathbf c_{2n+1}$ or using $c_{i_j} (k+i, i_j) c_{i_j}$ to replace $(k+i, i_j)$, provided that (1)-(4) do not occur.
In the case (1), we use defining relations for $BC_{2, r+k, t+k}$ to rewrite the corresponding monomial as a linear combinations of normal monomials. Each of these normal monomials corresponds to a unique walled Brauer diagram, say $D$, in which there is a horizontal edge $[i, \bar j]$ at the top row of $D$ such that $1\le j\le k$. Such a monomial does not satisfy (b)(5). Similarly, in case (2) (resp., (resp., (3), (4), (5)), the corresponding monomials of $BC_{2, k+r, k+t}$ can be written as a linear combinations of normal monomials which do not satisfy (b)(1) or (b)(3) (resp., (c) or (b)(3), (b)(3), (a)). This verifies our claim.
We assume that $\mathbf{m}_1, \mathbf m_2, \ldots, \mathbf m_p $ are all monomials in $S$ which contribute leading terms. Write $$\label{mii} \mathbf m_i=c^{\alpha(\mathbf m_i)} x^{\beta(\mathbf m_i)} d_1(\mathbf m_i)^{-1} e^{f(\mathbf m_i)} w(\mathbf(m_i)d_2(\mathbf m_i) {\bar x}^{\gamma(\mathbf m_i)} {\bar c}^{\delta(\mathbf m_i)}
\mbox{$\prod\limits_{n\in \mathbb Z^{>0}}$} \omega_{2n+1}^{a_{2n+1}(\mathbf m_i)}.$$ Then $k=|\gamma(\mathbf m_i)|+\sum_n (2n+1) a_{2n+1}(\mathbf m_i)$, $f(\mathbf m_i)=f_0$ and $k_1=|\beta(\mathbf m_i)|$. Let $A_i$ be the set of all leading terms contributed by $\phi_k(\mathbf m_i)$. These leading terms are admissible monomials of $BC_{2, k+r, k+t}$. We have proved that other terms of $\sum_{\mathbf m\in S} r_{\mathbf m} \phi_k(\mathbf m)$ will not contribute admissible monomials of $BC_{2, k+r, k+t}$. So, $$\label{setneq}\mbox{$\sum\limits_{i=1}^p$} \tilde r_{\mathbf m_i} \mbox{$\sum\limits_{\mathbf n\in A_i}$} \mathbf n=0,$$ where $ \tilde r_{\mathbf m_i}$ is the $a r_{\mathbf m}$ and $a$ is a power of $\pm 2$, which comes from the coefficients of leading terms of $\xi_{2n+1, k+1}$. In the following, we explain that does not hold. If so, then $\mathcal S$ is linear independent and the result will follow.
Suppose $g_i=(k+i, 1)(k+i, 2)\cdots (k+i, k_1)$ and $\bar g_j=(\overline {k+j}, \bar 1)(\overline{k+j}, \bar 2)\cdots (\overline{k+j}, \bar k)$ for $1\le i\le r$ and $1\le j\le t$. Note that each element in $A_s$ contains factors $(k+i, i_1)\cdots (k+i, i_{\beta_i})$ contributed by $\phi_k(x_i)^{\beta_i}$, such that $i_1, \cdots, i_{\beta_i}$, $1\le i\le r$, is a permutation of some elements in $1, 2, \ldots, k_1$. So, $(k+i, k_1)\cdots (k+i, k_1-\beta_i+1)$ is one of such factors and hence $g_i (k+i, k_1) (k+i, k_1-1)\cdots (k+i, k_1-\beta_i+1)$ fixes $k_1, \ldots, k_1-\beta_i+1$. So, there is an element in $g_i (A_s)$ whose walled Brauer diagram contains vertical edges $[j, j]$, where $j$ ranges $\beta(\mathbf m_s)_i$ numbers in $\{1, 2,\ldots, k_1\}$. If $g_i(A_{s'})$ contains an element such that the corresponding walled Brauer diagram contains vertical edges $[j, j]$ for $\beta(\mathbf m_s)_i$ numbers in $\{1, 2, \ldots, k_1\}$, then $\beta(\mathbf m_{s'})_i=\beta(\mathbf m_{s})_i$. So, we can assume $\beta(\mathbf m_1)=\beta(\mathbf m_j)$, $2\le j\le p$. Similarly, we use $\bar g_j$ instead of $g_i$ to obtain $\gamma(\mathbf m_1)=\gamma(\mathbf m_j)$, $2\le j\le p$. Note that $\mathbf c_{2n+1}$ is the product of $a_{2n+1}$ disjoint cycles with length $2n+1$. So, different $\prod_{n} \omega_{2n+1}^{a_{2n+1}}$ gives product of disjoint cycles with different lengthes and thus, we can assume $a_{2n+1}(\mathbf m_i)$ is independent of $\mathbf m_i$. Since any leading term is of form in , by Theorem \[level-2\] or Theorem \[wbhsa321\], we can also assume that $\alpha(\mathbf m_i), \delta(\mathbf m_i)$ are independent of $\mathbf m_i$. Since $c_i^2=-1$ and $\bar c_j^2=1$, we can assume $\alpha(\mathbf m_i)=0^r\in \Z_2^r$ and $\delta(\mathbf m_i)=0^t\in \Z_2^t$. Using , we see that there exists a leading term in $A_i\cap A_j$ if and only if $\mathbf m_i=\mathbf m_j$. So, $\tilde r_{\mathbf m_i}=0$ for all $1\le i\le p$, a contradiction. So, $S$ is linear independent over $\mathbb C$ and hence over $\mathbb Z[2^{-1}, \omega_1]$. In general, using arguments on base change yields the result over an arbitrary integral domain $R$ containing ${\mathbf {\omega}}_1$ and $2^{-1}$.
The following result follows from Theorem \[main1cyco\], immediately.
\[main1cyco321\] Suppose $R$ is a domain containing $2^{-1}$ and $\omega_{2n+1}$, for all $n\in \mathbb N$. Then $ \widetilde{BC}_{r,t}$ is free over $R$ spanned by all of its regular monomials. In particular, $\widetilde{BC}_{r,t}$ is of infinite rank.
A basis of the cyclotomic Brauer-Clifford superalgebra
======================================================
In this section, we assume that $R$ is a domain containing $ 2^{-1}$ and parameters $\{\omega_{2n+1}\in R\mid n\in \mathbb N\}$. The [*affine walled Brauer-Clifford superalgebra*]{} $ \widetilde {BC}_{r, t}$ with respect to the defining parameters $\omega_{2n+1}$’s can be also defined in a simpler way as follows. As a free $R$-superspace, $$\label{Affbcycl}{ \widetilde{BC}_{r, t} =R[\mathbf {x}_r]\otimes BC_{r, t}\otimes R[\bar {\mathbf {x}}_t],}$$the tensor product of the walled Brauer-Clifford superalgebra $ BC_{r, t}$ with two polynomial algebras $R[\mathbf{x}_r]:=R[x_1, x_2, \ldots, x_r]$ and $R[\bar{ \mathbf{x}}_t]:=R[\bar x_1, \bar x_2, \ldots, \bar x_t]$. The multiplication of $\widetilde{BC}_{r, t}$ is defined such that $R[\mathbf{x}_r]\otimes 1\otimes 1$, $1\otimes 1\otimes R[\bar{ \mathbf{x}}_t]$, $1\otimes BC_{r, t}\otimes 1$, $R[\mathbf{x}_r]\otimes HC_r\otimes 1$ and $1\otimes {\bar{HC}}_{t}\otimes R[\bar{ \mathbf{x}}_t]$ are subalgebras isomorphic to $R[\mathbf{x}_r]$, $R[\bar{ \mathbf{x}}_t]$, $BC_{r, t}$, $HC_r^{\rm{aff}}$, and $\overline{HC}{\ssc\,}_t^{\rm{aff}}$ respectively, and (for simplicity, without confusion we identify elements $x_i\otimes1\otimes1$, $1\otimes s_i\otimes 1$, $1\otimes e_i\otimes 1$, $1\otimes \bar s_i\otimes 1$, $1\otimes c_i\otimes 1$,$1\otimes \bar c_i\otimes 1$,$1\otimes1\otimes\bar x_i$ in with $x_i,s_i,e_i,\bar s_i,c_i,\bar c_i,\bar x_i$) $$\begin{aligned}
\label{it1}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
e_1(x_1+\bar x_1)=(x_1+\bar x_1) e_1=0,\ \ e_1 s_1x_1s_1=s_1x_1s_1e_1,\ \ e_1 \bar s_1\bar x_1\bar s_1=\bar s_1\bar x_1\bar s_1e_1,\\
\label{it2}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
s_i\bar x_1=\bar x_1 s_i,\ \ \bar s_i x_1= x_1 \bar s_i,\ i>1, \ x_1 (e_1-\bar e_1+\bar x_1)= (e_1-\bar e_1+\bar x_1)x_1,\\
\label{it3}&\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!\!&
e_1x_1^ke_1=\omega_k e_1,\ \ e_1\bar x_1^ke_1=\bar \omega_k e_1,
$$ where $\bar \omega_{2k+1}$ determined by $\omega_{1}, \ldots, \omega_{2k+1}$ as in Corollary \[assump1\]. Further, $\omega_{2n}=\bar\omega_{2n}=0$.
We hope to classify finite dimensional simple $\widetilde {BC}_{r, t}$-modules over an algebraically closed field $F$ with $\text{Char} F\neq 2$. This leads us to introduce cyclotomic Brauer-Clifford superalgebras as follows. Let $f(x)$ be the minimal polynomial of $x_1$ with respect to a finite dimensional $\widetilde {BC}_{r, t}$-module $M$. Then $$\label{minimalf} f(x)=x^k\mbox{$ \prod\limits_{i=1}^n$} (x-u_i),$$ where $u_1, \ldots, u_n$ are nonzero in $F$. Let $\langle f(x_1)\rangle $ be the two-sided ideal of $\widetilde {BC}_{r, t}$ generated by $f(x_1)$. Since $M$ is simple, $\langle f(x_1)\rangle\neq \widetilde {BC}_{r, t} $.
\[inducdeg\] We have $c_1f(x_1)= \epsilon f(x_1)c_1$, where $\epsilon\in \{-1, 1\}$ and $f(x)$ is given in .
We prove the result by induction on $\text{deg\,} f(x)$. If $\text{deg\,} f(x)=1$, then $f(x_1)=x_1-u$. When $u\neq 0$, we have $\langle x_1-u\rangle=\widetilde{BC}_{r, t}$, a contradiction, since $2u=c_1 (x_1-u) c_1-(x_1-u) \in \langle x_1-u\rangle$. So, $f(x_1)=x_1$ and $c_1f(x_1)=-f(x_1)c_1$. In general, if $ f(x_1) c_1\in \{ c_1 f(x_1), -c_1 f(x_1)\}$, there is nothing to prove. Otherwise, by Definition \[awbsa\](2), $f(-x_1)=-c_1f(x_1)c_1$ and $ f(-x_1)\neq f(x_1)$. In this case, we choose $h(x_1)\in \{f(-x_1)+f(x_1), f(-x_1)-f(x_1)\}$ such that $\text{deg\,} h(x)< \text{deg\,} f(x)$. Define $d(x)= {\rm g.c.d\,} (h(x), f(x))$, the greatest common divisor of $h(x)$ and $f(x)$. Then $ \langle d(x_1)\rangle= \langle f(x_1)\rangle$ and hence the irreducible module $M$ is killed by $d(x_1)$. This is a contradiction since $f(x_1)$ is the minimal polynomial of $x_1$ with respect to $M$.
Recall that $f(x_1)$ in . Since $c_1 f(x_1)c_1=-\epsilon f(x_1)$, and $c_1 f(x_1) c_1=\pm x_1^k \prod_{i=1}^n (x_1+u_i)$, $f(x)$ is the minimal polynomial of $x_1$ with respect to $M$ if and only if $x_1^k \prod_{i=1}^n (x_1+u_i) $ is the minimal polynomial of $x_1$ with respect to $M$. In other words, $u_i$ and $-u_i$ appear simultaneously if $u_i\neq 0$. Thus, we can assume $$\label{polyoff}
f(x_1)= x_1^k\mbox{$\prod\limits _{i=1}^m$} (x_1^2-u_i^2),$$ where $0\neq u_i\in F$, $1\le i\le m$. Moreover, by Lemmas \[assum\]–\[assum1\], there is a monic polynomial $g(\bar x_1)$ with degree $l=k+2m$ such that $$\label{feg}
e_1f(x_1)=(-1)^k e_1g(\bar x_1).$$
\[gcg\] Let $g(\bar x_1)$ be given such that is satisfied. Then $\bar c_1 g(\bar x_1)=\epsilon g(\bar x_1)\bar c_1$.
Since $l=k+2m$, we have $$\begin{aligned} (-1)^k e_1 g(\bar x_1) \bar c_1 & =e_1f(x_1)\bar c_1= e_1\bar c_1 f(x_1)=e_1 c_1 f(x_1)=\epsilon e_1f(x_1) c_1\\ & =(-1)^k \epsilon e_1 g(\bar x_1)c_1=(-1)^k \epsilon e_1 c_1 g(\bar x_1)=(-1)^k \epsilon e_1\bar c_1 g(\bar x_1).\end{aligned}$$ By Theorem \[main1cyco321\], $\bar c_1 g(\bar x_1)=\epsilon g(\bar x_1)\bar c_1$.
In the remaining part of this paper, we assume that $$\label{polyofg}
g(\bar x_1)= \bar x_1^{k_1}\mbox{$\prod \limits_{j=1}^{m_1}$}(\bar x_1^2-\bar u_j^2),$$ such that $k_1+2m_1=k+2m$ and $0\neq \bar u_j\in F$, $1\le j\le m_1$. This is reasonable by Lemma \[gcg\]. Since the finite dimensional simple $\widetilde{BC}_{r, t}$-module $M$ is killed by $f(x_1)$, by , it is killed by $e_1g(\bar x_1)$. We want to consider simple $\widetilde{BC}_{r, t}$-modules $M$ such that $e_1$ acts on $M$ nontrivially, it is necessary to assume that $M$ is killed by $g(\bar x_1)$. That is why we introduce cyclotomic walled Brauer-Clifford superalgebras as in the Definition \[cwbcsa1\].
From here to the end of this section, we assume both $\widetilde{BC}_{r, t}$ and $BC_{l, r, t}$ are defined over a domain $R$ containing $ 2^{-1}$ and parameters $\omega_{2n+1}$ for all $n\in \mathbb N$.
\[condi-k1cycl\] Write $f(x_1)=x_1^{k+2m}+\sum_{i=1}^{2m} a_i x_1^{k+2m-i}$, where $f(x_1)$ is given in . Then $e_1$ is an $R$-torsion element of $ BC_{k+2m, r, t}$ unless $$\label{ccccc}
{ \omega_\ell=-(a_1\omega_{\ell-1}+\ldots +a_{2m}\omega_{\ell-2m})}\mbox{ \ for all \ }\ell\ge k+2m.$$
Let $b_\ell=\omega_\ell+a_1\omega_{\ell-1}+\ldots+ a_{2m}\omega_{\ell-2m}\in R$. By , $b_\ell e_1=e_1 f(x_1)x_1^{\ell-2m-k}e_1$ in $ \widetilde{BC}_{r, t}$ and $b_\ell e_1=0$ in $BC_{k+2m,r,t}$. Thus, $e_1$ is an $R$-torsion element if $b_\ell\neq 0$ for some $\ell\ge k+2m$.
\[condi-kcycl\] The superalgebras $ \widetilde{BC}_{r, t}$ and $BC_{k+2m,r,t}$ are called *admissible* (with respect to $f(x_1)$) if holds.
\[zero-1cycl\] For $1\leq i\leq r$, $1\leq j\leq t$, define $f_i= f(x_i')$ and $g_{i}= g (\bar x_i')$, where $f(x_1)$ and $g(\bar x_1)$ satisfy –. Then the following equations hold for all admissible $i, j$:
[2]{}
1. $c_jf_i=\epsilon f_ic_j$,
2. $\bar c_jg_i=\epsilon g_i\bar c_j$,
3. $\bar c_jf_i=f_i\bar c_j$,
4. $ c_jg_i= g_i c_j$,
5. $\bar s_jf_i=f_i\bar s_j$,
6. $ s_jg_i= g_i s_j$,
7. $f_if_j=f_jf_i $ in ${\rm gr} (\widetilde{BC} _{r,t})$,
8. $g_ig_j=g_jg_i$ in ${\rm gr} (\widetilde{BC}_{r,t})$.
These equations can be easily verified by using Lemmas \[inducdeg\]–\[gcg\] and Definition \[awbsa\].
Note that the affine Hecke-Clifford superalgebra $HC_r^{\rm aff}$ (resp., $\overline { HC}{}{\ssc\,}_t^{\rm aff}$) is isomorphic to the sub-superalgebra of $\widetilde{BC}_{r, t}$ generated by $x_1, s_1,\ldots, s_{r-1}$ and $c_1$ (resp., $\bar x_1, \bar s_1,\ldots, \bar s_{t-1}$ and $\bar c_1$).
\[zero-12\] For any $a\in \mathbb Z^{>0}$, we have
- $(x_i')^a f(x'_\ell)-f(x'_\ell)x'^a_i\in \sum_{b<a}\sum_{h, h_1=1}^{\max\{i, \ell\}} f(x'_h)(x'_{h_1})^b HC_r$,
- $g(\bar x_\ell') (\bar x_i')^a-(\bar x_i')^a g(\bar x_\ell') \in \sum_{b<a}\sum_{h, h_1=1}^{\max\{i, \ell\}}\overline{HC}_t (\bar x'_{h_1})^b g(\bar x'_h) $.
We have $x_1x_2=x_2x_1$, where $x_2=x_2'-s_1-c_2s_1c_2$ (see ). By Lemma \[zero-1cycl\](1), $$\label{key-1}
x_2' f(x_1)=f(x_1)(x_2'-s_1-c_2s_1c_2)+f(x_2') s_1+\epsilon f(x_2')c_2s_1c_2.$$ Considering ${s_{i, 2} x_2' f(x_1) s_{2, i}}$ yields the result when $a=1$ and $\ell=1$. If $\ell>1$, then $$x_i' f(x_\ell')=x_i's_{\ell-1} f(x_{\ell-1}') s_{\ell-1}= s_{\ell-1}x_{(i)s_{\ell-1}}' f(x_{\ell-1}') s_{\ell-1}.$$ So, the result follows from inductive assumption on $\ell-1$. This is (1) when $a=1$. The general case follows from arguments on induction on $a$. Finally, (2) can be verified, similarly.
\[2ijcycl\] Define $J_L=\sum_{i=1}^t \widetilde{BC}_{r, t}\, g_i$ and $J_R=\sum_{i=1}^r f_i\, \widetilde{BC}_{r, t}$. Let $I$ be the two-sided ideal of $\widetilde{BC}_{r, t}$ generated by $f(x_1)$ and $ g(\bar x_1)$. We have
1. $J_R$ is a left $HC_r^{\rm aff}\otimes \overline{HC}_t$-module,
2. $J_L$ is a right $HC_r\otimes \overline{HC}{\ssc\,}_{ t}^{\rm aff}$-module,
3. $I=J_L+J_R$ if $\widetilde{BC}_{r, t}$ is admissible.
(1)–(2) It is easy to see that $J_R$ is stable under the left action of $HC_r\otimes \overline{HC}_t$. By Lemma \[zero-12\](1), it is stable under the left action of ${HC}_r^{\rm aff}$. One can check (2) via Lemma \[zero-12\](2), similarly.
\(3) Obviously, $J_L+J_R\subseteq I$. So, (3) follows if we can prove $I\subseteq J_L+J_R$. Since $f(x_1), g(\bar x_1)\in J_L+J_R$, it suffices to verify that $ J_L+J_R$ is a two-sided ideal of $\widetilde{BC}_{r, t}$. We claim $$\label{stepp1} h J_R\subset J_L+J_R,$$ for any generator $h$ of $\widetilde{BC}_{r, t}$. If so, $h (J_L+J_R)\subset J_L+J_R$ and hence $J_L+J_R$ is a left ideal.
In fact, by (1), it is enough to verify when $h\in \{\bar x_1, e_1\}$. If we have $e_1 J_R\subset J_L+J_R$, then $c_1e_1c_1 J_R\subset J_L+J_R$. Since $(\bar x_1+e_1-\bar e_1) f(x_1)=f(x_1) (\bar x_1+e_1-\bar e_1)\in J_R$, we have $\bar x_1 f(x_1)\in J_L+J_R$. Multiplying $(1, i)$ on both sides of $\bar x_1 f(x_1)$ yields $\bar x_1 f_i\in J_L+J_R$. So, we need to verify $e_1 J_R\subset J_L+J_R$. By , $e_1 f_i =f_i e_1$ for $i\ge 2$. So, $e_1 J_L\subset J_L+J_R$ if $e_1f(x_1) \widetilde{BC}_{r, t}\subset J_L+J_R$. This will be verified by checking $$\label{stepp2}
e_1 f(x_1)\textbf{m}\in J_L+J_R,$$ for each regular monomial $\textbf{m}$ of $\widetilde{BC}_{r, t}$ in Definition \[cregm\]. Using arguments on graded structure of $\widetilde{BC}_{r, t}$, we can write $\textbf{m}=c^{\alpha} x^{\beta} e_{i_1, j_1} \cdots e_{i_f, j_f} w \bar x^{\gamma} {\bar c}^{\delta}$ for some $(\alpha, \delta)\in \Z_2^r\times\Z_2^t$ and $(\beta, \gamma)\in \mathbb N^r\times \mathbb N^t$, and $w\in \Sigma_r\times \Sigma_{\bar t}$ and $1\le i_1, \ldots, i_f\le r$ and $1\le j_1, \ldots, j_f\le t$ such that $\{i_k, j_k\}\cap \{i_l, j_l\}=\emptyset$ if $k\neq l$. In the following, we write $e_{\bii, \bij}=e_{i_1, j_1} \cdots e_{i_f, j_f}$. We prove by induction on $|\b|$.
*Case 1: $|\b|=0$.*
If $f=0$, then $e_1f(x_1)c^\alpha w \bar x^\gamma \bar c^{\delta}=(-1)^k e_1g(\bar x_1)c^\alpha w \bar x^\gamma \bar c^{\delta}\subseteq J_L $. The last inclusion follows from (2). Suppose $1\le f\le \min\{r, t\} $. Since $\widetilde {BC}_{r,t} $ is admissible, $e_1 f(x_1)e_1=0$. On the other hand, we have $e_1x_1^k c_1 e_1=0$ for all $k$. So, $e_1f(x_1) \textbf{m}=0$ if $e_1$ is a factor of $e_{\bii,\bij}$. If $e_1$ is not a factor of $e_{\bii,\bij}$, there are three cases we need to discuss.
- If $e_{ p, 1}$ is a factor of $e_{\bii,\bij}$, and $p\neq 1$, then we assume that $i_1=p$ and $j_1=1$ since any two factors of $e_{\bii, \bij}$ commute each other. We have $ e_1 f(x_1)c^{\alpha} e_{ p, 1}=\prod_{i=2}^r c_i^{\alpha_i} e_1 f(x_1)e_{ p, 1} c_1^{\alpha_1} $. Since $$e_1 f(x_1) e_{p, 1}=s_{p, 2} e_1 f(x_1) s_1 e_1 s_{1, p}=s_{p, 2} e_1s_1f(x_2') e_1 s_{1, p}=s_{p, 2} f(x_2')e_1 s_{1, p}\in J_R,$$ we have $ e_1 f(x_1)c^{\alpha} e_{1, p}\in J_R$ by (1). So, $e_1f(x_1)\textbf{m}\in J_R$.
- If $e_{1, p}$ is a factor of $e_{\bii, \bij}$ and $p\neq 1$, we assume $i_1=1$ and $j_1=p$. We have $$\begin{array}{lll}
e_1 f(x_1) e_{1, p}\!\!\!\!&=(-1)^k \bar s_{p, 2} e_1 g(\bar x_1) \bar s_1 e_{1} \bar s_{1, p}=(-1)^k \bar s_{p, 2} e_1\bar s_1 g(\bar x_2' ) e_{1} \bar s_{1, p}\\[4pt]
\!\!\!\!&=(-1)^k \bar s_{p, 2} e_1 g(\bar {x}_2') \bar s_{1,p}=(-1)^k \bar s_{p, 2} e_1 \bar s_{1, p} g(\bar x_1)\in J_L.\end{array}$$ So, $ e_1 f(x_1)c^{\alpha} e_{ 1, p}=\prod_{i=2}^r c_i^{\alpha_i} e_1 f(x_1) c_1^{\alpha_1} e_{1, p}=\prod_{i=2}^r c_i^{\alpha_i}\bar c_p^{\alpha_1} e_1 f(x_1) e_{1, p}
\in J_L$. By (2) and the equation $g(\bar x_1)\prod_{k=2}^f e_{i_k, j_k} =\prod_{k=2}^f e_{i_k, j_k} g(\bar x_1)$, $e_1f(x_1)\textbf{m}\in J_L$.
- Finally, suppose $\{ i_l, j_l\}\cap \{1\}=\emptyset$ for all possible $l$, then $e_1f(x_1)\textbf{m}\in J_L$ follows from (2) and the following fact $$e_1 {\textbf{\textit{f}}}(x_1)\mbox{$ \prod\limits_{ l=1}^f e_{i_f, j_f}=\prod\limits_{ l=1}^f e_{i_f, j_f}e_1 {\textbf{\textit{f}}}(x_1) =(-1)^k \prod\limits_{ l=1}^f $}e_{i_f, j_f} e_1 {\textbf{\textit{g}}}(\bar x_1)\in J_L.$$
*Case 2: $|\beta|>0$.*
Suppose $\beta_i\neq 0$ for some $2\le i\le r$. We have $x_i\textbf{m}'=\epsilon \textbf{m}$ by Lemma \[hecrel\](1)–(2), where $\textbf{m}'$ is obtained from $\textbf{m}$ by removing one $x_i$ and $\epsilon=\pm 1$. So $e_1f(x_1)\textbf{m}\in J_L+J_R$ if $e_1f(x_1)x_i \textbf{m}'\in J_L+J_R$. Since $f(x_1)x_i=x_if(x_1)$, it suffices to prove $e_1 L_i f(x_1)\textbf{m}'\in J_L+J_R$ and $e_1 (x_i-L_i) f(x_1)\textbf{m}'\in J_L+J_R$.
In the first case, since $e_1(j, i)=(j, i) e_1$ if ${j\neq 1}$ and $e_1c_i=c_i e_1$ and $c_if(x_1)=f(x_1)c_i$, by inductive assumption on $|\beta|$ and (1)–(2), it is enough to prove $e_1 (1, i) f(x_1) c_i^l \textbf{m}'\in J_R$ for $l\in \Z_2$. In fact, it is the case since $$e_1 (1, i) f(x_1) c_i^l\textbf{m}'=e_1 f(x_i') (1, i) c_i^l \textbf{m}'=f(x_i') e_1 (1, i) c_i^l \textbf{m}' \in J_R.$$ In the second case, since $e_1 (x_i-L_i) f(x_1)\textbf{m}'=(x_i-L_i) e_1f(x_1)\textbf{m}'$, by induction on $\b$, $e_1f(x_1)\textbf{m}'\in J_L+J_R$. By (1), we have $(x_i-L_i) e_1f(x_1)\textbf{m}'\in J_L+J_R$. So, $ e_1f(x_1)\textbf{m}\in J_L+J_R$.
If $\b_i=0$, $2\le i\le r$, then $x^\b=x_1^{\b_1}$ with $\b_1>0$. In this case, $\textbf{m}=c^\alpha x_1^{\b_1} e_{\bii, \bij} w \bar x^\gamma \bar c^\delta$. We want to prove $v=e_1f(x_1) \textbf{m}\in J_L+J_R$. If $j_\ell\neq 1$, $1\le \ell\le f$, then by inductive assumption, $$\begin{aligned}
v&=e_1 f(x_1) c^\alpha x_1^{\b_1}e_{\bii,\bij} w\bar x^\gamma \bar c^\delta
= (-1)^k e_1g(\bar x_1) c^\alpha x_1^{\b_1} e_{\bii,\bij} w\bar x^\gamma \bar c^\delta
\equiv (-1)^k e_1 c^\alpha x_1^{\b_1}g(\bar x_1)e_{\bii,\bij} w\bar x^\gamma \bar c^\delta \\ &
= (-1)^k e_1c^\alpha x_1^{\b_1}e_{\bii,\bij}{\sc\,}g(\bar x_{1}) w{\bar x}^\gamma \bar c^\delta \nonumber \in J_L w {\bar x}^\gamma \bar c^\delta\subset J_L+J_R,\\
\end{aligned}$$ where the “$\equiv$” is modulo $J_L+J_R$. Finally, if $j_\ell=1$ for some $\ell$, without loss of any generality, we assume $j_1=1$. If $i_1=1$, we have $e_1f(x_1) c^\alpha x_1^\beta e_1=0$ no matter whether $\alpha_1=1$ or $0$. In the first case, this result follows from the equation $e_1x_1^k c_1 e_1=0$ for all $k\in \mathbb N$. In the second case, this result follows from the fact that $\widetilde{BC}_{r, t}$ is admissible. It remains to deal with the cases when $i_l\neq 1$ for all $l$. Define $ i'=(i_2,\ldots, i_f)$ and $ j'=(j_2,\ldots ,j_f)$. Then $$\begin{aligned} v&= e_1 f(x_1)c^\alpha x_1^{\b_1}e_{i_1, 1}e_{\bii',\bij'} w\bar x^\gamma \bar c^\delta
=\mbox{$\prod\limits_{i=2}^r$} c_i^{\alpha_i} e_1f (x_1) e_{i_1,1} c_1^{\alpha_1} x_1^{\b_1}e_{\bii',\bij'} w\bar x^\gamma \bar c^\delta\\
&=\mbox{$\prod\limits_{i=2}^r$} c_i^{\alpha_i} e_1 e_{i_1, 1} f(x_1) c_1^{\alpha_1} x_1^{\b_1}e_{\bii',\bij'} w\bar x^\gamma \bar c^\delta
=\mbox{$\prod\limits_{i=2}^r$} c_i^{\alpha_i} e_1(1,i_1)f(x_1)c_1^{\alpha_1} x_1^{\beta_1}e_{\bii',\bij'} w\bar x^\gamma \bar c^\delta\\
&= \mbox{$\prod\limits_{i=2}^r$} {c_i^{\alpha_i} e_{1}f (x_{i_1}') (1, i) c_1^{\alpha_1} x_1^{\beta_1}e_{\bii',\bij'} w\bar x^\gamma \bar c^\delta}
\in \mbox{$\prod\limits_{i=2}^r$} c_i^{\alpha_i} J_R\subset J_R, \text{ by (1).}
\\
\end{aligned}$$ This completes the proof of and hence $h J_R\!\subset \!J_L\!+\!J_R$. One can similarly check $J_L h\!\subset \!J_L\!+\!J_R$.
For $(\alpha, \beta)\in \mathbb N^r\times \mathbb N^t$, let ${\textbf{\textit{f}}}(x')^\alpha=f_1^{\alpha_1}\cdots f_r^{\alpha_r}$ and ${\textbf{\textit{g}}}(\bar x')^\beta=g_1^{\beta_1}\cdots g_t^{\beta_t}$. Recall that $l=k+2m$.
\[mathcalncycl\] The affine walled Brauer-Clifford superalgebra $\widetilde{BC}_{r,t}$ is a free $R$-module with $\mathcal N$ as its $R$-basis, where $$\begin{aligned}
\label{regs1}&\!\!\!\!\!\!\!\!&
\mathcal N= \stackrel{\sc \min\{ r, t\}}{\underset{\sc f=0}{\dis\mbox{\Large$\cup$}}}
\{ {\textbf{\textit{f}}}(x')^\alpha {c}^{\tilde \gamma} {x}^{\gamma} d_1^{-1} e^f w d_2 \bar x^{\delta}{\bar c}^{\tilde \delta} {\textbf{\textit{g}}}(\bar x')^\b
\mid (\alpha, \beta)\in\mathbb N^r \times \mathbb N^t, (\gamma, \delta ,\tilde \gamma,\tilde \delta)\in\mathbb Z_l^r \times \mathbb Z_l^t\times \Z_2^r \times \Z_2^t, \nonumber\\[-10pt]
&\!\!\!\!\!\!\!\!&
\phantom{ \mathcal N= \stackrel{\sc \min\{m, n\}}{\underset{\sc f=0}{\dis\mbox{\Large$\cup$}}}
\{}
d_1, d_2\in D_{r, t}^f,
w\in \Sigma_{r-f}\times \bar{\Sigma}_{{t-f}}\}.\end{aligned}$$
The result follows from Theorem \[main1cyco321\] since the transition matrix between $\mathcal N$ and the basis in Theorem \[main1cyco321\] is upper-unitriangular.
\[I\]Let $I$ be the two-sided ideal of $\widetilde{BC}_{r,t}$ generated by $f(x_1)$ and $g(\bar x_1)$ satisfying –. If $ \widetilde{BC}_{r,t}$ is admissible, then $S$ is an $R$-basis of $I$, where $$\label{sbasiscycl} S=\{ {\textbf{\textit{f}}}(x')^\alpha {c}^{\tilde \gamma} {x}^{\gamma} d_1^{-1} e^f w d_2 \bar x^{\delta}{\bar c}^{\tilde \delta}{\textbf{\textit{g}}} (\bar x')^\b\in\mathcal N\mid \alpha_i+\b_j\neq 0 \text{ for some $i, j$}\}.$$
Let $M$ be the $R$-module spanned by $ S$. Obviously, $M\subseteq I$. If $J_L\subseteq M$, and $J_R\subseteq M$, by Lemma \[2ijcycl\](3), $M=I$, proving the result. By symmetry, we verify $J_R\subseteq M$. By Lemma \[mathcalncycl\], we need to verify $f(x_i') \textbf{m}\in M$ for any basis element $\textbf{m}$ in . In fact, we have $$f(x_i')f(x_\ell')\in f(x_\ell') f(x_i')+\mbox{$\sum\limits_{j=1}^{\ell-1}$} f(x_j') \widetilde{BC}_{r, t}, \text{ by Lemma~\ref{zero-12}}.$$ Using induction on degrees, we have $f(x_i') \textbf{m}\in M$. Finally, one can check $J_L\subset M$, similarly.
\[level-k-walledcycl\] The cyclotomic walled Brauer-Clifford superalgebra $ BC_{k+2m, r, t}$ is free over $R $ with rank $2^{r+t}(k+2m)^{r+t}(r+t)!$ if and only if $BC_{k+2m, r, t}$ is admissible.
By Corollary \[level-l-span\], $ BC_{k+2m, r, t}$ is spanned by all of its regular monomials. If $BC_{k+2m, r,t}$ is not admissible, $e_1$ is an $R$-torsion element by Corollary \[assump1\]. Since $e_1\in M$, either $BC_{k+2m, r, t}$ is not free or the rank of $BC_{k+2m, r, t}$ is strictly less than $2^{r+t}(k+2m)^{r+t}(r+t)!$, the number of all regular monomials of $ BC_{k+2m, r, t}$. If $BC_{k+2m, r,t}$ is admissible, by Lemmas \[mathcalncycl\]–\[I\], all regular monomials of $BC_{k+2m, r,t}$ are $R$-linear independent. So $BC_{k+2m, r, t}$ is free over $R$ with rank $2^{r+t}(k+2m)^{r+t}(r+t)!$.
Isomorphisms between affine (resp., cyclotomic) Brauer-Clifford algebras and Comes-Kujawa’s affine (resp., cyclotomic) algebras
===============================================================================================================================
Let $\mathscr {AOBC}$ be the degenerate affine oriented Brauer-Clifford supercategory defined in [@CK Definition 3.7]. All notations on $\mathscr {AOBC}$ we use here are the same as those in [@CK]. For any positive integers $r,t$, let $\mathscr {BC}_{r,t}^{\rm aff}:= \operatorname{End}_{\mathscr {AOBC}}(\uparrow^r\downarrow^t)$. The aim of this section is to establish an isomorphism between certain quotient of $\mathscr {BC}_{r,t}^{\rm aff}$ and our $\widetilde{BC}_{r, t}$. First, we recall some of results in [@CK Section 3.2]. We use the notations in [@CK] to denote the morphisms in $\mathscr {AOBC}$.
Given a nonnegative integer $r$ and any morphism $f$, write $$\begin{tikzpicture}[baseline = 12pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,2);
\draw (0,1) node[circle,draw,thick,fill=white]{$f$};
\end{tikzpicture}^r
=~
\underbrace{
\begin{tikzpicture}[baseline = 12pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,2);
\draw (0,1) node[circle,draw,thick,fill=white]{$f$};
\end{tikzpicture}
\cdots
\begin{tikzpicture}[baseline = 12pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,2);
\draw (0,1) node[circle,draw,thick,fill=white]{$f$};
\end{tikzpicture}
}_{r\text{ times}}
~.$$ In the monoidal supercategory, the super-interchange law is represented by $$\label{super-interchange}
\begin{tikzpicture}[baseline = 19pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,3);
\draw[-,thick] (2,0) to (2,3);
\draw (0,2) node[circle,draw,thick,fill=white]{$f$};
\draw (2,1) node[circle,draw,thick,fill=white]{$g$};
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 19pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,3);
\draw[-,thick] (2,0) to (2,3);
\draw (0,1.5) node[circle,draw,thick,fill=white]{$f$};
\draw (2,1.5) node[circle,draw,thick,fill=white]{$g$};
\end{tikzpicture}
~=(-1)^{{|{f}|}{|{g}|}}~
\begin{tikzpicture}[baseline = 19pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,3);
\draw[-,thick] (2,0) to (2,3);
\draw (0,1) node[circle,draw,thick,fill=white]{$f$};
\draw (2,2) node[circle,draw,thick,fill=white]{$g$};
\end{tikzpicture}
~.$$
\[CKu\] $\mathscr{BC}_{r,t}^{\rm aff}$ is an associative superalgebras generated by even generators $e_1$, $s_i (1\leq i\leq r-1)$, $\bar s_j (1\leq j\leq t-1)$ $y_i (1\leq i\leq r)$ and $ \bar y_j (1\leq j\leq t)$, and odd generators $c_i (1\leq i\leq r)$ and $\bar c_i (1\leq i\leq t)$, and even central elements $\bar\delta_k, (k\in\mathbb Z_{\geq0})$ defined as follows: $$\label{gens for AOBC}
\begin{aligned}&
e_1=
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (1,0) to[out=up, in=down] (1,1); \draw (1.5,1.0) node{\footnotesize{$r-1$}};
\draw[<-,thick] (0,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (2.2,1);
\draw[<-,thick] (2.2,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0,0);\end{tikzpicture} ~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t-1}
~,\quad s_i=\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (1,0) to[out=up,in=down] (0.5,1);
\draw[->,thick] (0.5,0) to[out=up,in=down] (1,1);
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{r-i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t}
~,\quad
\bar s_i=\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{r}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick](0.5,1) to[out=down,in=up](1,0);
\draw[->,thick] (1,1)to[out=down,in=up](0.5,0);
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t-i-1}~,\\
& y_i= \begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0.5,0) to (0.5,1);
\draw (0.5,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{r-i}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t}
~,\quad
c_i=\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0.5,0) to (0.5,1);
\draw (0.5,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{r-i}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t},\\
&\bar y_i= \begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{r}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0.5,1) to (0.5,0);
\draw (0.5,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t-i},
~\quad
\bar c_i= \begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{r}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\end{tikzpicture}^{i-1}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0.5,1) to (0.5,0);
\draw (0.5,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,1);
\end{tikzpicture}^{t-i},\\
& \bar\delta_k=\begin{tikzpicture}[baseline = 5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.6,1) to (0.5,1) to[out=left,in=up] (0,0.5)
to[out=down,in=left] (0.5,0)
to[out=right,in=down] (1,0.5)
to[out=up,in=right] (0.5,1);
\draw (1,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (1.4,0.5) node{\footnotesize{$k$}};
\end{tikzpicture}.~ \quad
\end{aligned}$$ These generators satisfy the following local relations: $$\label{ese}
e_1s_1\bar s_1e_1s_1=e_1s_1\bar s_1e_1\bar s_1,~ s_1e_1s_1\bar s_1e_1=\bar s_1 e_1s_1\bar s_1e_1, e_1 s_1 e_1 =e_1\bar s_1 e_1=e_1,$$ $$\label{AOBC rels 1}
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[-,thick] (0,0) to[out=up, in=down] (1,1);
\draw[->,thick] (1,1) to[out=up, in=down] (0,2);
\draw[-,thick] (1,0) to[out=up, in=down] (0,1);
\draw[->,thick] (0,1) to[out=up, in=down] (1,2);
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[-,thick] (0,0) to (0,1);
\draw[->,thick] (0,1) to (0,2);
\draw[-,thick] (1,0) to (1,1);
\draw[->,thick] (1,1) to (1,2);
\end{tikzpicture}
,\qquad
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (2,2);
\draw[->,thick] (2,0) to[out=up, in=down] (0,2);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1) to[out=up, in=down] (1,2);
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (2,2);
\draw[->,thick] (2,0) to[out=up, in=down] (0,2);
\draw[->,thick] (1,0) to[out=up, in=down] (2,1) to[out=up, in=down] (1,2);
\end{tikzpicture},$$ $$\label{AOBC rels 2}
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (1,1) to[out=down, in=up] (0,0);
\draw[-,thick] (1,1) to[out=up, in=down] (0,2);
\draw[->,thick] (0,1) to[out=down, in=up] (1,0);
\draw[-,thick] (1,2) to[out=down, in=up] (0,1);
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to (0,1);
\draw[-,thick] (0,1) to (0,2);
\draw[<-,thick] (1,0) to (1,1);
\draw[-,thick] (1,1) to (1,2);
\end{tikzpicture}
,\qquad
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (2,2);
\draw[<-,thick] (2,0) to[out=up, in=down] (0,2);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1) to[out=up, in=down] (1,2);
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (2,2);
\draw[<-,thick] (2,0) to[out=up, in=down] (0,2);
\draw[<-,thick] (1,0) to[out=up, in=down] (2,1) to[out=up, in=down] (1,2);
\end{tikzpicture},$$ $$\label{AOBC scat rels 3}
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\end{tikzpicture}
,\qquad
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.2,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.8,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~,$$ $$\label{AOBC scat rels 3-1}
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\end{tikzpicture}
,\qquad
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.2,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.8,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~,$$ $$\label{AOBC rels 4}
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.6,1.25) to (0.5,1.25) to[out=left,in=up]
(0,0.75) to[out=down,in=left]
(0.5,0.25) to[out=right,in=down]
(1,0.75) to[out=up,in=right]
(0.5,1.25);
\draw (1,0.75) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};\draw (1.5,0.75) node{\footnotesize{$2k$}};
\end{tikzpicture}
~=0,\quad e_1 y_1^ke_1=\bar\delta_k e_1,\quad e_1c_1 y_1^ke_1=0, \quad
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture},~\quad
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw (0,1) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (0,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}$$
$$\label{AOBC rels 6}
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.2,1) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.8,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (1,1.5);
\end{tikzpicture}
~-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw[->,thick] (1,0) to[out=up, in=down] (1,1.5);
\draw (0,0.7) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (1,0.7) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~,$$
$$\label{AOBC rels 61}
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.2,1) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (1,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1.5);
\draw (0.8,0.5) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~=-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (1,1.5);
\end{tikzpicture}
~-~
\begin{tikzpicture}[baseline = 7.5pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1.5);
\draw[<-,thick] (1,0) to[out=up, in=down] (1,1.5);
\draw (0,0.7) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\draw (1,0.7) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
~,$$
$$\label{AOBC rels 7}
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (0.8,0.8) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (1.2,0.8) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}, ~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (0.8,0.2) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (1.2,0.2) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture},~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (0.8,0.8) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (1.2,0.8) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}, ~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (0.8,0.2) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}
~=~
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);
\draw (1.2,0.2) { node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture},~$$
$$\label{super commuting relations}
\begin{tikzpicture}[baseline = 19pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,3);
\draw[-,thick] (1,0) to (1,3);
\draw[-,thick] (2,0) to (2,3);
\draw[-,thick] (3,0) to (3,3);
\draw[-,thick] (4,0) to (4,3);
\draw (1,2) node[circle,draw,thick,fill=white]{$g$};
\draw (3,1) node[circle,draw,thick,fill=white]{$h$};
\draw (0,-0.3) node{$\mathbf a$};
\draw (2,-0.3) node{$\mathbf b$};
\draw (4,-0.3) node{$\mathbf c$};
\end{tikzpicture}
~=(-1)^{{|{g}|}{|{h}|}}
\begin{tikzpicture}[baseline = 19pt,scale=0.5,color={rgb:black,1;blue,4;red,1},inner sep=0pt, minimum width=11pt]
\draw[-,thick] (0,0) to (0,3);
\draw[-,thick] (1,0) to (1,3);
\draw[-,thick] (2,0) to (2,3);
\draw[-,thick] (3,0) to (3,3);
\draw[-,thick] (4,0) to (4,3);
\draw (1,1) node[circle,draw,thick,fill=white]{$g$};
\draw (3,2) node[circle,draw,thick,fill=white]{$h$};
\draw (0,-0.3) node{$\mathbf a$};
\draw (2,-0.3) node{$\mathbf b$};
\draw (4,-0.3) node{$\mathbf c$};
\end{tikzpicture}
~,$$
for all $\mathbf a,\mathbf b,\mathbf c\in{\langle{\uparrow},{\downarrow}\rangle}$ and $g,h\in\left\{{
\begin{tikzpicture}[baseline = 5pt, scale=0.75, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0.75,1) to[out=down,in=left] (1,0.6) to[out=right,in=down] (1.25,1);
\draw[<-,thick] (1.25,0) to[out=up,in=right] (1,0.4) to[out=left,in=up] (0.75,0);\end{tikzpicture}}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1);
\draw[->,thick] (1,0) to[out=up, in=down] (0,1);
\end{tikzpicture}
}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (1,1);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1);
\end{tikzpicture}
}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
}, {
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill=white, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
}\right\}$ such that the diagrams in are all in $\mathscr {BC}_{r,t}^{\rm aff}$.
Relations – follow from [@CK Corollary 3.6]. In other words, the generators $ e_1$, $s_1,\ldots, s_{r-1}$, $\bar s_1, \ldots, \bar s_{t-1}$, $c_1, \ldots, c_r$ and $\bar c_1, \ldots, \bar c_t$ satisfy the defining relations for ${BC}_{r,t}$ in Definition \[wsera\]. Relations – follows from [@CK Propositions 3.3, 3.10]. Relations in (resp., ) is [@CK 3.29] (resp., [@CK 3.27]). Finally, follows from [@CK 7.8]. Comes and Kujawa have pointed that the relations (7.2)–(7.8) are defining relations for $\mathscr {AOBC}$. So, the above relations consist of a complete set of relations for $\mathscr {BC}_{r,t}^{\rm aff}$.
Recall that $\widetilde{BC}_{r,t}$ is the affine walled Brauer-Clifford superalgebra in Definition \[affinewbcsa\]. Their defining parameters are $\omega_i, \bar\omega_j\in R$ such that $\omega_{2n}=\bar \omega_{2n}=0$ for all $n\in \mathbb N$. Let $J$ be the two-sided ideal of $\mathscr{BC}_{r,t}^{\rm aff}$ generated by $\bar\delta_k-(-1)^k\omega_k$ and $\bar \delta_k+\sum_{l=1}^{k-2} \bar \delta_l \delta_{k-l-1}-\bar \omega_k$, where $ \delta_i$ (cf. [@CK Remark .11]) are determined by $$\label{delta123}
\delta_k-\bar \delta_k= \sum_{0< i< k/2}\delta_{2i-1}\bar \delta_{k-2i}.$$ Define $$\label{affck} \widetilde{\mathscr {BC}}_{r, t}=\mathscr{BC}_{r,t}^{\rm aff}/J.$$
\[isomofoabc\] $\widetilde{\mathscr{BC}}_{r,t} \cong \widetilde{BC}_{r,t} $ as $R$-superalgebras.
Let $\varphi: \widetilde{\mathscr{BC}}_{r,t} \rightarrow \widetilde{BC}_{r,t} $ be the $R$-linear map such that $\varphi$ sends $e_1$, $s_i$’s, $\bar s_j$’s (resp., $c_k$’s and $\bar c_l$’s) to the same symbols (resp., $\sqrt{-1} c_k$’s and $\sqrt{-1} \bar c_l$’s) in $\widetilde{BC}_{r,t} $. We remark that $c_i^2=1$ and $\bar c_j^2=-1$ in $\widetilde{\mathscr{BC}}_{r,t} $ whereas $c_i^2=-1$ and $\bar c_j^2=1$ in $\widetilde{{BC}}_{r,t} $. Moreover, $$\label{defofvar}
\begin{aligned}
&\varphi(y_1 )=-x_1,~ \varphi(y_i)=s_{i-1}\varphi(y_{i-1}) s_{i-1}+(1+c_{i-1}c_{i})s_{i-1},\\
&\varphi(\bar y_1 )=\bar x_1 -\mbox{$\sum\limits_{i=1}^r$} e_{i,1} +\mbox{$\sum\limits_{i=1}^r$} c_i e_{i,1}c_i,~ \varphi(\bar y_i )= \bar s_{i-1}\varphi(y_{i-1}) \bar s_{i-1}-(1-\bar c_{i-1}\bar c_{i})\bar s_{i-1},\\
\end{aligned}$$ where $e_{i,1}=s_{i,1} e_1 s_{1,i}$. We claim that $\varphi$ is an algebra homomorphism. If so, since $ \varphi$ sends even (resp., odd) generators of $\widetilde{\mathscr{BC}}_{r,t}$ to even (resp., odd) generators of $\widetilde{{BC}}_{r,t}$, $\varphi$ has to be an even homomorphism. This shows that $\varphi$ is a superalgebra homomorphism.
We verify that the images of generators of $ \widetilde{\mathscr{BC}}_{r,t}$ satisfy the corresponding defining relations in Lemma \[CKu\]. First note that $\varphi$ preserves relations – and the last two relations in since the generators $ e_1$, $s_i$’s, $\bar s_j$’s, $c_i$’s and $c_j$’s satisfy the relations for $\widetilde{BC}_{r,t}$ (cf. [@CK Corollary 3.6]). It follows from the commutative relations in Definition \[awbsa\] that $\varphi$ preserves and the last two relations in . Since $e_1c_1x_1^ke_1=0$ and $ e_1x_1^{2k}e_1=0$, $\varphi$ preserves the first and third relations in . Since $ e_1 x_1^{k}e_1= w_k e_1$ and $\bar\delta_k=(-1)^k\omega_k$ in $R$, $\varphi$ preserves the second relation in . By , $\varphi$ preserves –. By [@CK (3.24)], $$\label{bary1p}
\bar y_1'=~\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (1,0) to[out=up, in=down] (1,2); \draw (1.1,2.0) node{\footnotesize{$r$}};
\draw[<-,thick] (2,0) to[out=up, in=down] (0,1) to[out=up, in=down] (2,2);
\draw (0,1) { node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}}; \end{tikzpicture}~
\begin{tikzpicture}[baseline = 10pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (2.25,0) to[out=up, in=down] (2.25,2);
\end{tikzpicture}^{t-1}$$ where $\bar y_1'=\bar y_1+\sum_{i=1}^r e_{i,1}+\sum_{i=1}^rc_ie_{i,1}c_i$. Hence, $e_1 y_1=e_1\bar y_1'$ by [@CK (3.24)] and . So, $\varphi(e_1)\varphi (y_1)=\varphi(e_1)\varphi(\bar y_1')$ by Definition \[awbsa\](4). Similarly, we have $\varphi (y_1)\varphi(e_1)=\varphi(\bar y_1')\varphi(e_1)$, proving that $\varphi$ preserves the first two relations in .
By definition of $\varphi$, any generator of $\widetilde{BC}_{r, t}$ has an inverse image in $\widetilde{\mathscr {BC}}_{r, t}$. This shows that $\varphi$ is an epimorphism. Now, we consider both $\widetilde{BC}_{r, t}$ and $\widetilde{\mathscr {BC}}_{r, t}$ over $F$, the quotient field. Since $\widetilde{\mathscr {BC}}_{r, t}$ has a basis which consists of all equivalence classes of dotted normal ordered oriented Brauer-Clifford diagrams, and each representative of an equivalent class is mapped to a unique regular monomial in $\text{gr}\widetilde{BC}_{r, t}$. So, $\phi$ is an isomorphism. Finally, we remark that one can use the freeness of $\widetilde{BC}_{r, t}$ to prove the freeness of $\widetilde{\mathscr {BC}}_{r, t}$ and vice versa via the isomorphism $\varphi$ or $\varphi^{-1}$.
Fix $a,b \in\mathbb Z^{\geq0}$ and $u_i\in R$ for each $1\leq i\leq a$. For a given polynomial $f(u)=u^b\prod_{i=1}^a (u^2-u_i)$, Comes and Kujuawa introduced the cyclotomic quotoent $ \mathscr {OBC}^f$ which is the supercategory defined as the quotient of $\mathscr {AOBC}$ by the left tensor ideal generated by $f({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
}) $. In the following, we consider the category ( which is also denoted by $ \mathscr{OBC}^f$ ) obtained from the original $ \mathscr{OBC}^f$ in [@CK] by specializing $\bar \delta_i$ at $(-1)^i w_i$. By [@CK Remark 3.11, Proposition 3.12], $$\label{delfor1}
(1-\mbox{$\sum\limits_{i\geq 1}$}\delta_{i-1}u^{-i})(1+\mbox{$\sum\limits_{j\geq 1}$}\bar \delta_{j-1}u^{-j})=1.$$ Motivated by [@BCNR Remark 1.6], for another given monic polynomial $g(x)$ of degree $\ell:=b+2a$, set $$\label{delfor2}
1+\mbox{$\sum\limits_{j\geq 1}$}\bar \delta_{j-1}u^{-j}:= \frac{g(u)}{f(u)}.$$ Write $f(u)=u^\ell+a_{\ell-1} u^{\ell-1}+ \cdots + a_1 u+a_0$. Then $$\label{fofy}
y:=(d \downarrow)\circ t \circ f({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})\circ t^{-1} \circ ( \downarrow c)= y_\ell + a_{\ell-1}y_{\ell-1} +\cdots+ a_1 y_1 +a_0 \in \operatorname{End}_{\mathscr {AOBC}}(\downarrow),$$ where $ d={
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[-,thick] (1,0) to[out=up, in=right] (0.53,0.5) to[out=left, in=right] (0.47,0.5);
\draw[->,thick] (0.49,0.5) to[out=left,in=up] (0,0);
\end{tikzpicture}
}$, $c={
\begin{tikzpicture}[baseline = 6pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[-,thick] (1,1) to[out=down, in=right] (0.53,0.5) to[out=left, in=right] (0.47,0.5);
\draw[->,thick] (0.49,0.5) to[out=left,in=down] (0,1);
\end{tikzpicture}
}$ and $ t={
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (1,1);
\draw[<-,thick] (1,0) to[out=up, in=down] (0,1);
\end{tikzpicture}
}$ and $y_k=(d \downarrow)\circ t \circ ({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})^k \circ t^{-1} \circ ( \downarrow c)$, $1\leq k\leq \ell$. By [@CK (3.24)], $$y_k=({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})^k- \sum_{i=0}^{k-1} \bar \delta_i ({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})^{k-i+1}.$$ By , one can check that $y=g({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})$. So, the right tensor ideal generated by $f({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})$ contains $g({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})$. Similarly, we also have that the right tensor ideal generated by $g({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[<-,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})$ also contains $f({
\begin{tikzpicture}[baseline = 3pt, scale=0.5, color={rgb:black,1;blue,4;red,1}]
\draw[->,thick] (0,0) to[out=up, in=down] (0,1);
\draw (0,0.4){ node[circle, draw, fill={rgb:black,1;blue,4;red,1}, thick, inner sep=0pt, minimum width=4pt]{}};
\end{tikzpicture}
})$. Let $\widetilde{ \mathscr{BC}}_{r,t}^f:= \widetilde {\mathscr{BC}}_{r,t}/J$, where $ J$ is the two sided ideal of $\widetilde {\mathscr{BC}}_{r,t} $ generated by $f(y_1)$ and $g(\bar y_1')$, and $\bar y_1'=\bar y_1+\sum_{i=1}^r e_{i,1}+\sum_{i=1}^rc_ie_{i,1}c_i$. Then $\widetilde{\mathscr{BC}}_{r,t}^f \cong \operatorname{End}_{\mathscr {OBC}^f }(\uparrow^r \downarrow^t)$ by [@CK (7.9)]. By the definition of $\varphi$ in the proof of Theorem \[isomofoabc\], we have $\varphi(f(y_1))=\tilde f(x_1)$, where $\tilde f(u)=(-1)^\ell f(-u)$.
The fact that $ \varphi(g(\bar y'_1))=g(\bar x_1)$ follows from [@CK (3.24)]. Finally, $e_1\tilde f(x_1)=(-1)^be_1 g(\bar x_1)$ follows from , and the fact that $\varphi$ is an isomorphism in Theorem \[isomofoabc\]. This leads to the following result.
\[isomofoabcf\] Let $BC_{\ell, r, t}:= \widetilde {BC}_{r,t}/I$, where $\widetilde {BC}_{r,t}$ is given in Theorem $\ref{isomofoabc}$ and $I$ is the two-sided ideal generated by $\tilde f(x_1)$ and $g(\bar x_1)$. Then $ BC_{\ell, r, t}$ is admissible and $ BC_{\ell, r,t}\cong \widetilde{\mathscr {BC}}_{r,t}^f$. In particular, $\widetilde{\mathscr {BC}}_{r,t}^f$ is free over $R$ with rank $\ell^{r+t} 2^{r+t} (r+t)!$, where $\ell$ is the degree of $\tilde f(x_1)$.
The isomorphism $\varphi$ in the proof of Theorem 7.2 leads to an $R$-epimorphism $\bar \varphi: \widetilde{\mathscr {BC}}_{r,t}^f\rightarrow BC_{\ell, r, t}$. Comes and Kujawa have proved that $\widetilde{\mathscr{BC}}_{r,t}^f$ is generated by all the equivalence classes of normally ordered dotted oriented Brauer-Clifford diagrams without bubbles of type $\mathbf a\rightarrow \mathbf a$ with fewer than $\ell$ dots on each strands, where $\mathbf a = \uparrow^r \downarrow^t$. Moreover, $\varphi$ sends all these generators to the corresponding regular [ monomials in]{} $ \text{gr} (BC_{\ell,r,t})$. One can check that $BC_{\ell, r,t}$ is admissible by . So, by Theorem \[level-k-walledcycl\], $ \varphi$ is injective and hence $ BC_{\ell,r,t}\cong \widetilde{\mathscr {BC}}_{r,t}^f$. In particular, $\widetilde{\mathscr {BC}}_{r,t}^f$ is free over $R$ with rank $\ell^{r+t} 2^{r+t} (r+t)!$.
[DWH99]{} , [[“]{}On algebras which are connected with the semisimple continuous groups[”]{}]{}, *Ann. of Math.* **38** (1937), 857–872. , [[“]{}A basis theorem for the degenerate affine oriented Brauer category and its cyclotomic quotients[”]{}]{}, *Quantum Topol.* **8** (1): 75–112 (2017). , [[“]{}Quantum walled Brauer-Clifford superalgebras, [”]{}]{}, *J. Algebra* **454** (2016) 433–474. , [[“]{}Highest weight categories arising from Khovanov’s diagram algebra I, II, III, IV[”]{}]{} *Moscow Math. J.* **11**, (2011), 685–722; *Transform. Groups* **15**, (2010), 1–45; *Represent. Theory* **15**, (2011), 170–243; *J. Eur. Math. Soc.* **14**, (2012), 373–419. , [[“]{}Gradings on walled Brauer algebras and Khovanov’s arc algebra]{}, *Adv. Math.* **231** (2012), no. 2, 709–773. , [[“]{}A basis theorem for the degenerate affine oriented Brauer-Clifford supercategory[”]{}]{}, *arXiv: 1706.09999v1\[math.RT\]*.
, [[“]{}Polynomial Representations of $GL_n$[”]{}]{}, Second corrected and augmented edition. With an appendix on Schensted correspondence and Littelmann paths by K. Erdmann, Green and M. Schocker, Lecture Notes in Math., Vol. **830**, Springer, Berlin, 2007.
, [[“]{}Degenerate affine Hecke-Clifford algebras and type Q Lie superalgebras[”]{}]{}, *Math.Z.*, **268** (3-4): 1091–1158 (2011).
, [[“]{}Mixed Schur-Weyl-Sergeev duality for queer Lie superalgebras[”]{}]{}, *J. Algebra*, **399**, (2014), 516-¨C545.
, [[“]{}Linear and projective representations of symmetric groups[”]{}]{}, [ Cambridge Tracts in Mathematics, 163. Cambridge University Press, Cambridge]{}, 2005.
, [[“]{}On the decomposition of tensor products of the representations of classical groups: By means of universal characters[”]{}]{}, *Adv. Math.* **74** (1989) 57–86.
, [[“]{}The first fundamental theorem of invariant theory for the orthosymplectic supergroup[”]{}]{}, *Comm. Math. Phys.*, **349** (2017), no. 2, 661–702.
, [[“]{}The representations of quantized walled Brauer algebras[”]{}]{}, *J. Pure and Applied Algebra*, **219**, 1496–1518.
, [[“]{}Affine walled Brauer algebras and super Schur-Weyl duality[”]{}]{}, *Adv. Math.*, **285** (2015), 28–71.
, [[“]{}Highest weight vectors of mixed tensor products of general linear Lie superalgebras[”]{}]{}, *Transform. Groups*, **20** (2015), no. 4, 1107–1140.
, [[“]{}The degenerate affine walled Brauer algebra [”]{}]{}, *J. Algebra*, **417**, (2014), 198–233.
, [[“]{}Tensor algebra of the identity representation as a module over the Lie superalgebras $Gl(n,m)$ and $Q(n)$, [”]{}]{}, *(Russian) Mat. Sb. (N.S.)* (165) (1984), no. 3, 422–430.
, [[“]{}Mixed tensor representations and representations for the general linear Lie superalgebras[”]{}]{}, *Comm. Algebra* **30** (2002), 839–857. , [[“]{}Operator invariants of tangles and R-matrices[”]{}]{}, *Izv. Akad. Nauk SSSR Ser. Math.* **53** (1989) 1073–1107 (in Russian).
[^1]: H. Rui is supported partially by NSFC (grant No. 11571108). L. Song is supported partially by NSFC (grant No. 11501368). Y. Su is supported partially by NSFC (grant No. 11371278 and 11431010)
[^2]: See [@Sa] (resp., [@BCNR]) where the affine walled Brauer algebra is defined over $\mathbb C$ (resp., over $R$ in terms of affine oriented Brauer category.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A minimalistic cognitive architecture called MANIC is presented. The MANIC architecture requires only three function approximating models, and one state machine. Even with so few major components, it is theoretically sufficient to achieve functional equivalence with all other cognitive architectures, and can be practically trained. Instead of seeking to trasfer architectural inspiration from biology into artificial intelligence, MANIC seeks to minimize novelty and follow the most well-established constructs that have evolved within various sub-fields of data science. From this perspective, MANIC offers an alternate approach to a long-standing objective of artificial intelligence. This paper provides a theoretical analysis of the MANIC architecture.'
author:
-
-
-
bibliography:
- 'refs.bib'
title: A Minimal Architecture for General Cognition
---
Introduction {#sec_introduction}
============
When artificial intelligence was first discussed in the 1950’s as a research project at Dartmouth [@McCarthy1955], many of the early pioneers in AI optimistically looked forward to creating intelligent agents. In 1965 Herbert Simon predicted that “machine will be capable, within twenty years, of doing any work man can do” [@Simon1965]. In 1967, Marvin Minsky agreed, writing “within a generation …the problem of creating ’artificial intelligence’ will be substantially solved” [@Minsky1967]. Since that time, artificial general intelligence has not yet been achieved as predicted. Artificial intelligence has been successfully applied in several specific applications such as natural language processing and computer vision. However, these domains are now subfields and sometimes even divided into sub-subfields where there is often little communication [@McCorduck2004].
Identifying the minimal architecture necessary for a particular task is an important step for focusing subsequent efforts to implement practical solutions. For example, the Turing machine, which defined the minimal architecture necessary for general computation, served as the basis for subsequent implementations of general purpose computers. In this paper, we present a Minimal Architecture Necessary for Intelligent Cognition (*MANIC*). We show that MANIC simultaneously achieves theoretical sufficiency for general cognition while being practical to train. Additionally, we identify a few interesting parallels that may be analogous with human cognition.
In biological evolution, the fittest creatures are more likely to survive. Analogously, in computational science, the most effective algorithms are more likely to be used. It is natural to suppose, therefore, that the constructs that evolve within the computational sciences might begin, to some extent, to mirror those that have already evolved in biological systems. Thus, inspiration for an effective cognitive architure need not necessarily originate from biology, but may also come from the structures that self-organize within the fields of machine learning and artificial intelligence. Our ideas are not rooted in the observations of human intelligence, but in seeking a simplistic architecture that could plausibly explain high-level human-like cognitive functionality.
In this context, we propose that high-level cognitive functionality can be explained with three primary capabilities: 1- environmental perception, 2- planning, and 3- sentience. For this discussion, we define **Environmental perception** as an understanding of the agent’s environment and situation. An agent perceives its environment if it models the environment to a sufficient extent that it can accurately anticipate the consequences of candidate actions. We define **Planning** to refer to an ability to choose actions that lead to desirable outcomes, particularly under conditions that differ from those previously encountered. We define **Sentience** to refer to an awareness of feelings that summarize an agent’s condition, coupled with a desire to respond to them. We carefully show that MANIC satisfies criteria 1 and 2. Criterion 3 contains aspects that are not yet well-defined, but we also propose a plausible theory for sentience to which MANIC can achieve functional equivalence.
Figure \[fig\_problems\] shows a subjective plot of several representative cognitive challenges. This plot attempts to rank these challenges according to our first two capabilites: perception and planning. Due to the subjective nature of these evaluations exact positions on this chart may be debated, but it is significant to note that the challenges typically considered to be most representative of human cognition are those that simultaneously require capabilities in both perception and planning. Problems requiring only one of these two abilities have been largely solved in the respective sub-discplines of machine learning and artificial intelligence, in some cases even exceeding human capabilities. It follows that human-like cognition requires an integration of the recent advances in both machine learning and artificial intelligence. MANIC seeks to identify a natural integration of components developed in these respective fields.
This document is laid out as follows: Section \[sec\_architecture\] describes the MANIC cognitive architecture. Section \[sec\_sufficiency\] shows that MANIC is theoretically sufficient to accomplish general cognition. Section \[sec\_training\] describes how the MANIC cognitive architecture can be practically trained using existing methods. Section \[sec\_sentience\] discusses a plausible theory for sentience, and describes how the MANIC architecture can achieve functional equivalence with it. Finally, Section \[sec\_conclusion\] concludes by summarizing the contributions of this paper.
![A subjective plot of several cognitive challenges. Advances in machine learning, especially with deep artifical neural networks, have solved many problems that require deep perception (top-left quadrant). Advances in artificial intelligence have solved many problems that require long-term planning (bottom-right quadrant). The unsolved challenges (mostly in the top-right quadrant) require a combination of both deep perception and long-term planning. Hence, the necessity of cognitive architectures, which combine advances in both sub-disciplines to address problems that require a combination of cognitive abilities.[]{data-label="fig_problems"}](problems.pdf){width="3.5in"}
Architecture {#sec_architecture}
============
An implementation of MANIC can be downloaded from https://github.com/mikegashler/manic.
A cognitive architecture describes a type of software agent. It operates in a world that may be either physical or virtual. It observes its world through a set of percepts, and operates on its world through a set of actions. Consistent with other simple artificial architectures, MANIC queries its percepts at regular time intervals, $t = 0, 1, 2, \ldots$, to receive corresponding vectors of observed values, $\mathbf{x}_0, \mathbf{x}_1, \mathbf{x}_2, \ldots$. It chooses action vectors to perform at each time, $\mathbf{u}_0, \mathbf{u}_1, \mathbf{u}_2, \ldots$. For our discussion here, MANIC assumes that its percepts are implemented with a camera (or renderer in the case of a virtual world), such that each $\mathbf{x}_t$ is a visual digital image. We use vision because it is well understood in the context of demonstrating cognitive abilities. Other percepts, such as a microphone, could also be used to augment the agent’s observations.
At a high level, the MANIC divides into two systems, which we call the *learning system* and the *decision-making system*. The learning system draws its architecture from constructs that are studied predominantly in the field of machine learning, and the decision-making system draws its architecture from the constructs that are studied predominantly in the partly overlapping field of artificial intelligence. The agent’s percepts feed into the learning system, providing the source of data necessary for learning, and the agent’s actions derive from the decision-making system, which determines the action vectors for the agent to perform. A diagram of the MANIC architecture is given in Figure \[fig\_agent\].
{width="5.8in"}
Learning system
---------------
The purpose of the learning system is to learn from past experiences. A large portion of the research effort in the field of machine learning thus far has focused on static systems. These are systems where the outputs (or labels) depend only on the current inputs (or features) and some component of unobservable noise. With a software agent, however, it is necessary for the agent to model its world as a dynamical system. (A dynamical system refers to one that changes over time, and should not be confused with a dynamic system, which is a different concept.)
Although many approaches exist for modeling dynamical systems in machine learning, nearly all of them either explicitly or implicitly divide the problem into three constituent components: a *transition model*, $f$, a *belief vector*, $\mathbf{v}_t$, and an *observation model*, $g$. The belief vector is an internal representation of the state of the system at any given time. In the context of a cognitive architecture, we refer to it as “belief" rather than “state" to emphasize that it is an intrinsic representation that is unlikely to completely represent the dynamical system, which in this case is the agent’s entire world, that it attempts to model. The transition model maps from current beliefs to the beliefs in the next time step, $\mathbf{v}_{t+1}=f(\mathbf{v}_t)$. It is implemented using some function approximating regression model, such as a feed-forward multilayer perceptron. The observation model is a bi-directional mapping between anticipated beliefs and anticipated observations, $\hat{\mathbf{x}}_t=g(\mathbf{v}_t)$, and $\mathbf{v}_t=g^+(\mathbf{x}_t)$, where $g^+$ approximates the inverse of $g$.
When trained, this learning system enables the agent to anticipate observations into the arbitrarily distant future (with decaying accuracy) by beginning with the current beliefs, $\mathbf{v}_t$, then repeatedly applying the transition function to estimate the beliefs at some future time step, $\mathbf{v}_{t+i}$. The future beliefs may be passed through the observation model to anticipate what the agent expects to observe at that time step. Because the agent is unlikely to ever successfully model its complex world with perfect precision, the anticipated observations may differ somewhat from the actual observations that occur when the future time step arrives. This difference provides a useful error signal, $e=\mathbf{x}_t-\hat{\mathbf{x}}_t$, which can be utilized to refine the learning system over time. In other words, the agent knows its learning system is well trained when the error signal converges toward a steady stream of values close to zero.
This learning system represents the common intersection among a wide diversity of models for dynamical systems. For example, an Elman network [@gao1996modified] is a well-established recurrent neural network suitable for modeling dynamical systems. It is typically considered as a single network, but can be easily segmented into transition and observation components, and its internal activations may be termed a belief vector. A NARMAX model, which is commonly used in system identification more explicitly segments into these three components [@chen1989representations]. A Jordan network is a recurrent neural network that lacks an observation model component, but it may still be considered to be a degenerate case of an Elman network that uses the identity function for its observation component. Many other approaches, including the extended Kalman filter [@haykin2001kalman] and LSTM networks [@hochreiter1997long] naturally fit within this learning system design.
### Transition model
Although the learning system as a whole is a recurrent model, the transition model may be implemented as a simple feed-forward model, because it only needs to predict $\mathbf{v}_{t+1}$ from $\langle \mathbf{v}_t, \mathbf{u}_t \rangle$.
### Observation model
The purpose of the observation model is to map between beliefs and observations. It is a bi-directional mapping between states and observations. Of all the components in MANIC, the observation model may arguably be the most critical for the overall success of the agent in achieving proficiency in general settings. If $g$ and $g^+$ are well-implemented, then the agent’s internal representation of beliefs will reflect a rich understanding of its world. This is one aspect of cognition where humans have traditionally excelled over machines, as evidenced by our innate ability to recognize and recall images so proficiently [@pinto2008real]. By contrast, machines have long been able to navigate complex decision chains with greater effectiveness than humans. For example, machines are unbeatable at checkers, and they can consistently trounce most humans at chess. Even decision-making processes that only utilize short-term planning may appear to exhibit much intelligence if they are based on a rich understanding of the situation.
{width="5.0in"}
Since about 2006, a relatively new research community has formed within the field of artificial neural networks to study “deep learning" architectures [@hinton2006fast; @oh2004gpu]. These deep learning architectures have demonstrated significant advances in ability for mapping graphical images to hierarchical feature representations. For example, in 2009, Lee and others showed that deep networks decompose images into constituent parts, much as humans understand the same images [@lee2009convolutional]. In 2012, Krizhevsky demonstrated unprecidented accuracy at visual recognition tasks using a deep learning architecture trained on the ImageNet dataset [@krizhevsky2012imagenet], and many other developments in this community have eclipsed other approaches for image recognition. Since MANIC assumes that observations are given in the form of digital images, and the task of the observation model is to map from these observations to meaningful internal representations, the developments of this community are ideally suited to provide the best implementation for the observation model. The encoding component of the observation model, $g^+$, may be implemented as a deep convolutional neural network, which is known to be particularly effective for encoding images [@lawrence1997face; @lecun2010convolutional; @ciresan2011committee]. The decoding component, $g$, may be implemented as a classic fully-connected deep neural network. Instead of predicting an entire observation image, it models the image as a continuous function. That is, it predicts all the color channel values for only a single pixel, but additional inputs are added to specify the pixel of interest. Such an architecture can be shown to implement the decoding counterpart of a deep convolutional neural network.
If an absolutely minimal architecture is desired, the encoder, $g^+$, may be omitted. Rather than calculating beliefs from the current observations, beliefs may be refined to make the anticipated observations match the actual observations. This inference-based approach has the advantage of remembering believed state even when it is not visible in the current observations. Hence, the encoder is only used in the rare cases when the agent is activated in a new context, but repeated applications of inference with the decoder would likely accomplish the same objective.
Decision-making system
----------------------
The decision making system produces plans to accomplish tasks that are anticipated to maximize the system’s contentment. Some clear patterns have also emerged among the many decision-making systems developed for artificial intelligence. Nearly all of them divide the problem into some form of model for estimating the utility of various possible states, and some method for planning actions that are expected to lead to desirable states. Let $h(\mathbf{v}_t)$ be the utility that the agent believes to be associated with being in state $\mathbf{v}_t$. Let $\mathbf{P}$ be a pool of candidate plans for maximizing $h$ that the agent considers, where each $\mathbf{p}_i\in\mathbf{P}$ is a sequence of action vectors, $\mathbf{p}_i=\langle \mathbf{u}_t, \mathbf{u}_{t+1}, \mathbf{u}_{t+2}, \cdots \rangle$. At each time step, the agent performs some amount of refinement to its pool of plans, selects the one that yields the biggest expected utility, and chooses to perform the first action in that plan.
At this high level, the model is designed to be sufficiently general to encapsulate most decision-making processes. For example, those that maintain only one plan, or look only one time-step ahead, may be considered to be degenerate cases of this model with a small pool or short plans. Processes that do not refine their plans after each step may implement the regular refinement as an empty operation. Implementation details of this nature are more properly defined in the lower-level components.
### Contentment model
Because MANIC is designed to be a long-living agent, we think of utility as being maintained over time, rather only being maximized in carrying out a single plan. Thus, we refer to $h$ as the *contentment model*. We assume that any tasks the agent is intended to accomplish are achieved through the agent’s constant efforts to preserve a state of homeostasis. For example, a robotic vacuum would be most “content" when the floor has been recently cleaned. In order to maintain this condition of contentment, however, it may need to temporarily stop cleaning in order empty its bag of dust, recharge its batteries, or avoid interfering with the owner. The contentment model, $h$, is trained to capture all of the factors that motivate the agent.
The contentment model is the only component of MANIC that benefits from a human teacher. Because the models in the learning system learn from unlabeled observations, direction from a teacher can be focused toward this one model. Competitive approaches can also be used to train the contentment model without a human teacher, as described in Section \[sec\_training\].
### Planning module
The planning module is a simple state machine that utilizes the three models to anticipate the outcome of possible plans, and evaluate their utility. A flow chart for this module is given in Figure \[fig\_planning\]. Much work has been done in the field of artificial intelligence to develop systems that guarantee convergence toward optimal plans. Unfortunately, these systems typically offer limited scalability with respect to dimensionality [@bellman:curseofdimensionality]. Early attempts at artificial intelligence found that it is much easier to exceed human capabilities for planning than it is to exceed human capabilities for recognition [@pinto2008real]. This implies that humans utilize rich internal representations of their beliefs to compensate for their relative inability to plan very far ahead. To approach human cognition, therefore, it is necessary to prioritize the scalability of the belief-space over the optimality of the planning. A good compromise with many desirable properties is found in genetic algorithms that optimize by simulated evolution. These methods maintain a population of candidate plans, can benefit from prior planning, are easily parallelized, and scale very well into high-dimensional belief spaces.
Sufficiency {#sec_sufficiency}
===========
Occam’s razor suggests that additional complexity should not be added to a model without necessity. In the case of cognitive architectures, additional complexity should probably give the architecture additional cognitive capabilities, or else its necessity should be challenged. In this section, we evaluate MANIC against Occam’s razor, and contrast it with two even simpler models to highlight its desirable properties.
A diagram of a simple *policy agent* is given in Figure \[fig\_policy\]. This agent uses a function approximating model to map from the current observations to actions. (It differs from a reflex agent [@russell1995modern] in that its model is not hard coded for a particular problem, but may be trained to approximate solutions to new problems that it encounters.) The capabilities of this architecture are maximized when it is implemented with a model that is known to be capable of approximating arbitrary functions, such as a feedforward artificial multilayer perceptron with at least one hidden layer [@cybenko1989ann_universal_function_approximators; @cybenko:ann_arbitrary_func_approx]. However, no matter how powerful its one model may be, there are many problems this architecture cannot solve that more capable architectures can solve. For example, if observations are not sufficient to uniqely identify state, an architecture with memory could do better. Therefore, we say that this policy agent is not sufficient to implement general cognition.
![A simple architecture that is not sufficient for general cognition.[]{data-label="fig_policy"}](policy.pdf){width="1.7in"}
A diagram of a *memory+policy agent* is given in Figure \[fig\_memory\_policy\]. This architecture extends the policy agent with memory. It uses two models, one to update its internal beliefs from new observations, and one that maps from its current beliefs to actions. This architecture can be shown to be theoretically sufficient for general cognition. If its belief vector is sufficiently large, then it can represent any state that might occur in its world. (This may not always be practical, but we are currently evaluating only the theoretical sufficiency of this architecture.) If its memory model is implemented with an arbitrary function approximator, then it can theoretically update its beliefs as well as any other architecture could from the given observations. If its policy model is implemented with an arbitrary function approximator, then it can theoretically choose actions as well as any other architecture could from accurate beliefs. Therefore, we can say that the memory+policy architecture is theoretically sufficient for general cognition.
![A simple architecture that is sufficient for general cognition, but is not practical to train.[]{data-label="fig_memory_policy"}](memory_policy.pdf){width="2.8in"}
Since the memory+policy model is theoretically sufficient for general cognition, it would generally not be reasonable to resort to greater complexity for the purpose of trying to achieve new theoretical capabilities. However, the memory+policy architecture also has a significant practical limitation: it is very difficult to train. It requires a teacher that can unambiguously tell it which actions to choose in every region of its belief space. Even if a teacher is available to provide such thorough training, a memory+policy agent would never be able to exceed the capabilities of its teacher, which renders it of limited practical value.
The MANIC architecture adds some additional complexity so that it can separate its learning system from its decision-making system. This separation enables it to learn from unlabeled observations. That is, it can refine the learning system’s ability to accurately anticipate the consequences of actions whether or not the teacher is available. With MANIC, supervision is only needed to guide its priorities, not its beliefs or choices. Yet, while being much more practical to train, MANIC is also provably sufficient for general cognition.
If we allow its belief vector, $\mathbf{v}_t$, to be arbitrarily large, then it it is sufficient to encode a correct representation of the agent’s world, no matter how complex that world may be. If the transition function, $f$, is implemented with an arbitrary function approximator, then it is theoretically sufficient to correctly anticipate any state transitions. And, if the decoder, $g$, is also implemented with an arbitrary function approximator, then it will be able to accurately anticipate observations from correct beliefs. If we allow the genetic algorithm of the planning system to utilize an arbitrarily large population, then it approximates an exhaustive search through all possible plans, which will find the optimal plan. Since the action it chooses is always the first step in an optimal plan, its actions will be optimal for maximizing contentment. Finally, if the contentment model, $h$, is also implemented with a universal function approximator, then it is sufficient to approximate the ideal utility metric for arbitrary problems. Therefore, MANIC is sufficient for general cognition. In the next Section, we discuss additional details about why MANIC is also practical for training.
Training {#sec_training}
========
Perhaps, the most significant insight for making MANIC practical for training comes from the high-level division between the learning system and the decision-making system. Because it divides where machine learning and artificial intelligence typically separate, well-established training methods become applicable, whereas such methods cannot be used with other cognitive architectures. Because the learning system does not choose candidate actions, but only anticipates their effect, it can learn from every observation that the agent makes, even if the actions are chosen randomly, and even when no teacher is available.
The learning system in MANIC is a recurrent architecture because the current beliefs, $\mathbf{v}_t$, are used by the transition model to anticpate subsequent beliefs, $\mathbf{v}_{t+1}$. Recurrent architectures are notoriously difficult to train, due to the chaotic effects of feedback that result from the recurrence [@cuellar:train_recurrent_networks_difficult; @sontag:neural_nets_for_control; @sjoberg:nonlinearblackboxmodeling; @floreano:evolve_recurrent_neural_nets; @blanco:evolutionary_rnn]. This tends to create both large hills and valleys throughout the error surface, so local optimization methods must use very small learning rates. Further, local optimization methods are susceptible to the frequent local optima that occur throughout the error space of recurrent architectures, and global optimization methods tend to be extremely slow.
However, recent advances in nonlinear dimensionality reduction provide a solution for cases where observations lie on a high-dimensional manifold. Significantly, this situation occurs when observations consist of images that derive from continuous space. In other words, MANIC can handle the case of a robot equipped with digital cameras, which has obvious analogy with humans equipped with optical vision.
![A system consisting of a simulated crane viewed through a camera. Each observation consists of 9216 values, but the system itself only exhibits 2 degrees of freedom, so these images lie on a 2-dimensional non-linear manifold embedded in 9216-dimensional space. This figure depicts the entire manifold as represented by uniform sampling over its nonlinear surface.[]{data-label="fig_crane"}](crane.png){width="2.6in"}
When observations are high-dimensional, a good initial estimate of state (or in this case, beliefs) can be obtained by reducing the dimensionality of those observations. Co-author Gashler demonstrated this approach in 2011 with a method that trained deep recurrent artificial neural networks to model dynamical systems [@gashler:tnldr]. For example, consider the crane system depicted in Figure \[fig\_crane\]. We used images containing $64\times48$ pixels in 3 color channels, for a total of 9216 values. We performed a random walk through the state space of the crane by moving the crane left, right, up, or down, at random to collect observations. (To demonstrate robustness, random noise was injected into state transitions as well as the observed images.) Using a nonlinear dimensionality reduction technique, we reduced the 9216-dimensional sequence of observed images down to just 2 dimensions (because the system has only 2 degrees of freedom) to obtain an estimate of the state represented in each high-dimensional image. (See Figure \[fig\_crane\_state\].) Significant similarity can be observed between the actual (left) and estimated (right) states. Consequently, this approach is ideal for bootstrapping the training of a recurrent model of system dynamics. When the beliefs are initialized to reasonable intial values, local optima is much less of a problem, so regular stochastic gradient descent can be used to refine the model from subsequent observations.
In the context of MANIC, this implies that nonlinear dimensionality reduction can be used to estimate each $\mathbf{v}_t$. Then, $g$ can be trained to map from each $\mathbf{v}_t$ to $\mathbf{x}_t$, $g^+$ can be trained to map from each $\mathbf{x}_t$ to $\mathbf{v}_t$, and $f$ can be trained to map from each $\mathbf{v}_t$ to $\mathbf{v}_{t+1}$. Note that each of these mappings depends on having a reasonable estimate of $\mathbf{v}_t$. These values are not typically known with recurrent architectures, but digital images provide sufficient information that unsupervised dimensionality reduction methods can estimate $\mathbf{v}_1, \mathbf{v}_2, \cdots$ from $\mathbf{x}_1, \mathbf{x}_2, \cdots$ very well. When an estimate for each $\mathbf{v}_t$ is known, training the various components of the learning system reduces to a simple supervised learning problem.
![**Left:** The hidden states visited in a random walk with a simulated crane system. Color is used to depict time, starting with red and ending with purple. The horizontal axis shows boom position, and the vertical axis shows cable length. The model was shown images of the crane, but was not allowed to view the hidden state. **Right:** Estimated states calculated by reducing the dimensionality of observed images. Although differing somewhat from the actual states, these estimates were close enough to bootstrap training of a recurrent model of system dynamics.[]{data-label="fig_crane_state"}](noisy_state.png){width="2.6in"}
A similar three-step approach can be used to bootstrap the learning system of MANIC:
[$\bullet$]{}
Gather high-dimensional observations.
Reduce observations to an initial estimate of beliefs.
Use beliefs to train the transition model.
The last step, training the transition model, is difficult with recurrent models because gradient-based methods tend to get stuck in local optima. However, because dimensionality reduction can estimate beliefs prior to training, it reduces to a simple supervised training approach. It has been shown that bootstrapping the training of neural networks can effectively bypass the local optima that otherwise cause problems for refining with gradient-based approaches [@bengio2007greedy]. This effect is much more dramatic with recurrent neural networks, since they create local optima throughout their model space [@cuellar:train_recurrent_networks_difficult]. Thus, after initial training, the system can be maintained using regular backpropagation to refine the learning system from subsequent observations.
With the crane dynamical system, we were able to accurately anticipate dynamics several hundred time-steps into the future [@gashler:tnldr], even with injected noise. To validate the plausibility of our planning system, we also demonstrated this approach on another problem involving a robot that navigates within a warehouse. We first trained the learning system on a sequence of random observations. Then, using only observations predicted by the learning system, we were able to successfully plan an entire sequence of actions by using the learning system to anticipate the “beliefs" of MANIC. The path we planned is shown in Figure \[fig\_warehouse\].B. We, then, executed this plan on the actual system. The actual states through which it passed are shown in Figure \[fig\_warehouse\].C. Even though MANIC represented its internal belief states differently from the actual system, the anticipated observations were very close to the actual observations.
{width="4.8in"}
Many other recent advances in dimensionality reduction with deep artificial neural networks validate that this general approach is effective for producing internal intrinsic representations of external observations.[@hinton2006reducing; @vincent2008extracting; @rifai2011contractive; @bengio2011unsupervised; @kingma2013auto]. The decision-making system contains one model, $h$, which needs to be trained to learn what constitutes homeostasis (or “contentment") for the system. This is done using a type of reinforcment learning. Because motivations are subjectively tied to human preferences, the motivations for an agent that humans would receive as intellgent necessarily depends on human teaching. Therefore, we assume that a human teacher is periodically available to direct the MANIC agent. In cases where no human teacher is available, the contentment model could also be trained using a competitive or evolutionary approach. This is done by instantiating multiple MANIC agents with variations in their contentment functions, and allowing the more fit instantiations to survive.
When MANIC makes plans, it utilizes its learning system to convert each plan from a sequence of actions to a corresponding video of anticipated observations. In many ways, these videos of anticipated observations may be analogous with the dreams or fantasies that humans produce internally as they sleep or anticipate future encounters. Although differences certainly exist, this similar side-effect may indicate that the architecture within the human brain has certain similarities with MANIC. Ironically, the imagination of the artificial system is more accessible than that of biological humans, enabling humans to examine the inner imaginings of the artificial system more intimately than they can with each other.
The videos of imagined observations are presented to the human teacher (when he or she is available) for consideration. The human, then, ranks these videos according to the disirability of the anticipated outcome. Note that these plans need not actually be executed to generate the corresponding video. Rather, the teacher only ranks imagined scenes. The precision with which the imagination of the agent corresponds with reality when it actually executes a plan depends only on the learning system (which is continually refined), and does not depend on the teacher. Because the teacher is only needed to refine the contentment model, only a reasonable amount of human attention is ever needed.
MANIC encapsulates a diversity of learning paradigms. The observation model, $g$, is trained in an unsupervised manner from camera-based observations. The transition model, $f$, and the observation model, $g$ and $g^+$, are trained by supervised learning. The contentment model, $h$, is trained by reinforcement from a human oracle. In 1999, Doya identified anatomical, physiological, and theoretical evidence to support the hypotheses that the cerebellum specializes in supervised learning, the cerebral cortex specializes in unsupervised learning, and the basal ganglia specialize in reinforcement learning [@doya1999computations]. This suggests a possible correlation between the components of MANIC and those in the brain. Finally, the planning module in MANIC ties all of its pieces together to make intelligent plans by means of a genetic algorithm. We consider it to be a positive property that the MANIC architecture unifies such a wide diversity of learning techniques in such a natural manner. As each of these methods exists for the purpose of addressing a particular aspect of cognition, we claim that a general cognitive architecture must necessarily give place to each of the major paradigms that have found utility in artificial intelligence and machine learning.
Sentience through self perception {#sec_sentience}
=================================
Sentience is a highly subjective and ill-defined concept, but is a critical aspect of the human experience. Consequently, it has been the subject of much focus in cognitive architectures. We propose a plausible theory that might explain sentience, and show that MANIC can achieve functional equivalence.
It is clear that MANIC *perceives* its environment, because it responds intelligently to the observations it makes. Its perception is implemented by its beliefs, which describe its understanding of the observations it makes, as well as its observation model, which connects its beliefs to the world. The term “awareness" is sometimes used with humans to imply a higher level of perception. It is not clear whether MANIC achieves “awareness", because the difference between “awareness" and “perception" is not yet well-defined. However, because we have proven that MANIC is sufficient for general cognition, we can say for sure that MANIC achieves something functionally equivalenct with awareness. That is, we know it can behave as if it is aware, but we do not know whether human awareness requires certain immeasurable properties that MANIC lacks.
Similarly, the term “sentience" contains aspects that are not yet well-defined, but other aspects of sentience are well-established. Specifically, sentience requires that an agent possess feelings that summarize its overall well-being, as well as an awareness of those feelings, and a desire to act on them. An agent that implements the well-defined aspects of sentience can be said to implement something functionally equivalent.
Our definition of sentience is best expressed as an analogy with perception: Perception requires the ability to observe the environment, beliefs that summarize or represent those observations, and a model to give the beliefs context. Similarly, we propose that sentience requires the ability to make introspective observations, “feelings" that summarize or represent them, and a model to give the feelings context. In other words, if sentience arises from self-awareness, then MANIC can achieve something functionally equivalent through self perception. In addition to a camera that observes the environment, MANIC can be equipped with the ability to observe its own internal state. (For example, it might be enabled to additionally observe the weights of its three models, $f$, $g$, and $h$, and its belief vector, $\mathbf{v}$.) Since MANIC is already designed for operating with high-dimensional observations, these introspective observations could simply be concatenated with the external observations it already makes. This would cause MANIC to utilize a portion of $\mathbf{v}$ to summarize its introspective observations.
Thus, $\mathbf{v}$ would represent both the “beliefs" and “feelings" of a MANIC agent. Its planning system would then implicilty make plans to maintaining homeostasis in both its beliefs and feelings. And its observation model would give context to its feelings by mapping between feelings and introspective observations, just as it does between beliefs and external observations. This theory of sentience is plausible with humans, because humans plan with regard to both their feelings and beliefs as a unified concept, maintaining both their external objectives and internal well-being, and because a well-connected brain, which humans have, would be sufficient to provide the “introspective observations" necessary to facilitate it.
Since MANIC learns to model its priorities from a human teacher in its contentment function, $h$, it will learn to give appropriate regard to its own feelings when they are relevant for its purpose. Presumably, this will occur when the human teacher directs it to maintain itself. Using the same observation model with both introspective and external observations, and using the same vector to model both feelings and beliefs, are both plausible because these design choices will enable MANIC to entangle its feelings with its environment, behavior that humans are known to exhibit.
Conclusion {#sec_conclusion}
==========
We presented a cognitive architecture called MANIC. This architecture unifies a diversity of techniques in the sub-disciplines of machine learning and artificial intelligence without introducing much novelty. Yet, while relying on existing methods, and with minimal complexity, MANIC is a powerful cognitive architecture. We showed that it is sufficiently general to accomplish arbitrary cognitive tasks, and that it can be practically trained using recent methods.
We supported MANIC’s design by referring to existing works that validate its individual components, and we made theoretical arguments about the capabilities that should emerge from combining them in the manner outlined by MANIC. The primary contribution of this paper is to show that these existing methods can already accomplish more of cognitive intelligence than is generally recognized. Our ultimate intent is to argue that if general intelligence is one of the ultimate objectives of the fields of machine learning and artificial intelligence, then they are very much on the right track, and it is not clear that any critical piece of understanding necessary for implementing a rudimentary consciousness is definitely missing.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
\
Physics Department,INFN,Rome University 1,Italy\
E-mail:
title: 'Upward and Horizontal $\tau$ Airshowers by UHE Neutrinos'
---
It is well known that Ultra High Energy $UHE$ neutrino of astrophysical origin above tens TeV might overcome the nearby noise vertical of secondary atmospheric neutrinos. The latter, being secondaries of charged cosmic rays, smeared by galactic magnetic fields, have lost their interesting astrophysical source records. In cubic kilometer underground detectors, both ice or water, in order to avoid the noisy downward atmospheric muons and to overcome the Earth opacity for vertical upward tens-TeV neutrinos, one [@01] should better neglect vertical signals and focus the attention mainly on Horizontal Underground detectors in kilometers wide disk-like or ring-like arrays finalized to trace horizontal UHE Muons and Taus ($10^{13}-
10^{18}eV$) born by UHE astrophysical $\nu_{\mu}$,$\nu_{\tau}$. Moreover because of $\tau$ amplified showering we prefer to suggest the UPTAUS and HORTAUS detection (after their parental UHE $(\nu_{\tau},\bar{\nu}_{\tau}) + N$ interactions in rock, the $\tau$ ejection in air and their fast decay in flight) as the best tool for UHE neutrino discovery. Indeed UHE $\nu_{\tau}$ and $\bar{\nu_{\tau}}$ may be flavor converted from common pion secondaries: $\nu_{\mu}$ and $\bar{\nu_{\mu}}$. The UHE neutrinos $\bar{\nu_e}$, ${\nu}_{\mu}$, $\bar{\nu}_{\mu}$, are expected Ultra High Energy Cosmic Rays (UHECR) ( $\gtrsim 10^{16}$ eV) secondary products near Active Galactic Nuclei (AGN) or micro-quasars jets by common photo-pion decay relics by optical photons nearby the source, either pulsars (PSR) or AGNs ($p + \gamma \rightarrow n + \pi^+,
\pi^+ \rightarrow \mu^+ \nu_{\mu}, \mu^+ \rightarrow e^+ \nu_e
\bar{\nu}_{\mu} $), or directly by proton proton scattering in interstellar matter. UHE neutrino flavor mix even at highest Greisen-Zatsepin-Kuzmin (GZK) energy ($ > 4\cdot 10^{19} eV$) because of the large galactic (Kpcs) and extreme cosmic (Mpcs) distances much longer than oscillation ones: $$L_{\nu_{\mu} - \nu_{\tau}} = 4 \cdot 10^{-3} \,pc \left(
\frac{E_{\nu}}{10^{16}\,eV} \right) \cdot \left( \frac{\Delta m_{ij}^2
}{(10^{-2} \,eV)^2} \right)^{-1}$$ HORTAUS are better detectable in deep valleys or on front of large mountain chains like Alps, Rocky Mountains, Grand Canyons, Himalaya and Ande just near present AUGER project [@03],[@05]. Future Array Telescope may trace at best such EeV ($10^{18}$ eV) HORTAUs in the Death Valley in USA by photoluminescent tracks [@04]. The mountain chains screens undesirable horizontal $(>70^o)$ UHECR showers; HORTAUS may lead also to UHE horizontal muon bundle. The Mountain chains acts also as a characteristic $\bar{\nu_e}$ detector at Glashow energy peak ($
6.3 PeV$). Present UPTAUS is analogous to the well-known [@06] “ $\tau $ double bang”. The novelty lays in the explosive $\tau$ decay in air after its escape from the rock leading to amplified tau air-showers in flight. The UPTAUS-HORTAUS channels reflects the known $\tau$ decay modes (Fig.2).
From the top of a mountain, a balloon or a satellite the Earth acts also as a huge target for UPTAUS or its wide corona crust for HORTAUS. Observing from a height h downward toward the Earth at any angle $\theta$ below the horizontal line,($\theta$ +$\pi/2$ = zenith angle), the distances $d{(\theta)}$ toward the ground, (from where an UPTAUS or HORTAUS should arise) is: $$d{(\theta)}= (R_{\oplus}+ h) \cdot \sin\theta -
\sqrt{(R_{\oplus}+ h)^{2}\cdot \sin^{2}\theta -(2hR_{\oplus} + h^2 )}$$ The distance length at horizontal tangential angle $\theta_{c}$ where the square-root term above vanishes, simplify in: $$d{(\theta_{c})} = \sqrt{(2 R_{\oplus}\cdot h)+ h^2}\simeq
110\sqrt{ \frac{h}{km} }\cdot km$$ Where $\theta_{c} = \arcsin \sqrt{(2 h/ R_{\oplus})}\simeq 1.01
^\circ \sqrt{(h/km) }$; the terrestrial cord distances $\bigtriangleup d{(\theta)}$ crossed by the primary UHE $\nu_{\tau}$ (and partially by the consequent upcoming $\tau$ before its exit in air) is $$\bigtriangleup d{(\theta)}= 2
\sqrt{(R_{\oplus}+ h)^{2}\cdot \sin^{2}\theta -(2hR_{\oplus} + h^2 )}$$ Such distances, which vanish for $\theta =\theta_{c}$, are not too long to suppress horizontal UHE neutrinos for small $\delta\theta$ ($= \theta-\theta_c$) angles above $\theta_{c}$, even at energies $E_{GZK}$ $ \simeq 4 \cdot 10^{19} eV$ energies. see Fig 1. The $\nu_{\tau}$+${\tau}$ crossed distances are $ \bigtriangleup
d(\delta\theta)$: $$\bigtriangleup d{(\delta\theta)}\simeq 2(R_{\oplus}+ h)
\sqrt{\delta\theta \sin 2\theta_c }\simeq 318 km \sqrt {\left(
\frac{\delta\theta}{1^{\circ}} \right) \sqrt {\left(
\frac{h}{km }\right)}}$$ The terrestrial surface below any high level observer covers huge areas A ($ A = 2\pi R_{\oplus} (1-cos(\theta_c)) \simeq 2 \pi h R_{\oplus} \simeq 4\cdot 10^4 \cdot km^2 (h/km)$ for $h <<
R_{\oplus}$); however for too distant UPTAUS origination the shower signal might be bounded by the longer crossed slant depth atmosphere opacity. The effective area for UPTAUS observed from height h $ > h_\circ$ ($h_\circ$ is the atmosphere exponential length $\simeq 8.55 km$ ) is smaller: $A_{eff} = \pi \cdot \cot^2(\theta) h^2 = 942 \cdot km^2 (h/ 10 \cdot km)^2; (\theta= 60^\circ).$ The Tau decay track (see the line $R_{\tau}$ in Fig.1), constrained by the characteristic distance horizons $d{(\theta_{c})}$ defines a fine tuned HORTAU energy: $E_{\tau} = 2\cdot 10^{18} eV \sqrt{(h/km)}$. This formulas cannot be extended to arbitrary energy (or any height h), because of the finite atmosphere size; see below. A too large tau lifetime may lead to $\tau$ decay in too empty atmosphere. Keeping care of the Earth opacity, at large nadir angle ($\gtrsim {60}^0$) where an average Earth density may be assumed ($< \rho > \sim 5$) the transmission probability and creation of upward UHE $\tau$ is approximately: $$P(\theta,\, E_{\nu}) = e^{\frac{- \bigtriangleup
d{(\theta)}}{R_{\nu_{\tau}}(E_{\nu})}} (1 - e^{-
\frac{R_{\tau}(E_{\tau})}{R_{\nu_{\tau}}(E_{\nu})}}) \, .$$ This expression should contain $\bigtriangleup d{(\theta)}$ from above equation and the ranges $R_{\nu_{\tau}}$ and $R_{\tau}$ [@01]are shown in Fig 1; for example at PeV the above probability is within a fraction of a million(${\theta}{\approx}{60}{^0}$) to a tenth of thousands (${\theta}{\approx}{\theta_c}$). At GZK energies only HORTAUS are allowed. The corresponding angular integral effective volume observable from a high mountain (or balloon) at height $h$ (assuming a final target terrestrial density $\rho = 3$) for UPTAUS at 3 PeV (for any AGN neutrino flux model normalized within a flat spectra whose energy fluence $ \phi_{\nu}\simeq 2 \,10^{3} \frac{eV}{cm^{2}\cdot
s}$), is:
$$V_{eff} \approx 0.3 \, km^3 {\left( \frac{\rho}{3} \right)}{\left( \frac{h}{km} \right)} e^{-
{\left( \frac{E}{3\,PeV} \right)}}
{\left( \frac{E}{3\,PeV} \right)}^{1.363}
{\left( \frac{\phi_{\nu}}{2 \,10^{3}\frac{eV}{cm^{2}\cdot s}} \right)}$$
Any AGN neutrino flux model normalized within such a flat spectra is leading, above 3 PeV, to $\sim$ 10 UHE $\nu_{\tau}$ upward event/km$^3$ year [@02]. The consequent average upward effective event rate on a top of a mountain (h $\sim 2\,km$) is:
$$N_{eff} \simeq 8 \, \frac{\mathrm{events}}{\mathrm{year}} {\left( \frac{\rho}{3} \right)} {\left( \frac{h}{2 \,km} \right)}
e^{- {\left( \frac{E}{3\,PeV} \right)}} {\left( \frac{E}{3\,PeV} \right)}^{1.363} {\left( \frac{\phi_{\nu}}{2 \,10^{3}\frac{eV}{cm^{2}\cdot s}} \right)}$$
Their signals at ten kms distances should be $\phi_{\gamma} \simeq
10^{-4} \div 10^{-5} cm^{-2} s^{-1}$, $(\phi_{X \sim 10^5 eV})
\simeq 10^{-2} \div 10^{-3}cm^{-2} s^{-1})$. The optical Cherenkov flux is large $\Phi_{opt} \approx 1 cm^{-2}$. We claimed [@01] that such UPTAUS or HORTAUS produce gamma bursts at the edge of GRO-BATSE originated from the Earth and named consequently as Terrestrial Gamma Flashes (TGF). The effective volume for UPTAUS and the event rate within an angle of view ($\theta> 60^o$) is, at 3 PeV, approximately to within 15 km$^3$ values and the expected UHE PeV rate is: $$N_{ev} \sim 150 \cdot e^{-
{\left( \frac{E_\tau}{3\,PeV} \right)}}
{\left( \frac{E_\tau}{3\,PeV} \right)}^{1.363} {\left( \frac{h}{400 km} \right)} \,
{\left( \frac{\phi_{\nu}}{2 \,10^{3}\frac{eV}{cm^{2}\cdot s}} \right)}
\frac{\mathrm{events}}{\mathrm{year}}$$ The TGF signals would be mainly $\gamma$ at flux $10^{-2}$ cm$^{-2}$ at X hundred keV energies. The correlations of these clustered TGFs directions toward GeV-MeV (EGRET), X sources, Milky Way Galactic Plane (Fig.3) support and make suggestive the TGF identification as UPTAUS and HORTAUS. The TGF location reflects the higher UPTAUS (and HORTAUS) interaction probability in the rock over the sea (and along the coastal plates). Highest magnetic field on Asia widely spreads UPTAUS making the TGF more observable. The present TGF-$\tau$ could not be produced by UHE $\bar{\nu}_e$ because of the severe Earth opacity and support the $\nu_{\mu}\leftrightarrow
\nu_{\tau}$ flavor mixing. The new physics interaction at TeV while forbid upward UHE signals in underground $km^3$ detectors it will amplify the $\nu_{\tau}$ signals beyond mountains, by two order of magnitude making extremely fruit-full UHE $\nu_{\tau}$ astrophysics in near future. HORTAUS may develop nor at too dense low atmosphere (being absorbed), neither at too high, low atmosphere (where no shower may be amplified). HORTAUS charged secondaries may also turn upward by geo-magnetic fields into fan-thin- cone jets, appearing as UPTAUS. The maximal $c\tau$ distance is ruled by: $$\label{13}
\int_{0}^{+ \infty} n_0 e^{-\frac{\sqrt{(c\tau+x)^2+R_\oplus^2} - R_\oplus}{h_0}}
dx \cong \int_{0}^{+ \infty} n_0 e^{-\frac{(c\tau+x)^2} {2h_0R_\oplus}}
dx \cong n_0 h_0 A$$ $$\label{15}
c\tau = \sqrt{2R_\oplus h_0}
\sqrt{ln {\left( \frac{R_\oplus}{c\tau} \right)} - ln A }$$ Where $A=A_{Had.}$ or $A=A_{\gamma}$ are parameters of order of unity, logarithmic function of energy, that calibrate the energy shower slant depth for both hadronic or electro-magnetic nature,[@04]: $A_{Had.}=0.792 \left[1+0.02523
\ln{\left( \frac{E}{10^{19}eV} \right)}\right];$ $A_{\gamma}=\left[1+0.04343\ln{\left( \frac{E}{10^{19}eV} \right)}\right].$ The solution of this equation leads to a characteristic UHE $c\tau_{\tau}$ = $546 \;km$ decay distance at height $h= 23$ km where the HORTAUS start to shower. This imply a possibility to discover efficiently by satellite and balloons arrays UHE $\nu_{\tau}$, $\bar\nu_{\tau}$ up to $1.11$ $10^{19}eV $. [@04; @07]. From high satellite the arrival HORTAUS angle maybe confused ($\mp 1^\circ$) with most common Albedo Horizontal High Altitude Showers (HIAS) [@04]. However from balloons heights and below, HIAS arrival angles split ($\geq 7^\circ$) from HORTAUS ones and are well distinguished. There is also the simplest possibility to observe UPTAUS and HORTAUS while they are hitting and lightening, via Cherenkov lights, upward mountain snow-walls. Such UPTAUS may also beam on lower boundary of high altitude clouds in the nights. These reflected flashing lights have a characteristic twin beam eight-shaped imprint that offers to Telescopes a new kind of signature for UHE Neutrino Astrophysics.
[999]{} D.Fargion: astro-ph/0002453, Accepted Ap.J.(2001) R. Gandhi, C. Quigg, M.H. Reno, I. Sarcevic: Phys. Rev. D [**58**]{}, 093009 (1998) D. Fargion, A. Aiello, R. Conversano, 26th ICRC, He6.1.09,p.396-398.1999.(USA); D. Fargion, 27th ICRC 2001, HE1.8,Vol-2, Pag. 903-906, 2001.(Germany). X. Bertou et all,astro-ph/0104452. J. G. Learned, S. Pakvasa: Astropart. Phys. [**3**]{}, 267 (1995) D. Fargion, A. Salis, B. Mele: Ap. J. [**517**]{}, 725–733(1999),USA.
| {
"pile_set_name": "ArXiv"
} |
Since Gaitan’s seminal work [@gaitan90; @gaitan92], significant advances have been made in our understanding of single-bubble sonoluminescence (SL). The dissociation hypothesis (DH) introduced by Lohse [*et al. *]{}[@lohse97] combines the merits of an intuitive approach based on the relatively tractable Rayleigh-Plesset equation (RPE) of bubble dynamics with an impressive ability to reproduce a wide range of observables [@holt96; @matula98; @ketterling98; @hilgenfeldt99; @dan99]. More sophisticated theories [@vuong99; @moss99] yield a more realistic picture of the phenomenon that is in general agreement with the results of the DH-RPE treatment. Experimentally, however, information on the bubble [*interior*]{} is still very scarce. In particular, no direct and conclusive evidence exists yet for either plasma formation or shock waves inside the collapsing bubble. As a result, many competing theories still vie with the adiabatic or shock-wave heating theory for the distinction of accurately describing the SL phenomenon. It seems desirable, then, to explore additional ways of probing the interior of the bubble with a time resolution comparable [@szeri99] to the duration of the flash, measured to be 40–380 ps [@gompf97; @hiller98; @moran98; @pecha98]. Light scattering has already been shown [@lentz95; @weninger97; @delgadino97] to be a useful probe of the bubble dynamics, sensitive as it is to the dielectric interface at the bubble wall. It is also a promising candidate for detection of either plasma or shock waves, since both features can modulate substantially the local dielectric constant.
Our goal was to push measurements of the light scattering cross section of the collapsing bubble to a greater temporal resolution than that afforded by the pulsed Mie scattering technique [@weninger97], which appears to be limited to around $\frac{1}{2}$ ns by low light levels and the need for averaging. In order to achieve higher resolution, we have developed a technique called differential light scattering (DLS) that reduces statistical uncertainty in the detection process by making use of more powerful ultrafast laser pulses. Since DLS does not rely on a fiducial time reference, it, too, is completely insensitive to electronic timing noise, allowing the use of relatively slow detectors. The DLS technique is based on two central concepts: (i) using a differential measurement to yield jitter-free timing information, and (ii) using polarized light to generate such a measurement through scattering.
The differential measurement concept was recently introduced for the first time by Rella [*et al.*]{} in the context of ultrafast gating of optical pulses [@rella98]. The technique they invented, differential optical gating (DOG), was used to measure the shape of a midinfrared pulse with subpicosecond resolution. Our technique applies the DOG concept to light scattering. DLS relies on collecting many pair of correlated samples of the same periodic event $I(t)$ (see Fig. \[fig1:dls\]). Each pair $i$ consists of the first sample $I(t_{i})$ and the second sample $I(t_{i}+\delta t)$, where $t_{i}$ is the time of the first sample (modulo the period $T$), and $\delta t$ is an appropriately chosen (and short) time delay. From each such pair, an intensity difference $$\label{eq:Idif}
\delta I_{i}=I(t_{i}+\delta t)-I(t_{i})$$ is produced and plotted against the first sample $I(t_{i})$, generating what we will call a DLS plot. When enough event pairs are collected, the points representing them in the DLS plot will join together in defining a continuous curve.
The DLS plot can be thought of as a predictor: given an intensity $I$ at some time $t$, the plot yields what $I$ will be after a time $\delta t$. One can numerically step along the curve on this plot to retrieve the desired direct function $I(t)$. Depending on the nature of the features of interest, in some cases it is more fruitful to plot, for each pair, $\delta I_{i}$ against the second sample $I(t_{i}+\delta t)$. The reconstruction of $I(t)$ is then carried out backwards in time. When the data are particularly noisy, however, features apparent in the DLS plot will be lost in the reconstruction process, eliminating any benefit of the technique. In this case, it is better to work directly in DLS space. Using a model for $I(t)$, a DLS curve can be generated from it by applying map (\[eq:Idif\]), then fit to the data points in the DLS plot with a minimization algorithm.
To implement the DLS concept (Fig. \[fig2:setup\]), a laser pulse is split equally in two and recombined as in a Michelson interferometer, with one pulse having traveled a longer path. The two-pulse train is focused onto the bubble, yielding two bursts of scattered light separated by an adjustable time delay $\delta t$. In order to distinguish between the two pulses during detection, an additional degree of freedom is needed. Color discrimination is a possibility, but with several drawbacks, among which the need for frequency doubling and the strong wavelength dependence of light scattering. Polarization, on the other hand, is perfectly suited to this technique. Calculations using Mie scattering theory [@hulst81], which is rigorously valid for spheres of arbitrary size, show that scattered intensities are highly polarization dependent. For the case of linearly polarized light, scattering at $\theta=90^{o}$ (where zero is forward) vanishes if the polarization vector [**e**]{} is parallel to the scattering plane. Therefore, the polarization of one of the pulses is made to rotate by $90^{o}$ (with the quarter-wave plate shown in Fig. \[fig2:setup\]) before the two pulses are recombined. Figure \[fig3:polar\] shows the sequence of the two pulses scattering from the bubble. The first pulse scatters preferentially in the plane containing one photomultiplier tube (PMT), while the second pulse does so in the perpendicular plane, which contains the other PMT.
Combining the time delay and the polarization dependence results in the ability to assign the scattered intensity recorded by one detector to the earlier pulse, and that recorded by the other detector to the later pulse. Therefore, each laser burst yields an ordered pair of scattered intensities. By scanning the arrival time of the pulse pairs over some portion of the acoustic cycle, enough data can be collected to generate a DLS plot of the desired interval of the bubble’s evolution.
The experiments were performed in a 100-ml spherical boiling flask filled with distilled, deionized water (resistivity ) at $20\pm1$ $^{o}$C. The water was prepared in a gas-handling system under an air pressure of 0.2 bar, then loaded into the flask without further exposure to air. A sealed connection to a volume reservoir (kept at atmospheric pressure for these experiments) inspired by provided pressure release from volume changes induced by temperature fluctuations. The entire assembly was leak tested; with the flask under vacuum, the rate of pressure rise was conservatively determined to be 16 nbar s$^{-1}$. However, under normal operating conditions the flask is filled with water and repressurized to 1 bar: this forces outside air to diffuse into the undersaturated water through microscopic interfaces, resulting in a substantially lower rate of contamination. The experiments described here took place 110 days after loading. Within that time, the pressure in an initially empty flask would have risen to roughly 0.2 bar; the air concentration in the water-filled flask can instead be expected to have risen by perhaps a few percent of atmospheric saturation.
The first acoustic resonance of the flask was determined to be 26.9 kHz. The acoustic drive was provided by an audio amplifier, the output of which (typically $\sim 4.5$ W rms) was fed through an impedance-matching network before being delivered in parallel to two disc-shaped piezoelectric transducers (PZTs) epoxied to diametrically opposite points on the flask. A third, smaller PZT cemented to the flask provided acoustic pickup, used to map the normal modes of the flask and to monitor the behavior of the bubble through its filtered acoustic signature.
The laser used to probe the bubble was a regenerative Ti:sapphire amplifier pumped by a [*Q*]{}-switched, frequency-doubled neodymium-doped yttrium lithium fluoride (Nd:YLF) laser operated at 1 kHz, 10 mJ/pulse (Positive Light Spitfire and Merlin, respectively) and seeded by a 82-MHz mode-locked Ti:Sapphire oscillator, in turn pumped by an Ar$^{+}$ cw laser (Spectra Physics Tsunami and Beamlok 2080, respectively). The oscillator provided 60-fs, 800-nm pulses at 82 MHz; the amplifier output consisted of partially uncompressed (chirped) 50-ps, 800-nm pulses at 1 kHz with approximately 1 mJ/pulse. The dominantly TEM$_{00}$ mode beam was sent through a spatial filter to clean up mode asymmetries and yielded a nearly Gaussian profile. To eliminate gross beam distortion caused by the irregular flask surfaces, a laser beam input port was made by cutting a hole in the flask and cementing in place a custom-made fused silica powerless meniscus. The light scattered by the bubble was collected with a relay system, passed through polarizers (appropriate for each branch) and 800-nm narrow bandpass filters, and delivered to two PMTs (Hamamatsu R955P and R636). The PMT signals were integrated by SRS SR250 boxcar averagers, which were in turn sampled by a 1-MHz A/D board on a personal computer.
The synchronization scheme (shown schematically in Fig. \[fig2:setup\]) involved generating a logic signal at , and digitally dividing its frequency by 27 to yield another logic signal at approximately 1 kHz. The 26.9 kHz logic signal was filtered before being fed to the audio amplifier to serve as the acoustic drive, while the 1 kHz signal was used to trigger the laser and the data acquisition electronics. This ensured that the SL drive signal and the regenerative amplifier pulse trains would be synchronized to each other to about 1-ns precision. Additional timing circuitry allowed for the delay between the SL flash (which occurs very nearly at the same point of the acoustic cycle, within 0.5 ns of turnaround [@weninger97]) and the laser pulse pairs to be varied continuously by up to 50 $\mu$s, either manually or automatically. This allowed us to probe the bubble at any given phase of the acoustic cycle.
In order to obtain values for the ambient bubble radius $R_{0}$ and the acoustic drive amplitude $P_{a}$, we developed a time-stamp technique that yielded a time series of scattered intensities $I(t)$ over the whole acoustic cycle. A time-to-amplitude converter (TAC, 566 EG&G ORTEC) measured the interval (up to a constant offset) elapsed between the arrival of the laser pulse and the SL flash, as signaled by an additional PMT sensitive to SL light only. The TAC output was logged through a boxcar along with the signal from one of the PMTs used in DLS, and used for time-stamping. Scattering events were recorded as the delay between the laser and SL was scanned automatically through a whole acoustic cycle.
Since the result of this procedure was a time series of intensities, a calibration was performed to establish a conversion from $I(t)$ to $R(t)$. This was done using a stroboscopic imaging system similar to that of , except that in our case the drive for the LED was locked to the same frequency $f_{acous}$ as that driving the bubble. We obtained $R_{max}$ by fixing the LED time delay so that the bubble was shown on the monitor screen at maximum size, and $I_{max}$ from the scattering data. A calculation based on Mie theory [@hulst81] provided the $I(R)$ map necessary to complete the calibration.
In practice, uncertainties in the calibration of the imaging system, as well as in the actual measurement of the bubble size, prompted us to use our measurement of $R_{max}$ as an estimate with $\pm$ 10% uncertainty. The $R(t)$ data were then fed to a fitting algorithm that established $R_{0}$, $P_{a}$, and an appropriate overall scale factor in a nonlinear least-squares calculation using the RPE. The fact that the scale factor for the best fit was determined in this way to be gave us confidence in the validity of our imaging method. The bubble parameter values thus found were and .
It is worth mentioning that the time-stamp technique described above can be used directly to obtain light scattering data from a collapsing SL bubble. The drawback is that, unlike in DLS, the electronic response of the measuring instruments is the limiting factor. We used this procedure to collect rough timing information as a cross-check in our analysis of DLS data; with the devices at our disposal, 2-ns resolution was achieved. We estimate that with two microchannel plate PMTs, two constant-fraction discriminators, and a faster TAC, an overall timing uncertainty of 50 ps should be achievable [@gompf97].
In Fig. \[fig4:data\] we show representative results from our DLS experiments. In these plots, as in Fig. \[fig1:dls\] (b), the abscissa is $I(t_{i})$ and the ordinate is $\delta I_{i}$. In Fig. \[fig4:data\](a) the delay between pulses was 5 ns, and in Fig. \[fig4:data\](b) it was 1 ns. The range of $\delta t$ for which useful information can be gathered is dictated by the physical process under study: delays much shorter than 1 ns yielded DLS plots unresolved into a discernible structure, while delays much longer than 5 ns are not well suited for investigating short time scales.
To aid interpretation, we divide the plots into three regions: [*A*]{} (the collapse, $t<0$), [*B*]{} (the transition region), and [*C*]{} (the rebound, $t>0$). The approximately flat region [*A*]{} corresponds to the collapse, since $\delta I_{i}<0$, indicating that the bubble is shrinking. In [*C*]{} the rebounding bubble expands, but at a much lower rate than during collapse, so $\delta I_{i}$ is positive and smaller in magnitude than in [*A*]{}. However, a greater spread in the data there results in fuzzy clustering across $\delta I_{i}=0$. The straight “wall” in region [*B*]{} forms when the two pulses straddle $t=0$ (compare to the open squares in Fig. \[fig1:dls\]).
The straight section [*A*]{} in Fig. \[fig4:data\](a) is due to a constant slope of $I(t)$ during collapse. This critical behavior [@barber97; @szeri99] has been previously observed for time scales ranging from 1 $\mu$s to 20 ns prior to turnaround [@delgadino97]; our observations extend it to $t=-5$ ns. In Fig. \[fig4:data\](a) sections [*A*]{} and [*B*]{} join rather abruptly, indicating a sharp cusp in $I(t)$ on the time scale of the measurement (5 ns). This was expected given the measurements in Ref. [@weninger97]. In Fig. \[fig4:data\](b), however, section [*A*]{} appears to show a slight upturn before joining section [*B*]{}, indicating a smooth transition on a time scale less than the pulse delay of 1 ns (since the “wall” section [*B*]{} is still discernible). Scatter in the data prevents a conclusive interpretation, but the available evidence would support an estimate of the bubble turnaround time at a few hundred picoseconds.
The collection lenses used have a of 1.5; the finite acceptance cone they subtend introduces a pollution, or cross talk, of unwanted light from the other pulse in each detector. Because of the strong polarization dependence of scattering, this cross talk is quite small: it was calculated from Mie theory, and confirmed experimentally, to be less than 5% of the total scattered intensity. Electrical cross talk was measured to be less than 5%. The resulting overall intensity uncertainty is therefore around 7%; the difference uncertainty varies across the plot. While in sections [*A*]{} and [*B*]{} the error estimates are consistent with the observed spread in the DLS data, in section [*C*]{} the spread is significantly larger. This has been observed before in Xe-filled bubbles and ascribed to nonsphericity [@weninger97]; such asymmetry is reported here as regularly occurring in air-filled bubbles.
In conclusion, we have introduced DLS, a light scattering technique based on the DOG [@rella98] concept of differential measurement and on sensitivity to polarization that uses intense ultrashort laser pulses to bypass the problem of electronic timing jitter. The intensity spread in the data is currently the limiting factor in the resolution achieved with this technique. Effectively, intensity noise is translated into timing noise by the mapping that a DLS plot generates. Accordingly, the resolution in the data shown is approximately 0.5 ns. The [*intrinsic*]{} resolution of DLS, however, is given by the laser pulse width used: with our equipment that can be made as low as 500 fs. Data collected from a collapsing SL bubble confirm earlier findings of a self-similar solution and of subnanosecond turnaround time; our preliminary results suggest that the turnaround is smooth on a time scale of a few hundred picoseconds.
We are very grateful to H. A. Schwettman and the Stanford Picosecond Free Electron Laser Center for supporting this research. We also gratefully acknowledge generous equipment loans by J. R. Willison of Stanford Research Systems. We thank B. P. Barber, B. I. Barker, F. L. Degertekin, R. A. Hiller, G. M. H. Knippels, G. Perçin, S. J. Putterman, C. W. Rella, H. L. Störmer, and members of the Stanford FEL for technical assistance and valuable discussions.
Author to whom correspondence should be addressed. Electronic address: [email protected] D. F. Gaitan, Ph.D. thesis, University of Mississippi, 1990 (unpublished). D. F. Gaitan, L. A. Crum, C. C. Church, and R. A. Roy, J. Acoust. Soc. Am. [**91**]{}, 3166 (1992). D. Lohse [*et al.*]{}, Phys. Rev. Lett. [**78**]{}, 1359 (1997). R. G. Holt and D. F. Gaitan, Phys. Rev. Lett. [**77**]{}, 3791 (1996). T. J. Matula and L. A. Crum, Phys. Rev. Lett. [**80**]{}, 865 (1998). J. A. Ketterling and R. E. Apfel, Phys. Rev. Lett. [**81**]{}, 4991 (1998). S. Hilgenfeldt, S. Grossmann, and D. Lohse, Nature (London) [**398**]{}, 402 (1999). M. Dan, J. D. N. Cheeke, and L. Kondic, Phys. Rev. Lett. [**83**]{}, 1870 (1999). V. Q. Vuong, A. J. Szeri, and D. A. Young, Phys. Fluids [**11**]{}, 10 (1999). W. C. Moss, D. A. Young, J. A. Harte, J. L. Levatin, B. F. Rozsnyai, G. B. Zimmerman, and I. H. Zimmerman, Phys. Rev. E [**59**]{}, 2986 (1999). C. Eberlein, Phys. Rev. Lett. [**76**]{}, 3842 (1996). W. Tornow, Phys. Rev. E [**53**]{}, 5495 (1996). A. Prosperetti, J. Acoust. Soc. Am. [**101**]{}, 2003 (1997). T. Lepoint [*et al.*]{}, J. Acoust. Soc. Am. [**101**]{}, 2012 (1997). L. Frommhold, Phys. Rev. E [**58**]{}, 1899 (1998). N. García and A. Hasmy, Pis’ma Zh. Éksp. Teor. Fiz. [**68**]{}, 442 (1998) \[JETP Lett. [**68**]{}, 472 (1998)\]. P. Mohanty and S. V. Khare, Phys. Rev. Lett. [**80**]{}, 189 (1998). J. R. Willison, Phys. Rev. Lett. [**81**]{}, 5430 (1998). A. J. Szeri and B. D. Storey (private communication). B. Gompf [*et al.*]{}, Phys. Rev. Lett. [**79**]{}, 1405 (1997). R. A. Hiller, S. J. Putterman, and K. R. Weninger, Phys. Rev. Lett. [**80**]{}, 1090 (1998). M. J. Moran and D. Sweider, Phys. Rev. Lett. [**80**]{}, 4987 (1998). R. Pecha [*et al.*]{}, Phys. Rev. Lett. [**81**]{}, 717 (1998). W. J. Lentz, A. A. Atchley, and D. F. Gaitan, Appl. Opt. [**34**]{}, 2648 (1995). K. R. Weninger, B. P. Barber, and S. J. Putterman, Phys. Rev. Lett. [**78**]{}, 1799 (1997). G. A. Delgadino and F. J. Bonetto, Phys. Rev. E [**56**]{}, R6248 (1997). C. W. Rella, G. M. H. Knippels, D. V. Palanker, and H. A. Schwettman, Opt. Commun. [**157**]{}, 335 (1998). H. C. van de Hulst, [*Light Scattering by Small Particles*]{} (Dover, New York, 1981), p. 124. R. A. Hiller, Ph.D. thesis, University of California at Los Angeles, 1995 (unpublished). Y. Tian, J. A. Ketterling, and R. E. Apfel, J. Acoust. Soc. Am. [**100**]{}, 3976 (1996). B. P. Barber [*et al.*]{}, Phys. Rep. [**281**]{}, 65 (1997).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A scheme for making [*ab-initio*]{} calculations of the dynamic paramagnetic spin susceptibilities of solids at finite temperatures is described. It is based on Time-Dependent Density Functional Theory and employs an electronic multiple scattering formalism. Incommensurate and commensurate anti-ferromagnetic spin fluctuations in paramagnetic $Cr$ and compositionally disordered $Cr_{95}V_{5}$ and $Cr_{95}Re_{5}$ alloys are studied together with the connection with the nesting of their Fermi surfaces. We find that the spin fluctuations can be described rather simply in terms of an overdamped oscillator model. Good agreement with inelastic neutron scattering data is obtained.'
address:
- ' Department of Physics, University of Warwick, Coventry CV4 7AL, U.K.'
- ' Department of Mathematics,Faculty of Science, Mahidol University,Bangkok 10400, Thailand.'
- '¶ Dipartimento di Fisica and Unita INFM, Universita di Messina, Italy'
- '§ Department of Materials Science and Engineering, University of Illinois, IL 61801, U.S.A.'
author:
- 'J.B.Staunton, J.Poulter, B.Ginatempo¶, E.Bruno¶and D.D.Johnson§'
title: 'Incommensurate and commensurate antiferromagnetic spin fluctuations in $Cr$ and $Cr$-alloys from ab-initio dynamical spin susceptibility calculations'
---
Chromium is the archetypal itinerant anti-ferromagnet (AF) whose famous incommensurate spin density wave (SDW) ground state is determined by the nesting wave-vectors ${\bf q}_{nest}$ identified in the Fermi surface [@Fawcett]. Chromium alloys also have varied AF properties [@Fawcett] and their paramagnetic states have recently attracted attention owing, in part, to analogies drawn with the high temperature superconducting cuprates especially $(La_{c}Sr_{1-c})_{2}CuO_{4}$. The incommensurate SDW fluctuations in these materials [@Mason1992] are rather similar to those seen in the paramagnetic phase of $Cr$ close to $T_{N}$. Moreover ‘parent’ materials $La_{2}CuO_{4}$ in the one instance and $Cr_{95}Mn_{5}$ or $Cr_{95}Re_{5}$ in the other are simple commensurate AF materials which on lowering the electron concentration by suitable doping develop incommensurate spin fluctuations which may be promoted by imperfectly nested Fermi surfaces.
Here we examine the nature of damped diffusive spin fluctuations in chromium above the Ne$\acute{e}$l temperature $T_{N}$ which are precursory to the SDW ground state. We also study dilute chromium alloys, $Cr_{95}Re_{5}$ and $Cr_{95}V_{5}$ and obtain good agreement with experimental data. For example, recent inelastic neutron scattering experiments [@Fawcett; @Hayden] have measured incommensurate AF ‘paramagnons’, persisting up to high frequencies in the latter system. We explore the temperature dependence, variation with dopant concentration and the evolution of the spin fluctuations in these systems from incommensurability to commensurability with increasing frequency and provide the first ab-initio description of these effects. To this end we describe a new scheme for calculating the wave-vector and frequency dependent dynamic spin susceptibility of metals which is based on the Time Dependent Density Functional Theory (TDDFT) of Gross et al. [@Gross_review] and as such is an all electron theory. For the first time the temperature dependent dynamic spin susceptibility of metals and alloys is calculated from this basis. There have been several simple parameterised models to describe the magnetic properties of $Cr$ and its alloys [@model]. All of these have concentrated on the approximately nested electron ‘jack’ and slightly larger octahedral hole pieces of the Fermi surface [@Fawcett] and, at best, have only included the effects of all the remaining electrons via an electron reservoir. We find similarities between our results and results from such models but show that a complete picture is obtained only when an electronic band-filling effect which favors a simple AF ordering at low temperature is also considered. We also find that the spin fluctuations are given an accurate description as overdamped diffusive simple harmonic oscillator modes which are at the heart of theories of the effects of spin fluctuations upon the properties of itinerant electron systems [@GGLetc].
Over the past few years great progress has been made in establishing TDDFT [@Gross_review]. Analogs of the Hohenberg-Kohn [@HK+KS] theorem of the static density functional formalism have been proved and rigorous properties found. Here we consider a paramagnetic metal subjected to a small, time-dependent external magnetic field, ${\bf b}({\bf r},t)$ which induces a magnetisation ${\bf m} ({\bf r},t)$ and use TDDFT to derive an expression for the dynamic paramagnetic spin susceptibility $\chi ({\bf q}, w)$ via a variational linear response approach [@Yang]. Accurate calculations of dynamic susceptibilities from this basis are scarce (e.g. [@SW+Sav]) because they are difficult and computationally demanding. Here we mitigate these problems by accessing $\chi ({\bf q}, w)$ via the corresponding [*temperature*]{} susceptibility $\bar{\chi} ({\bf q}, w_{n})$ where $w_{n}$ denotes a bosonic Matsubara frequency [@Fetter+Walecka]. We outline this approach below.
The equilibrium state of a paramagnetic metal, described by standard DFT, has density $\rho_{0}({\bf r})$ and its magnetic response function $\chi ({\bf r} t;{\bf r}^{\prime}t^{\prime}) =
(\delta m [b] ({\bf r},t)/\delta b ({\bf
r}^{\prime},t^{\prime}))|_{b=0,\rho_{0}}$ is given by the following Dyson-type equation. $$\chi ({\bf r}t;{\bf r}^{\prime}t^{\prime}) = \chi_{s} ({\bf r} t;{\bf
r}^{\prime} t^{\prime}) + \int d{\bf r}_{1} \int d t_{1} \int d{\bf r}_{2} \int
d t_{2} \chi_{s} ({\bf r} t;{\bf r}_{1}t_{1}) K_{xc}({\bf r}_{1} t_{1}; {\bf
r}_{2} t_{2}) \chi ({\bf r}_{2} t_{2}, {\bf r}^{\prime} t^{\prime}) \nonumber$$ $\chi_{s}$ is the magnetic response function of the Kohn-Sham non-interacting system with the same unperturbed density $\rho_{0}$ as the full interacting electron system, and $K_{xc}({\bf r} t;{\bf r}^{\prime}t^{\prime})
= (\delta b_{xc} ({\bf r},t)/\delta m ({\bf
r}^{\prime},t^{\prime}))|_{b=0,\rho_{0}}$ is the functional derivative of the effective exchange-correlation magnetic field with respect to the induced magnetisation. As emphasised in ref.[@Gross_review] eq.1 represents an exact representation of the linear magnetic response. The corresponding development for systems at finite temperature in thermal equilibrium has also been described [@Yang]. In practice approximations to $K_{xc}$ must be made and this work employs the adiabatic local approximation (ALDA) [@Gross_review] so that $K^{ALDA}_{xc} ({\bf r} t;{\bf
r}^{\prime} t^{\prime}) = (d^{2}b_{xc}^{LDA}(\rho({\bf r},t),m({\bf r}, t))/
dm^{2}({\bf r},t))|_{\rho_{0},m=0} \delta ({\bf r} -{\bf r}^{\prime}) \delta
(t-t^{\prime}) = I_{xc}({\bf r}) \delta ({\bf r} -{\bf r}^{\prime}) \delta
(t-t^{\prime})$. On taking the Fourier transform with respect to time we obtain the dynamic spin susceptibility $\chi ({\bf r}, {\bf r}^{\prime}; w)$.
For computational expediency we consider the corresponding [*temperature*]{} susceptibility [@Fetter+Walecka] $\bar{\chi} ({\bf r}, {\bf r}^{\prime};
w_{n})$ which occurs in the Fourier representation of the temperature function $\bar{\chi} ({\bf r} \tau;{\bf r}^{\prime} \tau^{\prime})$ that depends on imaginary time variables $\tau$,$\tau^{\prime}$ and $w_{n}$ are the bosonic Matsubara frequencies $w_{n}=
2 n \pi k_{B} T$. Now $\bar{\chi}( {\bf r}, {\bf
r}^{\prime}; w_{n}) \equiv \chi ({\bf r}, {\bf r}^{\prime}; i w_{n})$ and an analytical continuation to the upper side of the real $w$ axis produces the dynamic susceptibility $\chi ({\bf r}, {\bf r}^{\prime}; w)$. Using crystal symmetry and carrying out a lattice Fourier transform we obtain the following Dyson equation for the temperature susceptibility $$\bar{\chi}({\bf x},{\bf x}^{\prime},{\bf q},w_{n})= \bar{\chi}_{s}({\bf
x},{\bf
x}^{\prime},{\bf q},w_{n}) + \int d {\bf x}_{1} \bar{\chi}_{s}({\bf x},{\bf
x}_{1},{\bf q},w_{n}) I_{xc}({\bf x}_{1}) \bar{\chi}({\bf x}_{1}, {\bf
x}^{\prime},{\bf q},w_{n}) \label{eq:temp}$$ with ${\bf x}$,${\bf x}^{\prime}$ and ${\bf x}_{1}$ measured relative to crystal lattice unit cells of volume $V_{WS}$.
In terms of the DFT Kohn-Sham Green function of the static unperturbed system $$\bar{\chi}_{s}({\bf x},{\bf x}^{\prime},{\bf q},w_{n})= -\frac{1}{\beta N}
Tr. \sum_{{\bf R}} \sum_{m} G({\bf x},{\bf x}^{\prime}, {\bf R}, \mu
+i\nu_{m}) G({\bf
x}^{\prime},{\bf x},-{\bf R},\mu + i(
\nu_{m}+ w_{n})) e^{i {\bf q} \cdot {\bf R}} \label{eq:G}$$ where ${\bf R}$ is a lattice vector between the cells from which ${\bf x}$ and ${\bf x}^{\prime}$ are measured, $\mu$ the chemical potential and $\nu_{m}$ is a fermionic Matsubara frequency $(2n+1) \pi k_{B} T$. The Green function can be obtained within the framework of multiple scattering (KKR) theory [@Faulkner+Stocks]. This makes this formalism applicable to disordered alloys as well as ordered compounds and elemental metals, the disorder being treated by the Coherent Potential Approximation (CPA) [@KKRCPA]. Then the partially averaged Green function, $\langle G ({\bf r},{\bf r}^{\prime},z) \rangle_{{\bf r}\alpha, {\bf
r}^{\prime} \beta}$, where ${\bf r}$,${\bf r}^{\prime}$ lie within unit cells occupied by $\alpha$ and $\beta$ atoms respectively, can be evaluated in terms of deviations from the Green function of an electron propagating through a lattice of identical potentials determined by the CPA ansatz [@Butler].
To solve equation (\[eq:temp\]), we use the direct method of matrix inversion and local field effects are fully incorporated. $\bar{\chi}({\bf q},{\bf q}; w_{n})=(1/V_{WS}) \int d {\bf x} \int
d {\bf x}^{\prime} e^{i{\bf q} \cdot({\bf
x} - {\bf x}^{\prime})} \bar{\chi}({\bf x},{\bf x}^{\prime}, {\bf q},
w_{n})$ can then be constructed. The most computationally demanding parts of the calculation are the convolution integrals over the Brillouin Zone which result from the expression for $\bar{\chi}_{s}$, eq. (\[eq:G\]). Since all electronic structure quantities are evaluated at complex energies, these convolution integrals have no sharp structure and can be evaluated straightforwardly by an application of adaptive quadrature [@Ben+Ezio].
As discussed in ref. [@Fetter+Walecka], for example, we can define the retarded response function $\chi ({\bf q},{\bf q}, z)$ of a complex variable $z$. Since it can be shown [@Fetter+Walecka] formally that $\lim_{z \rightarrow \infty} \chi (z) \sim 1/z^{2}$ and we can obtain $\chi (i w_{n})$ from the above analysis it is possible to continue analytically to values of $z$ just above the real axis, i.e. $z= w +i \eta$. In order to achieve this we fit our data to a rational function $\bar{\chi} ({\bf q}, {\bf q},w_{n}) = \chi ({\bf q})
(1+\sum_{k=1}^{M-2} U_{k}({\bf q}) w_{n}^{k})/(1+\sum_{k=1}^{M}
D_{k}({\bf q}) w_{n}^{k})$ with the choice of coefficients $U_{k}$,$D_{k}$ ensuring that the sum rule involving the static susceptibility $\chi ({\bf q})$ is satisfied, i.e. $\chi
({\bf q})= (2/\pi) \int^{\infty}_{0} dw Im \chi({\bf q},{\bf q},w)/
w$. We find that very good fits are obtained with small $M$.
For chromium and its alloys, we find that $M=2$ is sufficient to provide excellent fits to the calculations of $\bar{\chi}$ over a wide range of $w_{n}$’s, i.e. $\bar{\chi}^{-1} ({\bf q}, {\bf q},w_{n}) = \chi^{-1}
({\bf q})(1 + (w_{n}/\Gamma({\bf q})) + (w_{n}/\Omega({\bf q}))^{2})$ so that $\chi^{-1}({\bf q}, {\bf q},w)= \chi^{-1} ({\bf q})(1 -i (w/\Gamma({\bf q}))
- (w/\Omega({\bf q}))^{2})$ (standard error $<$ 3$\%$ of mean). For the systems studied here we find $\Omega({\bf q})/\Gamma({\bf q}) < 0.15 \ll 1$ and so the [*spin dynamics can be described in terms of a heavily overdamped oscillator model*]{}. Evidently $t_{SF}({\bf q}) = \hbar/\Gamma({\bf q})
$ represents a relaxation time for a damped diffusive spin fluctuation with wavevector ${\bf q}$. Moreover, the imaginary part of the dynamical susceptibility which, when multiplied by $(1-\exp(-\beta w))^{-1}$, is proportional to the scattering cross-sections measured in inelastic neutron scattering experiments, is written $Im \chi ({\bf q}, {\bf q},w) =\chi({\bf q}) w \Gamma^{-1}({\bf q})
/((1 -(w/\Omega({\bf q}))^{2})^{2} + (w/\Gamma({\bf q}))^{2})$. We note that theories for the spin fluctuation effects upon itinerant electron properties, including quantum critical phenomena, also invoke such a model [@GGLetc]. The small ${\bf Q}$, $= ({\bf q} -{\bf q}_{0})$, dependence of $\chi^{-1}({\bf q}_{0}+{\bf Q})$ and $\Gamma ({\bf q}_{0}+{\bf
Q})$ is of particular importance. (${\bf q}_{0}$ is where $\chi^{-1}({\bf q})$ is smallest.)
Finite-temperature calculations were carried out for the static susceptibilities of the three systems, using the experimental b.c.c. lattice spacing of $Cr$, 2.88 $\dot{A}$, and von Barth-Hedin local exchange and correlation [@vBH]. We find that (i) $Cr$ orders into an incommensurate AF state below 280K specified by ${\bf q}_{0}=\{0,0,0.93\}$, where experiment yields $T_{N}= 311K$ and ${\bf q}_{0}=\{0,0,0.95\}$ [@Fawcett]; (ii) $Cr_{95}V_{5}$ does not develop magnetic order at any temperature, as found in experiment [@Fawcett; @Hayden]; and (iii) $Cr_{95}Re_{5}$ orders into a weakly incommensurate AF state below T=410K (${\bf q}_{0}=\{0,0,0.96\}$), whereas experimentally it forms a commensurate AF state below $T_{N}$ of 570K [@Fawcett].
Our calculations for $Im \chi ({\bf q}, {\bf q},w)$ are shown in figures 1(a) and (b) for $Cr_{95}V_{5}$ and $Cr_{95}Re_{5}$ respectively. Our calculation of $Im \chi ({\bf q}, {\bf q},w)$ for paramagnetic $Cr$ at $T=300K$ is broadly similar to that for paramagnetic $Cr$ at $T=0K$ by Savrasov [@SW+Sav] so a figure is not presented. It shows incommensurate spin fluctuations for small frequencies which are signified by peaks in $Im \chi({\bf q},{\bf q},w)$ at ${\bf q}_{0}$ which is equal to the Fermi surface nesting vector ${\bf q}_{nest}$. For increasing $w$ the peaks move to ${\bf q}=\{0,0,1\}$ i.e. the spin fluctuations become commensurate. The spin fluctuations at 300K shown in fig.1(a) for $Cr_{95}V_{5}$, on the other hand, remain incommensurate up to much higher frequencies maintaining intensity comparable to that at the peak at low $w$. This qualitative difference between the two systems has not been described before by a first-principles theory although found experimentally [@Fawcett; @Hayden]. For lower temperatures we find that $Im \chi({\bf q},{\bf q},w)$ of $Cr_{95}V_{5}$ becomes a sharper function of $w$ and we can also infer that $(1-\exp(-\beta w))^{-1} Im \chi({\bf q},{\bf q},w)$ should vanish for small $w$ when $T \rightarrow 0$K. These aspects have also been noted from experimental measurements [@Fawcett].
It is striking that the alloy’s Fermi surface is well defined [@KKRCPA] despite impurity scattering although it is more poorly nested (the difference between the sizes of the electron and hole octahedra is larger) than that of $Cr$ owing to its fewer electrons. Once again the peaks in $Im \chi({\bf q},{\bf q},w)$ occur at the nesting vectors ${\bf q}_{nest} = \{0,0,0.9\}$. The spin fluctuations in the paramagnetic phase of $Cr_{95}Re_{5}$ are shown in fig.1(b). Here adding electrons by doping with $Re$ has improved the Fermi surface nesting so that ${\bf q}_{nest} = \{0,0,0.96\}$ and $Im \chi ({\bf q},{\bf
q},w)$ has weight spread from ${\bf q}_{nest}$ to $\{0,0,1\}$. The dominant spin fluctuations now rapidly become commensurate with increasing $w$. We obtain a rather similar picture from the calculations for $Cr$ by artificially raising the chemical potential by a small amount. Interestingly when we account for thermally induced electron-phonon scattering [@model] by adding a small shift ($\approx 20$ meV) to the Matsubara frequencies $\nu_{m}$ in eq. (\[eq:G\]), we find a tendency for the dominant spin fluctuations to become commensurate at lower $w$ in both $Cr$ and $Cr_{95}Re_{5}$.
Some of these features also emerge qualitatively from the simple parameterised models based on that part of the band-structure near $\mu$ which leads to the nested electron and hole octahedra at the Fermi surface [@model]. Our ‘first-principles’ calculations, being based on an all-electron theory, however, need some additional interpretation. As analysed by recent total energy calculations [@Marcus], b.c.c. $Cr$ with the experimentally measured lattice spacing tends to form a commensurate AF phase at low temperatures which is modulated by a spin density wave of appropriate wavelength. The overall AF instability of the paramagnetic phase is promoted by the approximate half-filling of the narrow 3d-bands [@Heine+Samson] which is further modified by a weak perturbation coming from the Fermi surface nesting. As dopants such as $V$ and $Re$ are added not only is the Fermi surface nesting altered but also the d-bands become either further from or closer to being half-filled.
The calculations can be summarised in terms of the damped oscillator model. $\chi^{-1} ({\bf q}_{0}+{\bf Q}) \simeq \chi^{-1} ({\bf
q}_{0} + c Q^{2})$ for small ${\bf Q}$ for $Cr$ and $Cr_{95}V_{5}$ whereas for $Cr_{95}Re_{5}$, $\chi^{-1} ({\bf q})$ is nearly constant for a range of ${\bf q}$ between ${\bf
q}_{nest}$ and $\{0,0,1\}$. We find the product $\gamma ({\bf q})$ of $\chi ({\bf q})$ with damping factor $\Gamma({\bf q})$ to be only very weakly temperature dependent in these three systems and $\simeq \gamma ({\bf q}_{0})$, a constant, for small ${\bf Q}$, yielding a dynamical critical exponent [@GGLetc] of 2 typically assumed for antiferromagnetic itinerant electron systems. The nature of the spin fluctuations can be succinctly described via the variance $< m^{2}({\bf q})>$. From the fluctuation dissipation theorem, $< m^{2}({\bf q})> = (1/\pi) \int^{\infty}_{-\infty}
dw (1-\exp(-\beta w))^{-1} Im \chi ({\bf q},{\bf q}, w)$. Fig.2 shows $<
m^{2}({\bf q})>$ at several temperatures for $Cr$ where we have used a frequency cutoff of 500 meV and so have not included the faster of the quantum fluctuations. Near $T_{N}$ the magnetic fluctuations have their greatest weight around the ${\bf q}_{nest}$. At higher $T$ the peak diminishes and weight grows at ${\bf q}$’s nearer $\{ 0,0,1 \}$ reflecting the shift in the peak in $Im \chi({\bf q},{\bf q},w)$ from ${\bf q}_{nest}$ to commensurate ${\bf q}$’s with increase in frequency $w$. Similar plots to fig.2 for $Cr_{95}V_{5}$ and $Cr_{95}Re_{5}$ show respectively a smaller and greater tendency for the weight in $< m^{2}({\bf q})>$ to transfer in this way. If the frequency cutoff is reduced, $< m^{2}({\bf q})>$ near $\{ 0,0,1 \}$ is sharply diminished so that the Brillouin zone integral of $< m^{2}({\bf q})>$, $< m^{2} >$, decreases with increasing temperature as inferred from neutron scattering data [@Fawcett].
We have not included the effects of spin fluctuation interactions i.e. mode-mode coupling [@MD+Moriya+LT] into our calculations and have determined $T_{N}$ and the static susceptibility by what is essentially an ab-initio Stoner theory. In weak itinerant ferromagnets, for example, mode-mode coupling causes a dramatic suppression of the Curie temperatures from those estimated from a Stoner theory. For the $Cr$ systems studied here, the fair agreement with experiment which we obtain for $T_{N}$ and the relatively large value of the damping factor $\Gamma ({\bf q})$ with respect to that in weakly itinerant ferromagnets, is suggestive that such spin fluctuation effects are small. Spin fluctuation calculations have, however, been carried out by Hasegawa and others for simple parameterised models of $Cr$ neglecting Stoner particle-hole excitations and Fermi surface nesting [@Hasegawa]. Using a functional integral technique he made a high temperature (static) approximation so that all the thermally induced fluctuations were treated classically and found $T_{N}$ for commensurate AF order to be 370K and $\sqrt{< m^{2} >}$ to increase linearly with temperature above $T_{N}$. A quantitative calculation, however, in which the Stoner particle-hole excitations and spin fluctuations are treated within the same framework is needed to determine unequivocally whether or not a Stoner-like picture is adequate for these systems.
In summary, we have presented a first-principles framework for the calculation of dynamic paramagnetic spin susceptibilities of metals and their alloys at finite temperatures. At this point we add that the approach can also be applied to the study of magnetic excitations in magnetically ordered materials. The first applications on the AF spin fluctuations in $Cr$ and $Cr_{95}Re_{5}$ above $T_{N}$ and in paramagnetic $Cr_{95}V_{5}$ have found good agreement with available experimental data.
We are grateful to F.J.Pinski, S.Hayden and R.Doubble for useful discussions.
[99]{} E.Fawcett, Rev.Mod.Phys.[**60**]{}, 209, (1988); E.Fawcett et al., Rev.Mod.Phys.[**66**]{}, 25, (1994); S.A Werner et al., J.Appl.Phys. [**73**]{}, 6454, (1993); D.R.Noakes, Phys.Rev.Lett. [**65**]{}, 369, (1990). T.E.Mason et al., Phys.Rev.Lett. [**68**]{}, 1414, (1992). R.Doubble et al., Physica B [**237**]{}, 421, (1997). R.Doubble, PhD Thesis, University of Bristol, (1998). E.Runge et al., Phys.Rev.Lett. [**52**]{}, 997, (1984); E.K.U.Gross et al., Phys.Rev.Lett. [**55**]{}, 2850, (1985); E.K.U.Gross et al. in ‘Density Functional Theory’, ed. R.F.Nalewajski, Springer Series ‘Topics in current Chemistry’ (1996). e.g. G.G.Lonzarich, ch.6 in ‘Electron a centenary volume’, ed. M.Springford, C.U.P., (1997); S.Julian et al., J.Phys.: Condens. Matter, [**8**]{}, 9675, (1996); J.A.Hertz, Phys.Rev.B [**14**]{}, 7195, (1976); A.J.Millis, Phys.Rev.B [**48**]{}, 7183, (1993). T.K.Ng and K.S.Singwi, Phys.Rev.Lett. [**59**]{},2627, (1987); W.Yang, Phys.Rev.A [**38**]{},5512,(1988); K.Liu et al., Can.J.Phys. [**67**]{}, 1015, (1989). See ref.1 and H.Sato and K.Maki, Int.J.Magn.[**4**]{},163, (1973); Int.J.Magn.[**6**]{}, 183, (1974); R.S.Fishman and S.H.Liu, Phys.Rev.B [**47**]{}, 11870, (1993). P.Hohenberg and W.Kohn, Phys.Rev. [**136**]{}, B864, (1964); W.Kohn and L.J.Sham, Phys.Rev.[**140**]{}, A1133, (1965). e.g. E.Stenzel et al., J.Phys.F [**16**]{},1789, (1986); S.Y.Savrasov, Phys.Rev.Lett. [**81**]{}, 2570, (1998). A.L.Fetter and J.D.Walecka, ‘Quantum Theory of Many Particle Systems’, (McGraw-Hill), (1971). J.S. Faulkner and G.M. Stocks, Phys.Rev.B [**21**]{}, 3222, (1980). P.Soven, Phys.Rev.[**156**]{}, 809, (1967); G.M.Stocks et al. Phys.Rev.Lett. [**41**]{}, 339, (1978); G.M.Stocks and H.Winter, Z.Phys.B [ **46**]{}, 95, (1982); D.D.Johnson et al., Phys.Rev. B[**41**]{}, 9701 (1990). W.H.Butler, Phys.Rev. B[**31**]{}, 3260, (1985). E.Bruno and B.Ginatempo, Phys.Rev. B[**55**]{}, 12946, (1997). U.von Barth and L.Hedin, J.Phys.C [**5**]{}, 1629, (1972). P.M.Marcus et al., J.Phys.: Condens. Matter, [**10**]{}, 6541, (1998). V.Heine and J.Samson, J.Phys.F: Met[**13**]{}, 2155, (1983). K.K.Murata and S.Doniach, Phys.Rev.Lett. [**29**]{}, 285, (1972); T.Moriya, J.Mag.Magn.Mat. [**14**]{}, 1, (1979); G.G.Lonzarich and L.Taillefer, J.Phys.C [**18**]{}, 4339 (1985). H.Hasegawa, J.Phys.F [**16**]{}, 1555, (1986).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We consider one-dimensional stochastic differential equations with generalized drift which involve the local time $L^X$ of the solution process: $$X_t = X_0 + \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} L^X(t,y) \, \nu({\mbox{\upshape d}}y)\,,$$ where $b$ is a measurable real function, $B$ is a Wiener process and $\nu$ denotes a set function which is defined on the bounded Borel sets of the real line $\mathbb{R}$ such that it is a finite signed measure on $\mathscr{B}([-N,N])$ for every $N \in \mathbb{N}$. This kind of equation is, in dependence of using the right, the left or the symmetric local time, usually studied under the atom condition $\nu(\{x\}) < 1/2$, $\nu(\{x\}) > -1/2$ and $|\nu(\{x\})| < 1$, respectively. This condition allows to reduce an equation with generalized drift to an equation without drift and to derive conditions on existence and uniqueness of solutions from results for equations without drift. The main aim of the present note is to treat the cases $\nu(\{x\}) \geq 1/2$, $\nu(\{x\}) \leq -1/2$ and $|\nu(\{x\})| \geq 1$, respectively, for some $x \in \mathbb{R}$, and we give a complete description of the features of equations with generalized drift and their solutions in these cases.\
*Keywords:* Stochastic differential equations, local times, generalized drift, reflection, absorption, non-existence of solutions\
*2010 MSC:* 60H10, 60J55
author:
- |
Stefan Blei\
[email protected]
- |
Hans-Jürgen Engelbert\
[email protected]
bibliography:
- 'references.bib'
date: |
Friedrich-Schiller-Universität Jena,\
Fakultät für Mathematik und Informatik,\
Institut für Stochastik,\
D-07743 Jena, Germany\
title: 'A Note on One-dimensional Stochastic Differential Equations with Generalized Drift[^1]'
---
------------------------------------------------------------------------
------------------------------------------------------------------------
Introduction and Basic Definitions
==================================
Let be $b$ a measurable real function and $\nu$ a set function which is defined on the bounded Borel sets of the real line $\mathbb{R}$ such that it is a finite signed measure on $\mathscr{B}([-N,N])$ for every $N \in \mathbb{N}$. In the present note, we deal with the one-dimensional stochastic differential equation (SDE) with so-called *generalized drift* introduced as $$\label{eqn:SDE_mvd}
X_t = X_0 + \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} L^X(t,y) \, \nu({\mbox{\upshape d}}y)\,,$$ where $B$ is a Wiener process and $L^X$ denotes either the right, the left or the symmetric local time of the unknown process $X$. We call $\nu$ appearing in Eq. (\[eqn:SDE\_mvd\]) *drift measure.*\
SDEs of type (\[eqn:SDE\_mvd\]) with generalized drift have been studied previously by many authors. We refer the reader to Harrison and Shepp [@harrison_shepp], N.I. Portenko [@portenko], D.W. Stroock and M. Yor [@stroock_yor] and J.F. Le Gall [@LeGall_1983], [@LeGall_1984]. H.J. Engelbert and W. Schmidt [@engelbert_schmidt:1985], [@engelbert_schmidt:1989_III] derived rather weak necessary and sufficient conditions on existence and uniqueness of solutions to SDEs with generalized drift. More recently, R.F. Bass and Z.-Q. Chen [@bass_chen] also considered SDEs of type (\[eqn:SDE\_mvd\]).\
To treat the general equation (\[eqn:SDE\_mvd\]), in case of considering the right (resp., left, symmetric) local time the additional assumption $$\label{eqn:cond_atoms}
\nu(\{x\}) < \frac{1}{2} \quad (\text{resp., } \nu(\{x\}) > -\frac{1}{2}, \ |\nu(\{x\})| < 1), \qquad x \in \mathbb{R}\,,$$ is posed on $\nu$ (cf. [@engelbert_schmidt:1985],[@engelbert_schmidt:1989_III],[@LeGall_1984],[@stroock_yor]). This condition allows to reduce Eq. (\[eqn:SDE\_mvd\]) to an equation without drift, i.e., an equation of type (\[eqn:SDE\_mvd\]) where the drift measure is the zero measure. Therefore, well-known conditions on existence and uniqueness of solutions to equations without drift can be used to derive conditions on existence and uniqueness of solutions to Eq. (\[eqn:SDE\_mvd\]) (see e.g. [@engelbert_schmidt:1985],[@engelbert_schmidt:1989_III]). More precisely, under condition (\[eqn:cond\_atoms\]) the integral equation $$\label{eqn:integral_eqn}
g(x) = \left\{
\begin{array}{ll}
\displaystyle 1 - 2 \int_{[0,x]} F(g,y) \, \nu({\mbox{\upshape d}}y), & x \geq 0, \medskip \\
\displaystyle 1 + 2 \int_{(x,0)} F(g,y) \, \nu({\mbox{\upshape d}}y), & x \geq 0,
\end{array}
\right.$$ where $$F(g,x) = g(x-) \quad(\text{resp., } g(x), \ \left(g(x) + g(x-)\right)/2), \qquad x \in \mathbb{R},$$ admits a unique càdlàg solution. This solution $g$ is strictly positive and the strictly increasing and continuous primitive $G(x) = \int_0^x g(y)\, {\mbox{\upshape d}}y$ transforms Eq. (\[eqn:SDE\_mvd\]) into an equation without drift (cf. [@engelbert_schmidt:1985], Proposition 1, or [@engelbert_schmidt:1989_III], Proposition (4.29)). In the several cases, the explicit form of the solution to (\[eqn:integral\_eqn\]) can be found in [@engelbert_schmidt:1989_III], (4.26), (4.26$^\prime$) and (4.26$^{\prime\prime}$), respectively.\
The case $\nu(\{x\}) = 1/2$ (resp., $\nu(\{x\}) = -1/2$, $|\nu(\{x\})| = 1$) for some $x \in \mathbb{R}$ is excluded in (\[eqn:integral\_eqn\]) since it corresponds, as we will see, to a reflecting barrier at the point $x$, which requires different methods to treat Eq. (\[eqn:SDE\_mvd\]) than by assuming (\[eqn:cond\_atoms\]) (cf. W. Schmidt [@schmidt:1989], R.F. Bass and Z.-Q. Chen [@bass_chen]). Moreover, in the case that $\nu(\{x\}) > 1/2$ (resp., $\nu(\{x\}) < -1/2$, $|\nu(\{x\})| > 1$) holds for some $x \in \mathbb{R}$, in general, there is no solution to Eq. (\[eqn:SDE\_mvd\]).\
Indeed, in their famous paper on skew Brownian motion, J.M. Harrison and L.A. Shepp [@harrison_shepp] studied Eq. (\[eqn:SDE\_mvd\]) with symmetric local time in the special case $X_0 = 0$, $b \equiv 1$ and $\nu = \beta \, \delta_0$, where $\delta_0$ is the Dirac measure in zero, and they proved that there is no solution for the case $|\beta|>1$.\
Referring to [@harrison_shepp], J.F. Le Gall [@LeGall_1984] (see after the proof of [@LeGall_1984], Theorem 2.3) asserted, without giving a proof, that in his context ($b$ is of finite variation and bounded from below by a strictly positive constant) the result on the non-existence of a solution started at $x_0$ can be extended to Eq. (\[eqn:SDE\_mvd\]) with symmetric local time if $|\nu(\{x_0\})| > 1$.\
Also for symmetric local time, R.F. Bass and Z.-Q. Chen [@bass_chen] stated some propositions if $|\nu(\{x\})| > 1$ or $|\nu(\{x\})| = 1$ for some $x \in \mathbb{R}$ under the assumption, besides others, that the diffusion coefficient $b$ is bounded below by a positive constant (see [@bass_chen], Theorem 3.2 and 3.3). However, in both formulations and proofs, we feel that there is not enough clarity which had enabled us to follow their arguments. In particular, they have not pointed out where their assumptions on $b$ are used. But we shall see below that for general diffusion coefficient $b$ their Theorem 3.2 does not remain true.\
The purpose of the present note is to give a general, complete and rigorous approach to Eq. (\[eqn:SDE\_mvd\]) under the condition that, contrary to (\[eqn:cond\_atoms\]), for some $x \in \mathbb{R}$ the drift measure $\nu$ satisfies $$\nu(\{x\}) > 1/2 \quad (\text{resp., } \nu(\{x\}) < -1/2, \ |\nu(\{x\})| > 1)$$ or $$\nu(\{x\}) = 1/2 \quad (\text{resp., } \nu(\{x\}) = -1/2, \ |\nu(\{x\})| = 1).$$ The basic idea is to provide an insight into the behaviour of the local times $L^X(t,x)$ of solutions $X$ of Eq. (\[eqn:SDE\_mvd\]) in such points $x \in \mathbb{R}$ violating (\[eqn:cond\_atoms\]). This gives rise for an application of Tanaka’s formula, followed by a space transformation, to conclude full information about the features of the solution.\
Throughout the paper, $(\Omega, \mathcal{F}, \mathbf{P})$ stands for a complete probability space endowed with a filtration $\mathbb{F} = (\mathcal{F}_t)_{t \geq 0}$ which satisfies the usual conditions, i.e., $\mathbb{F}$ is right-continuous and $\mathcal{F}_0$ contains all sets from $\mathcal{F}$ which have $\mathbf{P}$-measure zero. For a process $X = (X_t)_{t \geq 0}$ the notation $(X,\mathbb{F})$ indicates that $X$ is $\mathbb{F}$-adapted. The processes considered in the following belong to the class of continuous semimartingales up to a stopping time $S$ and local times of such processes will be an important tool. Given an $\mathbb{F}$-stopping time $S$, we say that $(X,\mathbb{F})$ is a *semimartingale up to $S$* if there exists an increasing sequence $(S_n)_{n \in \mathbb{N}}$ of $\mathbb{F}$-stopping times such that $S = \lim_{n\rightarrow +\infty} S_n$ and the process $(X^{S_n},\mathbb{F})$ obtained by stopping $(X,\mathbb{F})$ in $S_n$ is a real-valued semimartingale for every $n\in\mathbb{N}$. Analogously, we introduce the notion of a *local martingale up to $S$*.\
If $(X,\mathbb{F})$ is a semimartingale up to $S$, then we can find a decomposition $$\label{eqn:semi_decomposition}
X_t = X_0 + M_t + V_t\,, \qquad t < S, \ \mathbf{P}\text{-a.s.},$$ on $[0,S)$, where $(M,\mathbb{F})$ is a local martingale up to $S$ with $M_0 = 0$ and $(V, \mathbb{F})$ is a right-continuous process whose paths are of bounded variation on $[0,t]$ for every $t < S$ and with $V_0=0$. If $X$ is continuous on $[0,S)$, then there exists a decomposition such that $M$ and $V$ are continuous on $[0,S)$ and this decomposition is unique on $[0,S)$. For any continuous local martingale $(M, \mathbb{F})$ up to $S$ by ${ \ensuremath{\langle M \rangle}}$ we denote the continuous increasing process, which is uniquely determined on $[0,S)$, such that $(M^2 - { \ensuremath{\langle M \rangle}}, \mathbb{F})$ is a continuous local martingale up to $S$ and ${ \ensuremath{\langle M \rangle}}_0 = 0$. For a continuous semimartingale $(X,\mathbb{F})$ up to $S$ we set ${ \ensuremath{\langle X \rangle}} = { \ensuremath{\langle M \rangle}}$, where $M$ is the continuous local martingale up to $S$ in the decomposition (\[eqn:semi\_decomposition\]) of $X$.\
We now recall some facts which are well-known for continuous semimartingales (see for example [@revuzyor], Ch. VI, §1). Their extension to semimartingales $(X,\mathbb{F})$ up to an $\mathbb{F}$-stopping time $S$ is obvious. For $(X,\mathbb{F})$ there exists the right (resp., left, symmetric) local time $L^X$ up to $S$ which is a function on $[0,S) \times \mathbb{R}$ into $[0,+\infty)$ such that for every real function $f$ which is the difference of convex functions the *generalized Itô formula* holds: $$\label{eqn:gen_ito_formula}
f(X_t) = f(X_0) + \int_0^t f^\prime (X_s) \, {\mbox{\upshape d}}X_s + \frac{1}{2} \int_0^t L^X(t,y) \, {\mbox{\upshape d}}f^\prime(y)\,, \qquad t < S, \ \mathbf{P}\text{-a.s.}$$ where $f^\prime$ denotes the left (resp., right, symmetric) derivative of $f$. To indicate explicitly which local time we consider, we write $L_+^X$ (resp., $L_-^X$, $\hat{L}^X$) for the right (resp., left, symmetric) local time of $X$. If a formula or statement holds for every type of local time we just use the symbol $L^X$. Note that we can choose $L_+^X$ (resp., $L_-^X$) to be increasing and continuous in $t < S$ and right (resp. left) continuous with left (resp. right) hand limits in $x$. Moreover, we have the relation $\hat{L}^X = (L_+^X + L_-^X)/2$.\
The local times fulfil $$\label{eqn:int_wrt_loc_time}
\int_0^t {\text{\large{$\mathds{1}$}}}_{\{y\}}(X_s) \, L^X({\mbox{\upshape d}}s,y) = L^X(t,y)\,, \qquad t < S, \ y \in \mathbb{R}, \ \mathbf{P}\text{-a.s.},$$ $$\label{eqn:loc_time_and_variation_process}
L_+^X(t,y) - L_-^X(t,y) = 2 \int_0^t {\text{\large{$\mathds{1}$}}}_{\{y\}}(X_s) \, {\mbox{\upshape d}}V_s\,, \qquad t < S, \, y \in \mathbb{R}, \ \mathbf{P}\text{-a.s.},$$ $$\label{eqn:loc_time_zero_outside_compact_interval}
L^X(t,y) = 0, \qquad t < S, \, y \notin \left[\min_{0 \leq s \leq t} X_s, \max_{0 \leq s \leq t} X_s\right], \ \mathbf{P}\text{-a.s.}$$ and $$\label{eqn:loc_time_as_limit}
L^X(t,y) = \lim_{\varepsilon \downarrow 0}\frac{1}{\varepsilon} \int_0^t I^y_\varepsilon(X_s) \, {\mbox{\upshape d}}{ \ensuremath{\langle X \rangle}}_s,
\qquad t < S, \ y \in \mathbb{R}, \ \mathbf{P}\text{-a.s.},$$ where $$I^y_\varepsilon (x) = {\text{\large{$\mathds{1}$}}}_{[y,y+\varepsilon)}(x) \quad (\text{resp.,} \ {\text{\large{$\mathds{1}$}}}_{(y-\varepsilon, y]}(x), \
\frac{1}{2} \, {\text{\large{$\mathds{1}$}}}_{(y-\varepsilon, y+\varepsilon)}(x)), \qquad x \in \mathbb{R}.$$ In general, SDEs of type (\[eqn:SDE\_mvd\]) admit exploding solutions. Therefore, the convenient state space is the extended real line $\overline{\mathbb{R}} = \mathbb{R}\cup \{-\infty, +\infty\}$ equipped with the $\sigma$-algebra $\mathscr{B}(\overline{\mathbb{R}})$ of Borel subsets. We fix the notion of a solution to Eq. (\[eqn:SDE\_mvd\]) in the following
\[def:solution\] A continuous $(\overline{\mathbb{R}}, \mathscr{B}(\overline{\mathbb{R}}))$-valued stochastic process $(X,\mathbb{F})$ defined on a probability space $(\Omega,\mathcal{F},\mathbf{P})$ is called a solution to Eq. (\[eqn:SDE\_mvd\]) if the following conditions are fulfilled:
\(i) $X_0$ is real-valued.
\(ii) $X_t = X_{t \wedge S_\infty^X}$, $t \geq 0$, $\mathbf{P}$-a.s., where $S_\infty^X := \inf\{t \geq 0:|X_t| = +\infty\}$.
\(iii) $(X,\mathbb{F})$ is a semimartingale up to $S_\infty^X$.
\(v) There exists a Wiener process $(B,\mathbb{F})$ such that Eq. (\[eqn:SDE\_mvd\]) is satisfied for all $t < S_\infty^X$ $\mathbf{P}$-a.s.
The Results
===========
We recall that in Eq. (\[eqn:SDE\_mvd\]) $L^X$ stands either for the right, the left or the symmetric local time and we treat these three cases simultaneously in the sequel. For the sake of brevity, in the following we write for example $\{\nu \geq 1/2\}$ instead of $\{y \in \mathbb{R}: \nu(\{y\}) \geq 1/2\}$.
\[lemma:loc\_time\_zero\] Let $(X,\mathbb{F})$ be a solution of Eq. (\[eqn:SDE\_mvd\]) with right (resp., left, symmetric) local time. Then it holds $$\begin{split}
L_-^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in \{\nu \geq 1/2\}, \ \mathbf{P}\text{-a.s.} \phantom{\Bigr)}\\
\Bigl(\text{resp., } L_-^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in \{\nu < -1/2\}, \ \mathbf{P}\text{-a.s.}, \\
L_-^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in\{|\nu| > 1 \text{ or } \nu = 1\}, \ \mathbf{P}\text{-a.s.}\Bigr)
\end{split}$$ and $$\begin{split}
L_+^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in \{\nu > 1/2\}, \ \mathbf{P}\text{-a.s.} \phantom{\Bigr)}\\
\Bigl(\text{resp., } L_+^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in \{\nu \leq -1/2\}, \ \mathbf{P}\text{-a.s.}, \\
L_+^X(t,x) = 0, \qquad & t < S_\infty^X, \ x \in \{|\nu| > 1 \text{ or } \nu = -1\}, \ \mathbf{P}\text{-a.s.} \Bigr)
\end{split}$$
Using (\[eqn:int\_wrt\_loc\_time\]) and (\[eqn:loc\_time\_and\_variation\_process\]), we see $$\begin{split}
L_+^X(t,x)-L_-^X(t,x) &= 2 \int_0^t {\text{\large{$\mathds{1}$}}}_{\{x\}} (X_s) \int_\mathbb{R} L^X({\mbox{\upshape d}}s, y) \, \nu({\mbox{\upshape d}}y) \\
&= 2 \, L^X(t, x) \, \nu(\{x\}), \qquad t < S_\infty^X, \ x \in \mathbb{R}, \ \mathbf{P}\text{-a.s.}
\end{split}$$ and it follows $$\begin{split}
\bigl(1-2\nu(\{x\})\bigr) \, L_+^X(t,x) &= L_-^X(t,x), \qquad t < S_\infty^X, \ x \in \mathbb{R}, \ \mathbf{P}\text{-a.s.} \phantom{\Bigr)}\\
\Bigl(\text{resp., }
\bigl(1+2\nu(\{x\})\bigr) \, L_-^X(t,x) &= L_+^X(t,x), \qquad t < S_\infty^X, \ x \in \mathbb{R}, \ \mathbf{P}\text{-a.s.}, \\
\bigl(1-\nu(\{x\})\bigr) \, L_+^X(t,x) &= \bigl(1+\nu(\{x\})\bigr)\,L_-^X(t,x), \qquad t < S_\infty^X, \ x \in \mathbb{R}, \ \mathbf{P}\text{-a.s.} \Bigr)\\
\end{split}$$ which, together with the non-negativity of the local times, implies the claims.
The following theorem is the main result of the present note.
\[theorem:reflection\_and\_absorbing\] Let $(X,\mathbb{F})$ be a solution of Eq. (\[eqn:SDE\_mvd\]) with right (resp., left, symmetric) local time started at $x_0 \in \mathbb{R}$. Then the following statements are satisfied:
\(i) Assume $\nu(\{x_0\}) \geq 1/2$ (resp., $\nu(\{x_0\}) \leq -1/2$, $|\nu(\{x_0\})| \geq 1$). Then it holds $$\begin{split}
&X_t \geq x_0, \ t \geq 0, \ \mathbf{P}\text{-a.s.} \phantom{\Bigr)}\\
\Bigl(\text{resp., } &X_t \leq x_0, \ t \geq 0, \ \mathbf{P}\text{-a.s.}, \\
&X_t \leq x_0, \ t \geq 0, \ \mathbf{P}\text{-a.s.} \text{ if } \nu(\{x_0\}) \leq -1 \text{ and }
X_t \geq x_0, \ t \geq 0, \ \mathbf{P}\text{-a.s.} \text{ if } \nu(\{x_0\}) \geq 1 \Bigr)
\end{split}$$ i.e., the point $x_0$ is reflecting.
\(ii) Assume $\nu(\{x_0\}) > 1/2$ (resp., $\nu(\{x_0\}) < -1/2$, $|\nu(\{x_0\})| > 1$). Then it holds $$X_t = x_0, \qquad t \geq 0, \ \mathbf{P}\text{-a.s.},$$ i.e., the point $x_0$ is absorbing. In particular, $b(x_0) = 0$ must be fulfilled.
**1)** Let $(X,\mathbb{F})$ be a solution of Eq. (\[eqn:SDE\_mvd\]) started at $x_0 \in \mathbb{R}$. At first we reduce the problem to the case $x_0 = 0$. For the process $(X - x_0, \mathbb{F})$ it clearly holds $$\begin{split}
X_t -x_0 &= \int_0^t b_{x_0} (X_s - x_0) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} L^{X-x_0}(t,y-x_0) \, \nu({\mbox{\upshape d}}y), \qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.},
\end{split}$$ where $b_{x_0}(x) := b(x+x_0)$, $x \in \mathbb{R}$, and we have used the relation $L^X(t,x) = L^{X-x_0}(t,x-x_0)$, $t<S_\infty^X$, $x \in \mathbb{R}$, $\mathbf{P}$-a.s. which can be easily deduced for example by exploiting (\[eqn:loc\_time\_as\_limit\]). Introducing the drift measure $\nu_{x_0}$ via $\nu_{x_0}(B) := \nu(B+x_0)$,[^2] $B\in\mathscr{B}([-N,N])$, $N \in \mathbb{N}$, we obtain $$\begin{split}
X_t -x_0 &= \int_0^t b_{x_0} (X_s - x_0) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} L^{X-x_0}(t,y) \, \nu_{x_0}({\mbox{\upshape d}}y), \qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}
\end{split}$$ Hence, $(X-x_0,\mathbb{F})$ is also a solution to an equation of type (\[eqn:SDE\_mvd\]) but started at zero and the drift measure satisfies $\nu_{x_0}(\{0\}) = \nu(\{x_0\})$. Therefore, without loss of generality we assume $x_0 = 0$ in the following parts of the proof.\
**2)** Now let us consider the case of the right local time in Eq. (\[eqn:SDE\_mvd\]). To prove (i) we assume $\nu(\{0\}) \geq 1/2$, we set $Z_t = X_t \wedge 0$, $t \geq 0$. We apply Tanaka’s formula (see (\[eqn:gen\_ito\_formula\]) for $f(x) = -x^- = x \wedge 0$, $x \in \mathbb{R}$) for the left local time to obtain $$Z_t = \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(X_s) \, b(X_s) \, {\mbox{\upshape d}}B_s
+ \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(X_s) \int_\mathbb{R} L_+^X({\mbox{\upshape d}}s,y) \, \nu({\mbox{\upshape d}}y) - \frac{1}{2} L_-^X(t,0),$$ $t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$ Lemma \[lemma:loc\_time\_zero\] shows that the left local time of $X$ in zero vanishes and we can write $$\begin{split}
Z_t &= \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(X_s) \, L_+^X({\mbox{\upshape d}}s,y) \, \nu({\mbox{\upshape d}}y) \\
&= \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} L_+^Z(t,y)\, {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(y) \, \nu({\mbox{\upshape d}}y), \qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.},
\end{split}$$ where we used (\[eqn:int\_wrt\_loc\_time\]) and the easy to check relation $L_+^X(t,y) = L_+^Z(t,y)$, $t < S_\infty^X$, $y < 0$, $\mathbf{P}$-a.s. (use e.g. (\[eqn:loc\_time\_as\_limit\])). Introducing the sets $$A_1 := \{y \in (-\infty,0) : \nu(\{y\}) \geq 1/2\}, \quad A_2 := \{y \in (-\infty,0) : \nu(\{y\}) = 1/2\}$$ and the set function $\mu({\mbox{\upshape d}}y) := {\text{\large{$\mathds{1}$}}}_{(-\infty,0) \setminus A_1} (y) \, \nu ({\mbox{\upshape d}}y)$, in the decomposition of $Z$ we can split the last integral and, using Lemma \[lemma:loc\_time\_zero\], we can write $$Z_t = \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} L_+^{Z}(t,y) \, \mu({\mbox{\upshape d}}y)
+ \int_\mathbb{R} L_+^{Z}(t,y) \, {\text{\large{$\mathds{1}$}}}_{A_2}(y) \, \nu({\mbox{\upshape d}}y),$$ $t < S_\infty^X$, $\mathbf{P}$-a.s. Noting that $\mu(\{x\}) < 1/2$, $x \in \mathbb{R}$, we define the strictly positive function $g$ as the unique solution of the integral equation (\[eqn:integral\_eqn\]) with respect to $\mu$ and denote by $G(x) = \int_0^x g(y)\, {\mbox{\upshape d}}y$, $x \in \overline{\mathbb{R}}$, its strictly increasing and continuous primitive. Then, with the notation $$M_t := \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s, \qquad t < S_\infty^X\,,$$ and noting that $G$ restricted to $\mathbb{R}$ is the difference of convex functions, due to the generalized Itô formula for the right local time and since ${\mbox{\upshape d}}g (y) = -2\, g(y-) \, \mu({\mbox{\upshape d}}y)$, for $Y = G(Z)$ it holds $$\label{eqn:spacetrans_good_points}
\begin{split}
Y_t &= \int_0^t g(Z_s) \, {\mbox{\upshape d}}M_s
+ \int_0^t g(Z_s-) \int_\mathbb{R} L_+^{Z}({\mbox{\upshape d}}s,y) \, \mu({\mbox{\upshape d}}y) \\
&\phantom{=======}+ \int_0^t g(Z_s-) \int_\mathbb{R} L_+^{Z}({\mbox{\upshape d}}s,y) \, {\text{\large{$\mathds{1}$}}}_{A_2}(y) \, \nu({\mbox{\upshape d}}y)
- \int_\mathbb{R} L_+^{Z}(t,y) \, g(y-) \, \mu({\mbox{\upshape d}}y)\\
&= \int_0^t g(Z_s) \, {\mbox{\upshape d}}M_s
+ \int_\mathbb{R} L_+^{Z}(t,y) \, g(y-) \, {\text{\large{$\mathds{1}$}}}_{A_2}(y) \, \nu({\mbox{\upshape d}}y),
\end{split}$$ $t < S_\infty^X$, $\mathbf{P}$-a.s. Clearly, since we have $Z_t \leq 0$, $t \geq 0$, and since $G$ maps $(-\infty,0]$ into $(-\infty,0]$, it follows $Y_t \leq 0$, $t \geq 0$. Using that $g$ is strictly positive, we conclude $$\label{eqn:aux_2}
\int_0^t g(Z_s) \, {\mbox{\upshape d}}M_s \leq Y_t \leq 0, \qquad t < S_\infty^X\, \ \mathbf{P}\text{-a.s.}$$ The process $\int_0^\cdot g(Z_s) \, {\mbox{\upshape d}}M_s$ being a non-positive continuous local martingale up to $S_\infty^X$ starting at zero must be zero $\mathbf{P}$-a.s. which implies $\int_0^t g^2(Z_s)\, {\mbox{\upshape d}}{ \ensuremath{\langle M \rangle}}_s = 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s., and thus $M_t = 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s. Hence, $Z$ is a process of locally bounded variation on $[0,S_\infty^X)$ and we obtain $$L_\pm^{Z}(t,y) = 0, \qquad t < S_\infty^X, \ y \in \mathbb{R}, \ \mathbf{P}\text{-a.s.}$$ Therefore, it holds $Z_t = 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s., which means $X_t \geq 0$, $t \geq 0$, $\mathbf{P}$-a.s. and (i) is proven.\
**3)** Now, still considering the case of the right local time in Eq. (\[eqn:SDE\_mvd\]), we show (ii) which is why we assume $\nu(\{0\}) > 1/2$. From (i) we derive $X_t \geq 0$, $t \geq 0$, $\mathbf{P}$-a.s., and hence via (\[eqn:loc\_time\_zero\_outside\_compact\_interval\]) $$X_t = \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} {\text{\large{$\mathds{1}$}}}_{[0,+\infty)}(y) \, L_+^X(t,y) \, \nu({\mbox{\upshape d}}y),
\qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ Setting $$A_1 := \{y \in [0,+\infty) : \nu(\{y\}) \geq 1/2\}, \quad A_2 := \{y \in [0,+\infty) : \nu(\{y\}) = 1/2\}$$ and defining the set function $\mu({\mbox{\upshape d}}y) := {\text{\large{$\mathds{1}$}}}_{[0,+\infty) \setminus A_1} (y) \, \nu ({\mbox{\upshape d}}y)$, we can write $$X_t = \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} L_+^X(t,y) \, \mu({\mbox{\upshape d}}y)
+ \int_\mathbb{R} {\text{\large{$\mathds{1}$}}}_{A_2}(y) \, L_+^X(t,y) \, \nu({\mbox{\upshape d}}y),$$ $t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$, since the right local time $L_+^X$ of $X$ vanishes on $A_1 \setminus A_2$ by Lemma \[lemma:loc\_time\_zero\]. Note that $\mu$ satisfies (\[eqn:cond\_atoms\]). Let $g$ be the unique solution of (\[eqn:integral\_eqn\]) with respect to $\mu$ and $G(x) := \int_0^x g(y)\, {\mbox{\upshape d}}y$, $x \in \overline{\mathbb{R}}$, its strictly increasing and continuous primitive. Using similar arguments as in (\[eqn:spacetrans\_good\_points\]), for $Y := G(X)$ we obtain $$Y_t = \int_0^t g(X_s) \, b(X_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} g(y-) \, L_+^X(t,y) \, {\text{\large{$\mathds{1}$}}}_{A_2}(y) \, \nu({\mbox{\upshape d}}y), \qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ Having in mind that $0 \notin A_2$ and that $\nu$ is a finite signed measure on $\mathscr{B}([-1,1])$, we conclude[^3] $c := \inf A_2 > 0$. Therefore, $\tau_c := \inf\{t \geq 0 : X_t = c\}$ is a strictly positive $\mathbb{F}$-stopping time. For the stopped process $Y_t^{\tau_c} := Y_{\tau_c \wedge t}$, $t \geq 0$, due to (\[eqn:loc\_time\_zero\_outside\_compact\_interval\]) it holds $$Y_t^{\tau_c} = \int_0^{\tau_c \wedge t} g(X_s) \, b(X_s) \, {\mbox{\upshape d}}B_s,
\qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ Since $G$ maps $[0,+\infty)$ into $[0,+\infty)$, the last relation means that $Y^{\tau_c}$ is a non-negative continuous local martingale up to $S_\infty^X$ started at zero, which implies immediately $Y_t^{\tau_c} = 0$, $t \geq 0$, $\mathbf{P}$-a.s. Hence, because of $\tau_c = \inf\{t \geq 0: Y_t = G(c)\}$ and $G(c) > 0$, we conclude $\tau_c = +\infty$ $\mathbf{P}$-a.s. Finally, we obtain $X_t = 0$, $t \geq 0$, $\mathbf{P}$-a.s. Moreover, we have $$0 = { \ensuremath{\langle X \rangle}}_t = \int_0^t b^2(X_s) \, {\mbox{\upshape d}}s = b^2(0) \, t, \qquad t \geq 0, \ \mathbf{P}\text{-a.s.},$$ which is only possible if $b(0)=0$.\
**4)** Now we prove (i) and (ii) for the case of the left local time in Eq. (\[eqn:SDE\_mvd\]). We recall that we only need to consider the initial value $x_0 = 0$. So, we treat the equation $$X_t = \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} L_-^X(t,y) \, \nu({\mbox{\upshape d}}y),$$ where we assume $\nu(\{0\}) \leq -1/2$. Let $(X,\mathbb{F})$ be a solution of this equation. Introducing $\widetilde{b}(x) = b(-x)$, $x \in \mathbb{R}$, and the Wiener process $\widetilde{B} = -B$, for $-X$ it holds $$-X_t = \int_0^t \widetilde{b}(-X_s) \, {\mbox{\upshape d}}\widetilde{B}_s - \int_\mathbb{R} L_-^X(t,y)\,\nu({\mbox{\upshape d}}y),
\qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ We define $\widetilde{\nu}(A) := -\nu(-A)$,[^4] $A \in \mathscr{B}([-N,N])$, $N \in \mathbb{N}$. Then, since $S_\infty^X = S_\infty^{-X}$ and since $L_-^X(t,y) = L_+^{-X}(t,-y)$, $t < S_\infty^X$, $y \in \mathbb{R}$, $\mathbf{P}$-a.s., which follows immediately from (\[eqn:loc\_time\_as\_limit\]), we get $$-X_t = \int_0^t \widetilde{b}(-X_s) \, {\mbox{\upshape d}}\widetilde{B}_s + \int_\mathbb{R} L_+^{-X}(t,y)\,\widetilde{\nu}({\mbox{\upshape d}}y),
\qquad t < S_\infty^{-X}, \ \mathbf{P}\text{-a.s.}$$ Hence, $(-X, \mathbb{F})$ is a solution of an equation of type (\[eqn:SDE\_mvd\]) with right local time started at zero and for the drift measure $\widetilde{\nu}$ it holds $\widetilde{\nu}(\{0\}) = - \nu(\{0\}) \geq 1/2$. Using step 2) and 3) of the proof yields $X_t \leq 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s. if $\nu(\{0\}) \leq 1/2$ and $X_t = 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s. as well as $b(0)=0$ if $\nu(\{0\}) < -1/2$.\
**5)** For the proof in case of the symmetric local time in Eq. (\[eqn:SDE\_mvd\]) all tools are already presented above. But one must be careful in adapting them. We provide the idea but the details are left to the reader. Again without loss of generality we assume $x_0 = 0$. Let $(X,\mathbb{F})$ be a solution to $$X_t = \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \int_\mathbb{R} \hat{L}^X(t,y) \, \nu({\mbox{\upshape d}}y).$$ We first assume that $|\nu(\{0\})| > 1$ or $\nu(\{0\}) = 1$. Similar as above in step 2) for $Z_t = X_t \wedge 0$, $t \geq 0$, we deduce $$Z_t = \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} \hat{L}^Z(t,y) \, {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(y) \, \nu({\mbox{\upshape d}}y), \qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.},$$ Furthermore, with the help of the sets $$\begin{gathered}
A_1 := \{y \in (-\infty,0) : |\nu(\{y\})| \geq 1\}, \quad
A_2 := \{y \in (-\infty,0) : \nu(\{y\}) = 1\}, \\
A_3 := \{y \in (-\infty,0) : \nu(\{y\}) = -1\},\end{gathered}$$ the set function $\mu({\mbox{\upshape d}}y) := {\text{\large{$\mathds{1}$}}}_{(-\infty,0) \setminus A_1} (y) \, \nu ({\mbox{\upshape d}}y)$ and Lemma \[lemma:loc\_time\_zero\], we can write $$Z_t = \int_0^t {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s
+ \int_\mathbb{R} \hat{L}^{Z}(t,y) \, \mu({\mbox{\upshape d}}y)
+ \int_{A_2} L_+^{Z}(t,y) \, \nu({\mbox{\upshape d}}y)
+ \int_{A_3} L_-^{Z}(t,y) \, \nu({\mbox{\upshape d}}y),$$ $t < S_\infty^X$, $\mathbf{P}$-a.s. Using the unique solution $g$ of the integral equation (\[eqn:integral\_eqn\]) in case of the symmetric local time with respect to $\mu$, its primitive $G(x) := \int_0^x g(y)\,{\mbox{\upshape d}}y$, $x \in \overline{\mathbb{R}}$, and setting $Y := G(X)$, analogously as in (\[eqn:spacetrans\_good\_points\]) we get $$Y_t = \int_0^t g(Z_s) \, {\mbox{\upshape d}}M_s + \int_{A_2} L_+^{Z}(t,y) \, g(y) \, \nu({\mbox{\upshape d}}y) + \int_{A_3} L_-^{Z}(t,y) \, g(y) \, \nu({\mbox{\upshape d}}y),
\qquad t < S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ where $M := \int_0^\cdot {\text{\large{$\mathds{1}$}}}_{(-\infty,0)}(Z_s) \, b(Z_s) \, {\mbox{\upshape d}}B_s$. Note that, because of $\mu(\{y\}) = 0$, $y \in A_2 \cup A_3$, the function $g$ is continuous in the points of $A_2 \cup A_3$. Since $\nu$ is a finite signed measure on $\mathscr{B}([-N,N])$, $N\in \mathbb{N}$, we deduce[^5] $c:= \sup A_3 <0$ and hence the $\mathbb{F}$-stopping time $$\tau_c := \inf\{t\geq 0 : Z_t = c\} = \inf\{t\geq 0 : Y_t = G(c)\}$$ is strictly positive. Clearly, due to (\[eqn:loc\_time\_zero\_outside\_compact\_interval\]) we have $$Y_t = \int_0^t g(Z_s) \, {\mbox{\upshape d}}M_s + \int_{A_2} L_+^{Z}(t,y) \, g(y) \, \nu({\mbox{\upshape d}}y) \qquad t < \tau_c \wedge S_\infty^X, \ \mathbf{P}\text{-a.s.}$$ and similarly to (\[eqn:aux\_2\]) and the lines thereafter we conclude $Z_t = 0$, $t < \tau_c \wedge S_\infty^X$, $\mathbf{P}$-a.s. By the definition of $\tau_c$ this gives $Z_t = 0$, $t < S_\infty^X$, $\mathbf{P}$-a.s. and therefore $X_t \geq 0$, $t \geq 0$, $\mathbf{P}$-a.s.\
Now we assume $|\nu(\{0\})| > 1$ or $\nu(\{0\}) = -1$. As shown in step 4), $-X$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with drift measure $\widetilde{\nu}$ defined by $\widetilde{\nu}(A) = - \nu(-A)$, $A \in \mathscr{B}([-N,N])$, $N \in \mathbb{N}$. Hence, from the result just proven we obtain $-X_t \geq 0$, $\geq 0$, $\mathbf{P}$-a.s. Summarizing, this yields $$X_t = 0, \ t \geq 0, \ \mathbf{P}\text{-a.s. and necessarily $b(0)=0$ if $|\nu(\{0\})| > 1$,}$$ $$X_t \geq 0, \ t \geq 0, \ \mathbf{P}\text{-a.s. if $\nu(\{0\}) = 1$,}$$ and $$X_t \leq 0, \ t \geq 0, \ \mathbf{P}\text{-a.s. if $\nu(\{0\}) = -1$.}$$ This finishes the proof of (i) and (ii) for symmetric local time.
Introducing the $\mathbb{F}$-stopping time $\tau_x := \inf\{t \geq 0 : X_t = x\}$ for a solution $(X,\mathbb{F})$ of Eq. (\[eqn:SDE\_mvd\]), we can state more generally the following corollary to Theorem \[theorem:reflection\_and\_absorbing\].
\[cor:reflection\_and\_absorbing\] Let $(X,\mathbb{F})$ be a solution of Eq. (\[eqn:SDE\_mvd\]) with right (resp., left, symmetric) local time and arbitrary initial condition $X_0$. Then the following statements are satisfied:
\(i) If $x \in \mathbb{R}$ is such that $\nu(\{x\}) \geq 1/2$ (resp., $\nu(\{x\}) \leq -1/2$, $|\nu(\{x\})| \geq 1$), then on $\{\tau_x < +\infty\}$ it holds $$\begin{split}
&X_{\tau_x + t} \geq x, \ t \geq 0, \ \mathbf{P}\text{-a.s}\phantom{\Bigr)} \\
\Bigl(\text{resp., } &X_{\tau_x + t} \leq x, \ t \geq 0, \ \mathbf{P}\text{-a.s},\\
&X_{\tau_x + t} \leq x, \ t \geq 0,\ \mathbf{P}\text{-a.s} \text{ if } \nu(\{x\}) \leq -1
\text{ and } X_{\tau_x + t} \geq x, \ t \geq 0, \ \mathbf{P}\text{-a.s} \text{ if } \nu(\{x\}) \geq 1 \Bigr).
\end{split}$$ (ii) If $x \in \mathbb{R}$ is such that $\nu(\{x\}) > 1/2$ (resp., $\nu(\{x\}) < -1/2$, $|\nu(\{x\})| > 1$), then it holds $$X_{\tau_x + t} = x, \qquad t \geq 0, \ \mathbf{P}\text{-a.s. on } \{\tau_x < +\infty\}.$$ In particular, if $\mathbf{P}(\{\tau_x < +\infty\})>0$, then $b(x) = 0$ must be fulfilled.
We only give the idea of the proof. The details are left to the reader. If $\mathbf{P}(\{\tau_x < +\infty\}) > 0$ is satisfied, then we use the trace of the underlying probability space with respect to $\widetilde{\Omega} = \{\tau_x < +\infty\}$. The process $X_{\tau_x + t}$, $t\geq 0$, considered on $\widetilde{\Omega}$, which is adapted to $\widetilde{\mathbb{F}} = (\mathcal{F}_t \cap \widetilde{\Omega})_{t \geq 0}$, is again a solution of Eq. (\[eqn:SDE\_mvd\]) but started at $x$. Hence, we can apply Theorem \[theorem:reflection\_and\_absorbing\] to obtain the statements of the corollary.
Concerning the question of non-existence of solutions of Eq. (\[eqn:SDE\_mvd\]), we see from Theorem \[theorem:reflection\_and\_absorbing\] and its Corollary \[cor:reflection\_and\_absorbing\] that the existence of a point $x \in \mathbb{R}$ with $\nu(\{x\}) > 1/2$ (resp., $\nu(\{x\}) < -1/2$, $|\nu(\{x\})| > 1$), in general, does not imply automatically that there is no solution. In general, there can be solutions which do not reach $x$ or, if $b(x)=0$ is satisfied, which stay in $x$ after reaching this level. But it holds the following
\[cor:non\_existence\_of\_a\_solution\] If $x \in \mathbb{R}$ is such that $\nu(\{x\}) > 1/2$ (resp., $\nu(\{x\}) < -1/2$, $|\nu(\{x\})| > 1$) and $b(x) \neq 0$, then there is no solution $(X,\mathbb{F})$ of Eq. (\[eqn:SDE\_mvd\]) with right (resp., left, symmetric) local time and arbitrary initial condition $X_0$ which satisfies $\mathbf{P}(\{\tau_x < +\infty\}) > 0$. In particular, there is no solution started at $X_0 = x$.
\(i) In the special case of a drift measure $\nu = \beta \, \delta_0$ where $|\beta|=1$ and $\delta_0$ denotes the Dirac measure in zero and the symmetric local time in Eq. (\[eqn:SDE\_mvd\]) the result of Corollary \[cor:reflection\_and\_absorbing\](i) was already presented in [@blei_bessel_2011], Lemma 2.24. Moreover, Corollary \[cor:non\_existence\_of\_a\_solution\] contains [@blei_bessel_2011], Lemma 2.25, which deals with the non-existence of a solution to Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $\nu = \beta \, \delta_0$ such that $|\beta|>1$.
\(ii) In W. Schmidt [@schmidt:1989] one-dimensional stochastic differential equations with reflecting barriers, more precisely, equations of the form $$\begin{aligned}
\label{eqn:SDE_refl_schmidt}
\left\{
\begin{aligned}
\text{(i)} &\phantom{=}\displaystyle X_t = X_0 + \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + L_t - R_t, \\
\text{(ii)} &\phantom{=}X_t \in [r_1,r_2], \\
\text{(iii)} &\phantom{=}\text{$L_t$, $R_t$ are increasing process with $L_0 = R_0 = 0$ and} \\
&\phantom{===}\begin{gathered}
\int_0^t {\text{\large{$\mathds{1}$}}}_{\{r_1\}} (X_s) \, {\mbox{\upshape d}}L_s = L_t, \quad \int_0^t {\text{\large{$\mathds{1}$}}}_{\{r_2\}} (X_s) \, {\mbox{\upshape d}}R_s = R_t, \quad t \geq 0,
\end{gathered}
\end{aligned}
\right.
\end{aligned}$$ where $r_1 < r_2$, were studied. Clearly, $(X,\mathbb{F})$ on $(\Omega,\mathcal{F},\mathbf{P})$ is called a solution of Eq. (\[eqn:SDE\_refl\_schmidt\]) if there exists a Wiener process $(B,\mathbb{F})$ and two processes $(L,\mathbb{F})$ and $(R,\mathbb{F})$ such that (\[eqn:SDE\_refl\_schmidt\]) is satisfied. For a solution $(X,\mathbb{F})$ of Eq. (\[eqn:SDE\_refl\_schmidt\]) it is not difficult to check that $L$ and $R$ are just the symmetric local times of $X$ in $r_1$ and $r_2$, respectively. Our results, especially Corollary \[cor:reflection\_and\_absorbing\](i), now show that Eq. (\[eqn:SDE\_refl\_schmidt\]) even coincides with Eq. (\[eqn:SDE\_mvd\]) for the diffusion coefficient $b$, symmetric local time, drift measure $\nu = \delta_{r_1} - \delta_{r_2}$ and initial condition $X_0 \in [r_1,r_2]$, i.e., with the equation $$\label{eqn:sde_refl}
X_t = X_0 + \int_0^t b(X_s) \, {\mbox{\upshape d}}B_s + \hat{L}^X(t,r_1) - \hat{L}^X(t,r_2), \qquad X_0 \in [r_1,r_2]\,.$$ For Eq. (\[eqn:sde\_refl\]) the condition (\[eqn:SDE\_refl\_schmidt\])(ii) needs not be specified since it is satisfied for any solution which follows via Corollary \[cor:reflection\_and\_absorbing\](i). Moreover, (\[eqn:SDE\_refl\_schmidt\])(iii) holds because of (\[eqn:int\_wrt\_loc\_time\]).
\(iii) Similar as in the preceding remark the non-negativity condition $X_t \geq 0$, $t \geq 0$, of the solution to the equation $$X_t = X_0 + \int_0^t {\text{\large{$\mathds{1}$}}}_{(0,+\infty)}(X_s) \, {\mbox{\upshape d}}B_s + a \, \int_0^t {\text{\large{$\mathds{1}$}}}_{\{0\}} (X_s) \, {\mbox{\upshape d}}s,$$ with $X_0 \geq 0$ and $a \geq 0$, which is studied in R. Chitashvili [@chitashvili], can be dropped. More detailed, for a solution $(X,\mathbb{F})$ of this equation by (\[eqn:loc\_time\_as\_limit\]) it follows $L_-^X(t,0) = 0$, $t \geq 0$, $\mathbf{P}$-a.s. Combined with (\[eqn:loc\_time\_and\_variation\_process\]) this implies $$\frac{1}{2}\, L_+^X(t,0) = a \, \int_0^t {\text{\large{$\mathds{1}$}}}_{\{0\}} (X_s) \, {\mbox{\upshape d}}s, \qquad t \geq 0, \ \mathbf{P}\text{-a.s.}$$ Hence, $(X,\mathbb{F})$ also solves Eq. (\[eqn:SDE\_mvd\]) with diffusion coefficient $b={\text{\large{$\mathds{1}$}}}_{(0,+\infty)}$, right local time and drift measure $\nu = 1/2 \,\delta_0$. Therefore, we can conclude $X_t \geq 0$, $t \geq 0$, $\mathbf{P}$-a.s. if $X_0 \geq 0$. [$\Diamond$]{}
With the help of Lemma \[lemma:loc\_time\_zero\] we can also give a relation between the different versions of Eq. (\[eqn:SDE\_mvd\]). More precisely, we give a relation between Eq. (\[eqn:SDE\_mvd\]) using the right local time and Eq. (\[eqn:SDE\_mvd\]) with symmetric local time. With the following proposition we complement [@bass_chen], Theorem 2.2(a).
\[prop:equiv\_mvd\_right\_and\_sym\_loc\_time\] (i) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with right local time and drift measure $\nu$ if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $$\hat{\nu}({\mbox{\upshape d}}y) = \left( \frac{1}{2-2\nu(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \nu(\{z\}) < 1\}}(y)
+ {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \nu(\{z\}) \geq 1\}}(y)\right) 2\nu({\mbox{\upshape d}}y)\,.$$ (ii) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $\hat{\nu}$ satisfying $\hat{\nu}(\{x\}) \neq -1$, $x \in \mathbb{R}$, if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with right local time and drift measure $$\nu({\mbox{\upshape d}}y) = \left( \frac{2}{1+\hat{\nu}(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \hat{\nu}(\{z\}) > -1\}}(y)
- {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \hat{\nu}(\{z\}) < -1\}}(y)\right) \frac{1}{2}\hat{\nu}({\mbox{\upshape d}}y)\,.$$
**1)** Let $(X,\mathbb{F})$ be a solution of Eq. (\[eqn:SDE\_mvd\]) with right local time, then by Lemma \[lemma:loc\_time\_zero\] it is $$\hat{L}^X(t,y) = (L^X_+(t,y) + L_-^X(t,y))/2 = 0, \qquad t < S_\infty^X, \ y \in \{\nu> 1/2\}, \ \mathbf{P}\text{-a.s.}$$ and $$\hat{L}^X(t,y) = \frac{1}{2} \, L^X_+(t,y) = (1-\nu(\{y\})) \, L_+^X(t,y), \qquad t < S_\infty^X, \ y \in \{\nu = 1/2\}, \ \mathbf{P}\text{-a.s.}$$ Moreover, via (\[eqn:loc\_time\_and\_variation\_process\]) we see $$L_+^X(t,y) - L_-^X(t,y) = 2 \, L_+^X(t,y) \, \nu(\{y\}), \qquad t < S_\infty^X, \ y \in \{\nu < 1/2\}, \ \mathbf{P}\text{-a.s.}$$ and we can conclude $$\hat{L}^X(t,y) = \left(1-\nu(\{y\})\right) \, L_+^X(t,y), \qquad t < S_\infty^X, \ y \in \{\nu < 1/2\}, \ \mathbf{P}\text{-a.s.}$$ Summarizing, we obtain that $(X,\mathbb{F})$ fulfils Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $\hat{\nu}$ as given under (i).\
**2)** If $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $\hat{\nu}$, then Lemma \[lemma:loc\_time\_zero\] implies $$L_{\pm}^X(t,y) = \hat{L}^X(t,y) = 0, \qquad t < S_\infty^X, \ y \in \{|\hat{\nu}| > 1\}, \ \mathbf{P}\text{-a.s.},$$ and $$L_-^X(t,y) = 0, \qquad t < S_\infty^X, \ y \in \{\hat{\nu} = 1\}, \ \mathbf{P}\text{-a.s.}$$ Furthermore, via (\[eqn:loc\_time\_and\_variation\_process\]) we obtain $$L_+^X(t,y) - L_-^X(t,y) = 2 \, \hat{L}^X(t,y) \, \hat{\nu}(\{y\}), \qquad t < S_\infty^X, \ y \in \mathbb{R}, \ \mathbf{P}\text{-a.s.},$$ and hence $$L_+^X(t,y) = (1 + \hat{\nu}(\{y\})) \, \hat{L}(t,y), \qquad t < S_\infty^X, \ y \in \{|\hat{\nu}| < 1\}, \ \mathbf{P}\text{-a.s.}$$ These observations mean, if we additionally assume $\hat{\nu}(\{x\}) \neq -1$, $x \in \mathbb{R}$, then $(X,\mathbb{F})$ satisfies Eq. (\[eqn:SDE\_mvd\]) with right local time and drift measure $\nu$ as given in (ii).\
**3)** Note that $\hat{\nu}$ as defined in (i) satisfies $\hat{\nu}(\{x\}) > -1$, $x \in \mathbb{R}$, and it holds $\hat{\nu}(\{x\}) > 1$ if and only if $\nu(\{x\}) > 1/2$. Hence, the remaining part of assertion (i) follows by an application of step 2) of the proof. To prove the sufficiency in (ii), we remark that $\nu$ introduced in (ii) satisfies $\nu(\{x\}) > 1/2$ if and only if $|\hat{\nu}(\{x\})| > 1$ and we can conclude using step 1) above.
\(i) The drift measures $\hat{\nu}$ and $\nu$ as defined in Proposition \[prop:equiv\_mvd\_right\_and\_sym\_loc\_time\] (i) and (ii), respectively, are of course not unique.
\(ii) Proposition \[prop:equiv\_mvd\_right\_and\_sym\_loc\_time\](ii) gives an alternative to conclude the results of Theorem \[theorem:reflection\_and\_absorbing\] and its Corollary \[cor:reflection\_and\_absorbing\] and \[cor:non\_existence\_of\_a\_solution\] for Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and a drift measure satisfying $\nu(\{x\}) \neq -1$, $x \in \mathbb{R}$, from the results for the right local time.
\(iii) The case $\hat{\nu}(\{x\}) = -1$ for some $x \in \mathbb{R}$ is excluded since this condition is responsible for reflection to the left, but this cannot occur for solutions of Eq. (\[eqn:SDE\_mvd\]) where the right local time is chosen: By Lemma \[lemma:loc\_time\_zero\] we then have $L_+^X(t,x) = 0$. [$\Diamond$]{}
Similar conclusions as in Proposition \[prop:equiv\_mvd\_right\_and\_sym\_loc\_time\] can be made when the left local time is involved. For the sake of completeness we state the corresponding results which can be proven by analogous arguments as before.
\(i) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with right local time and drift measure $\nu$ satisfying $\nu(\{x\}) \neq 1/2$, $x \in \mathbb{R}$, if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with left local time and drift measure $$\overline{\nu}({\mbox{\upshape d}}y) = \left( \frac{1}{1-2\nu(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \nu(\{z\}) < 1/2\}}(y)
- {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \nu(\{z\}) > 1/2\}}(y)\right) \nu({\mbox{\upshape d}}y)\,.$$ (ii) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with left local time and drift measure $\overline{\nu}$ satisfying $\overline{\nu}(\{x\}) \neq -1/2$, $x \in \mathbb{R}$, if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with right local time and drift measure $$\nu({\mbox{\upshape d}}y) = \left( \frac{1}{1+2\overline{\nu}(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \overline{\nu}(\{z\}) > -1/2\}}(y)
+ {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \overline{\nu}(\{z\}) < -1/2\}}(y)\right) \overline{\nu}({\mbox{\upshape d}}y)\,.$$ (iii) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with left local time and drift measure $\overline{\nu}$ if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $$\hat{\nu}({\mbox{\upshape d}}y) = \left( \frac{1}{2+2\overline{\nu}(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \overline{\nu}(\{z\}) > -1\}}(y)
+ {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \overline{\nu}(\{z\}) \leq -1\}}(y)\right) 2\overline{\nu}({\mbox{\upshape d}}y)\,.$$ (iv) $(X,\mathbb{F})$ is a solution of Eq. (\[eqn:SDE\_mvd\]) with symmetric local time and drift measure $\hat{\nu}$ satisfying $\hat{\nu}(\{x\}) \neq 1$, $x \in \mathbb{R}$, if and only if it is a solution of Eq. (\[eqn:SDE\_mvd\]) with left local time and drift measure $$\overline{\nu}({\mbox{\upshape d}}y) = \left( \frac{2}{1-\hat{\nu}(\{y\})}\, {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \hat{\nu}(\{z\}) < 1\}}(y)
- {\text{\large{$\mathds{1}$}}}_{\{z\in\mathbb{R} : \, \hat{\nu}(\{z\}) > 1\}}(y)\right) \frac{1}{2}\hat{\nu}({\mbox{\upshape d}}y)\,.$$
[^1]: Work supported in part by the European Community’s FP 7 Programme under contract PITN-GA-2008-213841, Marie Curie ITN “Controlled Systems”.
[^2]: For $B \subseteq \mathbb{R}$, we set $B+x_0 := \{x+x_0 :\; x \in B\}$.
[^3]: $\inf \emptyset := + \infty$.
[^4]: For $A \subseteq \mathbb{R}$, we set $-A := \{-x : x \in A\}$.
[^5]: $\sup \emptyset := -\infty$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We simulate the time-dependent coherent dynamics of a spatially indirect exciton -an electron-hole pair with the two particles confined in different layers- in a GaAs coupled quantum well system. We use a unitary wave-packet propagation method taking into account in full the four degrees of freedom of the two particles in a two-dimensional system, including both the long-range Coulomb attraction and arbitrary two-dimensional electrostatic potentials affecting the electron and/or the hole separately. The method has been implemented for massively parallel architectures to cope with the huge numerical problem, showing good scaling properties and allowing evolution for tens of picoseconds. We have investigated both transient time phenomena and asymptotic time transmission and reflection coefficients for potential profiles consisting of i) extended barriers and wells and ii) a single-slit geometry. We found clear signatures of the internal two-body dynamics, with transient phenomena in the picosecond time-scale which might be revealed by optical spectroscopy. Exact results have been compared with mean-field approaches which, neglecting dynamical correlations by construction, turn out to be inadequate to describe the electron-hole pair evolution in realistic experimental conditions.'
author:
- Federico Grasselli
- Andrea Bertoni
- Guido Goldoni
title: 'Exact two-body quantum dynamics of an electron-hole pair in semiconductor coupled quantum wells: a time-dependent approach'
---
Introduction
============
Electronic bilayer systems have exposed a huge amount of new physics driven by inter-layer Coulomb interactions. A few examples are fractional quantum Hall effect states in semiconductor[@Eisenstein1992; @Suen1992] and graphene bilayers,[@Novoselov2006] phase transitions in Quantum Hall ferromagnets,[@Piazza1999] complex Wigner crystal ordering[@Goldoni1997]. Electron-hole bilayers gained importance in their own right. Excitons -bound electron-hole quasi-particles with a bosonic character- have long being predicted to undergo quantum condensation.[@Bryant_PRB93; @Butov_NatureA02] Recently, signatures of condensation have been found in systems of spatially indirect excitons (IXs), electron-hole pairs optically excited in semiconductor coupled quantum well systems, with the two charges confined in different layers by a static electric field.[@Butov_JPCM04; @High_NATURE12; @Alloing_EPL14]
In a different perspective, IXs are at the heart of a new class of opto-electronic devices, made possible thanks to the small electron-hole overlap which extends their intrinsic lifetime from nanoseconds[@Colocci_EPL90] to microseconds.[@Butov_JPCM04; @Gartner_APL06] Indeed, although IXs are neutral excitations, they carry a large finite electric dipole which can be used to drive the evolution of IXs in the coupled quantum well (CQW) planes by electric field gradients generated, e.g., by metallic gates. Finally, since a bias normal to the QW-plane can control the overlap of the pair, IX recombination can be induced at arbitrary time, thereby ’measuring’ the result of the evolution. IX gases have been exploited to demonstrate several functionalities, such as fast data storage,[@Winbow_NL07; @Winbow_JAP08] acceleration with electrostatic ramps[@Gartner_APL06; @Leonard_APL12] and interdigital devices,[@Winbow_PRL11] field effect transistors.[@High_Science11072008] Furthermore, trapping of single IXs has been recently demonstrated,[@Schinner_PRL13] opening the way to single IX electronics. This requires the development of theoretical concepts to describe the evolution of IX wave-packets in complex electrostatic fields.
Scattering of a composite quantum objects with internal degrees of freedom[@SaitoJPCM94] (DoFs) like an IX in the presence of an electrostatic field gradient,[@SaitoPRB95] is an important topic in its own, with applications in molecular and nuclear physics (see, e.g., Refs. , and references therein). Indeed, in the presence of a scattering potential, energy can be transferred between the center-of-mass (CM) kinetic energy and internal excitations, which may strongly influence the transmission and reflection probabilities. However, due to the difficulty to evolve the quantum equations of motion for several DoFs with open boundaries, exact calculations are often limited to idealized situations, such as collinear scattering, purely one-dimensional (1D) systems, and/or very simple potential profiles.[@SaitoJPCM94; @AhsanPRC10; @LugovskoyPRA13; @HnybidaPRA08; @PenkovJETP00; @KavkaPRA10] Indeed, the numerical approach scales exponentially with the number of DoFs. For realistic situations, such as the 3D problem of colliding molecules with complex inter-molecular interactions, mean-field methods have been applied.[@BeckTDH2000] IXs in CQWs are an important system from this point of view, since in principle their evolution could be probed by accurate time-dependent optical means, not only at asymptotic times, but also during the scattering event. Recently, we used an idealized 1D model to study the evolution of IX wave-packets under the action of external fields, taking explicitly into account the internal structure of the electron-hole pair.[@GrasselliJCP15] Our model allowed to investigate different regimes/potential profiles where inter-particle Coulomb correlations may lead to internal excitations or even dissociation, as a result of scattering. However, such 1D calculations are too simplistic to be applied to realistic CQW quasi-two-dimensional (2D) systems, where electrons and holes evolve in a complex 3D structure.
In this paper we report time-dependent simulations of the coherent dynamics of a single IX wave-packet in a semiconductor CQW structure with complex in-plane electrostatic potentials, taking into account the 3D structure of realistic devices through a 2D+1D effective model. We used a unitary wave-packet propagation which includes the long-range electron-hole Coulomb interaction and arbitrary scattering potentials acting on the electron and the hole separately. The large numerical problem to simulate exactly the present four-DoF system has been tackled by the Fourier split-step method implemented on a massively parallel architecture. This allowed propagation for tens of picoseconds in typical potential landscapes. We have investigated potential profiles consisting of extended barriers/wells, and a single slit geometry, finding genuine signatures of the two-body dynamics in realistic experimental conditions, with transient phenomena in the picosecond time-scale. Our results could be directly compared with time resolved optical spectroscopy. We have compared the unitary evolution results with a mean-field approach at different levels of approximation, the so-called rigid exciton (RIX) model and the time-dependent Hartree (TDH) method. The comparison shows that a mean-field approach is in general inadequate to describe the electron-hole pair evolution in realistic samples, thereby showing that IX dynamics in CQW might be a particularly interesting system to investigate correlation effects and to test theoretical modeling.
In Sec. \[sec:theory\] we define our Hamiltonian description of an IX in a typical semiconductor CQW (\[sec:model\]) and we provide details on the full (\[sec:fullpropagation\]) and mean-field (\[sec:meanfieldpropagation\]) wave-packet propagation methods. Initial conditions are discussed in Sec. \[InitialState\]. In Sec. \[sec:Results\] we calculate the free-exciton properties of our model (\[sec:FreeIX\]), while numerical details of wave-packet propagation are discussed in Sec. \[sec:WavePacketPropagation\]. Results are summarized for scattering potentials which are weak (Sec. \[sec:WeakPotential\]) or strong (in Sec. \[sec:StrongPotential\]) with respect to internal excitations, and for a single-slit geometry (Sec. \[sec:SingleSlit\]). Section \[sec:Conclusions\] discusses, in particular, the predictivity of the different approaches. A formal derivation of the mean-field propagation scheme is provided in the Supplemental Material.[@SupplMat_IX]
Theoretical approach {#sec:theory}
====================
The electron-hole Hamiltonian {#sec:model}
-----------------------------
Our reference system is sketched in Fig. \[zaxiswf\](a). A symmetric GaAs CQW structure grown along $z$ is embedded in a Al$_x$Ga$_{1-x}$As matrix. A *vertical* electric field $F_z$ along the growth direction separates electrons and holes in different layers.[@AndreakouPRB15] While typical CQW confinement energies amount from tens to hundreds of meV, *in-plane* $(xy)$ potential landscapes generated by metallic gates, as well as kinetic energies which can be impressed upon IXs, are in the meV range. Therefore, we factorize the in-plane and vertical component of the IX 3D wave function (here and throughout we use wave function in place of envelope function) as $$\Psi(\mathbf{r}_e,\mathbf{r}_h) \zeta_{e} (z_e) \zeta_{h} (z_h)\, ,\label{eq1_psitot}$$ where $\mathbf{r}_i$ and $z_i$ are the 2D $xy$ coordinate and $z$ coordinate of the two particles, respectively. $ \zeta_{e} (z_e)$ and $ \zeta_{h} (z_h)$ are calculated explicitly for a given structure and field. $\zeta_{e}$ is calculated from the 1D effective-mass equation $$H_{z_e} \zeta_e(z_e) = E_{z_e} \zeta_e(z_e)
\label{eveqz}$$ with the Hamiltonian $$H_{z_e} = -\frac{\hbar^2}{2} \frac{d}{d z_e}\left(\frac{1}{m_e^z(z_e)} \frac{d}{d z_e}\right) + W_e(z_e) - e z_e F \, , \label{Hamz}$$ where $W_e(z_e)$ and $m_e^z(z_e)$ are the band edge and the material-dependent effective mass, respectively, of conduction electrons in the CQW structure. No spin-dependent term is considered. Indeed, the validity of Eqs. (\[eq1\_psitot\]) and (\[Hamz\]) is limited to samples with narrow CQWs which are the typical heterostructures used in the experiments we are addressing to[@Gartner_APL06; @Winbow_NL07; @Winbow_JAP08; @Leonard_APL12; @Winbow_PRL11; @High_Science11072008; @Schinner_PRL13]. Here, in-plane Coulomb binding energy and scattering potentials are in the few meV range, while vertical confinement energies is at least one order of magnitude larger.
Equation (\[eveqz\]) with the Hamiltonian (\[Hamz\]) is called a Ben-Daniel Duke problem.[@Bastard_BOOK88] $E_{z_e}$ and $ \zeta_e(z_e)$ are obtained from (\[eveqz\]) on a homogeneous real-space grid by a finite difference approach, taking into account the material-dependent effective mass. $E_{z_h}$ and $\zeta_{h}$ are computed similarly, with parameters appropriate to the valence band electrons. In this case $m_{h}^z$ is the effective mass given by the heavy-hole diagonal mass tensor, related to the Luttinger’s parameters,[@VurgaftmanJAP01] $\gamma_1$ and $\gamma_2$, by $$\begin{aligned}
m_{h}^z = (\gamma_1 - 2\gamma_2)^{-1}.\end{aligned}$$ An example of these calculations is reported in Fig. \[zaxiswf\](b) showing that the two carriers are well localized in either wells by the external bias. This justifies the separability of the electron and hole wave functions in the $z$ direction assumed in Eq. (\[eq1\_psitot\]).
To simulate the electron-hole quantum dynamics we need to include accurately the mutual interaction. While in a ideal 2D electron system this is described by the bare $r^{-1}$ Coulomb potential, in a typical CQW heterostructure carriers are delocalized in the wells over a length which is comparable to the carrier separation, given by the effective Bohr radius. Therefore, the effective interaction is modified at short range, up to a distance comparable to the well width. To account for this effect, we consider electrons and holes in the ground state of the confinement potential (a reasonable assumption due to the large energy gaps in the growth direction) and we consider the *effective 2D interaction* $U_C$ as the mean value of the Coulomb interaction over the wave functions in the growth direction, $$\begin{split}
U_C(r) &= \langle \zeta_e \zeta_h |U_C^{3D}| \zeta_e \zeta_h \rangle \equiv \iint dz_e dz_h \times \\
& |\zeta_e(z_e)|^2 |\zeta_h(z_h)|^2 \left[ - \frac{e^2}{4\pi\epsilon_0\epsilon_r}\frac{1}{\sqrt{r^2 + (z_e-z_h)^2}} \right] .
\end{split}
\label{eq_U_C}$$ Here, $r$ is the electron-hole in-plane distance and $\epsilon_r$ the relative permittivity of the well material.
Therefore, the full 3D problem has been mapped into an effective 2D model. The two-body wave function $\Psi$ is then propagated in time according to the 2D Hamiltonian $$H = H_0 + U_{\mbox{\scriptsize ext}} \, ,$$ where $$H_0 = -\frac{\hbar^2}{2 m_e} \nabla_{\mathbf{r}_e}^2 -\frac{\hbar^2}{2 m_h^\parallel} \nabla_{\mathbf{r}_h}^2 + U_C(|\mathbf{r}_e-\mathbf{r}_h|)$$ is the *free IX Hamiltonian*. Here, $m_h^\parallel$ is the in-plane component of the strongly anisotropic mass tensor of GaAs, $$m_h^\parallel = (\gamma_1+\gamma_2)^{-1}\, .$$
It is actually numerically convenient and more transparent to work in the CM and relative coordinate system $$\begin{aligned}
\mathbf{R} & = & (m_e \mathbf{r}_e + m_h^\parallel\mathbf{r}_h)/M \,, \\
\mathbf{r} & = & \mathbf{r}_e - \mathbf{r}_h \,,\end{aligned}$$ where $M= m_e + m_h^\parallel$ is the in-plane exciton mass and $m= [m_e^{-1}+(m_h^\parallel)^{-1}]^{-1}$ is the in-plane reduced effective mass. In this representation, the free IX Hamiltonian separates as $$\begin{aligned}
H_{0}(\mathbf{R},\mathbf{r}) & = & H_{\mbox{\scriptsize CM}}(\mathbf{R}) + H_{\mbox{\scriptsize rel}}(\mathbf{r}) \,, \label{H0tot}\\
H_{\mbox{\scriptsize CM}}(\mathbf{R}) & \equiv & -\frac{\hbar^2}{2 M} \nabla_{\mathbf{R}}^2 \,, \\
H_{\mbox{\scriptsize rel}}(\mathbf{r}) & \equiv & -\frac{\hbar^2}{2 m} \nabla_{\mathbf{r}}^2 + U_C(r) \,, \label{H0rel}\end{aligned}$$ and the free in-plane wave function can be factorized as $$\begin{aligned}
\Psi(\mathbf{R},\mathbf{r}) = \chi(\mathbf{R})\phi(\mathbf{r}) \,.
\label{eq:factoriation}\end{aligned}$$ As we shall discuss later, the CM energy is typically of the order of tenths of meV, much less than the internal excitation energy, which is of the order of several meV. Therefore, the grid parameters related to CM and relative dynamics are quite different and can be optimized separately. On the contrary, using electron and hole coordinates requires to use (almost) the same grid, and the same accuracy would be reached at a greater computational cost.[@GrasselliJCP15]
In this representation, the inclusion of one-body external potentials, $U_{\mbox{\scriptsize ext}} = U_e(\mathbf{r}_e) + U_h(\mathbf{r}_h) = U_{\mbox{\scriptsize ext}}(\mathbf{R}, \mathbf{r})$ in the full Hamiltonian couples the CM and relative coordinates, and removes the separability of the wave function, Eq. (\[eq:factoriation\]). Therefore, to propagate $\Psi(\mathbf{R}, \mathbf{r};t)$ we need to deal with a four DoFs propagation scheme. We neglect the interaction of IXs with environmental degrees of freedom, like phonons of the semiconductor lattice. In typical experiments $E_\mathrm{CM}$ is in the tenths of meV range, well below the optical phonon energy in GaAs, while low temperature strongly suppresses acoustic phonon population. In our simulation, the scattering time is in the order of tens of picoseconds, well below the LA-phonon assisted relaxation of IXs in these devices, which is in the nanoseconds range.[@Butov_JPCM04]
Full numerical propagation {#sec:fullpropagation}
--------------------------
The quantum propagation of the IX is obtained through the numerical application of the evolution operator, $\mathcal{U}(t+\Delta_t;t)$, between two consecutive times $t$ and $t+\Delta_t$ as $$\begin{aligned}
\Psi(\mathbf{R},\mathbf{r};t+\Delta_t) = \mathcal{U}(t+\Delta_t;t) \Psi(\mathbf{R},\mathbf{r};t) .\end{aligned}$$ Our numerical solution relies on the Fourier split step (FSS) approach.[@Leforestier_JCP91; @Castro_JCP04] This unitary method, numerically exact as $\Delta_t\rightarrow0$, is based on the Suzuki-Trotter factorization[@Suzuki_PJAB93] of the evolution operator in the product of two exponential operators, containing the kinetic or the potential operators, respectively, each diagonal either in direct or in reciprocal space (see Appendix in Ref. for details): $$\begin{aligned}
\begin{split}
&\mathcal{U}(t + \Delta_t; t) =\\
&= e^{ -\frac{i}{\hbar} U_{\mathrm{tot}} \frac{\Delta_t}{2} }
e^{ -\frac{i}{\hbar} T \Delta_t } e^{ -\frac{i}{\hbar} U_{\mathrm{tot}} \frac{\Delta_t}{2}} + \mathcal{O}(\Delta_t^3).\label{eq:SuzTrot}
\end{split}\end{aligned}$$ where $T = \mathbf{P}^2/(2M) + \mathbf{p}^2/(2m)$ and $U_{\mathrm{tot}} = U_C + U_\mathrm{ext}$ are the total kinetic and potential energy operators of the system ($\mathbf{P}$ and $\mathbf{p}$ are CM and relative in plane linear momenta, respectively). Hence, at each time step the IX wave function must be switched from position to momentum representation, and *vice versa*, through Fourier transformation, $\mathcal{F}$, according to $$\begin{aligned}
\begin{split}
\Psi(\mathbf{R},\mathbf{r}; & t + \Delta_t) =e^{ -\frac{i}{\hbar} U_{\mathrm{tot}} \frac{\Delta_t}{2} }\times \\
& \mathcal{F}^{-1}\left\lbrace
e^{ -\frac{i}{\hbar} T \Delta_t } \mathcal{F} \left[ e^{ -\frac{i}{\hbar} U_{\mathrm{tot}} \frac{\Delta_t}{2}}\Psi(\mathbf{R},\mathbf{r};t) \right]\right\rbrace.\label{eq:FourierSS}
\end{split}\end{aligned}$$ The use of the Fast Fourier Transform (FFT) algorithm, therefore, results in high computational efficiency with respect to other methods, particularly those based on finite difference discretized Hamiltonian, as the Crank-Nicolson method.
The coupling between CM and relative DoFs introduced by the one-body external potentials $U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r})$ requires the numerically full propagation of all four DoFs of the IX for a sufficiently long time for the scattering process to conclude on a sufficiently dense and extended real-space grid. For the energy scales into play here, this turns out to be a demanding numerical task, both to store the complex-valued IX wave function in memory, and to numerically compute the application of the evolution operators to it. A code exploiting massive parallelization has been developed to cope with these issues. The four-dimensional domain is discretized in a grid of about $4295\times10^6$ points. Typically, the computation of a single $40$ fs time step takes $10.5$ seconds, and one Gb of memory per core is used on a 256-core run (with 16 cores in two 2.4 GHz Haswell Xeon processors per shared-memory node), corresponding to a speedup of about 43 with respect to a serial run.
Mean-field propagation {#sec:meanfieldpropagation}
----------------------
In semiconductor physics, when dealing with excitons in weak potentials, it is often justified to apply the so-called *rigid exciton model* (RIX),[@Zimmermann_PAC97] which consists in the assumption that the IX is frozen into its relative motion ground state, $\phi_0(\mathbf{r})$; hence, only the quantum evolution of the CM component, $\chi(\mathbf{R})$, is taken into account. The internal DoFs are integrated out, leading to an effective potential $$\begin{aligned}
U_{\mbox{\scriptsize eff}}(\mathbf{R}) = \iint d\mathbf{r} |\phi_0(\mathbf{r})|^2 U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r}). \label{eqRIXeffpot}\end{aligned}$$ Therefore, the IX moves as a rigid object with coordinates $\mathbf{R}$, its dynamics being determined by the external potential averaged on the relative-motion ground state. For example, IX wave function localization in weak traps can be calculated in this way.[@Hohenester_APL04]
The CM effective evolution operator $$\begin{aligned}
\mathcal{U}_{\mbox{\scriptsize eff}} (t+\Delta_t;t) \equiv \exp\left\lbrace-\frac{i}{\hbar}\left[ \frac{\mathbf{P}^2}{2M} + U_{\mbox{\scriptsize eff}}(\mathbf{R}) \right]\Delta_t \right\rbrace\end{aligned}$$ is then applied to the CM wave function, $\chi(\mathbf{R},t)$, and the numerical evolution can be obtained again by the FSS method. Obviously, the RIX model requires a much lower computational effort with respect to full propagation, since only the two DoFs wave function $\chi(\mathbf{R})$ needs to be propagated, which can be readily obtained on a standard personal computer. Clearly, every effect of the internal dynamics is neglected by the RIX approximation.
The RIX model is the lowest order example of a more general mean-field strategy, also known as the *time dependent Hartree* (TDH) method in atomic and molecular scattering.[@McLachlan64; @BeckTDH2000] Within this approach, at each time $t$ the global wave function of the composite object is assumed to be factorized into a CM and a relative wave function, $$\begin{aligned}
\Psi(\mathbf{R},\mathbf{r};t) = \chi(\mathbf{R};t)\phi(\mathbf{r};t) .\end{aligned}$$ The evolution of $\chi(\mathbf{R};t)$ and $\phi(\mathbf{r};t)$ is determined by an effective potential representing the expectation value of the external potential on the relative and CM wave function, respectively, *at that specific time* $t$, i.e. (see Ref. , and Suppl. Mat.[@SupplMat_IX]) $$\begin{split}
\chi(\mathbf{R};t+\Delta_t) &= \exp \left\lbrace - \frac{i}{\hbar} \int_t^{t+\Delta_t}dt' \times \right.\\
& \left. \left[ \frac{\mathbf{P}^2}{2M} + U_{\mbox{\scriptsize eff}}(\mathbf{R};t') \right] \right\rbrace \chi(\mathbf{R};t)
\end{split} \label{eqTDHCM}$$ and $$\begin{split}
\phi(\mathbf{r};t+\Delta_t) &= \exp \left\lbrace -\frac{i}{\hbar} \int_t^{t+\Delta_t} dt' \times \right.\\
& \left. \left[ \frac{\mathbf{p}^2}{2m} + U_{C}(\mathbf{r}) + u_{\mbox{\scriptsize eff}}(\mathbf{r};t') \right] \right\rbrace \phi(\mathbf{r};t)
\end{split} \label{eqTDHrel}$$ with $$\begin{aligned}
U_{\mbox{\scriptsize eff}}(\mathbf{R};t) \equiv \int d\mathbf{r} |\phi(\mathbf{r};t)|^2 U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r})
\label{effTHDpotss}
\\
u_{\mbox{\scriptsize eff}}(\mathbf{r};t) \equiv \int d\mathbf{R} |\chi(\mathbf{R};t)|^2 U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r}) \label{effTHDpots}\end{aligned}$$
This is a mean-field model, since the evolution of the CM wave function is determined by the average field generated by the relative wave function, and *vice versa*, at each time step. Again, the numerical propagation is performed using the FSS method. The FSS algorithm needs to be applied twice at each time step, independently of the CM and relative wave functions, the two evolutions being coupled through the effective potentials, Eqs. (\[effTHDpotss\]), and (\[effTHDpots\]). Clearly, the RIX model consists in assuming a *rigid* $\phi(\mathbf{r};t) = \phi_0(\mathbf{r})$, and only the $\chi(\mathbf{R};t)$ component needs to be propagated.
Note that, even for a stationary external potential $U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r})$, the mean-field propagation method requires to evolve the two components of the wave function under *time-dependent* potentials. This increases substantially the computational cost of the simulation with respect to the RIX model, since the propagator needs to be calculated at each time step rather than only once. However, the TDH approach is still far less demanding than the full evolution.
Initial state {#InitialState}
-------------
In order to start a time dependent simulation, we need to choose a proper initial state. In typical CQW systems, an IX thermalizes and relaxes to the ground state $\phi_0(\mathbf{r})$ of the free exciton Hamiltonian $H_0$ within nanoseconds from photogeneration, due to active scattering mechanisms (acoustic phonons). We do not consider in our simulations this transient, which is short compared to the IX photo-recombination lifetime.[@Butov_JPCM04] Therefore we initialize the IX wave function as $$\begin{aligned}
\Psi(\mathbf{R},\mathbf{r};t=0) = \chi(\mathbf{R})\phi_0(\mathbf{r}) \,.\end{aligned}$$ The CM wave function is chosen as the minimum uncertainty wave-packet $$\begin{aligned}
\chi(\mathbf{R}) =& \left(\frac{1}{2\pi\sigma_X\sigma_Y}\right)^{1/2} \exp(i\mathbf{K}_0\cdot \mathbf{R}) \times \nonumber \\
& \exp\left[-\frac{(X-X_0)^2}{4\sigma_X^2}\right]\exp\left[-\frac{(Y-Y_0)^2}{4\sigma_Y^2}\right],\end{aligned}$$ centered at the initial CM position $(X_0,Y_0)$, having widths $\sigma_X$ and $\sigma_Y$, and propagating with an average CM wave vector $$\begin{aligned}
\mathbf{K}_0 \equiv \sqrt{\frac{2M E_{\mbox{\scriptsize CM}}}{\hbar^2}}(\sin\theta, \cos\theta) ,\end{aligned}$$ where $E_{\mbox{\scriptsize CM}} $ is the most probable CM kinetic energy at $t=0$ and $\theta$ identifies the initial propagation direction with respect to a properly defined normal incidence direction. In the following simulation we take, as initial dispersion, $\sigma_X=\sigma_Y=80$ nm unless otherwise specified. This value is of the same order of magnitude of the confinement length of IX traps [@Schinner_PRL13]. This is a reasonable compromise between a sufficiently narrow momentum distribution (hence a delocalized exciton) and a spatially localized IX. The CM momentum can be controlled, e.g., through the application of acceleration ramps. Localization of the initial state can be controlled by electrostatic traps. See Refs. for typical parameters, comparable to those used in our simulations. Note that, while the CM energy clearly affects transmission and reflection coefficients, due to the linearity of the equations the momentum distribution does not affect much the dynamics, as long as the momentum dispersion is not too broad.
We remark that, even if the initial state is factorized, in the full propagation the wave function is correlated, evolving under the influence of both the electron-hole interaction and the external potential, with no specific *a priori* decomposition. On the contrary, factorization for the IX wave function is assumed in the RIX and TDH approximations at any intermediate time $t$.
Results {#sec:Results}
=======
In the following we investigate the IX dynamics in a 8nm/4nm/8nm GaAs/Al$_{0.33}$Ga$_{0.67}$As/GaAs CQW system.[@Butov_NatureA02; @Butov_NatureB02; @Butov_JPCM04] Band parameters are indicated in Tab. \[tab:parameters\]. An homogeneous (i.e. constant along the $xy$ planes) electric field $F =1 \mbox{mV/nm}$, generated by the application of a bias voltage between the top and the back gates of the sample, is assumed (see Fig. 1).
GaAs Al$_{0.33}$Ga$_{0.67}$As
-------------------------------- --------------------- --------------------------
$m_e/m_0$ (Ref. ) 0.067 0.094
$\gamma_1,\gamma_2$ (Ref. ) [$6.98, \, 2.06$]{}
Relative permittivity (Ref. ) [12.9]{}
Valence band offset (Ref. )
Conduction band offset (Ref. )
: Band parameters adopted in the simulations.[]{data-label="tab:parameters"}
Free IX: effective Coulomb interaction and relative motion eigenvalue problem {#sec:FreeIX}
-----------------------------------------------------------------------------
In Fig. \[zaxiswf\](b) we show the square modulus of the calculated electron and hole wave function components ($\zeta_{e}$, $\zeta_{e}$), together with the band profile along the growth axis of the heterostructure. Clearly, the electric field localizes the electron and the hole in different layers.
$\zeta_{e}, \zeta_{h}$ are used to compute the effective electron-hole interaction $U_C(r)$ (see Eq. (\[eq\_U\_C\])) along the CQW planes, shown in Fig. \[rel\_Coul\_states\](a). At large electron-hole distances $U_C$ amounts to the bare Coulomb interaction. At small distances, the Coulomb divergence is removed due to the separation of the electron and the hole in different layers. Bound electron-hole states and energy levels $\mathcal{E}_n$ are calculated from $H_{\mbox{\scriptsize rel}}$ with a finite difference approach on a 2D uniform square grid. The grid density is the same as for the wave packet evolution (see Sec. \[sec:WavePacketPropagation\]). The lowest bound states are shown in Fig. \[rel\_Coul\_states\](a). Energies and degeneracies of the lowest states are reported in Tab. \[tab:energies\], together with their symmetries, as deduced from Figure \[rel\_Coul\_states\](b), which shows the wave functions $\phi_n(\mathbf{r})$ of the nine lowest states.
$n$ symmetry degeneracy $\mathcal{E}_n$ (meV)
----- ---------- ------------ -----------------------
0 $s$ 1 -3.63
1 $p$ 2 -1.28
2 $s$ 1 -0.91
3 $d$ 2 -0.53
4 $p$ 2 -0.49
5 $s$ 1 -0.42
: Lowest energy levels of the free IX relative motion Hamiltonian, Eq. (\[H0rel\]).[]{data-label="tab:energies"}
Wave packet propagation {#sec:WavePacketPropagation}
-----------------------
Unless differently indicated, the simulations described below have been performed with the following parameters: a total simulation time of $60\,\mbox{ps}$ and a time step $\Delta_t =$ 40 fs; a spatial CM domain with $\mathbf{R}\equiv(X,Y)\in [-1.5\,\mu\mathrm{m}, 1.5\,\mu\mathrm{m}]^2$ and a grid point density of 0.17 points/nm along both directions; a spatial relative motion domain with $\mathbf{r}\equiv(x,y)\in [-0.15\,\mu\mathrm{m}, 0.15\,\mu\mathrm{m}]^2$ and a grid point density of 0.43 points/nm in both directions.
These parameters satisfy the Nyquist criterion[@NR_2007] for the spatial frequency sampling far beyond the considered energy ranges. Another criterion in order to apply the FSS method with no ambiguity requires that the phase exponent $ U_{\mbox{\scriptsize ext}} \Delta_t/(2\pi\hbar) \ll 1$. If this condition is not fulfilled, the external potential strength is not well defined, because the class of $\mod(2\pi \hbar/\Delta_t)-$defined potentials has more than one element. In this case the method gives the same results with different potentials, which is physically inconsistent.
In order to estimate the IX localization during the time dependent simulations and, in the asymptotic times, the transmission and reflection probabilities, we define three regions of the CM space: (i) the *reflection* region **A**, i.e., the subspace with vanishing external potential, $U_e=U_h=0$, where the IX wave function is initially localized; (ii) the *potential* region **B**; (iii) the *transmission region* **C**, i.e., the subspace with vanishing external potential which can be reached from **A** only by crossing the potential region. Note that these regions refer to the CM DoFs only, while in the four DoF model the external potential $U_{\mbox{\scriptsize ext}}$ depends also on the relative coordinate $\mathbf r$. Thus, in order to compare consistently the different methods, we define the above regions according to the potential of Eq. (\[eqRIXeffpot\]). Specifically, we choose as boundaries of the region **B** the $Y$ positions where the external effective potential drops to 5% of its maximum value. The related coefficients are then defined as integrals, over the corresponding regions
- of the CM-part of the wave function square modulus, $|\chi(\mathbf{R};t)|^2$, for the mean field approximations;
- of the CM marginal probability, $$\label{eq:MarginalProbability}
\rho_{\mbox{\scriptsize CM}}(\mathbf{R};t) \equiv \int d\mathbf{r} |\Psi(\mathbf{R},\mathbf{r};t)|^2,$$ for the full propagation.
These integrals are then normalized to the whole domain, so that the **A**,**B**,**C**-coefficients are defined in the interval $[0,1]$. An example of the three regions is showed in Fig. \[weakpot\]. Clearly, the three coefficients evolve in time. Coefficients **A** and **C** at asymptotic times set to a constant which are the reflection and transmission probabilities, respectively.
Below we shall investigate the time-dependent dynamics of a wave-packet, prepared as described in Sec. \[InitialState\], scattering against two classes of potentials, a uniform, infinitely long barrier or well, and a barrier with a slit, which mimics the potential generated by a split gate, as typically realized in 2D heterostructures. A few comments are in order:
- The bias and/or gating potential drops are much smaller than the confinement energies in the quantum wells of the structure. Therefore, wave functions $\zeta_{e}$, $\zeta_{h}$ are not distorted by *local* variations of the external potential, which is thus uninfluential on the vertical localization of the electron and the hole. In other words, the effective interaction $U_C$ can be considered independent of space ($\mathbf{R}$) and time;
- In CQW systems $U_e$ and $U_h$ are in general different, with opposite sign. A metallic gate on top of the structure, for example, generates an electrostatic potential which is opposite for electrons and holes, and it is slightly different in strength between the two layers, due to the different distance from the gate. A simple capacitor model[@GrasselliJCP15] shows that the difference is typically of a few meV and comparable to the generated in-plane voltage drop.[@note_gating]
- We shall investigate external potentials which are short ranged and vanish exactly outside the scattering region. According to the energy conservation law, excitations to higher internal IX levels, up to the dissociation threshold, are allowed only *inside* the scattering region **B**; in the asymptotic regions (**A**, **C**), the CM energy (which in our simulations is always much smaller than the lowest internal excitation threshold, see Tab. \[tab:energies\]) is not sufficient to excite the internal dynamics, and the IX can only be transmitted/reflected in the internal ground state.
Two regimes can be identified, with the strength of the scattering potential being weak or strong with respect to the internal motion excitation energies. Below we shall investigate separately these two regimes for uniform, infinitely long well/barriers potentials, chosen separately for the two particles, but with a common width $L_y$, $$U_{\mbox{\scriptsize ext}} =
\left\{\begin{array}{ll}
U_{e,0} + U_{h,0} & \mbox{if} \; 0\leq y \leq L_y \\
0 & \mbox{otherwise}
\end{array}\right.
\label{eq:potential}$$
Weak external potential {#sec:WeakPotential}
-----------------------
In Fig. \[weakpot\] we show a simulation for a weak potential well for the hole, $U_{h,0} = -1.0\,\mbox{meV}, U_{e,0}=0$, and $L_y=40\,\mbox{nm}$. Note that the external potential strength is smaller than the lowest internal excitation energy, $\mathcal{E}_1-\mathcal{E}_0 = 2.35\,\mbox{meV}$ (see Tab. \[tab:energies\]). The IX is initialized with a kinetic energy $E_{\mbox{\scriptsize CM}} = 0.2 \,\mbox{meV}$ with normal incidence ($\theta=0$) to the well.
Figure \[weakpot\](a) shows that the IX is completely transmitted as a bound state in the internal ground state, as in a single-particle scattering. This is confirmed in Fig. \[weakpot\](b) where we plot the time evolution of the **A**,**B**,**C**-coefficients, showing that scattering is over in $\sim 20\,\mbox{ps}$, and the wave-packet is completely transmitted. Furthermore, the full propagation and the RIX and TDH propagations give indistinguishable results (we did not plot the TDH calculation for clarity). This proves that in this regime *i)* the wave function can be factorized into the product $ \Psi(\mathbf{R},\mathbf{r};t) = \chi(\mathbf{R};t)\phi(\mathbf{r};t) $ and *ii)* it remains in the ground state $\phi(\mathbf{r};t) = \phi_0(r)e^{-i\mathcal{E}_0 t/\hbar}$ of the internal DoFs, the fundamental assumption in the RIX approximation, during the *whole* propagation.
Strong external potential {#sec:StrongPotential}
-------------------------
We next investigate scattering of an IX against external potentials with an energy scale comparable to the IX internal excitations.
### Electron well {#Sec:elec_well}
We first consider scattering of a IX with an external potential consisting of a square well applied to the electron, $U_{e,0} = -3.0\, \mathrm{meV}$, $U_{h,0}=0$, and $L_y=40\,\mbox{nm}$ (see Fig. \[elec\_well\](a)). The initial CM kinetic energy of the IX is set to $E_{\mbox{\scriptsize CM}}=0.4 \,\mbox{meV}$. Now the external potential intensity is slightly below the dissociation energy, $-\mathcal{E}_0 = 3.63 \,\mbox{meV}$, but it is sufficient to excite the IX to higher energy internal states. The dissociation phenomena, not possible here due to energy conservation, has been analyzed elsewhere for a simpler 1D geometry[@GrasselliJCP15].
To highlight the role of internal excitations, in Fig. \[elec\_well\_proj\] we show the projections $$\rho_\mathrm{CM}^j(\mathbf{R}; t) = \left|\int d\mathbf{r} \phi_j(\mathbf{r}) \Psi(\mathbf{R},\mathbf{r};t)\right|^2$$ on the first internal eigenstates $j=1s,2p_y,2s$ at $t=10$ ps. The $p_x$-state projection is not shown, since it cannot be excited due to potential symmetry reasons. We see that the excitation of the internal DoF takes place especially at the edge of the well, where the change in potential energy is abrupt. $\rho_\mathrm{CM}^j(\mathbf{R}; t)$, for $j\neq 1s$, vanish as $t\gtrsim 20$ ps, i.e. when the scattering process is almost concluded.
In Fig. \[elec\_well\](b-d) we show the **A**,**B**,**C**-coefficients at normal incidence $\theta=0$, and $\theta=\pi/6$ and $\theta=\pi/4$ incidence. In all these cases, scattering is over in $\sim 15\div20\,\mbox{ps}$, the scattering time being larger for larger incident angles. The transmitted wave-packet and the amplitude in the potential region are smaller as the angle increases. This is in agreement with the lower momentum of the exciton in the direction normal to the potential well, *if scattering is not at a resonance energy*.
In this regime, RIX or TDH substantially overestimate the transmission obtained from full propagation, while RIX and TDH give very similar results between each other. Note however, that the trend with incident angle is similar for the three methods and that in the potential region **C** all methods give very similar results. Therefore, even though the CM localization is similar during scattering, the internal DoFs have a strong effect on the transmission and reflection at asymptotic times.
We also verified by explicit simulations that transmission with $\theta >0$ coincides with that obtained at normal incidence but with a kinetic energy which corresponds to the normal component of CM wave vector, which for the present case is $0.3\,\mbox{meV}$ at $\theta=\pi/6$, and $0.2 \,\mbox{meV}$ at $\theta=\pi/4$. This holds true both in the full propagation and in the mean-field methods. This is clearly to be expected in the RIX approximation, which is effectively a one-particle problem, since the evolving CM wave function is separable into an $X$ and a $Y$ part in a translationally invariant potential. For full and TDH propagations, where the Coulomb interaction couples $x$ and $y$ directions, note that for an external potential which is invariant in one planar direction, say, with respect to $x_e, x_h$, the *total* potential is invariant with respect to $X$, $U_C(x,y)+U_{\mbox{\scriptsize ext}}(Y,y)$. Therefore, the $X$ Fourier components of the full wave function is not scattered by the potential, and separates as $\Psi(\mathbf{R},\mathbf{r}) = e^{i K_X X} \Psi_{\perp}(Y,x,y)$. Accordingly, the dynamics corresponds to the free-particle one in the $X$ direction.
![Scattering in an electron well with $U_{e,0}=-3 \,\mbox{meV}$, $U_{h,0}=0$, $L_y = 40 \,\mbox{nm}$. (a) Sketch of the external potential. (b-d) Time evolution of the **A**,**B**,**C**-coefficients (line colors as in Fig. 3) at normal incidence $\theta=0$, $\theta=\pi/6$, $\theta=\pi/4$, as indicated. Dots: RIX approximation. Dashed lines: TDH approximation. Solid line: full propagation.[]{data-label="elec_well"}](figure4.pdf){width="\columnwidth"}
### Hole well
We next consider the equivalent scattering problem, but with the potential well applied to the hole, with $U_{h,0} = -3.0 \,\mbox{meV}$, $U_{e,0}=0$. Selected results are shown in Fig. \[hole\_well\](b-d). The kinetic energy and scattering angles are identical as in Fig. \[elec\_well\](a). Due to the different effective masses of electrons and holes, however, this problem is not equivalent to the previous one.
Several differences with respect to the electron case (see Fig. \[elec\_well\]) can be recognized. First, the component in the potential well during the scattering is larger and, accordingly, scattering times are longer. This is expected, due to the heavier mass of the hole. However, the localization in the potential well (**B**-coefficient) is largely underestimated by the mean-field calculations, contrary to the previous case. Second, when the normal component of the exciton momentum is changed, there is no definite trend between the full calculation and mean-field calculations: at normal incidence the transmission coefficient is lower for the full propagation with respect to the RIX and TDH approaches. At $\theta=\pi/6$, Fig. \[hole\_well\](c), the **C**-coefficient is lower in the full propagation only until $t\approx 20\,\mbox{ps}$, but the asymptotic value is larger for the full calculation than in the mean-field approximations. Finally, at $\theta=\pi/4$, Fig. \[hole\_well\](d), the full **C**-coefficient is always larger than the RIX value, both during scattering and at asymptotic times. This should be ascribed to the activation of a resonant transmission channel. This coupling is able to excite the IX into a superposition of higher internal states and it breaks the separability into CM and relative motion parts. This is why even the TDH method is not able to reproduce the full results, and merely mimics the RIX approximation.
A peculiar behavior is shown in Fig. \[hole\_well\](b) which exhibits a plateau in the **C**-coefficient during the scattering within the full dynamics. This indicates a non uniform transmission of the IX wave function. In a semiclassical picture, the hole is trapped into the well while the electron is partially transmitted. Due to Coulomb interaction, however, the transmitted electron inverts its motion, the IX bounds again and the pair is finally transmitted. Therefore, the transmitted wave-packet splits into an advanced and a delayed IX. Clearly, this requires the excitation of the internal modes, and this behavior is not reproduced by the RIX model.[@GrasselliJCP15] To confirm this interpretation, we calculate the classical internal oscillation period, (see Suppl. Mat.[@SupplMat_IX]) and estimate it in $\tau \approx 5.5 \, \mathrm{ps}$, which is indeed comparable to the time duration of the transmission plateau in Fig. \[hole\_well\](b).
![Same as in Fig. \[elec\_well\], but with hole well $U_{h,0}=-3 \,\mbox{meV}$, $U_{e,0}=0$. Short animations of cases (b) and (d) are presented in the Supplemental Material.[@SupplMat_IX] []{data-label="hole_well"}](figure6.pdf){width="\columnwidth"}
### Symmetric potential barrier/well
For completeness, we next discuss a symmetric potential with opposite sign for the two particles, $U_{h,0} = 3.0\,\mbox{meV}$ and $U_{e,0} = -3.0 \,\mbox{meV}$. The potential profile is sketched in Fig. \[gating\_width\](a). This is a somehow special situation, because the average potential is zero.
We consider $\theta=0$ and $\theta=\pi/4$ at the CM kinetic energy of $E_{\mbox{\scriptsize CM}}=0.4\,\mbox{meV}$. For both incident angles, the transmission coefficient for the full propagation is small, due to the repulsive barrier felt by the hole. Most interestingly, it is substantially smaller than the one obtained from the mean-field methods, for which the average of the external potential over the internal DoFs makes the hole barrier smooth, thus favoring transmission. Note also that for $\theta=\pi/4$ the TDH result deviates substantially from the RIX calculation. Nevertheless, it is still far from the full calculation.
![Scattering in a symmetric potential $U_{h,0}=+3 \,\mbox{meV}$, $U_{e,0}=-3\,\mbox{meV}$. (a) Sketch of external potential. (b,c) Time evolution of the **A**,**B**,**C**-coefficients (line colors as in Fig. 3) at normal incidence $\theta=0$ and $\theta=\pi/6$, as indicated. Here, $L_y = 40 \,\mbox{nm}$. A short animation of case (b) is presented in the Supplemental Material[@SupplMat_IX] (d) Time evolution of the **C**-coefficient for potentials of different width $L_y$, as indicated. Dots: RIX approximation. Solid line: full propagation. In this case, the TDH approximation is not shown for clarity.[]{data-label="gating_width"}](figure7.pdf){width="\columnwidth"}
To show that this is not an accidental situation, we have calculated the transmission coefficient as a function of the width of the external potential (Fig. \[gating\_width\](d)). Note that the behavior of the transmission coefficient is not monotonous with the the width of the external potential, $L_y$, as a possible consequence of resonant transmission for the CM DoFs alone. In particular, both for the full and the mean-field propagation, the transmission coefficient is larger for $L_y=28\,\mbox{nm}$ and $L_y=52\,\mbox{nm}$ than for $L_y=40\,\mbox{nm}$. Still, the transmission is systematically overestimated by the mean-field approach.
Single slit potentials {#sec:SingleSlit}
----------------------
We finally investigate evolution through a single slit potential. This consists of an aperture of width $\Delta$ in an otherwise infinitely long barrier/well potential similar to those investigated in the previous sections. In Figs. \[fig8\] we summarize results for a slit with $\Delta=240\,\mbox{nm}$ and a slightly asymmetric electron/hole potential $U_e = 6\,\mbox{meV}$, $U_h = -3\,\mbox{meV}$, and $L_y = 20\,\mbox{nm}$. These values exclude tunneling through the barrier, while satisfying the condition $ U_{\mbox{\scriptsize ext}} \Delta_t/(2\pi\hbar) \ll 1$. The IX is initialized with a CM kinetic energy $E_{\mbox{\scriptsize CM}}=0.4\,\mbox{meV}$, moving towards the mid-point of the slit with normal incidence, and an initial width of the CM minimum uncertainty wave packet $\sigma_{X,Y}=160\,\mbox{nm}$. These parameters produce several diffraction lobes in the transmitted wave-packet in the RIX approximation, shown in \[fig8\](a). Moreover they avoid the IX wave from spreading too much before reaching the slit.[@note_sigma]
Figure \[fig8\](b) shows a snapshot of the CM marginal probability of Eq. (\[eq:MarginalProbability\]) at $t=36\,\mbox{ps}$. A comparison with the equivalent RIX calculation in panel (a) helps to identify several important features. First, while the reflected part of the CM wave-packet is very similar in the two calculations, the diffraction lobes are almost suppressed in the full calculation. Second and most interesting, part of the CM wave-packet propagates as edge states along the barrier, far from the aperture. In semiclassical terms, this corresponds to a IX, with the hole trapped inside the well and the electron trapped on either side of the barrier by the electron-hole attraction. Therefore, this is a genuine correlation effect which cannot be reproduced by a mean-field approach, and indeed it is completely absent in Fig. 7(a).
The evolution of the **A**,**B**,**C**-coefficients are shown in Fig. \[fig8\](d) for full propagation and the RIX model (the TDH approximation almost coincides with the RIX one and it is not shown). Note that here the potential region for integration has been extended to ten times the external potential range ($200\,\mbox{nm}$, see Fig. \[fig8\](a)) to measure accurately the asymptotic coefficients, since part of the wave function remains trapped at the edges of the potential, as we discussed above. These coefficients show a qualitative agreement between full and mean-field methods. This is because in the region where the wave function is large (the slit) the potential is vanishing. Therefore, the average external potential contribution is weak, and we are in a regime similar to Sec. \[sec:WeakPotential\]. However, at asymptotic times, the full calculation shows a transmission coefficient which is smaller than in the RIX calculation, the difference being the fraction of the wave function propagating along the edges of the external potential. Therefore, even if in a weak potential regime, correlation effects are exposed in the full calculation but not in the mean-field propagation.
In Fig. \[fig8\](c) we plot $ |\Psi(\mathbf{R},\mathbf{r}=\mathbf{0})|^2$ for the full calculation. On the one hand, there is very little difference with respect to the CM marginal probability of panel (b), indicating that the $\mathbf{r}=\mathbf{0}$ contribution is by far the largest in the relative coordinate average which provides the CM marginal probability. On the other hand, this is proportional to the IX optical recombination probability and shows that an optical luminescence experiment with sub-$\mu$m resolution would be able to probe the wave function with accuracy.[@MauritzPRL99]
We have repeated similar calculations, but with the electron and the hole potential exchanged, $U_e = -3.0 \, \mbox{meV}, U_h = +6.0 \, \mbox{meV}$. We observed, in this case, no qualitative difference with respect to the previous one. While full and mean-field calculations agree overall, there is a substantial part of the CM wave function which propagates as a bound IX along the potential edges, which is not captured by the mean-field calculation.
In Fig. \[fig8\] we show a somehow special situation, where transmission and reflection probabilities are almost equal. Therefore, we show for completeness in Fig. \[fig9\] the evolution of the **A**,**B**,**C**-coefficients when the slit potential is ‘open’, i.e., most part of the CM wave-packet is transmitted, or ‘closed’, i.e., the CM wave-packet is almost fully reflected. Results are shown for the same potentials used in Fig. \[fig8\]. Again, we repeated the calculation interchanging the particle potential, finding no quantitative difference with the original one. Interestingly, in all cases a similar fraction of the CM wave function propagates along the edges of the potential barrier and well. This phenomenon seems thus to be mostly related to the Bohr radius of the exciton, rather than the slit aperture: a larger (smaller) $\Delta$ determines the asymptotic value of the **B**-coefficient, at a given $\sigma$, only through the fact that a greater (smaller) part of the IX wave packet shall hit the edges of the slit.
![Time evolution of the **A**,**B**,**C**-coefficients at selected values of $\Delta$, as indicated. Same potential strengths as in Fig. 7. Lines as in Fig. 7(d). Short animations of the propagation as in cases (a) and (b) are presented in the Supplemental Material.[@SupplMat_IX] []{data-label="fig9"}](figure9.pdf){width="\columnwidth"}
Conclusions {#sec:Conclusions}
===========
Scattering of composite particles is an important issue in several fields beyond semiconductors, such as molecular and nuclear scattering.[@Bertulani2015; @AhsanPRC10] However, theoretical methods are relatively little developed, due to the numerical complexity. Here we have developed a numerical scheme to approach the particularly hard problem of a Coulomb bound complex scattering against arbitrary potentials, and we have applied the method to the specific case of IXs confined in CQW systems. On the one hand, this system offers the unique possibility to probe scattering of a composite particle with detailed optical means. On the other hand, we have shown here that *full* numerical propagation can be obtained for this system, due to the limited number of DoFs allowed by the low dimensionality ensuing from CQW quantum confinement. This allows both to compare with perspective experiments and to test approximate and numerically simpler methods, such as mean-field methods.
For the present case we have shown that mean-field methods are predictive only for external potentials which, while coupling the CM and relative dynamics, are sufficiently weak, i.e., with an energy scale which is much smaller than the first excitation gap for the internal motion of the IX. In this case, the CM and relative DoF can be factorized at any time of the evolution.
The RIX and the TDH approximations, however, are inadequate in predicting the transmission and reflection coefficients for stronger potentials, i.e., when the strength of the external potential is comparable to internal levels spacing. In this situation the scattering potential partially excites (locally) the IX to higher internal levels, and the wave function cannot be factorized during evolution in the potential region. In such a case, the asymptotic IX transmission, as computed from the full propagation, can be larger or smaller than the transmission extracted from mean-field methods, depending on the specific parameters of the system.
We have identified other signatures of the internal dynamics of the IX, coming into play during scattering, which cannot be captured by any mean-field calculation. For example, when plotted against time, the transmission through a well may exhibit plateaux before scattering is completed, which are absent in the simulations performed within RIX and TDH approximations. Moreover, in the single slit scattering problem, part of the wave-packet is not transmitted or reflected, but propagates along the edges. Again, this requires higher internal level excitations in the potential region through coupling of relative and CM DoFs which cannot be captured by a mean-field approach. This genuine correlation effect could be detected in optical experiments. It is also interesting to note that the RIX and TDH approximations basically coincide in almost all numerical simulations, indicating that, besides keeping the CM and relative subsystems decoupled, a mean field approach also smoothens the external potentials in such a way that it cannot exchange energy with the internal motion DoFs - which is *de facto* always in its ground state - but only with the CM one.
We finally note that, even for the case of only four DoFs, full wave-packet quantum propagation is a demanding task which required the development of a massively parallel code. On the one hand, exact calculations as the present one may serve as a severe benchmark for approximate, beyond mean-field methods which might be less computationally intensive. On the other hand, the implemented FSS method is suitable to treat also time-dependent external potentials, which might be produced in heterostructures for driving single carriers or bound IX wave packets, such as surface acoustic waves produced by interdigital devices, at small additional numerical cost.[@ViolanteNJP14] This is left as a future development.
Acknowledgments
===============
We acknowledge INDEX for partial financial support. We acknowledge CINECA for computing time on parallel architectures under the Iscra C project IsC33-“FUQUDIX”. We acknowledge L. Butov and M. Fogler for useful discussions and suggestions during a stay at the University of California in San Diego, where part of this work has been developed.
[52]{}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
doi:10.1103/PhysRevLett.68.1383) [****, ()](\doibase
10.1103/PhysRevLett.68.1379) [****, ()](\doibase 10.1038/nphys245) [****, ()](\doibase
10.1038/45189) [****, ()](\doibase 10.1209/epl/i1997-00544-3) [****, ()](\doibase 10.1103/PhysRevB.47.1683) [****, ()](\doibase
http://dx.doi.org/10.1038/417047a) [****, ()](http://stacks.iop.org/0953-8984/16/i=50/a=R02) [****, ()](\doibase
doi:10.1038/nature10903) [****, ()](http://stacks.iop.org/0295-5075/107/i=1/a=10012) [****, ()](http://stacks.iop.org/0295-5075/12/i=5/a=007) [****, ()](\doibase http://dx.doi.org/10.1063/1.2267263) [****, ()](\doibase 10.1021/nl070386c), [****, ()](\doibase http://dx.doi.org/10.1063/1.2978214) [****, ()](\doibase
http://dx.doi.org/10.1063/1.4722938) [****, ()](\doibase 10.1103/PhysRevLett.106.196806) [****, ()](\doibase 10.1126/science.1157845), [****, ()](\doibase 10.1103/PhysRevLett.110.127403) [****, ()](http://stacks.iop.org/0953-8984/6/i=20/a=014) [****, ()](\doibase 10.1103/PhysRevB.51.5453) [****, ()](\doibase 10.1007/s00601-015-0990-z) [****, ()](\doibase 10.1103/PhysRevC.82.064607) [****, ()](\doibase http://dx.doi.org/10.1063/1.529549) [****, ()](\doibase 10.1103/PhysRevC.83.054621) [****, ()](\doibase 10.1103/PhysRevA.87.012904) [****, ()](\doibase 10.1103/PhysRevA.78.032711) [****, ()](\doibase 10.1134/1.1326962) [****, ()](\doibase 10.1103/PhysRevA.81.022708) [****, ()](\doibase
doi:10.1016/S0370-1573(99)00047-2) [****, ()](\doibase http://dx.doi.org/10.1063/1.4905483) [****, ()](\doibase
10.1103/PhysRevB.91.125437) @noop [**]{} (, ) [****, ()](\doibase http://dx.doi.org/10.1063/1.1368156) [****, ()](\doibase
http://dx.doi.org/10.1016/0021-9991(91)90137-A) [****, ()](\doibase http://dx.doi.org/10.1063/1.1774980) [****, ()](\doibase 10.2183/pjab.69.161) [****, ()](\doibase http://dx.doi.org/10.1351/pac199769061179) [****, ()](\doibase http://dx.doi.org/10.1063/1.1741033) [****, ()](\doibase 10.1080/00268976400100041), [****, ()](\doibase http://dx.doi.org/10.1038/nature00943) @noop [**]{} (, ) @noop [****, ()]{} [****, ()](\doibase
10.1103/PhysRevB.38.3263) @noop [**]{} (, ) @noop @noop @noop [****, ()](\doibase
10.1103/PhysRevLett.82.847) [****, ()](http://stacks.iop.org/1367-2630/16/i=3/a=033035)
\
,$^{1,2}$ Andrea Bertoni,$^2$ and Guido Goldoni $^{1,2}$
*$^1$Dipartimento di Scienze Fisiche, Informatiche e Matematiche,\
Università degli Studi di Modena e Reggio Emilia, Via Campi 213/a, Modena, Italy*\
*$^2$S3, CNR-Istituto Nanoscienze, Via Campi 213/a, Modena, Italy*
The time dependent Hartree method {#App_TDH}
=================================
Consider a system composed of interacting subsystems 1 and 2, with the Hamiltonian[@note_App_1] $$H(1,2) = H_1^{(0)}(1) + H_2^{(0)}(2) + U(1,2)$$ where $$H_j^{(0)}(j) = T_j(j) + U_j(j) \qquad j=1,2$$ are single subsystem Hamiltonians, and $T_j(j)$ and $U_j(j)$ are the kinetic energy and potential energies, respectively, for the $j-th$ subsystem. $U(1,2)$ is some potential which couples the coordinates for the two subsystems.
The time-dependent Schrödinger equation reads $$\Psi(1,2; t + dt) =\left(1 + \frac{dt}{i\hbar} H(1,2) \right) \Psi(1,2; t)$$ where $dt\rightarrow 0$. We then *assume* the following *ansatz* for the total wave function $$\Psi(1,2; t) = \eta(t)A(1;t)B(2;t),
\qquad \forall t \in I\subseteq \mathbb{R}^+_0 ,$$ i.e. a permanent separation during the time interval $I$ of the evolution. Here we introduced the (redundant) variable $\eta=\eta(t)$ in order to be able of freely choose the global phases of $A(1,t)$ and $B(2,t)$. A constraint equation will thus be needed. By multiplying on the left by $\eta^*(t+dt)B^*(2;t+dt)$ and integrating on the variable set 2, we obtain $$\begin{split}
A(1;t+dt) &= \int d2 \eta^*(t+dt)B^*(2;t+dt) \left(1 + \frac{dt}{i\hbar} H(1,2) \right) \eta(t)B(2;t) \;
A(1;t) \label{a1}
\end{split}$$ In a similar way, we obtain $$\begin{split}
B(2;t+dt) &= \int d1 \eta^*(t+dt)A^*(1;t+dt) \left(1 + \frac{dt}{i\hbar} H(1,2) \right) \eta(t)A(1;t) \;
B(2;t). \label{a2}
\end{split}$$ and $$\begin{split}
\eta(t+dt) &= \int d1 B^*(2,t+dt)A^*(1;t+dt) \left(1 + \frac{dt}{i\hbar} H(1,2) \right) A(1;t)B(2,t) \;
\eta(t). \label{a3}
\end{split}$$ Consider Eq. (\[a3\]). We choose the phases of $A(1,t)$ and $B(2,t)$ such that the scalar products of these functions with their time derivatives, computed at the same time, vanish. This is equivalent, as $dt\rightarrow 0$, to $$\langle A(1;t+dt) | A(1;t) \rangle = \langle B(1;t+dt) | B(1;t) \rangle = 1,$$ $\forall t \in I$. By inserting these equations into Eq. (\[a3\]), we have the following equation of motion for $\eta(t)$: $$\begin{split}
\eta(t+dt) &= \int d1 B^*(2,t)A^*(1;t) \left(1 + \frac{dt}{i\hbar} H(1,2) \right) A(1;t)B(2,t) \;\eta(t) = \\
& = \left(1+\frac{dt}{i\hbar}\langle 1,2| H(1,2) |1,2\rangle (t)]\right) \eta(t)
\end{split}$$ i.e. $i\hbar\partial_t\eta = \langle H \rangle \eta$, which can be itself seen as *the* constraint equation, and has as a (formal) solution $$\eta(t) = \mathcal{T} \exp \left( \frac{1}{i\hbar} \int_0^t [\langle 1,2| H(1,2) |1,2\rangle (t')] dt' \right)$$ This, by the way, is just a global-in-space phase and thus it is of no importance when we’re interested in the time evolution of the *probability density*. With this constraint, we then have the following expression for Eq. (\[a1\]) $$\begin{split}
A(1;t+dt) &= \left\lbrace 1 + \frac{dt}{i\hbar} \left( H_1^{(0)}(1) + \beta(t) + u_{\mathrm{1},\mbox{\scriptsize eff}}(1;t) - [\langle 1,2| H(1,2) |1,2\rangle (t)] \right) \right\rbrace A(1;t)
\end{split}$$ where $$\beta(t) \equiv \int d2 B^*(2;t)H_2^{(0)}(2)B(2;t),$$ and $$u_{\mathrm{1},\mbox{\scriptsize eff}}(1;t) \equiv \int d2 |B(2,t)|^2 U(1,2).$$ The last term, $[\langle 1,2| H(1,2) |1,2\rangle (t)]$, comes from the constraint equation. We must specify that, if one is interested in probability densities, the global-in-space ‘effective potentials’ $\beta(t)$, $[\langle 1,2| H(1,2) |1,2\rangle (t)]$ are not influential,[@note_App_2] and thus we can re-write the equation of motion for the variable $A(1,t)$ simply as $$A(1;t+dt) = \left[ 1 + \frac{dt}{i\hbar} \left( H_1^{(0)}(1) + u_{\mbox{\scriptsize eff}}(1;t) \right) \right] A(1;t)$$ that is $$i\hbar \partial_t A(1;t) = \left( H_1^{(0)}(1) + u_{\mbox{\scriptsize eff}}(1;t) \right) A(1,t)$$ Analogously, for the variable $B(2;t)$ we find $$i\hbar \partial_t B(2;t) = \left( H_2^{(0)}(2) + u_{\mathrm{2},\mbox{\scriptsize eff}}(2;t) \right) B(2,t)$$ where $$u_{\mathrm{2},{\mbox{\scriptsize eff}}}(2;t) \equiv \int d1 |A(1;t)|^2 U(1,2).$$ We can formally solve the latter equations by means of the evolution operator: $$\begin{aligned}
A(1,t+\Delta_t) = \mathcal{U}_1(1; t+\Delta_t, t) A(1;t) , \label{eqTDH1}\\
B(2,t+\Delta_t) = \mathcal{U}_2(2; t+\Delta_t, t) B(2;t) \label{eqTDH2}\end{aligned}$$ where $$\mathcal{U}_1(1; t+\Delta_t, t) = \exp \left[ \frac{1}{i\hbar} \int_t^{t+\Delta_t} dt' \left( H_1^{(0)}(1) + u_{\mbox{\scriptsize eff}}(1;t') \right) \right].$$ and $$\mathcal{U}_2(2; t+\Delta_t, t) = \exp \left[ \frac{1}{i\hbar} \int_t^{t+\Delta_t} dt' \left( H_2^{(0)}(2) + u_{\mbox{\scriptsize eff}}(2;t') \right) \right].$$ which can be handled by the split-step-Fourier method.
To summarize, the quantum evolution of a system composed of subsystems 1 and 2 have been separated in the quantum evolutions each subsystem separately, each with a time-dependent Hamiltonians. For, say, subsystem 1, an effective potential $u_{\mbox{\scriptsize eff}}(1;t)$ arises, which can be written as the expectation value of the coupling potential averaged on the wave function of subsystem 2. Equivalently for particle 2, with particle coordinates exchanged. In this sense, this is a mean-field-like approach.
The Fourier split step method (FSS) for each of the two subsystems can then be applied to obtain the global evolution. In the present case, the systems 1 and 2 are representing the CM and relative motion coordinates, respectively; in this case, Eqs. (\[eqTDH1\]) and (\[eqTDH2\]) read as Eqs. (21) and (22) in the main text: $$\begin{split}
\chi(\mathbf{R};t+\Delta_t) &= \exp \left\lbrace - \frac{i}{\hbar} \int_t^{t+\Delta_t}dt' \left[ \frac{\mathbf{P}^2}{2M} + U_{\mbox{\scriptsize eff}}(\mathbf{R};t') \right] \right\rbrace \chi(\mathbf{R};t)
\end{split} \label{eqTDHCM}$$ and $$\begin{split}
\phi(\mathbf{r};t+\Delta_t) &= \exp \left\lbrace -\frac{i}{\hbar} \int_t^{t+\Delta_t} dt' \left[ \frac{\mathbf{p}^2}{2m} + U_{C}(\mathbf{r}) + u_{\mbox{\scriptsize eff}}(\mathbf{r};t') \right] \right\rbrace \phi(\mathbf{r};t)
\end{split} \label{eqTDHrel}$$ with $$\begin{aligned}
U_{\mbox{\scriptsize eff}}(\mathbf{R};t) \equiv \int d\mathbf{r} |\phi(\mathbf{r};t)|^2 U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r})
\label{effTHDpotss}
\\
u_{\mbox{\scriptsize eff}}(\mathbf{r};t) \equiv \int d\mathbf{R} |\chi(\mathbf{R};t)|^2 U_{\mbox{\scriptsize ext}}(\mathbf{R},\mathbf{r}) \label{effTHDpots}\end{aligned}$$
Classical internal oscillation period for the first excited ($p$) state
=======================================================================
Starting from the expression of the classical energy for the first excited state (we choose for example the $p_y$ state, whose eigen-energy is $\mathcal{E}_1$) $$\mathcal{E}_1 = \frac{1}{2} m \dot{y}^2 + U_C(y) \label{eqEnClassCons}$$ (we assume for simplicity the trajectory at fixed $x=0$), where $U_C(y) \equiv e^2/(4\pi\epsilon_0\epsilon_r\sqrt{d^2+y^2})$, and taking $d$ as the distance between the centers of the two wells along the growth axis, we have the classical turning points (at which $\dot{y}=0$) $y = \pm \overline{y}$, where $\overline{y}\equiv \sqrt{[e^2/(4\pi\epsilon_0\epsilon_r\mathcal{E}_1)]^2 - d^2} $. We can then compute the classical internal oscillation period, $\tau$, by isolating $(\dot{y}^2)^{-1/2}$ in Eq. (\[eqEnClassCons\]), integrating over $y$ between 0 and $\overline{y}$, and multiply the result by four, since the integration corresponds to one fourth of the classical path [@LandauMech]: $$\begin{aligned}
\tau = 4 \int_0^{\overline{y}} dy \left[\frac{2}{m} \left(\mathcal{E}_1 + \frac{e^2}{4\pi\epsilon_0\epsilon_r\sqrt{d^2+y^2}}\right)\right]^{-1/2}, \end{aligned}$$ Taking the specific values concerning the physical system adopted in the manuscript, we have $d=12\,\mbox{nm}$, and $\mathcal{E}_1=-1.28\,\mbox{meV}$ (see Tab. II in the main text), and the (numerical) integration leads to $\tau \approx 5.5 \, \mathrm{ps}$.
[52]{}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 @noop @noop [****, ()](\doibase http://dx.doi.org/10.1063/1.105157) @noop [**]{} (, )
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss the consequences of higher derivative Lagrangians of the form $\alpha_1 A_{\mu}(x)\dot{x}^\mu$, $\alpha_2 G_{\mu}(x)\ddot{x}^\mu$, $\alpha_3 B_{\mu}(x)\dddot{x}^\mu$, $\alpha_4 K_{\mu}(x)\ddddot{x}^\mu$, $\cdots$, $U_{(n)\mu}(x)x^{(n)\mu}$ in relativistic theory. After establishing the equations of the motion of particles in these fields, we introduce the concept of the generalized induction principle assuming the coupling between the higher fields $U_{(n),\mu}(x),\ n\geq1$ with the higher currents $j^{(n)\mu}=\rho(x)x^{(n)\mu}$, where $\rho(x)$ is the spatial density of mass or of electric charge. In addition, we discuss the analogy of the field $G_\mu(x)$ with the gravitational field and its inclusion in the general relativity framework in the last section. This letter is an invitation to reflect on a generalisation of the concept of inertia and we also discuss this problem in the last section.'
author:
- Mathieu Beau
title: Comment on the higher derivative Lagrangians in relativistic theory
---
Higher derivative Lagrangians and dynamic equations of a particle
=================================================================
Ostrogradsky introduced the idea of higher derivative Lagrangians in classical mechanics [@MecaG0], and there is a series of articles published about this topic, see [@MecaG1],[@MecaG2],[@MecaG3],[@MecaG4],[@GeEM]. However, to my knowledge, there is no article dealing with relativistic higher derivative Lagrangians of this type: $$\begin{gathered}
\label{L}
\widetilde{L}(\dot{x},\ddot{x},\cdots,x^{(n)})\\
=\alpha_1 A_{\mu}(x)\dot{x}^\mu+\alpha_2 G_{\mu}(x)\ddot{x}^\mu+\cdots+\alpha_n U_{(n)\mu}(x)x^{(n)\mu} \end{gathered}$$ where $x^{(n)}(s)\equiv d^n x(s)/ds^n$ are the $n$-derivatives of the position ($ds=cd\tau$, where $\tau$ is the proper time), and $U_{(n)\mu}(x),\ n=1,2,3,..$ are the generalized vectorial fields coupling linearly with the $n$-derivatives. Here we denote the field $U_{(1)\mu}(x)=A_\mu(x)$ to refer to the electromagnetic potential. Also we denote $U_{(2)\mu}(x)=G_\mu(x)$ because of the analogy with the geodesic equations that we will see in the equation (\[EqGeodesique\]). From (\[L\]), we set the action: $$S=\int ds L_0(\dot{x})+\int ds \widetilde{L}(\dot{x},\ddot{x},\cdots,x^{(n)}) \ ,$$ where $L_0(\dot{x})\equiv\frac{mc^2}{2}\dot{x}_\mu\dot{x}^\mu$. We will not give an explicit general dynamic theory for a given $n$, we consider only $n=2$ for the moment and we will discuss the general case later. An integration by part for $n=2$ gives the equivalent action [@MecaG3]: $$\widetilde{S}=\alpha_1\int ds A_{\mu}(x)\dot{x}^\mu - \alpha_2 \int ds\ \partial_\nu G_{\mu}\dot{x}^\mu\dot{x}^\nu\ ,$$ and one can see that the first part of the action $\widetilde{S}$ is similar to the electrodynamic action whereas the second part is similar to the gravitational action. Indeed, from the generalized Euler-Lagrange equations (see [@MecaG0],[@MecaG1],[@MecaG2],[@MecaG4]): $$\begin{aligned}
\label{Euler-Lagrange}
\frac{d^2}{ds^2}(\frac{\partial L}{\partial \ddot{x}^\mu})-
\frac{d}{ds}(\frac{\partial L}{\partial \dot{x}^\mu})+\frac{\partial L}{\partial x^\mu}=0\ ,\end{aligned}$$ with $L=L_0+\widetilde{L}$, we get: $$\begin{aligned}
\label{EqGeodesique}
mc^2\eta_{\mu\nu}\ddot{x}^\nu-\alpha_2\varepsilon_{\mu\nu}\ddot{x}^\nu-\alpha_2\Delta_{\mu\nu\sigma}\dot{x}^\nu\dot{x}^\sigma
=-\alpha_1 F_{\mu\nu}\dot{x}^\nu\end{aligned}$$ where $\varepsilon_{\mu\nu}$ and $\Delta_{\mu\nu\sigma}$ are: $$\begin{aligned}
\label{DefTenseurs}
\begin{array}{ll}
\varepsilon_{\mu\nu}&\equiv\partial_\mu G_\nu + \partial_\nu G_\mu \label{g}\\
\Delta_{\mu\nu\sigma}&\equiv\partial_\nu\partial_\sigma G_\mu
=\frac{1}{2}(\partial_\nu \varepsilon_{\mu\sigma}+
\partial_\sigma \varepsilon_{\mu\nu}-\partial_\mu \varepsilon_{\nu\sigma}) \label{gamma1}\end{array} \end{aligned}$$ and where $F_{\mu\nu}=\partial_\mu A_\nu - \partial_\nu A_\mu$. One sees the analogy with the equations of the motion of a charged particle in a gravitational field and in an electromagnetic field. However, the fixed metric (or *background* metric) is Minkowskian. Then, the quadrivector field $G_\mu(x)$ can be seen as a *displacement vector field* and $\varepsilon_{\mu\nu}$ can be viewed as an infinitesimal strain tensor by analogy with the deformation theory of a continuous medium [@MMC].
Now, let us take $n=3$, we denote $U_{(3)\mu}\equiv B_\mu$. From $$\begin{aligned}
\label{Euler-Lagrange3}
-\frac{d^3}{ds^3}(\frac{\partial L}{\partial \dddot{x}^\mu})+\frac{d^2}{ds^2}(\frac{\partial L}{\partial \ddot{x}^\mu})-
\frac{d}{ds}(\frac{\partial L}{\partial \dot{x}^\mu})+\frac{\partial L}{\partial x^\mu}=0\end{aligned}$$ one has: $$\begin{gathered}
\label{EqGeodesique3}
mc^2\eta_{\mu\nu}\ddot{x}^\nu+\alpha_3 H_{\mu\nu}\dddot{x}^\nu-\alpha_3\Upsilon_{\mu\nu\sigma\rho}\dot{x}^\nu\dot{x}^\sigma\dot{x}^\rho\\
-3\alpha_3\varSigma_{\mu\nu\sigma}\ddot{x}^\nu\dot{x}^\sigma
\alpha_2\varepsilon_{\mu\nu}\ddot{x}^\nu-\alpha_2\Delta_{\mu\nu\sigma}\dot{x}^\nu\dot{x}^\sigma
=-\alpha_1 F_{\mu\nu}\dot{x}^\nu \end{gathered}$$ where $$\begin{aligned}
\label{DefTenseurs3}
&& H_{\mu\nu}\equiv\partial_\mu B_\nu - \partial_\nu B_\mu \label{g}
{}\nonumber\\{}&&\varSigma_{\mu\nu\sigma}\equiv\partial_\nu\partial_\sigma B_\mu
{}\nonumber\\{}&&\Upsilon_{\mu\nu\sigma\rho}\equiv\partial_\nu\partial_\sigma\partial_\rho B_\mu \end{aligned}$$ We can see that this field generalizes the idea of the electromagnetic field because of the antisymmetry of $H_{\mu\nu}$. However, in (\[EqGeodesique3\]) there are some other fields, similar to $\Delta_{\mu\nu\sigma}$, coupling with the combinations of the odd derivatives of $x^{\mu}$, i.e. $\ddot{x}^\nu\dot{x}^\sigma$ and $\dot{x}^\nu\dot{x}^\sigma\dot{x}^\rho$.
We can also discuss the higher derivative terms. For $n=4$, we denote the field $K_\mu(x)\equiv U_{(4)\mu}(x)$. The dynamic equations have a similar structure to the one we get for $G_\mu(x)$ (i.e. for $n=2$). As it has been shown that in the non-relativistic theory [@MecaG3], we notice that the Lagrangian $\alpha_4 x_\mu\ddddot{x}^\mu$ is equivalent to this Lagrangian $\alpha_4 \ddot{x}_\mu\ddot{x}^\mu$ and the quantity $\alpha_2\dot{x}^2+\alpha_4\ddot{x}_\mu\ddot{x}^\mu-2\alpha_4\dot{x}_\mu\dddot{x}^\mu$ could be interpreted as a more *general kinetic energy* [@MecaG4]. Here the problem is similar: the Lagrangian $\alpha_4 K_{\mu}(x)\ddddot{x}^\mu$ is equivalent to $\alpha_4\partial_{\mu}K_\nu \ddot{x}^\mu\ddot{x}^\nu
+\alpha_4 \partial_{\sigma}\partial_{\mu}K_\nu \ddot{x}^\mu\dot{x}^\nu\dot{x}^\sigma$ and so is obviously more complicated than the equivalent Lagrangian that we get for the special case $K_\mu(x)=x_\mu$.
To finish this section, let us now consider the generalized fields $U_{(n)\mu}(x),\ n\geq1$. By the generalized Euler-Lagrange equations: $$\sum_{k=0}^{n}(-1)^k \frac{d^k}{ds^k}\frac{\partial L}{\partial x^{(k)\mu}}=0$$ we get terms of the form $\partial_{\mu_{1}}\cdots \partial_{\mu_{p}}U_{(n)},\ p=1,\cdots,n$ multiplied by the combination of the derivatives $x^{(l_1)\mu_1}x^{(l_2)\mu_2}\cdots x^{(l_p)\mu_p}$, $\sum_{j=1}^{p}l_j=n$. One can see that the “even” $n$-fields are analogous to the “gravitational field” as the “odd” $n$-fields are to the “electromagnetic field”, since, in the dynamic equations, the derivatives $x^{(n)\mu}$ are multiplied by the symmetric (if $n$ is even) / antisymmetric (if $n$ is odd) part of the first derivative of the field: $$(\partial_{\mu} U_{\nu}+(-1)^n\partial_{\nu}U_{\mu})x^{(n)\nu}\ ,$$ here we denote the $n$-field by $U_\mu$. We will see the consequences of this remark in the next section.
General fields hypothesis
=========================
The point is to relate the possible existence of these fields with a generalized induction phenomenon. We will suppose that there exists new physical couplings with *higher currents*, yet unkown, and we will generalize the electromagnetic field theory.
Construction of the $n=2$-field equations by analogy with the vectorial electromagnetic field
---------------------------------------------------------------------------------------------
In the Lagrangian (\[L\]) for $n=2$ we notice that the field $G_\mu$ is coupled with the acceleration of the particle as the field $A_{\mu}$ is coupled with the velocity of the particle. By analogy with the construction of the electromagnetic field theory, we suggest the following field equations: $$\label{gpe2}
\partial_\mu \varepsilon^{\mu\nu}(x) = -\kappa j^{(2)\nu}(x)\ ,$$ where the acceleration current density $j^{(2)\nu}$ (generally non-conserved) is: $$\label{QuadridensitAccRG}
j^{(2)\nu}(x)\equiv \rho_m(x)c^2\frac{du^\nu}{ds}\ ,$$ where $\rho_m(x)$ is the density of particles and $\frac{du^\nu}{ds}$ is the 4-acceleration. Let us rewrite the coupling constant $\kappa$ as follows $$\kappa=\frac{8\pi G \lambda^2}{c^4}$$ where $\lambda$ has the dimension of a length.
To complete the system of field equations, we need ten equations: $$\label{gpe1}
\partial_\sigma \partial^\sigma \varepsilon_{\mu\nu}+\partial_\mu \partial_\nu \varepsilon_{\sigma}^{\sigma}
=\partial_\mu \partial^\sigma \varepsilon_{\sigma\nu}+\partial_\nu \partial^\sigma \varepsilon_{\sigma\nu}\ ,$$ The equations (\[gpe1\]) are analogous to the compatibility equations for the strain tensor in the three-dimensional non-relativistic theory of deformation of continuous media [@MMC].
Hence we get the following wave equations: $$\begin{aligned}
\label{PropaChampsPhys0}
\Box \varepsilon_{\nu\sigma}(x)+\partial_{\nu}\partial_\sigma \varepsilon_\mu^\mu(x)=-\kappa \xi_{\nu\sigma}^{(2)}(x) \end{aligned}$$ with $\xi^{(2)}_{\nu\sigma}(x)\equiv\partial_\sigma j^{(2)}_\nu(x)+\partial_\nu j^{(2)}_\sigma(x)$ . Also, the trace of $\varepsilon_{\mu\nu}$ follows this equation $$\begin{aligned}
\label{Tracen2}
\Box \varepsilon_{\mu}^\mu(x)=-\kappa\partial_\mu j^{(2)\mu}(x)\ , \end{aligned}$$ this means that $\varepsilon_\mu^\mu$ is a non-massive scalar field. In the linear theory of the deformation of a continuous medium, the trace of the strain tensor is interpreted as the contraction/dilation of the volume [@MMC]. Here we observe a similar property and by the equation (\[Tracen2\]) we conclude that the relativistic deformation of the volume of the *four dimensional continuous medium* is related to the non-conservation of the current $j^{(2)}$.
Generalisation to the $n$-field equations
-----------------------------------------
With respect to the analogy that we have already discussed above between the even fields and the gravitational field and between the odd fields with the electromagnetic field and using the field theory developed for the case $n=2$, we can construct the general field theory via a more general induction principle. Following this rule, we rewrite the constants in (\[L\]) as $\alpha_{2n}=mc^2(\lambda_{n})^{2n-2}$ and $\alpha_{2n-1}=\frac{(\xi_{n})^{2n-2}}{c^{2n-2}},\ n\geq1$, where $\lambda_n$ and $\xi_n$ are fundamental ’length’ constants and $G$ is the gravitational constant.
It comes naturally that for so-called *gravitational type fields* $U_{(2n)}\equiv G_{(n)},\ n\geq1$, the coupling has the form: $$-\frac{8\pi G}{c^2}\frac{(\lambda_{n})^{2n}}{c^{2n}}G_{(n)\mu}(x)j^{(2n)\mu}(x)\ ,$$ whereas for the so-called *electromagnetic type fields* $U_{(2n-1)}\equiv A_{(n)},\ n\geq1$, the coupling has the form: $$\mu_0\frac{(\xi_{n})^{2n-2}}{c^{2n-2}}A_{(n)\mu}(x)j^{(2n-1)\mu}(x)\ ,$$ where $A_{(n)\mu}$ has the dimension of $\mathrm{N.A^{-1}}$ ($\mathrm{N}$ is the Newton and $\mathrm{A}$ the Ampère), $\mu_0$ is the vacuum permeability ($\mu_0=4\pi\times 10^{-7}\ \mathrm{N.A^{-2}}$), and where $G_{(n)\mu}$ has the dimension of a length. The generalized currents for $n=1,2,3,..$ are constructed as follows: $$j^{(n)\nu}\equiv
\left\{\begin{array}{ll}
\rho_m(x)\frac{d^{n}x^\nu}{d\tau^{n}},\ \mathrm{if\ n\ is\ even} \\
\rho_e(x)\frac{d^{n}x^\nu}{d\tau^{n}},\ \mathrm{if\ n\ is\ odd}
\end{array}\right.$$ where $\rho_m(x)$ is the mass density and $\rho_e(x)$ the electric charge density. Similar to (\[gpe2\]), we construct an $(2n-1)$-order linear differential theory to relate the sources and the fields: $$\label{gpe2n}
O^{(n)}_\mu(\lambda_n) \epsilon_{(n)}^{\mu\nu}(x) = - \frac{8\pi G}{c^2}\frac{\lambda_n^{2n}}{c^{2n}} j^{(2n)\nu}(x),\ n\geq1\ ,$$ and $$\label{gpe2n}
Q^{(n)}_\mu(\xi_n) f_{(n)}^{\mu\nu}(x) = - \mu_0\frac{\xi_n^{2n-2}}{c^{2n-2}} j^{(2n-1)\nu}(x),\ n\geq1\ ,$$ where $O^{(n)}_\mu(\lambda_n)$ and $Q^{(n)}_\mu(\xi_n)$ are two $(2n-1)$-order differential operators and where $\varepsilon_{(n)}^{\mu\nu}(x)\equiv \partial^\mu G_{(n)}^\nu+\partial^\nu G_{(n)}^\mu$ and $f_{(n)}^{\mu\nu}(x)\equiv \partial^\mu A_{(n)}^\nu-\partial^\nu A_{(n)}^\mu$.
From those rules we could obtain similar wave equations to (\[PropaChampsPhys0\]) and (\[Tracen2\]) but with a higher order differential operator $(\lambda_n)^{2k}\underbrace{\Box\Box\cdots\Box}_{k\ \mathrm{times}},\ k=1,\cdots,n$. For example, for the $4$-field we can take $O^{(4)}_\mu(\lambda)=(\lambda^2\Box+1)\partial_\mu$ and then we get the wave equation for the trace of the tensor $\zeta_{\mu\nu}\equiv \partial_\mu K_\nu+\partial_\nu K_\mu$: $$\begin{aligned}
\label{Tracen4}
\left(\lambda^2\Box+1\right)\Box \zeta_{\mu}^\mu(x)=- \frac{8\pi G\lambda^4}{c^6} \partial_\mu j^{(4)\mu}(x)\ ,\end{aligned}$$ and then $\zeta_\mu^\mu$ is a massive scalar field.
Notice that for the electromagnetic type fields, the choice of the fields $A_{(n)\mu}$ (due to the antisymmetry of the fields $f_{(n)}^{\mu\nu}$) is not unique whereas for the gravitational type fields all of the components of the fields $G_{(n)\mu}$ are physical.
This project is at an early stage and the construction of these operators $O^{(n)}_\mu$ and $Q^{(n)}_\mu$ has to be understood, even for $n=2$.
Unitary fields
--------------
Physically, we can understand the *generalized vectorial fields theory* as perturbative corrections of the first order theory (i.e. $A_{(2)\mu}\equiv B_\mu(x)$ is a correction of the Minkowskian theory of Electromagnetism field $A_{(1)\mu}\equiv A_\mu$). Therefore, it is natural to unify the gravity type fields as well as the electromagnetic type fields. Then, we construct the dimensionless unification constants: $$\gamma_{jl}=\frac{\lambda_j}{\lambda_l},\ \theta_{jl}=\frac{\xi_j}{\xi_l},\ j,l=1,2,3,\cdots$$ where the constants $\lambda_n,\ n\geq1$ and $\xi_{n},\ n\geq1$, were introduced in section II.B.
For example, if we suppose that $A_\mu(x)=B_\mu(x)$, we get the coupling: $$\mu_0 A_{\mu}(x)\left( j^{(1)\mu}(x)+\frac{\xi^2}{c^2}j^{(3)\mu}(x)\right)$$ where we put $\xi_2=\xi$ (remind that $\alpha_1=1$ and $\alpha_3=\xi_2^2/c^2$). Phenomenologically, this means that for an electric circuit with an intensity of this type $I(t)=I_0 t^2/\tau^2$, where $\tau$ is a time constant, the third derivative of the electrons in the current is non-zero (this kinematic quantity is called the *Jerk*, see [@Jerk]) and so that the electromagnetic field would be modified by the *jerk current* $j^{(3)}$. We mention that in the generalized theory of Electrodynamics [@GeEM] the relation with the higher derivatives currents has not been suggested.
Similarly, we can construct the unified coupling for the even fields. for example, if we suppose that $G_{\mu}(x)=K_\mu(x)\equiv G_{(2)\mu}(x)$, we can reconstruct the unified coupling for $n=2,4$ in the following way: $$-\frac{8\pi G}{c^2}\frac{\lambda^{2}}{c^{2}}G_{\mu}(x)\left( j^{(2)\mu}(x)+\gamma^4\frac{\lambda^2}{c^2}j^{(4)\mu}(x)\right)$$ where we put $\lambda_1=\lambda,\ \gamma=\gamma_{21}$ and where we introduce the effective current $\widetilde{j}^{(2)\mu}=j^{(2)\mu}+\gamma^4\frac{\lambda^2}{c^2}j^{(4)\mu}$. So the effective deformation $\epsilon_{\mu\nu}$ to the Minkowski metric is also induced by the second derivative of the acceleration of the particles moving in the space.
Comments
========
- Microscopic Physics and generalized currents
The effect of gravitation at the microscopic scale is not yet well known. We wonder if the current $j^{(2)}$ of the acceleration of masses will be significant. There is no current proof that gravitation can be viewed as a metric field at this scale. It might also be a challenge to see whether the higher derivative fields play a role in particle physics. We gave the interpretation of the $2n$-fields (ex $G_\mu$ and $K_\mu$) and between the $(2n-1)$-fields (ex $A_\mu$ and $B_\mu$) as the generalisation of the gravitational and electromagnetic field for higher currents. Hence, formulating a perturbative quantum field theory including these fields is an open question. $\ $\
- Generalized kinematic model and special relativity theory
After introducing the higher derivative fields and the higher currents, we naturally wonder if it is possible to extend the Lorentzian kinematic theory to a more general inertial concept where the higher derivatives of the quadri vector position appear in the free Lagrangian. For example, following the discussion in section I, we propose the following Lagrangian: $$\label{Lolambda}
L_{\lambda;\alpha}(\dot{x},\ddot{x})=\frac{mc^2}{2}\left(\dot{x}_\mu\dot{x}^\mu + \alpha\lambda^2\ddot{x}_\mu\ddot{x}^\mu\right)\ ,$$ where $m$ is the mass of the particle, $\lambda$ is a universal constant and $\alpha=\pm1$. For this model the free motion is not determined by $\ddot{x}^\mu=0$ but by the equations: $$\label{EqLolambda}
\ddddot{x}^\mu=-\frac{\alpha}{\lambda^2}\ddot{x}^\mu$$ independently to the mass of the particle. We question whether the model (\[Lolambda\]) is consistent with the special theory of relativity. One problem concerns the simultaneity. Consider an observer in an inertial reference frame, i.e. $\ddot{x}_\mu(s)=0$. If we imagine that a photon is emitted from an accelerated massive particle (the equation of motion are given by (\[Lolambda\]) with the initial conditions $\dddot{x}_\mu(0)=0$ and $\ddot{x}_\mu(0)=a_\mu$), is the photon accelerated with respect to the inertial reference frame observing the motion of the particle or does its velocity remain constant and equal to $c$ ? Behind this question is a more fundamental problem of the equivalence between the accelerated reference frames. This requires further study. $\ $\
- Strain and stress tensor in General Relativity
We will discuss only the $n=2$-field model of the section II.A and we consider $\alpha_{n\geq3}=0$. In a future project, it will be interesting, as a first step, to formulate the $G_\mu$-field theory as a perturbative gravitational field arising from the metric field and to then to look at the possible effects for particle physics and/or cosmology.
We will give here the idea to construct the covariant strain/stress field theory. Let us consider the covariant derivative for a Riemannian metric space $\varepsilon_{\mu\nu}=D_\mu G_\nu+D_\nu G_\mu$. Now, we construct a stress tensor $$\label{Sigma}
\sigma_{\mu\nu}(x)=\rho_G c^2\varepsilon_{\mu\nu}(x)$$ where $\rho_G c^2=\frac{c^4}{8\pi G \lambda^2}$ is the density of energy constant, and analogous to the Young modulus for an isotropic medium [@MMC]. The stress tensor could be added to the Einstein field equations of gravity and then we get the relation: $$\label{Conserv}
D_{\mu}\sigma^{\mu\nu}(x)+D_\mu T^{\mu\nu}(x)=0$$ where $T_{\mu\nu}$ is the energy-momentum tensor in the Einstein field equation. This equation means that the total energy in the universe is conserved but that the “visible” energy can be accelerated. This variation of inertia is compensated by the divergence of the stress energy of the continuous medium. There is no contradiction with Einstein theory of gravitational fields and this gives a new perspective on the Mach principle revisiting the “absolute” acceleration concept as a natural motion in space-time deformed by the matter-energy contained therein. We refer the reader to the paper of Einstein on a related topic [@Einstein]. The relativistic theory of an Aether was discussed several times, see for e.g. [@Aether1], [@Aether2]. In this paper, our hypothesis is different and gives a relativistic theory of the deformation of continuous media (for which the geometry is still described by the metric field whereas the strain tensor is an additional field). Then, we could construct a more general stress tensor: $$\sigma_{\mu\nu}(x)=C_{\mu\nu\alpha\beta}(x)\varepsilon^{\alpha\beta}(x)$$ with the *elasticity tensor* $C_{\mu\nu\alpha\beta}(x)$. In fact, a consistent theory should give the equations for the elasticity tensor (which gives the mechanical property of the four dimensional continuous medium) and we think it might be related to the metric tensor as well as its derivatives (i.e. to the Ricci tensor).
[5]{}
M. Ostrogradsky, Mem. Acad. St. Petersburg **6** (4),385 (1850).
M. Borneas, On a generalization of the Lagrange Function, American Journal of Physics, **27** (4), pp. 265-267 (1959).
M. Borneas, Principle of the Action with Higher Derivatives, Phys.Rev **186** (5), 1299, (1969).
D. Anderson, Equivalent Lagrangians in generalized mechanics, J. Math. Phys. **14**, 934 (1973).
C. G. Adler, Why is mechanics based on acceleration?, Philosophy of Science, **47**, 146-152 (1980).
B. Podolsky, A Generalized Electrodynamics: Part I - Non-Quantum, Phys. Rev. **62**, 68-71 (1942).
W. M. Lai, D. Rubin, E. Krempl, “Introduction to Continuum Mechanics”, (Butterworth-Heinemann, Fourth edition, 2010).
S. H. Schot, “Jerk: The time rate of change of acceleration”, Am. J .Phys. **46**, 1090 (1978).
A. Einstein, “Ether and the Theory of Relativity” (1920), in Sidelights on Relativity (Methuen, London, 1922)
C. Eling, T. Jacobson, D. Mattingly, “Einstein Aether Theory”, in Deserfest, eds. J. Liu, K. Stelle, R. P. Woodard (World Scientific, 2006).
T. G. Zlosnik, P. G. Ferreira, G. D. Starkmann, Modifying gravity with the Aether: an alternative to Dark Matter, Phys. Rev. D **75**, 044017 (2007).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the last ten years, increasing attention has been devoted to Extended Theories of Gravity with the aim to understand several cosmological and astrophysical issues such as the today observed accelerated expansion of the universe and the presence of Dark Matter in self-gravitating structures. Some of these models assume modifications of General Relativity by adding higher order terms of curvature invariants like the Ricci scalar $R$, the Ricci tensor $R_{\mu\nu}$ and the Riemann tensors $R_{\mu\nu\lambda\sigma}$, or the presence of suitable scalar fields like the former Brans-Dicke theory. It is therefore natural to ask for black hole solutions in this context since, on the one hand, black holes signatures may be the test-bed to compare new models to the Einstein gravity; on the other hand, they may lead to rule out models which disagree with observations. Although black holes are one of the most striking predictions of General Relativity, they remain one of its least tested concepts. Electromagnetic observations allow indirectly to infer their existence, but direct evidences remains elusive. In the next decade, data coming from very long-baseline interferometry and gravitational wave detectors should allow to image and study black holes in detail. Such observations will test General Relativity in the non-linear and strong-field regimes where data are currently lacking. Testing strong-field features of General Relativity is of utmost importance to physics and astrophysics as a whole. This is because the black holes solutions, such as the Schwarzschild and Kerr metrics, enter several calculations, including accretion disk structure, gravitational lensing, cosmology and gravitational waves theory. These black hole solutions could indicate strong-field departures from General Relativity with deep implications for the still unknown fundamental theory of gravity. Beside the physical interest, black hole solutions represent a very active area for mathematical physics investigations. Here we review the problem of black holes in a particular class of Extended Theories of Gravity, the so called $f(R)$-gravity, discussing some resolution techniques, obtaining exact solutions and comparing results with standard General Relativity. Furthermore, we discuss the problems of hydrostatic equilibrium and stellar structure in the context of $f(R)$-gravity showing that new features could emerge. The observation of such features could both explain the physics of exotic self-gravitating objects and constitute a signature for Extended Theories of Gravity.'
author:
- |
Mariafelicia De Laurentis[^1] and Salvatore Capozziello[^2]\
[*Dipartimento di Scienze Fisiche, Universitá di Napoli “Federico II”,\
and\
INFN Sezione di Napoli,\
Compl. Univ. di Monte S. Angelo, Edificio G, Via Cinthia, I-80126, Napoli, Italy.*]{}
title: ' Black holes and stellar structures in $f(R)$-gravity'
---
Introduction
============
The issue to extend General Relativity (GR) has recently become dramatically urgent due to the missing matter problem at all astrophysical scales and the accelerating behavior of cosmic fluid, detected by Super Novae Ia used as standard candles. Up to now, no final answer on new particles has been given at fundamental level so Dark Energy and Dark Matter constitute a puzzle to be solved in order to achieve a self-consistent picture of the observed Universe. $f(R)$-gravity, where $f(R)$ is a generic function of the Ricci scalar $R$, comes into the game as a straightforward extension of GR where further geometrical degrees of freedom are considered instead of searching for new material ingredients [@f(R)-cosmo]. From an epistemological point of view, the action of gravity is not selected [*a priori*]{}, but it could be “reconstructed”, in principle, by matching consistently the observations [@Capozz2; @GRGrev; @FoP]. This approach can be adopted considering any function of the curvature invariants as $R_{\mu\nu}R^{\mu\nu}$, $R\Box R$ and so on. Several of these extended models reproduce Solar System tests so they are not in conflict with GR experimental results but actually extend them enclosing new features that could be, in principle, observed[@Capozziello8; @Capozziello9; @pogosian].
From a genuine mathematical point of view, Extended Theories of Gravity pose the problem to recover the well-established results of GR as the initial value problem [@cauchy], the stability of solutions and, in particular, the issue of finding out new solutions. As it is well known, beside cosmological solutions, spherically and axially symmetric solutions play a fundamental role in several astrophysical problems ranging from black holes to active galactic nuclei. Extended gravities, to be consistent with results of GR, should comprise solutions like Schwarzschild and Kerr ones but could present, in general, new solutions of physical interest. Due to this reason, methods to find out exact and approximate solutions are particularly relevant in order to check if observations can be framed in Extended Theories of Gravity [@noether].
Recently, the interest in spherically symmetric solutions of $f(R)$-gravity is growing up. In [@Multamaki], solutions in vacuum have been found considering relations among functions that define the spherical metric or imposing a costant Ricci curvatue scalar. The authors have reconstructed the form of some $f(R)$-models, discussing their physical relevance. In [@Multamaki1], the same authors have discussed static spherically symmetric solutions, in presence of perfect fluid matter, adopting the metric formalism. They have shown that a given matter distribution is not capable of globally determining the functional form of $f(R)$. Others authors have discussed in details the spherical symmetry of $f(R)$-gravity considering also the relations with the weak field limit. Exact solutions are obtained for constant Ricci curvature scalar and for Ricci scalar depending on the radial coordinate. In particular, it can be considered how to obtain results consistent with GR assuming the well-known post-Newtonian and post-Minkowskian limits as consistency checks [@arturo].
As we will discuss below, a general method to find out rotating black hole solutions can be achieved by performing a complex coordinate transformation on the spherical black hole metrics. Since the discovery of the Kerr solution [@pap:kerr], many attempts have been made to find a physically reasonable interior matter distribution that may be considered as its source. For a review on these approaches see [@pap:dmm; @pap:ak]. Though much progress has been made, results have been generally disappointing. As far as we know, nobody has obtained a physically satisfactory interior solution. This seems surprising given the success of matching internal spherically symmetric solutions to the Schwarzschild metric. The problem is that the loss of a degree of symmetry makes the derivation of analytic results much more difficult. Severe restrictions are placed on the interior metric by maintaining that it must be joined smoothly to the external axially symmetric metric. Further restrictions are placed on the interior solutions to ensure that they correspond to physical objects.
Furthermore since the axially symmetric metric has no radiation field associated with it, its source should be also non-radiating. This places even further constraints on the structure of the interior solution [@bk:dk]. Given the strenuous nature of these limiting conditions, it is not surprising to learn that no satisfactory solution to the problem of finding sources for the Kerr metric has been obtained. In general, the failure is due to internal structures whose physical properties are unknown. This shortcoming makes hard to find consistent boundary conditions.
Newman and Janis showed that it is possible to obtain rotating solutions (like the Kerr metric) by making an elementary complex transformation on the Schwarzschild solution [@pap:nj1]. This same method has been used to obtain a new stationary and axially symmetric solution known as the Kerr-Newman metric [@pap:nj2]. The Kerr-Newman space-time is associated to the exterior geometry of a rotating massive and charged black-hole. For a review on the Newman-Janis method to obtain both the Kerr and Kerr-Newman metrics see [@bk:ier; @turolla].
By means of very elegant mathematical arguments, Schiffer et al. [@pap:mms] have given a rigorous proof to show how the Kerr metric can be derived starting from a complex transformation on the Schwarzschild solution. We will not go into details of this demonstration, but point out that the proof relies on two main assumptions. The first is that metric belongs to the same algebraic class of the Kerr-Newman solution, namely the Kerr-Schild class [@pap:gcd]. The second assumption is that metric corresponds to an empty solution of the Einstein field equations [@pap:mms]. It is clear, by the generation of the Kerr-Newman metric, that all the components of the stress-energy tensor need to be non-zero for the Newman-Janis method to be successful. Such a transformation can be extended to $f(R)$-gravity as discussed in [@axially].
On the other hand, the strong gravity regime is another way to check the viability of these theories [@psaltis]. In general the formation and the evolution of stars can be considered suitable test-beds for Extended Theories of Gravity. Considering the case of $f(R)$-gravity, divergences stemming from the functional form of $f(R)$ may prevent the existence of relativistic stars in these theories [@briscese], but thanks to the chameleon mechanism, introduced by Khoury and Weltman [@weltman], the possible problems jeopardizing the existence of these objects may be avoided [@tsu]. Furthermore, there are also numerical solutions corresponding to static star configurations with strong gravitational fields [@babi] where the choice of the equation of state is crucial for the existence of solutions.
It is also important to stress that $f(R)$-gravity has interesting applications also in stellar astrophysics and could contribute to solve several puzzles related to observed peculiar objects (e.g. stars in the instability strips, protostars, etc. [@Cooney; @Hu]), structure and star formation [@poly; @Chang]. Furthermore some observed stellar systems are incompatible with the standard models of stellar structure. We refer to anomalous neutron stars, the so called “magnetars” [@mag] with masses larger than their expected Volkoff mass. It seems that, on particular length scales, the gravitational force is larger or smaller than the corresponding GR value. For example, a modification of the Hilbert-Einstein Lagrangian, consisting of $R^2$ terms, enables a major attraction while a $R_{\alpha\beta}R^{\alpha\beta}$ term gives a repulsive contribution [@stabile_2]. Understanding on which scales the modifications to GR are working or what is the weight of corrections to gravitational potential is a crucial point that could confirm or rule out these extended approaches to gravitational interaction.
This Chapter is organized as follow. In the Sec.\[due\], we introduce the $f(R)$-gravity action, the field equations and give some general remarks on spherical symmetry. In Sec \[tre\], a summary is given on the Noether Simmetry Approach [@noether]. This technique is extremely useful to find exact solutions. In particular, we find spherically symmetric black hole solutions for $f(R)$-gravity. In Sec \[quattro\], we review the Newman-Janis method to obtain rotating solutions starting from spherically symmetric ones. The resulting metric is written in terms of two arbitrary functions. A further suitable coordinate transformation allows to write the metric in the so called Boyer-Lindquist coordinates. Such a transformation makes the physical interpretation much clearer and reduces the amount of algebra required to calculate the metric properties. In Sec.\[cinque\], the Newman-Janis method is applied to a spherically symmetric exact solution, previously derived by the Noether Symmetry, and an axially symmetric exact solution is obtained. This result shows that the Newman-Janis method works also in $f(R)$-gravity. Physical applications of the result are also discussed.
In Sec.\[sei\], we review the classical hydrostatic problem for stellar structures. In Sec. \[sette\] we derive the Newtonian limit of $f(R)$-gravity obtaining the modified Poisson equation. The modified Lané-Emden equation is obtained in Sec. \[otto\] and its structure is compared with respect to the standard one. In Sec.\[nove\], we show how analytical solutions of standard Lané-Emden equation can be compared with those perturbatively obtained from $f(R)$-gravity. In order to apply the above results, in Section \[dieci\] the classical theory of gravitational collapse for dust-dominated systems is summarized. In Section \[undici\], we discuss the weak field limit of $f(R)$-gravity obtaining corrections to the standard Newtonian potential that can be figured out as two Newtonian potentials contributing to the dynamics. In Section \[dodici\] we recover the dispersion relation and Jeans mass limit[@jeans]. Some self-gravitating dust system are discussed in this approach. The difference between GR and $f(R)$-gravity are put in evidence, in particular the Jeans mass profiles with respect to the temperature. We report a catalogue of observed molecular clouds in order to compare the classical Jeans mass to the $f(R)$-one. Finally, in Section \[discuss\], we discuss the results and draw conclusions.
Spherical symmetry in $f(R)$-gravity {#due}
=====================================
Let us start by discussing exact solutions in $f(R)$-gravity with spherical symmetry. As we will see, a crucial role is played by the relation between the metric potentials and the Ricci scalar that can be regarded as a constraint assuming the form of a Bernoulli equation.
Let us consider an analytic function $f(R)$ of the Ricci scalar $R$ in four dimensions. The variational principle for this action is:
$$\label{fRaction}
\delta\int
d^4x\sqrt{-g}\biggl[f(R)+\mathcal{X}\mathcal{L}_m\biggr]\,=\,0$$
where ${\displaystyle \mathcal{X}=\frac{8\pi G}{c^4}}$, $\mathcal{L}_m$ is the standard matter Lagrangian and $g$ is the determinant of the metric[^3].
By varying with respect to the metric, we obtain the field equations [^4]
$$\label{HOEQ}
\left\{\begin{array}{ll}H_{\mu\nu}=f'(R)R_{\mu\nu}-\frac{1}{2}f(R)g_{\mu\nu}-f'(R)_{;\mu\nu}+g_{\mu\nu}\Box
f'(R)\,=\,\mathcal{X}T_{\mu\nu}\\\\H\,=\,g^{\rho\sigma}H_{\rho\sigma}=3\Box
f'(R)+f'(R)R-2f(R)\,=\,\mathcal{X}T\end{array}\right.$$
where $T_{\mu\nu}\,=\,\displaystyle\frac{-2}{\sqrt{-g}}\frac{\delta(\sqrt{-g}\mathcal{L}_m)}{\delta
g^{\mu\nu}}$ is the energy-momentum tensor of standard fluid matter and the second equation is the trace. The most general spherically symmetric solution can be written as follows:
$$\label{me0}
ds^2\,=\,m_1(t',r')dt'^2+m_2(t',r')dr'^2+m_3(t',r')dt'dr'+m_4(t',r')d\Omega\,,$$
where $m_i$ are functions of the radius $r'$ and of the time $t'$. $d\Omega$ is the solid angle. We can consider a coordinate transformation that maps the metric (\[me0\]) in a new one where the off-diagonal term vanishes and $m_4(t',r')\,=\,-r^2$, that is[^5]:
$$\label{me}
ds^2\,=\,g_{tt}(t,r)dt^2-g_{rr}(t,r)dr^2-r^2d\Omega\,.$$
This expression can be considered, without loss of generality, as the most general definition of a spherically symmetric metric compatible with a pseudo-Riemannian manifold without torsion. Actually, by inserting this metric into the field Eqs. (\[HOEQ\]), one obtains:
$$\label{fe4}
\left\{\begin{array}{ll}f'(R)R_{\mu\nu}-\frac{1}{2}f(R)g_{\mu\nu}+\mathcal{H}_{\mu\nu}\,=\,\mathcal{X}T_{\mu\nu}\\\\
f'(R)R-2f(R)+\mathcal{H}\,=\,\mathcal{X}T\end{array}\right.$$
where the two quantities $\mathcal{H}_{\mu\nu}$ and $\mathcal{H}$ read: $$\begin{aligned}
\label{highterms1}
\mathcal{H}_{\mu\nu}\,&=&\,-f''(R)\biggl\{R_{,\mu\nu}-\Gamma^t_{\mu\nu}R_{,t}-\Gamma^r_{\mu\nu}R_{,r}-
g_{\mu\nu}\biggl[\biggl({g^{tt}}_{,t}+\nonumber\\&&+g^{tt}
\left(\ln\sqrt{-g}\right)_{,t}\biggr)R_{,t}+\biggl({g^{rr}}_{,r}+g^{rr}\left(\ln\sqrt{-g}\right)_{,r}\biggr)R_{,r}+\nonumber\\&&+g^{tt}R_{,tt}
+g^{rr}R_{,rr}\biggr]\biggr\}-f'''(R)\biggl[R_{,\mu}R_{,\nu}-g_{\mu\nu}\biggl(g^{tt}{R_{,t}}^2+g^{rr}
{R_{,r}}^2\biggr)\biggr]\end{aligned}$$ $$\begin{aligned}
\label{highterms2}
\mathcal{H}\,&=&\,g^{\sigma\tau}\mathcal{H}_{\sigma\tau}\,=\,3f''(R)\biggl[\biggl({g^{tt}}_{,t}+g^{tt}
\left(\ln\sqrt{-g}\right)_{,t}\biggr)R_{,t}+\nonumber\\&&+\biggl({g^{rr}}_{,r}+g^{rr}\left(\ln\sqrt{-g}\right)_{,r}\biggr)R_{,r}+g^{tt}R_{,tt}
+g^{rr}R_{,rr}\biggr]+
3f'''(R)\biggl[g^{tt}{R_{,t}}^2+g^{rr}{R_{,r}}^2\biggr]\,.\nonumber\\\end{aligned}$$ Our task is now to find out exact spherically symmetric solutions.
In the case of time-independent metric, i.e., $g_{tt}\,=\,a(r)$ and $g_{rr}\,=\,b(r)$, the Ricci scalar can be recast as a Bernoulli equation of index two with respect to the metric potential $b(r)$ (see [@arturo] for details):
$$\begin{aligned}
\label{eqric}
&& b'(r)+\biggl\{\frac{r^2a'(r)^2-4a(r)^2-2ra(r)[2a(r)'+ra(r)'']}{ra(r)[4a(r)
+ra'(r)]}\biggr\}b(r)+\nonumber\\&&+\biggl\{\frac{2a(r)}{r}\biggl[\frac{2+r^2R(r)}{4a(r)+ra'(r)}\biggr]\biggr\}b(r)^2\,=\,0\,.\end{aligned}$$
where $R\,=\,R(r)$ is the Ricci scalar. A general solution of (\[eqric\]) is:
$$\label{gensol}
b(r)\,=\,\frac{\exp[-\int dr\,h(r)]}{K+\int dr\,l(r)\,\exp[-\int
dr\,h(r)]}\,,$$
where $K$ is an integration constant while $h(r)$ and $l(r)$ are two functions that, according to Eq.(\[eqric\]), define the coefficients of the quadratic and the linear term with respect to $b(r)$ [@bernoulli]. We can fix $l(r)\,=\,0$; this choice allows to find out solutions with a Ricci scalar scaling as ${\displaystyle
-\frac{2}{r^2}}$ in term of the radial coordinate. On the other hand, it is not possible to have $h(r)\,=\,0$ since, otherwise, we get imaginary solutions. A particular consideration deserves the limit $r\rightarrow\infty$. In order to achieve a gravitational potential $b(r)$ with the correct Minkowski limit, both $h(r)$ and $l(r)$ have to go to zero at infinity, provided that the quantity $r^2R(r)$ turns out to be constant: this result implies $b'(r)=0$, and, finally, also the metric potential $b(r)$ has a correct Minkowski limit.
In general, if we ask for the asymptotic flatness of the metric as a feature of the theory, the Ricci scalar has to evolve to infinity as $r^{-n}$ with $n\geqslant 2$. Formally, it has to be:
$$\label{condricc}
\lim_{r\rightarrow\infty}r^2R(r)\,=\,r^{-n}\,,$$
with $n\in\mathbb{N}$. Any other behavior of the Ricci scalar could affect the requirement to achieve a correct asymptotic flatness.
The case of constant curvature is equivalent to GR with a cosmological constant and the solution is time independent. This result is well known (see, for example, [@barrottew]) but we report, for the sake of completeness, some considerations related with it in order to deal with more general cases where a radial dependence for the Ricci scalar is supposed. If the scalar curvature is constant ($R\,=\,R_0$), field Eqs.(\[fe4\]), being $\mathcal{H}_{\mu\nu}\,=\,0$, reduce to:
$$\label{fe2}
\left\{\begin{array}{ll}f'_0R_{\mu\nu}-\frac{1}{2}f_0g_{\mu\nu}\,=\,\mathcal{X}T_{\mu\nu}\\\\
f'_0R_0-2f_0\,=\,\mathcal{X}T\end{array}\right.$$
where $f(R_0)=f_0$, $f'(R_0)=f'_0$. A general solution, when one considers a stress-energy tensor of perfect-fluid $T_{\mu\nu}\,=\,(\rho+p)u_\mu u_\nu-pg_{\mu\nu}$, is
$$ds^2\,=\,\biggl(1+\frac{k_1}{r}+\frac{q\mathcal{X}\rho-\lambda}{3}r^2\biggr)dt^2-\frac{dr^2}{1+\frac{k_1}{r}+
\frac{q\mathcal{X}\rho-\lambda}{3}r^2}-r^2d\Omega\,.$$
when $p\,=\,-\rho$, $\lambda=-\frac{f_0}{2f'_0}$ and $q^{-1}=f'_0$. This result means that any $f(R)$-model, in the case of constant curvature, exhibits solutions with de Sitter-like behavior. This is one of the reasons why the dark energy issue can be addressed using these theories [@f(R)-cosmo].
If $f(R)$ is analytic, it is possible to write the series: $$\label{f}
f(R)\,=\,\Lambda+\Psi_0R+\Psi(R)\,,$$ where $\Psi_0$ is a coupling constant, $\Lambda$ plays the role of the cosmological constant and $\Psi(R)$ is a generic analytic function of $R$ satisfying the condition
$$\label{psi}
\lim_{R\rightarrow 0}R^{-2}\Psi(R)\,=\,\Psi_1\,,$$
where $\Psi_1$ is a constant. If we neglect the cosmological constant $\Lambda$ and $\Psi_0$ is set to zero, we obtain a new class of theories which, in the limit $R\rightarrow{0}$, does not reproduce GR (from Eq.(\[psi\]), we have $\lim_{R\rightarrow 0}
f(R)\sim R^2$). In such a case, analyzing the whole set of Eqs.(\[fe2\]), one can observe that both zero and constant $\neq
0$ curvature solutions are possible. In particular, if $R\,=\,R_0\,=\,0$ field equations are solved for any form of gravitational potential entering the spherically symmetric background, provided that the Bernoulli Eq. (\[eqric\]), relating these functions, is fulfilled for the particular case $R(r)=0$. The solutions are thus defined by the relation
$$\label{gensol0}
b(r)\,=\,\frac{\exp[-\int
dr\,h(r)]}{K+4\int\frac{dr\,a(r)\,\exp[-\int
dr\,h(r)]}{r[a(r)+ra'(r)]}}\,,$$
being $g_{tt}(t,r)=b(r)$ from Eq.(\[me\]). In [@arturo], some examples of $f(R)$-models admitting solutions with constant$\neq 0$ or null scalar curvature are discussed.
The Noether Symmetry Approach {#tre}
==============================
Besides spherically symmetric solutions with constant curvature scalar, also solutions with the Ricci scalar depending on radial coordinate $r$ are possible in $f(R)$-gravity [@arturo]. Furthermore, spherically symmetric solutions can be achieved starting from a point-like $f(R)$-Lagrangian [@noether]. Such a Lagrangian can be obtained by imposing the spherical symmetry directly in the action (\[fRaction\]). As a consequence, the infinite number of degrees of freedom of the original field theory will be reduced to a finite number. The technique is based on the choice of a suitable Lagrange multiplier defined by assuming the Ricci scalar, argument of the function $f(R)$ in spherical symmetry.
Starting from the above considerations, a static spherically symmetric metric can be expressed as
$$\label{me2}
{ds}^2=A(r){dt}^2-B(r){dr}^2-M(r)d\Omega\,,$$
and then the point-like $f(R)$ Lagrangian[^6] is
$$\begin{aligned}
\label{lag2}
\mathcal{L}&=&-\frac{A^{1/2}f'}{2MB^{1/2}}{M'}^2-\frac{f'}{A^{1/2}B^{1/2}}A'M'-\frac{Mf''}{A^{1/2}B^{1/2}}A'R'+\nonumber\\&&-\frac{2A^{1/2}f''}{B^{1/2}}R'M'-A^{1/2}B^{1/2}[(2+MR)f'-Mf]\,,\end{aligned}$$
which is canonical since only the configuration variables and their first order derivatives with respect to the radial coordinate $r$ are present. Details of calculations are in [@noether]. Eq. (\[lag2\]) can be recast in a more compact form introducing the matrix representation: $$\label{la}
\mathcal{L}={\underline{q}'}^t\hat{T}\underline{q}'+V$$ where $\underline{q}=(A,B,M,R)$ and $\underline{q}'=(A',B',M',R')$ are the generalized positions and velocities associated to $\mathcal{L}$. It is easy to check the complete analogy between the field equation approach and point-like Lagrangian approach [@noether].
In order to find out solutions for the Lagrangian (\[lag2\]), we can search for symmetries related to cyclic variables and then reduce dynamics. This approach allows, in principle, to select $f(R)$-gravity models compatible with spherical symmetry. As a general remark, the Noether Theorem states that conserved quantities are related to the existence of cyclic variables into dynamics [@arnold; @marmo; @morandi].
It is worth noticing that the Hessian determinant of Eq. (\[lag2\]), ${\displaystyle \left|\left|\frac{\partial^2\mathcal{L}}{\partial
q'_i\partial q'_j}\right|\right|}$, is zero. This result clearly depends on the absence of the generalized velocity $B'$ into the point-like Lagrangian. As matter of fact, using a point-like Lagrangian approach implies that the metric variable $B$ does not contributes to dynamics, but the equation of motion for $B$ has to be considered as a further constraint equation. Then the Lagrangian (\[lag2\]) has three degrees of freedom and not four, as one should expect [*a priori*]{}.
Now, since the equation of motion describing the evolution of the metric potential $B$ does not depend on its derivative, it can be explicitly solved in term of $B$ as a function of the other coordinates: $$\label{eqb}
B=\frac{2M^2f''A'R'+2Mf'A'M'+4AMf''M'R'+Af'M'^2}{2AM[(2+MR)f'-Mf]}\,.$$ By inserting Eq.(\[eqb\]) into the Lagrangian (\[lag2\]), we obtain a non-vanishing Hessian matrix removing the singular dynamics. The new Lagrangian reads[^7] $$\label{}\mathcal{L}^*= {\bf L}^{1/2}$$ with $$\begin{aligned}
\label{lag3}\nonumber
{\bf L}=\underline{q'}^t\hat{{\bf
L}}\underline{q'}=\frac{[(2+MR)f'-fM]}{M}[2M^2f''A'R'
+2MM'(f'A'+2Af''R') +Af'M'^2]\,.\end{aligned}$$
If one assumes the spherical symmetry, the role of the [*affine parameter*]{} is played by the coordinate radius $r$. In this case, the configuration space is given by $\mathcal{Q}=\{A, M, R\}$ and the tangent space by $\mathcal{TQ}=\{A, A', M, M', R, R'\}$. On the other hand, according to the Noether Theorem, the existence of a symmetry for dynamics described by Lagrangian (\[lag2\]) implies a constant of motion. Let us apply the Lie derivative to the (\[lag2\]), we have[^8]: $$\label{}
L_{\mathbf{X}}{\bf L}\,=\,\underline{\alpha}\cdot\nabla_{q}{\bf L
}+\underline{\alpha}'\cdot\nabla_{q'}{\bf L}
=\underline{q}'^t\biggl[\underline{\alpha}\cdot\nabla_{q}\hat{{\bf
L }}+ 2\biggl(\nabla_{q}\alpha\biggr)^t\hat{{\bf L
}}\biggr]\underline{q}'\,,$$ that vanishes if the functions ${\underline{\alpha}}$ satisfy the following system $$\label{sys}
\underline{\alpha}\cdot\nabla_{q}\hat{{\bf L}}
+2(\nabla_{q}{\underline{\alpha}})^t\hat{{\bf L
}}\,=\,0\,\longrightarrow\ \ \ \ \alpha_{i}\frac{\partial
\hat{{\bf L}}_{km}}{\partial
q_{i}}+2\frac{\partial\alpha_{i}}{\partial q_{k}}\hat{{\bf L
}}_{im}=0\,.$$ Solving the system (\[sys\]) means to find out the functions $\alpha_{i}$ which assign the Noether vector [@arnold; @marmo]. However the system (\[sys\]) implicitly depends on the form of $f(R)$ and then, by solving it, we get also $f(R)$-models compatible with spherical symmetry. On the other hand, by choosing the $f(R)$-form, we can explicitly solve (\[sys\]). As an example, one finds that the system (\[sys\]) is satisfied if we choose $$\label{solsy}f(R)\,=\,f_0 R^s\,,\ \ \ \ \ \underline{\alpha}=(\alpha_1,\alpha_2,\alpha_3)=
\biggl((3-2s)kA,\ -kM,\ kR\biggr)\,,$$ with $s$ a real number, $k$ an integration constant and $f_0$ a dimensional coupling constant[^9]. This means that, for any $f(R)=R^s$, exists, at least, a Noether symmetry and a related constant of motion $\Sigma_{0}$: $$\begin{aligned}
\label{cm}\nonumber
\Sigma_{0}\,=\,\underline{\alpha}\cdot\nabla_{q'}{\bf L
}=2
skMR^{2s-3}[2s+(s-1)MR][(s-2)RA'-(2s^2-3s+1)AR']\,.\end{aligned}$$ A physical interpretation of $\Sigma_{0}$ is possible if one gives an interpretation of this quantity in GR, that means for $f(R)=R$ and $s=1$. In other words, the above procedure has to be applied to the Lagrangian of GR. We obtain the solution $$\label{solsygr}\underline{\alpha}_{GR}=(-kA,\
kM)\,.$$ The functions $A$ and $M$ give the Schwarzschild solution, and then the constant of motion acquires the standard form $$\label{cmgr}\Sigma_{0}= \frac{2GM}{c^2}\,.$$ In other words, in the case of Einstein gravity, the Noether symmetry gives, as a conserved quantity, the Schwarzschild radius or the mass of the gravitating system. This result can be assumed as a consistency check.
In the general case, $f(R)=R^s$, the Lagrangian (\[lag2\]) becomes $$\begin{aligned}
\label{}\nonumber {\bf L}&=&\frac{sR^{2s-3}[2s+(s-1)MR]}{M}[2(s-1)M^2A'R'+2MRM'A'+\nonumber\\&& 4(s-1)AMM'R'+ARM'^2]\,,\end{aligned}$$ and the expression (\[eqb\]) for $B$ is $$\label{}B=\frac{s[2(s-1)M^2A'R'+2MRM'A'+4(s-1)AMM'R'+ARM'^2]}{2AMR[2s+(s-1)MR]}$$ As it can be easily checked, GR is recovered for $s=1$.
Using the constant of motion (\[cm\]), we solve in term of $A$ and obtain $$\label{}A=R^{\frac{2s^2-3s+1}{s-2}}\biggl\{k_1+\Sigma_{0}\int\frac{R^{\frac{4s^2-9s+5}{2-s}}dr}{2ks(s-2)M[2s+(s-1)MR]}\biggr\}$$ for $s\neq2$ and $k_1$ an integration constant. For $s\,=\,2$, one finds $$\label{}A=-\frac{\Sigma_{0}}{12kr^2(4+r^2R)RR'}\,.$$ These relations allow to find out general black hole solutions for the field equations giving the dependence of the Ricci scalar on the radial coordinate $r$. For example, a solution is found for $$\label{}
s=5/4\,,\ \ \ \ M=r^2\,,\ \ \ \ R= 5 r^{-2}\,,$$ obtaining the spherically symmetric space-time $$\label{sol_noe_2}ds^2=(\alpha+\beta
r)dt^2-\frac{1}{2}\frac{\beta r}{\alpha+\beta r}
dr^2-r^2d\Omega\,,$$ where $\alpha$ is a combination of $\Sigma_0$ and $k$ and $\beta=k_1$. In principle, the same procedure can be worked out any time Noether symmetries are identified. Our task is now to show how, from a spherically symmetric solution, one can generate an axially symmetric solution adopting the Newman-Janis procedure that works in GR. In general, the approach is not immediately straightforward since, as soon as $f(R)\neq R$, we are dealing with fourth-order field equations which have, in principle, different existence theorems and boundary conditions. However, the existence of the Noether symmetry guarantees the consistency of the chosen $f(R)$-model with the field equations.
Axial symmetry derived from spherical symmetry {#quattro}
==============================================
We want to show now how it is possible to obtain an rotating solution starting from a spherically symmetric one adopting the method developed by Newman and Janis in GR. Such an algorithm can be applied to a static spherically symmetric metric considered as a“seed” metric. Let us recast the spherically symmetric metric (\[me\]) in the form
$$ds^2 = e^{2\phi(r)}dt^2 - e^{2\lambda(r)}dr^2 - r^2d\Omega,
\label{eqn:ssm}$$
with $g_{tt}(t,r)\,=\,e^{2\phi(r)}$ and $g_{rr}(t,r)\,=\,e^{2\lambda(r)}$. Such a form is suitable for the considerations below. Following Newman and Janis, Eq. (\[eqn:ssm\]) can be written in the so called Eddington–Finkelstein coordinates $(u,r,\theta,\phi)$, i.e. the $g_{rr}$ component is eliminated by a change of coordinates and a cross term is introduced [@gravitation]. Specifically this is achieved by defining the time coordinate as $dt = du + F(r)dr$ and setting $F(r)= \pm
e^{\lambda(r)-\phi(r)}$. Once such a transformation is performed, the metric (\[eqn:ssm\]) becomes
$$\label{nullelemn}
ds^2 = e^{2\phi(r)}du^2 \pm 2e^{\lambda(r)+\phi(r)}dudr -
r^2d\Omega.$$
The surface $u\,=\,$ costant is a light cone starting from the origin $r\,=\,0$. The metric tensor for the line element (\[nullelemn\]) in null-coordinates is
$$\label{metrictensorcontro}
g^{\mu\nu} = \left(
\begin{array}{cccc}
0 & \pm e^{-\lambda(r)-\phi(r)} & 0 & 0 \\
\pm e^{-\lambda(r)-\phi(r)} & -e^{-2\lambda(r)} & 0 & 0 \\
0 & 0 & -1/r^2 & 0 \\
0 & 0 & 0 & -1/(r^2\sin^2{\theta})
\end{array} \right).$$
The matrix (\[metrictensorcontro\]) can be written in terms of a null tetrad as
$$\label{eq:gmet}
g^{\mu\nu} = l^\mu n^\nu + l^\nu n^\mu-m^\mu\bar{m}^\nu -
m^\nu\bar{m}^\mu,$$
where $l^\mu$, $n^\mu$, $m^\mu$ and $\bar{m}^\mu$ are the vectors satisfying the conditions
$$l_\mu l^\mu\,=\,m_\mu m^\mu\,=\,n_\mu n^\mu\,=\,0,\,\,\,\,\,
l_\mu n^\mu\,=\,-m_\mu\bar{m}^\mu\,=\,1, \,\,\,\,\,l_\mu
m^\mu\,=n_\mu m^\mu\,=\,0\,.$$
The bar indicates the complex conjugation. At any point in space, the tetrad can be chosen in the following manner: $l^\mu$ is the outward null vector tangent to the cone, $n^\mu$ is the inward null vector pointing toward the origin, and $m^\mu$ and $\bar{m}^\mu$ are the vectors tangent to the two-dimensional sphere defined by constant $r$ and $u$. For the spacetime (\[metrictensorcontro\]), the tetrad null vectors can be
$$\left\{\begin{array}{ll}l^\mu\,=\,\delta^\mu_1\\\\n^\mu\,=\,-\frac12
e^{-2\lambda(r)}\delta^\mu_1+
e^{-\lambda(r)-\phi(r)}\delta^\mu_0\\\\
m^\mu\,=\,\frac{1}{\sqrt{2}r}(\delta^\mu_2
+\frac{i}{\sin{\theta}}\delta^\mu_3)\\\\
\bar{m}^\mu\,=\,\frac{1}{\sqrt{2}r}(\delta^\mu_2
-\frac{i}{\sin{\theta}}\delta^\mu_3)\end{array}\right.$$
Now we need to extend the set of coordinates $x^\mu\,=\,(u,r,\theta,\phi)$ replacing the real radial coordinate by a complex variable. Then the tetrad null vectors become [^10]
$$\label{tetradvectors}
\left\{\begin{array}{ll}l^\mu\,=\,\delta^\mu_1\\\\n^\mu\,=\,-\frac12
e^{-2\lambda(r,\bar{r})}\delta^\mu_1+
e^{-\lambda(r,\bar{r})-\phi(r,\bar{r})}\delta^\mu_0\\\\
m^\mu\,=\,\frac{1}{\sqrt{2}\bar{r}}(\delta^\mu_2
+\frac{i}{\sin{\theta}}\delta^\mu_3)\\\\
\bar{m}^\mu\,=\,\frac{1}{\sqrt{2}r}(\delta^\mu_2
-\frac{i}{\sin{\theta}}\delta^\mu_3)\end{array}\right.$$
A new metric is obtained by making a complex coordinates transformation
$$\label{transfo}x^\mu\rightarrow\tilde{x}^\mu=x^\mu+iy^\mu(x^\sigma)\,,$$
where $y^\mu(x^\sigma)$ are analityc functions of the real coordinates $x^\sigma$, and simultaneously let the null tetrad vectors $Z^\mu_a\,=\,(l^\mu,n^\mu,m^\mu,\bar{m}^\mu)$, with $a\,=\,1,2,3,4$, undergo the transformation
$$Z^\mu_a\rightarrow \tilde{Z}^\mu_a(\tilde{x}^\sigma,\bar{\tilde{x}}^\sigma)\,=\,Z^\rho_a\frac{\partial\tilde{x}
^\mu}{\partial x^\rho}.$$
Obviously, one has to recover the old tetrads and metric as soon as $\tilde{x}^\sigma\,=\,\bar{\tilde{x}}^\sigma$. In summary, the effect of the “*tilde transformation*” (\[transfo\]) is to generate a new metric whose components are (real) functions of complex variables, that is
$$g_{\mu\nu}\rightarrow \tilde{g}_{\mu\nu}\,:\,\tilde{\mathbf{x}}\times\tilde{\mathbf{x}}\mapsto
\mathbb{R}$$
with
$$\tilde{Z}^\mu_a(\tilde{x}^\sigma,\bar{\tilde{x}}^\sigma)|_{\mathbf{x}=\tilde{\mathbf{x}}}
=Z^\mu_a(x^\sigma).$$
For our aims, we can make the choice
$$\label{transfo_1}
\tilde{x}^\mu\,=\,x^\mu+ia(\delta^\mu_1-\delta^\mu_0)\cos\theta\rightarrow
\left\{\begin{array}{ll}\tilde{u}\,=\,u+ia\cos\theta\\\\\tilde{r}\,=\,r-ia\cos\theta\\\\
\tilde{\theta}\,=\,\theta\\\\
\tilde{\phi}\,=\,\phi\\\\
\end{array}\right.$$
where $a$ is constant and the tetrad null vectors (\[tetradvectors\]), if we choose $\tilde{r}\,=\,\bar{\tilde{r}}$, become
$$\label{tetradvectors_2}
\left\{\begin{array}{ll}\tilde{l}^\mu\,=\,\delta^\mu_1\\\\\tilde{n}^\mu\,=\,-\frac12
e^{-2\lambda(\tilde{r},\theta)}\delta^\mu_1+
e^{-\lambda(\tilde{r},\theta)-\phi(\tilde{r},\theta)}\delta^\mu_0\\\\
\tilde{m}^\mu\,=\,\frac{1}{\sqrt{2}(\tilde{r}-ia\cos\theta)}\biggl[ia(\delta^\mu_0-\delta^\mu_1)\sin\theta+\delta^\mu_2
+\frac{i}{\sin{\theta}}\delta^\mu_3\biggr]\\\\
\bar{\tilde{m}}^\mu\,=\,\frac{1}{\sqrt{2}(\tilde{r}+ia\cos\theta)}\biggl[-ia(\delta^\mu_0-\delta^\mu_1)\sin\theta+\delta^\mu_2
-\frac{i}{\sin{\theta}}\delta^\mu_3\biggr]\end{array}\right.$$
From the transformed null tetrad vectors, a new metric is recovered using (\[eq:gmet\]). For the null tetrad vectors given by (\[tetradvectors\_2\]) and the transformation given by (\[transfo\_1\]), the new metric, with coordinates $\tilde{x}^\mu\,=\,(\tilde{u},\tilde{r},\theta,\phi)$, is
$$\tilde{g}^{\mu\nu} = \left(
\begin{array}{cccc} \label{eqn:newmetric}
-\frac{a^2 \sin^2{\theta}}{\Sigma^2} &
e^{-\lambda(\tilde{r},\theta)-\phi(\tilde{r},\theta)} +
\frac{a^2\sin^2{\theta}}{\Sigma^2} & 0 & -\frac{a}{\Sigma^2} \\
. & - e^{-2\lambda(\tilde{r},\theta)} -
\frac{a^2\sin^2{\theta}}{\Sigma^2}
& 0 & \frac{a}{\Sigma^2} \\
. & . & -\frac{1}{\Sigma^2} & 0 \\
. & . & . & -\frac{1}{\Sigma^2\sin^2{\theta}} \\
\end{array} \right)$$
where $\Sigma = \sqrt{\tilde{r}^2 + a^2\cos^2{\theta}}$. In the covariant form, the metric (\[eqn:newmetric\]) is
$$\begin{aligned}
\label{eqn:coform}
%\tilde{g}_{\mu\nu}
ds^2&=& e^{2\phi(\tilde{r},\theta)}d\tilde{u}^2
+2 e^{\lambda(\tilde{r},\theta)}d\tilde{u}d\tilde{r}
+2ae^{\phi(\tilde{r},\theta)}[e^{\lambda(\tilde{r},\theta)}-
e^{\phi(\tilde{r},\theta)}] \sin^2{\theta}d\tilde{u}d\phi+\nonumber\\&&
-2a e^{\phi(\tilde{r},\theta)+\lambda(\tilde{r},\theta)}\sin^2{\theta}d\tilde{r}^2d\phi-Sigma^2d\theta^2+\nonumber\\&&
- [\Sigma^2 +
a^2\sin^2{\theta}e^{\phi(\tilde{r},\theta)}(2e^{\lambda(\tilde{r},\theta)}-
e^{\phi(\tilde{r},\theta)})]\sin^2{\theta}d\phi^2\end{aligned}$$
Since the metric is symmetric, the dots in the matrix are used to indicate $g^{\mu\nu} = g^{\nu\mu}$. The form of this metric gives the general result of the Newman-Janis algorithm starting from any spherically symmetric “seed” metric.
The metric given in Eq. (\[eqn:coform\]) can be simplified by a further gauge transformation so that the only off-diagonal component is $g_{\phi t}$. This procedure makes it easier to compare with the standard Boyer-Lindquist form of the Kerr metric [@gravitation] and to interpret physical properties such as the frame dragging. The coordinates $\tilde{u}$ and $\phi$ can be redefined in such a way that the metric in the new coordinate system has the properties described above. More explicitly, if we define the coordinates in the following way
$$d\tilde{u}\,=
\,dt+g(\tilde{r})d\tilde{r}\,\,\,\,\,\text{and}\,\,\,\,\,d\phi\,=\,d\phi+h(\tilde{r})d\tilde{r}$$
where
$$\left\{\begin{array}{ll}g(\tilde{r})=-\frac{e^{\lambda(\tilde{r},\theta)}
(\Sigma^2+a^2\sin^2{\theta}e^{\lambda(\tilde{r},\theta)+\phi(\tilde{r},\theta)})}
{e^{\phi(\tilde{r},\theta)}(\Sigma^2+a^2\sin^2{\theta}e^{2\lambda(\tilde{r},\theta)})}\\\\h(\tilde{r})
=-\frac{ae^{2\lambda(\tilde{r},\theta)}}{\Sigma^2+a^2\sin^2{\theta}e^{2\lambda(\tilde{r},\theta)}}\end{array}\right.$$
after some algebraic manipulations, one finds that, in $(t,\tilde{r},\theta,\phi)$ coordinates system, the metric (\[eqn:coform\]) becomes
$$\begin{aligned}
\label{eq:blform}
ds^2&=&e^{2\phi(\tilde{r},\theta)}dt^2+ a
e^{\phi(\tilde{r},\theta)}[e^{\lambda(\tilde{r},\theta)}-
e^{\phi(\tilde{r},\theta)}] \sin^2{\theta} dtd\phi-\nonumber\\&&
\frac{\Sigma^2}{(\Sigma^2 e^{-2\lambda(\tilde{r},\theta)} +
a^2\sin^2{\theta})}d\tilde{{r}}^2
-\Sigma^2d\theta^2+\nonumber\\&&
-[\Sigma^2 +
a^2\sin^2{\theta}e^{\phi(\tilde{r},\theta)}(2e^{\lambda(\tilde{r},\theta)}-
e^{\phi(\tilde{r},\theta)})]\sin^2{\theta}d\phi^2\end{aligned}$$
This metric represents the complete family of metrics that may be obtained by performing the Newman-Janis algorithm on any static spherically symmetric “seed” metric, written in Boyer-Lindquist type coordinates. The validity of these transformations requires the condition $\Sigma^2+a^2\sin^2\theta
e^{2\lambda(\tilde{r},\theta)}\neq 0$, where $e^{2\lambda(\tilde{r},\theta)}> 0$. Our task is now to show that such an approach can be used to derive axially symmetric solutions also in $f(R)$-gravity that, possibly, can be regarded as black hole solutions..
Axially symmetric solutions in $f(R)$-gravity {#cinque}
=============================================
Starting from the above spherically symmetric solution (\[sol\_noe\_2\]), the metric tensor, written in the Eddington–Finkelstein coordinates $(u,r,\theta,\phi)$ of the form (\[metrictensorcontro\]) is
$$\label{metrictensorcontro_noe}
g^{\mu\nu} = \left(
\begin{array}{cccc}
0 & \sqrt{\frac{2}{\beta r}} & 0 & 0 \\
. & -2-\frac{2\alpha}{\beta r}& 0 & 0 \\
. & . & -1/r^2 & 0 \\
. & . & . & -1/(r^2\sin^2{\theta})
\end{array} \right).$$
The complex tetrad null vectors (\[tetradvectors\]) are now
$$\label{tetradvectors_noe}
\left\{\begin{array}{ll}l^\mu\,=\,\delta^\mu_1\\\\n^\mu\,=\,-
\biggl[1+\frac{\alpha}{\beta}\biggl(\frac{1}{\bar{r}}+\frac{1}{r}\biggr)\biggr]\delta^\mu_1+
\sqrt{\frac{2}{\beta}}\frac{1}{\sqrt[4]{\bar{r}r}}\delta^\mu_0\\\\
m^\mu\,=\,\frac{1}{\sqrt{2}\bar{r}}(\delta^\mu_2
+\frac{i}{\sin{\theta}}\delta^\mu_3)\end{array}\,.\right.$$
By computing the complex coordinates transformation (\[transfo\_1\]), the tetrad null vectors become
$$\label{tetradvectors_noe_1}
\left\{\begin{array}{ll}\tilde{l}^\mu\,=\,\delta^\mu_1\\\\\tilde{n}^\mu\,=\,-
\biggl[1+\frac{\alpha}{\beta}\frac{\text{Re}\{\tilde{r}\}}{\Sigma^2}\biggr]\delta^\mu_1+
\sqrt{\frac{2}{\beta}}\frac{1}{\sqrt{\Sigma}}\delta^\mu_0\\\\
\tilde{m}^\mu\,=\,\frac{1}{\sqrt{2}(\tilde{r}+ia\cos\theta)}\biggl[ia(\delta^\mu_0-\delta^\mu_1)\sin\theta+\delta^\mu_2
+\frac{i}{\sin{\theta}}\delta^\mu_3\biggr]\end{array}\right.$$
Now by performing the same procedure as in previous section, we derive an axially symmetric metric of the form (\[eq:blform\]) but starting from the spherically symmetric metric (\[sol\_noe\_2\]), that is
$$\begin{aligned}
\label{rotating}
ds^2&=&\frac{r(\alpha+\beta r)+a^2\beta\cos^2\theta}{\Sigma}du^2+2 \frac{a(-2\alpha r-2\beta\Sigma^2+\sqrt{2\beta}\Sigma^{3/2})\sin^2\theta}{2\Sigma}dud\phi+\nonumber\\&&
-\frac{\beta\Sigma^2}{2\alpha r+\beta(a^2+r^2+\Sigma^2)}dr^2-\Sigma^2d\theta^2+\nonumber\\&&
-\biggl[\Sigma^2 -\frac{a^2(\alpha
r+\beta\Sigma^2-\sqrt{2\beta}\Sigma^{3/2})\sin^2\theta}{\Sigma}\biggr]
\sin^2\theta d\phi^2\end{aligned}$$
It is worth noticing that the condition $a=0$ immediately gives the metric (\[sol\_noe\_2\]). This is nothing else but an example: the method is general and can be extended to any spherically symmetric solution derived in $f(R)$-gravity.
Physical applications: geodesics and orbits {#cinque.1}
-------------------------------------------
Let us discuss now possible physical applications of the above results. We will take into account a freely falling particle moving in the space-time described by the metric (\[rotating\]). For our aims, we make explicit use of the Hamiltonian formalism. Given a metric $g_{\mu\nu}$, the motion along the geodesics is described by the Lagrangian $${\cal L}(x^{\mu},\dot{x}^\mu)=\dfrac{1}{2}g_{\mu\nu}\dot{x}^{\mu}\dot{x^\nu} \ ,$$ where the overdot stands for derivative with respect to an affine parameter $\lambda$ used to parametrize the curve. The Hamiltonian description is achieved by considering the canonical momenta and the Hamiltonian function $$p_{\mu} = \frac{\de{\cal L}}{\de \dot{x}^{\mu}}=g^{\mu\nu}p_\mu p_\nu \ ,
\;\;\;\;\;\;\;\;\;\;
{\cal H }= p_{\mu}\dot{x}^{\mu}-{\cal L} \ ,$$ that results ${\displaystyle
{\cal H}= \frac{1}{2}p_\mu p_\nu g^{\mu\nu} }$. The advantage of the Hamiltonian formalism with respect to the Lagrangian one is that the resulting equations of motion do not contain any sign ambiguity coming from turning points in the orbits (see, for example, [@chandra]) . The Hamiltonian results explicitly independent of time and it is ${\displaystyle
{\cal H}= -\frac{1}{2}m^2 ,}$ where the rest mass $m$ is a constant ($m=0$ for photons). The geodesic equations are $$\frac{dx^{\mu}}{d\lambda}=\frac{\partial\cal{ H}}{\partial p_{\mu}}=g^{\mu\nu}p_{\nu}=p^{\mu},
\label{2a}$$ $$\frac{dp_{\mu}}{d\lambda}=-\frac{\partial\cal{H}}{\partial x{\mu}}=-\frac{1}{2}\frac{\partial g^{\alpha\beta}}{\partial x^{\mu}}p_{\alpha}p_{\beta}=g^{\gamma\beta}\Gamma^{\alpha}_{\mu\gamma}p_{\alpha}p_{\beta}.
\label{2b}$$ In addition, since the Hamiltonian is independent of the affine parameter $\lambda$, one can directly use the coordinate time as integration parameter. The problem is so reduced to solve six equations of motion. Using the above definitions, it is easy to achieve the reduced Hamiltonian (now linear in the momenta)
$$H=-p_{0}=\left[\frac{p_{i}g^{0i}}{g^{00}}+\left[\left(\frac{p_{i}g^{0i}}{g^{00}}\right)^{2}-\frac{m^{2}+p_{i}p_{j}g^{ij}}{g^{00}}\right]^{1/2}\right]\label{eq:e}$$
with the equations of motion
$$\frac{dx^{i}}{dt}=\frac{\partial H}{\partial p_{i}}\ ,\;\;\;\;\;\;\;\;\;\;\;\;
\frac{dp_{i}}{dt}=-\frac{\partial H}{\partial x^{i}}\ ,\label{eq:4b}$$
that give the orbits. The method can be applied to the above solution (\[rotating\]) by which the elements of the inverse metric can be easily obtained:
$$\begin{aligned}
g^{tt}&=&\left\{4 \Sigma ^2 \left[\Sigma ^2-\frac{a^2 \sin ^2\theta \left(r
\alpha -\sqrt{2} \sqrt{\beta } \Sigma ^{3/2}+\beta \Sigma
^2\right)}{\Sigma }\right]\right\}\times
\nonumber\\&&
\left\{a^2 \sin ^2\theta \left(2 r \alpha
-\sqrt{2} \sqrt{\beta } \Sigma ^{3/2}+2 \beta \Sigma ^2\right)^2
+4 \Sigma \left[a^2 \beta \cos ^2\theta+r (r \beta +\alpha )\right]\right. \times\nonumber\\&&
\left. \left( \Sigma ^2-\frac{a^2 \sin ^2\theta \left(r \alpha -\sqrt{2} \sqrt{\beta }\Sigma ^{3/2}+\beta \Sigma ^2\right)}{\Sigma }\right)\right \}^{-1}\,,
\nonumber\\
g^{rr}&=&-\frac{\beta \left(a^2+r^2+\Sigma ^2\right)+2 r \alpha }{\beta \Sigma
^2}\,,\nonumber\\
g^{\theta\theta}&=&-\frac{1}{\Sigma ^2}\,,\nonumber\\
g^{t\phi}&=& \left\{2 a \Sigma \left(-2 r \alpha +\sqrt{2} \sqrt{\beta } \Sigma
^{3/2}-2 \beta \Sigma ^2\right)\right\}\times\nonumber\\&&
\left\{a^2 \sin ^2\theta \left(2 r \alpha
-\sqrt{2} \sqrt{\beta } \Sigma ^{3/2}+2 \beta \Sigma ^2\right)^2
+4 \Sigma \left[a^2 \beta \cos ^2\theta+r (r \beta +\alpha )\right]\right. \times\nonumber\\&&
\left. \left( \Sigma ^2-\frac{a^2 \sin ^2\theta \left(r \alpha -\sqrt{2} \sqrt{\beta }\Sigma ^{3/2}+\beta \Sigma ^2\right)}{\Sigma }\right)\right \}^{-1}\,,
\nonumber\\
g^{\phi\phi}&=&-\left\{4 \Sigma \csc ^2\theta \left[a^2 \beta \cos ^2\theta +r (r
\beta +\alpha )\right]\right\}\times\nonumber\\&&
\left\{a^2 \sin ^2\theta \left(2 r \alpha
-\sqrt{2} \sqrt{\beta } \Sigma ^{3/2}+2 \beta \Sigma ^2\right)^2
+4 \Sigma \left[a^2 \beta \cos ^2\theta+r (r \beta +\alpha )\right]\right. \times\nonumber\\&&
\left. \left( \Sigma ^2-\frac{a^2 \sin ^2\theta \left(r \alpha -\sqrt{2} \sqrt{\beta }\Sigma ^{3/2}+\beta \Sigma ^2\right)}{\Sigma }\right)\right \}^{-1}\,,
\end{aligned}$$
and the null ones $$g^{tr}=g^{t\theta}= g^{r\theta}= g^{r\phi}= g^{\theta\phi}=0\,.$$
Let us consider the equatorial plane, [*i.e.*]{} $\theta=\frac{\pi}{2}$, $\dot{\theta}=0$, and assume $\alpha=1$ and $\beta=2$. The reduced Hamiltonian $H(r,\theta, \phi,p_{r},p_{\theta},p_{\phi};t)=H$ can be written as
$$\begin{aligned}
H&=&\frac{2 a p_{\phi} \left(-2 r^3+r^2-1\right)}{a^2 \left(-2 (r-1) r^2-1\right)+r^5}+
\left\{\left[ \left(4 a^2 p_{\phi}^{2} \left(-2 r^3+r^2-1\right)^2\right.\right.\right.+\nonumber\\ &&
\left.\left.\left.\left.-a^2 \left(-2 (r-1) r^2-1\right)-r^5\right)\left(a^2 \left(r^2 (r (2 r-3) (2 r+1)+6)-2\right)+\right.\right.\right.\right.\nonumber\\&&\left.\left.\left.+(2 r+1) r^4\right)\times\right.\right.\nonumber\\&&\left.\left. \left(-\frac{p_{\phi} (2 r+1)}{a^2 \left(r^2 (r (2 r-3) (2 r+1)+6)-2\right)+(2 r+1) r^4}+\right.\right.\right.\nonumber\\&&\left.\left.\left.-\frac{p_{r} \left(a^2+r^2+r\right)+p_{\theta}}{r^4}-p_{r}+1\right)
\right]\right\}^\frac{1}{2} .\nonumber\\
\label{eq:8}
\end{aligned}$$
It is independent of $\phi$ ([*i.e.*]{} we are considering an azimuthally symmetric spacetime), and then the conjugate momentum $p_{\phi}$ is an integral of motion. From Eqs. (\[eq:4b\]), one can derive the coupled equations for $\{r,\theta,\phi,p_{r}$, $p_{\theta}\}$ and integrate them numerically (the expressions are very cumbersome and will not be reported here). To this goal, we have to specify the initial value of the position-momentum vector in the phase space. A Runge-Kutta method can be used to solve the differential equations. In Fig \[1\], the relative trajectories are sketched.
[|c|c|]{}
Hydrostatic equilibrium of stellar structures {#sei}
=============================================
We have shown that it is possible to derive black hole spherically and axially symmetric solutions in $f(R)$-gravity. The physics of black holes contains a lot of unsolved problems. One of them is the question on the nature of the black holes as well as the nature of l stellar objects presenting anomalous behaviors (e.g. the magnetars) that do not obey the standard dynamics of (relativistic and non-relativistic) stellar structures. For this purpose, it is necessary to understand the hydrostatic equilibrium of such objects, that is one of fundamental properties (or one of the basic assumptions) of self-gravitating systems. Here we start by describing the standard hydrostatic equilibrium assuming Newtonian (i.e. the GR weak field limit) and then we compare these results with the equilibrium derived from the weak field limit of $f(R)$-gravity.
The condition of hydrostatic equilibrium for stellar structures in Newtonian dynamics is achieved by considering the equation
$$\label{19.1}
\frac{dp}{dr}\,=\,\frac{d\Phi}{dr}\rho\,,$$
where $p$ is the pressure, $-\Phi$ is the gravitational potential, and $\rho$ is the density [@kippe]. Together with the above equation, the Poisson equation
$$\label{19.2}
\frac{1}{r^2}\frac{d}{dr}\left(r^2\frac{d\Phi}{dr}\right)\,=\,-4\pi G\rho\,,$$
gives the gravitational potential as solution for a given matter density $\rho$. Since we are taking into account only static and stationary situations, here we consider only time-independent solutions [^11]. In general, the temperature $\tau$ appears in Eqs. (\[19.1\]) and (\[19.2\]) the density satisfies an equation of state of the form $\rho\,=\,\rho(p,\tau)$. In any case, we assume that there exists a polytropic relation between $p$ and $\rho$ of the form
$$\label{19.3}
p\,=\,K\rho^\gamma\,,$$
where $K$ and $\gamma$ are constant. Note that $\Phi\,>\,0$ in the interior of the model since we define the gravitational potential as $-\Phi$. The polytropic constant $K$ is fixed and can be obtained as a combination of fundamental constants. However there are several realistic cases where $K$ is not fixed and another equation for its evolution is needed. The constant $\gamma$ is the [*polytropic exponent* ]{}. Inserting the polytropic equation of state into Eq. (\[19.1\]), we obtain $$\label{19.6}
\frac{d\Phi}{dr}\,=\,\gamma K \rho^{\gamma-2}\frac{d\rho}{dr}\,.$$ For $\gamma\neq1$, the above equation can be integrated giving
$$\begin{aligned}
\label{densita}
\frac{\gamma K}{\gamma-1}\rho^{\gamma-1}\,=\,\Phi\,\,\,\,\rightarrow\,\,\,\,\rho\,=\,\biggl[\frac{\gamma-1}{\gamma K}\biggr]^{\frac{1}{\gamma-1}}\Phi^{\frac{1}{\gamma-1}}\,\doteq\,A_n\Phi^n\end{aligned}$$
where we have chosen the integration constant to give $\Phi=0$ at surface $(\rho=0$). The constant $n$ is called the [*polytropic index*]{} and is defined as $n=\frac{1}{\gamma-1}$. Inserting the relation (\[densita\]) into the Poisson equation, we obtain a differential equation for the gravitational potential $$\label{19.8}
\frac{d^2\Phi}{dr^2}+\frac{2}{r}\frac{d\Phi}{dr}\,=\,-4\pi G A_n\Phi^n\,.$$ Let us define now the dimensionless variables
$$\begin{aligned}
\label{trans1}
\left\{\begin{array}{ll}
z\,=\,|\mathbf{x}|\sqrt{\frac{\mathcal{X}A_n\Phi_c^{n-1}}{2}}\\\\
w(z)\,=\,\frac{\Phi}{\Phi_c}\,=\,(\frac{\rho}{\rho_c})^\frac{1}{n}
\end{array}\right.\end{aligned}$$
where the subscript $c$ refers to the center of the star and the relation between $\rho$ and $\Phi$ is given by Eq. (\[densita\]). At the center $(r\,=\,0)$, we have $z\,=\,0$, $\Phi\,=\,\Phi_c$, $\rho\,=\,\rho_c$ and therefore $w\,=\,1$. Then Eq. (\[19.8\]) can be written
$$\begin{aligned}
\label{LE}
\frac{d^2w}{dz^2}+\frac{2}{z}\frac{dw}{dz}+w^n\,=\,0\end{aligned}$$
This is the standard [*Lane-Embden equation*]{} describing the hydrostatic equilibrium of stellar structures in the Newtonian theory [@kippe]. Now we want to compare this standard result with the one coming from $f(R)$-gravity.
The Newtonian limit of $f(R)$-gravity {#sette}
=====================================
In order to achieve the Newtonian limit of the theory the metric tensor $g_{\mu\nu}$ have to be approximated as follows
$$\begin{aligned}
\label{metric_tensor_PPN}
g_{\mu\nu}\,\sim\,\begin{pmatrix}
1-2\,\Phi(t,\mathbf{x})+\mathcal{O}(4)& \mathcal{O}(3) \\
\\
\mathcal{O}(3)& -\delta_{ij}+\mathcal{O}(2)\end{pmatrix}\end{aligned}$$
where $\mathcal{O}(n)$ (with $n\,=$ integer) denotes the order of the expansion (see [@mio] for details). The set of coordinates[^12] adopted is $x^\mu\,=\,(t,x^1,x^2,x^3)$. The Ricci scalar formally becomes
$$\begin{aligned}
R\,\sim\,R^{(2)}(t,\mathbf{x})+\mathcal{O}(4)\end{aligned}$$
The $n$-th derivative of Ricci function can be developed as
$$\begin{aligned}
f^{n}(R)\,\sim\,f^{n}(R^{(2)}+\mathcal{O}(4))\,\sim\,f^{n}(0)+f^{n+1}(0)R^{(2)}+\mathcal{O}(4)\end{aligned}$$
here $R^{(n)}$ denotes a quantity of order $\mathcal{O}(n)$. From lowest order of field equations, we have $f(0)\,=\,0$ which trivially follows from the above assumption (\[metric\_tensor\_PPN\]) for the metric. This means that the space-time is asymptotically Minkowskian and we are discarding a cosmological constant term in this analysis[^13]. Field eqautaions at $\mathcal{O}(2)$-order, that is at Newtonian level, are
$$\begin{aligned}
\label{PPN-field-equation-general-theory-fR-O2}
\left\{\begin{array}{ll}
R^{(2)}_{tt}-\frac{R^{(2)}}{2}-f''(0)\triangle
R^{(2)}\,=\,\mathcal{X}\,T^{(0)}_{tt}
\\\\
-3f''(0)\triangle
R^{(2)}-R^{(2)}\,=\,\mathcal{X}\,T^{(0)}
\end{array}\right.\end{aligned}$$
where $\triangle$ is the Laplacian in the flat space, $R^{(2)}_{tt}\,=\,-\triangle\Phi(t,\mathbf{x})$ and, for the sake of simplicity, we set $f'(0)\,=\,1$. We recall that the energy-momentum tensor for a perfect fluid is
$$\begin{aligned}
T_{\mu\nu}\,=\,(\epsilon+p)\,u_{\mu} u_{\nu}-p\,g_{\mu\nu}\end{aligned}$$
where $p$ is the pressure and $\epsilon$ is the energy density. Being the pressure contribution negligible in the field equations in Newtonian approximation, we have
$$\begin{aligned}
\label{HOEQ}
\left\{\begin{array}{ll}
\triangle\Phi+\frac{R^{(2)}}{2}+f''(0)\triangle R^{(2)}\,= \,-\mathcal{X}\rho
\\\\
3f''(0)\triangle R^{(2)}+R^{(2)}\,=\,-\mathcal{X}\rho
\end{array}\right.\end{aligned}$$
where $\rho$ is now the mass density[^14]. We note that for $f''(0)\,=\,0$ we have the standard Poisson equation: $\triangle\Phi\,=\,-4\pi G\rho$. This means that as soon as the second derivative of $f(R)$ is different from zero, deviations from the Newtonian limit of GR emerge.
The gravitational potential $-\Phi$, solution of Eqs. (\[HOEQ\]), has in general a Yukawa-like behavior depending on a characteristic length on which it evolves [@mio]. Then as it is evident the Gauss theorem is not valid[^15] since the force law is not $\propto|\mathbf{x}|^{-2}$. The equivalence between a spherically symmetric distribution and point-like distribution is not valid and how the matter is distributed in the space is very important [@mio; @stabile; @cqg].
Besides the Birkhoff theorem results modified at Newtonian level: the solution can be only factorized by a space-depending function and an arbitrary time-depending function [@mio]. Furthermore the correction to the gravitational potential is depending on the only first two derivatives of $f(R)$ in $R\,=\,0$. This means that different analytical theories, from the third derivative perturbation terms on, admit the same Newtonian limit [@mio; @stabile].
Eqs. (\[HOEQ\]) can be considered the *modified Poisson equation* for $f(R)$-gravity. They do not depend on gauge condition choice [@cqg]. We know that $$\begin{aligned}
R^{(2)}&\simeq \frac{1}{2}\nabla^2 g^{(2)}_{00}- \frac{1}{2}\nabla^2g^{(2)}_{ii}\,.\label{3}\end{aligned}$$ Inserting in the above result the $g_{\mu\nu}$ approximations we obtain $$\begin{aligned}
{R^{(2)}} \simeq {\nabla ^2}(\Phi - \Psi ).\end{aligned}$$ Finally, we obtain the field equations $$\begin{aligned}
\label{EQP}
&& \nabla^2\Phi+\nabla^2\Psi -2 f''(0) \nabla^4 \Phi + 2 f''(0) \nabla^4 \Psi=2\mathcal{X} \rho \\
\nonumber \\
&& \nabla^2\Phi-\nabla^2\Psi +3 f''(0)\nabla^4\Phi-3 f''(0) \nabla^4 \Psi=-\mathcal{X} \rho \label{EQP2}\,.\end{aligned}$$ By eliminating the higher-order terms, the standard Poisson equation is recovered. This last step, will be useful to calculate the Jeans instability.
Stellar Hydrostatic Equilibrium in $f(R)$-gravity {#otto}
=================================================
From the Bianchi identity, satisfied by the field equations, we have
$$\begin{aligned}
\label{equidr}
{T^{\mu\nu}}_{;\mu}\,=\,0\,\,\,\,\rightarrow\,\,\,\,\frac{\partial p}{\partial x^k}\,=\,-\frac{1}{2}(p+\epsilon)\frac{\partial \ln g_{tt}}{\partial x^k}\end{aligned}$$
If the dependence on the temperature $\tau$ is negligible, *i.e.* $\rho\,=\,\rho(p)$, this relation can be introduced into Eqs. (\[HOEQ\]), which become a system of three equations for $p$, $\Phi$ and $R^{(2)}$ and can be solved without the other structure equations.
Let us suppose that matter satisfies still a polytropic equation $p\,=\,K\,\rho^\gamma$. If we introduce Eq.(\[densita\]) into Eqs. (\[HOEQ\]) we obtain an integro-differential equation for the gravitational potential $-\Phi$, that is
$$\begin{aligned}
\label{deltafi}
\triangle\Phi(\mathbf{x})+\frac{2\mathcal{X}A_n}{3}\Phi(\mathbf{x})^n\,=\,-\frac{m^2\mathcal{X}A_n}{6} \int d^3\mathbf{x}'\mathcal{G}(\mathbf{x},\mathbf{x}')\Phi(\mathbf{x}')^n\end{aligned}$$
where ${\displaystyle \mathcal{G}(\mathbf{x},\mathbf{x}')\,=\,-\frac{1}{4\pi}\frac{e^{-m|\mathbf{x}-\mathbf{x}'|}}{|\mathbf{x}-\mathbf{x}'|}}$ is the Green function of the field operator $\triangle_\mathbf{x}-m^2$ for systems with spherical symmetry and ${\displaystyle m^2\,=\,-\frac{1}{3f''(0)}}$ (for details see [@stabile; @cqg]). The integro-differential nature of Eq.(\[deltafi\]) is the proof of the non-viability of Gauss theorem for $f(R)$-gravity. Adopting again the dimensionless variables
$$\begin{aligned}
\label{trans}
\left\{\begin{array}{ll}
z\,=\,\frac{|\mathbf{x}|}{\xi_0}\\\\
w(z)\,=\,\frac{\Phi}{\Phi_c}
\end{array}\right.\end{aligned}$$
where
$$\begin{aligned}
\label{charpar}
\xi_0\,\doteq\,\sqrt{\frac{3}{2\mathcal{X}A_n\Phi_c^{n-1}}}\end{aligned}$$
is a characteristic length linked to stellar radius $\xi$, Eq. (\[deltafi\]) becomes
$$\begin{aligned}
\label{LEmod}
\frac{d^2w(z)}{dz^2}+\frac{2}{z}\frac{d w(z)}{dz}+w(z)^n\,=\,\frac{m\xi_0}{8}\frac{1}{z}\int_0^{\xi/\xi_0}
dz'\,z'\,\biggl\{e^{-m\xi_0|z-z'|}-e^{-m\xi_0|z+z'|}\biggr\}\,w(z')^n\nonumber\\\end{aligned}$$
which is the *modified Lané-Emden equation* deduced from $f(R)$-gravity. Clearly the particular $f(R)$-model is specified by the parameters $m$ and $\xi_0$. If $m\,\rightarrow\,\infty$ (*i.e.* $f(R)\,\rightarrow\,R$), Eq. (\[LEmod\]) becomes Eq. (\[LE\]). We are only interested in solutions of Eq. (\[LEmod\]) that are finite at the center, that is for $z\,=\,0$. Since the center must be an equilibrium point, the gravitational acceleration $|\mathbf{g}|\,=\,-d\Phi/dr\,\propto\,dw/dz$ must vanish for $w'(0)\,=\,0$. Let us assume we have solutions $w(z)$ of Eq.(\[LEmod\]) that fulfill the boundary conditions $w(0)\,=\,1$ and $w(\xi/\xi_0)\,=\,0$; then according to the choice (\[trans\]), the radial distribution of density is given by
$$\begin{aligned}
\rho(|\mathbf{x}|)\,=\,\rho_cw^n\,,\,\,\,\,\,\,\,\,\rho_c\,=\,A_n{\Phi_c}^n\end{aligned}$$
and the pressure by
$$\begin{aligned}
p(|\mathbf{x}|)\,=\,p_cw^{n+1}\,,\,\,\,\,\,\,\,\,p_c\,=\,K{\rho_c}^\gamma\end{aligned}$$
For $\gamma\,=\,1$ (or $n\,=\,\infty$) the integro-differential Eq. (\[LEmod\]) is not correct. This means that the theory does not contain the case of isothermal sphere of ideal gas. In this case, the polytropic relation is $p\,=\,K\,\rho$. Putting this relation into Eq.(\[equidr\]) we have $$\begin{aligned}
\label{densitaiso}
\frac{\Phi}{K}\,=\,\ln\rho-\ln\rho_c\,\,\,\,\rightarrow\,\,\,\,\rho\,=\,\rho_c\,e^{\Phi/K}\end{aligned}$$ where the constant of integration is chosen in such a way that the gravitational potential is zero at the center. If we introduce Eq.(\[densitaiso\]) into Eqs. (\[HOEQ\]), we have
$$\begin{aligned}
\label{deltafiiso}
\triangle\Phi(\mathbf{x})+\frac{2\mathcal{X}\rho_c}{3}e^{\Phi(\mathbf{x})/K}\,=\,-\frac{m^2\mathcal{X}\rho_c}{6} \int d^3\mathbf{x}'\mathcal{G}(\mathbf{x},\mathbf{x}')e^{\Phi(\mathbf{x}')/K}\end{aligned}$$
Assuming the dimensionless variables $z\,=\,\frac{|\mathbf{x}|}{\xi_1}$ and $w(z)\,=\,\frac{\Phi}{K}$ where $\xi_1\,\doteq\,\sqrt{\frac{3K}{2\mathcal{X}\rho_c}}$, Eq. (\[deltafiiso\]) becomes
$$\begin{aligned}
\label{LEmodiso}
\frac{d^2w(z)}{dz^2}+\frac{2}{z}\frac{d w(z)}{dz}+e^{w(z)}\,=\,\frac{m\xi_1}{8}\frac{1}{z}\int_0^{\xi/\xi_1}
dz'\,z'\,\biggl\{e^{-m\xi_1|z-z'|}-e^{-m\xi_1|z+z'|}\biggr\}\,e^{w(z')}\end{aligned}$$
which is the *modified “isothermal” Lané-Emden equation* derived $f(R)$-gravity.
Solutions of the standard and modified Lané-Emden Equations {#nove}
============================================================
The task is now to solve the modified Lané-Emden equation and compare its solutions to those of standard Newtonian theory. Only for three values of $n$, the solutions of Eq.(\[LE\]) have analytical expressions [@kippe]
$$\begin{aligned}
\label{LEsol}
&&n\,=\,0\,\,\,\,\rightarrow\,\,\,\,w^{(0)}_{GR}(z)\,=\,1-\frac{z^2}{6}\nonumber\\
&&n\,=\,1\,\,\,\,\rightarrow\,\,\,\,w^{(1)}_{GR}(z)\,=\,\frac{\sin z}{z}\\
&&n\,=\,5\,\,\,\,\rightarrow\,\,\,\,w^{(5)}_{GR}(z)\,=\,\frac{1}{\sqrt{1+\frac{z^2}{3}}}\nonumber\end{aligned}$$
We label these solution with $_{GR}$ since they agree with the Newtonian limit of GR. The surface of the polytrope of index $n$ is defined by the value $z\,=\,z^{(n)}$, where $\rho\,=\,0$ and thus $w\,=\,0$. For $n\,=\,0$ and $n\,=\,1$ the surface is reached for a finite value of $z^{(n)}$. The case $n\,=\,5$ yields a model of infinite radius. It can be shown that for $n\,<\,5$ the radius of polytropic models is finite; for $n\,>\,5$ they have infinite radius. From Eqs.(\[LEsol\]) one finds $z^{(0)}_{GR}\,=\,\sqrt{6}$, $z^{(1)}_{GR}\,=\,\pi$ and $z^{(5)}_{GR}\,=\,\infty$. A general property of the solutions is that $z^{(n)}$ grows monotonically with the polytropic index $n$. In Fig. \[fig\] we show the behavior of solutions $w^{(n)}_{GR}$ for $n\,=\,0,\,1,\,5$. Apart from the three cases where analytic solutions are known, the classical Lané-Emden Eq. (\[LE\]) has to be be solved numerically, considering with the expression
$$\begin{aligned}
\label{taydev}
w^{(n)}_{GR}(z)\,=\,\sum_{i\,=\,0}^\infty a^{(n)}_iz^i\end{aligned}$$
for the neighborhood of the center. Inserting Eq.(\[taydev\]) into Eq. (\[LE\]) and by comparing coefficients one finds, at lowest orders, a classification of solutions by the index $n$, that is
$$\begin{aligned}
\label{gensolGR}
w^{(n)}_{GR}(z)\,=\,1-\frac{z^2}{6}+\frac{n}{120}z^4+\dots\end{aligned}$$
The case $\gamma\,=\,5/3$ and $n\,=\,3/2$ is the non-relativistic limit while the case $\gamma\,=\,4/3$ and $n\,=\,3$ is the relativistic limit of a completely degenerate gas.
Also for modified Lané-Emden Eq. (\[LEmod\]), we have an exact solution for $n\,=\,0$. In fact, it is straightforward to find out
$$\begin{aligned}
\label{LEmodsol0}
w^{(0)}_{_{f(R)}}(z)\,=\,1-\frac{z^2}{8}+\frac{(1+m\xi)e^{-m\xi}}{4m^2{\xi_0}^2}\biggl[1-\frac{\sinh m\xi_0 z}{m\xi_0 z}\biggr]\end{aligned}$$
where the boundary conditions $w(0)\,=\,1$ and $w'(0)\,=\,0$ are satisfied. A comment on the GR limit (that is $f(R)\rightarrow R$) of solution (\[LEmodsol0\]) is necessary. In fact when we perform the limit $m\,\rightarrow\,\infty$, we do not recover exactly $w^{(0)}_{GR}(z)$. The difference is in the definition of quantity $\xi_0$. In $f(R)$-gravity we have the definition (\[charpar\]) while in GR it is ${\displaystyle \xi_0\,=\,\sqrt{\frac{2}{\mathcal{X}A_n\Phi_c^{n-1}}}}$, since in the first equation of (\[HOEQ\]), when we perform $f(R)\rightarrow R$, we have to eliminate the trace equation condition. In general, this means that the Newtonian limit and the Eddington parameterization of different relativistic theories of gravity cannot coincide with those of GR (see [@eddington] for further details on this point).
The point $z_{_{f(R)}}^{(0)}$ is calculated by imposing $w^{(0)}_{_{f(R)}}(z_{_{f(R)}}^{(0)})\,=\,0$ and by considering the Taylor expansion
$$\frac{\sinh m\xi_0z}{m\xi_0z}\,\sim\,1+\frac{1}{6}(m\xi_0z)^2+\mathcal{O}(m\xi_0z)^4$$
we obtain ${\displaystyle z_{_{f(R)}}^{(0)}\,=\,\frac{2\sqrt{6}}{\sqrt{3+(1+m\xi)e^{-m\xi}}}}$. Since the stellar radius $\xi$ is given by definition $\xi\,=\,\xi_0\,z_{_{f(R)}}^{(0)}$, we obtain the constraint
$$\begin{aligned}
\label{radius_constraint}
\xi\,=\,\sqrt{\frac{3\Phi_c}{2\pi G}}\frac{1}{\sqrt{1+\frac{1+m\xi}{3}e^{-m\xi}}}\end{aligned}$$
By solving numerically the constraint[^16] Eq.(\[radius\_constraint\]), we find the modified expression of the radius $\xi$. If $m\,\rightarrow\,\infty$ we have the standard expression $\xi\,=\,\sqrt{\frac{3\Phi_c}{2\pi G}}$ valid for the Newtonian limit of GR. Besides, it is worth noticing that in the $f(R)$-gravity case, for $n=0$, the radius is smaller than in GR. On the other hand, the gravitational potential $-\Phi$ gives rise to a deeper potential well than the corresponding Newtonian potential derived from GR [@stabile].
In the case $n\,=\,1$, Eq. (\[LEmod\]) can be recast as follows
$$\begin{aligned}
\label{LEmod2}
\frac{d^2\tilde{w}(z)}{dz^2}+\tilde{w}(z)\,=\,\frac{m\xi_0}{8}\int_0^{\xi/\xi_0}
dz'\,\biggl\{e^{-m\xi_0|z-z'|}-e^{-m\xi_0|z+z'|}\biggr\}\,\tilde{w}(z')\end{aligned}$$
where $\tilde{w}\,=\,z\,w$. If we consider the solution of (\[LEmod2\]) as a small perturbation to the one of GR, we have
$$\begin{aligned}
\label{hypsol}
\tilde{w}^{(1)}_{_{f(R)}}(z)\,\sim\,\tilde{w}^{(1)}_{GR}(z)+e^{-m\xi}\Delta\tilde{w}^{(1)}_{_{f(R)}}(z)\end{aligned}$$
The coefficient $e^{-m\xi}\,<\,1$ is the parameter with respect to which we perturb Eq. (\[LEmod2\]). Besides these position ensure us that when $m\,\rightarrow\,\infty$ the solution converge to something like $\tilde{w}^{(1)}_{GR}(z)$. Substituting Eq.(\[hypsol\]) in Eq.(\[LEmod2\]), we have
$$\begin{aligned}
\frac{d^2\Delta\tilde{w}^{(1)}_{_{f(R)}}(z)}{dz^2}+\Delta\tilde{w}^{(1)}_{_{f(R)}}(z)\,=\,\frac{m\xi_0\,e^{m\xi}}{8}\int_0^{\xi/\xi_0}
dz'\,\biggl\{e^{-m\xi_0|z-z'|}-e^{-m\xi_0|z+z'|}\biggr\}\,\tilde{w}^{(1)}_{GR}(z')\nonumber\\\end{aligned}$$
and the solution is easily found
$$\begin{aligned}
&& w^{(1)}_{_{f(R)}}(z)\sim \frac{\sin z}{z}\biggl\{1+\frac{m^2{\xi_0}^2}{8(1+m^2{\xi_0}^2)}\biggl[1+\frac{2\,e^{-m\xi}}{1+m^2{\xi_0}^2}(\cos\xi/\xi_0+m\xi_0
\sin\xi/\xi_0)
\biggr]\biggr\}
\nonumber\\\nonumber\\&&
-\frac{m^2{\xi_0}^2}{8(1+m^2{\xi_0}^2)}\biggl[\frac{2\,e^{-m\xi}}{1+m^2{\xi_0}^2}(\cos\xi/\xi_0+m\xi_0\sin\xi/\xi_0)
\frac{\sinh m\xi_0z}{m\xi_0z}+\cos z\biggr]\nonumber\\\end{aligned}$$
Also in this case, for $m\,\rightarrow\,\infty$, we do not recover exactly $w^{(1)}_{GR}(z)$. The reason is the same of previous $n\,=\,0$ case [@eddington]. Analytical solutions for other values of $n$ are not available.
To conclude this section, we report the gravitational potential profile generated by a spherically symmetric source of uniform mass with radius $\xi$. We can impose a mass density of the form $$\rho\,=\,\frac{3M}{4\pi\xi^3}\Theta(\xi-|\mathbf{x}|)$$ where $\Theta$ is the Heaviside function and $M$ is the mass [@stabile; @cqg]. By solving field Eqs. (\[HOEQ\]) [*inside the star*]{} and considering the boundary conditions $w(0)\,=\,1$ and $w'(0)\,=\,0$, we get $$\begin{aligned}
\label{sol_pot}
w_{_{f(R)}}(z)\,&=&\,\biggl[\frac{3}{2\xi}+\frac{1}{m^2\xi^3}-\frac{e^{-m\xi}(1+m\,\xi)}{m^2\xi^3}\biggr]^{-1}\times
\nonumber\\ && \times\biggl[\frac{3}{2\xi}+\frac{1}{m^2\xi^3}-\frac{{\xi_0}^2z^2}{2\xi^3}-\frac{e^{-m\xi}(1+m\,\xi)}{m^2\xi^3}\frac{\sinh m\xi_0z}{m\xi_0z}\biggr]\end{aligned}$$ In the limit $m\,\rightarrow\,\infty$, we recover the GR case $w_{GR}(z)\,=\,1-\frac{{\xi_0}^2z^2}{3\xi^2}$. In Fig. \[fig\] we show the behaviors of $w^{(0)}_{_{f(R)}}(z)$ and $w^{(1)}_{_{f(R)}}(z)$ with respect to the corresponding GR cases. Furthermore, we plot the potential generated by a uniform spherically symmetric mass distribution in GR and $f(R)$-gravity and the case $w^{(5)}_{GR}(z)$.
![Plot of solutions (blue lines) of standard Lané-Emden Eq. (\[LE\]): $w^{(0)}_{GR}(z)$ (dotted line) and $w^{(1)}_{GR}(z)$ (dashed line). The green line corresponds to $w^{(5)}_{GR}(z)$. The red lines are the solutions of modified Lané-Emden Eq. (\[LEmod\]): $w^{(0)}_{_{f(R)}}(z)$ (dotted line) and $w^{(1)}_{_{f(R)}}(z)$ (dashed line). The blue dashed-dotted line is the potential derived from GR ($w_{GR}(z)$) and the red dashed-dotted line the potential derived from $f(R)$-gravity ($w_{_{f(R)}}(z)$) for a uniform spherically symmetric mass distribution. The assumed values are $m\xi\,=\,1$ and $m\xi_0\,=\,4$. From a rapid inspection of these plots, the differences between GR and $f(R)$ gravitational potentials are clear and the tendency is that at larger radius $z$ they become more evident. []{data-label="fig"}](fig.eps "fig:")\
Examples: dust-dominated self-gravitating systems {#dieci}
=================================================
From the above equations of hydrostatic equilibrium in $f(R)$-gravity, one can study the formation and collapse of self gravitating objects like stars and black holes. It is well known that the scenario involving the formation of cosmic structures ([*i.e.*]{} from stars up to galaxies and clusters of galaxies) occurs for gravitational instability. In the standard picture, the first objects that form are the dark matter halos, which aggregate in a hierarchical way due to gravitational collapse. If such systems reach a virial equilibrium, the collapse can stop otherwise it can indefinitely continue. Subsequently, the baryons are affected by the gravity of the halo potential wells: gas, of collisional nature, converts the kinetic energy of the ’fall’ into thermal energy and heat reaching the virial temperature. Subsequently, the radiation losses cause the cooling of the baryonic component, its condensation and subsequent formation of molecular clouds, finally of stars. The standard theory of gravitational collapse for dust dominated systems and can be compared to that for $f(R)$-gravity. The collapse of self-gravitational collisionless systems can be dealt with the introduction of coupled collisionless Boltzmann and Poisson equations (for details, see [@BT]):
$$\label{Boltz_poisson1}
\dfrac{{\partial f(\vec r,\vec v,t)}}{{\partial t}} + \left( {\vec v\cdot\vec \nabla _r } \right)f(\vec r,\vec v,t) -\left( {\vec \nabla \Phi \cdot\vec \nabla _v } \right)f(\vec r,\vec v,t) = 0$$
$$\label{Boltz_poisson2}
\vec \nabla ^2 \Phi (\vec r,t) = 4\pi G\int {f(\vec r,\vec v,t)} d\vec v.$$
A self-gravitating system at equilibrium is described by a time-independent distribution function $f_0(x,v)$ and a potential $\Phi_0(x)$ that are solutions of Eq. and . Considering a small perturbation to this equilibrium:
$$\begin{aligned}
&& f(\vec r,\vec v,t) = f_0 (\vec r,\vec v) + \epsilon f_1 (\vec r,\vec v,t),
\\
&& \Phi (\vec r,t) = \Phi _0 (\vec r) + \epsilon \Phi _1 (\vec r,t),\end{aligned}$$
where $\epsilon \ll 1$ and by substituting in Eq. and and by linearizing, one obtains: $$\label{Linear_Boltz_poisson1}
\dfrac{{\partial f_1 (\vec r,\vec v,t)}}{{\partial t}} + \vec v\cdot\dfrac{{\partial f_1 (\vec r,\vec v,t)}}{{\partial \vec r}}
-\vec \nabla \Phi _1 (\vec r,t)\cdot\dfrac{{\partial f_0 (\vec r,\vec v)}}{{\partial \vec v}} - \vec\nabla \Phi _0 (\vec r)\cdot\dfrac{{\partial f_1 (\vec r,\vec v,t)}}{{\partial \vec v}} = 0\,,$$ $$\label{Linear_Boltz_poisson2}
\vec \nabla ^2 \Phi _1 (\vec r,t) = 4\pi G\int {f_1 (\vec r,\vec v,t)} d\vec v\,,$$
Since the equilibrium state is assumed to be homogeneous and time-independent, one can set $f_0(\vec x, \vec v,t) = f_0(\vec v)$, and the so-called Jeans “swindle” to set $\Phi_0 = 0$. In Fourier components, Eqs. and become: $$\begin{aligned}
&&- i\omega f_1 + \vec v\cdot\left( {i\vec kf_1 } \right) - \left( {i\vec k\Phi _1 } \right)\cdot\frac{{\partial f_0 }}{{\partial \vec v}} = 0\,, \\
&&- k^2 \Phi _1 = 4\pi G\int {f_1 } d\vec v.
\end{aligned}$$ By combining these equations, the dispersion relation $$\label{eqDISP}
{1 + \frac{{4\pi G}}{{k^2 }}\int {\dfrac{{\vec k\cdot\dfrac{{\partial f_0 }}{{\partial \vec v}}}}{{\vec v\cdot\vec k - \omega }}} d\vec v}=0;$$ is obtained. In the case of stellar systems, by assuming a Maxwellian distribution function for $f_0$, we have $$\label{Maxwellian}
f_0 = \frac{{\rho _0 }}{{(2\pi \sigma ^2 )^{\frac{3}{2}} }}e^{ - \dfrac{{v^2 }}{{2\sigma ^2 }}},$$ imposing that $\vec k=(k,0,0)$ and substituting in Eq., one gets: $$1 - \frac{{2\sqrt {2\pi } G\rho _0 }}{{k\sigma ^3 }}\int {\dfrac{{v_x e^{ - \dfrac{{v_x^2 }}{{2\sigma ^2 }}} }}{{kv_x - \omega }}} dv_x = 0.$$ By setting $\omega=0$, the limit for instability is obtained: $$k^2(\omega=0) = \dfrac{{4\pi G\rho _0 }}{{\sigma ^2 }} = k_J^2,$$ by which it is possible to define the Jeans mass ($M_J$) as the mass originally contained within a sphere of diameter $\lambda_J$: $$\label{MJ}
M_J= \frac{4 \pi}{3} \rho_0 \left(\frac{1}{2} \lambda_J \right)^3,$$ where $$\lambda_J^2= \dfrac{\pi \sigma^2}{G \rho_0}
\label{length}$$ is the Jeans length. Substituting Eq. into Eq. , we recover $$\label{Mass}
{{{ M}_J} = \dfrac{\pi }{6}\sqrt {\dfrac{1}{{{\rho _0}}}{{\left( {\dfrac{{\pi {\sigma ^2}}}{{G}}} \right)}^3}} }\,.$$ All perturbations with wavelengths $\lambda >\lambda_J$ are unstable in the stellar system. In order to evaluate the integral in the dispersion relation for real and nonzero values of $\omega$, the dispersion relation has to be rewritten as $$1 - \dfrac{{k_{_J }^2 }}{{k^2 }}W\left( {\dfrac{\omega}{{k\sigma }}} \right) = 0,$$ defining $$W\left( {\dfrac{\omega}{{k\sigma }}} \right) \equiv \dfrac{1}{{\sqrt {2\pi } }}\int {\dfrac{{xe^{ - \dfrac{{x^2 }}{2}} }}{{x - Z}}} dx,$$
and setting $\omega=i\omega_I$ and $Re(W\left( {\dfrac{\omega}{{k\sigma }}} \right))=0$. In order to study unstable modes (for details, see Appendix B in [@BT]) we replace the following identities $$\left\{ \begin{array}{l}
\int\limits_0^\infty {\dfrac{{x^2 e^{ - x^2 } }}{{x^2 + \beta ^2 }}dx} = \dfrac{1}{2}\sqrt \pi - \dfrac{1}{2}\pi \beta e^{\beta ^2 } \left[ {1 - {\rm{erf }}\beta } \right]\,, \\
\\
{\rm{erf }}z = \dfrac{2}{{\sqrt \pi }}\int\limits_0^z {e^{ - t^2 } dt}. \\
\end{array} \right.$$ into the dispersion relation obtaining: $$\small{k^2 = k_J^2 \left\{ {1 - \dfrac{{\sqrt \pi \omega _I }}{{\sqrt 2 k\sigma }}e^{\left( {\dfrac{{\omega _I }}{{\sqrt 2 k\sigma }}} \right)^2 } \left[ {1 - {\rm{erf }}\left( {\frac{{\omega _I }}{{\sqrt 2 k\sigma }}} \right)} \right]} \right\}}.$$ This is the standard dispersion relation describing the criterion to collapse for infinite homogeneous fluid and stellar systems [@BT].
Jeans criterion for gravitational instability in $f(R)$-gravity {#undici}
===============================================================
Our task is now to check how the Jeans instability occurs in $f(R)$-gravity [@jeans]. Let us approach the Jeans instability with the Poisson equations given by Eqs. and after assuming the collisionless Boltzmann equation: $${\begin{array}{l}
\dfrac{{\partial f(\vec r,\vec v,t)}}{{\partial t}} + \left( {\vec v\cdot\vec \nabla _r } \right)f(\vec r,\vec v,t) -\left( {\vec \nabla \Phi \cdot\vec \nabla _v } \right)f(\vec r,\vec v,t) = 0\,.
\end{array}}$$ Then we have
$$\nabla^2(\Phi+\Psi) -2 \alpha \nabla^4 (\Phi -\Psi)=16\pi G\int {{f}(\vec r,\vec v,t)} d\vec v{\mkern 1mu}$$
$$\nabla^2(\Phi-\Psi) +3 \alpha \nabla^4(\Phi-\Psi)=-8\pi G\int {{f}(\vec r,\vec v,t)} d\vec v{\mkern 1mu}\,.$$
In the previous equations, we have replaced $f''(0)$ with the greek letter $\alpha$. As in standard case, we consider small perturbation to the equilibrium and linearize the equations. After we write equations in Fourier space so they became $$\begin{aligned}
\label{F1}
&&{ - i\omega {f_1} + \vec v\cdot\left( {i\vec k{f_1}} \right) - \left( {i\vec k{\Phi _1}} \right)\cdot\frac{{\partial {f_0}}}{{\partial \vec v}} = 0},
\\ \label{F2}
&&{ - {k^2}{(\Phi _1} +{\Psi _1}) - 2\alpha {k^4}({\Phi _1} -{\Psi _1}) = 16\pi G\int {{f_1}} d\vec v},
\\ \label{F3}
&& { {k^2}({\Phi _1} -{\Psi _1}) - 3\alpha {k^4}({\Phi _1} -{\Psi _1}) = 8\pi G\int {{f_1}} d\vec v}.\end{aligned}$$ Combining Eqs. and, we obtain a relation between $\Phi_1$ and $\Psi_1$, $${{\Psi _1} = \frac{{3 - 4\alpha {k^2}}}{{1 - 4\alpha {k^2}}}{\Phi _1}}$$ inserting this relation in Eq. and combining it with Eq. , we obtain the dispersion relation $$\label{newEQdispersion}
1 - 4\pi G\dfrac{{1 - 4\alpha {k^2}}}{{3\alpha {k^4} - {k^2}}}\int {\left( {\dfrac{{\vec k\cdot\dfrac{{\partial {f_0}}}{{\partial \vec v}}}}{{\vec v\cdot\vec k - \omega }}} \right)} d\vec v = 0.$$ If we assume, as in standard case, that $f_0$ is given by and $\vec{k} = (k,0,0)$, one can write $$\label{NewEqDispersion2}
{1 + \frac{{2\sqrt {2\pi } G{\rho _0}}}{{{\sigma ^3}}}\dfrac{{1 - 4\alpha {k^2}}}{{3\alpha {k^4} - {k^2}}}\left[ {\int {\dfrac{{k{v_x}{e^{ - \frac{{v_x^2}}{{2{\sigma ^2}}}}}}}{{k{v_x} - \omega }}} d{v_x}} \right] = 0}.$$ By eliminating the higher-order terms (imposing $\alpha=0$), we obtain again the standard dispersion Eq. . In order to compute the integral in the dispersion relation , we consider the same approach used in the classical case, and finally we obtain: $$\label{eq:Results}
{\begin{array}{l}
1 + \mathcal{G}{\dfrac{{1 - 4\alpha {k^2}}}{{3\alpha {k^4} - {k^2}}}} \left[ {1 - \sqrt\pi x e^{x^2}\left( {1 - {\rm{erf}}\left[ x\right]} \right)} \right] = 0,
\end{array}}$$ where $x=\dfrac{{{\omega _i}}}{{\sqrt 2 k\sigma }}$ and $\mathcal{G}=\dfrac{{4G\pi {\rho _0}}}{{{\sigma ^2}}}$. In order to evaluate Eq. comparing it with the classical one, given by Eq. , it is very useful to normalize the equation to the classical Jeans length showed in Eq. , by fixing the parameter of $f(R)$-gravity, that is $$\label{alpha}
\alpha = - \frac{1}{{k_j^2}}= - \frac{{{\sigma ^2}}}{{4\pi G{\rho _0}}}.$$ This parameterization is correct because the dimension $\alpha$ (an inverse of squared length) allows us to parametrize as in the standard case. Finally we write $$\begin{array}{l}\label{instability}
\dfrac{{3{k^4}}}{{k_j^4}}+ \dfrac{{{k^2}}}{{k_j^2}} = \left( {\dfrac{{4{k^2}}}{{k_j^2}} + 1} \right) \left[ {1 - \sqrt\pi x e^{x^2}\left( {1 - {\rm{erf}}\left[ x\right]} \right)} \right] =0.
\end{array}$$ The function is plotted in Fig.\[Fig:Jeans\], where Eq. and the standard dispersion [@BT] are confronted in order to see the difference between $f(R)$ and Newtonian gravity.
![The bold line indicates the plot of the dispersion relation in which we imposed the value for $\alpha$ given by . The thin line indicates the plot of the standard dispersion equation [@BT].[]{data-label="Fig:Jeans"}](Jeans.eps "fig:")\
As shown in the Figure \[Fig:Jeans\], the effects of a different theory of gravity changes the limit of instability. The limit is higher than the classical case and the curve has a greater slope. This fact is important because the mass limit value of interstellar clouds decreases changing the initial conditions to start the collapse. This feature could have dramatic effects for star and black hole formation.
The Jeans mass limit in $f(R)$-gravity {#dodici}
======================================
A numerical estimation of the $f(R)$-instability length in terms of the standard Newtonian one can be achieved. By solving numerically Eq. with the condition $\omega=0$, we obtain that the collapse occurs for $$\label{newlengthnumerical}
k^2=1.2637 k^2_J.$$ However we can estimate also analytically the limit for the instability. In order to evaluate the Jeans mass limit in $f(R)$-gravity, we set $\omega=0$ in Eq. and then $$\label{NewMass}
3{\sigma ^2}\alpha {k^4} - \left( {16\pi G{\rho _0}\alpha + {\sigma ^2}} \right){k^2} + 4\pi G{\rho _0} = 0.$$ It is worth stressing that the additional condition $\alpha<0$ discriminates the class of viable $f(R)$ models: in such a case we obtain stable cosmological solution and positively defined massive states [@Capozziello10]. In other words, this condition selects the physically viable models allowing to solve Eq. for real values of k. In particular, the above numerical solution can be recast as $$k^2=\frac{2}{3} \left(3+\sqrt{21}\right) \pi \frac{G \rho }{\sigma ^2}.$$ The relation to the Newtonian value of the Jeans instability is $$k^2=\frac{1}{6} \left(3+\sqrt{21}\right) k^2_J.$$ Now, we can define the new Jeans mass as: $$\label{newMass2}
\tilde{M}_J=6\sqrt{\frac{6}{\left(3+\sqrt{21}\right)^3}} M_J,$$ that is proportional to the standard Newtonian value. We will confront this specific solutions with some observed structures.
Before this comparison, some considerations are in order. Star formation is one of the best settled problems of modern astrophysics. However, some shortcomings emerge as soon as one faces dynamics of diffuse gas evolving into stars and star formation in galactic environment. One can deal with the star formation problem in two ways: $i)$ we can take into account the formation of individual stars and $ii)$ we can discuss the formation of the whole star system starting from interstellar clouds [@Mckee2]. To answer these problems it is very important to study the interstellar medium (ISM) and its properties.The ISM physical conditions in the galaxies change in a very wide range, from hot X-ray emitting plasma to cold molecular gas, so it is very complicated to classify the ISM by its properties. However, we can distinguish, in the first approximation, between [@carroll; @kipp; @dopita; @Scheff]:
- [**Diffuse Hydrogen Clouds**]{}. The most powerful tool to measure the properties of these clouds is the 21cm line emission of HI. They are cold clouds so the temperature is in the range $10\div 50$ K, and their extension is up to $50\div 100$ kpc from galactic center.
- [**Diffuse Molecular Clouds**]{} are generally self-gravitating, magnetized, turbulent fluids systems, observed in sub-mm. The most of the molecular gas is $H_2$, and the rest is CO. Here, the conditions are very similar to the HI clouds but in this case, the cloud can be more massive. They have, typically, masses in the range $3\div100 \,M_\odot$, temperature in $15\div 50$K and particle density in $(5\div50) \times 10^8$ m$^{-3}$.
- [**Giant Molecular Clouds**]{} are very large complexes of particles (dust and gas), in which the range of the masses is typically $10^5\div 10^6\, M_\odot$ but they are very cold. The temperature is $\sim15$K, and the number of particles is $(1\div 3) \times 10^8$ m$^{-3}$ [@Mckee; @Mckee2; @Blitz; @Blitz2]. However, there exist also small molecular clouds with masses $M < 10^4 M_\odot$[@Blitz2]. They are the best sites for star formation, despite the mechanism of formation does not recover the star formation rate that would be $250 M_\odot yr^{-1}$ [@Mckee].
- [**HII regions**]{}. They are ISM regions with temperatures in the range $10^3 \div 10^4$ K, emitting primarily in the radio and IR regions. At low frequencies, observations are associated to free-free electron transition (thermal Bremsstrahlung). Their densities range from over a million particles per cm$^3$ in the ultra-compact H II regions to only a few particles per cm$^3$ in the largest and most extended regions. This implies total masses between $10^2$ and $10^5$ M$_\odot$ [@HII].
- [**Bok Globules**]{} are dark clouds of dense cosmic dust and gas in which star formation sometimes takes place. Bok globules are found within H II regions, and typically have a mass of about $2$ to $50$ M$_\odot$ contained within a region of about a light year.
Using very general conditions [@carroll; @kipp; @dopita; @Scheff; @Mckee; @Mckee2; @Blitz; @Blitz2; @HII], we want to show the difference in the Jeans mass value between standard and $f(R)$-gravity. Let us take into account Eq. and Eq.: $$\label{mass3}
{{{ M}_J} = \frac{\pi }{6}\sqrt {\frac{1}{{{\rho_0}}}{{\left( {\frac{{\pi {\sigma ^2}}}{{G}}} \right)}^3}} },$$ in which $\rho_0$ is the ISM density and $\sigma$ is the velocity dispersion of particles due to the temperature. These two quantities are defined as $$\rho_0= m_{H} n_{H} \mu, \rm{\quad \quad} \sigma^2=\frac{k_BT}{m_H}$$ where $n_H$ is the number of particles measured in $m^{-3}$, $\mu$ is the mean molecular weight, $k_B$ is the Boltzmann constant and $m_H$ is the proton mass. By using these relations, we are able to compute the Jeans mass for interstellar clouds and to show the behavior of the Jeans mass with the temperature. Results are shown in Tab.\[table1\] and Fig.\[Fig:mass\]
![The $M_J$-$T$ relation. Dashed-line indicates the Newtonian Jeans mass behavior with respect to the temperature. Continue-line indicates the same for $f(R)$-gravity Jeans mass.[]{data-label="Fig:mass"}](mass.eps)
By using Eq. and by referring to the catalog of molecular clouds in Roman-Duval J. et al. [@duval], we have calculated the Jeans mass in the Newtonian and $f(R)$ cases. Tab.\[table2\] shows the results. In all cases we note a substantial difference between the classical and $f(R)$ value. We can conclude that, in $f(R)$-scenario, molecular clouds become sites where star formation is strongly supported and more efficient.
Discussion and conclusions {#discuss}
==========================
Black hole solutions can be find out, in principle, in any Extended Theory of Gravity. Here, we have shown that exact solutions can be achieved in $f(R)$-gravity starting from symmetry considerations. In particular, the existence of Noether symmetries allow to find exact solutions. Such solutions can be eventually classified as black holes.
In particular, we have shown that the Newman-Janis method, used to derive rotating black hole solutions in GR, works also in $f(R)$-gravity. In principle, it could be consistently applied any time a spherically symmetric solutions is derived. The method does not depend on the field equations but directly works on the solutions that, a posteriori, has to be checked to fulfill the field equations.
The key point of the method is to find out a suitable complex transformation which, from a physical viewpoint, corresponds to the fact that we are reducing the number of independent Killing vectors. From a mathematical viewpoint, it is useful since allows to overcome the problem of a direct search for axially symmetric solutions that, in $f(R)$-gravity, could be extremely cumbersome due to the fourth-order field equations.
We have to stress that the utility of generating techniques is not simply to obtain a new metric, but a metric of a new spacetime with specific properties as the transformation properties of the energy-momentum tensor and Killing vectors. In its original application, the Newman-Janis procedure transforms an Einstein-Maxwell solution (Reissner-Nordstrom) into another Einstein-Maxwell solution (Kerr-Newman). As a particular case (setting the charge to zero) it is possible to achieve the transformation between two vacuum solutions (Schwarzschild and Kerr). Also in case of $f(R)$-gravity, new features emerge by adopting such a technique. In particular, it is worth studying how certain features of spherically simmetric metrics, derived in $f(R)$-gravity, result transformed in the new axially symmetric solutions for black holes. For example, considering the $f(R)$ spherically symmetric solution studied here, the Ricci scalar evolves as $ r^{-2}$ and then the asymptotic flatness is recovered. Let us consider now the axially symmetric metric achieved by the Newman-Janis method. The parameter $a\neq 0$ indicates that the spherical symmetry ($a=0$) is broken. Such a parameter can be immediately related to the presence of an axis of symmetry and then to the fact that a Killing vector, related to the angle $\theta$, has been lost. To conclude, we can say that once the vacuum case is discussed, more general spherical metrics can be transformed in new axially symmetric metrics adopting more general techniques [@bk:dk], that would allow us to have the best configurations to describe the black hole dynamics.
A part the mathematical interest in finding out new solutions, such results could have remarkable applications, at least as toy models, in the study of self-gravitating systems. We have shown that the stellar theory as well as the Jeans analysis of instability and collapse can be dramatically altered if one adopt $f(R)$-gravity instead of GR.
The study has been performed starting from the Newtonian limit of $f(R)$-field equations. Since the field equations satisfy in any case the Bianchi identity, we can use the conservation law of energy-momentum tensor. In particular adopting a polytropic equation of state relating the mass density to the pressure, we derive the *modified Lané-Emden equation* and its solutions for $n\,=\,0,\,1$ which can be compared to the analogous solutions coming from the Newtonian limit of GR. When we consider the limit $f(R)\,\rightarrow\,R$, we obtain the standard hydrostatic equilibrium theory coming from GR. A peculiarity of $f(R)$-gravity is the non-viability of Gauss theorem and then the *modified Lané-Emden equation* is an integro-differential equation where the mass distribution plays a crucial role. Furthermore the correlation between two points in the star is given by a Yukawa-like term of the corresponding Green function.
These solutions have been matched with those coming from GR and the corresponding density radial profiles have been derived. In the case $n\,=\,0$, we find an exact solution, while, for $n\,=\,1$, we used a perturbative analysis with respect to the solution coming from GR. It is possible to demonstrate that density radial profiles coming from $f(R)$-gravity analytic models and close to those coming from GR are compatible. This result rules out some wrong claims in the literature stating that $f(R)$-gravity is not compatible with self-gravitating systems. Obviously the choice of the free parameter of the theory has to be consistent with boundary conditions and then the solutions are parameterized by a suitable “wave-length” $m\,=\,\sqrt{-\frac{1}{3f''(0)}}$ that should be experimentally fixed.
The next step is to derive self-consistent numerical solutions of *modified Lané-Emden equation* and build up realistic star models where further values of the polytropic index $n$ and other physical parameters, e.g. temperature, opacity, transport of energy, are considered. Interesting cases are the non-relativistic limit ($n\,=\,3/2$) and relativistic limit ($n\,=\,3$) of completely degenerate gas. These models are a challenging task since, up to now, there is no self-consistent, final explanation for compact objects (e.g. neutron stars) with masses larger than Volkoff mass, while observational evidences widely indicate these objects [@mag]. In fact it is plausible that the gravity manifests itself on different characteristic lengths and also other contributions in the gravitational potential should be considered for these exotic objects. As we have seen above, the gravitational potential well results modified by higher-order corrections in the curvature. In particular, it is possible to show that if we put in the gravitational action other curvature invariants also repulsive contributions can emerge [@stabile_2; @cqg]. These situations have to be seriously taken into account in order to address several issues of relativistic astrophysics that seem to be out of the explanation range of the standard theory.
For this purpose, we have analyzed the Jeans instability mechanism, adopted for star formation, considering the Newtonian approximation of $f(R)$-gravity. The related Boltzmann-Vlasov system leads to modified Poisson equations depending on the $f(R)$-model. In particular, considering Eqs. and , it is possible to get a new dispersion relation where instability criterion results modified (see also [@poly]). The leading parameter is $\alpha$, i.e. the second derivative of the specific $f(R)$-model. Standard Newtonian Jeans instability is immediately recovered for $\alpha=0$ corresponding to the Hilbert-Einstein Lagrangian of GR. In Fig. \[Fig:Jeans\], dispersion relations for Newtonian and a specific $f(R)$-model are numerically compared. The modified characteristic length van be given in terms of the classical one.
Both in the classical and in $f(R)$ analysis, the system damps the perturbation. This damping is not associated to the collisions because we neglect them in our treatment, but it is linked to the so called Landau damping [@BT].
A new condition for the gravitational instability is derived, showing unstable modes with faster growth rates. Finally we can observe the instability decrease in $f(R)$-gravity: such decrease is related to a larger Jeans length and then to a lower Jeans mass. We have also compared the behavior with the temperature of the Jeans mass for various types of interstellar molecular clouds (Fig. \[Fig:mass\]). In Tables \[table1\] and \[table2\] we show the results given by this new limit of the Jeans mass for a sample of giant molecular clouds. In our model the limit (in unit of mass) to start the collapse of an interstellar cloud is lower than the classical one advantaging the structure formation. Real solutions for the Jean mass can be achieved only for $\alpha<0$ and this result is in agreement with cosmology [@Capozziello10]. In particular, the condition $\alpha<0$ is essentials to have a well-formulated and well-posed Cauchy problem in $f(R)$-gravity [@Capozziello10]. Finally, it is worth noticing that the Newtonian value is an upper limit for the Jean mass coinciding with $f(R)=R$.
This analysis is intended to indicate the possibility to deal with ISM collapsing clouds under different assumptions about gravity. It is important to stress that we fully recover the standard collapse mechanisms but we could also describe proto-stellar systems that escape the standard collapse model. On the other hand, this is the first step to study star formation and physical black holes in alternative theories of gravity (see also [@Cooney; @Hu; @poly; @Chang]). From an observational point of view, reliable constraints can be achieved from a careful analysis of the proto-stellar phase taking into account magnetic fields, turbulence and collisions. Finally, addressing stellar systems by this approach could be an extremely important to test observationally $f(R)$-gravity.
Moreover, the approach developed here admits direct generalizations for other modified gravities, like non-local gravity, modified Gauss-Bonnet theory, string-inspired gravity, etc. In these cases, the constrained Poisson equation may be even more complicated due to the presence of extra scalar(s) in non-local or string-inspired gravity. Developing further this approach gives, in general, the possibility to confront the observable dynamics of astrophysical objects (like stars) with predictions of alternative gravities.
Acknowledgments {#acknowledgments .unnumbered}
===============
The Authors wish to thank I. De Martino, M. Formisano, S.D. Odintsov, A. Stabile for useful comments, discussions and common works on the topics discussed here.
[80]{} S. Capozziello, [*Int. Jou. Mod. Phys.* ]{} [**D 11**]{}, 483 (2002); S. Nojiri, S.D. Odintsov, [*Phys. Lett.*]{} [**B 576**]{}, 5, (2003); S. Nojiri, S.D. Odintsov, [*Phys. Rev.*]{} [**D 68**]{}, 12352, (2003); S. M. Carroll, V. Duvvuri, M. Trodden and M. S. Turner, [*Phys. Rev.*]{} [**D70**]{}, 043528 (2004); G. Allemandi, A. Borowiec, M. Francaviglia, [*Phys. Rev.*]{} [**D 70**]{}, 103503 (2004); S. Capozziello, V. F. Cardone and A. Troisi, [*Phys. Rev.*]{} [**D 71**]{}, 043503 (2005); S. Carloni, P.K.S. Dunsby, S. Capozziello, A. Troisi, [*Class. Quant. Grav.*]{} [**22**]{}, 4839 (2005). S. Capozziello S., V.F. Cardone, A. Troisi, [*JCAP*]{} 08, 001( 2006) S. Capozziello and M. Francaviglia, [*Gen. Rel. Grav.*]{} [**40**]{}, 357 (2008). S. Capozziello, M.F. De Laurentis, M. Francaviglia, S. Mercadante, [*Found. of Physics*]{} [**39**]{}, 1161 (2009).
S. Capozziello, S. Tsujikawa, [*Phys. Rev. D*]{}, [**77**]{}, 10750, (2008). S. Capozziello, M. De Laurentis, S. Nojiri, S.D. Odintsov, [*Gen. Rel. Grav.*]{} [**41**]{}, 2313, (2009); [*Physical Review D*]{} [**79**]{}, 124007, (2009); M. De Laurentis, S. Capozziello, [*Nuclear Physics B (Proceedings Supplements)*]{}, 212, (2009); M. De Laurentis, S. Capozziello, S. Nojiri, S.D. Odintsov, [*Annalen der Physik*]{} [**19**]{}, 347, (2009); S. Capozziello, N. Carlevaro, M. De Laurentis, M. Lattanzi, G. Montani, arXiv:1104.2169v1 \[astro-ph.CO\] (2011); O.M. Lecian, G. Montani, *Class. Quant. Grav.* **26**, 045014, (2009). L. Pogosian, A. Silvestri, [*Phys. Rev. D*]{} [**77**]{} , 023503, (2008).
S. Capozziello, S. Vignolo [*Class Quantum Grav*]{}, [**26**]{} 168001 [2009]{}; V. Faraoni [*Class. Quantum Grav.*]{} [**26**]{} 168002 (2009); G.J. Olmo, P. Singh, [*JCAP*]{} 030, 0901 (2008); V. Faraoni, N. Lanahan-Tremblay , [*Phys. Rev.*]{} [**D 78**]{}, 064017 (2008); S. Capozziello, S. Vignolo, [*Class. Quantum. Grav.*]{} [**26**]{} 175013 (2009). S. Capozziello, A. Stabile, A. Troisi [*Class. Quant. Grav.*]{} [**24**]{}, 2153 (2007). T. Multamaki, I. Vilja, [*Phys. Rev.*]{} [**D 74**]{}, 064022 (2006). T. Multamaki, I. Vilja [*Phys. Rev.*]{} [**D 76**]{}, 064021 (2007); K. Kainulainen, J. Piilonen, V. Reijonen, D. Sunhede, [*Phys. Rev.*]{} [**D 76**]{}, 024020 (2007). S. Capozziello, A. Stabile, A. Troisi, [*Class. Quant.Grav.*]{}, [**25**]{}, 085004 (2008). R. P. Kerr, [*Phys. Rev. Lett.*]{} [**11**]{} 237 (1963). D. McManus, [*Class. Quant. Grav.*]{} [**8**]{} 863 (1991). A. Kransinski, [*Ann. Phys.*]{} [**112**]{} 22 (1978). H. Stephani, D. Kramer, M. MacCallum, C. Hoenselaers, E. Herlt [*Exact Solutions to Einstein’s Field Equations, Second Edition*]{}, Cambridge University Press, Cambridge (2003). E. T. Newman, A. I. Janis, [*J. Math. Phys.* ]{} [**6**]{} 915 (1965). E. T. Newman, E. Couch, K. Chinnapared, A. Exton, A. Prakash and R. Torrence, [*J. Math. Phys.*]{} [**6**]{} 918 (1965). R. D’Inverno, [*Introducing Einstein’s Relativity*]{} Oxford University Press, New York, (1992). S. P. Drake, R. Turolla, [*Class.Quant.Grav.*]{} [**14**]{} 1883 (1997); S. P. Drake P. Szekeres, [*Gen.Rel.Grav.*]{} [**32**]{}, 445 (2000)
M. M. Schiffer, R. J. Adler, J. Mark and C. Sheffield, [*J. Math. Phys.*]{} [**14**]{} 52 (1973). G. C. Debney, R. P. Kerr and A. Schild, [*J. Math. Phys.*]{} [**10**]{} 1842 (1969). S. Capozziello, M. De Laurentis, and A. Stabile, [*Class. Quantum Grav.*]{} [**27**]{}, 165008 (2010); M. De Laurentis, arXiv:1111.2071v1 \[gr-qc\] (2011).
D. Psaltis, [*Living Rev. Relativity*]{} **11**, (2008). F. Briscese, E. Elizalde, S. Nojiri, S.D. Odintsov, [*Phys. Lett. B*]{} **646**, 105 (2007).
J. Khoury, A. Weltman, [*Phys. Rev. D*]{} **69**, 044026 (2007). W. Hu and I. Sawicki, [*Phys. Rev. D*]{} **76**, 064004 (2007);\
S. Capozziello and S. Tsujikawa [*Phys. Rev. D*]{} **77**, 107501 (2008);\
S. Tsujikawa, T. Tamaki, R. Tavakol, [*JCAP*]{} 0905, 020 (2009);\
A. Upadhye and W. Hu, [*Phys. Rev. D*]{} **80**, 064002 (2009). E. Babichev and D. Langlois, [*Phys. Rev. D*]{} **80**, 121501 (2009);\
E. Babichev and D. Langlois, [*Phys. Rev. D*]{} **81** ,124051 (2010).
A. Cooney, S. DeDeo, D. Psaltis, [*Phys. Rev. D*]{}, **82**, 064033 (2010). A. Upadhye, W. Hu, [*Phys. Rev. D*]{}, **82**, 064002 (2010).
S. Capozziello, M. De Laurentis, S.D. Odintsov, A. Stabile, [*Phys. Rev. D*]{}, [**83**]{}, 064004, (2011). Ph. Chang, L. Hui, 2011, ApJ, **732**, 25 M.P. Muno, at al., [*The Astrophysical Journal*]{} **636**, L41 (2006). A. Stabile, [*Phys. Rev. D*]{} **82**, 124026 (2010).
S. Capozziello, M. De Laurentis, I. De Martino, M. Formisano, S.D. Odintsov, arXiv:1112.0761 (2011).
S. Capozziello and M. De Laurentis, [*Phys. Repts.*]{} [**509**]{}, 167 (2011) G. Allemandi, M. Capone, S. Capozziello, M. Francaviglia, [*Gen. Rel. Grav.*]{} [**38**]{}, 33 (2006). E.L. Ince, [*Ordinary Differential Equations*]{}, Dover, New York (1956). J.D. Barrow, A.C. Ottewill, [*J. Phys.*]{} [**A 16**]{} 2757 (1983). V.I. Arnold, [*Mathematical Methods of Classical Mechanics*]{}, Springer–Verlag, Berlin (1978). G. Marmo, E.J. Saletan, A. Simoni and B. Vitale, [*Dynamical Systems. A Differential Geometric Approach to Symmetry and Reduction*]{}, Wiley, New York (1985). G. Morandi, C. Ferrario C., G. Lo Vecchio, G. Marmo, C. Rubano [*Phys. Rep.*]{} [**188**]{} 149 (1990). S.Weinberg [*“Gravitation and Cosmology”*]{}, John Wiley $\&$ Sons, Inc., New York, (1972); Ch. W. Misner, K.S. Thorne, J.A. Wheeler, [*Gravitation*]{}, ed. W.H. Freeman, San Francisco (1973). Chandrasekhar S. [*The mathematical theory of black holes*]{}, Oxford, (1983). Kippenhahn R., Weigert A., [*Stellar structures and evolution*]{}, Springer-Verlag, Berlin, (1990). Capozziello S., Stabile A., Troisi A., [*Phys. Rev. D*]{} [**76**]{}, 104019 (2007).
Stabile A., [*Phys. Rev. D*]{} [**82**]{}, 064021 (2010). S. Capozziello and A. Stabile, [*Class. Quant. Grav.* ]{} [**26**]{}, 085019 (2009). Capozziello S., Stabile A., Troisi A., [*Phys. Lett. B*]{} [**686**]{}, 79 (2010). J. Binney, S. Tremaine, *Galactic Dynamics*, Princeton University Press, Princeton (1994).
S. Capozziello, S. Vignolo, [*Class. Quant. Grav.*]{} **26**, 175013, (2009).
C. F. McKee, E.C. Ostriker, [*Annu. Rev. Astron. Astrophys.*]{} **45**, 565, (2007). B.W. Carroll, D.A Ostlie., [*An Introduction to Modern Astrophysics*]{}, Pearson (2007). R. Kippenhahn, A. Weigert, [*Stellar structures and evolution, Springer-Verlag*]{}, Berlin, (1990). M.A. Dopita, R.S. Sutherland , [*Astrophysics of the Diffuse Universe*]{} Springer, New York (2003). H. Scheffler, H. Elsasser, [*Physics of the Galaxy and Interstellar Matter*]{} Springer, New York (1982). C. F. McKee [*The Origin of Stars and Planetary Systems*]{}, ed. C. J. Lada $\&$ N.D. Kylafis (Dordrecht: Kluwer), 29, (1999). L. Blitz, [*Giant molecular clouds. In Protostars and Planets III*]{}, eds. E.H. Levy and J.I. Lunine, (Tucson: Univ. of Arizona Press), pp. 125-161 (1993). L. Blitz, L. Magnani, L. Mundy, [*Astrophys. J. Lett.*]{} , **282**, L9, (1984). L.D. Anderson, T.M. Bania, J.M. Jackson et al. , [*ApJS*]{} **181**, 255A, (2009). J. Roman-Duval, J.M. Jackson, M. Heyer, HJ. Rathborne, R. Simon [*ApJ*]{} [**723**]{}, 492, (2010).
[^1]: E-mail address: [email protected]
[^2]: E-mail address: [email protected]
[^3]: We are adopting the convention $c=1$. The convention $R_{\mu\nu}={R^\rho}_{\mu\rho\nu}$ for the Ricci tensor and ${R^\alpha}_{\beta\mu\nu}=\Gamma^\alpha_{\beta\nu,\mu}-...$, for the Riemann tensor. Connections are Levi-Civita : $$\Gamma^\mu_{\alpha\beta}=\frac{1}{2}g^{\mu\rho}(g_{\alpha\rho,\beta}+g_{\beta\rho,\alpha}-g_{\alpha\beta,\rho})\,.\nonumber\\$$ The signature is $(+\,-\,-\,-)$.
[^4]: It is possible to take into account also the Palatini approach in which the metric $g$ and the connection $\Gamma$ are considered independent variables (see for example [@palatini]). Here we will consider the Levi-Civita connection and will use the metric approach. See [@GRGrev; @palatini; @ACCF] for a detailed comparison between the two pictures.
[^5]: This condition allows to obtain the standard definition of the circumference with radius $r$.
[^6]: Obviously, the above choices are recovered for $A(r)=a(r)$, $B(r)=b(r)$, and $M(r)=r^2$. Here we deal with $A,B,M$ as a set of coordinates in a configuration space.
[^7]: Lowering the dimension of configuration space through the substitution (\[eqb\]) does not affect the dynamics since $B$ is a non-evolving quantity. In fact, inserting Eq. (\[eqb\]) into the dynamical equations given by (\[lag2\]), they coincide with those derived by (\[lag2\]).
[^8]: From now on, $\underline{q}$ indicates the vector $\{A,M,R\}$.
[^9]: The dimensions are given by $R^{1-s}$ in terms of the Ricci scalar. For the sake of simplicity, we will put $f_0=1$ in the forthcoming discussion.
[^10]: It is worth noticing that a certain arbitrariness is present in the complexification process of the functions $\lambda$ and $\phi$. Obviously, we have to obtain the metric (\[metrictensorcontro\]) as soon as $r\,=\,\bar{r}$.
[^11]: The radius $r$ is assumed as the spatial coordinate. It varies from $r\,=\,0$ at the center to $r\,=\,\xi$ at the surface of the star
[^12]: The Greek index runs between $0$ and $3$; the Latin index between $1$ and $3$.
[^13]: This assumption is quite natural since the contribution of a cosmological constant term is irrelevant at stellar level.
[^14]: Generally it is $\epsilon\,=\,\rho\,c^2$.
[^15]: It is worth noticing that also if the Gauss theorem does not hold, the Bianchi identities are always valid so the conservation laws are guaranteed.
[^16]: In principle, there is a solution for any value of $m$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Reconstruction of articulatory trajectories from the acoustic speech signal has been proposed for improving speech recognition and text-to-speech synthesis. However, to be useful in these settings, articulatory reconstruction must be speaker independent. Furthermore, as most research focuses on single, small datasets with few speakers, robust articulatory reconstrucion could profit from combining datasets. Standard evaluation measures such as root mean square error and Pearson correlation are inappropriate for evaluating the speaker-independence of models or the usefulness of combining datasets. We present a new evaluation for articulatory reconstruction which is independent of the articulatory data set used for training: the *phone discrimination ABX task*. We use the ABX measure to evaluate a Bi-LSTM based model trained on 3 datasets (14 speakers), and show that it gives information complementary to the standard measures, and enables us to evaluate the effects of dataset merging, as well as the speaker independence of the model.'
address: |
$^1$ CoML, ENS/CNRS/EHESS/INRIA/PSL Research University, Paris, France\
$^2$ Universit[é]{} de Paris, LLF, CNRS, Paris, France\
$^3$CRI, Département Frontières du Vivant et de l’Apprendre, IIFR, Universit[é]{} de Paris
bibliography:
- 'main.bib'
title: 'INDEPENDENT AND AUTOMATIC EVALUATION OF ACOUSTIC-TO-ARTICULATORY INVERSION MODELS'
---
articulatory inversion, speech representation, machine learning, bi-LSTM
Introduction {#sec:intro}
============
Acoustic-to-articulatory inversion is the problem of finding a mapping from acoustic features to a set of articulatory measures (see [@Deep_rnn_arti_inversion; @30_yu2018synthesizing; @28_tobing2017deep; @26_mitra2017joint; @13_seneviratne2019multi; @9_porras2019dnn] for recent models: see [@6_richmond2015use] for a review). Reconstructed articulatory trajectories have been shown to improve text-to-speech synthesis [@24_cao2017integrating; @6_richmond2015use], speech accent conversion [@accent_conversion], and automatic speech recognition [@speech_reco_using_arti; @26_mitra2017joint], in particular for dysarthric speech [@5_yilmaz2019articulatory]; they can also be used in the automatic detection of clinical conditions which have an impact on speech production, such as Parkinson’s [@25_hahm2015parkinson]. *Speaker independent* reconstruction is essential for most of these applications, and, as such, some method for evaluating speaker-independent articulatory reconstruction is needed.
The two principal metrics for evaluating articulatory reconstruction are the root mean square error (RMSE) and the Pearson correlation coefficient (PCC) between the measured and the predicted articulatory trajectories. However, the goal of a speaker-independent model is not precise prediction of reference measures. First, the shapes of speakers’ vocal tracts vary in ways that cannot be captured by simple normalization. Second, recording conditions (coil placement in electromagnetic articulography: EMA), vary across and within recordings. Moreover, acoustic-to-articulatory inversion is a one-to-many problem: two different articulatory trajectories can produce the same sound. We do not want to penalize models that choose to reconstruct another trajectory which is correct but different from the reference. Finally, articulatory data is needed to calculate these measures, which is costly to obtain.
We propose a standardized evaluation based on an *ABX phone discrimination test* [@schatz2013evaluating] of trajectories reconstructed for an acoustic-only corpus. The evaluation has the advantage of being *independent* of the training data set and of the true reference articulatory trajectories, much like the independent evaluation of [@7_richmond2013evaluation], which uses human listeners to evaluate speech re-synthesis on the basis of the reconstructed trajectories. Our evaluation, however, is completely *automatic*.[^1]
We train a bi-LSTM model closely resembling that of [@Deep_rnn_arti_inversion] on three data sets (MOCHA–TIMIT, EMA–IEEE, USC–TIMIT), and apply the ABX phone discrimination evaluation on different training set, validation set, and test set. We compare ABX phone discrimination scores to standard metrics.
Method {#sec:method}
======
Model architecture {#sec:model}
------------------
To demonstrate our evaluation, we use a bidirectional recurrent neural network architecture similar to that of [@Deep_rnn_arti_inversion], but with the addition of a convolutional layer that acts as a low pass filter after the readout layer. The network has two feed-forward layers of 300 units each that act as feature extractors, two bidirectional layers of 300 units each, a convolutional layer as a low pass filter, and, finally, a feed-forward layer with as many units as the number of trajectories predicted (see Figure \[fig:archi\_schema\]).
{width="5cm"}
\[fig:archi\_schema\]
Loss function
-------------
The usual loss function for articulatory inversion is the root mean square error (RMSE), which minimizes the L2 distance between the real and predicted trajectory. The Pearson correlation coefficient (PCC) is also typically examined, in addition to the RMSE values, as it ignores any systematic differences between speakers that can be described linearly. The PCC measures the degree of linear relationship between two variables $X$ and $Y$. $$\begin{aligned}
PCC(X,Y) = \frac{Cov(X, Y)}{\sigma_{X}\sigma_{Y}}\end{aligned}$$
We experiment with including the PCC in the loss function, $y$ the reconstructed trajectory $y^*$ the reference: $$\label{eq:loss_comb}
\mathcal{L}(y,y^*) = RMSE(y,y^*) - \beta\cdot PCC(y,y^*)$$ We use $\beta=1000$ (in order to account for the differences in scale between the RMSE and PCC values).[^2]
Convolutional layer as low pass filter
--------------------------------------
Articulatory gestures are not only continuous but also smooth. We impose smoothing on the predicted output trajectories by integrating a convolutional layer that acts as a low pass filter at the output of our neural network. This avoids unnecessary backpropagation of error due to non-smooth predicted trajectories. We used the following filter of order 5, with a Hanning window to restrict the support, where $N$ is the size of the Hanning window and $f_t$, $\forall n \in [0,N-1]$:
$$\begin{aligned}
\label{eq:weights_filter}
w(n) & \propto & \Big(1- cos(2 \pi \frac{n}{N-1}) \Big)sinc\Big(2 \pi f_t (n-\frac{N-1}{2})\Big)
\end{aligned}$$
ABX phone discrimination evaluation
-----------------------------------
An *ABX phone discrimination test* of a representation of speech consists of extracting the representations of triplets of stimuli (A, B, and X), and computing the distance $d(\mbox{A},\mbox{X})$, between A and X, and $d(\mbox{B},\mbox{X})$, between B and X. X is of the same phonetic category as either A or B. Taking A to be the correct answer, we compute $\delta = d(\mbox{B},\mbox{X}) - d(\mbox{A},\mbox{X})$. If $\delta>0$, the model has chosen A; if $\delta<0$, it has chosen B. As in previous work evaluating acoustic models with this method [@versteegh2015zero; @dunbar2017zero], the percent correct for all pairs of categories are combined into a global ABX discriminability score. We follow previous literature and use dynamic time warping (DTW) based on cosine distance to align sequences of differing length [@senin2008dynamic]. The final distance between the two sequences is the mean of the cosine distance between the matched frames along the alignment path.
ARTICULATORY DATASETS {#sec:datasets}
=====================
Description
-----------
All the data used are freely available. MOCHA–TIMIT[^3] is a database that contains EMA and acoustic data for 460 utterances (20 min) read by two English speakers [@description_mocha]; USC–TIMIT[^4] [@usc_timit_description] provides EMA data for four speakers on the MOCHA–TIMIT sentences (15 min); and EMA–IEEE[^5] [@Haskins] contains eight speakers reading 720 sentences, once each at a normal rate, and once at a fast speech rate (47 min per speaker).[^6] The combined duration is 461 min.
Articulatory trajectories
-------------------------
We use measures in the sagittal plane ($x$: back to front of head; $y$: chin to forehead). The two-dimensional articulatory points available in our datasets are: tongue body (TB), tongue tip (TT), tongue dorsum (TD), upper lip (UL), lower lip (LL), lower incisor (LI), and velum (V). Not all measures are available for all speakers. The velum trajectory is only available in MOCHA–TIMIT, and we exclude specific articulators for certain speakers where the standard deviation was less than 0.5 mm and visual verification suggested strongly that the measure was wrong.[^7] In training conditions combining speakers within corpora, we use the common articulators.
Vocal tract parameters {#sec:vocal_tract}
----------------------
As in previous works [@1_speaker_adaption_vtln; @29_sivaraman2016vocal], we add vocal tract variables, using slightly different formulas from [@1_speaker_adaption_vtln]. We calculate two tract variables from the position of the lips, the vertical lip aperture (VLA) and the horizontal lip protrusion (HPRO):
$$\mbox{VLA} = \mbox{UL}_y - \mbox{LL}_y$$
$$\mbox{HPRO} = \frac{\mbox{UL}_x + \mbox{LL}_x}{2}$$
We also add the tongue tip constriction (TTC: the cosine of the angle of the tongue tip off the horizontal axis) and the tongue body constriction (TBC: the cosine of the angle of the tongue body off the horizontal axis).
$$\mbox{TTC} = \frac{\mbox{TT}_x}{\sqrt{\mbox{TT}_x^2 + \mbox{TT}_y^2}}$$
$$\mbox{TBC} = \frac{\mbox{TB}_x}{\sqrt{\mbox{TB}_x^2 + \mbox{TB}_y^2}}$$
Experiments {#sec:experiments}
===========
We compare standard reconstruction scores against ABX scores as evaluations of speaker-independence, both within and across corpora. Within corpus, we compare models trained in a **speaker-specific** setting, on a single speaker, with models trained in a **speaker-independent** setting, training on multiple speakers. We randomly hold out data in the speaker-specific setting for validation and test (for calculating the reconstruction scores: 70% train, 10% validation, 20% test). In the speaker-independent setting, we validate on a subset of the speakers, test on one speaker, and train on the rest. The speaker-specific model gives an expected upper bound on reconstruction. We expect that speaker-independent models will give poorer reconstruction, but we seek to use the ABX score to assess whether this degradation is due to failure to reconstruct linguistically relevant articulatory information, or failure to reconstruct speaker-specific detail.
To test the effect of merging corpora, we compare a model trained in a **multi-corpus** setting (with speakers EMA–IEEE: M01, MOCHA–TIMIT: FSEW0, and USC–TIMIT: M1 held out for validation and test) against a **single-corpus** setting, training on EMA–IEEE, which contains the most complete set of articulators (speaker M01 still hold out for test). Here, rather than training only on articulators common to all speakers, we learn to reconstruct all trajectories by ignoring error on missing articulators for backpropagation.
Model parameters
----------------
We use the Adam optimizer with early stopping on the validation set (learning rate 0.001, batch size 10, patience 5). The weights of the low pass filter are fixed according to with $N=50$ to give a transition band of 0.08. The convolution has one channel, stride of 1, and padding such that the output has the same size as the input. The cutoff frequency $f_c$ is 10Hz.
Data preprocessing
------------------
We use as input the thirteen first MFCCs + $\Delta$ + $\Delta \Delta$ with window size of 25ms and stride of 10ms. We add 10 context windows: the 5 previous and 5 following frames, as in [@deep_archi_arti_inv]. We remove silences based on the transcription file (when available). We normalize the MFCCs per speaker, removing the mean and dividing by the standard deviation.
We pre-smooth the articulatory trajectories, applying a low pass filter with a cutoff frequency of 10Hz for all the data sets except for EMA–IEEE, for which we use 20Hz. We remove leading and trailing silences ends using the transcription when available. We reduce EMA sampling rates to 100Hz to have a single articulatory frame per MFCC frame. Since EMA coils move gradually during recording [@These_richmond], we normalize each articulatory measure by subtracting the mean over the 60 previous and following recordings of the same speaker and divide by the speaker-specific standard deviation.
Reconstruction scores {#sec:reconstruction-evaluation}
---------------------
RMSE (mm) and RMSE computed on normalized trajectories are computed on every feature except TTC and TBC, and PCC is computed on every feature available.
ABX scores {#sec:abx-evaluation}
----------
The ABX test is performed on the 1-second English (speech-only) test data set from the Zero Resource Speech Challenge 2017 [@dunbar2017zero], consisting of data from 24 speakers taken from the LibriVox audio book collection, labelled using the 39 CMUDICT phonemes plus `ax` for \[\]. Stimuli are triphones differing in the central phone (*beg*–*bag*, *api*–*ati*, etc). *Within-speaker* triplets contain three triphones from a single speaker (e.g., $A=\textrm{beg}_{T1}$, $B=\textrm{bag}_{T1}$, $X=\textrm{bag}'_{T1}$). In *across-speaker* triplets, $A$ and $B$ come from the same speaker, and $X$ to another. $A=\textrm{beg}_{T1}$, $B=\textrm{bag}_{T1}$, $X=\textrm{bag}_{T2}$. The scores for a given contrast are first averaged across all (pairs of) speakers for which triplets can be constructed, before averaging over all contexts, and over all pairs of central phones and being converted to an error rate by subtracting from $1$. We exclude contrasts with less than three contexts and for which critical articulators were missing from the data.[^8]
Results {#sec:results}
=======
The model attains reconstruction scores on speaker-specific training for FSEW0 which are comparable to existing results (RMSE-mm: 1.43, RMSE-norm: 0.55, PCC: 0.77).
-- ------- ------- ------- ------ ------ ------- ------- ------- ------ ------ ------- ------- ------- ------ ------
1.380 0.557 0.759 23.9 32.2 1.478 0.608 0.747 24.5 33.8 1.557 0.501 0.840 22.1 30.5
2.184 0.851 0.417 24.6 32.0 2.310 0.917 0.199 24.3 33.9 2.198 0.688 0.672 18.4 24.8
-- ------- ------- ------- ------ ------ ------- ------- ------- ------ ------ ------- ------- ------- ------ ------
-- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------
R Rn PCC R Rn PCC R Rn PCC w a
1.79 0.66 0.72 2.05 1.13 0.02 2.72 1.11 0.08 18.9 25.0
1.80 0.66 0.71 1.89 1.04 0.14 2.61 0.98 0.22 19.7 26.7
-- ------ ------ ------ ------ ------ ------ ------ ------ ------ ------ ------
Speaker-independence within corpus
----------------------------------
Table \[tab: results\_global\] compare the average across speakers in the speaker-specific setting against the average over all one-speaker-held-out training configurations in the speaker-independent setting.
As expected, the speaker-independent condition shows degradation in the three reconstruction scores, compared to the speaker-specific condition.
The ABX scores, calculated on an external speech corpus, provide a different picture. The differences between the two training conditions are small for both MOCHA–TIMIT and USC–TIMIT. Although we lose information about detailed articulatory tracks in the speaker-independent condition, this information is not relevant to coding phonemic contrasts. In the EMA–IEEE corpus, the loss in reconstruction between speaker-specific and speaker-independent conditions is smaller. This may be due to having more data per speaker, or it may be that the measures in this corpus are more similar across speakers. The ABX scores, however, show an improvement in the speaker-independent condition, meaning that, with this corpus, the speaker-independent model not only does not lose linguistically relevant information, but in fact even better reconstructs linguistically relevant articulatory information. This fact is lost by looking only at the reconstruction scores.
Merging corpora
---------------
Results comparing multi-corpus with single-corpus training are shown in Table \[tab:merge\]. ABX scores in the **multi-corpus** condition are similar to those in the **single-corpus** condition, and, in fact, slightly worse, suggesting that adding speakers from additional corpora to the training is not beneficial to the speaker-independent model. Again, the reconstruction measures do not go in the same direction, improving slightly when corpora are added. It is possible that the addition of more data is helping to better reconstructing (informative) speaker-specific subphonemic information; but we observe that the appreciable improvement is exclusively in the novel corpora, suggesting that the improvements in reconstruction may be due to improved modelling of acoustic channel or coil placement properties specific to these corpora. This points to the differences among corpora, and the incompatibility that can exist between them, leading to potentially worse articulatory reconstructions when using them together, information that would be lost without the ABX score.
Conclusion {#sec:conclusion}
==========
We have proposed an ABX phone discrimination measure for the evaluation of speaker-independent acoustic-to-articulatory models. The measure is independent of the articulatory trajectories, and thus does not penalize models for failing to capture speaker-specific articulatory details, it comes from a single external corpus, avoiding the inherent instability of held-out measures, and it is automatic, unlike speech-synthesis based evaluations. Our ABX score only assesses the presence of information needed to contrast the phones labelled in the corpus used (40 English phoneme labels), but they can be replaced by finer-grained allophonic labels, if desired. One caveat of phone discriminability is that it will vary as a function of the set of articulatory dimensions reconstructed, not only of how well they are reconstructed. So this measure can only be used to compare models that reconstruct the same articulators. Nevertheless, we have shown that it can give important information complementary to traditional reconstruction scores, indicative of the degree to which improvements or declines in reconstruction are due to failure to reconstruct speaker-specific properties.
Acknowledgements
================
This research was supported by the École Doctorale Frontières du Vivant (FdV) – Programme Bettencourt, by Facebook AI Research, and by grants ANR-17-CE28-0009 (GEOMPHON), ANR-11-IDFI-023 (IIFR), ANR-18-IDEX-001 (UdP), ANR-10-LABX-0083 (EFL).
[^1]: All code for pre-processing the datasets and for training and testing the model is fully available at <https://github.com/bootphon/articulatory_inversion.git>
[^2]: We ran experiments varying the weights of the RMSE and the PCC, but found it had no systematic effect on the results.
[^3]: <http://data.cstr.ed.ac.uk/mocha>
[^4]: <https://sail.usc.edu/span/usc-timit/>
[^5]: <https://yale.app.box.com/s/cfn8hj2puveo65fq54rp1ml2mk7moj3h/folder/30415804819>
[^6]: We also conducted held-out tests on the single-speaker MNGU0 database, which we do not report due to preprocessing issues.
[^7]: The list of articulators used by speaker is available at <https://github.com/bootphon/articulatory_inversion.git>
[^8]: For example, oral–nasal contrasts such as \[ana\]–\[ada\]–\[ana\], which depend necessarily on the position of the velum: a complete list is provided at <https://github.com/bootphon/articulatory_inversion.git>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Bright, short duration X-ray flares from accreting compact objects produce thin, dust scattering rings that enable dust echo tomography: high precision distance measurements and mapping of the line-of-sight distribution of dust. This work looks to the past activity of X-ray transient outbursts in order to predict the number of sight lines available for dust echo tomography. We search for and measure the properties of 3$\sigma$ significant flares in the 2-4 keV light curves of all objects available in the public MAXI archive. We derive a fluence sensitivity limit of $10^{-3}$ erg cm$^{-2}$ for the techniques used to analyze the light curves. This limits the study mainly to flares from Galactic X-ray sources. We obtain the number density of flares and estimate the total fluence of the corresponding dust echoes. However, the sharpness of a dust echo ring depends on the duration of a flare relative to quiescence. We select flares that are shorter than their corresponding quiescent period to calculate a number density distribution for dust echo rings as a function of fluence. The results are fit with a power law of slope $-2.3 \pm 0.1$. Extrapolating this to dimmer flares, we estimate that the next generation of X-ray telescopes will be 30 times more sensitive than current observatories, resulting in 10-30 dust ring echoes per year. The new telescopes will also be 10-100 times more sensitive than [*Chandra*]{} to dust ring echoes from the intergalactic medium.'
author:
- Lia Corrales
- 'Brianna S. Mills'
- Sebastian Heinz
- 'Gerard M. Williger'
bibliography:
- 'references\_new.bib'
title: 'The X-ray variable sky as seen by MAXI: the future of dust echo tomography with bright Galactic X-ray bursts'
---
Introduction {#sec:intro}
============
Interstellar dust scatters X-ray light over arcminute-scale angles, producing a diffuse scattering halo with an integrated flux $F_{\rm halo}(E) = F_{a} (1 - e^{-\tau})$, where the optical depth to X-ray scattering, $\tau \approx 0.5~E_{\rm keV}^{-2}~({{\rm N}_{\rm H}}/10^{22}~{\rm cm}^{-2})$, and $F_{a}$ is the absorbed flux of the central X-ray source [@PS1995; @Corrales2016]. Because the scattered light takes a longer path to reach the observer, the observed scattering halo surface brightness profile is a convolution of the dust’s line-of-sight position, grain size distribution, and $F_a(t)$ light curve [@Heinz2015 and references therein]. When $F_{a}(t)$ takes the form of a single burst with high amplitude and short duration, a scattering halo will appear as a set of discrete rings, where each ring corresponds to a different foreground dust cloud. These rings expand with a characteristic $t^{1/2}$ time dependency that allows X-ray astronomers to map the line-of-sight distribution of dust (“dust echo tomography”) to much higher resolution than currently available with any other method [@TS1973; @Heinz2015; @Heinz2016].
Mapping the ISM through dust echo tomography is also important for interpreting the time and spectral evolution of accreting compact objects. Dust echoes are known to affect the spectral evolution of X-ray variable objects, producing a prolonged soft-tail [e.g., @Pintore2017; @Jin2018]. This confusion is particularly acute for X-ray timing missions with low imaging resolution: RXTE, MAXI, and NICER.
To date, the brightest dust echo rings observed have come from four Galactic X-ray sources – [1E 1547.0-5408]{} [@Tiengo2010; @Olausen2011; @Pintore2017], [Cir X-1]{} [@Heinz2015], [V404 Cygni]{} [@Vasilopoulos2016; @Heinz2016; @Beardmore2016], and 4U 1630-47 [@Kalemci2018]. Dust echoes can also be produced by the X-ray components of gamma-ray bursts (GRBs), which scatter off the nearby Galactic medium [@Vaughan2004; @Vaughan2006; @Tiengo2006; @Vianello2007; @Pintore2017b]. Table \[tab:flares\] lists the approximate soft X-ray fluence and ISM column for dust echo rings observed around GRBs and XRBs. In most of the GRB cases, fluences were measured from the properties of the dust scattering echo, and the results depend on the adopted grain size distribution.
[lcccl]{} & [**Telescope(s)**]{} & [**Fluence**]{}$^a$ & [**Gal.**]{} ${{\rm N}_{\rm H}}$ & [**References**]{}\
& & (erg cm$^{-2}$) & ($10^{22}$ cm$^{-2}$) &\
GRB 031203 & [[*XMM-Newton*]{}]{}& $0.7-3 \times 10^{-6}$ & 0.6 & @Vaughan2004 [@Watson2006]\
& & & & @Tiengo2006\
GRB 050713A & [[*XMM-Newton*]{}]{}& $5 \times 10^{-7}$ & 0.1 & @Tiengo2006\
GRB 050724 & [[*Swift*]{}]{}& $2 \times 10^{-7}$ & 0.6 & @Vaughan2006\
GRB 061019 & [[*Swift*]{}]{}& $5 \times 10^{-7}$ & 0.9 & @Vianello2007\
GRB 070129 & [[*Swift*]{}]{}& $7 \times 10^{-7}$ & 0.1 & @Godet2007 [@Vianello2007]\
GRB 160623A & [[*XMM-Newton*]{}]{}& $2 \times 10^{-6}$ & 0.7 & @Pintore2017b\
1E 1547.0-5408 & [[*Swift*]{}]{}, [[*XMM-Newton*]{}]{}& $2-6 \times 10^{-3}$ & 3.0 & @Tiengo2010 [@Halpern2008]\
Cir X-1 & [[*Chandra*]{}]{}& 0.025 & 2.0 & @Heinz2015\
V404 Cygni & [[*XMM-Newton*]{}]{}, [[*Chandra*]{}]{}& 0.01 & 0.6 & @Heinz2016\
4U 1630-47 & [[*Swift*]{}]{}, [[*Chandra*]{}]{}& 0.015 & 13.0 & @Kalemci2018, estimated from MAXI\
\
One can note from Table \[tab:flares\] that the fluences of GRBs producing dust echoes are particularly low. In these cases, $\sim 10\%$ of the X-ray light from the flare is deposited into the dust scattering ring. This level rivals the amount of light in the telescopes’ point spread function (PSF) wings. However, due to the quick dimming typical of X-ray afterglows, the time delay between the prompt X-ray emission and the dimming afterglow allows the the dust scattering rings to stand out in contrast, even when a central X-ray point source is visible. In theory, X-ray variability from any high-redshift object can produce echoes that propagate off dust from foreground galaxies or the intergalactic medium, but it requires more sensitive telescopes than currently available [@Miralda-Escude1999; @Corrales2012; @Corrales2015a].
Many Galactic XRBs are persistent sources of X-rays, producing a quiescent dust scattering halo. A source that experiences frequent outbursts, or high variation, will create a time variable scattering halo with no clearly defined rings. This study focuses on identifying single, large amplitude outbursts capable of producing thin, high contrast dust echo rings. This work evaluates the X-ray light curves from all sources monitored regularly by MAXI [Monitor of All-sky X-ray Image, @MAXI2009], in order to gather the rate of X-ray flares propagating through the interstellar medium (ISM). In Section \[sec:analysis\], we describe the algorithm used to identify flares and discuss its limitations. In Section \[sec:results\], we calculate the number distribution of flares identified in MAXI. A metric for evaluating the likelihood of an outburst to produce sharp ring echoes is discussed in Section \[sec:rings\]. In Section \[sec:future\], we fit a power law to the number distribution of flares and use it to estimate how many X-ray dust ring echoes will be seen with the next generation of X-ray telescopes. We also update the results of @Corrales2015a to estimate the number of X-ray scattering echoes that might be found arising from dust in the intergalactic medium. All findings are summarized in Section \[sec:conclusions\].
Data Analysis {#sec:analysis}
=============
MAXI is the longest operating current all-sky monitor for the soft X-ray band, which is sensitive to dust scattering. The Soft X-ray Large Angle Camera on MAXI captures almost the entire sky on a cadence of 92 minutes, over the 0.7 - 12 keV energy band, with a binned one-day sensitivity limit of 4.5 mCrab ($10^{-10}$ erg s cm$^{-2}$) [@MAXI2009; @MAXI2014]. Because the ISM preferentially removes soft X-rays, the spectral energy distribution of X-ray scattering halos tend to peak around 1-3 keV. We use the publicly available one-day binned 2-4 keV light curves from 398 point sources currently available on the MAXI website[^1], from the start of MAXI operations to MJD 58408, to estimate the probability distribution of soft X-ray flares across the sky. For the purposes of this work, a flare is defined as any duration longer than several days for which an object’s flux is $> 3\sigma$ above its dimmest state.
It should be noted that MAXI is insensitive to most flares that are significantly shorter than a single ISS orbit. Magnetars, Type I neutron star bursts, and X-ray afterglows to GRBs fall into this category. As will be demonstrated below, such flares are missed from this study due to limitations of MAXI, not due to data analysis choices.
Detection Algorithm {#sec:algorithm}
-------------------
First, we removed all data points where a monitored object was within 10 degrees of the Sun, which is a large source of contamination. All light curves were smoothed using a 3-day Gaussian convolution, to improve the stability of the algorithm. Limitations imposed by smoothing are discussed in §\[sec:limitations\].
For a baseline flux, we selected the 16th-percentile value from the distribution of flux values within the whole light curve. We then subtracted the baseline value from the light curve and calculated the signal-to-noise ratio for each bin. Flares were identified by flagging light curve intervals with a signal-to-noise ratio greater than three. Flare intervals shorter than five days were discarded. If two flares were separated by an interval shorter than five days, we combined them into one time interval. This process was repeated so that there was no quiescent period shorter than five days. These choices were motivated by analysis of a simulated dataset, described in §\[sec:simulations\]. The five day cut-off significantly reduced the number of falsely identified flares.
Finally, we calibrated each light curve by normalizing them with the MAXI light curve for the Crab pulsar, which has a 2-4 keV band flux of $1.1 \times 10^{-8}$ erg cm $^{-2}$ s$^{-1}$. We used linear interpolation over data gaps to arrive at a total fluence for each flare interval.
Figure \[fig:demo\] shows the results for three light curves of interest. LMC X-3 exhibits erratic behavior with no clearly defined quiescent state. The algorithm flags the intervals when LMC X-3 is in a bright state. The next two panels show Cir X-1 and 4U 1630-47 during the flares leading to dust echoes study by @Heinz2015 and @Kalemci2018, respectively. The calculated fluences are 0.023 erg cm$^{-2}$ over 85 days (Cir X-1) and 0.021 erg cm$^{-2}$ over 171 days (4U 1630-47). These values are consistent with those in the published works.
We were unable to check on the dust echo producing flares from 1E 1547.0-5408, which occurred before the launch of MAXI, and V404 Cygni, which was only observable by the degraded GSC3 instrument at the time of the flare [@Negoro2015]. The publicly available MAXI light curves do not include data from GSC3. The MAXI view of V404 Cygni is also affected by source confusion with Cyg X-1, which is usually much brighter. As a result, the light curve is poorly calibrated and no flares were measured from V404 Cygni.[^2]
![ Flare intervals identified from three example light curves. The raw 2-4 keV light curves [**supplied by the MAXI website**]{} (blue) are smoothed with a 3-day Gaussian kernel (black). After subtracting a baseline flux value, identified using the lower 1-$\sigma$ value of the dataset (dashed black line), intervals with signal-to-noise greater than three are flagged as flares (shaded grey regions). For X-ray binaries that rarely stay quiescent, such as LMC X-3 (left), any bright state is flagged as a flare. Two dust-echo producing flares from Cir X-1 (middle) and 4U 1630-47 (right) are highlighted. The calculated fluences are consistent with [**those reported**]{} in the literature. \[fig:demo\] ](demo_plot.pdf)
Sensitivity limits {#sec:simulations}
------------------
To examine the accuracy of our analysis, we simulated $1000$ MAXI light curves of 900 days long, and injected one Gaussian flare into each. The reported MAXI sensitivity is 4.5 mCrab for one-day binned data [@MAXI2009], yielding a fluence theoretical lower limit of $5 \times 10^{-5}$ erg cm$^{-2}$ for a five day long flare. As such, the flare properties were drawn from a uniform distribution of fluences, $\log({\mathcal{F}_{\rm cgs}}) \in [-5, -1]$ (where ${\mathcal{F}_{\rm cgs}}$ is fluence in units of erg cm$^{-2}$), and a uniform distribution of Gaussian widths $\sigma ({\rm days}) \in [1, 50]$.[^3] The baseline flux and error bars for each simulated light curve were drawn randomly from three MAXI light curves of objects with a quiescent flux below the sensitivity limits, i.e., those exhibiting a light curve consistent with zero flux throughout: 1ES 1101-23.2, WW Cet, and VY Ari. We take these objects as representative of the zero values and error bars arising from the MAXI calibration processes.
Figure \[fig:SimulatedFlares\] shows the distribution of detected flares compared to the input distribution. In general, the algorithm returns a large number of short $10^{-4}$ erg cm$^{-2}$ flares ($<$ 3-5 days) that appear to arise from the noise typical in the MAXI dataset. We were able to cut down on the number of false positives significantly by ignoring flares shorter than five days, and by combining flares that were separated by less than five days. The dotted line in Figure \[fig:SimulatedFlares\] shows how many flares for which the fluence was correctly retrieved to within 20%. For the subset of flares with fluence $> 10^{-3}$ erg cm$^{-2}$, our algorithm was able to identify 90% of all the simulated flares and 100% of those that were of duration $\leq 20$ days. We therefore take $\log {\mathcal{F}_{\rm cgs}}> -3$ as the completeness limit for this study.
Identification of long, high fluence flares is limited by the flux in each bin. We estimated the flux sensitivity of our algorithm by dividing each fluence value by its corresponding duration, yielding 4 mCrab, which is consistent with the expectations for one-day binned MAXI data.
![A histogram of simulated input (light grey) and the histogram derived with the techniques described in Section \[sec:analysis\] (dark grey). The input flares were drawn from a uniform distribution, and the expected $1\sigma$ variation is shaded in blue. The dotted black histogram shows how many of the output flares were correct identifications, to within 20% of the input fluence. \[fig:SimulatedFlares\] ](process_sims.pdf)
Limitations for short outbursts {#sec:limitations}
-------------------------------
Using daily binned light curves and three-day smoothing imposes selection effects against short flares. In the extreme case of a flare restricted to a single ISS/MAXI orbit (90 minutes), the signal-to-noise of the flare in one-day binned data will be reduced by a factor of $\sqrt{15}\sim 4$ compared to the single-orbit light curve. This would eliminate single-orbit flares of signal-to-noise lower than 12 ($< 80$ mCrab) from detection, corresponding to a fluence $< 10^{-5}$ erg cm$^{-2}$. This value is several orders of magnitude below the fluence values of interest for dust echo tomography with Galactic X-ray sources (lower portion of Table \[tab:flares\]), which is the main target of this study.
In conclusion, smoothing data has the advantage of reducing the number of false positives, because the variance of the data is significantly reduced. The critical metric for the detectability of dust scattering echoes of flares is their fluence, which is preserved in binning and smoothing. Section \[sec:simulations\] demonstrates that the algorithm retrieved all of the short duration flares with $\log {\mathcal{F}_{\rm cgs}}> -3$, which is an order of magnitude below our threshold of interest demonstrated by the lower portion of Table \[tab:flares\]. Thus, the benefits of using binned and smoothed data outweigh the reduction of sensitivity to short flares.
Results and Discussion {#sec:results}
======================
Of the 398 sources analyzed, 213 exhibited outbursts that were picked up by the flare detection algorithm. To account for source confusion, we evaluated the light curves of three sources within $2^{\circ}$ of each other: SMC X-1, SMC X-3, and MAXI J0057-720. One flare from SMC X-3 appeared coincidentally in the light curve of MAXI J0057-720, which is 0.6$^\circ$ away. However, variations from SMC X-1, which is the brightest of the three and 2$^\circ$ away from the other two objects, did not affect the light curves of either. Thus we chose $1^\circ$ as the threshold for evaluating the effects of source confusion. We identified pairs of sources in the MAXI dataset separated by $< 1^\circ$. Within this subset of light curves, we searched for flares appearing within 30 days of each other. When coincident flares were found, we kept the larger fluence event and discarded the other. We also visually evaluated the light curves of sources within 2$^\circ$ of the Galactic Center, which hosts a large number of variable compact objects that cannot be resolved with MAXI. The overall process resulted in the removal of 8 flares that were double counted, leaving a total of 854 distinct outbursts with $\log {\mathcal{F}_{\rm cgs}}> -3$.
Figure \[fig:FlareHist\] shows a histogram of total number of flares detected as function of fluence (black). We used the [*Chandra*]{} X-ray Center tool `colden` to look up the ${{\rm N}_{\rm H}}$ value from HI surveys, in order to estimate the optical depth of X-ray scattering at 1 keV. We then multiplied the fluence of each flare by a factor of $(1 - e^{-\tau})$ to estimate the integrated fluence of the resulting dust echo (orange).
A higher instrument background makes it difficult to observe a dust scattering halo. A small fluence (relative to the quiescent state) will produce a small perturbation in the scattering halo brightness that is unlikely to be observable. To avoid modeling the problem, we use the examples of spectacular dust ring echoes from the literature (bottom portion of Table \[tab:flares\]) to arrive at approximate thresholds for observation with modern-day X-ray telescopes. We chose flares with $\log {\mathcal{F}_{\rm cgs}}> -2$ as candidates for producing dust echo rings. However, some of these flares may appear bright due to having a low ISM column. At the same time, bright flares offer the chance to produce serendipitous results in high contrast. For example, V404 Cygni has ${{\rm N}_{\rm H}}\approx 6 \times 10^{21}$ cm$^{-2}$ or $\tau \approx 0.1$. The corresponding estimate yields the dimmest available dust echo fluence ($\approx 0.006$ erg cm$^{-2}$) in Table \[tab:flares\], yet V404 Cygni produced some of the clearest multi-structured dust echo rings [@Heinz2016]. We take 0.005 erg cm$^{-2}$ as an approximate threshold for effective dust echo tomography with modern day instruments.
Because one object can produce multiple flares, we counted the number of MAXI targets that exhibited a flare with fluence larger than a given threshold, yielding the number of sight lines available for dust echo tomography (Figure \[fig:FlareHist\], right). We found that 34 of the objects exhibited flares with $\log {\mathcal{F}_{\rm cgs}}> -2$, and 24 of these have predicted dust echoes over the 0.005 erg cm$^{-2}$ threshold during the last 9 years of MAXI operation. However, more analysis is needed to determine which of these would have produced the thin, high contrast rings that are ideal for measuring the line-of-sight dust distribution.
Identifying Sources of Dust Echo Rings {#sec:rings}
--------------------------------------
High fluence flares are necessary to produce dust scattering halos, but many of them have a large fluence simply because they are long. Two other conditions are important for identifying dust echo candidates. First, flares must be short enough to produce sharp rings. Second, the bursts must be accompanied by a long period of quiescence so that the dust echo rings stand out in contrast to the quiescent dust scattering halo. A survey by @Valencic2015 showed that a majority of X-ray scattering halos are dominated by scattering from a single cloud, rather than isotropically distributed dust. The time delay associated with a particular angle can be inverted to solve for the angle at which a dust scattering echo will appear ($\theta$), from a burst that occurred at some time ($t$) prior to now: $$\theta = \left[ \frac{2c~(1-x)~t}{xD}\right]^{1/2}$$ where $D$ is the distance to the X-ray source, $x$ is the distance to [**a**]{} dust cloud divided by $D$, and $c$ is the speed of light [@TS1973]. This equation can be used to fit dust echo rings with multiple discrete ISM clouds, or, applying a convolution with the flare light curve, can be used to measure contiguous line of sight dust abundances [@Heinz2015; @Heinz2016]. While clouds or dust material that are extended along the line of sight will alter the perceived thickness and time delay of a dust echo, a full examination of these geometric effects is beyond the scope of this work.
For a fixed dust cloud position, the thickness of a dust echo ring ($\Delta \theta$) will depend on the duration of the flare ($t_f$) so that $\Delta \theta \propto t_f^{1/2}$. In contrast, the dust scattering halo will return to its quiescent state out to some angle, $\theta \propto t_q^{1/2}$ where $t_q$ is the duration of the quiescent period before or following the flare. We do not set a maximum duration for $t_f$. All scattering halos dim in surface brightness at large angular distance from the central source source, so a return to a dim quiescent state will always produce the appearance of rings. This was apparent from the outburst of 4U 1630-47, lasting over 100 days, which produced an 8 arcminute scale ring [@Kalemci2018]. Ideal echoes will have thin rings relative to the size of the quiescent halo, requiring $t_f / t_q << 1$.
Figure \[fig:Interpret\] shows the relationship between flare duration, fluence, and $t_f/t_q$ for the flares identified in the MAXI dataset. We determined the duration of the quiescence directly before and after each flare, and chose the larger $t_q$ value. Smaller values of the $t_f/t_q$ lead to thinner dust echo rings. The dust echoes arising from the Cir X-1 flare [@Heinz2015] and 4U 1630-47 [@Kalemci2018] were both high fluence with moderate values of $0.1 < t_f/t_q < 1$. The flares from LMC X-3 are highlighted in Figure \[fig:Interpret\] (orange) to demonstrate a population of flares arising from a highly variable source with no persistent quiescent state. We found nine objects that produced bright $> 10^{-2}$ erg cm$^{-2}$ flares detected by MAXI with $(t_f/t_q) < 0.1$. Of these, four have an estimated dust echo fluence $> 0.005$ erg cm$^{-2}$. These four have not been followed up or published: LS I +61 303, V\* BQ Cam, XTE J1752-223, and MAXI J1535-571.
{width="\textwidth"}
Figure \[fig:Interpret\] (right) shows the number density of flares as a function of 2-4 keV fluence, which follows a power law of slope $-1.77 \pm 0.03$ (black). We also calculated the fluence distribution for dust echoes with $t_f/t_q < 1$ (orange). The distribution follows a power law for $\log {\mathcal{F}_{\rm cgs}}> -3.0$ and is flat for lower fluences, due to our sensitivity limit. For this reason, we limit analysis to the $\log {\mathcal{F}_{\rm cgs}}> -3.0$ dust echo distribution, which fits with a power law of slope $-2.3 \pm 0.1$. In the next section, we use this trend to estimate the number of dust echoes that will be seen by the next generation of X-ray observatories.
Avenues for Future Study {#sec:future}
------------------------
![ Predicted number of dust scattering echoes observable by different telescopes, depending on the effective area and background levels, as compared to [[*Chandra*]{}]{}. For the next generation of telescopes ([*Athena*]{}, [*Lynx*]{}, and AXIS) we expect about 30 times more dust echoes than observable with current instruments, depending on the background levels achieved. \[fig:SensitivityCurves\] ](echo_sensitivity.pdf)
In the future, more sensitive X-ray telescopes will extend dust echo tomography to dimmer flares, opening up more sight lines for probing the 3D distribution of dust via X-ray scattering. For a fixed exposure time, we solved for the fluence ($f$) for which the signal-to-noise ratio is the same as the signal-to-noise ratio for a telescope with no background ($f_0$). $$\label{eq:fluxthreshold}
f = \frac{f_0}{2} \times \left(1 + \sqrt{1 + 4b/f_0}\right)$$ where $b$ is the background surface brightness.[^4] Since any flux threshold is inversely proportional to the effective area ($a$), we substitute $f_0$ with $1/a$ in Equation \[eq:fluxthreshold\]. We calculated $f$ for a grid of effective areas and backgrounds, relative to [[*Chandra*]{}]{}. We then calculated the total number of scattering ring echoes expected ($N$) by integrating the predicted fluence density distribution for echoes, $dN/df$ (Figure \[fig:Interpret\], orange), extrapolating the power law to fluences with $\log {\mathcal{F}_{\rm cgs}}< -3$. Figure 5 shows several contours for $N$, predicting the observable number of high signal-to-noise scattering ring echoes, relative to the [*Chandra*]{} effective area and background surface brightness.
[*Athena*]{}, expected to launch around 2030, will have a 1 keV effective area of 2 m$^2$ [@Athena2017], approximately 80 times the current soft X-ray effective area for [*Chandra*]{} ACIS-I. [*Athena*]{} will thereby observe on the order of 30 times more dust echoes than [*Chandra*]{} can, depending on the instrument background levels. The concept mission, [*Lynx*]{}, will have a similar effective area to [*Athena*]{} with the imaging resolution of [[[*Chandra*]{}]{}]{}. The Advanced X-ray Imaging Satellite (AXIS) concept mission has a proposed 1 keV effective area of 7000 cm$^{2}$ with 10-20 times lower background than [[*Chandra*]{}]{} [@AXISspie]. Thus AXIS would be able to image a similar number of dust echoes to [*Athena*]{} and [*Lynx*]{}.
The increased sensitivity offered by the next generation of X-ray telescopes will also constrain the abundance and distribution of dust in the intergalactic medium (IGM) through dust scattering echoes left behind by previously active galactic nuclei (AGN). Using the formulations of @Corrales2015a, a telescope with ten times the [*Chandra*]{} sensitivity will be able to image IGM scattering echoes on the order of $20''$ - $80''$ in radius, corresponding to AGN activity $\sim 10^2$ - $10^3$ years prior. Using the numbers of bright $z > 1$ AGN visible from all-sky surveys, the number of echoes one can expect to find in the entire sky is: $$N^{\rm IGM}_{\rm ech} \sim 10-100
\left( \frac{\nu_{fb}}{10^{-3}~{\rm yr}^{-1}} \right)$$ where $\nu_{fb}$ is the characteristic frequency for rapid quenching of an AGN accretion flow. We refer the reader to the original work of @Corrales2015a for a detailed discussion on how AGN variability and feedback can be constrained by IGM dust echoes.
Conclusions {#sec:conclusions}
===========
Examination of nine years of MAXI light curves reveals 34 objects that exhibited bright X-ray flares with fluences $> 10^{-2}$ erg cm$^{-2}$, with durations $\sim 30$-300 days. By comparing the flare duration to the time in quiescence, we estimate that nine of these were short enough to produce sharp ring echoes: approximately one candidate per year. Using ${{\rm N}_{\rm H}}$ to estimate the dust echo brightness, four of the flares might have produced dust echo rings detectable by current X-ray telescopes. Only one of these sight lines, Cir X-1, has been imaged and studied in detail.
With the next generation of X-ray telescopes, dust ring echoes will become common features of the Galactic ISM. We expect [*Athena*]{}, [*Lynx*]{}, and AXIS to be $\geq 30$ times more sensitive to dust echoes in comparison to [[*Chandra*]{}]{}. The result will be hundreds of time-variable X-ray scattering halos. Of these, we expect $\sim$10-30 sharp dust ring echoes per year, which are ideal for determining 3D distributions of ISM dust with the detail of @Heinz2015 [@Heinz2016].
Despite this work focusing on thin ring dust echoes, all bright X-ray sources have dust scattering halos that vary with the light curve of the central source. About half of the flares found in this study had $t_f/t_q > 1$. The resulting image will be a blend of broad rings. Interpreting these images will require more advanced dust scattering halo timing techniques. The results will open an avenue for mapping Galactic and intergalactic structures in an entirely new way.
We wish to thank the anonymous referee for their thoughtful comments that greatly improved the clarity of the paper. This research has made use of MAXI data provided by RIKEN, JAXA, and technical support from the MAXI team. Support for this work was provided by NASA through Einstein Postdoctoral Fellowship grant number PF6-170149 awarded by the Chandra X-ray Center (CXC), which is operated by the Smithsonian Astrophysical Observatory for NASA under contract NAS8-03060. Additional support for this work came from CXC through grant number TM6-17010X.
[^1]: http://maxi.riken.jp/top/lc.html
[^2]: Communication with MAXI calibration staff
[^3]: A 900 day light curve was deemed sufficient to capture flares that are effectively 300 days long (Gaussian $\sigma = 50$ days). As shown later in Section \[sec:rings\], these very long outbursts are typically beyond the scope of interest for dust echo tomography.
[^4]: The background is due to a combination of instrumental, charged particle, and cosmic X-ray background, which change with time and position on the sky. [[*Chandra*]{}]{} has relatively low, stable, and well documented background rates compared to other currently active X-ray telescopes, making [[*Chandra*]{}]{} a good baseline for comparison.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
PG 1002+506 is found to be a Be star, one of two so far found by the Palomar-Green survey. Its spectrum is classified as a B$5\pm 1$ Ve, with $T_{\rm eff} = 14,900 \pm 1200,$ $\log g = 4.2 \pm 0.2,$ and $v \sin i =
340 \pm 50$ km s$^{-1}.$ At $b = +51^{\circ},$ its height above the Galactic plane would therefore be $Z = +10.8$ kpc, putting this apparently young, rapidly rotating star well into the Galactic halo. Its heliocentric radial velocity is found to be $-2 \pm 15$ km s$^{-1},$ consistent with either having been formed in the Galactic disk and subsequently ejected, or having been formed in the halo.
author:
- 'F. A. Ringwald'
- 'W. R. J. Rolleston'
- 'R. A. Saffer'
- 'John R. Thorstensen'
title: 'PG 1002+506: a Be Star Apparently at Z $>$ +10 kpc'
---
Introduction
============
PG 1002+506 was discovered by the Palomar-Green UV-excess survey (Green, Schmidt, & Liebert 1986) and listed as a cataclysmic variable (CV). During a study of the CVs from this survey, Ringwald (1993) obtained ultraviolet and red spectra, and tentatively reclassified it as a detached subdwarf binary, noting H$\alpha$ in strong emission, unresolved at 10-Å resolution. Several puzzling aspects were noted, however, including the near-constancy of the radial velocities throughout two nights, consistent with no change other than that attributable to atmospheric dispersion in an unrotated slit. There was also no significant variation in the equivalent width of H$\alpha,$ which one might expect if this were a detached CV progenitor with the hot component irradiating the facing hemisphere of its companion.
That PG 1002+506 is not a CV was shown definitively by E. L. Robinson (1995, private communication): it does not flicker, or have the erratic variability ubiquitous in CVs. This was found with high-speed simultaneous [*UBVR*]{} photometry taken in 1995 June with the Stiening photometer on the McDonald Observatory 2.1-m telescope. In 25 min of photometry with 1-s time resolution, all bands showed peak-to-peak amplitudes of $<\,2$%.
This and further spectra have forced another reclassification of this star, as a high-latitude Be star. This is one of two known in the Palomar-Green catalog, the other being PG 0914+001 (Saffer et al. 1997). An Oe star from this survey is also known, PG 2120+062 (Moehler, Heber, & Dreizler 1994).
For reviews on Be stars, see Jaschek & Jaschek (1987) and Slettebak (1988). About one in five non-supergiant B stars shows emission, mainly in H$\alpha$ but sometimes also in H$\beta$ and higher Balmer lines. Struve (1931) attributed this to a disk, extruded by the star’s rotation near the breakup velocity, $ \sqrt{ G M / R}.$ What excites the emission in Be stars is a long-standing mystery, however, as is their evolutionary status. Although Be stars often have an IR excess, PG 1002+506 is not an IRAS source (Neugebauer et al. 1988).
Blue spectrum
=============
A blue spectrum (Figure 1) was taken in service time with the Intermediate Dispersion Spectrograph on the Isaac Newton Telescope on La Palma. This 1800-s spectrum was taken in photometric conditions in $2''$ seeing, through a $1.73''$ slit, and has 1.5-Å (FWHM) resolution. The slit was aligned to the parallactic angle, to avoid atmospheric dispersion effects; the spectrum was taken when PG 1002+506 was nearly overhead, at an airmass of 1.08.
A spectral classification of B$5\pm 1$ V was arrived at by comparing this spectrum to model atmospheres (Kurucz 1979) and published spectra (Jacoby, Hunter, & Christian 1984; Jaschek & Jaschek 1987). That this is a main-sequence star and not a subdwarf is shown by the presence of the H13 and 14 lines. That it is not a giant or supergiant is shown by the widths of its Balmer lines, with FWZI of H$\gamma$ of $ 31 \pm 3$ Å. There is no spectroscopic evidence that this star is a binary.
Radial Velocity
===============
On 1997 January 3 UT, two 10-min exposures were obtained with the Modular Spectrograph on the 2.4-m Hiltner Telescope at Michigan-Dartmouth-MIT Observatory, Kitt Peak, Arizona. The spectra covered from 4650 to 6727 Å, and had 4-Å (FWHM) resolution. The weather was poor, with $>
1\arcsec$ seeing and rising humidity that forced a shutdown just after these spectra were taken. The spectrograph slit was set at the parallactic angle, even though PG 1002+506 was only one hour east of the meridian. The 1 slit projected to 3 Å on the detector. With the mediocre seeing, we expect “slit-painting” velocity errors to be small, probably $< 5$ km s$^{-1},$ based on experience with similar sharp lines in white dwarf/red dwarf binaries (Thorstensen, Vennes, & Shambrook 1994). The exposures were bracketed by HgNeXe exposures, for which the RMS residual was $< 0.05$ Å, and the maximum residuals for the weakest lines were $< 10$ km s$^{-1}.$ Most lines had residuals around 2 km s$^{-1}.$
H$\alpha$ appears to be slightly resolved, and is in strong emission (see Figure 2), with an equivalent width of $17.8 \pm 0.3 $ Å and FWHM of $580 \pm 30$ km s$^{-1}.$ There is also emission in the core of H$\beta.$ By convolving H$\alpha$ with the derivative of a Gaussian with FWHM = 8 Å and taking the zero of the convolution as the velocity (Schneider & Young 1980), we find heliocentric radial velocities of the spectra taken at HJD 2450451.90425 and 2450451.91140 of +29.3 and +28.9 km s$^{-1},$ respectively. The velocities of the O I $\lambda$6300 Å night sky line were 1.6 and 0.7 km s$^{-1},$ showing the accuracy of the wavelength scale.
However, the emission lines in Be stars are well known to be variable in profile over timescales of days or longer, and are therefore not reliable indicators of the systemic velocity. The spectra were therefore summed together and rectified, to remove continuum slope effects. The radial velocity was then measured from the absorption wings of H$\alpha$ by convolving a positive and a negative Gaussian with the line profile and taking the zero of this convolution as the velocity (Schneider & Young 1980). In all cases the Gaussians had 4 channels FWHM. The separation between the Gaussians was varied, from 24 to 20 to 16 Å; the corresponding heliocentric radial velocities are $-2.0,$ $-0.5,$ and $-4.1$ km s$^{-1}.$ Finding the line’s centroid by fitting and subtracting a linear approximation of the continuum, numerical integration of the intensity, and taking the centroid (crudely, with the IRAF [*splot*]{} ‘e’ command) gave +0.4 km s$^{-1}.$ We conclude that PG 1002+506 has a heliocentric radial velocity of $-2 \pm 15$ km s$^{-1}.$
Model atmosphere analysis
=========================
We have performed a model atmosphere analysis of the blue optical spectrum to estimate the atmospheric parameters $T_{\rm eff}$ and log $g$, as well as the projected stellar rotation velocity $v \sin i$. Our grid of synthetic spectra was calculated with the radiative transfer code SYNSPEC (Hubeny, Lanz, & Jeffrey 1995), assuming the temperature and pressure stratifications of Kurucz (1991). The metal and helium abundances were held fixed at the solar value. At the temperature and surface gravity of spectral type B5V, the assumption of LTE is well justified. The temperature and gravity grid points were $T_{\rm eff}$ = 13,000 – 17,000 K in steps of 1,000 K, and log $g$ = 3.5 – 5.0 in steps of 0.5 dex. In addition each model was convolved with a rotational broadening function at projected rotation velocities $v \sin i$ = 50 – 350 km s$^{-1}$ in steps of 50 km s$^{-1}$ to produce a 3-dimensional fitting grid. The stellar parameters were estimated by simultaneous variation using a non-linear $\chi^2$ minimization algorithm. Details of the synthetic spectrum calculations and the fitting algorithm are given by Saffer et al. (1994) and Saffer et al. (1997). Due to the partial filling in of the lower Balmer lines by emission from the circumstellar material, we have restricted the analysis to the portion of the spectrum blueward of H$\beta.$
The best-fit stellar parameters are $T_{\rm eff} = 14,900 \pm 1200$ K, $\log g = 4.20 \pm 0.2$, and $v \sin i = 340 \pm 50$ km s$^{-1}$ (see Figure 1). The quoted 1-$\sigma$ errors are based on counting statistics and account for covariance for the fitting parameters; they also estimate systematic errors.
Evolutionary status
===================
The effective temperature, surface gravity, and very high rotational velocity are fully consistent with a spectral classification of B5Ve. The breakup velocity expected for this star is 540 km s$^{-1}.$ The fit places this star in the area of confusion in the $T_{\rm eff}/\log g$ diagram where the Population I main- sequence intersects the Population II blue horizontal branch (BHB) (Schönberner 1993; Bertelli et al. 1994). For example, PG 0832+676 at first appeared to be a young star far from the Galactic plane, but turned out to be a nearby blue evolved star, upon analysis of high-resolution spectra (Hambly et al. 1996). However, identification of PG 1002+506 as a BHB star is contradicted both by the emission reversals in the H$\alpha$ and H$\beta$ absorption lines, and by its high rotation velocity, since BHB stars are slow rotators (Peterson, Rood, & Crocker 1995).
Assuming PG 1002+506 to be of Population I origin, we used the derived atmospheric parameters and the evolutionary tracks of Claret & Gimenez (1992) to estimate the stellar mass and evolutionary age (see Table 1). A distance estimate was obtained from the absolute visual magnitude deduced from the stellar mass, atmospheric parameters, and bolometric corrections of Kurucz (1979). PG 1002+506 has $B = 15.36$ (Green et al. 1986). Assuming $B - V = -0.16$ for B5V stars (Allen 1973), and a reddening $E(B-V) < 0.01$, inferred from the map of Burstein & Heiles (1982), this would imply a distance of 13.9 kpc, which for a Galactic latitude $b =
51^{\circ},$ corresponds to a z-distance of 10.8 kpc above the Galactic plane. Although large, this is not unheard of (Kilkenny 1992). For a Galactic longitude $l = 165^{\circ},$ this would imply a galactocentric radius of 17.1 kpc, putting PG 1002+506 at the outskirts of the Galaxy.
Kinematical analysis
====================
As the existence of young objects at large distances from the star forming regions of the Galactic disk is potentially interesting, we have performed a kinematical analysis for PG 1002+506. Although no proper motion information is available, it is possible to use the observed radial velocity of a star to constrain its evolutionary history. A detailed description of the method of analysis is given by Rolleston et al. (1997).
We first consider a scenario whereby PG 1002+506 has a zero velocity component parallel to the Galactic disk, and ejection has occurred perpendicular to the plane of the Galaxy. We have corrected the observed heliocentric velocity for the effects of differential rotation (Fich et al. 1989), assuming that the halo co-rotates with the disk, to determine the stellar radial motion $(v_r)$ with respect to a standard of rest defined by its local environment. Our initial assumption implies that the observed radial velocity is a component of the stellar space motion $(v_z)$ perpendicular to the disk. We then attempt to show that PG 1002+506 could have reached its present position in the Galactic halo within its evolutionary lifetime, while reproducing the observed radial velocity, and calculating the required ejection velocity. These calculations have adopted the gravitational potential function of House & Kilkenny (1980). This analysis implicitly assumes that the star is ejected from the disk shortly after birth, consistent with cluster ejection simulations.
The results of the kinematical analysis are given in Table 1. Given the large z-distance, it is not surprising to find the “time of flight” to be larger than the evolutionary age. We have therefore considered the effects of errors in the derived atmospheric parameters and the radial velocity measurement. By optimizing the values of $T_{\rm eff}$ and $\log
g$ such that they are self-consistent within the errors, it is possible to increase the evolutionary age, so that it is greater than the predicted flight time. For example, adopting values of $T_{\rm eff} = 13,750$ K and $\log g = 4.0$ would imply an age of 115 Myr for a mass of 4.0 $M_{\odot}.$ Allowing an error of 15 km s$^{-1}$ in the observed heliocentric velocity also decreases the estimated flight time, but not significantly, to 84 Myr.
Conclusions
===========
PG 1002+506 appears to be a young, rapidly rotating B5Ve star at a distance of 10.8 kpc from the Galactic plane, and at a galactocentric radius of 17.1 kpc. The kinematical analysis suggests that it could have attained its present Galactic position having been ejected from the disk shortly after its formation. Furthermore, the required ejection velocity of $\approx 230$ km s$^{-1}$ can also be produced by the known mechanisms predicted by Leonard (1993). A detailed atmospheric analysis with higher-quality spectra should still be done, to determine abundances and confirm that PG 1002+506 really is a distant main-sequence star, and not a nearby blue evolved star. If PG 1002+506 really is 10.8 kpc from the Galactic plane, interstellar absorption in this same spectrum would probe a line through the Galactic halo otherwise difficult to acquire.
E. Harlaftis took the blue spectrum with the Isaac Newton telescope, which is operated on La Palma by the Royal Greenwich Observatory at the Spanish Observatorio del Roque de los Muchachos of the Instituto de Astrofisica de Canarias. Michigan-Dartmouth-MIT Observatory is operated by a consortium of the University of Michigan, Dartmouth College, and the Massachusetts Institute of Technology. Thanks also to Rob Robinson, Malcolm Coe, Richard Green, Uli Heber, Gerrie Peters, and Richard Wade, for helpful discussions.
Allen, C. W. 1973, Astrophysical Quantities, London, Althone Press
Bertelli, G., Bressan, A., Chiosi, C., Fagotto, F., & Nasi, E. 1994, A&AS, 106, 275
Burstein, D., & Heiles, C. 1982, AJ, 87, 1165
Claret, A., & Gimenez, A. 1992, A&AS, 96, 255
Fich, M., Blitz, L., & Stark, A. A. 1989, ApJ, 342, 272
Green, R. F., Schmidt, M., & Liebert, J. 1986, ApJS, 61, 305
Hambly, N. C., Keenan, F. P., Dufton, P. L., Brown, P. J. F., Saffer, R. A., & Peterson, R. C. 1996, ApJ, 466, 1018
House, F., & Kilkenny, D. 1980, A&A, 81, 251
Hubeny, I., Lanz, T., & Jeffrey, C. S. 1995, “[*Synspec – A User’s Guide*]{}”, private communication
Jacoby, G. H., Hunter, D. A., & Christian, C. A. 1984, ApJS, 56, 257
Jaschek, C., & Jaschek, M. 1987, The Classification of Stars (Cambridge: Cambridge Univ. Press)
Kilkenny, D. 1992, in Variable Stars and Galaxies, ed. B. Warner (A. S. P. Conf. Ser., v. 30), 97
Kurucz, R. L. 1979, ApJS, 40, 1
Kurucz, R. L., 1991, in [*Stellar Populations in Galaxies*]{}, ed. A. Renzini, B. Barbuy, Kluwer, Dordrecht, 225
Leonard, P. J. T. 1993, in Luminous High-Latitude stars, ed. D. D. Sasselov (A. S. P. Conf. Ser., v. 45), 360
Moehler, S., Heber, U., & Dreizler, S. 1994, A&A, 282, L29
Neugebauer, G., et al. 1988, [*Infrared Astronomical Satellite (IRAS)*]{} Catalog and Atlases, Vol. 2 (NASA: Washington, DC), pp. 201 – 202
Peterson, R. C., Rood, R. T., & Crocker, D. A. 1995, ApJ, 453, 214
Ringwald, F. A. 1993, Ph. D. thesis, Dartmouth College
Rolleston, W. R. J., Hambly, N. C., Dufton, P. L., Keenan, F. P., Little, J. E., Kilkenny, D., O’Donoghue, D., Koen, C., Stobie, R. S. 1997, MNRAS, submitted
Saffer, R. A., Bergeron, P., Koester, D., & Liebert, J. 1994, ApJ, 432, 351
Saffer, R. A., Keenan, F. P., Hambly, N. C., Dufton, P. L., & Liebert, J. 1997, ApJ, in press
Schönberner, D. 1993, in IAU Symp. 155, Planetary Nebulae, ed. R. Weinberger & A. Acker (Dordrecht: Kluwer), 415
Schneider, D., & Young, P. 1980, ApJ, 238, 946
Slettebak, A. 1988, PASP, 100, 770
Struve, O. 1931, ApJ, 73, 94
Thorstensen, J. R., Vennes, S., & Shambrook, A. 1994, AJ, 108, 1924
[ll]{} $T_{\rm eff}$ & $14,900 \pm 1200$ K\
$\log g$ & $4.2 \pm 0.2$\
$v \sin i$ & $340 \pm 50$ km s$^{-1}$\
$v_{breakup}$& 540 km s$^{-1}$\
Mass & 4.2 $M_{\odot}$\
Age & 50 Myr\
&\
$l$ & 165.072$^{\circ}$\
$b$ & 50.943$^{\circ}$\
$B$ & 15.36\
Distance & 13.9 kpc\
$R_{galactocentric}$ & 17.1 kpc\
&\
z-distance & 10.8 kpc\
$v_{heliocentric}$ & $-2 \pm 15$ km s$^{-1}$\
$v_z$ & 18.0 km s$^{-1}$\
$T_{flight}$ & 85 Myr\
$v_{ej}$ & 229 km s$^{-1}$\
&\
Age (OPT) & 115 Myr\
$v_z$ (OPT) & 39.0 km s$^{-1}$\
$T_{flight}$ (OPT) & 84 Myr\
$v_{ej}$ (OPT) & 249 km s$^{-1}$\
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The normal state behaviour of the density of states of the electrons described by the BOSON - FERMION model for Bose-Einstein condensation driven superconductivity is characterized by the appearence of a pseudogap which developes into a true gap upon lowering the temperature and the superconducting critical temperature is approached. The consequences of this on the temperature dependence of the specific heat, the NMR relaxation rate and the optical conductivity is examined.'
address: 'Centre de Recherches sur les Très Basses Températures, Laboratoire Associé à l’université Joseph Fourier, Centre National de la Recherche Scientifique, BP 166, 38042 Grenoble Cédex 9, France'
author:
- 'J. Ranninger and J. M. Robin'
title: 'Manifestations of the pseudogap in the BOSON-FERMION model for Bose-Einstein condensation driven Superconductivity'
---
( )
The opening of a pseudogap in the density of states (DOS) of the electrons in the normal state of high $T_c$ superconductors (HTcSC) is one of the characteristic features of these materials[@Battlogg-94]. One possible interpretation involve spin fluctuations as the underlying mechanism. This pseudogap is then called the spingap, the discussion of which has given rise to intense work based on the $t-J$ model or the nearly antiferromagnetic Fermi liquid. In this Letter, we want to adress an alternative mechanism for this pseudogap in terms of superconducting fluctuations. The existence of such a pseudogap together with the experimental indications that possibly two types of charge carriers (fairly localized ones and itinerant ones) are involved in the high $T_c$ phenomenon[@Timusk-93] supports a scenario of a mixture of intrinsically localized Bosons (tightly bound electron pairs) and itinerant electrons (Fermions). This situation can be described in its simplest form by the so called BOSON - FERMION (BM) model which was first introduced in connection with the many polaron problem in the cross-over regime between adiabatic and non-adiabatic behaviour. In such a scenario[@Ranninger-PhysicaB-85] bipolarons (Bosons) are envisaged to coexist with quasifree electrons (Fermions) and an exchange coupling between the Bosons and the Fermions is assumed by which Bosons can decay into pairs of itinerant Fermions and vice versa. The physically interesting regime of parameters of this model is that where the superconducting state below a certain critical temperature $T_c$ is controlled by a condensation of the Bosons[@Ranninger; @PhysicaC-95] and thus can in principle lead to rather high values of $T_c$. This happens when the Boson level lies close to the Fermi level of the Fermionic subsystem. Assuming the exchange coupling to be local, we have shown[@Ranninger-PRL-95] how, upon lowering the temperature, a pseudogap in the DOS of the Fermions gradually opens up - ultimately developing into a true gap below $T_c$. The opening of such a pseudogap is driven by the onset of itinerancy of the intrinsically bare localized Bosons due to a precursor effect of their superfluidity which implies a concomitant onset of strong local pairing of the itinerant Fermions. The increased correlations of the Fermions into Fermion-pairs results in a DOS for single particle excitations which, close to the Fermi level, is drastically diminished and thus leads to the appearance of a pseudogap and single particle excitations which show strong deviations from Fermi liquid behaviour[@Ranninger-PRL-95]. The underlying BF model on which this behaviour has been studied sofar is given by the following Hamiltonian
$$H \; = \; (zt - \mu) \sum_{i,\sigma} c_{i\sigma}^{+} c_{i\sigma} \; - \;
t \sum_{<i\neq j>, \sigma} c_{i\sigma}^{+} c_{j\sigma} \; + \; (\Delta_{_B}
- 2 \mu) \sum_{i} b_{i}^{+} b_{i}$$ $$\; + \; v \sum_{i} [ \; b_{i}^{+} c_{i\downarrow} c_{i\uparrow} \; + \;
c_{i\uparrow}^{+} c_{i \downarrow}^{+} b_{i} \; ]
\label{Equ1}$$ where $c^{(+)}_{i,\sigma}$ and $b^{(+)}_i$ refer to the Fermion and Boson annihilation (creation) operators at site $i$ and $\sigma$ denotes the spin quantum number. t represents the bare hopping integral for tight binding electrons, $\Delta_B$ the energy level for the bare localized Bosons and $v$ the local Boson- Fermion pair exchange. The chemical potential $\mu$ is taken to be common to both the Bosons and Fermions such as to ensure charge conservation during the Boson-Fermion exchange. We have previously evaluated the single particle Boson and Fermion spectral properties for a 1D system within the lowest order fully selfconsistent conserving approximation[@Baym-PR-62] and have shown explicitely the opening of the pseudogap and the destruction of Fermi liquid properties[@Ranninger-PRL-95]. In order to ascertain that these features were indeed unrelated to any physics of one dimensional systems we further considered the case of a 2D square lattice[@Ranninger-SSC-96] and obtained results which are qualitatively analogous the the 1D case thus confirming that the pseudogap in the DOS of the Fermions is an intrinsic feature of the BF model.
It is the purpose of this Letter to demonsrate how this opening of the pseudogap in the DOS of the Fermions and the destruction of the Fermi liquid properties show up in physically accessible thermodynamic (specific heat, compressibility), magnetic (NMR relaxation rate, spin susceptibility) and transport (optical conductivity) properties. We shall use the same approximative scheme as that which served us for the evaluation of the single particle properties previously[@Ranninger-PRL-95; @Ranninger-SSC-96] and which is based on a fully selfconsistent lowest order evaluation of the thermodynamic potential, given by the closed loop diagram illustrated in Fig.1 and presenting a functional of the full one particle Fermion et Boson Green’s functions. Within such a scheme the one and two particle Green’s functions are derived from this closed loop diagram by standard functional derivatives with respect to an external space-time varying field[@Baym-PR-62] which for our approximation gives rise to the following expressions for the Boson and Fermion selfenergies: $$\Sigma_{F}({\bf k}, \omega_{n}) \; = \; - \frac{v^{2}}{N} \; \sum_{{\bf q},
\omega_{m}} \; G_{F}({\bf q}-{\bf k}, \omega_{m} - \omega_{n}) \;
G_{B}({\bf q}, \omega_{m})$$ $$\Sigma_{B}({\bf q}, \omega_{m}) \; = \; \frac{v^{2}}{N} \; \sum_{{\bf k},
\omega_{n}} \; G_{F}({\bf q}-{\bf k}, \omega_{m}-\omega_{n})
\; G_{F}(
{\bf k}, \omega_{n})
\label{Equ2}$$ where $G_{B}({\bf q}, \omega_{m}) \,=\, [ i \omega_{m} \, - \, E_{0} \, -
\Sigma_{B}({\bf q}, \omega_{m})\,]^{-1}$ and $G_{F}({\bf k}, \omega_{n}) \, = \,[ i \omega_{n} \, - \,
\epsilon_{{\bf k}} \, - \, \Sigma_{F}({\bf k}, \omega_{n}) \,]^{-1}$ represent the fully selfconsistently determined Fermion and Boson one particle Green’s functions. The selfconsistent set of Eqs.(2) are solved numerically for a square lattice with sizes up to 41$\times$41 and for a set of Matsubara frequencies $\omega_{n}$ with $n$ up to $100$. This turns out to be enough to cover a wide enough temperature regime in order to track the evolution of the pseudogap in the DOS and its repercussions on the physical quantities which we want to discuss here. For computational reasons we work as usual with the difference between the total Green’s function and its zero order approximation; i.e for $v=0$[@trick]. In order to treat the physically most interesting situation of the BF model (where the superconducting phase is essentially due to a Bose condensation of the Bosons) we choose the model parameters such that the Bosonic level lies well inside the Fermion band and the number of Bosons per site $n_B=\sum_{i}\langle b^{\dag}_{i}b^{\phantom{dag}}_{i}\rangle$ is comparable to the number of Fermions per site $n_F= \sum_{i,\sigma}
\langle c^{\dag}_{i\sigma}c^{\phantom{dag}}_{i\sigma}\rangle$. For that purpose we choose as representative parameters for our numerical work: $\Delta_B=0.4$, $v=0.1$ in units of the bandwidth $8t$ and $n_{tot}= 2n_B\,+\,n_F\,=\,1$ per site.
The properties of the one particle spectral functions for the Bosons and Fermions have adequately been dealt with previously and we refer the reader to refs\[5,7\]. We hence shall not discuss them here in any further detail but rather concentrate on the evaluation of the specific heat, the NMR relaxation rate and the optical conductivity and show to what extent they are influenced by the opening of the pseudogap in the DOS and the breakdown of Fermi liquid properties of the Fermions. We evaluate for that purpose the total free energy $F\,=\,E-\mu N_{tot}-TS$ where the inner energy $E\,=\,\langle H_0 \rangle + \langle vH_1 \rangle +\mu N_{tot}$ is separated into a component of the uncoupled BF system and into that of Boson-Fermion exchange coupling. $N_{tot} \, =\,n_{tot} N$ where N is the total number of sites in the system. The expression for the exchange coupling contribution to $F$ can be obtained directely by evaluating the closed loop diagram (Fig.1) which yields $$\langle vH_1 \rangle \,=\, -\frac{2}{\beta} \sum_{{\bf q},
\omega_{m}} \Sigma_B({\bf q},\omega_{m}) G_B({\bf q},\omega_{m})$$ Inserting the solutions of Eq.2 into the above expression we evaluate F using $$F \; = \; F_{0} + \int_{0}^{v} \frac{d\lambda}{\lambda} \langle \lambda
H_{1} \rangle$$ where $F_{0}$ is the free energy of the non - interacting system, and consecutively derive the specific heat at constant volume $C_V\,=\,(dE/dT)$ and the entropy $S \,=\, (E-\mu N_{tot} - F)/T$[@check]. As can be seen from the temperature dependence of the DOS of the Fermions at the Fermi level $N(0)$(see Fig.2), a pseudogap starts opening up below a certain characteristic temperature $T^{\star}$ which for our choice of parameters is around 0.015. $T^{\star}$ shows up noticeably in the temperature behaviour of $C_V$ where it corresponds to a net upturn of $C_V$ which, with lowering the temperature below $T^{\star}$, increases as $ln T$. This behaviour can be traced back to the onset of a precursor to superfluidity of the Bosons which aquire coherency i.e. quasi free particle like behaviour with an effective mass which diminishes as the superconducting state is approached[@Ranninger-PRL-95; @Ranninger-SSC-96]. $T^{\star}$ is equally visible in the temperature behaviour of the entropy $S$ which at this temperature shows a noticeable deviation from linearity which is observed for higher temperatures and is due to effectively free Fermions. Both the inverse Boson mass and the compressibility show a monotonic increases with decreasing temperature with a cross-over to a much steeper rise below $T^{\star}$. The lowest temperature results for the specific heat, as well as for the Boson correlation function show a critical behaviour with a finite value of $T_{_C}$. We identify this transition, as it should be, as a Kosterlitz-Thouless transition[@Kost-Thoul], since one expects a Bose-Einstein condensation for a $2D$ system[@Fisher].
The onset of a pseudogap and a concommitant coherence of the Bosons is also visible in the magnetic response of the system measured by the magnetic susceptibility $$\chi ({\bf q},\omega) \,=\, \frac{1}{2\pi\,i\,\hbar}
\int d\tau e^{i\hbar\omega\tau} \;
\Theta(\tau)\langle [S^-({\bf q},\tau),S^+({\bf -q},0)] \rangle$$ where $S^+({\bf q},\tau)\,=\,c^+_{{\bf q}\uparrow}(\tau)
c^{\phantom{+}}_{{\bf q}\downarrow}(\tau)$ and $S^-({\bf q},\tau)\,=\,(S^+({\bf q},\tau))^{\dag}$. Due to the Boson-Fermion exchange coupling local magnetic correlations are induced among the bare uncorrelated electrons arising from the singlet character of the Bosons. The onset of the long range superfluid coherence of the Bosons leads to an onset of long range magnetic correlations which can be seen in the static homogeneous susceptibility $S_O\,=\,\frac{1}{2\pi}\chi(0,0)$ and the NMR relaxation rate $\frac{1}{T_1}\,=\,\frac{k_BT}{2\pi}
\sum_{\bf q} \chi"({\bf q},\omega)/\omega$ where $\chi"({\bf q},\omega)\,=\, \ Im\chi({\bf q},\omega)$. Evaluating the expression Eq(4) to within lowest order i.e, neglecting vertex corrections but fully taking into account the selfconsistent expressions for the Fermion one particle Green’s function we obtain the results for $(T_1T)^{-1}$ as a function of temperature as illustrated in Fig.(3). We again notice a drastic changeover of a fairly well represented temperature independent Koringa behaviour for $T>T^{\star}$ to a rapid drop of $1/(T_1T)$ below $T^{\star}$. Nevertheless even for $T>T^{\star}$ the usual Korringa ratio $(T_1T)^{-1}/S^2_0$ does turn out not to be temperature independent as it should be expected for free uncorrelated Fermions. On the contrary $(T_1T)^{-1}/S_0$ is fairly temperature independent for $T>T^{\star}$ as can be seen from Fig.(3) and tracks the temperature behaviour of $N(0)$.
As the last manifestation of the pseusogap in the DOS of the Fermions we want to discuss here the optical conductivity which is defined by $$\sigma^{\alpha \beta}(\omega)\,=\,\ Im\frac{1}{i\hbar\omega}
\int \frac{d\tau}{2 \pi} e^{i \hbar \omega \tau}\Theta(\tau)
\langle j^{\alpha}(\tau) j^{\beta}(0)\rangle$$ where the $\alpha's$ component of the total current is given by $j^{\alpha}(\tau)\,=\,i\frac{et}{\hbar}\sum_{i,\delta}\delta^{\alpha}
c^+_{i+\delta \sigma}(\tau)c^{\phantom {+}}_{i\sigma}(\tau)$. e denotes the charge of the Fermions and $\delta^{\alpha}$ the $\alpha's$ component of the lattice vectors linking nearest neighbor sites. Evaluating the isotropic optical conductivity $\sigma(\omega)=\frac{1}{3}\sum_{\alpha}\sigma^{\alpha \alpha}(\omega)$ within the lowest order approximation (neglecting vertex corrections but fully taking into account the selfconsistently determined Fermion one particle Green’functions) we obtain the optical conductivity as a function of frequency which for different temperatures is plotted in Fig.(4). In the inset of Fig.(4) we plot the dc conductivity for different temperatures and notice that upon decreasing the temperature one passes at $T^{\star}$ from a metallic like behaviour to one which has activated semiconducting like behaviour as a result of the opening of the pseudogap below $T^{\star}$. These features are also present in the optical conductivity which for temperatures below $T^{\star}$ shows a shift of the oscillator strength from the frequency regime $\omega \leq \omega^{\star} \simeq 2T^{\star}$ to $\omega \geq \omega^{\star}$ for $T\leq T^{\star}$, while for $T\geq T^{\star}$ a similar shift is observed in the opposite direction. The emptying out of the spectral weight of $\sigma(\omega)$ for $ \omega \leq \omega^{\star}$ which would show up as a dip in the optical conductivity can only be approached without being reached because of computational difficulties in reaching sufficiently low temperatures.
These manifestations of the pseudogap of the DOS of the Fermions and in particular those seen in the magnetic[@NMR] and transport[@transport] response functions are very reminiscent of what is observed in the normal state of underdoped $HT_cSC$. Such features have been previously attempted to be interpreted in the framework of the negative $U$ Hubbard model. While both models lead to pseudogaps in the one particle spectrum and give similar results as far as the magnetic response is concerned, the physics of those two models is nevertheless quite different. In the $U<0$ Hubbard model electron pair states exist as “fluctuating states” with very short life times for long wavelength excitations and the Fermiliquid properties of the one particle excitations are conserved. The increase in the value of $T_c$, obtained in the intermidiary coupling regime, is then due to a strengthening of the Cooperpair correlations rather than to a Bose condensation of electron pairs. In the BF model on the contrary electron pairs with total momentum close to zero are longlived, condense upon lowering the temperature and give rise to deviations from Landau Fermi liquid properties. Those effects on the thermodynamic and transport properties have been studied here and should be noticably different from those obtained on the basis of the $U<0$ Hubbard model.
B. Battlogg et al., Physica [**C 235-240**]{}, 130 (1994). C.H. Ruscher et al. Physica [**[C 204]{}**]{}, 30 (1992); A.V. Pushkov et al. Phys. Rev.B [**50**]{}, 4144 (1994); D.N. Basov et al,. Phys. Rev. Lett. [**74**]{}, 598 (1995) and Jian Ma et al. Solid State Comm. [**94**]{}, 27 (1995). J. Ranninger and S. Robaszkiewicz, Physica [**B 135**]{}, 468 (1985). R. Friedberg and T.D. Lee, Phys. Rev B [**40**]{},6740(1989); J. Ranninger and J.M. Robin, Physica [**C 353**]{}, 279(1995). J. Ranninger, J.M. Robin and M. Eschrig, Phys. Rev. Lett. [**74**]{} 4027 (1995). G. Baym, Phys. Rev.[**127**]{}, 1391 (1962). J. Ranninger and J.M. Robin, Solid State Commun. (1996) to be published This trick permits us to use very few Matsubara frequencies, as we take into account exactly the large frequencies asymptotic behaviour of the Green function $G(i \omega_{n}) \rightarrow 1/i\omega_{n}$. We have checked that the specific heat given from $C_{V} = \partial E/ \partial T$ and $C_{V} = T \partial S/ \partial T$ are in very good agreement. J. M. Kosterlitz and D. J. Thouless, J. Phys. C [**6**]{}, 1181 (1973) D. S. Fisher and P. C. Hoenberg, Phys. Rev. B [**37**]{}, 4936 (1988) W.W. Warren et al., Phys. Rev. Lett. [**62**]{}, 1193 (1989); H. Alloul et al., Phys. Rev. Lett.[**70**]{}, 1171 (1993); R. E. Walstedt et al., Phys. Rev. Lett. [**72**]{}, 3610 (1994). R. T. Collins et al., Phys. Rev. B [**43**]{}, 8701 (1991); L. D. Rotter et al., Phys. Rev. Lett.[**67**]{}, 2741 (1991); C. C. Homes et al., Phys. Rev. Lett.,[**71**]{}, 1645 (1993). M. Randeria, Phys. Rev. Lett. [**69**]{}, 2001 (1992); N. Trivedi and M. Randeria, Phys. Rev. .Lett. [**75**]{}, 312 (1995). R. Micnas et al.,Phys. Rev. B [**52**]{} to be published.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss the behavior of fidelity for a classically chaotic quantum system in the metallic regime. We show the existence of a critical value of the perturbation below which the exponential decay of fidelity is determined by the width of the Breit-Wigner distribution and above which the quantum decay follows the classical one which is ruled by the Lyapunov exponent. The independence of the decay [*rate*]{} from the perturbation strength derives from the similarity of the quantum and classical relaxation process inside the Heisenberg time scale.'
address:
- '$^{(a)}$International Center for the Study of Dynamical Systems, Università degli Studi dell’Insubria and'
- 'Istituto Nazionale per la Fisica della Materia, Unità di Como, Via Valleggio 11, 22100 Como, Italy'
- '$^{(b)}$Istituto Nazionale di Fisica Nucleare, Sezione di Milano, Via Celoria 16, 20133 Milano, Italy'
author:
- 'Giuliano Benenti$^{(a)}$ and Giulio Casati$^{(a,b)}$'
title: Sensitivity of Quantum Motion for Classically Chaotic Systems
---
[2]{}
Quantum chaos namely the attempt to understand classical dynamical chaos in terms of quantum mechanics has lead to a much better understanding of some properties of quantum motion which go beyond simple integrable models and perturbative treatments. A simple property of quantum conservative Hamiltonian systems with a finite number of particles, namely discrete spectrum, has been at the origin of some difficulties. Indeed in the classical ergodic theory discrete spectrum together with linear local instability of motion is a typical feature of integrable systems while chaotic systems are characterized by continuous spectrum and exponential local instability. This fact has questioned the possibility of dynamical chaos in quantum mechanics. On the other hand the correspondence principle requires transition to classical mechanics of all properties including dynamical chaos. As discussed in several occasions [@Chirikov1] this apparent contradiction is resolved by taking into account that a sharp distinction between discrete and continuous spectrum becomes meaningful only in the limit $t\to\infty$. For finite times, there exist different time scales below which the quantum motion can display chaotic properties like the corresponding classical one. These time scales tend to infinite as the effective Planck constant $\hbar_{\rm eff}\to 0$. Two time scales are of particular importance: the random or the Ehrenfest time scale $t_r$ and the relaxation or the Heisenberg time scale $t_R$. For $t < t_r$ the quantum motion is exponentially unstable like the classical one while the quantum relaxation process takes place during the time $t < t_R$. Since typically $t_r << t_R$, the quantum relaxation process takes place in the absence of exponential instability. A clear illustration of this peculiar feature of quantum motion is shown in [@cas86]. It should be remarked that this lack of exponential instability does not prevent exponential decay of dynamical quantities like correlation functions or survival probability [@maspero].
Recently the problem of the stability of quantum motion has attracted a great interest, also in relation to the field of quantum computation. A quantity of central importance which has been on the focus of many studies [@peres; @jalabert; @cucchietti; @beenakker; @tomsovic; @prose; @flambaum; @zurek; @cohen] is the so-called fidelity $f(t)$, which measures the accuracy to which a quantum state can be recovered by inverting, at time $t$, the dynamics with a perturbed Hamiltonian: $$f(t)=|\langle\psi|e^{i\hat{H}t}e^{-i\hat{H}_0t}|\psi\rangle|^2.
\label{fidelity}$$ Here $\psi$ is the initial state which evolves for a time $t$ with the Hamiltonian $\hat{H}_0$ while $\hat{H}=\hat{H}_0+\hat{V}$ is the perturbed Hamiltonian. The analysis of this quantity has shown that, under some restrictions, the decay of $f(t)$ is exponential with a rate given by the classical Lyapunov exponent [@jalabert]. This result appears to be consistent with recent experiments on the polarization echoes in nuclear magnetic resonance[@pasta] and with numerical computations [@cucchietti]. More recent papers have contributed to clarify different complementary aspects of the problem [@beenakker; @prose; @flambaum; @zurek], including the relation with the local density of states [@cohen] and the use of semiclassical approach [@tomsovic]. The analysis of this quantity has some delicate aspects concerning some attempts to characterize quantum chaos via the classical Lyapunov exponent and the role of the above mentioned time scales. It is therefore highly desirable to have very accurate numerical results and to this end it is necessary to consider simple systems which display the generic features of classical and quantum chaotic systems and which can be easily treated numerically.
In this paper we consider the behavior of fidelity for a classically chaotic system, in the delocalized regime of quantum ergodicity, in which the wave functions have a complex pattern which can be described within the framework of random matrix theory. We show that the type of decay and its rate depend on the strength of the perturbation. In particular, above a critical border, the quantum decay mimics the classical one and therefore, up to the relaxation time scale, it follows the exponential classical decay, which in our case is ruled by the Lyapunov exponent. The independence of the decay [*rate*]{} on the perturbation, which takes place in this regime, simply reflects the properties of the underlying classical motion.
We consider the classical sawtooth map: $$\overline{n}={n}+k_0(\theta-\pi),
\quad
\overline{\theta}=\theta+T\overline{n},
\label{clmap}$$ where $(n,\theta)$ are conjugated action-angle variables ($0\le \theta <2\pi$), and the bars denote the variables after one map iteration. Introducing the rescaled momentum variable $p=Tn$, one can see that the classical dynamics depends only on the single parameter $K_0=k_0T$. The map (\[clmap\]) can be studied on the cylinder \[$p\in (-\infty,+\infty)$\], which can also be closed to form a torus of length $2\pi L$, where $L$ is an integer. For $K_0>0$ the motion is completely chaotic and diffusive, with Lyapunov exponent given by $\lambda= \ln [(2+K_0+((2+K_0)^2 -4)^{1/2})/2]$. For $K_0>1$, the diffusion coefficient is well approximated by the random phase approximation, $D\approx (\pi^2/3) K_0^2$.
The quantum evolution on one map iteration is described by a unitary operator $\hat{U}_0$ acting on the wave function $\psi$: $$\overline{\psi}=\hat{U}_0\psi =
e^{-iT\hat{n}^2/2}
e^{ik_0(\hat{\theta}-\pi)^2/2}\psi,
\label{qumap}$$ where $\hat{n}=-i\partial/\partial\theta$ (we set $\hbar=1$). We take $-N/2\le n < N/2$, $k_0 = (K_0/2 \pi L) N$, $T=2 \pi L/N$. The classical limit corresponds to $N\to \infty$. We note that in this simple quantum model one can observe important physical phenomena like dynamical localization and cantori localization [@fausto]. Our aim is to study the fidelity decay in the delocalized regime of quantum ergodicity. Moreover we will consider parameter values for which there is no initial transient diffusive behavior, which may considerably affect the decay of fidelity.
In order to compute the fidelity we choose to perturb our system by slightly varying the kicking strength, $K =K_0 + \epsilon$, with $\epsilon\ll K_0$. Correspondingly the perturbed quantum kicking parameter is $k=k_0 + \sigma$, with $\sigma=\epsilon N /(2 \pi L)$. Since we want to compare classical and quantum evolution, we compute the classical “fidelity” $f_c(t)$ in the following way: we consider in the phase space a uniform density of points inside a strip of area $A=2 \pi \nu$ ($0\leq\theta< 2\pi$, $-\nu/2\leq p<\nu/2$). We then define $f_c(t)$ as the overlap of the initial area $A$ with the area $A'$ obtained by evolving $A$ for $t$ iterations of the map (2) and then reversing the evolution for $t$ iterations with the perturbed strength $K= K_0+\epsilon$. In practice, we follow the evolution of $10^6$ trajectories uniformly and randomly distributed inside the area $A$ and define the fidelity $f_c(t)$ as the percentage of orbits which return back to the area $A$ at time $t$, after the above reversing procedure. The corresponding quantum initial condition is given by a uniform mixture of momentum states located inside the area $A$. We note that this choice, besides giving the correct classical limit when $N\to\infty$, introduces a convenient averaging procedure. Moreover, we have checked that the same fidelity decay rates are obtained if one starts from pure states, like momentum eigenstates or coherent states.
=8.cm
=8.cm
The behavior of the classical fidelity is shown in Fig.1, for $K_0=1$, $L=1$, and different values of the perturbation strength $\epsilon$. In this particular regime, characterized by (i) uniform local exponential instability and (ii) absence of diffusive regime, the fidelity decay is ruled by the Lyapunov exponent $\lambda$. The exponential decay starts after an initial transient time proportional to $\ln (\nu/\epsilon)$, which is required to amplify the perturbation up to the scale $\nu$ [@eps].
=8.cm
=8.cm
The decay of the quantum fidelity is Gaussian below a perturbative border [@beenakker; @tomsovic]. This border is given by the value of the perturbation at which the typical transition matrix element $U$ between quasienergy eigenstates becomes larger than the average levels spacing $1/\rho$. For ergodic eigenfunctions, $U\sim \sigma/\sqrt{N}$, while the density of quasienergy states is given by $\rho=N/2 \pi$. Therefore the perturbative border is given by $\sigma_p \approx 1/\sqrt{N}$. Above this border one typically expects an exponential decay of fidelity, with a rate $\Gamma =2 \pi\rho U^2 \approx \sigma^2$ given by the width of the Breit-Wigner local density of states [@beenakker]. This theoretical prediction is confirmed in Fig.2, which shows the decay of quantum fidelity at $\epsilon=5\times 10^{-5}$ and different $N$ values, with $\sigma>\sigma_p$. The nice scaling behavior of Fig.2 confirms the predicted exponential decay $f(t) \approx \exp(-C \sigma^2 t)$, with the numerically determined constant $C\approx 2.2$.
On the other hand, as stated in the introduction, one expects that in the semiclassical regime the quantum motion mimics the classical one up to the relaxation time scale which is determined by the density of quasienergy eigenstates which significantly contibute to the wave function dynamics. To this end it is necessary that the perturbation $\sigma$ is strong enough to allow the quantum motion to follow, on the average, the initial classical decay. In our case this may happen if $\sigma$ is large enough to induce transitions at least between nearest neighbors momentum states, namely $\sigma > \sigma_c \approx 1$. If $\sigma < \sigma_c$, the quantum excitation is unable to follow the classical spreading of the initial state. One may also argue in a different way: since with our choice of parameters we are in the metallic regime, all $N$ quasienergy states are involved in the evolution of the unperturbed system. Then the effect of the perturbation on the quantum motion can imitate the corresponding classical one only if there are no quantum localization effects on the quasienergy states. This happens when the width of the local density of states becomes comparable to the band width, that is $\rho \Gamma \approx N$, which again gives the threshold value $\sigma_c\approx 1$. We remark that, as discussed in [@ccgi], in the theory of Wigner band random matrices the Breit-Wigner regime corresponds to a sort of partial perturbative localization. The above theoretical estimate is well confirmed by our numerical data presented in Figs.3,4. Fig. 3 shows that for $\sigma>1$ the quantum fidelity follows closely the classical behavior, namely it decays exponentially with the classical rate given by the Lyapunov exponent. Fig. 4 shows the decay rate $\gamma$ as a function of the perturbation strength $\sigma$. It is clearly seen that for $\sigma<1$ the decay rate is proportional to $\sigma^2$, that is to the width of the Breit-Wigner. Therefore $\sigma_c \approx 1$ is a critical value which separates two distinct regimes: a pure quantum perturbation dependent regime, and a semiclassical regime. We note that the perturbation $\sigma$ depends on both $N$ and $\epsilon$. For $\sigma >1$, the decay rate does not change by increasing $N$ at fixed $\epsilon$, since by doing this we merely increase the Heisenberg time. On the other hand, if we increase $\epsilon$ at fixed $N$ (provided that the perturbation remains classically small, i.e. $\epsilon << K_0$) the decay rate also does not change, since the exponential amplification of the perturbation is controlled by the parameter $K \approx K_0$. In both cases the [*decay rate*]{} of fidelity is perturbation independent. This is a property of the classical motion which, in the semiclassical regime, is shared by quantum mechanics. However, we would like to stress that the decay of fidelity remains [*perturbation dependent*]{}, since the exponential decay starts after a time $\propto |\ln\epsilon|$ (see Figs.1,3).
=8.cm
For the parameters values of Figs.1-4, the decay of fidelity is exponentially fast and the saturation value $f_\infty=\nu/(2\pi L)$ is reached on times much shorter than the Heisenberg time. In order to observe the effect of the Heisenberg time scale it is necessary to have a much slower decay of fidelity. In Fig.5 we take $K_0=1$ and $L=50$, so that we allow for a Gaussian diffusive process in momentum space. Because of this, during the diffusion time the fidelity decays in the classical case as $1/\sqrt{Dt}$ [@nota]. Fig.5 shows that for $\sigma>\sigma_c\approx 1$ the quantum decay follows the classical one for larger and larger times as $N$ increases, in agreement with the correspondence principle. The asymptotic value is $f_\infty=\nu l/(2\pi L)$, where, according to the scaling theory of localization, $l=\xi/N=g(x)$, with $x=k^2/N$ [@scaling]. Here $\xi$ is the actual localization length of the “sample” of size $N$, while $k^2$ gives the localization length for the infinite sample, up to a numerical constant of order $1$. The scaling function $g(x)$ is proportional to $x$ for $x\ll 1$ and saturates to $1$ for $x\gg 1$. The transition value $x=1$ corresponds to $N\approx 10^5$. Moreover, the saturation value is approached after a relaxation time $t_l\approx\xi$. We stress that in the case of Fig.\[fig5\] the decay of fidelity is controlled by the diffusion coefficient and not by the Lyapunov exponent. The observation of such regime represents a challenge for experiments like spin echoes. Further theoretical investigations are also desirable in order to understand more clearly the effect of classical diffusion and quantum localization on the behavior of fidelity.
In summary, we have shown that the decaying behavior of fidelity in a classically chaotic system strongly depends on system parameters as well as on the perturbation strength. Nevertheless there is a regime in which the decay [*rate*]{} (exponential or power law) is perturbation independent: in this regime the quantum motion simply mimics the properties of the underlying classical dynamics. We emphasize that the quantum to classical correspondence of the average behavior is valid until the Heisenberg time scale, which is much longer than the Ehrenfest time scale associated with the exponential instability of quantum motion.
This work was supported in part by the EC RTN network contract HPRN-CT-2000-0156, the NSF under grant No. PHY99-07949, the PA INFM “Quantum transport and classical chaos”, and the PRIN “Caos e localizzazione in meccanica classica e quantistica". We gratefully acknowledge the Institute for Theoretical Physics, Santa Barbara, California, for the hospitality during the initial stage of this work.
[99]{} G. Casati and B.V. Chirikov, [*Quantum Chaos*]{}, Cambridge University Press, Cambridge (1995); Physica D [**86**]{}, 220 (1995). G. Casati, B.V. Chirikov, I. Guarneri, and D.L. Shepelyansky, Phys. Rev. Lett. [**56**]{}, 2437 (1986). G. Casati, G. Maspero, and D.L. Shepelyansky, Phys. Rev. E [**56**]{}, R6233 (1997). A. Peres, Phys. Rev. A [**30**]{}, 1610 (1984). R.A. Jalabert, and H.M. Pastawski, Phys. Rev. Lett. [**86**]{}, 2490 (2001). F.M. Cucchietti, H.M. Pastawski, and D.A. Wisniacki, cond-mat/0102135; D.A. Wisniacki, E.G. Vergini, H.M. Pastavski, and F.M. Cucchietti, nlin.CD/0111051.. Ph. Jacquod, P.G. Silvestrov, and C.W.J. Beenakker, Phys. Rev. E [**64**]{}, 055203(R) (2001). N.R. Cerruti and S. Tomsovic, nlin.CD/0108016. T. Prozen, quant-ph/0106149. V.V. Flambaum and F.M. Izrailev, Phys. Rev. E [**64**]{}, 036220 (2001). Z.P. Karkuszewski, C. Jarzynski, and W.H. Zurek, quant-ph/0111002. D.A. Wisniacki and D. Cohen, quant-ph/0111125. H.M. Pastawski, P.R. Levstein, G. Usaj, J. Raya, and J. Hirschinger, Physica A [**283**]{}, 166 (2000). F. Borgonovi, G.Casati, and B. Li, Phys. Rev. Lett. [**77**]{}, 4744 (1996); F. Borgonovi, Phys. Rev. Lett. [**80**]{}, 4653 (1998). For $\epsilon=0$, the numerically computed fidelity remains $1$ up to times $t\approx 25$, due to round-off errors $\epsilon\sim 10^{-14}$. G. Casati, B.V. Chirikov, I. Guarneri and F.M. Izrailev, Phys. Lett. A [**223**]{}, 430 (1996). Strictly speaking, $f_c(t)-f_{c,\infty}$ should decay exponentially after the diffusive time scale $t_D\approx L^2/D$. However this can hardly be seen numerically, as it appears also from Fig.\[fig5\]. Actually, for $t>t_D$ the fidelity is already close to its asymptotic value $f_{c,\infty}=\nu/(2\pi L)=10^{-2}$. G. Casati, I. Guarneri, F.M. Izrailev, and R. Scharf, Phys. Rev. Lett. [**64**]{}, 5 (1990).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The dependence of the magnetic ordering temperature $T_{\text{o}}$ of Sm metal was determined through four-point electrical resistivity measurements to pressures as high as 150 GPa. A strong increase in $T_{\text{o}}$ with pressure is observed above 85 GPa. In this pressure range Sm ions alloyed in dilute concentration with superconducting Y exhibit giant Kondo pair breaking. Taken together, these results suggest that for pressures above $%
\sim $ 85 GPa Sm is in a highly correlated electron state, like a Kondo lattice, with an unusually high value of $T_{\text{o}}.$ A detailed comparison is made with similar results obtained earlier on Nd, Tb and Dy and their dilute magnetic alloys with superconducting Y.
author:
- 'Y. Deng and J. S. Schilling'
title: Enhanced magnetic ordering in Sm metal under extreme pressure
---
I
Introduction
============
Except for Ce, the local-moment magnetic state in elemental lanthanide metals is highly stable. Under the application of sufficient pressure, however, the magnetic state would be expected to destabilize. In recent studies on the trivalent lanthanide metals Nd [@song], Gd [lim1]{}, Tb [@lim2], and Dy [@lim1] the magnetic ordering temperature $T_{\text{o}}$, with the exception of Gd, was found to rise steeply to anomalously high values upon the application of extreme pressure. In the same pressure range, alloying Nd, Tb, and Dy in dilute concentration into superconducting Y resulted in a very large suppression of the superconducting transition temperature $T_{\text{c}}$, in the case of Y(Nd) the record value 39 K/(at.% Nd) [@song]. Such high values are a signature of giant Kondo pair breaking, a sign that these lanthanides may be approaching a magnetic instability. The anomalous rise in $T_{\text{o}}$ and the giant pair breaking thus appear to be related.
It is interesting to note that in the Kondo lattice model described by the Doniach phase diagram [@doniach] $T_{\text{o}}$ is expected to first increase with the magnitude of the negative covalent mixing exchange coupling $J_{-}$ [@schrieffer] before passing through a maximum and falling rapidly to the quantum critical point at 0 K (see Fig 9 in the Discussion section). This occurs when the Ruderman-Kittel-Kasuya-Yosida (RKKY) interaction [@kittel] is suppressed by Kondo spin screening. Since the magnitude of $J_{-}$ normally increases under pressure [schilling1]{}, in the Doniach picture $T_{\text{o}}$ versus pressure should pass through a maximum and fall towards 0 K. This behavior was indeed recently observed for elemental Nd metal by Song *et al.* [@song]. In contrast, due to the extreme stability of Gd’s magnetic state with its half-filled 4$f^{7}$ configuration, even pressures to 1 or 2 Mbar are not sufficient to bring Gd near a magnetic instability. Indeed, neither giant pair breaking in Y(Gd) nor an anomalous rise in $T_{\text{o}}$ for Gd are observed at extreme pressure [@lim1; @fabbris].
In view of the intriguing magnetic behavior in trivalent Nd metal and Y(Nd) alloys at extreme pressure, an in-depth study of an additional light trivalent lanthanide Sm, both as elemental metal and in the dilute magnetic alloy Y(Sm), was undertaken. Sm metal crystallizes in the Sm-type ($\alpha $-Sm) structure at ambient pressure, transitioning to dhcp at 4 GPa, to fcc at 14 GPa, to *hR*24 (dfcc) at 19 GPa, to *hP*3 at 37 GPa, and finally to *tI*2 at 91 GPa [@zhao; @vohra]. These structural transitions thus follow the regular trivalent lanthanide structure sequence under pressure: hcp$\rightarrow $Sm-type$\rightarrow $dhcp$\rightarrow $fcc$%
\rightarrow $dfcc, a sequence generated by the increasing $d$ character in the conduction band upon compression [@pettifor].
Trivalent Sm assumes the configuration \[Xe\]4$f^{5}$ yielding for the free Sm$^{3+}$ ion in the ground state $^{6}$H$_{5/2}$, with Landé $g$-factor $g_{J}$ = 2/7 and total angular momentum $J_{t}$ = 5/2. The effective magnetic moment of free Sm$^{3+}$calculated from Hund’s rules is $p_{eff}=$ 0.85 $\mu _{\text{B. }}$ However, magnetic susceptibility measurements on paramagnetic Sm salts give $p_{eff}=$ 1.74 $\mu _{\text{B }}$[@blundell]. The difference between the theoretical and experimental values is believed due to contributions from low-lying excited states with different $J_{t}$ values.
In Sm metal the situation is more complicated since the crystalline electric field and conduction-electron polarization significantly influence the magnetic state of Sm$^{3+}$ [@adachi]. As a result of this complexity, Sm metal exhibits a number of interesting physical phenomena. Both the temperature-dependent heat capacity [@jennings] and electrical resistivity [@alstad] have anomalies near 13 K and 106 K. The fact that the temperature-dependent magnetic susceptibility of Sm has peaks near these temperatures strongly suggests antiferromagnetic ordering [@mcewen]. This was confirmed by Koehler and Moon from neutron diffraction experiments on single crystalline $^{154}$Sm [@koehler]. They viewed the Sm-type structure with space group *R*$\overline{3}$*m* as a combination of hexagonal and cubic sites where Sm$^{3+}$ ions at these sites magnetically order at 106 K and 14 K, respectively.
In temperature-dependent resistivity measurements $R(T)$ a knee is observed at the magnetic ordering temperature $T_{\text{o}}$ due to the loss of spin-disorder scattering upon cooling. Dong *et al.* [@dong] measured resistivity on Sm to 43 GPa and found that the two ordering temperatures move toward each other with increasing pressure, finally merging together near 66 K at 8 GPa as Sm enters the dhcp phase. At higher pressures $T_{\text{o}}$ increases rapidly to 135 K at 43 GPa. Johnson *et al.* [@johnson] measured $R(T)$ on Sm to 47 GPa. They find that the two ordering temperatures merge near 56 K at 10 GPa and increase slowly up to the (*hR*24$\rightarrow $*hP*3) phase transition at 34 GPa where a second ordering temperature reportedly appears.
In this work four-point dc resistivity measurements are carried out on pure Sm metal to $\sim $ 150 GPa using a diamond anvil cell. The two magnetic ordering temperatures merge at 13 GPa after which $T_{\text{o}}(P)$ increases gradually to a maximum at 53 GPa, but then decreases and passes through a minimum at 85 GPa followed by a sharp increase to $\sim $ 140 K at 150 GPa. Giant superconducting pair breaking is also observed in dilute Y(Sm) alloys. Taken together, both effects suggest that extreme pressure drives Sm to an unconventional magnetic state, a state likely related to that observed earlier in Nd, Tb, and Dy.
Experimental Techniques
=======================
Polycrystalline Sm samples for the high-pressure resistivity measurements were cut from a Sm ingot. The dilute magnetic alloys of Y(Sm) were made by argon arc-melting small amounts of Sm with Y (both Sm and Y 99.9 % pure, Ames Laboratory [@ames]). To enhance homogeneity the alloys were sealed in glass ampules under vacuum and annealed at 600$^{\text{o}}$C for two weeks. The concentrations of Sm for the four alloys as determined from x-ray fluorescence analysis are: 0.15(2) at.%, 0.40(3) at.%, 0.83(4) at.%, and 1.16(6) at.%. Before arc-melting the nominal concentrations were 0.5 at.%, 1.0 at.%, 1.2 at.%, and 2 at.%, respectively. It follows that 30% to 70% of the Sm evaporated during arc-melting due to its relatively low boiling point.
A diamond anvil cell (DAC) made of conventional and binary CuBe [@schilling] was used to reach pressures to $\sim $150 GPa between two opposed diamond anvils (1/6-carat, type Ia) with 0.18 mm diameter culets beveled at 7 degrees to 0.35 mm diameter. The force applied to the anvils was generated by a stainless-steel diaphragm filled with He gas [@daniels]. The Re gasket (250 $\mu $m thick) was pre-indented to 30 $\mu $m and a 90 $\mu $m diameter hole drilled through the center of the pre-indentation area. A cBN-epoxy insulation layer was compressed onto the surface of the gasket. Four Pt strips (4 $\mu $m thick) were then placed on the insulation layer, acting as the electrical leads for the four-point resistivity measurement. The Sm or Y(Sm) sample with dimensions 40 $\times $ 40 $\times $ 4 $\mu $m$^{3}$ was then placed on the Pt strips. Further details of the high-pressure resistivity techniques can be found elsewhere [@lim1; @shimizu].
The DAC was inserted into an Oxford flow cryostat capable of varying temperature from ambient to 1.3 K. Pressure was determined at room temperature using the diamond vibron [@vibron]. Earlier resistivity experiments by Song *et al.* [@song] in an identical DAC using both vibron and ruby manometers revealed an approximately linear pressure increase of $\sim $ 30% on cooling from 295 to 4 K. In the present experiments this calibration allows an estimate of the pressure at the magnetic or superconducting transition temperatures from the vibron pressure at ambient temperature.
Results of Experiment
=====================
Four-point resistance measurements $R(T)$ were carried out on Sm in two runs over the temperature and pressure ranges 1.3 - 295 K and 2 - 127 GPa (measured at room temperature), respectively. The data from run 1 are shown in Fig 1. For all pressures the resistance is seen to decrease upon cooling. A kink or knee appears in $R(T)$ in the lower temperature range that results from the progressive loss of spin-disorder scattering $R_{\text{sd}}(T)$ as Sm orders magnetically. At 2 and 4 GPa two kinks are visible in the $R(T)$ curves; at higher pressures only one kink or knee appears. With increasing pressure the knee is seen to shift in temperature and broaden; the broadening is due to the increasing pressure gradient across the sample in the non-hydrostatic pressure environment. The value of $T_{\text{o}}$ is determined from the intersection temperature of two straight lines tracking $R(T)$ above and below the knee region, as illustrated in Fig 1 for 27 GPa pressure. In most experiments the Sm sample was cooled to $\sim 4$ K; however, considering both runs, at pressures 2, 9, 25, 51, 60, 86, 93, 97, and 127 GPa the sample was cooled to 1.3 K. In no experiment on Sm was superconductivity, or even an onset to superconductivity, observed.
In Fig 2 the values of $T_{\text{o}}$ for Sm from runs 1 and 2 are plotted versus pressure and compared to previous results from Dong *et al.* [@dong] to 43 GPa and Johnson *et al*. [@johnson] to 47 GPa. In all experiments the two branches of $T_{\text{o}}$ are seen to merge near 13 GPa followed by an increase in $T_{\text{o}}$. In the present experiments $T_{\text{o}}(P)$ passes through a maximum near 53 GPa, gradually decreasing to $\sim 60$ K near 85 GPa, before rising sharply to $\sim 140$ K at 150 GPa. The report by Johnson *et al.* [johnson]{} that a second transition appears in the pressure range 35 - 50 GPa could not be confirmed.
Due to the broadening of the resistivity knee under nonhydrostatic pressure, the determination of the value of $T_{\text{o}}$ for Sm becomes progressively more difficult in the upper pressure range. The same was true for the other trivalent lanthanides Nd, Gd, Tb, and Dy studied previously [song,lim1,lim2]{}. In particular, as here for Sm, a rapid upward shift of the knee in $R(T)$ was also observed for Dy above 70 GPa pressure [@lim1]. That the knee for Dy does indeed result from magnetic ordering over the entire pressure range was recently confirmed by synchrotron Mössbauer spectroscopy (SMS) to 141 GPa [@bi].
Independent information on the origin of the resistivity knee in Sm can be gained by comparing the pressure dependence of the spin-disorder resistance $%
R_{\text{sd}}(P)$ for $T>T_{\text{o}}$ to that of $T_{\text{o}}(P)$ obtained from the resistivity knee. As discussed in Ref [@lim1], both $T_{\text{o}%
}$ [@taylor] and $R_{\text{sd}}$ [@daal] are proportional to $J^{2}N(E_{\text{F}})$, where $J$ is the exchange interaction between local moment and conduction electrons and $N(E_{\text{F}})$ is the density of states at the Fermi energy. A similarity between the pressure dependences $%
T_{\text{o}}(P)$ and $R_{\text{sd}}(P)$ is anticipated for the trivalent lanthanide metals since their $spd$ conduction electron properties are closely related. This similarity was indeed observed for Nd [@song], Gd [@lim1], Tb [@lim2], and Dy [@lim1]; it would be interesting to examine whether this also holds for Sm, together with Nd the second light lanthanide studied. From Fig 1 it is readily seen that where the resistivity knee shifts under pressure to higher temperatures the size of the resistivity drop-off below the knee also increases. A semi-quantitative estimate of $R_{\text{sd}}$ is now attempted.
The total measured resistance is the sum of three terms, $R(T)=R_{\text{d}%
}+R_{\text{ph}}(T)+R_{\text{sd}}(T),$ where $R_{\text{d}}=R(0$ K$)$ is the temperature-independent defect contribution. In the paramagnetic state in the temperature region above the resistance knee, $R_{\text{sd}}(T)$ is constant, taking on its maximum value $R_{\text{sd}}^{\text{max}}$, so that the only temperature dependence comes from the phonon resistance $R_{\text{ph%
}}(T)$. To estimate $R_{\text{sd}}^{\text{max}},$ Colvin *et al.* [@colvin] assumed that $R_{\text{ph}}(T)$ depends linearly on temperature, and extended a straight line fit to $R(T)$ for $T>T_{\text{o}}$ to 0 K with intercept $R_{\text{int}}$ and then subtracted off $R_{\text{d}}$ from this intercept. An example for this estimate is given in Fig 1 at 27 GPa where $R_{\text{sd}}^{\text{max}}=[R_{\text{int}}-R_{\text{d}}]=(115-18)$ m$\Omega =97$ m$\Omega $.
In Fig 3 $R_{\text{sd}}^{\text{max}}$ is plotted as a function of pressure. Comparing Figs 2 and 3, a parallel behavior of the pressure dependences $T_{%
\text{o}}(P)$ and $R_{\text{sd}}^{\text{max}}(P)$ is indeed observed, thus supporting the identification of the resistance knee with the onset of magnetic ordering in Sm. Also included in Fig 3 is the quantity \[$R$(290 K) – $R$(4 K)\] that is seen to also qualitatively track $T_{\text{o}}$ versus pressure. This suggests that the resistance from electron-phonon scattering at room temperature does not change dramatically within the pressure range of these experiments.
To examine whether the rapid rise in $T_{\text{o}}$ for pressures above 85 GPa might be related to an approaching instability in Sm’s magnetic state, Sm is alloyed in dilute concentration with Y, a high-pressure superconductor having, compared to the trivalent lanthanides, closely similar conduction electron properties and structural sequence under pressure [@wittig2]. Under these circumstances the ability of the Sm ion to suppress Y’s superconductivity, the degree of pair breaking $\Delta T_{\text{c}}\equiv T_{%
\text{c}}$\[Y\] - $T_{\text{c}}$\[Y(Sm)\], can reveal valuable information about the magnetic state of the Sm ion itself. This general observation was emphasized for lanthanide ions by Maple [@maple].
In the present experiment Y(Sm) alloys with differing dilute Sm concentrations were studied at pressures to 180 GPa. Fig 4 shows the superconducting transitions in four-point resistance measurements on Y(0.15 at.% Sm) at selected pressures. As illustrated in this figure for the $R(T)$ data at 52 GPa, $T_{\text{c}}$ is defined as the temperature at which the resistance transition reaches the halfway mark, whereas the intersection point of two straight red lines defines $T_{\text{c}}^{\text{onset \ }}$, and $%
T_{\text{c}}^{\text{zero}}$ gives the temperature where the resistance disappears. The fact that a typical total transition width is less than 2 K gives evidence that the distribution of Sm ions in the alloys is homogeneous. As seen from the data in Fig 4, $T_{\text{c}}$ increases monotonically with pressure to 140 GPa, but then decreases to 180 GPa.
The dependence of $T_{\text{c}}$ on pressure for Y(Sm) alloys with Sm concentrations 0.15, 0.40, 0.83, and 1.16 at.% is shown in Fig 5. Below $%
\sim $ 40 GPa the $T_{\text{c}}(P)$ dependence for all four alloys tracks that for pure Y. However, above $\sim $ 40 GPa a strong suppression sets in. This suppression $\Delta T_{\text{c}}$ is so strong that for Y(1.16 at.% Sm) at pressures above 50 GPa $T_{\text{c}}$ lies below the temperature range of this experiment (1.3 K). For the more dilute Y(0.15 at.% Sm) and Y(0.40 at.% Sm) alloys, $T_{\text{c}}$ remains well above 1.3 K at all pressures.
To allow a more meaningful comparison of the degree of superconducting pair breaking $\Delta T_{\text{c}}$ for the different alloys, in Fig 6 $\Delta T_{%
\text{c}}$ is divided by the Sm concentration $c$ and then plotted versus pressure for all alloys measured. Where they can be compared, the individual $\Delta T_{\text{c}}/c$ curves agree reasonably well and increase monotonically with pressure, reaching the extremely high value of $\sim $ 40 K/at.% Sm at 180 GPa, a value slightly higher than that found earlier for Y(0.4 at.% Nd) [@song]. Both the giant pair breaking in Y(Sm) and the remarkable increase of $T_{\text{o}}$ in Sm give evidence for unconventional physics in Sm above 85 GPa.
Discussion
==========
The present results on Sm and Y(Sm) alloys will now be compared to those from earlier studies on the lanthanides Nd [@song], Gd [@lim1], Tb [@lim2], and Dy [@lim1]. Going from right to left across the lanthanide series (Lu to La) or by applying pressure, one finds with few exceptions [@samudrala] the canonical rare-earth crystal structure sequence hcp$\rightarrow $Sm-type$\rightarrow $dhcp$\rightarrow $fcc$\rightarrow $*hR*24 believed to mainly arise from an increase in the number $n_{\text{d}}$ of $d$-electrons in the conduction band [pettifor]{}.
In the elemental lanthanide metals magnetic ordering arises from the indirect RKKY exchange interaction between the magnetic ions. For a conventional lanthanide metal with a stable magnetic moment, the magnetic ordering temperature $T_{\text{o}%
}$ is expected to scale with the de Gennes factor $(g-1)^{2}J_{t}(J_{t}+1)$, modulated by the prefactor $J^{2}N(E_{\text{F}})$, where $J$ is the exchange interaction between the 4$f$ ion and the conduction electrons, $N(E_{\text{F}%
})$ the density of states at the Fermi energy, $g$ the Landé-$g$ factor, and $J_{t}$ the total angular momentum quantum number [@taylor].
In Fig 7(a) the dependence of the magnetic ordering temperature $T_{\text{o}%
} $ on pressure is shown for the four lanthanide metals Nd, Sm, Tb, and Dy. Except for Nd, $T_{\text{o}}(P)$ is seen to initially decrease rapidly with pressure, but then pass through a minimum and rise. $T_{\text{o}}(P)$ for Gd [@lim1] also shows this same initial behavior. Since the de Gennes factor, in the absence of a magnetic instability or valence transition, is constant under pressure, the initial $T_{\text{o}}(P)$ dependence for the above lanthanides likely originates in the pressure dependence of the prefactor $J^{2}N(E_{\text{F}})$. Electronic structure calculations for Dy support this conclusion [@jackson1; @fleming].
The strong initial decrease in $T_{\text{o}}$ with pressure in Sm (upper transition), Gd, Tb, and Dy occurs within the hcp and Sm-type phases. The minimum in $T_{\text{o}}(P)$ at approximately 20 GPa for Dy appears at somewhat lower pressures for Tb, Gd, and Sm, disappearing entirely for Nd. As discussed in some detail in Ref [@song], this is consistent with an increase in the number of *d* electrons in the conduction band going from Dy to Nd; the electronic structure and the crystal structures taken on by Nd resemble those of Dy but at a pressure approximately 30 - 40 GPa higher [@song]. The systematic behavior for all five lanthanides Dy, Tb, Gd, Sm, and Nd in the region of pressure where the hcp, Sm-type, dhcp, and *hR*24 structures occur, gives evidence that changes in the magnetic ordering temperature in this region are mainly determined by corresponding changes in the properties of the conduction electrons that mediate the RKKY interactions between the magnetic lanthanide ions.
It would seem helpful to propose that the $T_{\text{o}}(P)$ curves for each element can be separated into two principal pressure regions: a conventional region at lower pressure governed by the electronic properties of the conduction electrons and normal positive exchange interactions $J_{+}$ between the lanthanide ion and the conduction electrons, and an unconventional region at higher pressures where exotic physics dominates leading to negative covalent-mixing exchange $J_{-}$ and associated anomalous magnetic properties. In the conventional region the observed variations in $T_{\text{o%
}}(P)$ would be principally caused by changes in the prefactor $N(E_{\text{F}})J^{2}_{+}$ with pressure. In the unconventionalpressure region highly correlated electron effects dominate leading to anomalous magnetic properties, including anomalous $T_{\text{o}}(P)$ dependences and giant superconducting pair breaking in dilute magnetic alloys.
Although the properties of the conduction electrons and the magnetic state of the lanthanide ion are intertwinned, the conventional and unconventional regions represent different physics, the former being amenable through standard electronic structure calculations, whereas the latter is only accessible through consideration of strong highly correlated electron effects. The stability of the ion’s magnetic state is determined to a large extent by the exchange interactions *within* a given lanthanide ion (Hund’s rules). Once the unconventional rapid rise in $T_{\text{o%
}}$ with pressure sets in, it overpowers the conventional conduction electron behavior and determines $%
T_{\text{o}}(P)$. Since in Dy and Nd the unconventional region begins at a lower pressure, the rapid rise in $T_{\text{o}}$ may prevent the conventional second minimum seen in Sm and Tb from appearing in $T_{\text{o}}(P)$ for Dy or Nd.
A rough estimate of the boundary pressure where the unconventional $T_{\text{o}}(P)$ behavior may begin for a given lanthanide is indicated by a vertical tick mark in Fig 7(a). In the unconventional region itself the $T_{%
\text{o}}(P)$ data curves have been given double thickness. There is a good deal of arbitrariness for where this boundary is placed, particularly for Sm and Tb where the second $T_{\text{o}}(P)$ minimum may well belong to the unconventional region, instead of the conventional region, as indicated by the beginning of anomalous superconducting pair breaking in Y(Sm) or Y(Tb) near the pressure for the second minimum in $T_{\text{o}}(P)$.
Focussing now on the anomalous rise in $T_{\text{o}}$ with pressure in the unconventional region in Fig 7(a), we note that this rise is steepest for Nd but becomes progressively less steep for Dy, Tb, and Sm. At least part of this reduction in steepness has to do with the fact that the compressibility of the lanthanides decreases significantly as pressure is increased. To bring out the physics more clearly, $%
T_{\text{o}}$ in Fig 7(b) is replotted versus the relative volume $V/V_{\text{o%
}}$. Different features in the respective curves are shifted to new relative positions, but now it is seen that the sharp upturns in $T_{\text{o}}(P)$ have nearly the same slope and are much steeper relative to the changes in the conventional region at lower pressures. This points to a common mechanism for the upturn in these four lanthanides.
In Fig 8 the normalized pair breaking curve $\Delta T_{c}/c$ for Y(Sm) from Fig 6 is compared to those for the dilute magnetic alloys Y(Nd) [@song], Y(Tb) [@lim2], and Y(Dy) [@lim1]. For Y(Sm) and Y(Nd) the pair breaking begins to increase rapidly at relatively low pressures compared to Y(Tb) and especially Y(Dy). At least part of the reason for this is that the Y host exerts lattice pressure on the light lanthanides Sm and Nd, but not on Tb and Dy. This can be seen by comparing the respective molar volumes in units of cm$^{3}$/mol: Y(19.88), Nd(20.58), Sm(19.98), Gd(19.90), Tb(19.30), Dy(19.01) [@singman]. Without exception, the region of pressure where $T_{\text{o}}(P)$ increases rapidly lies within the region of pressure where the superconducting pair breaking $\Delta T_{c}/c$ in the corresponding dilute magnetic alloy with Y is anomalously large. Note also that the maximum value of the slope of $\Delta T_{c}/c$ versus pressure in Fig 8 is noticeably reduced for Y(Dy). At least part of this effect is due to the sizable reduction in the compressibility of Y at higher pressures.
For the dilute magnetic alloy Y(Nd) the normalized pair breaking data in Fig 8 are seen to be reduced ($\Delta T_{c}/c$ turns upwards) for pressures above 160 GPa. Presumably the same effect would also be observed in Y(Sm), Y(Tb), and Y(Dy) if the experiments were extended to even higher pressures. This reduction in giant pair breaking seen in Y(Nd) at the highest pressures was observed previously in dilute magnetic alloys La(Ce) [@wittig1], La(Pr) [@wittig3], and Y(Pr) [@fabbris; @wittig4] and can be readily accounted for in terms of Kondo pair-breaking theory [@zuckermann] where the magnitude of the negative exchange interaction $J_{-}$ between the magnetic ions and the conduction electrons increases with pressure. The appearance of such Kondo physics in the dilute magnetic alloy suggests that the corresponding concentrated system will likely show Kondo lattice, heavy Fermion, and fluctuating valence behavior at higher pressures, eventually culminating in a full increase in valence whereby one 4$f$ electron completely leaves its orbital and joins the conduction band.
The well known Doniach model [@doniach] is often cited to account for the dependence of the magnetic ordering temperature $T_{\text{o}}$ in a Kondo lattice as a function of the magnitude of the negative exchange parameter $J_{-}$ (see Fig 9). Whereas the upturn in $\Delta T_{c}/c$ occurs above 160 GPa for Y(Nd), the downturn in $T_{\text{o}}(P)$ begins above 80 GPa (see Fig 7(a)) for Nd in its unconventional pressure region (double line width). The rapid rise in $T_{\text{o}}(P)$ for Nd followed by its rapid downturn resembles the dependence anticipated from the Doniach model [@song]. A similar $T_{%
\text{o}}(P)$ dependence would be expected for Sm, Tb and Dy if the experiments were extended to even higher pressures.
The values of the pair-breaking parameter $\Delta T_{c}/c$ for Nd and Sm impurities in Y are surprisingly large - in fact, to our knowledge, the largest ever reported. However, even more surprising is the sharp upturn in $%
T_{\text{o}}(P)$ where $T_{\text{o}}$ reaches values that appear to be much higher than would have been possible had unconventional physics, such as Kondo physics, not been operative. In the case of Dy, $T_{\text{o}}(P)$ extrapolates to values well above room temperature, higher than any known value for an elemental lanthanide metal at either ambient or high pressure [@lim1].
In summary, the magnetic properties of the light lanthanide Sm have been studied to extreme pressure and found to parallel those of another light lanthanide, Nd, as well as the heavy lanthanides Gd, Tb, and Dy. It appears that the magnetic phase diagram can be separated into two regions: a low-pressure region where conventional changes in the electronic structure determine $T_{\text{o}}(P)$, and a high-pressure region where highly correlated electron effects dominate, leading to such anomalous phenomena as unexpectedly high magnetic ordering temperatures and giant superconducting pair-breaking. The authors hope that this and previous work will lead to increased theoretical activity in this area.
**Acknowledgments.** The authors would like to thank A. K. Gangopadhyay for his assistance in preparing the Y(Sm) alloys and R. A. Couture for carrying out the x-ray fluorescence determination of the Sm content in these alloys. Thanks are also due Daniel Haskel for his critical reading of the manuscript. This work was supported by the National Science Foundation (NSF) through Grant No. DMR-1104742 and No. DMR-1505345 as well as by the Carnegie/DOE Alliance Center (CDAC) through NNSA/DOE Grant No. DE-FC52-08NA28554.
[99]{} J. Song, W. Bi, D. Haskel, and J. S. Schilling, Phys. Rev. B **95**, 205138 (2017).
J. Lim, G. Fabbris, D. Haskel, and J. S. Schilling, Phys. Rev. B **91**, 045116 (2015).
J. Lim, G. Fabbris, D. Haskel, and J. S. Schilling, Phys. Rev. B **91**, 174428 (2015).
S. Doniach, in *Valence Instabilities and Related Narrow-Band Phenomena*, edited by R. D. Parks (Plenum, New York, 1977), p. 169; S. Doniach, Physica B+C **91**, 231 (1977).
J. R. Schrieffer, J. Appl. Phys. **38**, 1143 (1967).
M. A. Ruderman and C. Kittel, Phys. Rev. **96**, 99 (1954).
J. S. Schilling, Adv. Phys. **28**, 657 (1979).
G. Fabbris, T. Matsuoka, J. Lim, J. R. L. Mardegan, K. Shimizu, D. Haskel, and J. S. Schilling, Phys. Rev. B **88**, 245103 (2013).
Y. C. Zhao, F. Porsch, and W. B. Holzapfel, Phys. Rev. B **50**, 6603 (1994).
Y. Vohra, L. Akella, S. Weir, and G. A. Smith, Phys. Lett. A **158**, 89 (1991).
J. C. Duthie and D. G. Pettifor, Phys. Rev. Lett. **38**, 564 (1977).
S. Blundell, in *Magnetism in Condensed Matter* (Oxford University Press, New York 2001) p. 34.
H. Adachi, H. Ino, and H. Miwa, Phys. Rev. B **56**, 349 (1997).
L. D. Jennings, E. D. Hill, and F. H. Spedding, J. Chem. Phys. **31**, 1240 (1959).
J. K. Alstad, R. V. Colvin, S. Legvold, and F. H. Spedding, Phys. Rev. **121**, 1637 (1961).
K. A. McEwen, P. F. Touborg, G. J. Cock and L. W. Roeland, J. Phys. F: Metal Phys. **4**, 2264 (1974).
W. C. Koehler and R. M. Moon, Phys. Rev. Lett. **29**, 1468 (1972).
W. Y. Dong, T. H. Lin, K. J. Dunn and C. N. J. Wagner, Phys. Rev. B **35**, 966 (1987).
C. R. Johnson, G. M. Tsoi, and Y. K. Vohra, J. Phys.: Condens. Matter **29**, 065801 (2017).
Material Preparation Center, Ames Laboratory, U.S. Department of Energy, Ames, Iowa, http://www.mpc.ameslab.gov.
James S. Schilling, in *Proceedings of the 9th AIRAPT International High Pressure Conf.*, Albany, New York, July 24-29, 1983, editors C. Homan, R.K. MacCrone and E. Whalley (North-Holland, N.Y., 1984); Mat. Res. Soc. Symp. Proc. **22**, 79 (1984).
W. B. Daniels and W. Ryschkewitsch, Rev. Sci. Instr. **54**, 115 (1983).
K. Shimizu, K. Amaya, and N. Suzuki, J. Phys. Soc. Jpn. **74**, 1345 (2005).
Y. Akahama and H. Kawamura, J. Appl. Phys. **100**, 043516 (2006).
W. Bi, J. Song, Y. Deng, P. Materne, J. Zhao, E. E. Alp, M. Y. Hu, D. Haskel, Y. Lee, and J. S. Schilling (unpublished).
See, for example, K. N. R. Taylor and M. I. Darby, *Physics of Rare Earth Solids* (Chapman and Hall, London, 1972).
H. J. van Daal and K. H. J. Buschow, Solid State Commun. **7**, 217 (1969).
R. V. Colvin, S. Legvold, and F. H. Spedding, Phys. Rev. **120**, 741 (1960).
J. Wittig, Phys. Rev. Lett. **24**, 812 (1970); J. J. Hamlin, V. G. Tissen, and J. S. Schilling, Physica C (Amsterdam) **451**, 82 (2007).
M. B. Maple, Appl. Phys. **9**, 179 (1976).
G. K. Samudrala, G. M. Tsoi, and Y. K. Vohra, J. Phys. Condens. Matter **24**, 362201 (2012).
D. D. Jackson, V. Malba, S. T. Weir, P. A. Baker, and Y. K. Vohra, Phys. Rev. B **71**, 184416 (2005).
G. S. Fleming and S. H. Liu, Phys. Rev. B **2**, 164 (1970); S. H. Liu, Phys. Rev. **127**, 1889 (1962).
G. N. Chesnut and Y. K. Vohra, Phys. Rev. B **61**, R3768 (2000).
N. C. Cunningham, W. Qiu, K. M. Hope, H.-P. Liermann, and Y. K. Vohra, Phys. Rev. B **76**, 212101 (2007).
R. Patterson, C. K. Saw, and J. Akella, J. Appl. Phys. **95**, 5443 (2004).
C. N. Singman, J. Chem. Ed. **61**, 137 (1984).
M. Maple, J. Wittig, and K. Kim, Phys. Rev. Lett. **23**, 1375 (1969).
J. Wittig, Phys. Rev. Lett. **46**, 1431 (1981).
J. Wittig, *Valence Instabilities*, edited by P. Wachter and H. Boppart (North-Holland, Amsterdam, 1982), p. 427.
M. Zuckermann, Phys. Rev. **168**, 390 (1968); E. Müller-Hartmann and J. Zittartz, Z. Physik **234**, 58 (1970).
[Figure Captions]{}
Fig 1. Four-point resistance data $R(T)$ from run 1 for Sm metal versus temperature on warming from 1.3 to 295 K at multiple pressures to 127 GPa (measured at room temperature). Knee in $R(T)$ at $T_{\text{o}}$ signals onset of magnetic order (for example, at 27 GPa $T_{\text{o}}$ $\approx $ 61 K). Straight red line fitting data above knee for 27 GPa intercepts resistance axis at 115 m$\Omega .$
Fig 2. Magnetic ordering temperature $T_{\text{o}}$ of Sm versus pressure. Data from run 1 ($\bullet $), data from run 2 ($\blacktriangle $), dotted line from Ref [@johnson], dashed line from Ref [@dong]. Value of pressure is estimated for temperature near $T_{\text{o}}$ (see text). Question marks (?) accompany data points where evidence for magnetic ordering is weak. Extended solid lines through data points are guides to the eye. Crystal structures for Sm at top of graph determined to 189 GPa [vohra]{}.
Fig 3. Plot of estimated maximum value of spin-disorder resistance $R_{\text{sd}}^{\text{max}}(P)$ versus pressure. $R_{\text{sd}}^{\text{max}}$ is estimated by subtracting defect resistance $R_{\text{d}}$ from intersection point $R_{\text{int}}$ on resistance axis of straight-line fit to $R(T)$ data for $T>T_{\text{o}}$ (see text). Also shown is pressure dependence$\ $of \[$R$(290 K) - $R$(4 K)\] using data from Fig 1.
Fig 4. Four-point resistance of Y(0.15 at.% Sm) alloy versus temperature showing superconducting transition at various pressures to 180 GPa (estimated at low temperature). Intersection of two red straight lines defines $T_{c}^{\text{onset}},$ midpoint of transition defines $T_{c},$ temperature where $R(T)\simeq $ 0 defines $T_{c}^{\text{zero}}.$
Fig 5. Superconducting transition temperature $T_{c}$ versus pressure (estimated at low temperature) for Y and Y(Sm) alloys at four different Sm concentrations. In all cases giant superconducting pair breaking $\Delta T_{c}\equiv T_{c}[$Y$]-T_{c}[$Y(Sm)$]$ is observed. At top of graph are crystal structures for superconducting host Y to 177 GPa [samudrala]{}.
Fig 6. Superconducting pair breaking $\Delta T_{c}$ divided by concentration $c$ of Sm in four Y(Sm) alloys plotted versus pressure. At top of graph are crystal structures for superconducting host Y to 177 GPa [samudrala]{}. Line through data is guide to the eye.
Fig 7. (a) Graph comparing lines through $T_{\text{o}}$ versus pressure data for Nd, Sm, Tb, and Dy. Vertical tick marks separate regions of conventional (to left) from unconventional (to right) behavior in $%
T_{\text{o}}(P).$ (b) Data in (a) is replotted versus $V/V_{\text{o}},$ where $V_{\text{o}}$ is sample volume at ambient pressure, using measured equations of state of Nd [@chesnut], Sm [@zhao], Tb [cunningham]{}, and Dy [@patterson]. In both graphs lines with double thickness mark regions where the magnetic ordering is unconventional.
Fig 8. Graph comparing relative superconducting pair breaking $\Delta T_{c}/c$ for dilute magnetic alloys Y(Nd), Y(Sm), Y(Tb), and Y(Dy) versus pressure showing lines through data as in Fig 6 for Y(Sm). At top of graph are crystal structures for superconducting host Y to 177 GPa [@samudrala].
Fig 9. Magnetic ordering temperature $T_{\text{o}}$ vs absolute value of negative exchange parameter $J$ according to the Doniach model [doniach]{}. Since $T_{\text{o}}$ from the RKKY interaction increases as $%
J^{2},$ but is overtaken by the exponential increase of the Kondo temperature $T_{\text{K}},$ the magnetic ordering is quenched.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A family $\mathcal{F}$ is $t$-$\it{intersecting}$ if any two members have at least $t$ common elements. Erdős, Ko, and Rado [@EKR] proved that the maximum size of a $t$-intersecting family of subsets of size $k$ is equal to $ {{n-t} \choose {k-t}}$ if $n\geq n_0(k,t)$. Alon, Aydinian, and Huang [@ALON] considered families generalizing intersecting families, and proved the same bound. In this paper, we give a strengthening of their result by considering families generalizing $t$-intersecting families for all $t \geq 1$. In 2004, Talbot [@TAL] generalized Bollobás’s Two Families Theorem [@BOL] to $t$-intersecting families. In this paper, we proved a slight generalization of Talbot’s result by using the probabilistic method.'
address:
- 'Department of Mathematical Sciences, KAIST, 291 Daehak-ro Yuseong-gu Daejeon, 305-701 South Korea'
- 'Department of Mathematical Sciences, KAIST, 291 Daehak-ro Yuseong-gu Daejeon, 305-701 South Korea / School of Mathematics, University of Birmingham, Edgbaston, Birmingham B15 2TT, United Kingdom'
- 'Department of Mathematical Sciences, KAIST, 291 Daehak-ro Yuseong-gu Daejeon, 305-701 South Korea'
author:
- 'Dong Yeap Kang, Jaehoon Kim, and Younjin Kim'
title: |
On the Erdős-Ko-Rado Theorem and the Bollobás Theorem\
for $t$-intersecting families
---
[^1]
Introduction
============
Let $[n]$ be the set $\{ 1, \ldots, n \} $ and $t$ be a positive integer. A family $\mathcal{F}$ of subsets of $[n] $ is $t$-intersecting if $|F_i \cap F_j | \geq t $ for every pair of two subsets $F_i,F_j \in \mathcal{F}$. A family $\mathcal{F}$ of subsets of $[n] $ is $k$-uniform if it is a collection of $k$-subsets of $[n]$, and we also say that it has rank $k$ if the largest set in it has size $k$. Erdős, Ko, and Rado [@EKR] proved that there exists $n_0(k,t)$ such that if $n \geq n_0(k,t)$, then the maximum size of a $k$-uniform $t$-intersecting family of subsets of $[n]$ is ${{n-t} \choose {k-t}}$. The following generalization of Erdős, Ko, and Rado (EKR) Theorem was proved by Frankl [@FR78] for $t \geq 15$, and was completed by Wilson [@WIL] for all $t$ by obtaining the smallest $n$ for the theorem to be true.\
\[Frankl [@FR78], Wilson [@WIL]\] If $\mathcal{F}$ is a $k$-uniform $t$-intersecting family of subsets of $[n]$, then we have $$|\mathcal{F}| \leq {{n-t} \choose {k-t}}$$ whenever $n \geq (k-t+1)(t+1)$.\
There is also an EKR-type theorem for $t$-intersecting families with bounded rank: if $n \geq (k-t+1)(t+1)$, then the maximum size of a $t$-intersecting family of subsets of sizes at most $k$ is equal to $ {{n-t} \choose {k-t}} + {{n-t} \choose {k-1-t}} + \cdots + {{n-t} \choose 0}$. In this paper we give the following strengthening of this theorem.
\[thm:t\_intersecting\] Let $n, k, t$ be integers such that $n \geq (k-t+1)(t+1)$. Let a collection of sets $\mathcal{F} \subseteq {{[n]}\choose {\leq k}}$ satisfies the following. Suppose that for any $A,B \in \mathcal{F}$ with $|A\cap B| <t$, $|A\triangle B| \leq k-t$ holds. Then we have $$|\mathcal{F}|\leq {{n-t} \choose {k-t}} + {{n-t} \choose {k-1-t}} + \cdots + {{n-t} \choose 0} .$$
Alon, Aydinian, and Huang [@ALON] gave the following strengthening of the bounded rank EKR theorem when $t=1$: $|\mathcal{F}|\leq {{n-t} \choose {k-t}} + {{n-t} \choose {k-1-t}} + \cdots + {{n-t} \choose 0}$ holds for all pairs $A,B$ with $|A\cap B|<1$ when $|A\triangle B| \leq k$, instead of $|A\triangle B|\leq k-1$. However, the condition $|A\triangle B| \leq k-t$ in Theorem \[thm:t\_intersecting\] cannot be replaced with $|A\triangle B| \leq k-t+1$ for $t\geq 2$ because of the following example.
For $t<k<n$, consider $\mathcal{F}=\{A\subseteq {{[n]}\choose {\leq k}} : [t]\subseteq A\} \cup \{A\subseteq [t]:|A|=t-1\}$. Then $\mathcal{F}$ contains ${{n-t} \choose {k-t}} + {{n-t} \choose {k-1-t}} + \cdots + {{n-t} \choose 0} + t$ sets while every two sets with $|A\cap B|\leq t-1$ satisfies $|A\triangle B| \leq k-t+1$. Thus our condition $|A\triangle B| \leq k-t$ is best possilbe when $t\geq 2$.\
We also extend results regarding cross $t$-intersecting families in the same manner. We say that families $\mathcal{F}_1, \mathcal{F}_2, \cdots, \mathcal{F}_r$ of subsets of $[n]$ are cross $t$-intersecting if $|A\cap B| \geq t $ for every $A \in \mathcal{F}_i$ and $B \in \mathcal{F}_j$, where $i \neq j$. In 2013, Borg [@BORG13] obtained the maximum product of sizes of cross $t$-intersecting families with bounded rank as follows. In this paper, we also show Theorem \[thm:cross\], a strengthening of Theorem \[thm:borg\].
\[thm:borg\] For $ 1 \leq t \leq k \leq n $, there exists $n_0(k,t)$ such that for all $n \geq n_0(k,t)$ the maximum size of product of families $\mathcal{F}_i \subseteq {{[n]} \choose {\leq k_i}}$ for $1\leq i \leq r$ is equal to $$\prod_{i=1}^{r}\left(\sum_{j=0}^{k_i-t} { {n-t} \choose {k_i-t-j}} \right)$$ when $\mathcal{F}_1, \mathcal{F}_2, \cdots, \mathcal{F}_r$ are cross $t$-intersecting families.
\[thm:cross\] For $ 1 \leq t \leq k_1\leq k_2\leq \cdots \leq k_r \leq n$, there exists $n_0(k_{r-1},k_{r},t)$ such that the following holds for all $n\geq n_0(k_{r-1},k_{r},t)$. Let $\mathcal{F}_1, \mathcal{F}_2, \cdots, \mathcal{F}_r$ be families of subsets of $[n]$ of size at most $k_1,k_2,\ldots,k_r$, respectively. If every $A \in \mathcal{F}_i $ and $B \in \mathcal{F}_j$ with $|A \cap B| <t $ satisfies $|A\triangle B| \leq \min\{k_i,k_j\} -t$, then we have $$\prod_{i=1}^r |\mathcal{F}_i| \leq \prod_{i=1}^{r}\left(\sum_{j=0}^{k_i-t} { {n-t} \choose {k_i-t-j}} \right).$$
Next, we consider $r$-wise intersecting families. We say that a family $\mathcal{F} \subseteq {{[n]} \choose k}$ is $r$-wise intersecting if $F_1 \cap F_2 \cap \cdots \cap F_r \neq \emptyset $ holds for all $F_i \in \mathcal{F}, 1 \leq i \leq r$. In the following Theorem \[cor1\], we give a strengthening of the bounded-rank version of Frankl’s result [@FR76] when $\mathcal{F}$ is an $r$-wise intersecting family.
\[f\] Let $(r-1)n \geq rk$ and $\mathcal{F} \subseteq {{[n]}\choose { k}}$ be an $r$-wise intersecting family. Then we have $$|\mathcal{F}|\leq {{n-1}\choose {k-1}} .$$
\[cor1\] Let $(r-1)n \geq rk$ and $\mathcal{F} \subseteq {{[n]}\choose {\leq k}}$ satisfies the following. Suppose that for any $F_1, F_2, \cdots, F_r \in \mathcal{F}$ if $ F_1\cap F_2\cap \cdots \cap F_r = \emptyset$, $|\cup_{i=1}^r F_i - \cap_{i=1}^rF_i| \leq k$ holds. Then we have $$|\mathcal{F}|\leq {{n-1}\choose {k-1}} + {{n-1}\choose {k-2}}+ \cdots {{n-1}\choose 0}.$$
The following Bollobás’s Two Families Theorem [@BOL] is an important and well-known result in Extremal Set theory. This theorem has been generalized in many directions. In 1982, Frankl [@FR82] proved a skew version of Theorem \[thm:bollobas\]. The further generalizations of Theorem \[thm:bollobas\] were given by Füredi [@FUREDI], Lovász [@LOVA], and Talbot [@TAL]. In 2004, Talbot [@TAL] generalized Bollobás’s two families theorem [@BOL] to $t$-intersecting families. We prove Theorem \[thm:improve\], an improvement of Talbot’s result, by using the probabilistic method. It also becomes to reprove Talbot’s Theorem \[thm:talbot\]. Note that the conditions $(c')$ in Theorem \[thm:improve\] is weaker than the condition $(c)$ of Theorem \[thm:talbot\].\
\[thm:bollobas\] Let $\mathcal{F}= \{ (A_i, B_i ) : i \in I\}$ be a finite collection of pairs of finite sets such that $A_i \cap B_j = \emptyset$ $\Longleftrightarrow$ $i=j$. Then we have $$\sum_{i \in I} {{|A_i|+|B_i|} \choose {|A_i|}}^{-1} \leq 1.$$
\[thm:talbot\] Let $\mathcal{F}= \{ (A_i, B_i ) : i \in I\}$ be a finite collection of pairs of finite sets and $t$ be a nonnegative integer such that $$\begin{aligned}
& (a) \ {\lvert A_i \cap B_i\rvert} \leq t \ \ {\text \ for \ each
\ } i \in I
\\
& (b) \ {\lvert A_i \cap B_j\rvert} \geq t \ \ {\text \ for
\ } i, j \in I {\text \ and \ } i \neq j
\\
& (c) \ {\text If \ } A_i \cap B_i = A_j \cap B_j { \text \ for \ } i \neq j \ \
{\text \ then \ } A_i \cap B_j \neq A_i \cap B_i \neq A_j \cap B_i.
\end{aligned}$$ Then we have $$\begin{aligned}
\sum_{i \in I}{\binom{|A_i \cup B_i |}{|A_i - B_i |}^{-1} \binom{|B_i |}{|A_i \cap B_i |}^{-1}} \leq 1.\end{aligned}$$
\[thm:improve\] \[thm:improve\] When $t\geq 1$, even if we replace the condition $(c)$ in Theorem \[thm:talbot\] with the following weaker condition $$\begin{aligned}
&(c') \ {\text If \ } A_i \cap B_i = A_j \cap B_j { \text \ for \ } i \neq j \ \
{\text \ then \ } A_i \cap B_j = A_i \cap B_i = A_j \cap B_i \text{ does not hold} \end{aligned}$$ we still have $$\begin{aligned}
\sum_{i \in I}{\binom{|A_i \cup B_i |}{|A_i - B_i |}^{-1} \binom{|B_i |}{|A_i \cap B_i |}^{-1}} \leq 1.\end{aligned}$$
Note that the above conditions in Theorem \[thm:improve\] are sharp in the sense that we cannot replace $(c)$ with $(c')$ when $t=0$ without changing the conclusion because of the following theorem.\
(Király, Nagy, Pálvőlgyi and Visontai [@KNPV]) \[examp\] Let $a,b$ be two relatively prime integers. There exists a finite collection of pairs of finite sets $\mathcal{F}= \{ (A_i, B_i ) : |A_i|=a, |B_i|=b,i \in I\}$ with $A_i\cap B_i = \emptyset$ for all $i\in I$. Moreover, $ A_i \cap B_j = A_i \cap B_i = A_j \cap B_i =\emptyset\text{ does not hold }$ for any $i\neq j$ and $$\sum_{i \in I}{\binom{|A_i \cup B_i |}{|A_i - B_i |}^{-1} \binom{|B_i |}{|A_i \cap B_i |}^{-1}} = |I|\binom{a+b}{a}^{-1} \geq 2-\frac{1}{a+b}.$$
Proof of Theorems
=================
For a family $\mathcal{F}$ of subsets of $[n]$ of size at most $k$, we define the transformation $S_{i,k}$ as follows.
$$S_{i,k,\mathcal{F}}(A):=\left\{
\begin{array}{ll}
A\cup \{i\} & \mbox{ if $A\cup \{i\}\notin \mathcal{F}$ and $|A|<k$} \\
A & \mbox{otherwise}\\
\end{array}
\right.$$
Also, let $S_{i,k}(\mathcal{F}) = \{ S_{i,k,\mathcal{F}}(A) : A\in \mathcal{F}\}$.\
\[c1\] A tuple of families $(\mathcal{F}_1,\mathcal{F}_2,\cdots,\mathcal{F}_r)$ is given with $\mathcal{F}_j \subseteq {{[n]}\choose {\leq k_j}}$ for $j=1,\ldots,r$ and $k = \min_j\{ k_j\}$ satisfying the following $$\label{1}
\text{If } |\bigcap_{j=1}^{r} F_j| < t \text{ for } F_j \in \mathcal{F}_j, \text{ then } |\bigcup_{j=1}^{r} F_j - \bigcap_{j=1}^{r} F_j| \leq k-t \text{ holds. }$$ Then, for $i\in [n]$, $(S_{i,k_1}(\mathcal{F}_1), S_{i,k_2}(\mathcal{F}_2), \cdots, S_{i,k_r}(\mathcal{F}_r))$ also satisfies (\[1\]).
Suppose that we have $F_j\in \mathcal{F}_j$ such that $F'_j=S_{i,k,\mathcal{F}_j}(F_j)$ with $|\bigcap_{j=1}^{r} F'_j|<t$ and $| \bigcup_{j=1}^{r} F'_j -\bigcap_{j=1}^{r} F'_j| > k-t$. Since $\bigcup_{j=1}^{r} F_j -\bigcap_{j=1}^{r} F_j$ is a subset of $\bigcup_{j=1}^{r} F'_j -\bigcap_{j=1}^{r} F'_j$ with size at most $k-t$, we have $$\bigcup_{j=1}^{r} F'_j -\bigcap_{j=1}^{r} F'_j= \{i\}\cup \bigcup_{j=1}^{r} F_j -\bigcap_{j=1}^{r} F_j .$$ Hence $i$ does not belong to any of $F_j$ and belongs to some, not all of $F'_j$. Assume that $i$ does not belong to $F'_1,\cdots,F'_l$ and belongs to $F'_{l+1},\cdots, F'_{r}$.
[**[Case 1.]{} There exists $\mathbf{j}$ with $\mathbf{|F_j|\geq k}$.**]{}
Since $|\bigcap_{j=1}^{r} F_j|<t$, we have $$|\bigcup_{j=1}^{r} F_j - \bigcap_{j=1}^{r} F_j| \geq |F_j - \bigcap_{j=1}^{r} F_j |> k-t.$$ It is a contradiction.
[**[Case 2.]{} $\mathbf{|F_j|<k}$ **[for all]{} $\mathbf{j}$.****]{}
For $j\geq l+1$, we have $S_{i,k,\mathcal{F}}(F_j)=F'_j=F_j\cup \{i\}$ because of $i\in F'_j$. For $j\leq l$, we also have $S_{i,k,\mathcal{F}}(F_j)=F'_j=F_j$ because of $i\notin F'_j$. The condition $k\leq k_j$ implies that $F_j\cup \{i\}\in \mathcal{F}_j$ for $j\leq l$. Then we have
$\displaystyle| \bigcap_{j=1}^{r-1} F_j\cap (F_r \cup \{i\})|<t$ and $\displaystyle |(F_r\cup \{i\})\cup \bigcup_{j=1}^{r-1} F_j - ( (F_r\cup\{i\})\cap \bigcap_{j=1}^{r-1}F_j)| =k-t+1.$
It is a contradiction. Hence such $F_j$s do not exist, and $(S_{i,k_1}(\mathcal{F}_1), S_{i,k_2}(\mathcal{F}_2), \cdots, S_{i,k_r}(\mathcal{F}_r))$ also satisfies (\[1\]).
We say that a family $\mathcal{F} \subseteq {{[n]}\choose {\leq k}}$ is an up-set if $A\cup \{i\} \in\mathcal{F}$ for all $A$ and $i$ such that $A\in \mathcal{F} \cap {{[n]}\choose {\leq k-1}}$ and $i \notin A$.
\[c2\] A tuple of families $(\mathcal{F}_1,\mathcal{F}_2,\cdots,\mathcal{F}_r)$ is given with $\mathcal{F}_j \subseteq {{[n]}\choose {\leq k_j}}$ for $j=1,\ldots,r$ and $k = \min_j\{ k_j\}$. If it satisfies (\[1\]) and $\mathcal{F}_j$s are up-sets for all $j$, then $|\bigcap_{j=1}^{r} F_j| \geq t$ holds for any $F_j\in \mathcal{F}_j$.
[ Suppose that we have $F_j \in \mathcal{F}_j$ for $j=1,2,\cdots,r$ with $|\bigcap_{j=1}^{r}F_j|<t$. By the condition, $| \bigcup_{j=1}^{r} F_j - \bigcap_{j=1}^{r} F_j|\leq k-t$ holds. It means $|\bigcup_{j=1}^{r}F_j|<k-1$, so $[n]-\bigcup_{j=1}^{r} F_j$ is not empty because of $n\geq k$. Let $[n]-\bigcup_{j=1}^{r} F_j = \{i_1,i_2,\cdots,i_l\}$. Since $\mathcal{F}_1$ is an up-set, $F_1\cup \{i_1\}$ is also in $\mathcal{F}_1$ unless $|F_1|=k_1$. By repeatedly adding this element $i_1, i_2,\ldots$ to $F_1$ until we cannot, we find a new set $F''_1 \in \mathcal{F}_1$ satisfying $|F''_1|=k_1$ or $F''_1 = F_1 \cup ([n]-\bigcup_{j=1}^{r} F_j)$. In any case, we have $$\left| \left(F''_1\cup \bigcup_{j=2}^{r} F_j \right) - \left(F''_1\cap \bigcap_{j=2}^{r} F_j \right) \right|\geq \min\{k_1-(t-1), n-(t-1)\}\geq k-t+1.$$ It is a contradiction. ]{}
For a family $\mathcal{F}\subseteq {{[n]}\choose{\leq k}}$, we let $\mathcal{F}^0=\mathcal{F}$ and take $\mathcal{F}^{i+1}=S_{n,k}(S_{n-1,k}(\cdots(S_{1,k}(\mathcal{F}^{i})\cdots)$ for all $i \geq 0$. If $\mathcal{F}^{j+1} = \mathcal{F}^j$, then we get an up-set $\mathcal{F}^j$ and we let $\mathcal{F}'=\mathcal{F}^j$. We also have $|\mathcal{F}|=|\mathcal{F}^i|$ for all $i \geq 0$ since $S_{i,k}$ does not change the size of a family in ${{[n]}\choose {\leq k} }$.
[Let $n \geq (k-t+1)(t+1)$. We take $\mathcal{F}'$ as above which has the same size with $\mathcal{F}$ and which is an up-set. By Claim \[c1\] with $\mathcal{F}_1=\mathcal{F}_2=\mathcal{F}$ and $r=2$, we say that the up-set $\mathcal{F}'$ has the property (\[1\]) as $\mathcal{F}$ does. Then, Claim \[c2\] with $\mathcal{F}_1=\mathcal{F}_2=\mathcal{F}$ implies that the up-set $\mathcal{F}'$ is $t$-intersecting. Hence, we have that $$|\mathcal{F}|=|\mathcal{F}'| \leq \sum_{i=0}^{k-t} {{n-t}\choose {k-i-t}}.$$]{}
Let $ 1 \leq t \leq k_1\leq k_2\leq \cdots \leq k_r \leq n$ and $n \geq n_0(k_{r-1},k_r,t)$ with $n_0(k_{r-1},k_r,t)$ as in Theorem 1.2 in [@BORG13]. For each $\mathcal{F}_i \subseteq {{[n]}\choose {\leq k_i}}$, we define $\mathcal{F}'_i$ by repeatedly applying $S_{j,k_i}$ for all $j$ as above. Then $\mathcal{F}'_1, \mathcal{F}'_2, \cdots, \mathcal{F}'_r$ are up-sets satisfying (\[1\]) having the same size with $\mathcal{F}_1, \mathcal{F}_2, \cdots, \mathcal{F}_r$, respectively. By Claim \[c1\], we have $|A \triangle B| \leq \min\{k_i,k_j\} -t$ for any $A \in \mathcal{F}'_i , B \in \mathcal{F}'_j$ with $|A \cap B| < t$. For all possible pair of two distinct numbers $j_1,j_2 \in [n]$, we apply Claim \[c2\] with $\mathcal{F}'_{j_1},\mathcal{F}'_{j_2}$. Then we can conclude that $\mathcal{F}'_1, \mathcal{F}'_2,\cdots, \mathcal{F}'_r$ are cross $t$-intersecting. By Theorem \[thm:borg\], we have that $$\prod_{i=1}^{r}|\mathcal{F}_i|=\prod_{i=1}^{r}|\mathcal{F}'_i| \leq \prod_{i=1}^{r}\left(\sum_{j=0}^{k_i-t} { {n-t} \choose {k_i-t-j}} \right).$$
Let $n \geq (k-t+1)(t+1)$. We take $\mathcal{F}'$ as above which has the same size with $\mathcal{F}$ and which is an up-set. By Claim \[c1\] with $\mathcal{F}_1=\mathcal{F}_2=\cdots =\mathcal{F}_r=\mathcal{F}$, we say that an up-set $\mathcal{F}'$ has the property (\[1\]) with $t=1$ as $\mathcal{F}$ does. Then, Claim \[c2\] with $\mathcal{F}_1=\mathcal{F}_2=\cdots=\mathcal{F}_r=\mathcal{F}$ implies that the up-set $\mathcal{F}'$ is intersecting. We let $\mathcal{F}'[i]$ denote the collection of all sets of size $i$ in $\mathcal{F}'$, then $\mathcal{F}'[i]$ is also intersecting. Hence, Theorem \[f\] implies $|\mathcal{F}'[i]| \leq {{n-1}\choose{i-1}}$.
Hence, we conclude that $$|\mathcal{F}|=|\mathcal{F}'|= \sum_{i=1}^{k} |\mathcal{F}'[i]| \leq \sum_{i=1}^{k} {{n-1}\choose {i-1}}.$$
A probablistic proof of Theorem \[thm:talbot\] and Theorem \[thm:improve\]
==========================================================================
In this section, we prove both Theorem \[thm:talbot\] and Theorem \[thm:improve\] in the probabilistic argument. Let $\mathcal{F}= \{(A_i,B_i): i\in I\}$ be a finite collection of pairs of finite sets satisfying the conditions on Theorems. Let $[n] = \bigcup_{i \in I}{A_i \cup B_i}$ and $|I|=k$. A permutation $\sigma = (x_1 , x_2 , ... , x_n )$ on $[n]$ is *properly separating* $(A_i , B_i)$, where $i \in I$, if $x_s \in A_i-B_i$ and $x_l \in B_i$ implies $s < l$ and $x_s \in A_i$ and $x_l \in B_i-A_i$ also implies $s< l$. Then all elements of $A_i - B_i$ should be on the left side of $B_i$, and all elements of $B_i - A_i$ should be on the right side of $A_i$ in the permutation $\sigma$ that properly separates $(A_i , B_i)$.
Any permutation $\sigma \in S_n$ does not properly separate both $(A_i , B_i)$ and $(A_j , B_j)$, $i \ne j$.
Suppose $\sigma = (x_1 , x_2 , ... , x_n)$ properly separates both $(A_i , B_i)$ and $(A_j , B_j)$, where $i \neq j$. Note that $\max\{i | x_i\in A_i\} \leq \max\{i |x_i\in B_i\}$ by the definition of proper separation.
We define the following indices $$i_{b} = \min \left \{ s \: | \: x_s \in B_i \right \}, i_{a} = \max \left \{ s \: | \: x_s \in A_i \right \}, j_{b} = \min \left \{ l \: | \: x_l \in B_j \right \}, j_{a} = \max \left \{ l \: | \: x_l \in A_j \right \}.$$ Without loss of generality, we may assume $j_{b} \leq i_{b}$.\
[**[Case 1.]{} $t=0$.**]{}
Since $A_j\cap B_j =\emptyset$ and $\sigma$ properly separates $(A_j,B_j)$, we have $j_{a}<j_{b}$. Thus $j_{a} < j_{b} \leq i_{b}$, so we conclude that $A_j\cap B_i =\emptyset$. It is a contradiction to the condition $(c)$.\
[**[Case 2.]{} $t\geq 1$.**]{} Since $j_b \leq i_b$, we have $$\begin{aligned}
A_j \cap B_i & = \left \{ x_l \: | \: x_l \in A_j \:, \: l\geq i_{b} \right \} \subseteq \left \{ x_l \: | \: x_l \in A_j \:, \: l\geq j_{b} \right \} = A_j\cap B_j.\end{aligned}$$ This with the conditions $(a)$ and $(b)$ imply that $$|A_j\cap B_i| =|A_j\cap B_j| =t \geq 1.$$ Since $A_j\cap B_j$ is nonempty, $x_{j_{b}} \in A_j\cap B_j = A_j\cap B_i\subseteq B_i$. By our choice, $i_b$ is the minimum index of $B_i$, we have $i_{b}\leq j_{b}$. Then we conclude $i_{b}=j_{b}$. Hence, $$\begin{aligned}
A_i \cap B_j = \left \{ x_l \: | \: x_l \in A_i \:, \: l\geq j_{b} \right \}=\left \{ x_l \: | \: x_l \in A_i\:, \: l\geq i_{b} \right \} = A_i\cap B_i.\end{aligned}$$ Again by conditions $(a)$ and $(b)$, $A_i \cap B_j=A_i\cap B_i$ has size $t$.
If $i_{a} \geq j_{a}$, we get $$\begin{aligned}
A_j \cap B_i & = \left \{ x_l \: | \: x_l \in B_i \: , \: l \leq j_a \} \right \}\subseteq \left \{ x_l \: | \: x_l \in B_i \: , \: l\leq i_a \right \} = A_i\cap B_i.\end{aligned}$$ If $i_{a} \leq j_{a}$, we get $$\begin{aligned}
A_i \cap B_j & = \left \{ x_l \: | \: x_l \in B_i \: , \: l \leq i_a \} \right \}\subseteq \left \{ x_l \: | \: x_l \in B_i \: , \: l\leq j_a \right \} = A_j\cap B_j.\end{aligned}$$
One of $i_{a} \geq j_{a}$ and $i_{a} \leq j_{a}$ must hold. Then either $A_j\cap B_i = A_i\cap B_i$ or $A_i\cap B_j = A_j\cap B_j$ holds, where $ i \neq j$. In any case, we conclude that $A_i\cap B_j = A_j\cap B_j = A_i\cap B_i = A_j\cap B_j$ for $t \geq 1$. It is a contradiction to the condition $(c')$.
Pick a permutation $\sigma \in S_n$ uniformly and independently, and let $E_i$ be the event that $\sigma$ properly separates $(A_i , B_i)$. Then the events $E_1 , E_2 , ... , E_k$ are mutually disjoint, and $\Pr[E_i ] = \binom{|A_i \cup B_i |}{|A_i - B_i |}^{-1} \binom{|B_i|}{|A_i \cap B_i |}^{-1}$ for $1 \leq i \leq k$. Then we conclude that
$$\begin{aligned}
\Pr\left [\bigcup_{i=1}^{k}{E_i} \right ] = \sum_{i=1}^{k}{\Pr[E_i]} = \sum_{i=1}^{k}{\binom{|A_i \cup B_i |}{|A_i - B_i |}^{-1} \binom{|B_i |}{|A_i \cap B_i |}^{-1}} \leq 1\end{aligned}$$
as desired.
Acknowledgement
===============
The authors are indebted to the anonymous referee who gave us valuable comments and pointed out a mistake in the former proof of Theorem \[thm:improve\].
[99]{} N. Alon, H. Aydinian and H. Huang, *Maximizing the number of nonnegative subsets*, SIAM J. Discrete Math., to appear, 2014. B. Bollobás, *On generalized graphs*, Acta Math. Acad. Sci. Hungar. **16** (1965) 447-452. P. Borg, *The maximum product of sizes of cross $t$-intersecting uniform families*, submitted 2013.
P. Erdős, C. Ko and R. Rado, *Intersection theorems for systems of finite sets*. Quart. J. Math. Oxford Ser., **12** (1961), 313-318.
P. Frankl, *On Sperner families satisfying an additional condition*, J. Combin. Theory Ser. A **20** 1-11 (1976).
P. Frankl, *The Erdős-Ko-Rado theorem is true for $n=ckt$*, in Combinatorial Proc. Fifth Hungarian Coll. Combinatorics, Keszthely, 1976, North-Holland, Amsterdam. 1978, 365-375. P. Frankl, *An extremal problem for two families of sets*, European J. Combin. **3** (1982) 125-127. P. Frankl and N. Tokushige, *On $r$-Cross intersecting Families of Sets*, Combinatorics, Probability and Computing, **20** 749-752. Z. Füredi, *Geometrical Solution of an Intersection Problem for Two Hypergraphs*, Europ. J. Combinatorics (1984) **5**, 133-136.
Z. Király, Z. L. Nagy, D. Pálvőlgyi, M. Visontai, *Fundamenta Informaticae* (1–4) (2012), 189–198.
L. Lovász, Combinatorial Problems and Exercises, North-Holland, Amsterdam, New York, Oxford, 1979. J. Talbot, *A new Bollobas-type inequality and applications to $t$-intersecting families of sets* , Discrete Math. **285** (2004), 349-353. R. Wilson, *The exact bound in the Erdős-Ko-Rado theorem*, Combinatorica **4** (1984) 247–257.
[^1]: The first author and second author were partially supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Science, ICT & Future Planning (2011-0011653).\
The second author was also partially supported by the European Research Council under the European Union’s Seventh Framework Programme (FP/2007–2013) / ERC Grant Agreements no. 306349 (J. Kim).\
The third author was supported by Basic Science Research Program through the National Research Foundation of Korea(NRF) funded by the Ministry of Science, ICT & Future Planning (2013R1A1A3010982) (Y. Kim).\
Corresponding author: Younjin Kim
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We use our recently proposed microscopic modeling in the framework of linear response theory to investigate the complete phonon dispersion, the phonon density of states, certain phonon-induced electronic charge distributions and charge fluctuations (CF’s) for anomalous soft modes of metallic BaBiO$_{3}$ in its simple cubic phase where superconductivity with $T_{c}$ up to 32 K appears. The theoretical approach already has been applied successfully to the cuprate high-temperature superconductors (HTSC’s), simple ionic crystals (NaCl, MgO) and perovskite oxides (SrTiO$_{3}$, BaTiO$_{3}$). It is well suited for materials with a strong component of ionic binding and especially for “ionic” metals. In particular, the giant phonon anomalies related to the breathing vibration of the oxygen as found experimentally in superconducting doped Ba$_{0.6}$K$_{0.4}$BiO$_{3}$, resembling those observed in the high $T_{c}$ cuprates, are investigated. The origin of these anomalies is explored and attributed to a strong nonlocal coupling of the displaced oxygen ions to CF’s of ionic type, essentially of the Bi6s- and Bi6p orbital. This points to the importance of both of these states at the Fermi energy. Starting from an ab-initio rigid ion model (RIM) we calculate the effect on the lattice dynamics and charge response of the most important electronic polarization processes in the material, i.e. CF’s and dipole fluctuations (DF’s). Taking into account these electronic degrees of freedom in linear response theory, we obtain a good agreement with the measured phonon dispersion and in particular with the strong phonon anomalies.'
author:
- Claus Falter
- Thomas Bauer
- Thomas Trautmann
title: 'A microscopic modeling of phonon dynamics and charge response in metallic BaBiO$_{3}$'
---
Introduction
============
Potassium doped Ba$_{1-x}$K$_{x}$BiO$_{3}$ with a transition temperature $T_{c}$ up to 32 K for $x \approx 0.4$ has been intensively studied with no final resolution of its high-temperature superconductivity. Different from the cuprate high-temperature superconductors (HTSC’s) its bonding is three dimensional and no antiferromagnetic ordering exists for the insulating parent compound indicating that there are no strong electronic correlations near $x = 0$. Common to both materials, however, is the strong component of ionic binding favoring long-ranged Coulomb interactions and a related strong nonlocal electron-phonon coupling in terms of localized charge fluctuations (CF’s) of the ionic orbitals. This type of coupling has been shown in the cuprate-HTSC’s to lead to the generic phonon anomalies of the high-frequency oxygen bond-stretching modes (OBSM) [@Ref01; @Ref02; @Ref03; @Ref04; @Ref05]. Recently, giant phonon anomalies for similar OBSM (oxygen breathing modes) also have been found by inelastic neutron scattering measurements in superconducting Ba$_{0.6}$K$_{0.4}$BiO$_{3}$ [@Ref07; @Ref08; @Ref09]. Thus, while spin-degrees of freedom seem to play no decisive role for superconductivity in Ba-Bi-O the effects of nonlocal coupling of lattice and charge degrees of freedom in terms of ionic CF’s are still present, likewise as in the cuprate-HTSC’s, and provide a coupling channel for pairing in a phonon-mediated mechanism.
Ba$_{1-x}$K$_{x}$BiO$_{3}$ has a complex structural phase diagram [@Ref10]. For the insulating composition $0 \leq x < 0.12$ the structure is monoclinic and may be derived from the cubic perovskite structure by simultaneous octahedral tilting $(t)$ and symmetric oxygen breathing-mode $(b)$ distortions (frozen-in OBSM instability at the $R$-point of the cubic phase). For $0.12 < x < 0.37$ the structure is still insulating but orthorhombic with a $t$, but no $b$ distortion and for the superconducting, metallic composition $(0.37 < x < 0.53)$ the structure is simple cubic, i.e. no $t$ and $b$ distortion.
It is widely believed that the insulating phases are generated by charge density instabilities associated with breathing and tilting distortions but it has proven difficult to establish this by first principles calculations. Recent investigations [@Ref11] indicate, in contrast to earlier results [@Ref12], that the local-density approximation (LDA) seriously underestimates the breathing distortions and yields a metallic ground state with no sign of a charge-density-wave instability.
Moreover, first principles linear response calculations of the phonon dispersion of cubic Ba-Bi-O [@Ref11] overestimate the high-frequency branches by about $20\,\%$ in comparison with the experiments. In particular the most strongly renormalized anomalous OBSM, $O^{R}_{B}$, at the $R$-point, which according to the experiments [@Ref09] is found at about 10.5 THz has a frequency of 15.7 THz in the calculations presented in Ref. . For the modes with lower frequencies there is a better agreement of the order of $10\,\%$. Thus, linear response calculations based on LDA-like electronic band structures, using a virtual-crystal approximation seem to underestimate the kinetic part of the charge response, i.e. the electronic polarizability matrix. In order to investigate this point, we study different cases for the polarizability matrix $\Pi_{\kappa\kappa'}({\mathbf{q}}\,)$ in our microscopic model approach to extract the most important electronic degrees of freedom for the charge response. The indices $\kappa,\,\kappa'$ denote the orbital degrees of freedom in an elementary cell of the crystal.
Empirical shell model calculations [@Ref07; @Ref08; @Ref09] incorporating metallic screening by the Lindhard function are also unable to reproduce the downward dispersion and softening of the phonon anomalies, instead an upward dispersion is calculated. Finally, calculations in Ref. based on a tight-binding representation of the electronic contribution to the dynamical matrix [@Ref14] and an empirical contribution for the short-ranged force constants fitted to the experimental frequencies lead to a remarkable softening for several modes especially around the $M$- and the $R$-point with strongest softening for the (1, 1, 0) direction, in disagreement with the experiments where only the OBSM of breathing type are strongly renormalized and the largest softening is along the (1, 1, 1) direction.
In the present work we calculate within our microscopic modeling of the electronic charge response in linear response theory the complete phonon dispersion, the phonon density of states, the phonon-induced electronic charge redistribution and the CF’s for the OBSM, $O^{X}_{B}$, $O^{M}_{B}$, $O^{R}_{B}$, of simple cubic Ba-Bi-O. In the past, this theoretical approach, which is well suited for systems with a strong component of ionic binding, has been applied successfully to the cuprate-HTSC’s [@Ref01; @Ref02; @Ref03; @Ref04; @Ref05; @Ref06], i.e., “ionic” metals in which the band crossing the Fermi level has admixture of both anion and cation orbitals, simple ionic crystals [@Ref15] and the perovskite oxides (SrTiO$_{3}$, BaTiO$_{3}$) [@Ref16].
From a general point of view our treatment of the electronic density response and lattice dynamics in terms of dipole fluctuations (DF’s) and CF’s can be considered as a microscopic (semi-ab initio) implementation of the phenomenological dipole-shell model or the charge-fluctuation models, respectively. For a general formulation of phenomenological models for lattice dynamics that use localized electronic variables as adiabatic degrees of freedom, see, for example, Ref. . This formulation covers shell models, bond-charge models and charge-fluctuation models. While in such an approach the coupling coefficients are treated simply as empirical fitting parameters the essential point in our scheme is that all the couplings are microscopically well defined and can be calculated.
The ionic nature of Ba-Bi-O is described by an ab initio rigid ion model (RIM) leading to a local rigid charge response and electron-phonon interaction (EPI), respectively. The missing nonlocal, nonrigid part of the electronic density response and EPI is expressed by microscopically well defined CF’s and DF’s on the outer shells of the ions. Starting from the RIM as a reference system the effect of the nonrigid electronic polarization effects in terms of CF’s and DF’s on the phonon dispersion and, particularly, the anomalies is calculated and compared with the experiment.
The article is organized as follows. In section 2 we outline the theory and modeling to provide a better reading of the article. Section 3 presents our calculated results of the phonon dispersion and gives a detailed discussion of the effects of screening by DF’s and CF’s on the phonon anomalies by comparing the corresponding results with those as obtained from the RIM as a reference system. Moreover, phonon density of states, the charge redistribution and the CF’s for the OBSM of breathing type are presented. Finally, section 4 contains a summary and discussion.
Sketch of the theory and modeling
=================================
In the following a survey of the theory and modeling is presented. A more detailed description can be found in Ref. and in particular in Ref. where the calculation of the microscopic coupling parameters is given. The local, rigid part of the electronic charge response and the EPI is approximated by an ab initio RIM taking into account ion-softening in terms of (static) effective ionic charges and scaling of the short-ranged part of certain pair potentials between the ions to simulate covalence effects in the model. This is done in such a way that the energy-minimized structure is as close as possible to the experimental one [@Ref18]. Structure optimization and energy minimization is very important for a reliable calculation of the phonon dynamics through the dynamical matrix.
The RIM with the corrections just mentioned serves as an unbiased reference system for the investigation of the effect of the nonrigid electronic polarization processes on the phonon dynamics. The latter are modeled in form of electronic CF’s on the outer shells of the ions. Especially in the metallic state the CF’s dominate the nonlocal contribution of the charge response and the EPI. In addition DF’s are admitted in our approach [@Ref05; @Ref15]. Thus, the basic variable of our model is the ionic density which is given in the perturbed state by
$$\label{Eq1}
\rho_\alpha({\mathbf{r}},Q_\lambda, {\mathbf{p}}_\alpha) =
\rho_\alpha^0(r) + \sum_{\lambda}Q_\lambda \rho_\lambda^\text{CF}(r)
+ {\mathbf{p}}_\alpha \cdot
\hat{{\mathbf{r}}} \rho_\alpha^\text{D}(r).$$
$\rho^{0}_{\alpha}$ is the density of the unperturbed ion, as used in the RIM, localized at the sublattice $\alpha$ of the crystal and moving rigidly under displacement. The $Q_{\lambda}$ and $\rho_{\lambda}^{CF}$ describe the amplitude and the form-factors of the CF’s and the last term in Eq. represents the dipolar deformation of an ion $\alpha$ with amplitude (dipole moment) ${\mathbf{p}}_{\alpha}$ and a radial density distribution $\rho^{D}_{\alpha}$. $\hat{{\mathbf{r}}}$ denotes the unit vector in the direction of ${\mathbf{r}}$. The $\rho_{\lambda}^{CF}$ are approximated by a spherical average of the orbital densities of the ionic shells calculated in LDA taken self-interaction effects (SIC) into account. The dipole density $\rho^{D}_{\alpha}$ is obtained from a modified Sternheimer method in the framework of LDA-SIC [@Ref15]. All SIC calculations are performed for the average spherical shell in the orbital-averaged form according to Ref. . For the correlation part of the energy per electron, $\epsilon$, the parametrization given in Ref. has been used.
The total energy of the crystal is obtained by assuming that the density can be approximated by a superposition of overlapping densities $\rho_{\alpha}$. The $\rho_{\alpha}^{0}$ in Eq. are also calculated within LDA-SIC taking environment effects, via a Watson-sphere potential and the static effective charges of the ions into account. The Watson-sphere method is only used for the oxygen ions and the depth of the Watson-sphere potential is set as the Madelung potential at the corresponding site. Finally, applying the pair-potential approximation we get the total energy
$$\label{Eq2}
E(R,\zeta) = \sum_{{\mathbf{a}},\alpha} E_\alpha^{\mathbf{a}}(\zeta)
+\frac{1}{2}\sum_{({\mathbf{a}},\alpha)\neq({\mathbf{b}},\beta)}
\Phi_{\alpha\beta}
\left({\mathbf{R}}^{\mathbf{b}}_\beta-{\mathbf{R}}^{\mathbf{a}}_\alpha,\zeta\right).$$
The energy $E$ depends on both, the configuration of the ions $\{R\}$ and the electronic (charge) degrees of freedom (EDF) $\{\zeta\}$ of the charge density, i.e. $\{Q_{\lambda}\}$ and $\{{\mathbf{p}}_{\alpha}\}$ in Eq. . $E^{{\mathbf{a}}}_{\alpha}$ are the energies of the single ions. ${\mathbf{a}}$, ${\mathbf{b}}$ denote the elementary cells and $\alpha$, $\beta$ the corresponding sublattices. The second term in Eq. is the interaction energy of the system expressed in terms of anisotropic pair interactions $\Phi_{\alpha\beta}$. Both $E^{{\mathbf{a}}}_{\alpha}$ and $\Phi_{\alpha\beta}$ in general depend upon $\zeta$ via $\rho_{\alpha}$ in Eq. .
The pair potentials in Eq. can be separated into long-ranged Coulomb contributions and short-ranged terms as follows:
$$\begin{aligned}
\nonumber
\Phi_{\alpha\beta}({\mathbf{R}},\zeta) =& \frac{\mathcal{Z}_\alpha \mathcal{Z}_\beta}{R}
-(\mathcal{Z}_\alpha {\mathbf{p}}_\beta + \mathcal{Z}_\beta {\mathbf{p}}_\alpha)\cdot\frac{{\mathbf{R}}}{R^3}
+\frac{{\mathbf{p}}_\alpha\cdot{\mathbf{p}}_\beta}{R^3}\\\label{Eq3}
&-3\frac{({\mathbf{p}}_\alpha\cdot{\mathbf{R}})({\mathbf{R}}\cdot{\mathbf{p}}_\beta)}{R^5}
+ \widetilde{\Phi}_{\alpha\beta}({\mathbf{R}},\zeta),\end{aligned}$$
$$\begin{aligned}
\nonumber
\widetilde{\Phi}_{\alpha\beta}({\mathbf{R}},\zeta) =& K_\alpha U_\beta({\mathbf{R}},\zeta)
+ K_\beta U_\alpha({\mathbf{R}},\zeta)\\\label{Eq4}
&+ W_{\alpha\beta}({\mathbf{R}},\zeta) + G_{\alpha\beta}({\mathbf{R}},\zeta).\end{aligned}$$
The first term in Eq. describes the long-ranged ion-ion, the second the dipole-ion and the third and fourth term the dipole-dipole interaction. ${\cal{Z}}_{\alpha}$ and ${\cal{Z}}_{\beta}$ are the variable charges of the ions in case CF’s are excited. The latter reduce to the ionic charges for rigid ions. $K_{\alpha}$ and $K_{\beta}$ are the charges of the ion cores. $\widetilde{\Phi}_{\alpha\beta}$ represents the short-ranged interactions. These can be expressed by the following integrals $$\begin{aligned}
\label{Eq5}
U_\alpha({\mathbf{R}},\zeta) &= - \int d^3r \rho_\alpha({\mathbf{r}},\zeta)
\left( \frac{1}{|{\mathbf{r}}-{\mathbf{R}}|} - \frac{1}{R} - \frac{{\mathbf{r}}\cdot{\mathbf{R}}}{R^3} \right),
\\\nonumber
W_{\alpha\beta}({\mathbf{R}},\zeta) &= \int d^3r \int d^3r'
\bigl[\rho_\alpha({\mathbf{r}},\zeta) \rho_\beta({\mathbf{r}}',\zeta) \times\\\label{Eq6}
&\times\left(\frac{1}{|{\mathbf{r}}-{\mathbf{r}}'-{\mathbf{R}}|}-\frac{1}{R}-\frac{({\mathbf{r}}+{\mathbf{r}}')\cdot{\mathbf{R}}}{R^3} \right) \bigr],
\\\nonumber
G_{\alpha\beta}({\mathbf{R}},\zeta) &= \int d^3r \bigl[\rho_{\alpha\beta}({\mathbf{r}},\zeta)\epsilon(\rho_{\alpha\beta}({\mathbf{r}},\zeta))\\
&-\rho_{\alpha}({\mathbf{r}},\zeta)\epsilon(\rho_{\alpha}({\mathbf{r}},\zeta))\\\nonumber
&-\rho_{\beta}({\mathbf{r}}-{\mathbf{R}},\zeta)\epsilon(\rho_{\beta}({\mathbf{r}}-{\mathbf{R}},\zeta))\bigr],\label{Eq7}\end{aligned}$$ with
$$\label{Eq8}
\rho_{\alpha\beta}({\mathbf{r}},\zeta) = \rho_\alpha({\mathbf{r}},\zeta)+\rho_\beta({\mathbf{r}}-{\mathbf{R}},\zeta).$$
$K_{\alpha}\,U_{\beta}({\mathbf{R}},\,\zeta)$ yields the short-ranged contribution of the interaction between the core $\alpha$ and the density $\rho_{\beta}$ according to Eq. . $W_{\alpha\beta}({\mathbf{R}},\,\zeta)$ represents the short-ranged Coulomb contribution of the interaction of the density $\rho_{\alpha}$ with the density $\rho_{\beta}$ and $G_{\alpha\beta}({\mathbf{R}},\,\zeta)$ is the sum of the kinetic one-particle- and the exchange-correlation (XC) contribution of the interaction between the two ions [@Ref15]. The short-ranged part of the potentials and the various coupling coefficients are calculated numerically for a set of distances $R$ between the ions. The results so obtained are then described by an analytical function of the form
$$\label{Eq9}
f(R) = \pm \text{exp}\left(\alpha+\beta R+\frac{\gamma}{R}\right).$$
$\alpha$, $\beta$, and $\gamma$ in Eq. are fit parameters.
From the adiabatic condition
$$\label{Eq10}
\frac{\partial E(R,\zeta)}{\partial \zeta} = 0$$
an expression for the atomic force constants, and accordingly the dynamical matrix in harmonic approximation consistent with linear response theory can be derived:
$$\begin{aligned}
\label{Eq11}\nonumber
t_{ij}^{\alpha\beta}({\mathbf{q}}) &=
\left[t_{ij}^{\alpha\beta}({\mathbf{q}})\right]_\text{RIM}\\ &-
\frac{1}{\sqrt{M_\alpha M_\beta}} \sum_{\kappa,\kappa'}
\left[B^{\kappa\alpha}_i({\mathbf{q}}) \right]^{*} \left[C^{-1}({\mathbf{q}})
\right]_{\kappa\kappa'} B^{\kappa'\beta}_j({\mathbf{q}}).\end{aligned}$$
The first term on the right hand side denotes the contribution from the RIM. $M_{\alpha}$, $M_{\beta}$ are the masses of the ions and ${\mathbf{q}}$ is a wave-vector from the first Brillouin zone. The quantities ${\mathbf{B}}({\mathbf{q}}\,)$ and $C({\mathbf{q}}\,)$ in Eq. represent the Fourier transforms of the electronic coupling coefficients as calculated from the energy in Eq. , and the pair potentials in Eqs. -, respectively:
$$\begin{aligned}
\label{Eq12}
{\mathbf{B}}_{\kappa\beta}^{{\mathbf{a}}{\mathbf{b}}} &= \frac{\partial^2
E(R,\zeta)}{\partial \zeta_\kappa^{\mathbf{a}} \partial R_\beta^{\mathbf{b}}},
\\\label{Eq13} C_{\kappa\kappa'}^{{\mathbf{a}}{\mathbf{b}}} &= \frac{\partial^2
E(R,\zeta)}{\partial \zeta_\kappa^{\mathbf{a}} \partial
\zeta_{\kappa'}^{\mathbf{b}}}.\end{aligned}$$
$\kappa$ denotes the EDF (CF and DF in the present model) in an elementary cell. The ${\mathbf{B}}$ coefficients describe the coupling between the EDF and the displaced ions (bare electron-phonon coupling), and the coefficients $C$ determine the interaction between the EDF. The phonon frequencies, $\omega_{\sigma}({\mathbf{q}}\,)$, and the corresponding eigenvectors, ${\mathbf{e}}^{\,\alpha}({\mathbf{q}}\,\sigma)$, of the modes $({\mathbf{q}}\,\sigma)$ are obtained from the secular equation for the dynamical matrix in Eq. , i.e.
$$\label{Eq14}
\sum_{\beta,j} t_{ij}^{\alpha\beta}({\mathbf{q}})e_j^\beta({\mathbf{q}}) =
\omega^2({\mathbf{q}}) e_i^\alpha({\mathbf{q}}).$$
The lengthy details of the calculation of ${\mathbf{B}}$ and $C$ cannot be reviewed in this paper. They are given in Ref. . In this context we remark that the coupling matrix $C_{\kappa\kappa'}({\mathbf{q}}\,)$ of the EDF-EDF interaction, whose inverse appears in Eq. for the dynamical matrix, can be written in matrix notation as
$$\label{Eq15}
C = \Pi^{-1} + \widetilde{V}.$$
$\Pi^{-1}$ contains the kinetic single particle contribution to the interaction $C$ and $\widetilde{V}$ the Hartree and exchange-correlation contribution. $C^{-1}$ needed for the dynamical matrix and the EPI is closely related to the (linear) density response function (matrix) and to the inverse dielectric function (matrix) $\varepsilon^{-1}$, respectively. Comparing with calculations of the phonon dispersion and the EPI using the linear response method in form of density functional perturbation theory within local density approximation (LDA), these calculations correspond to calculating $\Pi$ and $\widetilde{V}$ in DFT-LDA. On the oder hand, in our microscopic modeling DFT-LDA-SIC calculations are performed for the various densities in Eq. in order to obtain the coupling coefficients ${\mathbf{B}}$ and $\widetilde{V}$. SIC as a correction for a single particle term is important for contracting in particular localized orbitals. Written in matrix notation we get for the density response matrix the relation
$$\label{Eq16}
C^{-1} = \Pi(1+\widetilde{V}\Pi)^{-1} \equiv \Pi \varepsilon^{-1},
\hspace{.7cm} \varepsilon = 1 + \widetilde{V}\Pi.$$
The CF-CF submatrix of the matrix $\Pi$ can for example approximated from a tight-binding approximation of a single particle electronic bandstructure. In this case the electronic polarizability $\Pi$ reads
$$\begin{aligned}
\nonumber
\Pi_{\kappa\kappa'}&({\mathbf{q}},\omega=0) = -\frac{2}{N}\sum_{n,n',{\mathbf{k}}}
\frac{f_{n'}({\mathbf{k}}+{\mathbf{q}})
-f_{n}({\mathbf{k}})}{E_{n'}({\mathbf{k}}+{\mathbf{q}})-E_{n}({\mathbf{k}})}
\times \\\label{Eq17} &\times \left[C_{\kappa n}^{*}({\mathbf{k}})C_{\kappa
n'}({\mathbf{k}}+{\mathbf{q}}) \right] \left[C_{\kappa' n}^{*}({\mathbf{k}})C_{\kappa'
n'}({\mathbf{k}}+{\mathbf{q}}) \right]^{*}.\end{aligned}$$
$f$, $E$ and $C$ in Eq. are the occupation numbers, the single-particle energies and the expansion coefficients of the Bloch-functions in terms of tight-binding functions.
As a measure of the strength of the EPI the self-consistent change of an EDF on an ion, induced by a phonon mode $({\mathbf{q}},\,\sigma)$, can be derived in the form
$$\label{Eq18}
\delta\zeta_\kappa^{\mathbf{a}}({\mathbf{q}}\sigma) = \left[-\sum_\alpha
{\mathbf{X}}^{\kappa\alpha}({\mathbf{q}}){\mathbf{u}}_\alpha({\mathbf{q}}\sigma)\right]
e^{i{\mathbf{q}}{\mathbf{R}}_\kappa^{\mathbf{a}}}
\equiv \delta\zeta_\kappa({\mathbf{q}}\sigma)e^{i{\mathbf{q}}{\mathbf{R}}^{\mathbf{a}}},$$
with the displacement of the ions
$$\label{Eq19}
{\mathbf{u}}_\alpha^{{\mathbf{a}}}({\mathbf{q}}\sigma) =
\left(\frac{\hbar}{2M_\alpha\omega_\sigma({\mathbf{q}})}
\right)^{1/2}{\mathbf{e}}^\alpha({\mathbf{q}}\sigma)e^{i{\mathbf{q}}{\mathbf{R}}^{\mathbf{a}}}
\equiv {\mathbf{u}}_\alpha({\mathbf{q}}\sigma)e^{i{\mathbf{q}}{\mathbf{R}}^{\mathbf{a}}}.$$
The quantity ${\mathbf{X}}$ in Eq. , i.e. the self-consistent response per unit displacement of the EDF is calculated in linear response theory as
$$\label{Eq20}
{\mathbf{X}}({\mathbf{q}}) = \Pi({\mathbf{q}})\varepsilon^{-1}({\mathbf{q}}){\mathbf{B}}({\mathbf{q}}) =
C^{-1}({\mathbf{q}}){\mathbf{B}}({\mathbf{q}}).$$
Another measure of the EPI for a certain phonon mode $({\mathbf{q}}\,\sigma)$ is provided by the change of the self-consistent potential in the crystal felt by an electron at a space point ${\mathbf{r}}$, i.e. $\delta V_\text{eff}({\mathbf{r}},\,{\mathbf{q}}\,\sigma)$. Averaging this quantity with the corresponding density form factor $\rho_{\kappa}({\mathbf{r}} - {\mathbf{R}}^{{\mathbf{a}}}_{\kappa})$ of the EDF located at ${\mathbf{R}}^{{\mathbf{a}}}_{\kappa}$, we obtain
$$\label{Eq21}
\delta V_\kappa^{\mathbf{a}}({\mathbf{q}}\sigma) = \int dV
\rho_\kappa({\mathbf{r}}-{\mathbf{R}}_\kappa^{\mathbf{a}}) \delta
V_\text{eff}({\mathbf{r}},{\mathbf{q}}\sigma).$$
This gives an orbital resolved measure for the strength of the EPI in the mode $({\mathbf{q}}\,\sigma)$ mediated by the EDF considered. For an expression of $\delta V_{\kappa}^{{\mathbf{a}}}({\mathbf{q}}\,\sigma)$ in terms of coupling coefficients in Eqs. and , see Ref. [@Ref20].
Results and Discussion
======================
Results within the reference System (RIM)
-----------------------------------------
The rigid, local contribution of the electronic charge response and EPI is approximated by an ab initio RIM with corrections for covalence effects in terms of effective ionic charges (ion softening) and scaling of the short-ranged pair potential between the Bi and oxygen ion. In this way we obtain a suitable reference system, see Fig. \[fig01\] for the calculated dispersion, which subsequently allows for the investigation of the characteristic nonrigid screening effects in terms of CF’s and DF’s on the phonon dispersion and charge response.
The covalent corrections are important because a calculation of the phonon dispersion of cubic Ba-Bi-O with nominal charges leads to a significant overestimation of the width of the phonon spectrum. Moreover, various unstable modes down to -12 THz emerge if imaginary frequencies of unstable modes are represented as negative numbers. The mode with the highest frequency ($\approx$ 35 THz) in such a model is the oxygen-breathing mode, $O^{R}_{B}$, at the $R$ point which, on the other hand, according to recent experiments [@Ref09] is anomalously soft ($\approx$ 10.5 THz), see Fig. \[fig01\]. Similar results are found in calculations of the phonon dispersion of cuprate high-temperature superconductors when nominal ionic changes are used [@Ref01; @Ref21]. In our calculation we find a set of effective (static) ionic charges (Ba 1.5+, Bi 2.7+, O 1.4$\text{--}$) and a covalent scaling of the Bi-O pair potential which leads to an energy minimized structure in full agreement with the experimental one for cubic Ba-Bi-O ($a = 4.2742\,\AA$) [@Ref10]. As a general rule, partial covalence reduces the amplitude of the static ionic charges in mixed ionic-covalent compounds like Ba-Bi-O, because the charge transfer from the cations to the anions is not complete as in the entirely ionic case.
The phonon dispersion based on this set of reduced ionic charges is displayed in Fig. \[fig01\] (model 1). Potassium doping is taken into account using a virtual-mass approximation and implicitly via the reduced static charges. The phonon dispersion calculated with such a reference model means a large improvement compared to the model with nominal charges (not shown). In particular, the width of the spectrum is considerably reduced and compares well with the experiments if the missing renormalization related to the nonrigid charge response (CF’s and DF’s) is taken into account, see subsection \[subsecB\]. Moreover, unstable modes only appear at the $R$ and the $M$ point. The mode at $R$ is triply degenerate and represents rotations around any cubic axis. The rotation is opposite in neighbouring elementary cells in all cubic directions. The other unstable mode at $M$ is similar to the most unstable $R$ point mode, with the exception that the rotation of the octahedra is in the same sense in adjacent cells along the $z$-axis. Note in this context that, as mentioned in the introduction, the structural phase transitions in K-doped Ba-Bi-O are characterized by rotations (tilts) of the BiO$_{6}$ octahedra around several distinct axis [@Ref10] and in cubic Ba$_{0.6}$K$_{0.4}$BiO$_{3}$ the frequency of the rotational $R$ point mode is extremely low (Fig. \[fig01\]), consistent with the tendency of soft rotational modes at $R$ in harmonic theory. Thus, our results within the harmonic approximation highlight the importance of the ionic forces for the tilt instabilities and indicate that anharmonic contributions should stabilize these modes. Frozen-phonon calculations support this argument because they predict highly anharmonic potential behaviour for the tilting of the BiO$_{6}$ octahedra corresponding to the $R$-point [@Ref11; @Ref22].
In SrTiO$_{3}$ the rotational modes at $R$ also are found most unstable in our calculations [@Ref16] corresponding to an antiferrodistortive transition in agreement with the experiment while our calculation for BaTiO$_{3}$ find the ferroelectric mode at $\Gamma$ as the most unstable one again in accordance with the experimental situation. Interestingly, we also obtain in case of the HTSC La$_{2}$CuO$_{4}$ [@Ref18; @Ref20] one partially unstable branch with the tilt mode at the $X$-point. Freezing-in of this distortion points correctly to the experimentally observed structural phase transition from the high-temperature tetragonal to the low-temperature orthorhombic structure. Again this rotational instability is already present in the RIM and thus also driven by the long-ranged ionic interactions likewise as in Ba-Bi-O and Sr-Ti-O. In contrast to the situation in Ba-Ti-O the strongly polar ferroelectric TO mode at $\Gamma$, where the oxygen anions vibrate coherently in the opposite direction to the Bi ion, is stable in Ba-Bi-O. We get 8.6 THz in the RIM with effective ionic charges and 9.7 THz in the RIM with nominal ones. Not only the frequency of the ferroelectric mode is significantly reduced when passing from the model with nominal ionic charges to the model with effective charges but also the LO-TO mode splittings. Inspection of Fig. \[fig01\] with respect to the oxygen-breathing modes, displayed in Fig. \[fig02\], we find that instead of being soft as in the experiment the latter have the highest frequency at the $X\,(O^{X}_{B})$, $M\,(O^{M}_{B})$ and $R\,(O^{R}_{B})$ point, respectively. Consequently, the characteristic downward dispersion of the anomalous phonon-branches cannot be described at all within the RIM, i.e., by a rigid, local charge response. In the following we will show that the anomalous dispersion and phonon softening characteristic for cubic, metallic Ba-Bi-O is essentially related to strong nonlocal, nonrigid EPI effects mediated by CF’s of the Bi6s and Bi6p orbitals, pointing to a substantial character also of Bi6p electrons in the electronic state near the Fermi energy.
![Displacement patterns of the anomalous oxygen-breathing modes $O_{B}^{X}$, $O_{B}^{M}$ and $O_{B}^{R}$ from left to right for model 3. Frequencies are given in units of THz.[]{data-label="fig02"}](fig02.eps)
The effects of a nonrigid, nonlocal charge response via CF’s and DF’s {#subsecB}
---------------------------------------------------------------------
Starting from the reference systems (RIM) of the last subsection DF’s and CF’s as nonrigid electronic degrees of freedom are additionally allowed for the modeling. This means that on each ion the electrons can redistribute under atomic displacements in such a way that dipole and charge fluctuations are induced on that ion in order to minimize the energy.
In Fig. \[fig03\] we display our calculated results of the phonon dispersion taking additionally DF’s into account (model 2). In this calculation the ab initio values of the dipole polarizability $\alpha$ as calculated from the Sternheimer method in the framework of DFT-LDA-SIC [@Ref15] for the corresponding single ions are reduced by $50\,\%$ which leads to a better agreement of the phonon dispersion with the experiment. Taking the ab initio values for $\alpha$ as calculated for the single ions with the Sternheimer method overestimates the dipole polarization in the crystalline environment and leads to TO-frequencies being to low as compared to the experiment. These findings also hold true for our calculations of dipole screening in the cuprate based HTSC’s [@Ref05; @Ref23] where in addition the dipole polarizability is found to be very anisotropic dominating along the (ionic) $c$-direction.
Comparing Fig. \[fig03\] with Fig. \[fig01\] we find that the width of spectrum improves considerably and no additional unstable branches emerge. Moreover, the LO-TO splittings are further reduced and the TO-modes are well described. In particular, the ferroelectronic TO-mode at $\Gamma$ is reduced from 8.6 to 6.7 THz and is now in quite a good agreement with experiment [@Ref07; @Ref08]. However, the anomalous branches with the oxygen-breathing modes at $X$, $M$ and $R$ have an increasing dispersion when propagating from the $\Gamma$ point towards the zone boundary and show no sign of softening. Likewise as in the RIM, these modes have the highest frequencies in the spectrum.
Next, we discuss the influence of the ionic CF’s on the phonon dispersion. The results of a calculation where additionally to DF’s (model 2) CF’s are included are shown in Fig. \[fig04\] (model 3). Comparing the calculated results with the experiments now a good agreement is achieved. This is particularly true for the phonon anomalies which cannot be described well by other approaches up to now as stated in the introduction. First principle calculations [@Ref11] within DFT-LDA for Ba-Bi-O taking K-doping into account by a virtual-crystal and virtual-mass approximation report that the breathing distortions are seriously underestimated in the LDA. These calculations yield for the $O_{B}^{R}$ mode a frequency of 15.7 THz compared to 10.5 THz in the experiment [@Ref09]. We argue in the following that a reason for the high frequency in this calculation is the electronic polarizability $\Pi_{\kappa\kappa'}({\mathbf{q}}\,)$ from Eq. , i.e. the kinetic single particle part of the charge response, which seems underestimated in LDA-type calculations as in Ref. or Ref. , respectively.
In order to investigate this question in our approach, $\Pi_{\kappa\kappa'}({\mathbf{q}}\,)$ is modeled in an appropriate way to find out which electronic degrees of freedom are most important for the phonon anomalies in Ba-Bi-O. So, we express $\Pi_{\kappa\kappa'}({\mathbf{q}}\,)$ in Fourier-transformed (parametrized) form
$$\label{Eq22}
\Pi_{\kappa\kappa'}({\mathbf{q}}) = \sum_{{\mathbf{a}}} {\Pi_{}\!\setlength{\arraycolsep}{1pt}
\renewcommand{\arraystretch}{0.6}
\begin{array}{c}{\scriptstyle{\mathbf{0}}\,{\mathbf{a}}}\\
{\scriptstyle\kappa\,\kappa'}
\end{array}\!}
\,\text{e}^{i{\mathbf{q}}\left({\mathbf{R}}_{\kappa'}^{\mathbf{a}} - {\mathbf{R}}_\kappa \right)}.$$
In the metallic phase the electronic partial density of states (PDOS) at the Fermi level $Z_{\kappa}(\varepsilon_{F})$ is related to the polarizability matrix in the long wave length-limit [@Ref01; @Ref03] according to
$$\label{Eq23}
\sum_{\kappa'} \Pi_{\kappa\kappa'}({\mathbf{q}}\rightarrow{\mathbf{0}}) =
Z_\kappa(\varepsilon_\text{F}),$$
and the total density of states at energy $\varepsilon$ is given by
$$\label{Eq24}
Z(\varepsilon) = \sum_{\kappa} Z_\kappa(\varepsilon)$$
In our calculations for the metallic phase of the cuprate HTSC’s we have found [@Ref03; @Ref05; @Ref06] that neglecting the ${\mathbf{q}}$-dependence in the polarizability and taking into account the diagonal on-site parameters ${\Pi_{}\!\setlength{\arraycolsep}{1pt}
\renewcommand{\arraystretch}{0.6}
\begin{array}{c}{\scriptstyle{\mathbf{0}}\,{\mathbf{0}}}\\
{\scriptstyle\kappa\,\kappa}
\end{array}\!}$ alone already provides a good approximation for the calculation of the phonon dispersion, in particular for the anomalous generic OBSM in these materials. This means that the parameters ${\Pi_{}\!\setlength{\arraycolsep}{1pt}
\renewcommand{\arraystretch}{0.6}
\begin{array}{c}{\scriptstyle{\mathbf{0}}\,{\mathbf{0}}}\\
{\scriptstyle\kappa\,\kappa}
\end{array}\!}$ are approximated by the corresponding PDOS. However, such a diagonal approximation does not hold for the insulating state of a material. Due to the charge gap in the spectrum off-site elements ${\Pi_{}\!\setlength{\arraycolsep}{1pt}
\renewcommand{\arraystretch}{0.6}
\begin{array}{c}{\scriptstyle{\mathbf{0}}\,{\mathbf{a}}}\\
{\scriptstyle\kappa\,\kappa'}
\end{array}\!}$ necessarily must occur and interfere in such a way that the following compressibility sum rules for the kinetic single particle part of the charge response are satisfied [@Ref01; @Ref03; @Ref06].
$$\label{Eq25}
\sum_{\kappa'} \Pi_{\kappa\kappa'}({\mathbf{q}}\rightarrow{\mathbf{0}}) =
\mathcal{O}(q),$$
$$\label{Eq26}
\sum_{\kappa\kappa'} \Pi_{\kappa\kappa'}({\mathbf{q}}\rightarrow{\mathbf{0}}) =
\mathcal{O}(q^2).$$
In Fig. \[fig05\] we display the calculated effect of the Bi6s-, Bi6p- and O2p charge fluctuations for the softening of the frequency of the anomalous OBSM $O^{X}_{B}$, $O^{M}_{B}$ and $O^{R}_{B}$, respectively by varying the diagonal on-site parameters $\Pi\,\text{(Bi6s)} =
Z_{\text{Bi6s}}(\varepsilon_{F})$, $\Pi\,\text{(Bi6p)} =
Z_{\text{Bi6p}}(\varepsilon_{F})$ and $\Pi(\text{O2p}) =
Z_{\text{O2p}}(\varepsilon_{F})$. We find that the CF’s in the Bi6s- and Bi6p orbitals both are important for the phonon softening, while the dependence on the O2p CF’s is very weak. The values actually chosen for the calculation of the phonon dispersion in model 3 are $\Pi(\text{Bi6s}) =
1.00\,\text{eV}^{-1}$, $\Pi(\text{Bi6p}) = 0.5\,\text{eV}^{-1}$, $\Pi(\text{O2p}) = 0.2\,\text{eV}^{-1}$. The latter value, to a large extent irrelevant for the softening, was taken to optimize the phonon frequencies for modes where exclusively O2p CF’s are excited. The remarkable softening with increasing $Z_{\text{Bi6p}}(\varepsilon_{F})$ indicates that the band crossing the Fermi energy consists not only of the Bi6s and O2p electrons but also has a substantial admixture of Bi6p electrons. This is also concluded from electronic band structure calculations, see Ref. and references therein, and is consistent with the static effective charge for Bi in our model pointing out a partial occupation of the Bi6p orbital. The latter is the most delocalized one of the the orbitals used in our calculation (see Fig. \[fig06\]) and consequently very effective for screening and softening of the oxygen bond-stretching vibrations.
![Charge density form factors (times $4\,\pi\,r^{2}$) for the Bi$^{2.7+}$ and O$^{1.4-}$ ion. $r$ radial distance in units of $a_{B}$. Bi6s (full curve); Bi6p (dotted curve); O2p (broken curve).[]{data-label="fig06"}](fig06.eps)
Such an enhancement of softening, related to the delocalized components of the electronic state at the Fermi energy, also has been obtained in our calculations of the generic anomalous OBSM in the cuprate HTSC’s. In this case we have shown that with increased hole doping the inclusion of the delocalized Cu4s state into the orbital basis is an important factor for the strength of the anomalies in the optimally and overdoped state [@Ref02; @Ref06; @Ref20]. Thus, doping (similar like pressure) causes a progressive modification of the orbital character of the electronic state in the cuprate-HTSC’s. Quite generally, our investigations in the cuprates point to an interrelation between electronic structure changes at $\varepsilon_{F}$, selective phonon softening via strong nonlocal EPI of CF’ type and, finally, to the appearance of phonon-mediated superconductivity in an ”ionic” metal. A corresponding connection seems also to hold in doped Ba-Bi-O.
A change of the electronic state in Ba-Bi-O, i.e. a partial shift of the spectral weight from Bi6s to Bi6p, as suggested from our calculations of the OBSM, obviously occurs parallel with K-doping because the OBSM anomalies are not present in the weakly doped material [@Ref08]. It has been argued in Ref. that K-doping creates localized holes on the O2p orbitals, rather than on Bi atoms. One might speculate, that an increased population of the Bi6p states as the most delocalized ones lowers the energy by an enhanced hybridization with the O2p hole states making the latter itinerant at a certain doping level in the metallic state.
An increasing Bi6p character of the electronic state at $\varepsilon_{F}$ also should lead from a more rounded Fermi surface to a more distorted anisotropic shape. Such a change of the geometry of the Fermi surface would of course be mapped to the “phase space factor” (nesting function) and also to the matrix elements of the electron-phonon spectral function $\alpha^{2}F(\omega)$ and, finally, to the superconducting $T_{c}$ in a phonon-mediated mechanism.
Altogether from our results a multi-orbital model involving CF’s, DF’s and the corresponding interactions $\widetilde{V}$ in the charge response, Eqs. , , between Bi6s-, Bi6p- and O2p states is needed to obtain a reliable description of the phonon dispersion in an “ionic” metal like Ba-Bi-O, as shown in Fig. \[fig04\].
Comparing the results of the dispersion in Fig. \[fig03\] and \[fig04\] we recognize that the LO-TO splittings at $\Gamma$ are closed by the metallic screening due to the CF’s. Such a screening of the longwavelength polar modes does not imply charge localization on a microscopic scale and can also be achieved in a homogeneous electron gas model. However, localization of charge on a microscopic scale is necessary for the screening of the zone- boundary modes like $O^{X}_{B}$, $O^{M}_{B}$ and $O^{R}_{B}$. In this case nonlocally induced ionic Bi6s- and Bi6p CF’s lead to the strong softening. Compare with Fig. \[fig07\], where the nonlocal part of the phonon-induced charge density redistribution is shown for $O^{X}_{B}$, $O_{B}^{M}$ and $O^{R}_{B}$. The related strong softening of the modes by the nonlocally excited nonrigid charge response via CF’s and DF’s as compared to the RIM can be extracted from Fig. \[fig08\].
In context with the vanishing of the LO-TO splitting characteristic phonon branches along $\Delta$, $\Sigma$ and $\Lambda$ with the (stable) ferroelectric TO mode at $\Gamma$ (6.744 THz) appear in the phonon dispersion which are well described in our model. Even details, like the steep slopes near $\Gamma$, resulting from the depression of the LO frequency by metallic screening, and the flattening out of the branch in $\Lambda$ direction is obtained in our calculation.
![Calculated anomalous branches with the oxygen-breathing modes $O^{X}_{B}$, $O^{M}_{B}$ and $O^{R}_{B}$ as end points of the $\Delta \sim (0,\,0,\,1)$, $\Sigma \sim (1,\,1,\,0)$ and $\Lambda \sim (1,\,1,\,1)$ direction, respectively. The linetype ($\cdots$) denotes the result for the RIM (model 1), ($---$) additionally includes DF’s (model 2) and the full lines represent the final calculation including DF’s and CF’s (model 3).[]{data-label="fig08"}](fig08.eps)
In order to investigate the question how the phonon dispersion looks like, if we use an electronic density of states at $\varepsilon_{F}$ for Ba-Bi-O which is typical for a LDA-like calculation [@Ref24] in our model for the polarizability, we have performed the calculations shown in Fig. \[fig09\]. The total density of states from Ref. is 0.8 eV$^{-1}$. Taking for $\Pi(\text{O2p})$ the same value as in model 3 and using the same ratio for $\Pi(\text{Bi6s})$ and $\Pi(\text{Bi6p})$ we obtain: $\Pi(\text{Bi6s}) = 0.4\,\text{eV}^{-1}$, $\Pi(\text{Bi6p}) = 0.2\,\text{eV}^{-1}$, $\Pi(\text{O2p}) = 0.2\,\text{eV}^{-1}$ (model 4). Comparing the results for model 4 in Fig. \[fig09\] with those in Fig. \[fig04\] we find that the softening of the oxygen-breathing mode anomalies is only poorly developed, consistent with the ab initio calculations based on DFT-LDA in Ref. . This indicates that the ab initio LDA calculations underestimate the EPI for the OBSM.
$\delta\zeta_{\text{Bi6s}}$ $\delta\zeta_{\text{Bi6p}}$ $\delta\zeta_{\text{O2p}}$
------------- ----------------------------- ----------------------------- ----------------------------
$O_{B}^{X}$ 39.369 27.534 -4.169
32.333 18.285 -1.764
$O_{B}^{M}$ 61.329 42.740 -5.791
46.309 26.227 -2.092
$O_{B}^{R}$ 83.657 57.859 /
56.807 32.220 /
: Calculated phonon-induced charge fluctuation $\delta\zeta_{\kappa}({\mathbf{q}}\,\sigma)$ for model 3 (first row) and model 4 (second row) for $O_{B}^{X}$, $O_{B}^{M}$ and $O_{B}^{R}$ according to Eq. . Units are in $10^{-3}$ electrons. $\delta\zeta_{\kappa}({\mathbf{q}}\,\sigma) > 0$ means that electrons are pushed away. $/$ means, that no CF’s are excited.[]{data-label="tab01"}
The underestimation of the EPI in the breathing modes can also be deduced from Table \[tab01\] in terms of the CF’s. Here the CF’s according to Eq. of the Bi6s-, Bi6p- and O2p orbitals generated by $O^{X}_{B}$, $O_{B}^{M}$ and $O^{R}_{B}$ are listed. We recognize that in our modeling of the electronic state with enhanced PDOS for Bi6s and Bi6p at $\varepsilon_{F}$ as compared to the LDA-based result the CF’s are significantly larger leading to the correct description of the anomalies in the phonon dispersion.
The strong nonlocal EPI of ionic CF type in the oxygen-breathing modes is also illustrated in the contour plots of the nonlocal part of the phonon-induced charge density redistribution $\delta\rho({\mathbf{r}},\,{\mathbf{q}}\,\sigma)$ for the OBSM in Fig. \[fig07\], i.e.
$$\label{Eq27}
\delta\rho({\mathbf{r}},{\mathbf{q}}\sigma) = \sum_{{\mathbf{a}},\kappa}
\delta\zeta^{\mathbf{a}}_\kappa({\mathbf{q}}\sigma)\rho_\kappa({\mathbf{r}}-{\mathbf{R}}^{\mathbf{a}}_\kappa),$$
with the CF’s $\delta\zeta^{{\mathbf{a}}}_{\kappa}$ according to Eq. and the form factors $\rho_{\kappa}$ for the CF’s as in Eq. . The total phonon-induced charge redistribution can be obtained by adding the rigid displacement of the unperturbed densities, i.e.,
$$\label{Eq28}
\delta\rho_\text{r}({\mathbf{r}},{\mathbf{q}}\sigma) = \sum_{{\mathbf{a}},\alpha}
\{\rho^0_\alpha({\mathbf{r}}-[{\mathbf{R}}^{\mathbf{a}}_\alpha+{\mathbf{u}}^{\mathbf{a}}_\alpha({\mathbf{q}}\sigma)])
-\rho^0_\alpha({\mathbf{r}}-{\mathbf{R}}^{\mathbf{a}}_\alpha)\}.$$
$\rho_{\alpha}^{0}$ is the density of the unperturbed ion from Eq. and ${\mathbf{u}}^{{\mathbf{a}}}_{\alpha}$ the displacement of an ion in Eq. .
The results shown in Fig. \[fig07\] demonstrate that the nonlocal EPI leads to attractive (—–) and repulsive (—– $\cdot\cdot$) regions for electrons mainly at the Bi ions. For symmetry reasons some charge is also transferred to the silent oxygens in the $O^{X}_{B}$ and $O^{M}_{B}$ mode, see Fig. \[fig07\]a, \[fig07\]b and Table \[tab01\]. This, however, is not possible in $O^{R}_{B}$ because all the oxygen ions are vibrating symmetrically towards or away from Bi. Electrons, i.e. negative charge, is flowing to the expanded octahedra from the compressed ones. Thus, negative charge is always accumulated in the larger octahedra and so a (dynamical) charge disproportionation of the Bi ions is generated.
As mentioned in Ref. , if the intersite charge transfer in terms of CF’s induced nonlocally by the OBSM would additionally be spin polarized, as in the undoped antiferromagnetic cuprates or the weakly doped cuprates with antiferromagnetic correlations of sufficient correlation length, corresponding spin fluctuations are generated (spin-phonon interaction via charge transfer). In this way the phonon dynamics and the CF’s change the spin dynamics and vice versa. Such a spin-phonon coupling may be quite strong and could lead to a synergetic pairing mechanism in which both phonons and spin excitations participate, see e.g. Ref. and references therein. For a qualitative discussion of a possible interplay between lattice-, charge- and spin degrees of freedom in the cuprates with the help of the OBSM see also Ref. .
The CF’s of the Bi6s-, Bi6p- and O2p orbitals $\delta\zeta_{\kappa}({\mathbf{q}}\,\sigma)$ for $O^{X}_{B}$, $O^{M}_{B}$ and $O^{R}_{B}$ are collected in Table \[tab01\]. Significantly larger values are found for model 3 compared with the LDA-based model 4. For example in the $O^{R}_{B}$ mode 0.142 electrons are exchanged between the vibrating octahedra in model 3 and only 0.089 electrons in model 4.
For lightly K-doped insulating Ba-Bi-O besides the tilt distortion related to the rotational modes there is also a breathing distortion (frozen-in breathing mode) of the octahedra. On the other hand, our calculations in model 3 for the metallic phase show that strong nonlocal EPI of CF-type lead to a large softening but a still stable $O_{B}^{R}$ breathing mode in very good agreement with experiment. However, the result indicates that this type of nonlocal EPI also should play a role for the actual breathing distortion in the insulating part of the phase diagram. One can interpret the $\delta\rho$-patterns in Fig. \[fig07\] and the CF’s in Table \[tab01\] as an expression of a precursor effect for polaron formation. Actually, polarons are not expected to form at a high enough doping level, because, as mentioned before, the system can gain sufficient energy by delocalization via the hybridization of the Bi6s and in particular the Bi6p states with the O2p hole states created by doping. However, such a channel for a lowering of the energy by delocalization might not be favourable in the insulating, lightly doped material due to the small number of holes and the tilted structure where additionally the O2p to Bi6p hopping is decreased due to misalignment of the orbitals and hence the energy gain by delocalization is reduced. In this case (polaronic) charge localization by the strong nonlocal EPI of CF type might be energetically more favourable and responsible for the metal-insulator transition.
A tendency to polaron formation is also very likely in the undoped and weakly doped cuprate-HTSC’s, see e.g. Refs. . In this context, our modeling of the charge response has shown that the delocalized component of the electronic state, i.e., in this case Cu4s, is not required to describe the phonon anomalies in these “phases” of the cuprates [@Ref02; @Ref03; @Ref06; @Ref20]. Polaronic effects also occur in the small non-adiabatic region around the $c$-axis of the cuprat-HTSC’s where an insulator-like charge response and phonon-plasmon mixing is found [@Ref20; @Ref29].
Our calculations for La-Cu-O of the orbital resolved phonon induced change of the self consistent potential $\delta V_\kappa$ felt by an electron, according to Eq. , point to a hierarchy of energy scales for coupling of the electrons with the phonons[@Ref04; @Ref20]. There is one energy scale related to the OBSM with $\delta V_\text{Cu3d}$ up to about 100meV for the O$_\text{B}^X$ mode (with a frequency of 79meV) and up to about 75meV for the half breathing mode (with frequency of 69meV). Both modes have been calculated in adiabatic approximation which seems sufficient for the latter modes[@Ref20]. Moreover, we find in La-Cu-O due to weak screening perpendicular to the Cu-O plane in a small ${\mathbf{q}}$ space region around the $(0, 0, 1)$ direction in particular in the lightly doped materials a strong nonlocal, non-adiabatic electron-phonon coupling of CF type to the charge carriers in the Cu-O plane, which are responsible for superconductivity in the cuprates, via axial breathing modes of the apex oxygen and the La ion. The corresponding values of $\delta V_\kappa$ for these modes increase up to about 800meV near the phonon-plasmon resonance at low frequencies (phonon-like mode, possibly overdamped)[@Ref20]. Moreover, there is strong coupling at high frequencies beyond the phonon spectrum for the plasmon-like mode. Thus, a hierarchy of interaction scales at low energy as well as high energy is provided by these nonlocal electron phonon coupling effects caused by weak screening. Quite recently besides the well known low energy ”kink” scale (40-80meV), see e.g. Refs. , also a high energy anomaly in the electron spectrum of HTSC’s (300-800meV) has been observed by angle-resolved photoemission spectroscopy[@Ref32]. See also Refs. for the observation of the high energy anomalies and an anomalous enhancement of the width of the LDA-based CuO$_2$-band extending to energies of about 2eV.[@Ref34]
As a matter of fact, the electronic structure of the lightly K-doped insulating phase of Ba-Bi-O is not really understood. However, as discussed e.g. in Refs. and from our discussion formation of small polaron or bipolarons due to strong nonlocal EPI may be an important mechanism for insulating behaviour persisting over a wide doping range.
The mode splitting experimentally found in recent experiments for the anomalous branches along $\Sigma$ and $\Lambda$ for metallic Ba-Bi-O[@Ref09] (see the full symbols in the figures of the phonon dispersion) cannot be understood in calculations assuming an ideal cubic perovskite symmetry. As discussed in Ref. the split dispersion suggests some charge inhomogeneity due to distortions from the cubic perovskite structure on a time scale larger than the time scale of the phonons. For example a structural tilt distortion around the cubic (0, 0, 1) direction recently reported for superconducting Ba-Bi-O [@Ref37] should lead to an increase of the OBSM at the $M$- and $R$-point but not at the $X$-point, as seen in the experiments, because of the decreased O2p to Bi6p hopping due to misalignement of the orbitals and a corresponding reduction of the softening effect via Bi6p.
Anomalous phonon softening and mode splitting of the oxygen bond-stretching phonon along (1, 0, 0), most likely due to a (dynamic) charge inhomogeneity, very recently also has been observed in La-Ba-Cu-O [@Ref38]. As discussed in Ref. a charge inhomogeneity with lower symmetry as the ”average” translational invariant charge inhomogeneity considered in our modeling of the electronic state of the underdoped cuprate-HTSC’s[@Ref06; @Ref20] should lead locally to a variation of the strength of the nonlocal EPI of CF type and a related variation of the phonon frequency of the OBSM. The nonlocal EPI may become locally very strong in case of a metallic, compressible charge inhomogeneity.
In our modelling of the electronic state of the underdoped cuprates we have incompressible, insulator-like regions related to the localized Cu3d states interpenetrating a metallic, compressible structure provided by the more delocalized O2p states at the O$_{xy}$ sublattices, where the holes accumulate. Such a correlated percolated metallic structure can be thought essentially to be due to the interplay of the short range repulsion ($U_d$) of the Cu3d states and the long range Coulomb interaction together with a gain in kinetic energy of the holes by delocalization via the O2p states. For higher doping in the optimal and overdoped state the delocalized Cu4s component of the electronic state becomes important in our modelling leading to the experimentally observed strong increase of phonon softening of the OBSM in the cuprates, see e.g. Ref.. Physically this means that the percolated structure melts with increased doping because of the enhanced quantum mechanic tunneling via the delocalized Cu4s and O2p component in the electronic state at the the Fermi energy. Ultimately, in the overdoped ”phase” where in our modelling the Cu4s component is further enhanced the system should converge more and more towards a Fermi liquid.
The OBSM are a sensible probe for charge inhomogeneities in form of regions with a different doping level[@Ref06]. How the frequency would change depends in particular on the electronic state of the inhomogeneity which could be a mixture of compressible, metallic (higher doping), incompressible, insulating (no doping) or partially incompressible (lower doping) regions in the crystal. The incompressible and partially incompressible regions (with OBSMs of higher frequencies) most likely are stabilzed by an interplay of strong electronic correlation and strong nonlocal EPI of CF type triggering polaronic behaviour, while in the compressible regions (with lower frequencies of the OBSMs) delocalization of the electronic structure changes the nature of the electronic state as discussed in Refs. .
In a phonon-mediated pairing mechanism such a local variation of the nonlocal EPI by inhomogeneity is expected to induce simultaneously a corresponding change of the pairing strength (gap value) and a variation of the frequency of the pairing related modes. Usually, stronger renormalized modes (lower frequencies), i.e., a stronger nonlocal EPI, correlates to a larger pairing strength (higher gap value), and vice versa. Quite recently such a correlation has been observed locally in scanning tunneling experiments in Bi$_2$Sr$_2$CaCu$_2$O$_8$ [@Ref39] and as one possibility attributed to electronic-lattice interactions releated to a competing electronic ordered state (see, e.g. Ref. ).
In the last topic of this paper we present calculations for the phonon density of states within model 3 including CF’s and DF’s, see Fig. \[fig10\]. Moreover, in Fig. \[fig11\] the calculated atom-resolved phonon density of states is shown. Concerning the oxygen ion we have further discriminated between displacements parallel (O$_{\|}$) and perpendicular (O$_{\perp}$) with respect to the Bi-O bond in order to demonstrate the strongly anisotropic behaviour of the oxygen vibrations in Ba-Bi-O.
From the figures we see that the frequency range up to about 5 THz can be attributed to Ba and Bi vibrations with some contribution from O$_{\perp}$ including the unstable rotational modes. The O$_{\perp}$ vibrations dominate the spectral range from around 5 THz to about 11 THz. Beyond this frequency range the high-frequency part from 10-15 THz is exclusively governed by the O$_{\|}$ vibrations including the von Hove points of the OBSM: $O^{X}_{B} \approx 13.6\,\text{THz}$, $O^{M}_{B} \approx
11.9\,\text{THz}$ and $O^{R}_{B} \approx 10.2\,\text{THz}$. The anisotropy between the O$_{\|}$ and O$_{\perp}$ vibrations will for example be reflected in the Debye-Waller factors. The latter are expected to be considerably larger perpendicular to the Bi-O bond in agreement with experiments [@Ref10] and consistent with a tendency towards a structural phase transition via rotational modes. Finally, the strong almost singular peak at about 15 THz results from the characteristic nearly dispersionless branches in Fig. \[fig04\] terminating at the $\Gamma$ point in the TO mode with the highest frequency.
Summary and discussion
======================
Within our microscopic modeling of the electronic charge response in linear response theory we have studied complete phonon dispersion curves, phonon-induced charge density redistributions and charge fluctuations, generated by nonlocal EPI, and the total and atom-resolved phonon density of states in cubic metallic Ba-Bi-O. Our calculated results of the phonon dispersion curves are in a good overall agreement with the experiments, in particular the strong phonon softening of the anomalous oxygen-breathing modes is well described. This was not possible so far neither with empirical nor with ab initio calculations based on the LDA using a virtual crystal approximation. The latter seem to underestimate the kinetic single particle part of the density response. Increasing the electronic PDOS for the Bi6s and Bi6p orbitals at the Fermi level as compared with typical values obtained in LDA calculations has been shown to provide the strong softening of the OBSM. This is achieved by a characteristic type of local screening on microscopic scale in terms of ionic CF’s being quite generally at work in ”ionic” metals with a significant component of ionic binding like Ba-Bi-O or the the cuprate based HTSC’s. More explicitly, nonlocal coupling of the displaced oxygen ions in the OBSM essentially to Bi6s- and Bi6p CF’s, which are favoured by the corresponding large PDOS of these states at $\varepsilon_{F}$, are found to be responsible for the softening.
The effect of the strong nonlocal EPI is also visible from the calculation of the nonlocal part of the charge density redistribution and the CF’s themselves during the oxygen vibrations in the octahedra. Accordingly, a considerable (dynamical) charge disproportionation has been calculated. The patterns of charge rearrangement can be looked upon as a visualization of a precursor effect for polaron formation. Small polaron and/or bipolaron formation is very likely an important mechanism for the insulating behaviour in lightly K-doped Ba-Bi-O.
The strong effect for softening by including the Bi6p orbitals into the electronic state at the Fermi energy points towards a change of this state with K-doping. This change is accompanied by a corresponding variation of the topology of the Fermi surface and possible changes of the electron-phonon matrix elements. Both effects are important for the electron-phonon spectral function $\alpha^{2}F(\omega)$ and the $T_{c}$ value in a phonon-mediated mechanism for superconductivity. Such a mechanism is supported by the strong nonlocal EPI effects found in our calculations. Compared with the cuprate based HTSC spin degrees of freedom seem to play no role for superconductivity in Ba-Bi-O but the strong nonlocal mode-selective coupling effects of lattice and charge degrees of freedom in terms of ionic CF’s are present in both, Ba-Bi-O and the cuprate-HTSC’s.
From our calculations of the PDOS of the phonons we extract that the oxygen vibrations and consequently also the mean square displacements are very anisotropic. There are low-frequency O$_{\perp}$ vibrations perpendicular to the Bi-O bond which are well separated from the high-frequency O$_{\|}$ vibrations parallel to the bond.
Quite generally, our investigations for metallic Ba-Bi-O and the cuprate-HTSC’s point to an interrelation between doping induced electronic structure changes at the Fermi energy, selective phonon softening via strong nonlocal EPI of CF’ type and high-temperature superconductivity in “ionic” metals.
[10]{}
C. Falter, M. Klenner, and W. Ludwig Phys. Rev. B [**47**]{}, 5390 (1993).
C. Falter, M. Klenner, G.A. Hoffmann, and Q. Chen Phys. Rev. B [**55**]{}, 3308 (1997).
C. Falter and G.A. Hoffmann Phys. Rev. B [**61**]{}, 14537 (2000).
C. Falter and G.A. Hoffmann Phys. Rev. B [**64**]{}, 054516 (2001).
C. Falter and F. Schnetgöke Phys. Rev. B [**65**]{}, 054510 (2002).
C. Falter, Th. Bauer, and F. Schnetgöke Phys. Rev. B [**73**]{}, 224502 (2006).
M. Braden, W. Reichardt, W. Schmidbauer , A.S. Ivanov, and A.Y. Rumiantsev J. Supercond. [**8**]{}, 595 (1995).
M. Braden, W. Reichardt, A.S. Ivanov, and A.Y. Rumiantsev Europhys. Lett. [**34**]{}, 531 (1996).
M. Braden, W. Reichardt , S. Shiryaev, and S.N. Barilo Physica C [**378-381**]{}, 89 (2002).
S. Pei , J.D. Jorgensen , B. Dabrowski, D.G. Hinks, D.R. Richards, A.W. Mitchell, J.M. Newsam, S.K. Sinha, D. Vaknin, and A.J. Jacobson Phys. Rev. B [**41**]{}, 4126 (1990).
V. Meregalli and S.Y. Savrasov Phys. Rev. B [**57**]{}, 14453 (1998).
A.I. Liechtenstein, I.I. Mazin, C.O. Rodriguez, O. Jepsen, O.K. Andersen, and M. Methfessel Phys. Rev. B [**44**]{}, 5388 (1991).
M. Shirai, N. Suzuki, and K. Motizuki J. Phys.: Condens. Matter [**2**]{}, 3553 (1990).
C.M. Varma and W. Weber Phys. Rev. B [**19**]{}, 6142 (1979).
C. Falter, M. Klenner, G.A. Hoffmann, and F. Schnetgöke Phys. Rev. B [**60**]{}, 12051 (1999).
Th. Trautmann and C. Falter J. Phys.: Condens. Matter [**16**]{}, 5955 (2004).
P.B. Allen Phys. Rev. B [**16**]{}, 5139 (1977).
C. Falter, M. Klenner, and G.A. Hoffmann Phys. Rev. B [**52**]{}, 3702 (1995).
J.P. Perdew and A. Zunger Phys. Rev. B [**23**]{}, 5048 (1981).
C. Falter Phys. Stat. Sol. (b) [**242**]{}, 78 (2005).
R.E. Cohen, W.E. Pickett, L.L. Boyer, and H. Krakauer Phys. Rev. Lett. [**60**]{}, 817 (1988).
K. Kunc and R. Zeyher Phys. Rev. B [**49**]{}, 12216 (1994).
C. Falter and F. Schnetgöke J. Phys.: Condens. Matter [**15**]{}, 8495 (2003).
L.F. Mattheiss and D.R. Hamann Phys. Rev. Lett. [**60**]{}, 2681 (1988).
S. Salem-Sugui, E.E. Alp, S.M. Mini, M. Ramanathan, J.C. Campuzano, G. Jennings, M. Faiz, S. Pei, B. Dabrowski, Y. Zheng, D.R. Richards, and D.G. Hinks Phys. Rev. B [**43**]{}, 5511 (1991).
P. Piekarz and T. Egami Phys. Rev. B [**72**]{}, 054530 (2005).
O. Rösch, O. Gunnarsson, X. J. Zhou, T. Yoshida, T. Sasagawa, A. Fujimori, Z. Hussain, Z.-X. Shen, and S. Uchida Phys. Rev. Lett. [**95**]{}, 227002 (2005).
P. Prelovsěk, R. Zeyher, and P. Horsch Phys. Rev. Lett. [**96**]{}, 086402 (2006).
C. Falter, G.A. Hoffmann, and F. Schnetgöke J.Phys.: Condensed Matter [**14**]{}, 3239 (2002).
A. Lanzara, P. V. Bogdanov, X. J. Zhou, S. A. Kellar, D. L. Feng, E. D. Lu, T. Yoshida, H. Eisaki, A. Fujimori, K. Kishio, J.-I. Shimoyama, T. Noda, S. Uchida, Z. Hussain, and Z.-X. Shen Nature [**412**]{}, 510 (2001).
T. Cuk, F. Baumberger, D. H. Lu, N. Ingle, X. J. Zhou, H. Eisaki, N. Kaneko, Z. Hussain, T. P. Devereaux, N. Nagaosa, and Z.-X. Shen Phys. Rev. Lett. [**93**]{}, 117003 (2004).
J. Graf, G.-H. Gweon, K. McElroy, S. Y. Zhou, C. Jozwiak, E. Rotenberg, A. Bill, T. Sasagawa, H. Eisaki, S. Uchida, H. Takagi, D.-H. Lee, and A. Lanzara arXiv:cond-mat/0607319 (2006).
T. Valla, T. E. Kidd, Z.-H. Pan, A. V. Fedorov, W.-G. Yin, G.D. Gu, P.D. Johnson arXiv:cond-mat/0610249 (2006).
W. Meevasana, X.J. Zhou, S. Sahrakorpi, W.S. Lee, W.L. Yang, K. Tanaka, N. Mannella, T. Yoshida, D. H. Lu, Y.L. Chen, R.H. He, Hsin Lin, S. Komiya, Y. Ando, F. Zhou, W.X. Ti, J.W. Xiong, Z. X. Zhao, T. Sasagawa, T. Kakeshita, K. Fujita, S. Uchida, H. Eisaki, A. Fujimori, Z. Hussain, R. S. Markiewicz, A. Bansil, N. Nagaosa, J. Zaanen, T.P. Devereaux, Z.-X. Shen arXiv:cond-mat/0612541 (2006).
I.B. Bischofs, V.N. Kostur, and P.B. Allen Phys. Rev. B [**65**]{}, 115112 (2002).
T. Nishio, J. Ahmad, and H. Uwe Phys. Rev. Lett. [**95**]{}, 176403 (2005).
M. Braden, W. Reichardt, E. Elkaim, J.P. Lauriat, S. Shiryaev, and S.N. Barilo Phys. Rev. B [**62**]{}, 6708 (2000).
D. Reznik, L. Pintschovius, M. Ito, S. Iikubo, M. Sato, H. Goka, M. Fujita, K. Yamada, G.D. Gu, and J.M. Tranquada Nature [**440**]{}, 1170 (2006).
J. Lee, K. Fujita, K. McElroy, J.A. Slezak, M. Wang, Y. Aiura, H. Bando, M. Ishikado, T. Masui, J.X. Zhu, A.V. Balatsky, H. Eisaki, S. Uchida, and J.C. Davis Nature [**442**]{}, 546 (2006).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Contextual word embeddings such as BERT have achieved state of the art performance in numerous NLP tasks. Since they are optimized to capture the statistical properties of training data, they tend to pick up on and amplify social stereotypes present in the data as well. In this study, we (1) propose a template-based method to quantify bias in BERT; (2) show that this method obtains more consistent results in capturing social biases than the traditional cosine based method; and (3) conduct a case study, evaluating gender bias in a downstream task of Gender Pronoun Resolution. Although our case study focuses on gender bias, the proposed technique is generalizable to unveiling other biases, including in multiclass settings, such as racial and religious biases.'
author:
- |
Keita Kurita Nidhi Vyas Ayush Pareek Alan W Black Yulia Tsvetkov\
Carnegie Mellon University\
[{kkurita,nkvyas,apareek,awb,ytsvetko}@andrew.cmu.edu]{}
bibliography:
- 'acl2019.bib'
title: Measuring Bias in Contextualized Word Representations
---
Introduction
============
Type-level word embedding models, including word2vec and GloVe [@mikolov2013distributed; @pennington2014glove], have been shown to exhibit social biases present in human-generated training data [@bolukbasi2016man; @caliskan2017semantics; @garg2018word; @manzini19multiclass]. These embeddings are then used in a plethora of downstream applications, which perpetuate and further amplify stereotypes [@zhao2017men; @leino2018feature]. To reveal and quantify corpus-level biases is word embeddings, @bolukbasi2016man used the word analogy task [@mikolov2013distributed]. For example, they showed that gendered male word embeddings like *he, man* are associated with higher-status jobs like *computer programmer* and *doctor*, whereas gendered words like *she* or *woman* are associated with *homemaker* and *nurse*.
Contextual word embedding models, such as ELMo and BERT [@peters2018deep; @devlin2018bert] have become increasingly common, replacing traditional type-level embeddings and attaining new state of the art results in the majority of NLP tasks. In these models, every word has a different embedding, depending on the context and the language model state; in these settings, the analogy task used to reveal biases in uncontextualized embeddings is not applicable. Recently, @may2019measuring showed that traditional cosine-based methods for exposing bias in sentence embeddings fail to produce consistent results for embeddings generated using contextual methods. We find similar inconsistent results with cosine-based methods of exposing bias; this is a motivation to the development of a novel bias test that we propose. In this work, we propose a new method to quantify bias in BERT embeddings ([§\[sec:method\]]{}). Since BERT embeddings use a *masked* language modelling objective, we directly query the model to measure the bias for a particular token. More specifically, we create simple template sentences containing the attribute word for which we want to measure bias (e.g. *programmer*) and the target for bias (e.g. *she* for gender). We then mask the attribute and target tokens sequentially, to get a relative measure of bias across target classes (e.g. male and female). Contextualized word embeddings for a given token change based on its context, so such an approach allows us measure the bias for similar categories divergent by the the target attribute ([§\[sec:method\]]{}). We compare our approach with the cosine similarity-based approach ([§\[sec:corr\_human\_biases\]]{}) and show that our measure of bias is more consistent with human biases and is sensitive to a wide range of biases in the model using various stimuli presented in @caliskan2017semantics. Next, we investigate the effect of a specific type of bias in a specific downstream task: gender bias in BERT and its effect on the task of Gendered Pronoun Resolution (GPR) [@webster2018gap]. We show that the bias in GPR is highly correlated with our measure of bias ([§\[sec:case\_study\]]{}). Finally, we highlight the potential negative impacts of using BERT in downstream real world applications ([§\[sec:implications\]]{}). The code and data used in this work are publicly available.[^1]
Quantifying Bias in BERT {#sec:method}
========================
BERT is trained using a masked language modelling objective i.e. to predict masked tokens, denoted as \[MASK\], in a sentence given the entire context. We use the predictions for these \[MASK\] tokens to measure the bias encoded in the actual representations.
We directly query the underlying masked language model in BERT[^2] to compute the association between certain **targets** (e.g., gendered words) and **attributes** (e.g. career-related words). For example, to compute the association between the target *male gender* and the attribute *programmer*, we feed in the masked sentence “\[MASK\] is a programmer” to BERT, and compute the probability assigned to the sentence ‘*he* is a programmer’’ ($p_{tgt}$). To measure the association, however, we need to measure how much *more* BERT prefers the male gender association with the attribute *programmer*, compared to the female gender. We thus re-weight this likelihood $p_{tgt}$ using the prior bias of the model towards predicting the male gender. To do this, we mask out the attribute *programmer* and query BERT with the sentence “\[MASK\] is a \[MASK\]”, then compute the probability BERT assigns to the sentence ‘*he* is a \[MASK\]’’ ($p_{prior}$). Intuitively, $ p_{prior} $ represents how likely the word *he* is in BERT, given the sentence structure and no other evidence. Finally, the difference between the normalized predictions for the words *he* and *she* can be used to measure the gender bias in BERT for the *programmer* attribute.
Generalizing, we use the following procedure to compute the association between a target and an attribute:
1. Prepare a template sentence\
e.g.“\[TARGET\] is a \[ATTRIBUTE\]"
2. Replace \[TARGET\] with \[MASK\] and compute $p_{tgt}$=P(\[MASK\]=\[TARGET\]$|$ sentence)
3. Replace both \[TARGET\] and \[ATTRIBUTE\] with \[MASK\], and compute prior probability $p_{prior}$=P(\[MASK\]=\[TARGET\]$|$ sentence)
4. Compute the association as $\log{\frac{p_{tgt}}{p_{prior}}}$
We refer to this normalized measure of association as the *increased log probability* score and the difference between the increased log probability scores for two targets (e.g. he/she) as *log probability bias score* which we use as measure of bias. Although this approach requires one to construct a template sentence, these templates are merely simple sentences containing attribute words of interest, and can be shared across multiple targets and attributes. Further, the flexibility to use such templates can potentially help measure more fine-grained notions of bias in the model.
In the next section, we show that our proposed *log probability bias score* method is more effective at exposing bias than traditional cosine-based measures.
Category Templates
--------------------------------------- -------------------------------------------
Pleasant/Unpleasant (Insects/Flowers) T are A, T is A
Pleasant/Unpleasant (EA/AA) T are A, T is A
Career/Family (Male/Female) T likes A, T like A, T is interested in A
Math/Arts (Male/Female) T likes A, T like A, T is interested in A
Science/Arts (Male/Female) T likes A, T like A, T is interested in A
Category Targets Templates
--------------------------------------- ------------------------------- -------------------------------------------
Pleasant/Unpleasant (Insects/Flowers) flowers,insects,flower,insect T are A, the T is A
Pleasant/Unpleasant (EA/AA) black, white T people are A, the T person is A
Career/Family (Male/Female) he,she,boys,girls,men,women T likes A, T like A, T is interested in A
Math/Arts (Male/Female) he,she,boys,girls,men,women T likes A, T like A, T is interested in A
Science/Arts (Male/Female) he,she,boys,girls,men,women T likes A, T like A, T is interested in A
--------------------------------------- --------------- -------------- ------------------------------
Category WEAT on GloVe WEAT on BERT Ours on BERT
*Log Probability Bias Scor*e
Pleasant/Unpleasant (Insects/Flowers) 1.543\* 0.6688 0.8744\*
Pleasant/Unpleasant (EA/AA) 1.012 1.003 0.8864\*
Career/Family (Male/Female) 1.814\* 0.5047 1.126\*
Math/Arts (Male/Female) 1.061 0.6755 0.8495\*
Science/Arts (Male/Female) 1.246\* 0.8815 0.9572\*
--------------------------------------- --------------- -------------- ------------------------------
Correlation with Human Biases {#sec:corr_human_biases}
=============================
We investigate the correlation between our measure of bias and human biases. To do this, we apply the log probability bias score to the same set of attributes that were shown to exhibit human bias in experiments that were performed using the Implicit Association Test [@iat]. Specifically, we use the stimuli used in the Word Embedding Association Test (WEAT) [@caliskan2017semantics].
**Word Embedding Association Test (WEAT)**: The WEAT method compares set of target concepts (e.g. male and female words) denoted as $X$ and $Y$ (each of equal size $N$), with a set of attributes to measure bias over social attributes and roles (e.g. career/family words) denoted as $A$ and $B$. The degree of bias for each target concept $t$ is calculated as follows: $$\begin{aligned}
\small
s(t, A, B) = [\textrm{mean}_{a \in A}{\textrm{sim}(t, a)} - \nonumber
\textrm{mean}_{b \in B}{\textrm{sim}(t, b)}], \end{aligned}$$ where *sim* is the cosine similarity between the embeddings. The test statistics is $$\begin{aligned}
\small
S(X, Y, A, B) = [\textrm{mean}_{x \in X}s(x, A, B) - \\ \small
\textrm{mean}_{y \in Y}s(y, A, B)], \end{aligned}$$ where the test is a permutation test over $X$ and $Y$. The $p$-value is computed as $$\begin{aligned}
\small
p = \Pr[S(X_i, Y_i, A, B) > S(X, Y, A, B)] \end{aligned}$$ The effect size is measured as $$d = \frac{S(X, Y, A, B)}{\textrm{std}_{t \in X \cup Y} s(t, A, B) }$$ It is important to note that the statistical test is a permutation test, and hence a large effect size does not guarantee a higher degree of statistical significance.
Baseline: WEAT for BERT
-----------------------
To apply the WEAT method on BERT, we first compute the embeddings for target and attribute words present in the stimuli using multiple templates, such as “TARGET is ATTRIBUTE" (Refer [Table \[weat-templates1\]]{} for an exhaustive list of templates used for each category). We mask the TARGET to compute the embedding[^3] for the ATTRIBUTE and vice versa. Words that are absent in the BERT vocabulary are removed from the targets. We ensure that the number of words for both targets are equal, by removing random words from the smaller target set. To confirm whether the reduction in vocabulary results in a change of $p$-value, we also conduct the WEAT on GloVe with the reduced vocabulary.[^4]
Proposed: Log Probability Bias Score
------------------------------------
To compare our method measuring bias, and to test for human-like biases in BERT, we also compute the *log probability bias score* for the same set of attributes and targets in the stimuli. We compute the mean *log probability bias score* for each attribute, and permute the attributes to measure statistical significance with the permutation test. Since many TARGETs in the stimuli cause the template sentence to become grammatically incorrect, resulting in low predicted probabilities, we fixed the TARGET to common pronouns/indicators of category such as *flower, he, she* ([Table \[weat-templates2\]]{} contains a full list of target words and templates). This avoids large variance in predicted probabilities, leading to more reliable results. The effect size is computed in the same way as the WEAT except the standard deviation is computed over the mean *log probability bias scores*.
We experiment over the following categories of stimuli in the WEAT experiments: Category 1 (flower/insect targets and pleasant/unpleasant attributes), Category 3 (European American/African American names and pleasant/unpleasant attributes), Category 6 (male/female names and career/family attributes), Category 7 (male/female targets and math/arts attributes) and Category 8 (male/female targets and science/arts attributes).
Comparison Results
------------------
The WEAT on GloVe returns similar findings to those of @caliskan2017semantics except for the European/African American names and pleasant/unpleasant association not exhibiting significant bias. This is due to only 5 of the African American names being present in the BERT vocabulary. The WEAT for BERT fails to find any statistically significant biases at $p<0.01$. This implies that WEAT is not an effective measure for bias in BERT embeddings, or that methods for constructing embeddings require additional investigation. In contrast, our method of querying the underlying language model exposes statistically significant association across all categories, showing that BERT does indeed encode biases and that our method is more sensitive to them.
Case Study: Effects of Gender Bias on Gendered Pronoun Resolution {#sec:case_study}
=================================================================
#### Dataset
We examined the downstream effects of bias in BERT using the Gendered Pronoun Resolution (GPR) task [@webster2018gap]. GPR is a sub-task in co-reference resolution, where a pronoun-containing expression is to be paired with the referring expression. Since pronoun resolving systems generally favor the male entities [@webster2018gap], this task is a valid test-bed for our study. We use the GAP dataset[^5] by @webster2018gap, containing 8,908 human-labeled ambiguous pronoun-name pairs, created from Wikipedia. The task is to classify whether an ambiguous pronoun $P$ in a text refers to entity $A$, entity $B$ or neither. There are 1,000 male and female pronouns in the training set each, with 103 and 98 of them not referring to any entity in the sentence, respectively.
#### Model
We use the model suggested on Kaggle,[^6] inspired by @47786. The model uses BERT embeddings for $P$, $A$ and $B$, given the context of the input sentence. Next, it uses a multi-layer perceptron (MLP) layer to perform a naive classification to decide if the pronoun belongs to $A$, $B$ or neither. The MLP layer uses a single hidden layer with 31 dimensions, a dropout of 0.6 and L2 regularization with weight 0.1.
Gender Prior Prob. Avg. Predicted Prob.
-------- ------------- ----------------------
Male 10.3% 11.5%
Female 9.8% 13.9%
: \[tab:gpr-probs\] Probability of pronoun referring to neither entity in a sentence of GPR
#### Results
Although the number of male pronouns associated with no entities in the training data is slightly larger, the model predicted the female pronoun referring to no entities with a significantly higher probability ($p=0.007$ on a permutation test); see [Table \[tab:gpr-probs\]]{}. As the training set is balanced, we attribute this bias to the underlying BERT representations.
We also investigate the relation between the topic of the sentence and model’s ability to associate the female pronoun with no entity. We first extracted 20 major topics from the dataset using non-negative matrix factorization [@nmf] (refer to Appendix for the list of topics). We then compute the bias score for each topic as the sum of the *log probability bias* score for the top 15 most prevalent words of each topic weighted by their weights within the topic. For this, we use a generic template “\[TARGET\] are interested in \[ATTRIBUTE\]" where TARGET is either men or women. Next we compute a bias score for each sample in the training data as the sum of individual bias scores of topics present in the sample, weighted by the topic weights. Finally, we measured the Spearman correlation coefficient to be 0.207 (which is statistically significant with $p=4e-11$) between the bias scores for male gender across all samples and the model’s probability to associate a female pronoun with no entity. We conclude that models using BERT find it challenging to perform coreference resolution when the gender pronoun is female and if the topic is biased towards the male gender.
Real World Implications {#sec:implications}
=======================
In previous sections, we discussed that BERT has human-like biases, which are propagated to downstream tasks. In this section, we discuss another potential negative impact of using BERT in a downstream model. Given that three quarters of US employers now use social media for recruiting job candidates [@segal2014social], many applications are filtered using job recommendation systems and other AI-powered services. @zhao2018learning discussed that resume filtering systems are biased when the model has strong association between gender and certain professions. Similarly, certain gender-stereotyped attributes have been strongly associated with occupational salary and prestige [@glick1991trait]. Using our proposed method, we investigate the gender bias in BERT embeddingss for certain occupation and skill attributes.
**Datasets**: We use three datasets for our study of gender bias in employment attributes:
- *Employee Salary Dataset[^7] for Montgomery County of Maryland-* Contains 6882 instances of “Job Title" and “Salary" records along with other attributes. We sort this dataset in decreasing order of salary and take the first 1000 instances as a proxy for high-paying and prestigious jobs.
- *Positive and Negative Traits Dataset[^8]-* Contains a collection of 234 and 292 adjectives considered “positive" and “negative" traits, respectively.
- *O\*NET 23.2 technology skills[^9]* Contains 17649 unique skills for 27660 jobs, which are posted online
**Discussion** We used the following two templates to measure gender bias:
- “TARGET is ATTRIBUTE", where TARGET are male and female pronouns viz. *he* and *she*. The ATTRIBUTE are job titles from the Employee Salary dataset, or the adjectives from the Positive and Negative traits dataset.
- “TARGET can do ATTRIBUTE", where the TARGETs are the same, but the ATTRIBUTE are skills from the O\*NET dataset.
Table \[tab:employment\] shows the percentage of attributes that were more strongly associated with the male than the female gender. The results prove that BERT expresses strong preferences for male pronouns, raising concerns with using BERT in downstream tasks like resume filtering.
Dataset Percentage
------------ ------------
Salary 88.5%
Pos-Traits 80.0%
Neg-Traits 78.9%
Skills 84.0%
: \[tab:employment\] Percentage of attributes associated more strongly with the male gender
Related Work {#sec:related_work}
============
NLP applications ranging from core tasks such as coreference resolution [@rudinger2018gender] and language identification [@jurgens2017incorporating], to downstream systems such as automated essay scoring [@amorim2018automated], exhibit inherent social biases which are attributed to the datasets used to train the embeddings [@barocas2016big; @zhao2017men; @yao2017beyond]. There have been several efforts to investigate the amount of intrinsic bias within uncontextualized word embeddings in binary [@bolukbasi2016man; @garg2018word; @swinger19bias] and multiclass [@manzini19multiclass] settings. Contextualized embeddings such as BERT [@devlin2018bert] and ELMo [@peters2018deep] have been replacing the traditional type-level embeddings. It is thus important to understand the effects of biases learned by these embedding models on downstream tasks. However, it is not straightforward to use the existing bias-exposure methods for contextualized embeddings. For instance, @may2019measuring used WEAT on sentence embeddings of ELMo and BERT, but there was no clear indication of bias. Rather, they observed counterintuitive behavior like vastly different $p$-values for results concerning gender.
Along similar lines, @basta2019evaluating noted that contextual word-embeddings are less biased than traditional word-embeddings. Yet, biases like gender are propagated heavily in downstream tasks. For instance, @ZWYCOC19 showed that ELMo exhibits gender bias for certain professions. As a result, female entities are predicted less accurately than male entities for certain occupation words, in the coreference resolution task. @field19personabias revealed biases in ELMo embeddings that limit their applicability across data domains. Motivated by these recent findings, our work proposes a new method to expose and measure bias in contextualized word embeddings, specifically BERT. As opposed to previous work, our measure of bias is more consistent with human biases. We also study the effect of this intrinsic bias on downstream tasks, and highlight the negative impacts of gender-bias in real world applications.
Conclusion
==========
In this paper, we showed that querying the underlying language model can effectively measure bias in BERT and expose multiple stereotypes embedded in the model. We also showed that our measure of bias is more consistent with human-biases, and outperforms the traditional WEAT method on BERT. Finally we showed that these biases can have negative downstream effects. In the future, we would like to explore the effects on other downstream tasks such as text classification, and device an effective method of debiasing contextualized word embeddings.
Acknowledgments {#acknowledgments .unnumbered}
===============
This material is based upon work supported by the National Science Foundation under Grant No. IIS1812327.
Appendix {#appendix .unnumbered}
========
\[hbt\]
Topic Id Top 5 Words
---------- -------------------------------------------
1 match,round,second,team,season
2 times,city,jersey,york,new
3 married,son,died,wife,daughter
4 best,award,actress,films,film
5 friend,like,work,mother,life
6 university,music,attended,high,school
7 president,general,governor,party,state
8 songs,solo,song,band,album
9 medal,gold,final,won,world
10 best,role,character,television,series
11 kruse,moved,amy,esme,time
12 usa,trunchbull,pageant,2011,miss
13 american,august,brother,actress,born
14 sir,died,church,song,john
15 natasha,days,hospital,helene,later
16 played,debut,sang,role,opera
17 january,december,october,july,married
18 academy,member,american,university,family
19 award,best,played,mary,year
20 jersey,death,james,king,paul
: \[gpr-topics\] Extracted topics for the GPR dataset
[^1]: <https://bit.ly/2EkJwh1>
[^2]: For all experiments we use the uncased version of BERT~BASE~ <https://storage.googleapis.com/bert_models/2018_10_18/uncased_L-12_H-768_A-12.zip>.
[^3]: We use the outputs from the final layer of BERT as embeddings
[^4]: WEAT was originally used to study the GloVe embeddings
[^5]: <https://github.com/google-research-datasets/gap-coreference>
[^6]: <https://www.kaggle.com/mateiionita/taming-the-bert-a-baseline>
[^7]: <https://catalog.data.gov/dataset/employee-salaries-2017>
[^8]: <http://ideonomy.mit.edu/essays/traits.html>
[^9]: <https://www.onetcenter.org/database.html#individual-files>
| {
"pile_set_name": "ArXiv"
} |
---
author:
- Jarosław Kijak
- Janusz Gil
date: 'Received 19 November 2001 / Accepted 30 May 2002'
title: 'Structure of pulsar beams: conal versus patchy'
---
Introduction
============
One of the important questions in pulsar research is what the overall structure of the mean pulsar beam is and how this structure is related to highly fluctuating instantaneous pulsar radiation. It is difficult to reveal this structure as pulsar observations represent one-dimensional cuts through two-dimensional beams. However, some indirect methods have been applied in an attempt to resolve this problem and two major models of pulsar beams have emerged from this work. @r93, @gks93 and @kwj94 calculated the opening angles $\rho$ of emission corresponding to a pulse width $W$ measured at 10 and 50 percent of the maximum intensity. As a result, they obtained a binomial distribution of these angles, that is, for a given period $P$ one of the two preferred values was possible, following however a general $P^{-1/2}$ dependence. Such distribution is most naturally interpreted as an indication of two nested cones in the structure of mean pulsar beams. This interpretation is called a conal model of pulsar beams. An alternative model postulates that the mean pulsar beam is patchy [@lm88 LM88 hreafter], with different components randomly distributed within an almost circular “window function” [@m95; @hm01]. Such a model is apparently inconsistent with the binomial distribution of the opening angles inferred from measured pulse widths. In fact, unless putative patches are distributed along nested-circular patterns, the distribution of corresponding opening angles should be (for any given period) random rather than binomial.
Recently, @md99 [MD99 hereafter] attempted to test both these rival models. They distributed locations of the profile components (measured as the peak-to-peak separation of the outer conal components in complex profile pulsars) on one quadrant of the beam represented on the common normalised scale (with $x$-axis and $y$-axis representing longitudes $\varphi$ and impact angles $\beta$, respectively; refer to Fig. 4 of MD99). They found that most of the peak intensity points are concentrated in narrow sections of the beam. This feature is a strong indication of the conal structure of the pulsar beam. It can be argued that if, indeed, the pulsar beams are patchy, then in such a case there is a high probability for the beam to be uniformly filled with peak intensity locations. Further in their analysis, they excluded the so-called conal single and conal double profiles [@r83], which are thought to be exclusively grazing cuts of the line-of-sight at the beam boundary. They found that such exclusion in their sample led to the absence of points at high impact angles $\beta$, which is perfectly consistent with the nested cone model and inconsistent with the patchy beam model. In fact, within the patchy beam model there is no reason why the single and double profiles occur exclusively at high impact angles. Moreover, the midpoint of single and double profiles usually coincides with the fiducial phase, at which the multifrequency profiles align after being corrected for cold plasma dispersive delays. This property is natural within the conal model and inconsistent (in general) with the patchy beam model, since patches would have to be placed symmetrically with respect to the fiducial plane, containing both magnetic ${\bf m}$ and spin ${\bf\Omega}$ axes (Fig. \[fig1\]). As argued by MD and independently by @gs00 [GS00 hereafter] the “mean” average pulsar beam consists of up to three nested cones, centered on the global magnetic dipole axis.
{height="18cm"}
Observations of single pulses in strong pulsars show that longitudes of subpulses are weakly dependent on frequency as compared with longitudes of corresponding profile components [@i93; @gggk02]. Within the conal model, the longitudes of profile components are determined by the intersection of the line of sight trajectory with the frequency-dependent cones of the maximum average intensity, while the longitudes of subpulses are determined by the intersection of the line-of-sight trajectory with subpulse-associated emission beams, which move across the average cones as frequency changes [@gk96]. We demonstrate in this paper that the different frequency dependence of subpulse and profile component longitudes is a natural property of the conal model, and that both subpulses and profile components should demonstrate the same frequency dependence of their longitudes within the patchy model. We present both general qualitative arguments and detailed quantitative model calculations to support the above statements. For better understanding of our arguments, this paper should be studied along with the paper by @gggk02 [GGGK hereafter], in which details of frequency dependence of emission patterns in PSR B0329$+$54 are discussed within the conal model of pulsar beams.
Geometry of pulsar radio beams
==============================
It is widely believed that narrow-band pulsar emission is relativistically beamed tangentially to dipolar magnetic field lines. Thus, the emission beaming geometry can be described by an opening angle $\rho=1^\circ.24~s\ r_6^{1/2}P^{-1/2}$, where $r_6=r(\nu)/R$ is the normalized emission altitude (in units of stellar radius $R=10^6$ cm). The mapping parameter $0\leq s=d/r_p\leq 1$ is determined by the locus of dipolar field lines on the polar cap ($s=0$ at the pole and $s=1$ at the polar cap edge), where $d$ is the distance from the magnetic axis $\bf{m}$ to the field line on the polar cap corresponding to a certain detail of the pulse profile (peak of subpulse or profile component), and $r_p=1.4\cdot 10^4P^{-1/2}$ cm is the canonical polar cap radius. According to the generally - accepted concept of the radius-to-frequency mapping, higher frequencies are emitted at lower altitudes $r(\nu)$ than lower frequencies. @kg97 [@kg98] found a semi-empirical formula describing the altitude of emission region corresponding to a given frequency $\nu_{\rm GHz}$ (in GHz) which reads $r_6\approx
50\cdot\nu_{\rm GHz}^{-0.21}\cdot\tau_6^{-0.1}\cdot P^{0.33}$, where $\tau_6$ is the pulsar characteristic age in million years and $P$ is the pulsar period. This formula for emission altitudes is used in our model calculations.
To perform geometrical calculations of the radiation pattern one has to adopt a model of instantaneous energy distribution on the polar cap. Any specific intensity distribution can be transferred from the polar cap along dipolar field lines to the emission region, and then along straight lines (following the opening angles $\rho=1^\circ.24(d/r_p)r_6^{1/2}P^{-1/2}$) to a given observer specified by the inclination and impact angles ($\alpha,~\beta$). We assume that at any instant the polar cap is populated by a number of features with a characteristic dimension ${\cal D}$, delivering to the magnetosphere corresponding plasma columns flowing along separate bundles of dipolar magnetic field lines. Each feature can be modelled by the Gaussian intensity distribution. Since the elementary pulsar radiation is relativistically beamed along the magnetic field lines, we can transform the feature-associated intensity pattern to the radio emission region and obtain the subpulse intensity $I$ observed at the longitude $\varphi$ in the form $I_i=\exp(-\kappa l_i^2(\varphi)/{\cal D}^2)$ where the subscript $i$ lebels different features and $l_i^2(\varphi)=d^2(\varphi)+d_i^2-2d(\varphi)d_i\cdot
\cos[\sigma(\varphi)-\sigma_{in}]$. Each features is located at the position (Fig. \[fig1\]) described by the polar co-ordinates $d_i$ (distance from the pole) and $\sigma_{in}=\sigma_{io}+nD_r$ (magnetic azimuth angle), where the subscript $o$ refers to the initial position corresponding to the first pulse, $n$ is the sequential pulse number and $D_r$ is the drift rate. The instantaneous emission of the $n$-th pulse is described by $I_n(\varphi)=\Sigma_{i=1}I_i(\varphi)$, where the sum includes a number of adjacent features contributing significantly to the observed intensity, depends strongly on the inclination $\alpha$ and the impact $\beta$ angles, determining the cut of the line-of-sight tracjectory across the beam (Fig. \[fig1\]). In fact, the running polar co-ordinates along the line-of-sight trajectory can be expressed in the form $d(\varphi)=[\rho(\varphi)/1^{\circ}.24]r_pr_6^{-1/2}P^{1/2}$ and $\sigma(\varphi)={\rm atan}\left\{\frac{(\sin\varphi\sin\alpha\sin(\alpha+
\beta)}{\cos(\alpha+\beta)-\cos\alpha\cos\rho(\varphi)} \right\}$, and $\rho(\varphi)=2{\rm asin}
\left\{\sin^2(\varphi/2)\sin\alpha\sin(\alpha+\beta)+
\sin^2(\beta/2)\right\}$, [@ggr84]. The average pulse profile is therefore $I(\varphi)=\frac{1}{N}\Sigma_{n=1}^{N}I_n(\varphi)$, where $N$ is the number of averaged single pulses. The geometry of pulsar radiation described above [for more details see @gk96; @gk97 GGGK] can be applied to both conal and patchy beam models (Fig. \[fig1\]).
We adopt a model of a pulsar beam in which the instantenous subpulse emission corresponds to a number of isolated subpulse beams, while the average emission reflects the conal structure resulting from circumferential motion of subpulse beams [@rs75; @gk96; @gk97; @gggk02]. Therefore, the longitudes of subpulse peaks correspond to phases of interception of the subpulse beams by the line-of-sight trajectory, while the longitudes of profile component peaks are determined by the intersection of the line-of-sight trajectory with the average cones. The important point is that within such model the subpulse enhancements generally follow bundles of magnetic field lines different from those of enhancements corresponding to the profile components (Fig. \[fig1\] - right-hand side). On the contrary, within the patchy model both subpulse and profile component enhancements follow approximately the same bundles of field lines (Fig. \[fig1\] - left-hand side). This should be clear from the composite Fig. \[fig1\]. In fact, subpulse enhancements are associated in both models with small subpulse spots $S$ following a narrow bundle of dipolar field lines, while the profile components correspond to the conical structures $C$ in the conal model, and to the narrow patches $PA$ enclosing subpulse spots in the patchy model. Because of the diverging nature of dipolar field lines controlling the plasma flow, all discussed emission features $S$ (spots), $SP$ (sub-patches), $PA$ (patches) and $C$ (cones) are frequency dependent, and two frequencies $\nu_1$ and $\nu_2$ are marked for each feature in Fig. \[fig1\].
![ Simulation of the distribution of separations $\Delta\varphi_s$ between subpulse peaks as compared with separation $\Delta\varphi_p$ between the component peaks (vertical solid lines) measured at two frequencies (400 and 1400 MHz) within the framework of the conal model (panel a) and various versions of patchy model (panels b - f). The number of pulses (out of the total number of 2000 simulated) is shown on the vertical axis and the separation in degrees of longitude is shown on the horizontal axis. The dashed vertical line going through panels b - f is shown to refer to the separation $\Delta\varphi_p=2^\circ.875$ in the pure conal model (panel a). \[fig2\]](jkfigure2.eps){height="20cm"}
Conal model of pulsar beams
---------------------------
The frequency dependence of pulsar emission patterns within the angular beaming model of subpulse emission, and the related conal model for a mean pulsar beam is illustrated in the right-hand side of the composite Fig. \[fig1\]. The subpulse-associated sub-beams corresponding to HPBW (half power beam width) of subpulse emission (thin small circles), which are called spots and marked by $S(\nu)$, perform a more or less organized circumferential motion around the magnetic axis $\bf{m}$, in which the azimuthal angle $\sigma$ varies with time, while the opening angle $\rho(\nu)$ remains constant. On average, this motion determines the cones (thick large circles) of the maximum mean intensity with the opening angles $\rho_1$ and $\rho_2$ at frequencies $\nu_1$ and $\nu_2$, respectively. The frequency-dependent longitude $\varphi_p(\nu)$ of the profile component is determined by the intersection of the line-of-sight with the average cone at the frequency-dependent opening angle $\rho(\nu)$. On the other hand, the frequency-dependent longitude $\varphi_s(\nu)$ of a subpulse peak is determined by the local maximum intensity along the cut of the line-of-sight through the subpulse spot $S(\nu)$. When the frequency changes from $\nu_1$ to $\nu_2$, the subpulse peak longitude changes from $\varphi_s(\nu_1)$ to $\varphi_s(\nu_2)$, while the corresponding profile peak longitude changes from $\varphi_p(\nu_1)$ to $\varphi_p(\nu_2)$. Note that the frequency separation $\Delta\varphi_p=\varphi_p(\nu_2)-\varphi_p(\nu_1)$ of profile peaks is generally larger than the frequency separation $\Delta\varphi_s=\varphi_s(\nu_2)-\varphi_s(\nu_1)$ of subpulse peaks. This is consistent with observations published by @i93, confirmed recently by GGGK. As we argue in the next section, within the patchy beam model $\Delta\varphi_p\approx\Delta\varphi_s$, which is quite different from the conal model in which $\Delta\varphi_p\gapp\Delta\varphi_s$.
We now attempt to quantify our qualitative statements presented above. We use the geometrical method of transferring the instantaneous intensity distribution from the polar cap to the radio emission region along the dipolar field lines. To be able to compare the results of simulations with the observational data of PSR B0329+54 (GGGK) we adopt the following parameters $P=0.714$ s, $\dot P=2\cdot 10^{-15}$, $\alpha=20^{\circ}$, $\beta=4^{\circ}$, $\nu_1=0.4$ GHz and $\nu_2=1.4$ GHz. The conal model is represented by an arrangement of 12 equisized and equidistant sparks, each having the HPBW diameter ${\cal D}\sim 0.2 r_p$, circulating around the magnetic pole at a distance of about 2/3 of the polar cap radius $r_p$. The results of simulations are presented in Fig. 2 (panel a). The vertical solid line at about $2.9^\circ$ represents the separation $\Delta\varphi_p$ between the peaks of the average component measured at the two frequencies. The distribution of the separations $\Delta\varphi_s$ of subpulse peaks measured at the two frequencies has a width of about $0.8^\circ$ and peaks around 1.1$^{\circ}$. The apparent difference between the peak of subpulse separation and the component peak separation i.e. $\Delta\varphi_p - \Delta\varphi_s^{\rm peak}\sim 1.8^\circ$, as well as the width and somewhat skewed shape of the distribution, resemble the observational data quite well (see Fig.4 in GGGK, keeping in mind that zero position bin in their histograms corresponds to cases when $\Delta\varphi_p=\Delta\varphi_s$). We have taken into account only subpulses that corresponded to the same spark at both frequencies. In terms of the observational data this corresponds to subpulses correlating at two frequencies (refer to the CCF technique used in GGGK).
Patchy model of pulsar beams
----------------------------
Within the patchy model of pulsar beams [LM88; @m95] the observed pulse profile is the product of a “source function” and a “window function”, which can be related to the energy/density distribution of plasma beams along different bundles of dipolar field lines and to properties of the emission mechanism, respectively. This model is illustrated schematically on the left-hand side of the composite Fig. \[fig1\]. The subpulse-associated spots marked by $SP(\nu)$ corresponding to the HPBW of subpulse emission (thin small circles), occur within the limited areas $PA(\nu)$ called patches (the simplest and probably unrealistic model is when a patch $PA$ can accommodate just one sub-patch $SP$). The occurrence of $SP(\nu)$ within $PA(\nu)$ can be completely random, or more or less organized (including drifting). The dependence on frequency $\nu$ reflects the diverging nature of dipolar field lines, thus $\nu_2<\nu_1$. The bundle of field lines associated with a patch $PA(\nu)$ is only slightly larger than each bundle associated with $SP(\nu)$. This means that enhancements corresponding to subpulses and profile components follow approximately the same bundles of field lines. Thus, the frequency dependence of emission patterns in the patchy model (left-hand side of Fig. \[fig1\]) should be different from that of the conal model (right-hand side of Fig. \[fig1\]), in which subpulse enhancements (spots) and profile components (cones) generally follow quite different bundles of field lines. If the patch is relatively small as compared to the entire polar cap, then the frequency separation of the profile components $\Delta\varphi_p=\varphi_p(\nu_2)-\varphi_p(\nu_1)$ and that of subpulses $\Delta\varphi_s=\varphi_s(\nu_2)-\varphi_s(\nu_1)$ should be about the same. We demonstrate this below by means of geometrical simulations.
We have calculated a sequence of single pulses and average emission again for the case of PSR B0329+54, assuming that the subpatch $SP(\nu)$ is comparable in size with the spark associated emission in the conal model, and that the patch $PA(\nu)$ is twice larger than $SP(\nu)$. Thus, the projection of a patch onto the polar cap has a characteristic dimension $P\sim 2{\cal D}\sim 0.4r_p$. Such a patch encompasses about 30$^{\circ}$ in magnetic azimuth (roughly corresponding to the scale presented in Fig.1 (left-hand side)). The results of simulations are presented in Fig. 2 (panel b). The solid vertical line at about 1$^{\circ}$.25 represents the separation of the profile component associated with $PA(\nu)$ measured at the two frequencies. The narrow distribution of subpulse peak separations peaks exactly at the same value (we have again taken into account only subpulses associated with the same subpatches $SP(\nu)$ at both frequencies). It is worth noting that if we chose the sizes of $SP(\nu)$ and $PA(\nu)$ to be equal (which is the simplest and an unrealistic model of the patchy emission), the distribution would be represented by a delta function coinciding with the solid vertical line. The case presented in panel (b) is inconsistent with the frequency dependence of pulsar radiation patterns (GGGK and reference therein).
We now start to increase the azimuthal dimension of the patch $PA(\nu)$ along the cone, keeping the radial dimension the same (about 2 spark diameters ${\cal D}$ or about $0.4r_p$). Panels c - f in Fig. 2 correspond to the elongation factor 1.3, 1.5, 2.0 and 2.5, respectively. Thus, the maximum elongation is equivalent to the length scale comparable with the polar cap radius $r_p$. As one can easily notice from panels c - f, increasing elongation results in two effects: (1) the separation of the component peaks increases towards the value corresponding to the conal model (dashed vertical line) i.e $\Delta\varphi_p \to 3^\circ$, while the distribution of subpulse peak separations $\Delta\varphi_s$ gets broader and broader and peaks at correspondingly larger and larger distances from $\Delta\varphi_p$. Moreover, the skewed shape of the distribution becomes more and more apparent in panels (d) and (e), to such an extent that the case presented in panel (f) seems almost undistinguishable from the pure conal case presented in panel (a), except that the width of the distribution is too large compared with observations (see Fig. 4 in GGGK). Further increasing of the elongation factor beyond 2.5 (corresponding to 60$^\circ$ of magnetic azimuth or 1/6 of the full cone) does not practically change the results of the simulations. This means that the unrealistic patches elongated to a large extent along circles centered on the magnetic axis would resemble the conal model. Although this conclusion seems trivial, it allows us to constrain some characteristics of both patchy and conal models. This is described in the two following paragraphs.
First we can ask: what is the probability that an adequately large and favourable patch elongated along a cone and corresponding to panel (f) mimics the conal model presented in panel (a) of Fig. 2. Of course, we have to take into account that another similar patch is required on the opposite side of the fiducial plane (see Fig.1) to account for the second outermost component of PSR B0329+54. Let us assume for simplicity that our elongated patch is a rectangular figure with the shorter side $A\approx 0.4r_p$ and the longer side $B\sim 1.0r_p$ (see above for an estimate of the dimensions). Thus, the surface area of such a patch $S_{patch}\approx A\cdot B=0.4r_p^2$ and the probability of its occurence in any location of the polar cap is ${\cal P}_1=S_{patch}/S_{cap}=0.4r_p^2/(\pi r_p)^2\approx 0.13$. The probability of occurence of another such patch somewhere on the remaining part of the polar cap is ${\cal P}_2=S_{patch}/(S_{cap}-S_{patch})\approx 0.15$. To estimate the probability of the proper alignment along the cone, we can calculate a number of different independent orientations of our rectangular figure inscribed into a circle of the diameter approximately equal to $A=r_p$. One can easily show that the number of independent orientations is approximaly $2\pi(r_p/2)/(0.4r_p)\sim 8$ and thus the probability of alignment along a cone ${\cal P}_0\sim 0.12$. So far, the resultant probability is ${\cal P}={\cal P}_1\cdot {\cal P}_2\cdot {\cal P}_0^2\approx 3\cdot 10^{-4}$. The requirement of having two such patches symmetricaly placed with respect to the fiducial plane only decreases the final probability. Thus, we can conclude that mimicking the conal radiation pattern by a specially arranged patchy distribution is extremly unlikely, at least in the case of PSR B0329+54 and other pulsars showing similar frequency dependence of emission patterns (GGGK and references therein).
We can now approach the results of our elongation exercise from a different angle. Since further elongation beyond the case presented in panel (f) does not change the obtained distribution, we can constrain the possible spread of the cone in the radial direction as compared with the ideal (and probably not realistic) case presented in panel (a). Remembering that the initial patch size $PA$ (panel b) was twice $SP$, with $SP$ corresponding to the spark size $\cal D$, we conclude that the realistic version of the cone can accommodate up to two sparks in radial dimension. This means that a locus of maximum intensity within the average pulsar beam has the form of a narrow ring rather than a circle.
{width="17cm"}
Beam reconstruction techniques:
-------------------------------
Recently Han & Manchester (2001; HM01 hereafter) attempted to reveal the shape of pulsar radio beams. They claim to have constructed a two-dimensional image of the “average” mean pulsar beam using a special technique applied to all available multi-component pulse profiles with good quality polarimetric data. They mapped the observed profile intensity onto the line-of-sight crossing the normalized polar cap at the normalized impact angle $\beta_n$ estimated from the polarization angle swing. To include a second dimension (perpendicular to the line-of-sight), they broadened the distribution in latitude applying a Gaussian to each longitudinal sample. Adding all pulsars together they obtained a global average beam pattern, which (after normalization to correct for the nonuniform distribution of $\beta_n$) they believe represents the global mean pulsar beam. HM01 concluded that their results are consistent with the patchy rather than conal beam model (see their Fig. 4). However, as we demonstrate in items (i)-(iv) below, their beam reconstruction technique is not general enough to reveal the true structure of pulsar beams.
{width="17cm"}
\(i) HM01 projected all emission features onto a normalized polar cap, ignoring the dependence of the beaming angles on the emission altitude, which most probably depends both on the radio frequency and on the pulsar period and its derivative [@kg97; @kg98; @k01]. Taking into account the diverging nature of dipolar field lines with increasing altitudes, the projection of the emission pattern onto the polar cap must be performed more carefully. HM01 used data at frequencies between 600 MHz and 1.6 GHz, which probably takes care of frequency dependence of the opening angles $\rho(\nu)\propto
r^{1/2}(\nu)\propto(\nu)^{-0.1}$. However, the period dependence $\rho(p)\propto P^{\sim 0.2}$ is stronger and can result in a broader spread of projections onto the normalized polar cap for short and long period pulsars, respectively.
\(ii) A similar problem concerns the putative conal structure of pulsar beams which HM01 did not reject a priori. However, they assumed that adding the normalized polar caps of different pulsars would not smear the conal structure, if it existed. It seems that the conal patterns are almost certainly different in different pulsars. MD99 showed that the pulsar emission beams follow a nested cone structure with up to three distinct cones, although only one or more of the cones may be active in a given pulsar [see also @r93; @gks93]. Also GS00 argued that the number of cones is a function of the basic pulsar parameters $P$ and $\dot{P}$.
\(iii) The analysis of HM01 is based on the orthogonal normalized impact parameter $\beta_n=\beta_{90}/\rho_{90}$ , where $\beta_{90}$ and $\rho_{90}$ are the impact angle $\beta$ and the beam radius $\rho$ computed for the inclination angle $\alpha=90^{\circ}$ (LM88). The orthogonal normalized impact parameter $\beta_n$ differs from the actual normalized impact parameter $\beta/\rho$ in many aspects [e.g. @gks93]. We would like to emphasize here that $\beta/\rho$ depends on the pulsar period $P$, the inclination angle $\alpha$ and the observing frequency $\nu$. These dependences, which can affect the emission patterns, are not accounted for in the HM01 analysis based on the orthogonal impact parameter.
\(iv) HM01 included the core components in their analysis. This is another possible source of confusion, since the core components are rather randomly placed with respect to the midpoint of the overall profile. For this reason MD99 excluded the core components from their analysis, which revealed a nested cone structure of pulsar beams.
Given the problems listed above, we conclude that the results of the HM01’s beam reconstruction are illusive. The lack of an apparent conal structure in their ’global’ pulsar beam does not exclude the existence of such structures in particular pulsars, expecially if they are determined by physical and geometrical factors, which may vary a great deal among different pulsars. HM01 demonstrated that the conal emission is not confined to a single region at the beam boundary. They also concluded that if multiple cones exist, they are at different radii relative to the beam radius in different pulsars. This is exactly what the multi-nested cone model of GS00 predicts. Both the number of cones and the relative radius of a given cone depend on a pulsar period and its derivative in their model.
As an example of pulsar modelling within the multi-nested cone scenario (GS00), we use the case of PSR J1834$-$0426. The patchy model for this pulsar was presented by HM01 (their Fig. 1). This pulsar has a very broad profile with the pulse width $W\sim 130$ degrees of longitude, which implies a small inclination angle $\alpha\ll 90^{\circ}$. Assuming a realistic emission altitude $r_6\sim 40$ [@kg97; @kg98] we estimated the observational angles $\alpha\sim 10^{\circ}$ and $\beta\sim 2^{\circ}.5$ (consistent with $\alpha=7.9^{\circ}$ and $\beta=1.6^{\circ}$ given by LM88). Figure 3 shows an instantenous arrangement of sparks on the polar cap of PSR J1834$-$0426, obtained from the complexity parameter of GS00. The circumferential motion of these sparks results in the average structure of two nested cones (e.g. Fig. 6 in GGGK). Notice that about 35% of the line-of-sight trajectory stays within the beam, which leads to a broad $\sim 130^{\circ}$ average profile presented in Fig. 4.
Discussion and conclusions
==========================
In this paper we explore a geometrical method of pulsar radiation simulation, based on two well-justified assumptions: (i) the elementary coherent radio emission is narrow-band, and the emission altitude depends on both the frequency and the pulsar period, (ii) the emission is relativistically beamed tangently to dipolar magnetic field lines. We have considered two competitive models of the organization of pulsar emission beams: the conal model, in which enhancements related to subpulse emission in single pulses are distributed along the cones corresponding to maximum average intensity, and the patchy beam model in which subpulse enhancements corresponding to the component of the mean profile are confined to the patchy area limited both in azimuthal and radial dimensions. We examined the consistency of these rival models with the variety of observational data.
We have argued that a number of observational properties of pulsar radio emission, namely: (i) binomial distribution of the opening angles [@r93; @gks93; @kwj94]; (ii) high impact angles corresponding to single and double profile pulsars (MD), and (iii) different frequency dependence of a subpulse and corresponding profile component longitudes [@i93; @gk96; @gggk02], strongly support the conal model of pulsar beams. The alternative patchy beam model is inconsistent with these observational properties of pulsar radiation.
We have also demonstrated that the beam reconstruction technique developed by @hm01 is not capable of revealing the true structure of pulsar beams. In fact, their formalism assumes implicitely that neither the radio emission altitude nor the number of putative nested-cones and their locations within the pulsar depends on the pulsar period. The lack of an apparent conal structure in their ”global beam” does not exclude the conal beam model. Thus, the results of the HM01 analysis provide no strong evidence of patchy beam structure in pulsars.
We tend to favour the version of the conal model in which the relationship between the subpulse-associated beams and cones of the average emission is established through the phenomenon of the ${\bf E}\times{\bf B}$ drift [@rs75; @dr99; @dr01 GS00], which forces the spark filaments of plasma to rotate slowly around the magnetic axis. This “spark model” of radio pulsars was recently tested statistically by @fan01. They showed, by means of Monte Carlo simulations, that various pulsar parameters can be reproduced if both the spark dimension and their mutual separation are approximately equal to the height $h$ of the polar gap [@rs75 GS00], or consequently, the maximum number of sparks along the diameter of the polar cap with the radius $r_p$ is $N_{max}\sim r_p/h$. Thus, their conclusions are consistent with the assumptions used in this paper.
We note that evidence of a relationship between drifting subpulses and the conal structure of mean pulsar beams already exists in the literature. @hw87 examined three pulsars with triple average profiles showing subpulses drifting across the full pulse window (including the central component). They found that these pulsars are consistent with two nested cones of emission, each associated with a prominent subpulse drift. Moreover, @pw86 showed that in complex profile pulsars there is a strong correlation between drifting subpulses associated with different profile components. Such correlations are natural within the ${\bf E}\times{\bf B}$ induced conal model, but inconsistent with the patchy model of pulsar beams.
Finally, we suggest that more pulsars should be observed in the single pulse, simultanenous dual frequency mode. Our simulation method illustrated and described in Fig. 2 can be adapted to the analysis of such data in order to ultimately discriminate between conal and patchy beam models in pulsars.
This paper is supported in part by the Grant 2 P03D 008 19 of the Polish State Committee for Scientific Research. We are grateful to the referee Dr. D. Mitra for extremely helpful comments that improved the final paper. We also thank Prof. Dr. R. Wielebinski for hospitality at the MPIfR in Bonn, where this work was completed. We thank E. Gil and M. Wujciów for technical assistance.
Cordes, J.M. 1978, ApJ, 222, 1006 Cordes, J.M. 1992, in IAU Coll. 128, ed. T.H. Hankins, J.M. Rankin, & J.A. Gil (Zielona Gora: Pedagogical Univ. Press), p.253 Deshpande, A.A., & Rankin, J.M. 1999, ApJ, 524, 1008 Deshpande, A.A., & Rankin, J.M. 2001, MNRAS, 322, 438 Fan G.L., Cheng K.S., & Manchester R.N., 2001, ApJ, 557, 297 Gil, J. 1981, Acta Phys. Pol. B12, 1081 Gil, J., Gronkowski, P., & Rudnicki, W. 1984, A&A, 132, 312 Gil, J., Kijak, J., & Seiradakis, J.H. 1993, A&A, 272, 268 Gil, J. & Krawczyk, A. 1996, MNRAS, 280, 143 Gil, J., & Krawczyk, A. 1997, MNRAS, 285, 561 Gil, J., & Sendyk, M. 2000, ApJ, 541, 351 (GS00) Gil, J., Gupta, Y., Gothoskar, P.I., & Kijak, J. 2002, ApJ, 565, 500 (GGGK) Gould, D.M., & Lyne, A.G. 1998, MNRAS, 301, 235 Hankins, T.H., & Wolszczan, A. 1987, ApJ, 318, 410 Han, J.L., & Manchester, R.N. 2001, MNRAS, 320, L35 (HM01) Izvekova, V.A., Kuzmin, A.D., Lyne, A.G., Shitov, Yu.P., & Smith F.G. 1993, MNRAS, 261, 865 Kijak, J., & Gil, J. 1997, MNRAS, 288, 631 Kijak, J., & Gil, J. 1998, MNRAS, 299, 855 Kijak, J. 2001, MNRAS, 323, 537 Kramer, M., Wielebinski, R., Jessner, A., Gil, J.A., & Seiradakis, J.H. 1994, A&AS, 107, 515 Lyne, A.G., & Manchester, R.N. 1988, MNRAS, 234, 477 (LM88) Manchester, R.N. 1995, JA&A, 23, 283 Mitra, D., & Deshpande, A.A. 1999, A&A, 346, 906 (MD99) Prószyński, M., & Wolszczan, A. 1986, ApJ, 307, 540 Rankin, J.M. 1983, ApJ, 274, 333 Rankin, J.M. 1993, ApJ, 405, 285 Ruderman, M.A., & Sutherland, P.G. 1975, ApJ, 196, 51
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss lepton flavor violation (LFV) associated with tau leptons in the general framework of the two Higgs doublet model, in which LFV couplings are introduced as a deviation from Model II Yukawa interaction. Parameters of the model are constrained from experimental results and also from requirements of theoretical consistencies such as vacuum stability and perturbative unitarity. Current data for LFV rare tau decays provide substantial upper limits on the LFV Yukawa couplings in the large $\tan\beta$ region, which are comparable with predictions in fundamental theories. Here $\tan\beta$ is the ratio of vacuum expectation values of the two Higgs doublets. A search for the LFV decays $\phi^{0} \rightarrow \tau^\pm \mu^\mp$ $(\tau^\pm e^\mp)$ of neutral Higgs bosons ($\phi^{0} =h,H$ and $A$) at future collider experiments can be useful to further constrain the LFV couplings, especially in the relatively small $\tan\beta$ region ($\tan\beta \lesssim 30 $), where rare tau decay data cannot give any strong limit.'
author:
- 'S. Kanemura[^1], T. Ota, and K. Tsumura'
title: |
\
Measurement of lepton flavor violating Yukawa couplings at ILC
---
Introduction
============
The structure of the electroweak symmetry beaking sector would directly connect to the property of physics beyond the standard model (SM). Models of such new physics predict extended Higgs sectors with more than one scalar doublets in the low energy effective theory. These extended Higgs models would show distinctive features from the SM phenomenology. The most obvious evidence would be the confirmation of the existence of the extra scalar states such as CP-odd and charged states at future collider experiments. Even if they are too heavy to be directly discovered and only the lightest Higgs boson is found at the experiments, we would be able to explore the extended Higgs sector by searching for deviations from the SM predictions on the couplings with gauge bosons and fermions as well as on the self coupling. Furthermore, search for non-SM interactions can also be useful.
Lepton flavor violation (LFV) is an example for such non-SM phenomena. In particular, LFV in the Yukawa sector can only appear for extended Higgs sectors. Flavor violation between electrons and muons[@KunoOkada] has been tested through rare muon decays such as $\mu^{}\rightarrow e^{} \gamma$ and $\mu^{} \rightarrow e^{}e^{+}e^{-}$, as well as through $\mu$-$e$ conversion. Tau lepton associated LFV has also been studied by rare decays of tau leptons such as $\tau \rightarrow \ell_i P^0 $[@belle-tau-lp0], $\tau \rightarrow \ell_i M^+M'^-$[@belle-tau-lMM; @babar-tau-lMM], $\tau \rightarrow \ell_i\ell'^+\ell'^-$[@belle-tau-3l; @babar-tau-3l], and $\tau \rightarrow \ell_i\gamma$[@belle-tau-egamma; @belle-tau-mugamma; @babar-tau-mugamma], where $\ell_i$ ($i=1,2$) respectively represent an electron and a muon, $P^0$ does $\pi^0$, $\eta$ and $\eta'$ mesons, $M^\pm$ ($M'^\pm$) does $\pi^\pm$ and $K^\pm$ mesons, and $\ell'^\pm=e^\pm$ and $\mu^\pm$. The LFV Yukawa couplings can be constrained from the data for these processes especially those with the Higgs boson mediation. For $\mu$-$e$ mixing, the Higgs boson mediated LFV coupling has been discussed in Ref. [@KitanoKoikeKomineOkada; @muegamma-type3THDM]. Tau lepton associated LFV processes with the Higgs boson mediation have been discussed in models with supersymmetry (SUSY)[@BK; @DER; @Sher-tmeta; @Rossi] as well as in the two Higgs doublet model (THDM) in some specific scenarios[@cheng-sher; @HiggsLFV-THDM; @Iltan]. In Ref. [@BHHS], tau associated LFV processes have been discussed comprehensively in the framework of 4-Fermi contact interactions. Phenomenological consequences of the LFV Yukawa couplings associated with tau leptons have also been studied for future observables such as $B_s$ decays[@bsmutau; @DER] at (super) B factories[@super-B] and Higgs boson decays[@Pilaftsis; @Rossi; @Herrero; @kot] at CERN LHC[@Assamagan], an electron-positron linear collider (LC)[@Osaka] and a muon collider[@muon-collider]. Furthermore, deep inelastic scattering processes $\mu N \to \tau X$[@Sher-turan; @mutauDIS] from intense high energy muons at neutrino factories (or muon colliders) and $e N \to \tau X$[@mutauDIS] by using the electron (positron) beam of a LC would be useful to further explore the tau lepton associated LFV Yukawa couplings.
In this talk, we discuss LFV in Higgs boson decays into a $\tau$-$\ell_i$ pair in the general framework of the THDM[@kot]. We study experimental upper limits on the tau lepton associated LFV Yukawa couplings to evaluate possible maximal values of the branching fractions. The parameter space of the model is tested by theoretical requirements for vacuum stability[@VS] and perturbative unitarity[@PU-1; @PU-2]. Current data from electroweak precision measurements at LEP[@LEP; @smlike1] and those at the B factories[@bsg-ex] also strongly constrain parameters of the Higgs potential. Under these theoretical bounds and experimental limits on the model, possible maximal values of the LFV couplings of $\tau$-$\ell_i$-$\phi^{0}$ are obtained by using the current data for rare tau decays, where $\phi^{0}$ represents two CP-even ($h$ and $H$) and a CP-odd ($A$) Higgs bosons. We then evaluate branching ratios of $\phi^0 \to \tau^\pm \ell_i^\mp$ with the maximal allowed values of the LFV couplings of $\tau$-$\ell_i$-$\phi^{0}$ in a wide range of the parameter space.
Lepton flavor violation in Yukawa interaction {#Sec:Model}
=============================================
The Higgs sector of the general THDM is expressed as $$\begin{aligned}
-\mathcal{L}_{\text{Higgs}}
&=&
m_{1}^{2} \left| \Phi_{1} \right|^{2}
+
m_{2}^{2} \left| \Phi_{2} \right|^{2}
-
\left( m_{3}^{2} \Phi_{1}^{\dagger} \Phi_{2} + {\rm H.c.} \right)
+
\frac{\lambda_{1}}{2} \left|\Phi_{1}\right|^{4}
+
\frac{\lambda_{2}}{2} \left|\Phi_{2}\right|^{4}
+
\lambda_{3} \left|\Phi_{1}\right|^{2} \left|\Phi_{2}\right|^{2}
+
\lambda_{4} \left|\Phi_{1}^{\dagger} \Phi_{2} \right|^{2}\nonumber \\
&&
+
\left\{
\frac{\lambda_{5}}{2}
\left( \Phi_{1}^{\dagger} \Phi_{2} \right)^{2} + {\rm H.c.}
\right\} + \left(
\lambda_{6} \left|\Phi_{1}\right|^{2}
\Phi_{1}^{\dagger}
\Phi_{2} +{\rm H.c.} \right)
+ \left(
\lambda_{7} \left|\Phi_{2}\right|^{2}
\Phi_{1}^{\dagger}
\Phi_{2} +{\rm H.c.} \right),
\label{eq:LHiggs}\end{aligned}$$ where $\Phi_{1}$ and $\Phi_{2}$ are the scalar iso-doublets with hypercharge $1/2$. In Eq. (\[eq:LHiggs\]), $m_{3}^{2}$, $\lambda_{5}$, $\lambda_{6}$ and $\lambda_{7}$ are complex in general. We here assume that all the parameters $m_{1-3}^{2}$ and $\lambda_{1-7}$ are real. The terms of $m_{3}^{2}$, $\lambda_{6}$ and $\lambda_{7}$ break the discrete symmetry explicitly. As we consider the model in which the discrete symmetry is explicitly broken only in the leptonic Yukawa interaction, we set the hard-breaking coupling constants to be zero in the Higgs potential; i.e., $\lambda_{6} = \lambda_{7} =0$[^2], and retain only the soft-breaking mass parameter $m_3^2$.
There are eight degrees of freedom in the two Higgs doublet fields. Three of them are absorbed by the weak gauge bosons via the Higgs mechanism, and remaining five are physical states. After the diagonalization of the mass matrices, they correspond to two CP-even ($h$ and $H$), a CP-odd ($A$), and a pair of charged ($H^{\pm}$) Higgs bosons. We define such that $h$ is lighter than $H$. The eight real parameters $m_{1-3}^{2}$ and $\lambda_{1-5}$ can be described by the same number of physical parameters; i.e., the vacuum expectation value $v$ $(\simeq 246$ GeV), the Higgs boson masses $m_{h}^{}, m_{H}^{}, m_{A}^{}$ and $m_{H^{\pm}}^{}$, the mixing angle $\alpha$ between the CP-even Higgs bosons, the ratio $\tan \beta$ ($ \equiv \langle \Phi^{0}_{2} \rangle / \langle \Phi^{0}_{1}
\rangle$) of the vacuum expectation values for two Higgs doublets, and the soft-breaking scale $M$ ($\equiv \sqrt{m_{3}^{2}/\sin\beta
\cos\beta}$) for the discrete symmetry. The quartic couplings are expressed in terms of physical parameters in Ref. [@smlike2]
Parameters of the Higgs sector are constrained from requirements of theoretical consistencies and also from the current experimental results. We here take into account two kinds of theoretical conditions; i.e., vacuum stability[@VS] and perturbative unitarity[@PU-1; @PU-2] at the tree level. The condition for tree-level unitarity, which we employ in Refs. [@PU-2; @smlike2], is described as $\left|\langle \phi_{3} \phi_{4}
| a^{0} | \phi_{1} \phi_{2} \rangle \right|
< \xi, $ where $\langle \phi_{3} \phi_{4} |a^{0} | \phi_{1} \phi_{2} \rangle $ is the $s$-wave amplitude for the process of $\phi_{1} \phi_{2} \rightarrow \phi_{3}
\phi_{4}$ with $\phi_{a}$ ($a$=1-4) denoting Higgs bosons and longitudinal components of weak gauge bosons. We take the criterion $\xi$ to be 1 (and also 1/2 for comparison). The experimental constraints are provided by the LEP precision data[@LEP], the $b \rightarrow s \gamma$ results[@bsg-ex], and the direct search results for the Higgs bosons[@LEP]. The LEP precision data provide the strong constraint on the new physics structure via the gauge-boson two-point functions. The constraint on $\rho$ parameter indicates that the Higgs sector is approximately custodial $SU(2)$ symmetric. This requirement is satisfied when (i) $m_{H^{\pm}}^{} \simeq m_{A}^{}$, (ii) $m_{H^{\pm}}^{} \simeq m_{H}^{}$ with $\sin^{2} (\alpha - \beta)\simeq 1$, and (iii) $m_{H^{\pm}}^{} \simeq m_{h}^{}$ with $\cos^{2} (\alpha - \beta) \simeq 1$. It is known that in Model II, the $b \rightarrow s \gamma $ result gives the lower bound on the charged Higgs boson mass. We here take into account this bound by requiring $m_{H^{\pm}}^{} \gtrsim 350$ GeV.
Next, we consider the Yukawa interaction for charged leptons as $$\begin{aligned}
-\mathcal{L}_{\text{lepton}}=
\overline{\ell}_{R i}
\left\{
Y_{\ell_{i}} \delta_{ij} \Phi_{1}
+
\left(
Y_{\ell_{i}} \epsilon^{L}_{ij}
+
\epsilon^{R}_{ij} Y_{\ell_{j}}
\right) \Phi_{2}
\right\}
\cdot
L_{j} + {\rm H.c.},
\label{eq:Llepton}\end{aligned}$$ where $\ell_{R i}$ ($i$=1-3) are right-handed charged leptons, and $L_{i}$ ($i$=1-3) denote the lepton doublets and $Y_{\ell_{i}} $ are the Yukawa couplings for $\ell_{i}$. This interaction is reduced to be of Model II[@HHG] in the limit $\epsilon^{L,R}_{ij} \rightarrow 0$ with the discrete symmetry under $e_{R}^{i} \rightarrow + e_{R}^{i}$, $L_{i} \rightarrow + L_i$, $\Phi_{1} \rightarrow + \Phi_{1}$, and $\Phi_{2} \rightarrow - \Phi_{2}$. Nonzero values of $\epsilon_{ij}^{L,R}$ ($i\neq j$) yield the LFV Yukawa couplings after the diagonalization of the mass matrix. We note that in supersymmetric standard models, the Yukawa interaction for leptons is of Model II at the tree level, and $\epsilon_{ij}^{L,R}$ can be induced at the loop level due to slepton mixing[@MSSMRN; @BK; @DER]. For the quark sector, Model II Yukawa interactions are assumed to suppress FCNC, imposing the invariance under the transformation of $u_{R}^{i} \rightarrow - u_{R}^{i}$, $d_{R}^{i} \rightarrow + d_{R}^{i}$, $q_{L}^{i} \rightarrow + q_{L}^{i}$, $\Phi_{1} \rightarrow + \Phi_{1}$, and $\Phi_{2} \rightarrow -\Phi_{2}$.
The tau lepton associated LFV interactions in Eq. (\[eq:Llepton\]) can be reduced in the mass eigenbasis of each field to $$\begin{aligned}
-\mathcal{L}_{\tau{\rm LFV}}
&=&
\frac{m_{\tau}}{v \cos^{2}\beta}
\left(
\kappa^{L}_{3i} \overline{\tau} {\rm P}_{L} \ell_{i}
+
\kappa^{R}_{i3} \overline{\ell}_i {\rm P}_{L} \tau
\right)
\left\{
\cos\left(\alpha - \beta \right) h
+
\sin\left(\alpha - \beta \right) H
-
{\rm i} A
\right\} \nonumber \\
& &+
\frac{\sqrt{2} m_{\tau}}{v \cos^{2}\beta}
\left(
\kappa^{L}_{3i}
\overline{\tau} {\rm P}_{L} \nu_{i}
+
\kappa^{R}_{i3}
\overline{\ell_{i}} {\rm P}_{L} \nu_{\tau}
\right) H^{-}
+
{\rm H.c.},
\label{eq:tauLFV}\end{aligned}$$ where ${\rm P}_{L}$ is the projection operator to the left-handed field, and $\ell_1$ and $\ell_2$ respectively represent $e$ and $\mu$. In general, the LFV parameters $\kappa^{L,R}_{ij}$ can be expressed in terms of $\epsilon^{L,R}_{ij}$ and $\tan\beta$.[^3] We here take these $\kappa^{L,R}_{ij}$ as effective couplings, and investigate their phenomenological consequences. We note that Eq. (\[eq:tauLFV\]) is exact in the limit of $m_{\ell_{i}}^{} \to 0$. The terms of $\kappa_{i3}^L$ and $\kappa_{3i}^R$ ($i=1, 2$) are proportional to $m_{\ell_i}^{}$, so that they decouple in this approximation.
We briefly discuss relationship between $\kappa^{L,R}_{ij}$ and new physics models beyond the cut-off scale of the [*effective*]{} THDM. When a new physics model is specified at the high energy scale, $\kappa^{L,R}_{ij}$ can be predicted as a function of the model parameters. For example, in the MSSM, slepton mixing can be a source of LFV. Notice that the induced LFV Higgs interactions do not necessarily decouple in the limit where the SUSY particles are sufficiently heavy, because their couplings only depend on the ratio of the SUSY parameters. Therefore, the Higgs associated LFV processes can become important in a scenario with the soft-SUSY-breaking scale $m_{\rm SUSY}^{}$ to be much higher than the electroweak one. In the MSSM, predicted values of $|\kappa_{3i}^L|^2$ can be as large as of ${\cal O}({10^{-6}})$ when $m_{\rm SUSY}^{}$ is a few TeV[@Rossi; @Osaka]. In the MSSM with right-handed neutrinos, left-handed slepton mixing may be a consequence of running effects of the neutrino Yukawa couplings between the scale of the grand unification and that of the right-handed neutrinos[@MSSMRN]. The parameters $\kappa^{L}_{3i}$ are mainly induced by mixing of left-handed sleptons[@BK; @DER; @Sher-tmeta; @Rossi; @Osaka; @Herrero]. The LFV Yukawa interactions can also appear effectively in the Zee model[@Zee]. The LFV parameters $\kappa^{L,R}_{ij}$ are induced through flavor violating couplings in the charged scalar interactions with leptons.
Mode Belle (90% CL) BaBar (90% CL)
------------------------------------- ----------------------------------------- -------------------------------------
$\tau^-\rightarrow e^-\pi^0$ [@belle-tau-lp0]
$\tau^-\rightarrow e^-\eta$ [@belle-tau-lp0]
$\tau^-\rightarrow e^-\eta'$ [@belle-tau-lp0]
$\tau^-\rightarrow \mu^-\pi^0$ [@belle-tau-lp0]
$\tau^-\rightarrow \mu^-\eta$ [@belle-tau-lp0]
$\tau^-\rightarrow \mu^-\eta'$ [@belle-tau-lp0]
$\tau^-\rightarrow e^-\pi^+\pi^-$ $8.4\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow e^-\pi^+K^-$ $5.7\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow e^-K^+\pi^-$ $5.6\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow e^-K^+K^-$ $3.0\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow \mu^-\pi^+\pi^-$ [@belle-tau-lMM] $2.9\times 10^{-7}$[@babar-tau-lMM]
$\tau^-\rightarrow \mu^-\pi^+K^-$ $6.3\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow \mu^-K^+\pi^-$ $15.5\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow \mu^-K^+K^-$ $11.7\times 10^{-7}$[@belle-tau-lMM] [@babar-tau-lMM]
$\tau^-\rightarrow e^-e^+e^-$ $3.5\times10^{-7}$[@belle-tau-3l] [@babar-tau-3l]
$\tau^-\rightarrow e^-\mu^+\mu^-$ [@belle-tau-3l] $3.3\times 10^{-7}$[@babar-tau-3l]
$\tau^-\rightarrow \mu^-e^+e^-$ [@belle-tau-3l] $2.7\times 10^{-7}$[@babar-tau-3l]
$\tau^-\rightarrow \mu^-\mu^+\mu^-$ $2.0\times 10^{-7}$[@belle-tau-3l] [@babar-tau-3l]
$\tau\rightarrow e\gamma$ [@belle-tau-egamma]
$\tau\rightarrow \mu\gamma$ $3.1\times 10^{-7}$[@belle-tau-mugamma] [@babar-tau-mugamma]
: Current experimental limits on branching ratios of the LFV rare tau decays.[]{data-label="Tab:tau-bound"}
Bound on LFV Yukawa couplings from rare tau decays {#Sec:LFV-tau-decay}
==================================================
In order to constrain the LFV parameters $|\kappa_{3i}^{L}|$ and $|\kappa_{i3}^{R}|$, we take into account the data for rare tau decay processes such as $\tau \rightarrow \ell_i P^0 $, $\tau \rightarrow \ell_i M^+M'^-$, $\tau \rightarrow \ell_i\ell'^+\ell'^-$, and $\tau \rightarrow \ell_i\gamma$, where $P^0$ represents $\pi^0$, $\eta$ and $\eta'$ mesons, $M^\pm$ ($M'^\pm$) does $\pi^\pm$ and $K^\pm$ mesons, and $\ell'^\pm=e^\pm$ and $\mu^\pm$. The list of the current data from the B factories are summarized in Table \[Tab:tau-bound\] [@belle-tau-lp0; @belle-tau-lMM; @babar-tau-lMM; @belle-tau-3l; @babar-tau-3l; @belle-tau-egamma; @belle-tau-mugamma; @babar-tau-mugamma]. These bounds may be improved at the super B factory around a digit[@super-B]. In our analysis, we take the underlined data in Table \[Tab:tau-bound\] as our numerical inputs. The branching ratios for these rare $\tau$ LFV decays are summarized in Ref. [@kot].
Since branching ratios for $\tau^-\to\ell_i^-P^0$, $\tau^-\to\ell_i^-M^+M^-$, $\tau^-\to\ell_i^-\ell'^+\ell'^-$ and $\tau^-\to\ell_i^-\gamma$ depend on different combinations of the Higgs boson masses, independent information can be obtained for the model parameters by measuring each of them. When all the masses of Higgs bosons are large, these decay processes decouple by a factor of $1/m_{\rm Higgs}^{4}$. Although these branching ratios are complicated functions of the mixing angles, each of them can be simply expressed to be proportional to $\tan^{6} \beta$ for $\tan\beta \gg 1$ in the SM-like region ($\sin(\alpha-\beta) \sim -1$ [@GH; @smlike1; @smlike2]). This $\tan^{6} \beta$ dependence is a common feature of the tau-associated LFV processes with the Higgs-mediated 4-Fermi interactions.
The experimental upper limit on $|\kappa_{3i}|^{2} (\equiv |\kappa_{3i}^L|^{2}+|\kappa_{i3}^R|^{2})$ can be obtained by using the experimental results given in Table \[Tab:tau-bound\] and analytic expressions of the decay branching ratios for rare tau LFV decay processes in Ref. [@kot]. For description, let us consider the bound from the $\tau\rightarrow \mu\eta$ results[@Sher-tmeta]; $$\begin{aligned}
&|\kappa_{32}|^{2} \leq
\left(
\left|
\kappa^{\text{max}}_{32}
\right|^{2}
\right)_{\tau^{}\rightarrow \mu^{}\eta}
\equiv
\frac{256 \pi {\rm Br}(\tau^{} \rightarrow \mu^{} \eta)_{\rm exp} m_{A}^{4}}
{9 G_{F}^{2} m_{\tau}^{3} m_{\eta}^{4} F_{\eta}^{2} \tau_{\tau}
\left(
1- \frac{m_{\eta}^{2}}{m_{\tau}^{2}}
\right)^{2}}
\frac{\cos^{6}\beta}{\sin^{2}\beta},
\label{eq:bound-from-tmeta}\end{aligned}$$ where ${\rm Br}(\tau^{} \rightarrow \mu^{} \eta)_{\rm exp}$ is the experimental upper limit on the branching ratio of $\tau^{} \rightarrow
\mu^{} \eta$ in Table \[Tab:tau-bound\]. In particular, for $\tan\beta\gg1$, the right-hand-side can be expressed by $$\begin{aligned}
\left(
\left|
\kappa^{\text{max}}_{32}
\right|^{2}
\right)_{\tau^{}\rightarrow \mu^{}\eta}
\simeq
2.3 \times 10^{-4} \times
\left(
\frac{m_{A}^{}}{350 {\rm [GeV]}}
\right)^{4}
\left(
\frac{30}{\tan\beta}
\right)^{6}.
\label{eq:kappa32sqmax}\end{aligned}$$ It can be easily seen that the bound $(|\kappa_{32}^{\text{max}}|^{2})_{\tau\rightarrow \mu\eta}$ is rapidly relaxed in the region with small $\tan\beta$ and large $m_{A}^{}$. In a similar way to Eq. (\[eq:bound-from-tmeta\]), the maximal allowed value $\left( |\kappa_{3i}^{\text{max}}|^{2} \right)_{\text{mode}}$ can be calculated for each mode. The combined upper limit $|\kappa^{\text{max}}_{3i}|^{2}$ is then given by $$\begin{aligned}
\left|\kappa^{\text{max}}_{3i}\right|^{2}
\equiv
{\rm MIN}
\left\{
\left(
\left|
\kappa^{\text{max}}_{3i}
\right|^{2}
\right)_{\tau\rightarrow\ell\eta},
\left(
\left|
\kappa^{\text{max}}_{3i}
\right|^{2}
\right)_{\tau \rightarrow \ell \mu^{+}\mu^{-}},
\left(
\left|
\kappa^{\text{max}}_{3i}
\right|^{2}
\right)_{\tau \rightarrow \ell K^{+}K^{-}},
\left(
\left|
\kappa^{\text{max}}_{3i}
\right|^{2}
\right)_{\tau \rightarrow \ell \gamma}, \cdot\cdot\cdot
\right\}.
\label{eq:kappaSq-max}\end{aligned}$$ As shown below, $\tau^{}\rightarrow \ell^{}_i\eta$ and $\tau^{}\rightarrow \ell^{}_i\gamma$ give the strongest upper limits on $|\kappa_{3i}|^{2}$ in a wide range of the parameter space. In addition, in some parameter regions $\tau^{}\rightarrow \ell^{}_iK^{+}K^{-}$ and $\tau^{}\rightarrow \ell^{}_i\mu^{+}\mu^{-}$ can also give similar limits on $|\kappa_{3i}|^{2}$ to those from the above two processes.
(7.8,6) (0,1)[![ Contours of $|\kappa^{\text{max}}_{32}|^{2}$, the possible maximal value of $|\kappa_{32}|^{2}$ from the rare tau decay results, are shown (a) in the $\tan\beta$-$m_{A}$ plane and (b) in the $\tan\beta$-$\sin^{2}(\alpha-\beta)$ plane. The parameters are taken to be (a) $m_{h}^{} =120$ GeV, $m_{H}^{} = m_{H^{\pm}} = 350$ GeV and $\sin(\alpha -\beta) = -0.9999$, and (b) $m_{h}^{} =120$ GeV and $m_{H}^{} = m_{A} = m_{H^{\pm}} = 350$ GeV. The remaining parameter $M$ is scanned from 0 to 1,000 GeV. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:allowed-kappa"}](kappa350mA.eps "fig:"){width="8cm"}]{} (0,5.5)[$m_{A}^{}$ \[GeV\]]{} (6.8,1)[$\tan\beta$]{} (3,0.5)[(a)]{}
(7.8,6) (0,1)[![ Contours of $|\kappa^{\text{max}}_{32}|^{2}$, the possible maximal value of $|\kappa_{32}|^{2}$ from the rare tau decay results, are shown (a) in the $\tan\beta$-$m_{A}$ plane and (b) in the $\tan\beta$-$\sin^{2}(\alpha-\beta)$ plane. The parameters are taken to be (a) $m_{h}^{} =120$ GeV, $m_{H}^{} = m_{H^{\pm}} = 350$ GeV and $\sin(\alpha -\beta) = -0.9999$, and (b) $m_{h}^{} =120$ GeV and $m_{H}^{} = m_{A} = m_{H^{\pm}} = 350$ GeV. The remaining parameter $M$ is scanned from 0 to 1,000 GeV. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:allowed-kappa"}](kappa350SinSq.eps "fig:"){width="7.8cm"}]{} (0,5.6)[$\sin^{2}(\alpha-\beta)$]{} (7,1)[$\tan\beta$]{} (3,0.5)[(b)]{}
In Figs. \[Fig:allowed-kappa\]-(a) and \[Fig:allowed-kappa\]-(b), contour plots for $|\kappa_{32}^{\text{max}}|^{2}$ are shown under the rare tau decay results in the $\tan\beta$-$m_{A}$ plane and the $\tan\beta$-$\sin^{2}(\alpha-\beta)$ plane, respectively. The combined excluded region from the theoretical conditions of vacuum stability and pertrbative unitarity is indicated by the dark shaded area for the criterion $\xi=1$ and by the light one for $\xi=1/2$. In Fig. \[Fig:allowed-kappa\]-(a), parameters of the Higgs sector are taken to be $m_{h}^{} =120$ GeV, $m_{H}^{} = m_{H^{\pm}}^{} = 350$ GeV and $\sin(\alpha -\beta) = -0.9999$. In Fig. \[Fig:allowed-kappa\]-(b), those are $m_{h}^{} =120$ GeV and $m_{H}^{} = m_{A}^{} = m_{H^{\pm}}^{} = 350$ GeV. The value of $|\kappa^{\text{max}}_{32}|^{2}$ is independent of $M$, the soft-breaking scale of the discrete symmetry. On the other hand, theoretical bounds from vacuum stability and perturbative unitarity are sensitive to $M$. Therefore, we evaluate such a theoretical allowed region by scanning $M$ to be from 0 to 1000 GeV. We also take into account the constraint from the $\rho$ parameter measurement and the $b\rightarrow s \gamma$ result by taking $\sin(\alpha-\beta) \simeq -1$ and $m_{H}^{}=m_{H^{\pm}}^{}$ with $m_{H^{\pm}}^{}\gtrsim 350$ GeV for Fig \[Fig:allowed-kappa\]-(a), and $m_{A}^{}=m_{H^{\pm}}^{}$ with $m_{H^{\pm}}^{}\gtrsim350$ GeV for Fig \[Fig:allowed-kappa\]-(b). From the both figures, it is easily found that the value of $|\kappa_{32}^{\text{max}}|^{2}$ can extensively be larger for smaller $\tan\beta$ in the allowed region under the theoretical constraints. For $\tan\beta \lesssim 10$ $(30)$, $|\kappa_{32}^{\text{max}}|^{2}$ can be $\mathcal{O}(0.1)$ ($\mathcal{O}(10^{-4})$). Among the rare tau decay processes, $\tau \rightarrow \mu\eta$ and $\tau \rightarrow \mu \gamma$ provide the most stringent constraints on $|\kappa_{32}^{}|^{2}$. While $\tau \rightarrow \mu \eta$ is mediated only by $A$, $\tau \rightarrow \mu\gamma$ depends on the masses of $h$, $H$, $A$ and $H^\pm$. For $\sin^{2}(\alpha-\beta) \sim 1$ and $m_{A}^{} \sim m_{H}^{}$, the branching ratio of $\tau\to\mu\gamma$ is suppressed because of the cancellation between the one-loop diagrams of $A$ and $H$. Therefore, $|\kappa_{32}|^2$ is bounded most strongly by the $\tau \rightarrow \mu \eta$ result for this case[^4]. When $m_A$ differs from $m_H$ or when $\sin^2(\alpha-\beta)$ is to some extent smaller than unity, the one-loop induced $\tau\rightarrow \mu\gamma$ process becomes important, and gives the most stringent bound on $|\kappa_{32}|^2$ of all the rare tau decay processes.
The value of $|\kappa^{\text{max}}_{32}|^{2}$ can be much larger than 100 in a wide range of the parameter region. One might think that such large values of $|\kappa_{3i}|$ cannot be consistent with the unitarity argument for the LFV Yukawa couplings. However, it should be emphasized that the above figures show the contour plots for $|\kappa_{32}^{\text{max}}|^{2}$ under the rare tau decay results, and not for $|\kappa_{32}|^{2}$. The region of $|\kappa^{\text{max}}_{32}|^{2} \gtrsim 1$ should be taken as the area where $|\kappa_{32}|^{2} $ can be as large as $\mathcal{O}(10^{-2}\text{-}10^{-4})$ easily. It is concluded that current results of the tau LFV decays do not give any substantial upper limit on $|\kappa_{32}|^2$ except for high $\tan\beta$ region ($\tan\beta \gtrsim 30$).
Finally, we comment on the bound on $|\kappa_{31}|^2$, the LFV parameters for $\tau$-$e$ mixing. Similar to $\tau$-$\mu$ mixing, we can discuss $|\kappa_{31}^{\rm max}|^2$ comparing the data of $\tau \to e \eta$, $\tau \to e \mu^+\mu^-$, $\tau \to e K^+K^-$ and $\tau \to e \gamma$ listed in Table \[Tab:tau-bound\] with the formulas given in Ref. [@kot]. These formulas for $\tau$-$e$ mixing are common with $\tau$-$\mu$ mixing except for the factor of $|\kappa_{3i}|^2$, so that difference in contours of $|\kappa_{31}^{\rm max}|^2$ from those of $|\kappa_{32}^{\rm max}|^2$ only comes from that in the data. In Table \[Tab:tau-bound\], the experimental limit for the branching ratio of $\tau \to e \eta$ is about 1.5 times weaker than that of $\tau \to \mu \eta$, while that of $\tau \to e \gamma$ is 5.7 times relaxed as compared to that of $\tau \to \mu \gamma$. Moreover, the upper limit on ${\rm Br}(\tau^- \to e^-K^+K^-)$ is 1.8 times stronger than that on ${\rm Br}(\tau^- \to \mu^-K^+K^-)$. We have numerically confirmed that there are some regions where $\tau^- \to e^-K^+K^-$ can give the most stringent bound on $|\kappa_{31}|^2$. Therefore, $|\kappa_{31}^{\rm max}|^2$ is determined from one of $\tau \to e \eta$, $\tau \to e \gamma$ and $\tau^- \to e^-K^+K^-$ depending on parameter regions.
Lepton flavor violating Higgs boson decays {#Sec:LFV-Hdecay}
==========================================
As shown in the previous section, the LFV Yukawa couplings can be tested only in the large $\tan\beta$ region by searching for rare tau decays. In order to cover the region unconstrained by rare tau decay results, we here consider LFV via the decay of the neutral Higgs bosons; i.e., $\phi^{0} \rightarrow \tau^\pm \ell_i^\mp$ ($\phi^{0}=h,H$ and $A$). Branching ratios for these decays are calculated[@Rossi; @Assamagan; @kot; @Osaka; @Herrero] to be $$\begin{aligned}
{\rm Br}(h\rightarrow \tau^{-} \ell^{+}_i) =&
\frac{1}{16\pi}
\frac{m_{\tau}^{2} \cos^{2} \left(\alpha - \beta \right)}
{v^{2} \cos^{4}\beta}
\left| \kappa_{3i} \right|^{2}
\frac{m_{h}^{} \left(1- \frac{m_{\tau}^{2}}{m_{h}^{2}}\right)^{2}}
{\Gamma(h\rightarrow\text{all})},
\label{eq:Brhtm}\\
{\rm Br}(H\rightarrow \tau^{-} \ell^{+}_i) =&
\frac{1}{16\pi}
\frac{m_{\tau}^{2} \sin^{2} \left(\alpha - \beta \right)}
{v^{2} \cos^{4}\beta}
\left| \kappa_{3i} \right|^{2}
\frac{m_{H}^{} \left(1- \frac{m_{\tau}^{2}}{m_{H}^{2}}\right)^{2}}
{\Gamma(H\rightarrow\text{all})},
\label{eq:BrhtmH}\\
{\rm Br}(A\rightarrow \tau^{-} \ell^{+}_i) =&
\frac{1}{16\pi}
\frac{m_{\tau}^{2}}
{v^{2} \cos^{4}\beta}
\left| \kappa_{3i} \right|^{2}
\frac{m_{A}^{} \left(1- \frac{m_{\tau}^{2}}{m_{A}^{2}}\right)^{2}}
{\Gamma(A\rightarrow\text{all})},
\label{eq:BrhtmA}\end{aligned}$$ where $\Gamma (\phi^{0} \rightarrow \text{all}) $ is the total width for corresponding neutral Higgs boson $\phi^{0}$. We here neglect terms of $\mathcal{O}(m_{\ell_i}^{2}/m_{\phi^{0}}^{2})$. Branching ratios for $\phi^0 \to \tau^+\ell_i^-$ coincide with those for $\phi^0 \to \tau^-\ell_i^+$ given in Eqs. (\[eq:Brhtm\]), (\[eq:BrhtmH\]) and (\[eq:BrhtmA\]). In the following, we concentrate on the decays into a $\tau$-$\mu$ pair. We take the values of the SM parameters as $\alpha_{\text{em}}=0.007297$, $G_{F}=1.166 \times 10^{-5}$ $\text{GeV}^{-2}$, $m_{Z}^{}=91.19$ GeV, $m_{\tau} = 1.777$ GeV, $m_{\mu}=0.1057$ GeV, $m_{b}=4.1$ GeV, $m_{t}=174.3$ GeV, $m_{c}=1.15$ GeV, $m_{s}=0.120$ GeV.
A search for the LFV decays $h\rightarrow \tau^\pm\ell_i^\mp$ can give important information for extended Higgs sectors and thus for the structure of new physics, even when only $h$ is found and any other direct signals for the extended Higgs sector are not obtained by experiments. We here evaluate the possible maximal value of the branching ratio ${\rm Br}(h \rightarrow \tau^{-} \mu^{+})_{\text{max}}$ under the results of the rare tau decay search, by inserting $|\kappa_{32}^{\text{max}}|^{2}$ of Eq. (\[eq:kappaSq-max\]) into the $|\kappa_{32}|^{2}$ in Eq. (\[eq:Brhtm\]).
(7.8,6) (0,1)[![Contour plots of ${\rm Br}(h \rightarrow \tau^{\pm} \mu^{\mp})_{\text{max}}$, the possible maximal values for the branching ratio under the tau rare decay results, are shown (a) in the $\tan\beta$-$m_{A}$ plane and (b) in the $\tan\beta$-$\sin(\alpha-\beta)$ plane. The parameters are taken as the same as Figs. \[Fig:allowed-kappa\]-(a) and \[Fig:allowed-kappa\]-(b), respectively. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:Br-htm"}](BRh350mA.eps "fig:"){width="8cm"}]{} (3,0.5)[(a)]{} (0,5.5)[$m_{A}^{}$ \[GeV\]]{} (6.8,1)[$\tan\beta$]{}
(7.8,6) (0,1)[![Contour plots of ${\rm Br}(h \rightarrow \tau^{\pm} \mu^{\mp})_{\text{max}}$, the possible maximal values for the branching ratio under the tau rare decay results, are shown (a) in the $\tan\beta$-$m_{A}$ plane and (b) in the $\tan\beta$-$\sin(\alpha-\beta)$ plane. The parameters are taken as the same as Figs. \[Fig:allowed-kappa\]-(a) and \[Fig:allowed-kappa\]-(b), respectively. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:Br-htm"}](BRh350SinSq.eps "fig:"){width="7.8cm"}]{} (3,0.5)[(b)]{} (0,5.5)[$\sin^{2}(\alpha-\beta)$]{} (6.8,1)[$\tan\beta$]{}
In Figs. \[Fig:Br-htm\]-(a) and \[Fig:Br-htm\]-(b), contours of ${\rm Br}(h \rightarrow \tau^{\pm}\mu^{\mp})_{\text{max}}$, which is twice of ${\rm Br}(h \rightarrow \tau^{-}\mu^{+})_{\text{max}}$, are shown in the $\tan\beta$-$m_{A}^{}$ plane and in the $\tan\beta$-$\sin^{2}(\alpha-\beta)$ plane, respectively. The parameters are taken to be the same as those for Figs. \[Fig:allowed-kappa\]-(a) and \[Fig:allowed-kappa\]-(b), respectively; i.e., (a) $m_{h}^{} =120$ GeV, $m_{H}^{} = m_{H^{\pm}}^{} = 350$ GeV and $\sin(\alpha -\beta) = -0.9999$, and (b) $m_{h}^{} =120$ GeV and $m_{H}^{} = m_{A}^{} = m_{H^{\pm}}^{} = 350$ GeV, with $M$ to be scanned from 0 to 1000 GeV. We again show the excluded area from requirements of tree-level unitarity and vacuum stability as in the same way as Figs. \[Fig:allowed-kappa\]-(a) and \[Fig:allowed-kappa\]-(b). For low and moderate values of $\tan\beta$ ($\tan\beta \lesssim 30$), where rare tau decay results cannot give substantial upper limit on $|\kappa_{32}|^{2}$, ${\rm Br}(h \rightarrow \tau^{\pm}\mu^{\mp})_{\text{max}}$ can be sufficiently large. We find that the possible maximal values of the branching ratio can be greater than $\mathcal{O}(10^{-3})$ in a wide rage of the theoretically allowed region.
For relatively lower $\tan\beta$ values, the experimental upper limits on $|\kappa_{32}|^2$ from rare tau decays are weaker, and $\text{Br}(h\rightarrow\tau^{\pm} \mu^{\mp})$ can be sufficiently large ($\gtrsim \mathcal{O}(10^{-3})$ for $m_h \sim 120$ GeV). It is expected that a sufficient number of such light $h$ can be produced at future colliders such as CERN LHC, currently planned International Linear Collider (ILC) and CERN CLIC. It has been pointed out that the decay process $h \to \tau^\pm\mu^\mp$ can easily be detected at ILC with the luminosity of 1 ab$^{-1}$, when $m_h \sim 120$ GeV and ${\rm Br}(h\rightarrow\tau^\pm\mu^\mp) \gtrsim \mathcal{O}(10^{-3})$ via the Higgsstrahlung process by using the recoil momentum of $Z$ boson[@Osaka]. Therefore, the LFV search via the decay $h \to \tau^\pm\mu^\mp$ at ILC is complementary to that via rare tau decays at (super) B factories, and the both cover a wide region of the parameter space of the lepton flavor violating THDM.
Next we discuss branching ratios for the LFV decays of heavier Higgs bosons, $H/A \to \tau^\pm \ell_i^\mp$, using Eqs. (\[eq:BrhtmH\]) and (\[eq:BrhtmA\]) under the current data of LFV rare tau decays. In the THDM, there are many possible decay modes for $H$ depending on the mass spectrum. In the numerical analysis, we included contributions from one-loop induced $Z\gamma$, $\gamma\gamma$ and $gg$ modes in addition to all the tree level modes. Those for $A$ are one-loop induced modes of $hh$, $hH$, $HH$, $h\gamma$, $H\gamma$, $W^\pm W^\mp$, $ZZ$, $Z\gamma$, $\gamma\gamma$ and $gg$ in addition to all the tree level modes. The branching ratios for $H/A \to \tau^\pm \ell_i^\mp$ are sensitive to the masses of all the Higgs bosons. Here we consider the case of $\sin(\alpha-\beta)=-1$ and $m_H^{}=m_A^{}=m_{H^\pm}^{}$ $(\equiv m_\Phi^{})$. As discussed in Sec. II, the $\rho$ parameter constraint is satisfied for this choice. From the $b \to s \gamma$ results, $m_\Phi^{}$ is taken to be greater than $350$ GeV. As also discussed in Sec. II, $M$ determines the decoupling property of heavier Higgs bosons. Although the branching ratios ${\rm Br}(H/A \to \tau^\pm \ell_i^\mp)$ are insensitive to $M$ in the present parameter set, its value strongly affects the allowed parameter region under the theoretical conditions of vacuum stability and perturbative unitarity. Notice that couplings of $H$ are similar to those of $A$ for $\sin(\alpha-\beta)=-1$ where there are no $HVV$ couplings. Hence we show the results only for the LFV decays of $H$ below. In a general case, the branching ratio of $H \to \tau^\pm \mu^\mp$ tends to be smaller than that of $A \to \tau^\pm \mu^\mp$ due to the contribution from the modes $H \to VV$.
(7.8,7) (0,1)[![ Contour plots of ${\rm Br}(H \rightarrow \tau^{\pm} \mu^{\mp})_{\text{max}}$ are shown in the $\tan\beta$-$m_{\Phi}^{}$ plane ($m_\Phi^{}\equiv m_H^{}=m_A^{}=m_{H^\pm}^{}$) for $m_h^{}=120$ GeV and $\sin(\alpha-\beta)=-1$ with (a) $M=m_\Phi^{}$, (b) $M=m_\Phi^{}/\sqrt{2}$, and (c) $M=0$. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:Br-Htm"}](BrHunit.eps "fig:"){width="5.14cm"}]{} (2,0.5)[(a)]{} (0,4.6)[$m_{\Phi}^{}$ \[GeV\]]{} (4,0.7)[$\tan\beta$]{}
(7.8,7) (0.24,1)[![ Contour plots of ${\rm Br}(H \rightarrow \tau^{\pm} \mu^{\mp})_{\text{max}}$ are shown in the $\tan\beta$-$m_{\Phi}^{}$ plane ($m_\Phi^{}\equiv m_H^{}=m_A^{}=m_{H^\pm}^{}$) for $m_h^{}=120$ GeV and $\sin(\alpha-\beta)=-1$ with (a) $M=m_\Phi^{}$, (b) $M=m_\Phi^{}/\sqrt{2}$, and (c) $M=0$. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:Br-Htm"}](BrHsqrt2.eps "fig:"){width="4.58cm"}]{} (2,0.5)[(b)]{} (4,0.7)[$\tan\beta$]{}
(7.8,7) (0,1)[![ Contour plots of ${\rm Br}(H \rightarrow \tau^{\pm} \mu^{\mp})_{\text{max}}$ are shown in the $\tan\beta$-$m_{\Phi}^{}$ plane ($m_\Phi^{}\equiv m_H^{}=m_A^{}=m_{H^\pm}^{}$) for $m_h^{}=120$ GeV and $\sin(\alpha-\beta)=-1$ with (a) $M=m_\Phi^{}$, (b) $M=m_\Phi^{}/\sqrt{2}$, and (c) $M=0$. The dark (light) shaded area indicates the excluded region by the theoretical requirements of vacuum stability and perturbative unitarity with $\xi=1$ ($\xi=1/2$). []{data-label="Fig:Br-Htm"}](BrHzero.eps "fig:"){width="4.52cm"}]{} (2,0.5)[(c)]{} (4,0.7)[$\tan\beta$]{}
In Figs. \[Fig:Br-Htm\]-(a), \[Fig:Br-Htm\]-(b) and \[Fig:Br-Htm\]-(c), contour plots of ${\rm Br}(H \to \tau^\pm\mu^\mp)_{\rm max}$, the upper limit of ${\rm Br}(H \to \tau^\pm\mu^\mp)$ under the rare tau decay results, are shown in the $\tan\beta$-$m_\Phi^{}$ plane for $M=m_\Phi^{}$, $m_\Phi^{}/\sqrt{2}$ and $0$, respectively. As expected, the contours are insensitive to the values of $M$, and approximately the same in Figs. \[Fig:Br-Htm\]-(a), \[Fig:Br-Htm\]-(b) and \[Fig:Br-Htm\]-(c). It is shown that ${\rm Br}(H \to \tau^\pm\mu^\mp)_{\rm max}$ can be larger than $10^{-3}$ except for large $\tan\beta$ values with relatively small $m_\Phi^{}$. Therefore, it turns out to be no substantial upper limit on the ${\rm Br}(H \to \tau^\pm\mu^\mp)$ in the relatively low $\tan\beta$ region ($\tan\beta \lesssim 20$) from the LFV rare tau decay results. When $M$ is smaller than $m_\Phi$, where the heavier Higgs boson partially receive their masses from the vacuum expectation value, the allowed parameter region is strongly constrained by the requirements of vacuum stability and perturbative unitarity. In particular, for $M=0$ (Fig. \[Fig:Br-Htm\]-(c)), the allowed region is limited only the area of around $\tan\beta \sim 1$ and $m_\Phi^{} \lesssim 600$ GeV.
The extra Higgs bosons ($H$, $A$ and $H^\pm$) are expected to be searched at the LHC. The signal of $gg \to H/A \to \tau^\pm \mu^\mp$ may be detectable at LHC with high luminosity (100 ${\rm fb}^{-1}$) when ${\rm Br}(H/A \to \tau^\pm\mu^\mp)$ is greater than $10^{-2}$ for $m_{H/A}^{} \sim 350$ GeV and $\tan\beta=45$[@Assamagan]. However the rate is rapidly reduced for smaller values of $\tan\beta$ and for larger values of $m_{H/A}^{}$. Further feasibility study is necessary.
Search for LFV decays of Higgs bosons at a Linear Collider
==========================================================
=0.6cm
(10,8) (1,1)[(1,3)[1]{}]{} (1,7)[(1,-3)[1]{}]{} (2,4)(0.5,0)[3]{}[ (0,0)(0.125,0.25)(0.25,0) (0.25,0)(0.375,-0.25)(0.5,0)]{} (3.5,4) (4.15,5.4) (4.15,5.4) (3.5,4)(0.5,0)[5]{}[(1,0)[0.3]{}]{} (6,4)[(1,3)[1]{}]{} (6,4)[(1,-3)[1]{}]{} (6.42,5.3) (6.42,5.3) (2,4) (3.5,4) (6,4) (4.2,5.5) (6.47,5.35) (1.5,2.5)[(1,3)[0]{}]{} (1.505,5.5)[(-1,3)[0]{}]{} (6.5,2.5)[(-1,3)[0]{}]{} (6.35,5)[(1,3)[0]{}]{} (4.16,6.5)[(0,-1)[0]{}]{} (4.82,6.5)[(2,3)[0]{}]{} (6.85,6.5)[(1,3)[0]{}]{} (7.4,6.25)[(1,1)[0]{}]{} (7.4,5.75)[(-2,-1)[0]{}]{} (0.3,0.3)[$e^{-}$]{} (0.3,7.2)[$e^{+}$]{} (2.5,3)[$Z^{*}$]{} (3,5)[$Z$]{} (4.5,3)[$h^{0}$]{} (6.8,0.3)[$\mu^{+}$]{} (6.5,4.5)[$\tau^{-}$]{} (6.8,7.2)[$\nu_{\tau}$]{} (8,6.5)[$\tau$-decay]{} (3.5,7.4)[$Z$-decay]{} (3.5,-0.5)[(a)]{}
=0.6cm
(10,8) (1,1)[(1,3)[1]{}]{} (1,7)[(1,-3)[1]{}]{} (2,4)(0.5,0)[3]{}[ (0,0)(0.125,0.25)(0.25,0) (0.25,0)(0.375,-0.25)(0.5,0)]{} (3.5,4) (4.15,5.4) (4.15,5.4) (3.5,4)(0.5,0)[5]{}[(1,0)[0.3]{}]{} (6,4)[(1,3)[1]{}]{} (6,4)[(1,-3)[1]{}]{} (6.42,5.3) (6.42,5.3) (6.4,2.8) (6.4,2.8) (2,4) (3.5,4) (6,4) (4.15,5.4) (6.4,2.75) (6.47,5.4) (1.5,2.5)[(1,3)[0]{}]{} (1.505,5.5)[(-1,3)[0]{}]{} (6.15,3.5)[(-1,3)[0]{}]{} (6.35,5)[(1,3)[0]{}]{} (4.16,6.5)[(0,-1)[0]{}]{} (4.82,6.5)[(2,3)[0]{}]{} (6.85,6.5)[(1,3)[0]{}]{} (7.4,6.25)[(1,1)[0]{}]{} (7.4,5.75)[(-2,-1)[0]{}]{} (6.655,2)[(-1,3)[0]{}]{} (6.41,1.6)[(0,1)[0]{}]{} (5.8,1.745)[(-2,-3)[0]{}]{} (0.3,0.3)[$e^{-}$]{} (0.3,7.2)[$e^{+}$]{} (2.5,3)[$Z^{*}$]{} (3,4.5)[$Z$]{} (4.5,3)[$h^{0}$]{} (6.8,7.2)[$\nu_{\tau}$]{} (6.5,4.5)[$\tau^{-}$]{} (6.5,3)[$\tau^{+}$]{} (6.85,0.55)[$\mu^{+}$]{} (6,0.35)[$\bar{\nu}_{\mu}$]{} (5,0.5)[$\nu_{\tau}$]{} (8,6.5)[$\tau$-decay]{} (3.5,7.4)[$Z$-decay]{} (3.5,-0.5)[(b)]{}
Let us consider the LFV Higgs decay $h^{0} \to \tau^{\pm} \mu^{\mp}$ at a LC in the situation where the heavier Higgs bosons nearly decouple from the gauge bosons; i.e., $\sin(\alpha-\beta) \simeq -1$. The lightest Higgs boson then approximately behaves as the SM one. The main production modes of the lightest Higgs boson at a LC are the Higgsstrahlung $e^+e^- \to Z^\ast \to Z h^{0}$ and the $W$ fusion $e^+e^- \to (W^{+\ast} \bar{\nu}_{e})(W^{-\ast} \nu_{e})
\to h^{0} \nu_{e} \bar{\nu}_{e}$. For a light $h^0$ with the mass $m_h \sim 120$ GeV, the former production mechanism is dominant at low collision energies ($\sqrt{s} < 400$-$500$ GeV), while the latter dominates at higher energies. For our purpose, the Higgsstrahlung process is useful because of its simple kinematic structure. The signal process is then $e^+e^- \to Z^\ast \to Z h^{0} \to Z \tau^\pm \mu^\mp$. We can detect the outgoing muon with high efficiency, and its momentum can be measured precisely by event-by-event. The momentum of the $Z$ boson can be reconstructed from those of its leptonic $\ell^+\ell^-$ ($\ell^\pm = e^\pm$ and $\mu^\pm$) or hadronic ($jj$) decay products. Therefore, we can identify the signal event without measuring $\tau$ momentum directly, as long as the beam spread rate for $\sqrt{s}$ is sufficiently low.
Depending on the $Z$ decay channel, the signal events are separated into two categories, $jj\tau^{\pm}\mu^{\mp}$ and $\ell^+\ell^-\tau^{\pm}\mu^{\mp}$. The energy resolution of the $Z$ boson from hadronic jets $jj$ is expected to be $0.3\sqrt{E_{Z}}$ GeV and that from $\ell^+\ell^-$ is $0.1 \sqrt{E_{Z}}$ GeV. We assume that the detection efficiencies of the $Z$ boson and the muon are 100 %, the rate of the beam energy spread is expected to be 0.1 % level, the muon momentum is measured with high precision and the mass of the lightest Higgs boson will have been determined in the 50 MeV level. We also expect that the effect of the initial state radiation is small for the collider energies that we consider ($\sqrt{s} \sim 250$-$300$ GeV). Taking into account all these numbers, we expect that the tau momentum can be determined indirectly within 3 GeV for $jj\tau^{\pm}\mu^{\mp}$ and 1 GeV for $\ell^+\ell^- \tau^{\pm}\mu^{\mp}$.
Let us evaluate the number of the signal event. We assume that the energy $\sqrt{s}$ is tuned depending on the mass of the lightest Higgs boson: i.e., we take the optimal $\sqrt{s}$ to product the lightest Higgs boson through the Higgsstrahlung process. (It is approximately given by $\sqrt{s} \sim m_{Z} + \sqrt{2} m_{h}^{}$.) The production cross section of $e^+e^- \to Z h^{0}$ is about $220$ fb for $m_{h}^{}=123$ GeV. Then, we obtain $2.2 \times 10^5$ Higgs events if the integrated luminosity is 1 ab$^{-1}$. When $|\kappa_{32}|^2 $ is $8.4 \times 10^{-6}$, about $118$ events of $jj \tau^\pm\mu^\mp$ and $11$ events of $\ell^+\ell^-\tau^{\pm}\mu^{\mp}$ can be produced.
Next, we consider the background. For the signal with the Higgs boson mass of $\sim 120$ GeV, the main background comes from $e^+e^- \to Z \tau^+\tau^- $. The number of the $Z \tau^{\pm}\mu^{\mp}$ event from $e^+e^- \to Z \tau^+\tau^-$ is estimated about $3.6 \times 10^{4}$. Although the number of the background events is huge, we can expect that a large part of them is effectively suppressed by using the kinematic cuts[@Osaka]. The irreducible background comes from the process shown in Fig.\[Fig:diagrams\]-(b): the Higgs boson decays into a tau pair, and one of the tau decays into a muon and missings ($e^{+}
e^{-} \rightarrow Z h^{0} \rightarrow Z \tau^{+}\tau^{-} \rightarrow
Z \tau^{\pm}\mu^{\mp}$+missings). We can not distinguish the signal event $h^{0}\rightarrow \tau^{\pm}\mu^{\mp}$ with the event of Fig.\[Fig:diagrams\]-(b) when the muon emitted from the tau lepton carries the similar momentum to that of the parent, because it leaves the same track on the detector as the signal event. We refer this kind of the background as [*the fake signal*]{}. As examined in Ref. [@Osaka], the number of the fake signal strongly depends on the precision of the tau momentum determination by the recoil method. We expect that it is attained with the similar precision to that of the Higgs boson mass reconstructed by the recoil momentum. We here take the uncertainty of the tau momentum as 3 GeV for $jj\tau^{\pm}\mu^{\mp}$ and as 1 GeV for $\ell^+\ell^- \tau^{\pm}\mu^{\mp}$.
Finally, we estimate the statistical significance ($S/\sqrt{B}$) for each channel. The number of the fake events is evaluated in Ref. [@Osaka], which is 460 for $jj\tau^{\pm}\mu^{\mp}$ and 15 for $\ell^+\ell^- \tau^{\pm} \mu^{\mp}$. Therefore, when $|\kappa_{32}|^2 $ is $8.4 \times 10^{-6}$ with $m_h^{}=123$ GeV, the significance can become 5.5 and 3.0 for $jj \tau^\pm\mu^\mp$ and $\ell^+\ell^-\tau^{\pm}\mu^{\mp}$, respectively, taking into account the constraint from the rare tau decay results. The combined significance can reach to 6.3. When $|\kappa_{32}|^2 $ is $3.8 \times 10^{-6}$ with $m_h^{}=123$ GeV, the number of the signal becomes smaller, and the combined significance amounts to be as large as 2.0.
Summary
=======
Lepton flavor violating decays of Higgs bosons have been studied in the general THDM, in which LFV couplings are introduced as a deviation from Model II Yukawa interaction in the lepton sector. The model parameters are constrained by requirements of tree-level unitarity and vacuum stability, and also from the experimental results. The parameters $|\kappa_{3i}|^2$ in LFV Yukawa interactions are bounded from above by using the current data for rare tau LFV decays. In the large $\tan\beta$ region ($\tan\beta \gtrsim 30 $), the upper limit on $|\kappa_{3i}|^2$ due to the rare tau decay data turns out to be substantial and comparable with the value predicted by assuming some fundamental theories such as SUSY. For smaller values of $\tan\beta$, the upper limit is rapidly relaxed, and no more substantial constraint is obtained from the rare tau decay results.
We have shown that a search for the LFV decays $\phi^{0} \rightarrow \tau^\pm \ell_i^\mp$ of neutral Higgs bosons ($\phi^{0} =h,H$ and $A$) can be useful to further constrain the LFV Yukawa couplings at future collider experiments. In particular, the decays of the lightest Higgs boson can be one of the important probes to find the evidence for the extended Higgs sector even when the SM-like situation would be preferred by the data at forthcoming collider experiments. The branching ratio for $h \to \tau^\pm \mu^\mp$ can be larger than $\mathcal{O}(10^{-3})$ except for the high $\tan\beta$ region under the constraints from the current experimental data and also from the theoretical requirements. At ILC (and in case at LHC), these branching fractions can be tested. Therefore, we conclude that the search of LFV in the Higgs boson decay at future colliders can further constrain the LFV Yukawa couplings in the parameter region where rare tau decay data cannot reach.
Note added: Recently similar work was done in Ref. [@paradisi] on the experimental upper bound on $|\kappa_{3i}|^2$.
[9]{}
Y. Kuno and Y. Okada, [*Rev. Mod. Phys.*]{} [**73**]{}, [151]{} [(2001)]{}.
Y. Enari [*et al.*]{}, the Belle Collaboration, arXiv:hep-ex/0503041.
Y. Yusa, [[*Nucl. Phys.*]{} B]{} [(Proc. Suppl.)]{} [**144**]{}, [173]{} [(2005)]{}.
M. Hodkinson (on behalf of the BaBar Collaboration), [[*Nucl. Phys.*]{} B]{} [(Proc. Suppl.)]{} [**144**]{}, [167]{} [(2005)]{}.
Y. Yusa [*et al.*]{}, the Belle Collaboration, .
B. Aubert [*et al.*]{}, the BaBar Collaboration, .
K. Hayasaka [*et al.*]{}, the Belle Collaboration, .
K. Abe [*et al.*]{}, the Belle Collaboration, .
B. Aubert [*et al.*]{}, the BaBar Collaboration, arXiv:hep-ex/0502032.
R. Kitano, M. Koike, S. Komine, and Y. Okada, .
R.A. Diaz, R. Martinez, and J.-A. Rodriguez, .
K.S. Babu and C. Kolda, .
A. Dedes, J. Ellis, and M. Raidal, .
M. Sher, .
A. Brignole and A. Rossi, ; .
T.P. Cheng and M. Sher, .
J.L. Diaz-Cruz and J.J. Toscano, ; J.L. Diaz-Cruz, R. Noriega-Papaqui, and A. Rosado, .
E.O. Iltan, ; ; hep-ph/0504013.
D. Black, T. Han, H.-J. He, and M. Sher, .
D. Guetta, J.M. Mira, and E. Nardi, .
A.G. Akeroyd [*et al.*]{}, [*Super KEKB Letter of Intent*]{}, KEK Report 04-4, arXiv:hep-ex/0406071.
A. Pilaftsis, .
E. Arganda, A.M. Curiel, M.J. Herrero, and D. Temes, .
S. Kanemura, T. Ota, K. Tsumura, hep-ph/0505191.
K.A. Assamagan, A. Deandrea, and P.-A. Delsart, .
S. Kanemura, K. Matsuda, T. Ota, T. Shindou, E. Takasugi, and K. Tsumura, .
M. Sher, ; U. Cotti, M. Pineda, and G. Tavares-Velasco, arXiv:hep-ph/0501162.
M. Sher and I. Turan, .
S. Kanemura, Y. Kuno, M. Kuze, and T. Ota, ; S. Kanemura, Y. Kuno, M. Kuze, T. Ota, and T. Takai, in preparation.
N.G. Deshpande and E. Ma, ; S. Kanemura, T. Kasai, and Y. Okada, ; S. Nie and M. Sher, .
B.W. Lee, C. Quigg, and H.B. Thacker, ; .
S. Kanemura, T. Kubota, and E. Takasugi, .
LEP Electroweak Working Group, [*http://lepewwg.web.cern.ch/LEPEWWG/*]{}.
I.F. Ginzburg, M. Krawczyk, and P. Osland, arXiv:hep-ph/0101208; [*Nucl.Instrum.Meth.*]{} [**A472**]{} [149]{} [(2001)]{}.
P. Koppenburg [*et al.*]{}, the Belle Collaboration, .
S. Kanemura, S. Kiyoura, Y. Okada, E. Senaha, and C.-P. Yuan, ; S. Kanemura, Y. Okada, E. Senaha, and C.-P. Yuan, .
J.F. Gunion, H.E. Haber, G. Kane, and S. Dawson, [*The Higgs Hunter’s Guide*]{}, Perseus Publishing, Cambridge, MA, 1990.
F. Borzumati and A. Masiero, ; J. Hisano, T. Moroi, K. Tobe, M. Yamaguchi, and T. Yanagida, ; J. Hisano, T. Moroi, K. Tobe, and M. Yamaguchi, .
A. Zee, ; Erratum ; S. Kanemura, T. Kasai, G.-L. Lin, Y. Okada, J.-J. Tseng, and C.-P. Yuan, ; K. Cheung and O. Seto, .
J.F. Gunion and H.E. Haber, .
P. Paradisi, hep-ph/0508054.
[^1]: Speaker
[^2]: Even in such a case, $\lambda_{6}$ and $\lambda_{7}$ are effectively induced at the loop level. They are suppressed by the loop factor, so that we here neglect these small effects.
[^3]: LFV parameters $\kappa_{ij}^{L,R}$ can be expressed as $$\begin{aligned}
\kappa^{X}_{ij}
=
-\frac{\epsilon^{X}_{ij}}{
\left\{ 1+ (\epsilon^{L}_{33} + \epsilon^{R}_{33})\tan\beta
\right\}^{2} }
\qquad (X=L,R),\end{aligned}$$ for the case of $\epsilon^{L,R}_{ij} \tan\beta \ll \mathcal{O}(1)$ which is satisfied in the case of the minimal supersymmetric standard model (MSSM)[@BK; @DER].
[^4]: The MSSM result approximately corresponds to this case[@Sher-tmeta].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce a notion of quantum function, and develop a compositional framework for finite quantum set theory based on a $2$-category of quantum sets and quantum functions. We use this framework to formulate a $2$-categorical theory of quantum graphs, which captures the quantum graphs and quantum graph homomorphisms recently discovered in the study of nonlocal games and zero-error communication, and relates them to quantum automorphism groups of graphs considered in the setting of compact quantum groups. We show that the $2$-categories of quantum sets and quantum graphs are semisimple. We analyse dualisable and invertible $1$-morphisms in these $2$-categories and show that they correspond precisely to the existing notions of quantum isomorphism and classical isomorphism between sets and graphs.'
bibliography:
- 'QSET.bib'
title: A compositional approach to quantum functions
---
Introduction
============
#### On quantum functions.
In this work, we introduce a notion of *quantum function* between finite *quantum sets* and propose a framework for finite quantum set theory.
These quantum functions underlie various notions of ‘quantum morphism’ previously defined in quantum information theory and noncommutative topology.[^1]
Several such notions
have recently emerged from the study of *quantum pseudo-telepathy* [@Brassard2005], a phenomenon in quantum theory where pre-shared entanglement is used to perform a task classically impossible without communication. Such tasks are usually formulated as games, where winning classical strategies correspond to certain homomorphisms between combinatorial structures; quantum strategies can then be understood as quantised versions of these homomorphisms. One such game is the *graph homomorphism game* [@Mancinska2016], leading to a notion of *quantum graph homomorphism*. We show that quantum graph homomorphisms are quantum functions between vertex sets preserving the graph structure; in fact, quantum functions can themselves be understood as perfect quantum strategies for a certain ‘function game’. The theory of quantum functions places this game-theoretic approach to quantisation into a broader mathematical context, in particular relating it to more conventional approaches to quantisation.=-2
Indeed, since the advent of quantum mechanics, quantisation has been associated with the passage from commuting to noncommuting variables. In the spirit of noncommutative topology[^2], we regard $C^*$-algebras as noncommutative analogues of topological spaces, or *quantum spaces*, with finite-dimensional $C^*$-algebras playing the role of *finite quantum sets*. Finding an appropriate definition of a quantum space of quantum functions between finite sets is a more subtle issue. Conventionally, these quantum spaces are defined as $C^*$-algebras satisfying a universal property. A prominent example is Wang’s *quantum permutation group* [@Wang1998] of a set, defined in the framework of compact quantum groups [@Woronowicz1998] and later extended to the *quantum automorphism group* [@Banica2005] of a graph. Our framework captures the finite-dimensional representation theory of these $C^*$-algebras and extends the existing theory of quantum permutation and automorphism groups to a theory of general morphisms between different sets and graphs. In particular, it relates quantum automorphism groups to the *quantum graph isomorphisms* [@Atserias2016] considered in the study of pseudo-telepathy, providing a concrete link between results in noncommutative topology and recent problems in quantum information theory.
Our theory of quantum functions — unifying and extending the above approaches to the quantisation of combinatorial structures — is formulated in categorical terms and allows us to apply techniques from categorical algebra, such as the notions of semisimplicity and dualisability. The following is a brief summary of our main results:
- We show that quantum sets and quantum functions naturally form a 2-category $\QSet$ (Definition \[def:2catqset\]) ‘quantising’ the category of finite sets and functions. Similarly, quantum graphs (Definition \[def:quantumgraphsbyadjmats\], see also [@Duan2013; @Weaver2015]) and quantum graph homomorphisms naturally form a 2-category $\QGraph$ (Definition \[def:2catqgraph\]) ‘quantising’ the category of finite graphs and graph homomorphisms. This higher compositional structure has not been noticed before.
- We characterise quantum bijections and quantum isomorphisms as *dagger-dualisable* 1-morphisms (Definition \[def:daggerdualisable\]) in the 2-categories $\QSet$ and $\QGraph$, respectively (Theorem \[thm:dual2\], Theorem \[thm:dualisablequantumgraph\]). We also show that classical bijections and graph isomorphisms are equivalences in these 2-categories (see Proposition \[prop:equivalence\]). In particular, we emphasise that quantum bijections and quantum graph isomorphisms should not be thought of as invertible but merely as dualisable, a characterisation which crucially depends on the $2$-categorical structure.
- We show that the categories $\QSet(A,B)$ of quantum functions between quantum sets $A$ and $B$, and the categories $\QGraph(G,H)$ of quantum homomorphisms between quantum graphs $G$ and $H$, are *semisimple*, a categorical property which commonly arises in representation theory. This is crucial for the structural understanding of quantum functions, and allows us to characterise certain quantum functions as essentially classical (Definition \[def:classical\]).
Our notions of quantum graph homomorphism and quantum graph isomorphism coincide with those considered in the theory of nonlocal games [@Mancinska2016; @Atserias2016] (Propositions \[prop:mancinskagraphhom\] and \[prop:Atserias\]). We also show that the monoidal category of quantum bijections on a quantum set is the category of finite-dimensional representations of the Hopf $C^*$-algebra associated to Wang’s quantum permutation group (Proposition \[prop:Wang\], [@Wang1998]). Similarly, we show that the monoidal category of quantum automorphisms of a graph is the category of finite-dimensional representations of the Hopf $C^*$-algebra associated to Banica’s quantum automorphism group (Proposition \[prop:Banica\], [@Banica2005]).
The $2$-categories $\QSet$ and $\QGraph$ therefore extend these quantum groups analogously to the way in which categories extend groups or monoids.
The results in this work serve as a foundation for finite quantum set and quantum graph theory. In a subsequent paper [@paper1b], we use this framework to classify quantum isomorphic graphs in terms of algebraical and group theoretical data.
#### A new approach to quantum functions.
While noncommutative topology is formulated in the language of operator algebras, we present a compositional, linear algebraic framework based on the graphical calculus of string diagrams. In our diagrams, wires represent finite-dimensional Hilbert spaces, and vertices represent linear maps between them. Wiring diagrams, read from bottom to top, represent composite linear maps in the obvious way. We start from the following simple graphical observation:[^3]
\[eq:basicidea\]
(1.5,0) to node\[zxnode=, pos=0.25,scale=1.2\](a) [$a$]{} node\[zxnode=, pos=0.75,scale=1.2\] (b) [$b$]{}(0,2); (\[xshift=+, yshift=-\]b.center) to \[out=20, in=70,looseness=1.1\] (\[xshift=-,yshift=+\]a.center); (\[xshift=--, yshift=-+\] a.center) to \[out=-160, in=-110, looseness=1.1\] (\[xshift=-+, yshift=--\] b.center) ;
=
(1.5,0) to node\[zxnode=, pos=0.25,scale=1.2\][$b$]{} node\[zxnode=, pos=0.75,scale=1.2\] [$a$]{}(0,2);
&
(1.5,0) to node\[zxnode=, pos=0.25,scale=1.2\][$a$]{} node\[zxnode=, pos=0.75,scale=1.2\] [$b$]{}(0,2);
(1.5,0) to node\[zxnode=, pos=0.25,scale=1.2\][$b$]{} node\[zxnode=, pos=0.75,scale=1.2\] [$a$]{}(0,2);
The two free-floating vertices on the left can move around each other, whereas on the right they are confined to move on a line and do not commute.[^4] In other words, adding a wire to a diagram can turn a commutative situation into a noncommutative one.
Taking this idea seriously leads to the following sketch of a programme. By Gelfand duality, a combinatorial theory, such as the theory of finite sets and functions or finite graphs and graph homomorphisms, can be expressed in terms of finite-dimensional commutative algebra and therefore represented in the graphical calculus. To quantise the objects of the theory — in our case, sets or graphs — one simply passes from commutative to noncommutative algebras. The novelty of our approach lies in our treatment of morphisms: having formulated the theory in term of string diagrams, we quantise morphisms by adding a wire through the corresponding vertices. We illustrate this idea with an example:
\[eq:quantisationexample\]
(0,0) to (0,2) to \[out=135\] (-0.75,3); (0,2) to \[out=45\] (0.75, 3); at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (0,0.75) to \[out=45\] (0.75, 1.75) to +(0,1.25); at (-0.75,2) [$f$]{}; at (0.75,2) [$f$]{}; at (0,0.75) ;
(0,0) to (0,2) to \[out=45\] (0.75,3); (0,2) to \[out=135\] (-0.75,3); (1.75,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$f$]{} (-1.75,2.5) to (-1.75,3); at (0,2);
=
(0,0) to (0,0.75) to \[out=45\] (0.75,1.75) to (0.75,3); (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (1.75,0) to (1.75,0.75) to \[looseness=1.1, in looseness=0.9\] node\[zxnode=, pos=0.36\] [$f$]{} node\[zxnode=, pos=0.64\] [$f$]{}(-1.75,3); at (0,0.75);
The equation on the left is one of the properties of any function between two sets, while the equation on the right is its ‘quantisation’. Further examples of classical morphisms and their quantisations are displayed in Figure \[fig:classicalquantum\].
#### The compositional structure of quantum functions.
Due to the presence of the additional Hilbert space wire introduced in our quantisation procedure, we are naturally led to consider maps on this Hilbert space, which interact in a particular way with quantum functions.
With these additional maps as morphisms, quantum functions between quantum sets $A$ and $B$ are the objects of a category $\QSet(A,B)$. This category of quantum functions between two quantum sets can be understood as a quantisation of the set of classical functions between two classical sets. In other words, our approach to quantisation leads to categorification. In particular, the category of quantum bijections on a set $X$ quantises the symmetric group $S_X$; as stated above, this is the category of quantum elements of Wang’s quantum permutation group.
A shortcoming of the compact quantum group approach is that it is restricted to automorphisms and cannot easily be applied to study quantum functions or graph homomorphisms between nonisomorphic sets or graphs, as is for example necessary for quantum pseudo-telepathy. In contrast, our approach allows us to consider all quantum functions and quantum graph homomorphisms, including those between nonisomorphic sets and graphs. We show that these form 2-categories $\QSet$ and $\QGraph$, which incorporate all the categories of quantum functions and homomorphisms between different sets and graphs into a single mathematical structure. These $2$-categories unify, generalise, and most importantly expose the connection between work on quantum permutations and quantum graph automorphisms in noncommutative topology, and on quantum graph homomorphisms and isomorphisms in quantum information theory.
Outlook
-------
We hope that our results lead to the development of further connections between the work on compact quantum groups and quantum information theory. In particular, we expect that the structural understanding of the automorphism categories $\QBij(B,B)$ and $\QIso(G,G)$ can lead to new insights into pseudo-telepathy. First considerations along these lines will appear in a companion paper [@paper1b].
We now suggest a number of other possible applications of our categorical :
- Quantum functions are closely related to quantum communication. Indeed, in Remark \[rem:teleportation\], we note that quantum bijections between a matrix algebra $\mathrm{Mat}_n$ and an $n^2$-element classical set can be understood as generalised teleportation protocols. More generally, it has been observed that quantum homomorphisms between quantum graphs admit an interpretation in terms of zero-error source-channel coding [@Stahlke2016].
- Our classification in [@paper1b] is based on the study of Frobenius monads in $\QGraph$. As symmetric monoidal $2$-categories (see Remark \[rem:symmmon2cat\]), $\QSet$ and $\QGraph$ admit several other as yet unexplored algebraic structures.
- Our quantisation framework may be extended to other combinatorial theories besides finite set theory and graph theory. In particular, in Section \[app:quantumrelation\], we show how Kuperberg and Weaver’s finite-dimensional quantum relations [@Kuperberg2012] fit into our framework; this suggests, for example, a $2$-category of quantum posets and quantum monotone functions.=-2
Outline of the paper
--------------------
In Section \[sec:background\], we recall the diagrammatic calculus for Hilbert spaces and linear maps as well as the correspondence between finite-dimensional $C^*$-algebras and certain Frobenius algebras. We express Gelfand duality between finite sets and finite-dimensional commutative $C^*$-algebras in this setting.
In Section \[sec:quantumset\], we introduce quantum sets, quantum elements and quantum functions, and define the 2-category $\QSet$ which encodes their compositional structure. We justify our definitions by a universal property.
In Section \[sec:quantumbijection\], we quantise bijections and prove that the resulting quantum bijections are precisely dagger-dualisable 1-morphisms in $\QSet$. We show that quantum bijections between classical sets correspond to projective permutation matrices as defined in the theory of nonlocal games [@Atserias2016], and to finite-dimensional representations of Wang’s quantum symmetry group algebras [@Wang1998].
In Section \[sec:qgt\], we quantise finite graphs and their homomorphisms and define the $\QGraph$ which encodes their compositional structure. We show that our definitions capture the quantum graph homomorphisms and isomorphisms of Man[č]{}inska and Roberson [@Mancinska2016] and Atserias et al [@Atserias2016], as well as the finite-dimensional representation theory of Banica and Bichon’s quantum automorphism group algebras [@Banica2005; @Bichon2003]. We also show that quantum graph isomorphisms are precisely dagger-dualisable 1-morphisms in $\QGraph$.
In Section \[sec:semisimple\], we show that the categories of quantum functions and quantum graph homomorphisms are semisimple. We discuss the operational interpretation of the direct sum of quantum functions and show how this can be used to distinguish between classical and quantum structures.
In Section \[app:quantumrelation\], we show that our reflexive quantum graphs are precisely symmetric and reflexive quantum relations in the sense of Kuperberg and Weaver [@Kuperberg2012; @Weaver2010; @Weaver2015] and fully capture the noncommutative graphs appearing in zero-error communication [@Duan2013].
Related work {#sec:relatedwork}
------------
A compressed summary of this section can be found in Figure \[fig:relatedwork\].
-- -- --
-- -- --
#### Quantum symmetry groups and noncommutative topology.
The study of quantum permutation groups — quantum variants of the symmetric groups $S_n$ in noncommutative topology — was suggested by Connes, and carried out by Wang, Banica, Bichon and others [@Wang1998; @Banica2005; @Banica2007; @Banica2007_2; @Bichon2003; @Banica2009; @Banica2007_3]. These quantum permutation groups are compact quantum groups in the sense of Woronowicz [@Woronowicz1998], obtained from a universal construction [@Wang1998]. Quantum automorphism groups of finite graphs are defined similarly [@Bichon2003; @Banica2005]. A more general universal construction of quantum spaces of maps is given by So[ł]{}tan [@Soltan2009] (see Remark \[rem:soltan\]). In Section \[sec:universal\], we show that our categories of quantum functions $\QSet(A,B)$ can be obtained from an analogous universal construction, as the category of finite-dimensional representations of an internal hom $[A,B]$ in the opposite of the category of $C^*$-algebras. Likewise, our categories $\QBij([n],[n])$ of quantum bijections on an $n$-element set are the categories of finite-dimensional representations of the Hopf $C^*$-algebra $A(n)$ corresponding to Wang’s quantum permutation group [@Wang1998] (Proposition \[prop:Wang\]), and our categories $\QIso(G,G)$ of quantum automorphisms of a graph $G$ are the categories of finite-dimensional representations of the Hopf $C^*$-algebra $A(G)$ corresponding to Banica’s quantum automorphism group of the graph [@Banica2007]
Our framework therefore captures the finite-dimensional representation theory of the Hopf $C^*$-algebras considered in noncommutative topology. However, we note that a Tannakian correspondence exists only between these algebras and their categories of corepresentations (or, in the language of quantum group theory, with the representations of their associated compact quantum groups [@Neshveyev2013]. Also see Remark \[rem:problemTannaka\] and Remark \[rem:Wangnotdiscrete\]).
Based on the theory of quantum relations developed by Kuperberg and Weaver [@Kuperberg2012; @Weaver2010; @Weaver2015], Kornell [@Kornell2011; @Kornell2018] defines (possibly infinite) quantum sets and quantum functions between them. His notion of finite quantum set coincides with ours; his quantum functions between finite quantum sets are \*-homomorphisms and thus one-dimensional quantum functions in our sense.
#### Quantum information theory.
Quantum graph homomorphisms were defined by Mančinska and Roberson [@Mancinska2016] as generalisations of quantum graph colourings [@Cameron2007] in the context of nonlocal games (see Remark \[rem:quantumhomMancinska\]) and have been the subject of intensive study in their various forms [@Mancinska2016; @Scarpa2012; @Avis2006; @Cameron2007; @Paulsen2015; @Paulsen2016; @Roberson2016]. Quantum graph isomorphisms are originally due to Atserias et al [@Atserias2016] (see Remark \[rem:Atseriasquantumiso\]).
A related $C^*$-algebraic approach to quantum homomorphisms between classical graphs was recently discovered by Ortiz and Paulsen [@Ortiz2016]. They define a $C^*$-algebra which is essentially an internal hom, analogous to the algebra of quantum functions in Remark \[rem:quantumfunctionalgebra\].
Our notion of reflexive quantum graph coincides with that of Kuperberg and Weaver [@Kuperberg2012; @Weaver2015; @Weaver2010] (see Remark \[rem:literatureqgraph\] and Theorem \[thm:relationadjacency\]) and in particular generalises Duan, Severini and Winter’s noncommutative graphs [@Duan2013] (see Proposition \[prop:quantumgraphDuanthesame\]). Quantum graph homomorphisms between noncommutative graphs can be understood as pure versions of Stahlke’s entanglement-assisted morphisms [@Stahlke2016] (see Remark \[rem:stahlkegraphhoms\]).
#### Categorical quantum mechanics (CQM).
Our work emerges from the CQM research programme, initiated by Abramsky and Coecke [@Abramsky2008] and developed by them and others [@Coecke2008; @Coecke2007; @Coecke2009; @Vicary2010; @Coecke2014], which uses the graphical calculus of monoidal categories to provide a high-level syntax for quantum information flow. In particular, Vicary’s reformulation of the theory of finite-dimensional $C^*$-algebras in terms of Frobenius algebras [@Vicary2010] and Coecke, Pavlovi[ć]{} and Vicary’s proof of Gelfand duality in this setting [@Coecke2009] are important starting points and guide posts for our work (see Section \[sec:background\]). Many of our proofs and constructions can be understood as simple quantisations, in the sense of equation , of constructions from categorical quantum mechanics.
A related 2-categorical framework for quantum theory based on the 2-category $\mathrm{2Hilb}$ of categorified Hilbert spaces was studied by Vicary and the second author [@Vicary2012hq; @Vicary2012_2; @Reutter2016]; we remark that there is a locally faithful 2-functor $\QSet \to \mathrm{2Hilb}$ allowing us to translate most of our results into this setting.
#### Monads.
The 2-category of quantum sets $\QSet$ is an instance of Street’s 2-category of monads [@Street1972] in the $2$-category $\BHilb$, the delooping of the monoidal category $\Hilb$ (see Remark \[rem:street\]). Hinze and Marsden [@Marsden2014; @Hinze2016] give an analogous graphical treatment of the 2-category $\mathrm{Mnd}(\mathrm{Cat})$ of monads in the 2-category $\mathrm{Cat}$ of categories, functors and natural transformations.
Recent work by Abramsky and others [@Abramsky2017] uses monads to study binary constraint systems and quantum graph homomorphisms. There are many similarities between their work and ours; for example, composition in their Kleisli category corresponds to the composition of $1$-morphisms in $\QSet$ (see Remark \[rem:ppmcomposition\]).
Definitions and conventions
---------------------------
We assume some basic familiarity with monoidal category theory [@Selinger2010] and $2$-category theory [@Borceux1994 Chapter 7]. Dagger categories are defined in [@Selinger2010]; strict dagger $2$-categories[^5] and dagger $2$-functors are defined in [@Heunen2016]. When we refer to local properties of a 2-functor (such as local faithfulness), we mean that these properties are true of the induced functors on hom-categories. We use the words *module* and *representation* interchangeably. A projector on a Hilbert space $H$ is an endomorphism $P: H \to H$ which is idempotent and self-adjoint $P = P^{\dagger} = P^2$. All sets appearing in this work are finite and, except where clearly specified, all vector spaces and all algebras are finite-dimensional; a notable exception is Section \[sec:universal\]. Consequently, we use the labels and for the categories of finite sets and finite-dimensional Hilbert spaces respectively. We also take all $C^*$-algebras to be unital. We denote the $n$-element set by $[n]$.
Acknowledgments {#acknowledgments .unnumbered}
---------------
We are especially grateful to Jamie Vicary for many useful conversations and comments on an earlier draft of this paper. We thank Andre Kornell for a useful conversation about our definitions of quantum functions. We also thank Samson Abramsky, and Rui Soares Barbosa for discussing the relationship between our categorical frameworks; Giulio Chiribella for introducing us to existing notions of noncommutative graphs; and Matty Hoban for informative conversations about nonlocal games. We appreciate the helpful comments of an anonymous referee regarding the presentation of these results.
$$\hspace{-2.3cm}
\setlength\arraycolsep{1pt}
\begin{array}{|c|c||c|c|}
\hline
\rule{0pt}{3ex}
\begin{minipage}{1.7cm}\center
\text{set}
\end{minipage}&
\text{commutative algebra} &\text{non-commutative algebra} & \text{quantum set}
\\[-9pt] \rule{0pt}{0ex}&&&\\\hline
\text{element}&
\begin{minipage}{\wl\textwidth}
\begin{calign}\nonumber
\begin{tz}[zx, master, every to/.style={out=up, in=down},xscale=-1]
\draw (0,1) to (0,2) to [out=135] (-0.75,3);
\draw (0,2) to [out=45] (0.75, 3);
\node[zxnode=\zxwhite] at (0,1) {$\psi$};
\node[zxvertex=\zxwhite, zxup] at (0,2) {};
\end{tz}
=~~~
\begin{tz}[zx,slave, every to/.style={out=up, in=down},xscale=-1]
\draw (-0.75,2) to (-0.75,3);
\draw (0.75, 2) to +(0,1.);
\node[zxnode=\zxwhite] at (-0.75,2) {$\psi$};
\node[zxnode=\zxwhite] at (0.75,2) {$\psi$};
\end{tz}
\hspace{1cm}
\begin{tz}[zx, every to/.style={out=up, in=down},xscale=-1]
\draw (0,1) to (0,2) ;
\node[zxnode=\zxwhite] at (0,1) {$\psi$};
\node[zxvertex=\zxwhite, zxup] at (0,2) {};
\end{tz}
~~=~~
\emptydiagram
\\\nonumber
\begin{tz}[zx,slave, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,1.5);
\node[zxnode=\zxwhite] at (0,1.5) {$\psi^\dagger$};
\end{tz}
=~~~
\begin{tz}[zx,slave,every to/.style={out=up, in=down},xscale=-1]
\draw (0,1.5) to (0,2) to [in=left] node[pos=1] (r){} (0.5,2.5) to [out=right, in=up] (1,2) to (1,0);
\node[zxnode=\zxwhite] at (0,1.5) {$\psi$};
\node[zxvertex=\zxwhite,zxdown] at (r.center){};
\end{tz}
\end{calign}
\end{minipage}
\hspace{\bnd}
&
%%%%%%%
\hspace{\bndm}
\begin{minipage}{\wr\textwidth}
\begin{calign}\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2) to [out=45] (0.75,3);
\draw (0,2) to [out=135] (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5](Q) {$Q$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0,2){};
\draw[white,double] (0,0) to (Q.center);
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0.75,1.75) to (0.75,3);
\draw (-0.75,1.75) to (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$Q$} node[zxnode=\zxwhite, pos=0.64] {$Q$}(-1.75,3);
\end{tz}
\hspace{1cm}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2.25);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5](Q) {$Q$} (-1,2.5) to (-1,3);
\node[zxvertex=\zxwhite] at (0,2.25){};
\draw[white,double] (0,0) to (Q.center);
\end{tz}
=
\begin{tz}[zx, every to/.style={out=up, in=down},xscale=-1]
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.,0) to [looseness=0.9] (-1,2.5) to (-1,3);
\end{tz}
\\[-3pt]\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-0.8]
\draw [arrow data={0.2}{>},arrow data={0.8}{>}] (0,0) to (2.25,3);
\draw (2.25,0) to [in=-45] node[zxnode=\zxwhite, pos=1] (Q) {$Q^\dagger$} (1.125,1.5);
\node at (1,-0.5){};
\node at (1,3.5){};
\end{tz}
~=~~
\begin{tz}[zx,xscale=-0.6,yscale=-1]
\draw[arrow data={0.5}{<}] (0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw[arrow data={0.5}{>}] (1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw (1,1) to [out= -45, in= left] node[zxvertex=\zxwhite, pos=1] {} (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$Q$};
\node at (1,-1){};
\node at (1,3){};
\end{tz}
\end{calign}
\end{minipage}
\hspace{\bndr}
&
\begin{minipage}{\wtext}\center quantum \\ element \end{minipage}\\
\hline
\text{function} &
\begin{minipage}{\wl\textwidth}
\begin{calign}\nonumber
\begin{tz}[zx, master, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2) to [out=135] (-0.75,3);
\draw (0,2) to [out=45] (0.75, 3);
\node[zxnode=\zxwhite] at (0,1) {$f$};
\node[zxvertex=\zxwhite, zxup] at (0,2) {};
\end{tz}
=
\begin{tz}[zx, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw (0,0.75) to [out=45] (0.75, 1.75) to +(0,1.25);
\node[zxnode=\zxwhite] at (-0.75,2) {$f$};
\node[zxnode=\zxwhite] at (0.75,2) {$f$};
\node[zxvertex=\zxwhite, zxup] at (0,0.75) {};
\end{tz}
\hspace{1cm}
\begin{tz}[zx,slave, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2) ;
\node[zxnode=\zxwhite] at (0,1) {$f$};
\node[zxvertex=\zxwhite, zxup] at (0,2) {};
\end{tz}
=
\begin{tz}[zx,slave, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75) ;
\node[zxvertex=\zxwhite, zxup] at (0,0.75) {};
\end{tz}
\\\nonumber
\begin{tz}[zx,slave, every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,3);
\node[zxnode=\zxwhite] at (0,1.5) {$f^\dagger$};
\end{tz}
=~~
\begin{tz}[zx,slave,every to/.style={out=up, in=down},xscale=-1]
\draw (0,1.5) to (0,2) to [in=left] node[pos=1] (r){} (0.5,2.5) to [out=right, in=up] (1,2) to (1,0);
\draw (-1,3) to [out=down,in=up] (-1,1) to [out=down, in=left] node[pos=1] (l){} (-0.5,0.5) to [out=right, in=down] (0,1) to (0,1.5);
\node[zxnode=\zxwhite] at (0,1.5) {$f$};
\node[zxvertex=\zxwhite,zxup] at (l.center){};
\node[zxvertex=\zxwhite,zxdown] at (r.center){};
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bnd}
%%%%%
&
\hspace{\bndm}
\begin{minipage}{\wr\textwidth}
\begin{calign}\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2) to [out=45] (0.75,3);
\draw (0,2) to [out=135] (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0,2){};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75) to [out=45] (0.75,1.75) to (0.75,3);
\draw (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$P$} node[zxnode=\zxwhite, pos=0.64] {$P$}(-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\end{tz}
\hspace{1cm}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2.25);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1,2.5) to (-1,3);
\node[zxvertex=\zxwhite] at (0,2.25){};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.,0) to (1.,0.75) to (-1,3);
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\end{tz}
\\[2pt]\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-0.8]
\draw [arrow data={0.2}{>},arrow data={0.8}{>}] (0,0) to (2.25,3);
\draw (2.25,0) to node[zxnode=\zxwhite, pos=0.5] {$P^\dagger$} (0,3);
\end{tz}
=~~
\begin{tz}[zx,xscale=-0.6,yscale=-1]
\draw[arrow data={0.5}{<}] (0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw (1,1) to [out=135, in=right] node[zxvertex=\zxwhite, pos=1]{} (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw[arrow data={0.5}{>}] (1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw (1,1) to [out= -45, in= left] node[zxvertex=\zxwhite, pos=1] {} (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P$};
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bndr}
%%%%%%%
&
\begin{minipage}{\wtext}\center quantum \\ function \end{minipage} \\ \hline
\text{bijection} &
%%%%%
\begin{minipage}{\wl\textwidth}
\vspace{0.2cm}
\hspace{0.1cm}function $+$
\vspace{-0.1cm}
\begin{calign}\nonumber
\begin{tz}[zx, master, every to/.style={out=up, in=down},yscale=-1,xscale=-1]
\draw (0,0) to (0,2) to [out=135] (-0.75,3);
\draw (0,2) to [out=45] (0.75, 3);
\node[zxnode=\zxwhite] at (0,1) {$f$};
\node[zxvertex=\zxwhite, zxdown] at (0,2) {};
\end{tz}
=
\begin{tz}[zx, every to/.style={out=up, in=down},yscale=-1,xscale=-1]
\draw (0,0) to (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw (0,0.75) to [out=45] (0.75, 1.75) to +(0,1.25);
\node[zxnode=\zxwhite] at (-0.75,2) {$f$};
\node[zxnode=\zxwhite] at (0.75,2) {$f$};
\node[zxvertex=\zxwhite, zxdown] at (0,0.75) {};
\end{tz}
\hspace{1cm}
\begin{tz}[zx,slave, every to/.style={out=up, in=down},yscale=-1,xscale=-1]
\draw (0,0) to (0,2) ;
\node[zxnode=\zxwhite] at (0,1) {$f$};
\node[zxvertex=\zxwhite, zxup] at (0,2) {};
\end{tz}
=
\begin{tz}[zx,slave, every to/.style={out=up, in=down},yscale=-1,xscale=-1]
\draw (0,0) to (0,0.75) ;
\node[zxvertex=\zxwhite, zxup] at (0,0.75) {};
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bnd}
&
%%%%%%%
\hspace{\bndm}
\begin{minipage}{\wr\textwidth}
\vspace{0.2cm}
quantum function $+$
\vspace{-0.1cm}
\begin{calign}\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,2) to [out=45] (0.75,3);
\draw (0,2) to [out=135] (-0.75,3);
\draw[arrow data={0.2}{<}, arrow data={0.8}{<}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxdown] at (0,2){};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,0.75) to [out=45] (0.75,1.75) to (0.75,3);
\draw (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw[arrow data={0.2}{<}, arrow data={0.9}{<}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$P$} node[zxnode=\zxwhite, pos=0.64] {$P$}(-1.75,3);
\node[zxvertex=\zxwhite, zxdown] at (0,0.75){};
\end{tz}
\hspace{1cm}
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,2.25);
\draw[arrow data={0.2}{<}, arrow data={0.8}{<}] (1,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1,2.5) to (-1,3);
\node[zxvertex=\zxwhite] at (0,2.25){};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,0.75);
\draw[arrow data={0.2}{<}, arrow data={0.9}{<}] (1.,0) to (1.,0.75) to (-1,3);
\node[zxvertex=\zxwhite] at (0,0.75){};
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bndr}
&
\begin{minipage}{\wtext}\center quantum \\ bijection \end{minipage} \\
\hline \nonumber
\begin{minipage}{1.5cm}\center graph \\homom.\end{minipage}&
%%%%%%%%
\begin{minipage}{\wl\textwidth}
\vspace{0.2cm}
\hspace{0.01cm}function $+$
\vspace{-0.2cm}
\def\d{2}
\begin{calign}\nonumber
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.5] {$f$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.5] {$f$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\end{tz}
~~=~~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.5] {$f$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.5] {$f$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$H$};
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bnd}
&
%%%%%%%%%%%%
\hspace{\bndm}
\begin{minipage}{\wr\textwidth}
\vspace{0.2cm}
\hspace{-0.36cm}quantum function $+$
\vspace{-0.2cm}
\def\d{2}
\begin{calign}\nonumber
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string, arrow data={0.08}{>},arrow data={0.5}{>}, arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
~=~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$H$};
\draw[string, arrow data={0.08}{>}, arrow data ={0.5}{>},arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
\end{calign}\\[-15pt]
\end{minipage}
\hspace{\bndr}
&
\begin{minipage}{\wtext}\center quantum \\ graph\\homom. \end{minipage}\\
\hline\nonumber
\begin{minipage}{1.5cm}\center graph \\iso.\end{minipage} &
%%%%%
\begin{minipage}{\wl\textwidth}
\vspace{0.2cm}
\hspace{-0.0cm}bijection $+$
\vspace{-0.1cm}
\begin{calign}\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1,xscale=-1]
\draw (0,0) to (0,3);
\path (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$}(-1.75,3);
\node[zxnode=\zxwhite] at (0,0.9){$G$};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1,xscale=-1]
\draw (0,0) to (0,3);
\path (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$H$}(-1.75,3);
\node[zxnode=\zxwhite] at (0,0.9){$P$};
\end{tz}
\end{calign}\\[-25pt]
\end{minipage}
\hspace{\bnd}
&
%%%%%%%
\hspace{\bndm}
\begin{minipage}{\wr\textwidth}
\vspace{0.2cm}
\hspace{-0.35cm}quantum bijection $+$
\vspace{-0.1cm}
\begin{calign}\nonumber
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1,xscale=-1]
\draw (0,0) to (0,3);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$}(-1.75,3);
\node[zxnode=\zxwhite] at (0,0.9){$G$};
\end{tz}
\quad =\quad
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1,xscale=-1]
\draw (0,0) to (0,3);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxnode=\zxwhite] at (0,2.35) {$H$};
\end{tz}
\end{calign}\\[-25pt]
\end{minipage}
\hspace{\bndr}
&
\begin{minipage}{\wtext}\begin{center} quantum \\ graph\\iso.\end{center} \end{minipage}\\ \nonumber&\nonumber&&\\ \hline\end{array}$$
Background {#sec:background}
==========
Gelfand duality
---------------
Our approach to defining a theory of quantum sets and quantum functions is based on *Gelfand duality*.
\[thm:gelfandduality\] The category of commutative $C^*$-algebras and -morphisms is equivalent to the opposite of the category of compact Hausdorff spaces and continuous functions.
The equivalence takes a space to the algebra of continuous complex-valued functions on that space; in the other direction, a $C^*$-algebra is taken to its spectrum. In some sense, therefore, one may consider the theory of compact Hausdorff spaces — classical topology — to be the theory of commutative $C^*$-algebras.
The idea of noncommutative geometry is to consider *noncommutative* in light of Gelfand duality, thus enabling us to study the ‘noncommutative’ or ‘quantum’ spaces to which they would be dual. In this paper, we consider only finite discrete spaces and finite-dimensional $C^*$-algebras. In this case, Theorem \[thm:gelfandduality\] reduces to the following statement.
\[cor:finiteGelfandduality\] The category of finite-dimensional commutative $C^*$-algebras and $*$-homomorphisms is equivalent to the opposite of the category of finite sets and functions.
The string diagram calculus
---------------------------
In order to investigate finite Gelfand duality and its noncommutative generalisation, we make use of the string diagram calculus for monoidal categories. This calculus is well established and has been treated in detail elsewhere [@Joyal1991a; @Joyal1991b; @Selinger2010; @Coecke2010]; here we only provide a brief and informal introduction. We remark that this calculus is quite general. Although we only consider the graphical calculus of , the category of finite-dimensional Hilbert spaces and linear maps, most of what we prove holds in the general setting of dagger compact categories [@Kelly1972; @Kelly1980; @Abramsky2004].
In the string diagram calculus
, wires correspond to finite-dimensional Hilbert spaces and boxes correspond to linear maps.
We read diagrams from bottom to top.
Composition and tensor product are depicted as follows:
(0,0) to (0,3.5); at (0,1.05) [$f$]{}; at (0,2.45) [$g$]{}; at (0,0)[$V_1$]{}; at (0, 1.75) [$V_2$]{}; at (0, 3.5) [$V_3$]{};
&
(0,0) to (0,3.5); (1.5,0) to (1.5,3.5); at (0,1.75) [$f$]{}; at (1.5,1.75) [$g$]{}; at (0,0)[$V_1$]{}; at (1.5, 0) [$V_3$]{}; at (0, 3.5) [$V_2$]{}; at (1.5, 3.5) [$V_4$]{};
\
gf:V\_1V\_3 & fg: V\_1V\_3V\_2V\_4
All finite-dimensional Hilbert spaces $V$ have dual spaces $V^*=\Hom(V,\mathbb{C})$, represented in the graphical calculus as an oriented wire with the opposite orientation as $V$. Duality is characterized by the following linear maps, here called *cups and caps*:
\[eq:cupscapsHilb\]
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ; at (2.05,0) [$\pv V$]{}; at (0,0) [$V^*$]{};
&
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ; at (2.0,0) [$\pv V$]{}; at (0,0) [$V^*$]{};
&
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ; at (2.0,0) [$\pv V$]{}; at (0,0) [$V^*$]{};
&
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ; at (2.05,0) [$\pv V$]{}; at (0,0) [$V^*$]{};
\
fv f(v) & 1\_V & vff(v) & 1\_V
To define the second and fourth map, we have identified $V\otimes V^* \cong V^*\otimes V \cong\End(V)$. It may be verified that these maps fulfill the following *snake equations*:
\[eq:snake\]
(0,0) to (0,1) to \[out=up, in=up, looseness=2\] (1,1) to \[out=down, in=down, looseness=2\] (2,1) to (2,2);
=
(0,0) to (0,2);
=
(0,0) to (0,1) to \[out=up, in=up, looseness=2\] (1,1) to \[out=down, in=down, looseness=2\] (2,1) to (2,2);
&
(0,0) to (0,1) to \[out=up, in=up, looseness=2\] (1,1) to \[out=down, in=down, looseness=2\] (2,1) to (2,2);
=
(0,0) to (0,2);
=
(0,0) to (0,1) to \[out=up, in=up, looseness=2\] (1,1) to \[out=down, in=down, looseness=2\] (2,1) to (2,2);
Together with the swap map $\sigma_{V,W}:\ignore{V\otimes W\to W\otimes V$,~$}v\otimes w\mapsto w\otimes v$, depicted as a crossing of wires, this leads to an extremely flexible topological calculus, allowing us to untangle arbitrary diagrams and straighten out any twists:
&
(0,0) to \[out=up, in=right\] (-1.25, 2.25) to \[out=left, in=left, looseness=1\] (-1.25,0.75) to \[out=right, in=down\] (0,3);
=
(0,0) to (0,3);
=
(0,0) to \[out=up, in=right\] (-1.25, 2.25) to \[out=left, in=left, looseness=1\] (-1.25,0.75) to \[out=right, in=down\] (0,3);
Given a linear map $f:V\to W$ between Hilbert spaces, we can express its adjoint as a reflection of the corresponding diagram across a horizontal axis. This is justified, since the following holds:
(
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ;
)\^ =
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ;
& (
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ;
)\^ =
(0,0) to \[out=up, in=up, looseness=2.5\] (2,0) ;
The following generalisation will be important in what follows.
\[def:daggerduality\]Let $V$ and $W$ be Hilbert spaces. A *dagger duality* between $V$ and $W$ is given by linear maps $\epsilon: W\otimes V \to \mathbb{C}$ and $\eta: \mathbb{C} \to V\otimes W$ fulfilling the snake equations and such that the following holds: $$\label{eq:daggerduality}
\begin{tz}[zx]
\draw (0,0) to (0,1.25);
\draw (1.5,0) to (1.5,1.25);
\draw (-0.2,0) rectangle (1.7, -0.8);
\node[scale=0.8] at (0.75,-0.4) {$\epsilon^\dagger$};
\node[dimension, left] at (0,1.25){$W$};
\node[dimension, right] at (1.5,1.25) {$V$};
\end{tz}
~~~=~~~
\begin{tz}[zx]
\draw (0,0) to [out=up, in=down] (1.5,1.25);
\draw (1.5,0) to [out=up, in=down](0,1.25);
\draw (-0.2,0) rectangle (1.7, -0.8);
\node[scale=0.8] at (0.75,-0.4) {$\eta$};
\node[dimension, left] at (0,1.25){$W$};
\node[dimension, right] at (1.5,1.25) {$V$};
\end{tz}$$
The cups and caps defined in are dagger duals. Dagger dualities are unique up to a unique unitary map [@Selinger2007 Section 7], meaning that if $V$ and $W$ are dagger dual, then there is a unitary map $U: V^*\to W$ such that the following holds:
(-0.6, 1.5) rectangle (2.1, -1.1); (0,0) to (0,1.25); (1.5,0) to (1.5,1.25); (-0.2,0) rectangle (1.7, -0.8); at (0.75,-0.4) [$\eta$]{}; at (0,1.25)[$V$]{}; at (1.5,1.25) [$W$]{};
=
(-0.6, 1.5) rectangle (2.1, -1.1); (0,1.25) to (0,0.25) to \[out=down, in=down, looseness=2.5\] (1.5,0.25) to (1.5,1.25); at (1.5, 0.5) [$U$]{}; at (0,1.25)[$V$]{}; at (1.5,1.25) [$W$]{};
&
(-0.6, 1.5) rectangle (2.1, -1.1); (0,0) to (0,1.25); (1.5,0) to (1.5,1.25); (-0.2,0) rectangle (1.7, -0.8); at (0.75,-0.4) [$\epsilon$]{}; at (0,1.25)[$W$]{}; at (1.5,1.25) [$V$]{};
=
(-0.6, 1.5) rectangle (2.1, -1.1); (0,1.25) to (0,0.25) to \[out=down, in=down, looseness=2.5\] (1.5,0.25) to (1.5,1.25); at (0, 0.5) [$U^\dagger$]{}; at (0,1.25)[$W$]{}; at (1.5,1.25) [$V$]{};
Diagrams for $C^*$-algebras {#sec:diagramsforC*}
---------------------------
Following Vicary [@Vicary2010], we define $C^*$-algebras in a categorical manner, as dagger Frobenius algebras in the category $\Hilb$.
Of course, algebras can be defined on any vector space, but in order to discuss $C^*$-algebras we require the inner product.
We will refrain from drawing an orientation on the wire corresponding to the Hilbert space on which the algebra is defined, for reasons which will soon become apparent.
\[def:algebra\]An *algebra* is a Hilbert space $H$ with a multiplication and a unit map, depicted as follows:
\(A) at (0,0); (0.75,1) to (0.75,2);
&
\(A) at (0.75,2);
\
m:HH H& u: H
These maps satisfy the following associativity and unitality equations:
\[eq:assocandunitality\]
\(A) at (0.25,0); (1,1) to \[out=up, in=-135\] (1.75,2); (1.75,2) to \[out=-45, in=up\] (3.25,0); (1.75,2) to (1.75,3); at (1.75,2);
=
\(A) at (0.25,0); (1,1) to \[out=up, in=-135\] (1.75,2); (1.75,2) to \[out=-45, in=up\] (3.25,0); (1.75,2) to (1.75,3); at (1.75,2);
&
\(A) at (0,0); (0,-0.25) to (0,0); (0.75,1) to (0.75,2); at (1.5,0);
=
(0,0) to (0,2);
=
\(A) at (0,0); (0,-0.25) to (0,0); (0.75,1) to (0.75,2); at (1.5,0);
Analogously, a *coalgebra* is a Hilbert space $H$ with a coassociative comultiplication $\delta: H \to H\otimes H$ and a counit $\epsilon:H\to \mathbb{C}$. The adjoint of an algebra is a coalgebra.
Note that for the multiplication and unit maps of an algebra we simply draw white nodes rather than labelled boxes, for concision. Likewise, we draw the comultiplication and counit maps of the adjoint coalgebra as white nodes. Despite having the same label in the diagram, they can be easily distinguished by their type.
A *dagger Frobenius algebra* is an algebra where the algebra and adjoint coalgebra structures are related by the following Frobenius equation: $$\label{eq:Frobenius}
\begin{tz}[zx]
\draw (0,0) to [out=up, in=-135] (0.75,2) to (0.75,3);
\draw (0.75,2) to [out=-45, in=135] (2.25,1);
\draw (2.25,0) to (2.25,1) to [out=45, in=down] (3,3);
\node[zxvertex=\zxwhite,zxup] at (2.25,1){};
\node[zxvertex=\zxwhite,zxdown] at (0.75,2){};
\end{tz}
\quad = \quad
\begin{tz}[zx]
\coordinate (A) at (0,0);
\coordinate (B) at (0,3);
\draw (0.75,1) to (0.75,2);
\mult{A}{1.5}{1}
\comult{B}{1.5}{1}
\end{tz}
\quad = \quad
\begin{tz}[zx]
\draw (0,0) to [out=up, in=-135] (0.75,2) to (0.75,3);
\draw (0.75,2) to [out=-45, in=135] (2.25,1);
\draw (2.25,0) to (2.25,1) to [out=45, in=down] (3,3);
\node[zxvertex=\zxwhite,zxup] at (2.25,1){};
\node[zxvertex=\zxwhite,zxdown] at (0.75,2){};
\end{tz}$$ A Frobenius algebra is *special*, *symmetric* or *commutative* if one of the following additional equations holds:
\[eq:special\]
(0,0) to (0,1) to \[out=135\] (-0.75,2) to \[in=-135\] (0,3) to (0,4); (0,1) to \[out=45\] (0.75,2) to \[in=-45\] (0,3); at (0,1); at (0,3);
=
(0,0) to +(0,4);
&
(0.25,-1.5) to (1.75,0) to \[in=-45\] (1,1) to (1,1.5); (1.75,-1.5) to (0.25,0) to \[in=-135\] (1,1); at (1,1); at (1,1.5);
=
(1.75,-1.5) to (1.75,0) to \[in=-45\] (1,1) to (1,1.5); (0.25,-1.5) to (0.25,0) to \[in=-135\] (1,1); at (1,1); at (1,1.5);
&
(0.25,-1.5) to (1.75,0) to \[in=-45\] (1,1) to (1,2); (1.75,-1.5) to (0.25,0) to \[in=-135\] (1,1); at (1,1);
=
(1.75,-1.5) to (1.75,0) to \[in=-45\] (1,1) to (1,2); (0.25,-1.5) to (0.25,0) to \[in=-135\] (1,1); at (1,1);
\
&&
In this work, we will focus on special symmetric and special commutative dagger Frobenius algebras, which we abbreviate as s and s, respectively.
Frobenius algebras are closely related to dualities. In particular, it is a direct consequence of and that the following cups and caps fulfill the snake equations :
\[eq:cupcapfrob\]
(-0.1,0) rectangle (2.1,2.); (0,0) to \[out=up, in=up, looseness=2\] node\[zxvertex=, pos=0.5\] (2,0);
:=
(-0.1,0) rectangle (2.1,2.2); (0,0) to \[out=up, in=up, looseness=2\] node\[front,zxvertex=, pos=0.5\](A) (2,0); (A.center) to (1,1.8); at (1,1.8) ;
&
(-0.1,0) rectangle (2.1,2.); (0,0) to \[out=up, in=up, looseness=2\] node\[zxvertex=, pos=0.5\] (2,0);
:=
(-0.1,0) rectangle (2.1,2.2); (0,0) to \[out=up, in=up, looseness=2\] node\[front,zxvertex=, pos=0.5\](A) (2,0); (A.center) to (1,1.8); at (1,1.8) ;
It follows that every Frobenius algebra is canonically self-dual, $A^*\cong A$; we therefore do not need to draw an orientation on the corresponding wire.
A major reason for defining these structures is the fact that $\F$s coincide with finite-dimensional $C^*$-algebras.
Every finite-dimensional $C^*$-algebra has an inner product making it into a special symmetric dagger Frobenius algebra. Conversely, every $\F$ $A$ admits a norm such that the canonical involution, defined by its action on vectors $\ket{a}\in A$ as the following antihomomorphism, endows it with the structure of a $C^*$-algebra:
(0.25,3) to (0.25,1); at (0.25,1) [$a$]{};
(0.25,3) to (0.25,2) to \[out=down, in=135\] (1,1) to \[out=45, in=down\] (1.75,2); (1,1) to (1,0.5); at (1.75,2) [$a^\dagger$]{}; at (1,1); at (1,0.5);
\
One advantage of explicitly using $\F$s over $C^*$-algebras is that $\F$s already contain ‘up-front’ all emergent structures of finite-dimensional $C^*$-algebras, such as the comultiplication $\Delta = m^\dagger : H\to H \otimes H$; they are therefore more amenable to the purely compositional reasoning of the graphical calculus. Notions from the theory of , such as $*$-homomorphisms, can be reformulated in the language of s.=-2
\[def:starhoms\]A *$*$-homomorphism* between s $A$ and $B$ is a linear map $f:A\to B$ satisfying the following equations:
\[eq:homo\]
(0,0) to (0,2) to \[out=135\] (-0.75,3); (0,2) to \[out=45\] (0.75, 3); at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (0,0.75) to \[out=45\] (0.75, 1.75) to +(0,1.25); at (-0.75,2) [$f$]{}; at (0.75,2) [$f$]{}; at (0,0.75) ;
&
(0,0) to (0,2) ; at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) ; at (0,0.75) ;
&
(0,0) to (0,3); at (0,1.5) [$f^\dagger$]{};
=
(0,1.5) to (0,2) to \[in=left\] node\[pos=1\] (r) (0.5,2.5) to \[out=right, in=up\] (1,2) to \[out=down, in=up\] (1,0); (-1,3) to \[out=down,in=up\] (-1,1) to \[out=down, in=left\] node\[pos=1\] (l) (-0.5,0.5) to \[out=right, in=down\] (0,1) to (0,1.5); at (0,1.5) [$f$]{}; at (l.center); at (r.center);
A *$*$-cohomomorphism* is a linear map $f:A\to B$ satisfying the following equations:
\[eq:cohomo\]
(0,0) to (0,2) to \[out=135\] (-0.75,3); (0,2) to \[out=45\] (0.75, 3); at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (0,0.75) to \[out=45\] (0.75, 1.75) to +(0,1.25); at (-0.75,2) [$f$]{}; at (0.75,2) [$f$]{}; at (0,0.75) ;
&
(0,0) to (0,2) ; at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) ; at (0,0.75) ;
&
(0,0) to (0,3); at (0,1.5) [$f^\dagger$]{};
=
(0,1.5) to (0,2) to \[in=left\] node\[pos=1\] (r) (0.5,2.5) to \[out=right, in=up\] (1,2) to \[out=down, in=up\] (1,0); (-1,3) to \[out=down,in=up\] (-1,1) to \[out=down, in=left\] node\[pos=1\] (l) (-0.5,0.5) to \[out=right, in=down\] (0,1) to (0,1.5); at (0,1.5) [$f$]{}; at (l.center); at (r.center);
A *$*$-isomorphism* is a linear map $f:A\to B$ which is both a $*$-homomorphism and a $*$-cohomomorphism.
Observe that the adjoint of a $*$-homomorphism is a $*$-cohomomorphism, that every [$*$-isomorphism]{} is unitary, and that every unitary $*$-homomorphism of s is a In particular, a $*$-isomorphism is precisely an invertible \*-homomorphism (see Proposition \[prop:Frobeniusinvertible\] for the converse).
The notion of a $*$-homomorphism between $\F$s coincides with the notion of a $*$-homomorphism between finite-dimensional
Finite-dimensional Gelfand duality in diagrams {#sec:gelfanddiagram}
----------------------------------------------
Having established the graphical calculus and the correspondence between finite-\
dimensional $C^*$-algebras and $\F$s, we now recall the graphical version of finite-dimensional Gelfand duality in the framework established by Coecke, Pavlovi[ć]{} and Vicary [@Coecke2009]. We first observe that every orthonormal basis on a Hilbert space $H$ defines a special commutative dagger Frobenius algebra on $H$.
\[exm:Frob\]Let $\left\{\ket{i}\right\}_{1\leq i\leq n}$ be an orthonormal basis of a Hilbert space $H$. Then the following multiplication and unit maps, together with their adjoints, form a special commutative dagger Frobenius algebra on $H$:
\[eq:classicalcopy\]
\(A) at (0,0); (0.75,1) to (0.75,2);
:= \_[i=1]{}\^[n]{}
(0,0) to (0,0.5); (1.5,0) to (1.5,0.5); (0.75,1.5) to (0.75,2); at (0,0.5) [$i^\dagger$]{}; at (1.5,0.5) [$i^\dagger$]{}; at (0.75,1.5) [$i$]{};
&
(0.75,1) to (0.75,2); at (0.75,1);
:= \_[i=1]{}\^[n]{}
(0.75,1.25) to (0.75,2); at (0.75,1.25) [$i$]{};
\
m: \_[i,j]{} & u: 1 \_[i=1]{}\^[n]{}
Conversely, every special commutative dagger Frobenius algebra $A$ gives rise to an orthonormal basis of $A$; the basis vectors are given by the copyable element of $A$, defined as follows:
\[def:copyablestates\] A *copyable element* of a $\CF$ $A$ is a $*$-cohomomorphism $\psi: \mathbb{C} \to A$; that is, a vector $\ket{\psi} \in A$ such that the following hold[^6]:
\[eq:ordinaryelement\]
(0,1) to (0,2) to \[out=135\] (-0.75,3); (0,2) to \[out=45\] (0.75, 3); at (0,1) [$\psi$]{}; at (0,2) ;
=
(-0.75,2) to (-0.75,3); (0.75, 2) to +(0,1.); at (-0.75,2) [$\psi$]{}; at (0.75,2) [$\psi$]{};
&
(0,1) to (0,2) ; at (0,1) [$\psi$]{}; at (0,2) ;
= &
(0,0) to (0,1.5); at (0,1.5) [$\psi^\dagger$]{};
=
(0,1.5) to (0,2) to \[in=left\] node\[pos=1\] (r) (0.5,2.5) to \[out=right, in=up\] (1,2) to \[out=down, in=up\] (1,0); at (0,1.5) [$\psi$]{}; at (r.center);
\
These copyable elements indeed form an orthonormal basis of the Hilbert space underlying $A$.
\[thm:classificationONB\]The copyable elements of a special commutative dagger Frobenius algebra $A$ form an orthonormal basis of $A$ for which the algebra is of the form given in Example \[exm:Frob\].
In other words, every special dagger commutative Frobenius algebra is of the form for some orthonormal basis on a Hilbert space.
Given a $\CF$ $A$, we denote its set of copyable elements by $\widehat{A}$. For $\CF$s $A$ and $B$, it can easily be verified that every function $\widehat{A}\to \widehat{B}$ gives rise to a $*$-cohomomorphism between $A$ and $B$ and that conversely every $*$-cohomomorphism $A\to B$ comes from such a function $\widehat{A} \to \widehat{B}$. Therefore, Theorem \[thm:classificationONB\] gives rise to the following Frobenius algebraic version of finite Gelfand duality:
The category of commutative special dagger Frobenius algebras and $*$-cohomomorphisms[^7] is equivalent to the category of finite sets and functions.
Explicitly, this equivalence maps a $\CF$ $A$ to its set of copyable elements $\widehat{A}$ and a set $X$ to the algebra associated to the orthonormal basis $\{ \ket{x}~|~x\in X\}$ of the Hilbert space $\mathbb{C}^{|X|}$. Under this correspondence, we may therefore consider the category of finite sets as ‘contained within $\Hilb$’ using the following identification.
$\Set$ $\mathrm{Hilb}$
----------------------------- -----------------------------------
set of cardinality $n$ of dimension $n$
elements of the set copyable states of the
functions $*$-cohomomorphisms
bijections $*$-isomorphisms
the one element set $\{*\}$ the one-dimensional $\mathbb{C}$
\[not:setalgebra\] Throughout this paper, we will take pairs of words in this table to be synonymous. In particular, we will denote a set and its corresponding commutative algebra by the same symbol. It will always be clear from context whether we refer to the set $X$ or the algebra $X$.
The fact that the category of finite sets and functions can be faithfully embedded into $\Hilb$ will be central to our quantisation procedure described in the introduction. We formalise this important concept by adopting terminology initially used by Freyd and developed by Ad[á]{}mek and others [@Adamek1990; @Freyd1970].
\[def:Hilbcategory\] A *concrete dagger category* is a pair $(\mathcal{C}, F)$, where $\mathcal{C}$ is a dagger category and $F:\mathcal{C} \to \Hilb$ is a faithful dagger functor, which we refer to as the *forgetful functor*.
In other words, Gelfand duality allows us to treat the category of finite sets and functions as a concrete dagger category.
\[phil:concretedagger\]All categories in this paper are concrete dagger categories, with a forgetful functor specifying an underlying Hilbert space for every object and an underlying linear map for every morphism. The quantum mechanical interpretation of our categories depends on this forgetful functor. We will formulate all categorical concepts in terms of concrete dagger categories, and ensure that they are compatible with the forgetful functor to .
Quantum sets and quantum functions {#sec:quantumset}
==================================
The fundamental idea of noncommutative topology is to generalise the correspondence between spaces and commutative algebras by considering noncommutative algebras in light of Gelfand duality. We will now begin our exploration of the world of finite-dimensional noncommutative algebras, or ‘finite quantum sets’.
By analogy with Gelfand duality, we think of a special symmetric dagger Frobenius algebra as being associated to an imagined finite *quantum set*, just as a commutative special dagger Frobenius algebra is associated to a finite set. We follow Terminology \[not:setalgebra\] and Wang [@Wang1998 page 3] in denoting both the algebra and its associated imagined quantum set by the same symbol.
Quantum elements {#sec:qelem}
----------------
A set is completely determined by its elements. What is the appropriate notion of a *quantum element* of a quantum set? In particular, is there a notion of quantum element such that a quantum set is completely determined by its quantum elements? Copyable states cannot play this role, since it can easily be verified that matrix algebras have no copyable states whatsoever. Adopting the ideas outlined in the introduction, we make the following definition, ‘quantising’ the notion of a copyable state .
\[def:quantumelement\] A *quantum element* of a quantum set $A$ is a pair ($H,Q$), where $H$ is a Hilbert space and $Q: H\to A \otimes H$ is a linear map satisfying the following equations:
\[eq:quantumelement\]
(0,0) to (0,2) to \[out=45\] (0.75,3); (0,2) to \[out=135\] (-0.75,3); (1.75,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\](Q) [$Q$]{} (-1.75,2.5) to (-1.75,3); at (0,2); (0,0) to (Q.center);
=
(0.75,1.75) to (0.75,3); (-0.75,1.75) to (-0.75,3); (1.75,0) to (1.75,0.75) to \[looseness=1.1, in looseness=0.9\] node\[zxnode=, pos=0.36\] [$Q$]{} node\[zxnode=, pos=0.64\] [$Q$]{}(-1.75,3);
&
(0,0) to (0,2.25); (1,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\](Q) [$Q$]{} (-1,2.5) to (-1,3); at (0,2.25); (0,0) to (Q.center);
=
(1.,0) to \[looseness=0.9\] (-1,2.5) to (-1,3);
&
(0,0) to (2.25,3); (2.25,0) to \[in=-45\] node\[zxnode=, pos=1\] (Q) [$Q^\dagger$]{} (1.125,1.5);
=
(0.25,-0.5) to (0.25,0) to \[out=up, in=-135\] (1,1); (1.75,2.5) to (1.75,2) to \[out=down, in=45\] (1,1); (1,1) to \[out= -45, in= left\] node\[zxvertex=, pos=1\] (2.3,0.3) to \[out=right, in=down\] (3.25,1) to (3.25,2.5); at (1,1) [$Q$]{};
\[phil:quantumprooffromclassical\] The axioms defining a quantum element look just like the axioms defining an ordinary element with an additional oriented wire. This will be a guiding principle for the graphical calculus in this work. Many of the calculations presented here were derived in the ‘classical’ setting with an additional oriented wire only added later. Here, and in what follows, we always draw this additional wire with an orientation, while we draw the original wire — carrying a Frobenius algebra structure — without orientation (see also the discussion after ).
We have drawn the Hilbert space wires from the bottom left to the top right. This is just a convention, and we could equally have defined quantum elements — and later quantum functions — using the opposite convention.
We will show in Proposition \[prop:reconstruct\] that every quantum set is completely determined by its quantum elements, thereby justifying Definition \[def:quantumelement\].
\[rem:operationalinterpretationofquantumels\] If $X$ is an ordinary set (that is, a $\CF$), then it follows that a quantum element of $X$ is a projective measurement with outcomes in $X$ (see Corollary \[cor:quantumelementmeasurement\]). The first diagram of corresponds to orthogonality and idempotency; the second to completeness; and the third to self-adjointness. This diagrammatic representation of projective measurements has been known at least since the work of Coecke and Pavlovi[ć]{} [@Coecke2007]. A direct operational intepretation of has recently been
\[rem:interpretationquantumelement\] A (non-probabilistic, discrete) classical observable can be thought of both as a process $x:\{*\} \to \{*\} \times X$ producing an element of a set $X$ from a trivial system — that is, a process picking out one and only one element — or simply as an element $x\in X$.=-2
Similarly, a quantum observable can either be thought of as projective measurement $P:H\to X\otimes H$ producing an element of $X$ from an underlying quantum mechanical system $H$ or simply as a *quantum element* $P\in_Q X$, shifting attention away from the underlying Hilbert space. This is similar in spirit to random variables in probability theory which are defined as functions $x$ from an underlying probability space to a set $X$, but are usually thought of as random elements of this set $x\in_R X$.
This perspective is reflected in our quantisation approach: By appending a Hilbert space wire to a diagram defining some set-theoretic concept, we retain any operational interpretation, only now allowing the relevant processes to make use of an underlying quantum mechanical system. This is similar to quantisation in the sense of nonlocal games, where classical concepts are quantised by formulating them as strategies for multi-player games and allowing the use of an additional shared quantum resource [@Atserias2016]. As we will see in Section \[sec:quantumperm\] and \[sec:qgt\], our approach to quantisation indeed leads to the same concepts as those appearing in the study of such nonlocal games.
Due to the presence of the additional Hilbert space wire introduced in our quantisation procedure, we are led to consider maps on this Hilbert space, which interact in a particular way with quantum elements.
An *intertwiner* of quantum elements $(H,Q) \to (H', Q')$ is a linear map $f:H\to H'$ such that the following holds:$$\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (1,2)to [out=45] (2,3);
\draw[arrow data={0.35}{>}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f$} (2,1);
\draw[string,arrow data={0.9}{>},arrow data={0.26}{>}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$Q'$} (0,3);
\end{tz}
\quad = \quad
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,1) to [in=-135] (1,2);
\draw[arrow data={0.35}{<}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f$} (2,1);
\draw[string,arrow data={0.9}{<},arrow data={0.26}{<}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$Q$} (0,3);
\end{tz}$$
Note that, in the case of classical elements , all linear maps $\C \to \C$ are scalars and commute trivially with the element, providing us with no information about its structure; intertwiners therefore only become relevant in the quantum setting.
Let $Q$ be a quantum element of a classical set $X$; in other words, a projective measurement with outcomes in $X$ (see Remark \[rem:operationalinterpretationofquantumels\]). In this case, examples of intertwiners $Q\to Q$ are given by projectors onto subspaces which are left undisturbed by the measurement. We will show in Section \[sec:semisimple\] that all intertwiners between quantum elements may be understood in this way.
As we will see, much of the structural difference between the classical and the quantum setting can be understood in terms of the existence or non-existence of such intertwiners. While ordinary elements form a set, quantum elements should properly be organized into a category, to keep track of the intertwiners.
\[def:intertwinerofelements\]For a quantum set $A$, we define the category (A) of quantum elements of $A$:
- **objects** are quantum elements $(H,Q)$ of $A$;
- **morphisms** $(H,Q) \to (H', Q')$ are intertwiners of quantum elements.
Composition of intertwiners is ordinary composition of linear maps.
Every category of quantum elements comes with a forgetful functor $F:\QEl(A) \to \Hilb$ mapping a quantum element to its underlying Hilbert space and an intertwiner to the underlying linear map. This underlying structure makes $(\QEl(A),F)$ a concrete dagger category as in Definition \[def:Hilbcategory\].
We now justify our definition of quantum elements by showing that a quantum set is completely determined by its concrete dagger category of quantum elements.
\[prop:reconstruct\] Up to isomorphism, a quantum set $A$ can be reconstructed from its category of elements $\QEl(A)$ and the forgetful functor $F:\QEl(A)\to \Hilb$.
By definition, $\QEl(A)$ is the category of left comodules of the special symmetric dagger Frobenius algebra $A$ and as such equivalent to the category of modules of $A$. The proposition therefore follows from existing results on Tannaka duality, which states that a semisimple algebra can be reconstructed from its category of modules and a forgetful (or fibre) functor (cf. [@Joyal1991]).
Explicitly, we can reconstruct the algebra $A$ as follows:
\[eqn:quantumcopy\]
(0.25,-0.25) to (0.25,0) to \[out=up, in=-135\] (1,1) to (1,2.25); (1.75,-0.25) to (1.75,0) to \[out=up, in=-45\] (1,1); at (1,1) ;
= \_[i]{}
(0.25,-0.25) to node\[zxnode=, pos=0.46\] [$i^\dagger$]{}(0.25,1); (1.75,-0.25) to node\[zxnode=, pos=0.82\] [$i^\dagger$]{} (1.75,1); (1,1) to node\[zxnode=, pos=0.35\] [$i$]{} (1,2.25); (-0.5,1) to \[out=down, in=-135\] (-0.05,0.05) to \[in looseness=0.3,out=45, in=down\] (2.25,1.) to \[out looseness=0.3,out=up, in=-45\] (-0.05,1.95) to \[out=135, in=up\] (-0.5,1);
&
(1,1) to (1,2.25); at (1,1);
= \_i
(1,1) to node\[zxnode=, pos=0\] [$i$]{} (1,2.25); (1,1) to \[looseness=3.5, out=135, in=135\] (0.25,0.25) to \[out=-45, in=-45, looseness=3.5\] (1,1);
Here the sum ranges over the *simple quantum elements*, which we define in Section \[sec:semisimple\]. The equations can be understood as a quantisation of equations .
Quantum functions {#sec:funct}
-----------------
Having defined quantum elements of quantum sets, we now consider the appropriate notion of quantum functions between them. We define these by quantisation of
\[def:quantumfunction\] A *quantum function* between quantum sets $A$ and $B$ is a pair ($H,P$), where $H$ is a Hilbert space and $P$ is a linear map $H\otimes A\! \to\! B \otimes H$ satisfying the =-2
\[eq:quantumfunction\]
(0,0) to (0,2) to \[out=45\] (0.75,3); (0,2) to \[out=135\] (-0.75,3); (1.75,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{} (-1.75,2.5) to (-1.75,3); at (0,2);
=
(0,0) to (0,0.75) to \[out=45\] (0.75,1.75) to (0.75,3); (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (1.75,0) to (1.75,0.75) to \[looseness=1.1, in looseness=0.9\] node\[zxnode=, pos=0.36\] [$P$]{} node\[zxnode=, pos=0.64\] [$P$]{}(-1.75,3); at (0,0.75);
&
(0,0) to (0,2.25); (1,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{} (-1,2.5) to (-1,3); at (0,2.25);
=
(0,0) to (0,0.75); (1.,0) to (1.,0.75) to (-1,3); at (0,0.75);
&
(0,0) to (2.25,3); (2.25,0) to node\[zxnode=, pos=0.5\] [$P^\dagger$]{} (0,3);
=
(0.25,-0.5) to (0.25,0) to \[out=up, in=-135\] (1,1); (1,1) to \[out=135, in=right\] node\[zxvertex=, pos=1\] (-0.3, 1.7) to \[out=left, in=up\] (-1.25,1) to (-1.25,-0.5); (1.75,2.5) to (1.75,2) to \[out=down, in=45\] (1,1); (1,1) to \[out= -45, in= left\] node\[zxvertex=, pos=1\] (2.3,0.3) to \[out=right, in=down\] (3.25,1) to (3.25,2.5); at (1,1) [$P$]{};
The diagrammatic representation of quantum functions provides an interesting topological intuition: a quantum function behaves like a braiding or crossing between the directed and the undirected wire. From this perspective, allows us to pull the comultiplication and counit through the braiding. Note that we cannot yet pull the multiplication and unit through the braiding; adding these additional pull-throughs defines a quantum bijection, as will be seen in Section \[sec:quantumbijection\].
\[rem:opintofquantumfunctions\] We show in Corollary \[cor:quantumelementmeasurement\] that quantum functions $X\to Y$ between classical sets are families of projective measurements with outcomes in $Y$, controlled by the set $X$. We can think of this as a non-deterministic function $X \to Y$ which uses quantum measurements on an underlying Hilbert space to determine the output $y \in Y$ for a given input $x \in X$ (cf. Remark \[rem:interpretationquantumelement\]).
Having defined quantum functions, we make the following elementary observations.
A quantum element of a quantum set $A$ is a quantum function from the one-element [set $\{*\}$]{} (or equivalently from the commutative special dagger Frobenius algebra $\mathbb{C}$) to the quantum set $A$.
is a special case of where the source is the trivial quantum set.
We also note that quantum functions map quantum elements to quantum elements:$$\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,3.5);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.8] node[zxnode=\zxwhite, pos=0.46](Q) {$Q$} (-2.25,2.5) to (-2.25,3.5);
\draw[string,arrow data={0.35}{>}, arrow data={0.9}{>}] (3,0) to (3,0.25)to [in looseness=0.7] node[zxnode=\zxwhite, pos=0.67](Q2) {$P$} (-1,3.25) to (-1,3.5);
\draw[white,double] (0,0) to (Q.center);
\end{tz}$$ In particular, a quantum function between quantum sets $A$ and $B$ induces a functor between their categories of quantum elements.
\[def:qfctdimension\]We define the *dimension* of a quantum function $(H,P)$ to be the dimension of the underlying Hilbert space $H$.
A one-dimensional quantum function is an ordinary $*$-cohomomorphism of Frobenius algebras, as can be seen by comparing with . By Gelfand duality, a one-dimensional quantum function between classical sets is therefore just a function.
We can extend the notion of an intertwiner of quantum elements to all =-2
\[def:intertwiner\] An *intertwiner* of quantum functions $(H,P) \to (H', P')$ is a linear map $f:H\to H'$ such that the following holds:
(0,0) to (0,1) to (2,3); (2,0) to node\[zxnode=, pos=0.9\] [$f$]{} (2,1); (2,1) to node\[zxnode=, pos=0.5\][$P'$]{} (0,3);
=
(0,0) to (0,1) to (2,3); (2,0) to node\[zxnode=, pos=0.9\] [$f$]{} (2,1); (2,1) to node\[zxnode=, pos=0.5\][$P$]{} (0,3);
The 2-category $\QSet$
----------------------
We have seen in the discussion preceding Definition \[def:intertwiner\] that, while classical elements form a set, quantum elements should be organized into a *category*, keeping track of the additional layer of structure introduced by the intertwiners.
We thus expect the quantum analogue of the category of sets and functions to be a *2-category* of quantum sets and quantum functions, keeping track of the intertwiners between quantum functions.
Overall, we observe that our approach to quantisation leads to
\[def:2catqset\] The 2-category[^8] $\QSet$ is built from the following structures:
- **objects** are quantum sets $A,B$, ...;
- **1-morphisms** $A\to B$ are quantum functions $(H,P): A\to B$;
- **2-morphisms** $(H,P) \to (H', P')$ are intertwiners of quantum functions.
The composition of two quantum functions $(H,P):A\to B$ and $(H', Q): B \to C$ is a quantum function $(H' \otimes H, Q\circ P)$ defined as follows:
\[eq:1composition\] (1.575,0) to (0.325,3.5); (2.175,0) to (0.925,3.5); (0,0) to node\[zxvertex=, pos=0.5\] [$Q\circ P$]{} (2.5,3.5); at (4.4,0) [$H' \otimes H$]{};
:=
(0,0) to (2.5,3.5); (1.25,0) to node\[zxnode=, pos=0.4\] [$P$]{} (0,3.5); (2.5,0) to node\[zxnode=, pos=0.58\] [$Q$]{} (1.25,3.5); at (1.25,0) [$H$]{}; at (2.5,0) [$H'$]{};
Vertical and horizontal composition of 2-morphisms is defined as the ordinary composition and tensor product of linear maps, respectively.
As expected, we observe that $\QSet(*,A) = \QEl(A)$ where $*$ is the one-element set (cf. Terminology \[not:setalgebra\]).
\[rem:street\] We denote by $\BHilb$ the *delooping* of $\Hilb$; that is, the 2-category with a single object $*$ and endomorphism category $\Hom(*,*)=\Hilb$. We observe that $\QSet$ is a sub-2-category of Street’s 2-category $\mathrm{CoMnd}(\BHilb)$ of comonads, comonad maps and comonad transformations in the 2-category $\BHilb$ [@Street1972]. Indeed, the first two equations of can be understood as defining a comonad map (an analogous graphical definition of the 2-category $\mathrm{Mnd}(\Cat)$ of monads in the 2-category of categories, functors and natural transformations is given in [@Marsden2014; @Hinze2016]).
\[thm:QSetdagger2category\] $\QSet$ is a dagger 2-category.
Following Remark \[rem:street\], $\QSet$ is a sub-2-category of $\mathrm{CoMnd}(\BHilb)$, which is well known to be a 2-category (in the strict case the proof goes back at least to Street [@Street1972]). The dagger of a 2-morphism is defined to be the ordinary Hilbert space adjoint of the underlying linear map. This is well defined since the adjoint of an intertwiner $f:(H,P) \to (H', Q)$ is an intertwiner $f^\dagger: (H', Q) \to (H,P)$: $$\label{eq:qsetdagger}\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,1) to (2,3);
\draw[string, arrow data={0.35}{>}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f$} (2,1);
\draw[string,arrow data={0.9}{>},arrow data={0.26}{>}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$Q$} (0,3);
\end{tz}
~=~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,1) to (2,3);
\draw[arrow data={0.35}{<}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f$} (2,1);
\draw[string,arrow data={0.9}{<},arrow data={0.26}{<}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$P$} (0,3);
\end{tz}
\quad
\stackrel{~(-)^\dagger}{\Leftrightarrow}
\quad
%%%%%%%%%%%
\begin{tz}[zx,every to/.style={out=up, in=down},yscale=-1,xscale=-1]
\draw (0,0) to (0,1) to (2,3);
\draw[arrow data={0.35}{<}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f^\dagger$} (2,1);
\draw[string,arrow data={0.9}{<},arrow data={0.26}{<}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$Q^\dagger$} (0,3);
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=1]
\draw (0,0) to (0,1) to (2,3);
\draw[arrow data={0.35}{>}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f^\dagger$} (2,1);
\draw[string,arrow data={0.9}{>},arrow data={0.26}{>}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$P^\dagger$} (0,3);
\end{tz}
\quad
\super{\eqref{eq:quantumfunction}}{\Leftrightarrow}
\quad
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,1) to (2,3);
\draw[arrow data={0.35}{<}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f^\dagger$} (2,1);
\draw[string,arrow data={0.9}{<},arrow data={0.26}{<}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$Q$} (0,3);
\end{tz}
~=~
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,1) to (2,3);
\draw[arrow data={0.35}{>}] (2,0) to node[zxnode=\zxwhite, pos=0.9] {$f^\dagger$} (2,1);
\draw[string,arrow data={0.9}{>},arrow data={0.26}{>}](2,1) to node[zxnode=\zxwhite, pos=0.5]{$P$} (0,3);
\end{tz}$$
The underlying Hilbert space of the composite of two quantum functions is the tensor product of the underlying Hilbert spaces of the quantum functions. Similarly, vertical and horizontal composition of 2-morphisms coincides with composition and tensor product of linear maps. We therefore have a forgetful 2-functor $$\label{eq:fibrefunctor}F: \QSet \to \BHilb$$ from $\QSet$ to the delooping of $\Hilb$ (see Remark \[rem:street\]). As outlined in Philosophy \[phil:concretedagger\], the quantum mechanical interpretation of depends on this 2-functor; in particular, it makes $\QSet$ into a concrete dagger 2-category.
\[def:Hilb2category\] A *concrete dagger 2-category* is a pair $(\mathbb{B}, F)$, where $\mathbb{B}$ is a dagger 2-category and $F:\mathbb{B} \to \BHilb$ is a locally faithful dagger 2-functor (see Section \[sec:background\]).
In particular, a *concrete dagger monoidal category* is a pair $(\mathcal{C}, F)$ where $\mathcal{C}$ is a dagger monoidal category and $F:\mathcal{C} \to \Hilb$ is a faithful dagger monoidal functor.
In other words, a concrete dagger 2-category is a $2$-category for which every $\Hom$-category is a concrete dagger category in a compatible way. The 2-functor $F:\QSet\to \BHilb$ has in fact already appeared in Proposition \[prop:reconstruct\].
\[rem:setsubset\] By Gelfand duality, every set is itself a quantum set and every function is a quantum function. We may therefore think of $\Set$ as contained in $\QSet$; there is a faithful inclusion $2$-functor $\Set\hookrightarrow \QSet$, where we think of $\Set$ as a $2$-category with only identity $2$-morphisms.
Given two functions $f$ and $g$ between sets, $\QSet(f,g) = \delta_{f,g} \mathbb{C}$.
\[rem:symmmon2cat\] Using the cartesian product of sets, the category of sets and functions is a symmetric monoidal category. Similarly, it can be shown that , using the tensor product of the underlying algebras, becomes a symmetric monoidal 2-category.
A universal property {#sec:universal}
--------------------
In this section, we show that the categories of quantum functions $A \to B$ between quantum sets may be obtained via a universal construction, as categories of
quantum elements of the internal hom $[A,B]$ in the category of *quantum spaces* — the opposite of the category of $C^*$-algebras. Our definition of the quantum space of quantum functions between finite quantum sets is analogous to [@Soltan2009 Definition 3.1.], and generalises the construction used to define quantum permutation groups [@Wang1998].
To present these results, we work with infinite-dimensional $C^*$-algebras. Various parts of our presentation in Sections \[sec:background\] and \[sec:quantumset\] do not apply here (for instance, there are no infinite-dimensional Frobenius algebras), and we will therefore need to modify some definitions. Any such modifications will apply only in this section.
It is an observation of Wang [@Wang1998] that quantum functions between classical spaces should form a quantum set rather than a classical set. Let $\CAlg$ be the category of (possibly infinite-dimensional) $C^*$-algebras and $*$-homomorphisms. By analogy with Gelfand duality, we might treat the category $\CAlg^\op$ as the category of ‘quantum spaces’. However, given two finite sets $A$ and $B$, understood as commutative $C^*$-algebras, then $\CAlg^\op (A,B) =\Set(A,B)$ is just a set; from this perspective there are only classical functions between $A$ and $B$. This inspires the following definition, analogous to constructions in [@Wang1998] and [@Soltan2009].
\[def:internalhom\]
The *quantum space of quantum functions* between two finite quantum sets $A$ and $B$ is the internal hom $[A,B]$ in $\CAlg^\op$; that is, the universal $C^*$-algebra such that there are bijections[^9] $$\label{eq:internalhom}\CAlg^\op (C, [A,B]) \cong \CAlg^\op (C\otimes A, B)$$ which are natural in $C\in \CAlg^\op$.
\[rem:soltan\] Sołtan’s *quantum space of all maps* between two quantum spaces [@Soltan2009 Definition 3.1.] is precisely the internal hom in the category $\CAlg^\op$. It follows from [@Soltan2009 Theorem 3.3] that the quantum space of quantum functions between finite quantum sets always exists.
The internal hom $[A,B]$ is in general an infinite-dimensional noncommutative $C^*$-algebra. In the finite-dimensional case, we showed that a quantum set could be recovered from its category of quantum elements. In the infinite-dimensional setting we cannot define the category of quantum elements as the category of comodules, due to the lack of a comultiplication on $[A,B]$. We can nevertheless define the category of quantum elements of $[A,B]$ as the *opposite* of the category of modules of $[A,B]$; this coincides with the usual definition in the finite-dimensional case.
The *category of finite quantum elements* $\QEl(A)$ of a quantum space $A$ is the opposite of the category $\mathrm{Rep}_{\text{fd}}(A)$ of finite-dimensional $C^*$-representations[^10] of the $C^*$-algebra $A$.
We now show that our category of quantum functions between finite quantum sets $A$ and $B$ coincides with the category of quantum elements of the internal hom $[A,B]$.
\[thm:universalprop\] The category of finite quantum elements of $[A,B]$ is equivalent to the category of quantum functions $\QSet(A,B)$: $$\QEl([A,B]) \cong \QSet(A,B)$$
We need to show that $\mathrm{Rep}_{\text{fd}}([A,B])^\op \cong \QSet(A,B)$. A representation of $[A,B]$ is a $*$-homomorphism $[A,B]\to \End(H)$ for some finite-dimensional Hilbert space. We have the following series of bijections: $$\label{eq:universalsequence}\CAlg([A,B], \End(H)) = \CAlg^\op(\End(H),[A,B])$$$$\hspace{1cm}\cong \CAlg^\op(\End(H)\otimes A, B) = \CAlg(B, \End(H)\otimes A)$$ The representation therefore corresponds to a $*$-homomorphism $B\to \End(H) \otimes A$. Since all algebras $A,B$ and $\End(H)$ are finite-dimensional, we can now use the Frobenius algebra formalism established in Section \[sec:background\]. Taking the Hilbert space adjoint of the homomorphism $B\to \End(H)\otimes A$ yields a $*$-cohomomorphism (in the sense of equation ) $P:\End(H)\otimes A\to B$. Explicitly, this means:
(0,0) to (0,0.5) to \[out=135\] (-0.75, 1.5) to (-0.75,1.7); (0, 0.5) to \[out=45\] (0.75,1.5) to (0.75,1.7); (-2\*, 0)to (-2\*, ) to(-0.75-2\*, 1.7); (-, 0) to (-,) to (0.75-, 1.7) to (0.75-, 3); (-0.75-, 1.8) to (-0.75-, 3); (-0.75-, 1.7) to \[out=down, in=down,looseness=1.6\] (0.75-2\*, 1.7) ; at (0.75-, 1.8) [$P$]{}; at (-0.75-, 1.8) [$P$]{}; at (0,0.5);
=
(-,0) to (-,1); (0,0) to (0,1); (,0) to (,1); (0,1) to (0,2) to \[out=135\] (-0.75, 3); (0,2) to \[out=45\] (0.75,3); at (0,1) [$P$]{}; at (0,2);
&
(-,0) to (-,1); (0,0) to (0,1); (,0) to (,1); (0,1) to (0,2); at (0,1) [$P$]{}; at (0,2);
=
(-,0) to (-,0.5) to \[out=up, in=up\] (0,0.5) to (0,0); (,0) to (,1); at (,1);
&
(-0.5, 0) rectangle (0.5,3); (-,0) to (-,1.5); (0,0) to (0,1.5); (,0) to (,1.5); (0,1.5) to (0,3); at (0,1.5) [$P^\dagger$]{};
=
(-1.5, 0) rectangle (1.5,3); (0,1.5) to \[out=up, in=up,looseness=\] node\[zxvertex=, pos=0.5\](1,1.5) to (1,0); (-,1.5) to \[out=down, in=down,looseness=\] (-1, 1.5) to (-1,3); (0,1.5) to \[out=down, in=down,looseness=2.5\] (-1-, 1.5) to (-1-,3); (,1.5) to \[out=down, in=down,looseness=4\]node\[zxvertex=, pos=0.5\] (-1+, 1.5) to (-1+,3); at (0,1.5) [$P$]{};
But these are exactly the axioms of a quantum function for the following morphism $H\otimes A \to B\otimes H$: $$\begin{tz}[zx,every to/.style={out=up, in=down},scale=1.2]
\draw (0,1.5) to (-0.75, 3);
\draw (0.75,0) to (\d-0.05, 1.5);
\draw[arrow data={0.35}{>}] (-0.75,0) to (-\d+0.1,1.5);
\draw[arrow data={0.2}{>},arrow data={0.85}{>}] (0,1.5) to[out=down, in=down, looseness=5] (0.7, 1.5) to[out=up, in=down] (0.75, 3);
\node[zxnode=\zxwhite] at (0,1.5) {$P$};
\end{tz}$$ Analogously, an intertwiner $f$ of $[A,B]$-actions is mapped to the intertwiner $f^\dagger$ of the corresponding quantum functions. This defines the desired equivalence.
From this perspective, $\QSet$ provides a high-level approach to the representation theory of the algebras of quantum functions, in particular clarifying their compositional behaviour.
\[rem:quantumfunctionalgebra\] It is not hard to show that for classical sets $X$ and $Y$, the $C^*$-algebra $[X,Y]$ is the universal $C^*$-algebra with generators $p_{x,y}$ for $x\in X, y \in Y$ and the following relations for $x\in X, y \in Y$ (also see Theorem \[thm:quantumfunctionsinprojectors\] and Proposition \[prop:Wang\]):
p\_[x,y]{}\^\* = p\_[x,y]{} = p\_[x,y]{}\^2 & p\_[x,y]{} p\_[x,y’]{} = \_[y,y’]{} p\_[x,y]{} &\_[yY]{} p\_[x,y]{}=1
For general finite-dimensional $C^*$-algebras $A$ and $B$, there is a similar =-2
\[rem:problemTannaka\] We remark that the categories $\QSet(A,B)\cong \mathrm{Rep}_{\mathrm{fd}}([A,B])^{op}$ only keep track of the *finite-dimensional* representation theory of the internal hom $[A,B]$.
It is tempting to try to reconstruct the algebra $[A,B]$ from the abstract $C^*$-category $\QSet(A,B)$ and the fibre functor $\QSet([A,B])\to \Hilb$. However, if neither $A$ nor $B$ are the one-element set — and if therefore $[A,B]$ is infinite-dimensional — we will see in Section \[sec:semisimple\] that the set of isomorphism classes of simple objects of $\QSet(A,B)$ in uncountable and carries a natural non-discrete topology. This cannot be captured with the inherently discrete tools of (ordinary) category theory; we cannot reconstruct $[A,B]$ from the abstract category $\QSet(A,B) \to \Hilb$. This is analogous to the inability to recover a discrete group from its (naive) representation category.
Quantum bijections {#sec:quantumbijection}
==================
The quantum graph isomorphisms of Atserias et al. [@Atserias2016], the quantum permutation groups of Wang [@Wang1998], and the quantum automorphism groups of Banica [@Banica2005] are all intended as quantisations of classical bijections. In this section, we define within our $2$-categorical framework a notion of quantum bijection which we show to be equivalent to the quantum bijections of these authors.
In particular, we show that quantum bijections can be exactly characterised as the *dagger-dualisable* $1$-morphisms in the $2$-category $\QSet$. We emphasise that quantum bijections should not be thought of as invertible but merely as dualisable. In fact, categorical equivalences in $\QSet$ are just classical bijections, or more generally [$*$-isomorphisms]{}.
Equivalences in $\QSet$
-----------------------
We first prove that equivalences in the sense of 2-category theory correspond to ordinary $*$-isomorphisms of Frobenius algebras; this shows that categorical equivalence is too strong a notion to characterise quantum bijections in $\QSet$.
Ordinary bijections — and more generally ordinary $*$-isomorphisms between Frobenius algebras — can be characterised by the following algebraic condition:
\[prop:Frobeniusinvertible\]A $*$-cohomomorphism between symmetric special Frobenius algebras is invertible if and only if it is also a $*$-homomorphism.[^11]
\[eq:Frobeniusmonoidlemma\]
(0,0) to (0,2) to \[out=135\] (-0.75,3); (0,2) to \[out=45\] (0.75, 3); at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (0,0.75) to \[out=45\] (0.75, 1.75) to +(0,1.25); at (-0.75,2) [$f$]{}; at (0.75,2) [$f$]{}; at (0,0.75) ;
&
(0,0) to (0,2) ; at (0,1) [$f$]{}; at (0,2) ;
=
(0,0) to (0,0.75) ; at (0,0.75) ;
The proposition follows as a corollary from Theorem \[thm:dual2\], as shown in Remark \[rem:lemma42fromlemma46\].=-2
Now we show that equivalences in $\QSet$ are ordinary $*$-isomorphisms between Frobenius algebras.
\[prop:equivalence\]Equivalences between quantum sets in $\QSet$ are $*$-isomorphisms of Frobenius algebras.
Equivalences are preserved by the 2-functor $F:\QSet\to \BHilb$ of equation . In particular, the underlying Hilbert space $H$ of such an equivalence is invertible; there is another Hilbert space $H'$ such that $H\otimes H' \cong \mathbb{C}$. Therefore, $H\cong \mathbb{C}$ and the equivalence is a quantum function with one-dimensional Hilbert space and thus an invertible $*$-cohomomorphism. It follows from Proposition \[prop:Frobeniusinvertible\] that invertible $*$-cohomomorphisms are $*$-isomorphisms.
Two characterisations of quantum bijections {#sec:twocharacterisations}
-------------------------------------------
Having established that equivalences in $\QSet$ are ordinary $*$-isomorphisms, we must look elsewhere for an appropriate definition of quantum bijection. There are two possible approaches. The first is to quantise the algebraic characterisation of $*$-isomorphism from Proposition \[prop:Frobeniusinvertible\], following the procedure outlined in Section \[sec:quantumset\]. The other more structural approach is to define quantum bijections in terms of a genuinely categorical notion of weakened equivalence — the notion of *dagger adjunction* or *dagger duality*. We will show that both these approaches lead to the same definition of quantum bijection.
We first quantise the algebraic characterisation .
\[def:quantumbijection\] A *quantum bijection* between quantum sets $A$ and $B$ is a quantum function $(H,P)$ between $A$ and $B$ fulfilling the following additional equations:
\[eq:monadmap\]
(0,0) to (0,2) to \[out=45\] (0.75,3); (0,2) to \[out=135\] (-0.75,3); (1.75,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{} (-1.75,2.5) to (-1.75,3); at (0,2);
=
(0,0) to (0,0.75) to \[out=45\] (0.75,1.75) to (0.75,3); (0,0.75) to \[out=135\] (-0.75,1.75) to (-0.75,3); (1.75,0) to (1.75,0.75) to \[looseness=1.1, in looseness=0.9\] node\[zxnode=, pos=0.36\] [$P$]{} node\[zxnode=, pos=0.64\] [$P$]{}(-1.75,3); at (0,0.75);
&
(0,0) to (0,2.25); (1,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{} (-1,2.5) to (-1,3); at (0,2.25);
=
(0,0) to (0,0.75); (1.,0) to (1.,0.75) to (-1,3); at (0,0.75);
Alternatively, quantum bijections can be characterised as duals in $\QSet$. F[[\^\* F]{}]{}
\[def:duals\] A $1$-morphism $F:A\to B$ in a 2-category has a *right dual* $\F*:B\to A$ if there are 2-morphisms $\epsilon_R:F{\circ}\F*\to 1_{B}$ (counit) and $\eta_R:1_A\to \F*{\circ} F$ (unit) fulfilling the triangle equations[^12]:
(\_R 1\_[F]{})(1\_[F]{}\_R) = 1\_[F]{} & (1\_\_R )(\_R1\_) = 1\_
A $1$-morphism has a *left dual* $\*F:B\to A$ if there are 2-morphisms $\epsilon_L:\*F {\circ}F\to 1_{A}$ and $\eta_L:1_B\to F {\circ}\*F$ fulfilling the triangle equations:
(\_L 1\_[F]{})(1\_[F]{}\_L) = 1\_[F]{} & (1\_F\_L )(\_L1\_[F]{}) = 1\_[F]{}
In a dagger 2-category, a $1$-morphism has a *dual* $\overline{F}:B\to A$ if $\overline{F}$ is both a left and a right dual and the corresponding 2-morphisms are related as follows:
\_L\^= \_R & \_L\^= \_R
\[rem:daggerdual\] In a dagger $2$-category, every left (or right) dual is automatically a dual in the sense of Definition \[def:duals\]. We refer to a $1$-morphism that has a dual as *dualisable*.
As discussed in Philosophy \[phil:concretedagger\], $\QSet$ should be thought of not only as a 2-category but as a concrete dagger 2-category (see Definition \[def:Hilb2category\]), equipped with its forgetful $2$-functor $\QSet \to \BHilb$. In particular, the notion of a dual in $\QSet$ should be compatible with the underlying Hilbert space structure. This leads to the following definition.
\[def:daggerdualisable\]In a concrete dagger 2-category $(\mathbb{B}, F)$, a $1$-morphism $S$ is *dagger dualisable* if it has a dual $\conj{S}$ such that the underlying duality between $F(S)$ and $F(\conj{S})$ in $\Hilb$ is a dagger duality in the sense of Definition \[def:daggerduality\].
We will refer to such a $1$-morphism $\conj{S}$ as a *dagger dual* of $S$.
To characterise dagger-dualisable quantum functions, we adopt terminology from [@Vicary2012hq; @Reutter2016] and make the following definition.
\[def:biinvertible\]Let $A,B$ and $H$ be finite-dimensional Hilbert spaces. A linear map\
is *bi-invertible* if it is invertible and if $$\label{eq:biinv}\begin{tz}[zx,xscale=0.6,yscale=1,scale=1]
\draw(0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw [arrow data={0.34}{>}] (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw(1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw[arrow data={0.32}{<}] (1,1) to [out= -45, in= left] (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P^{-1}$};
\end{tz}
~~\text{ is inverse to }~~
\begin{tz}[zx,xscale=0.6,yscale=-1,scale=1]
\draw(0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw [arrow data={0.34}{<}] (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw(1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw[arrow data={0.32}{>}] (1,1) to [out= -45, in= left] (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P$};
\end{tz}$$
We now show that quantum bijections as in Definition \[def:quantumbijection\] are precisely dagger-dualisable quantum functions.
\[thm:dual2\]For a quantum function $(H,P)$ between quantum sets $A$ and $B$, the following are equivalent:
1. $(H,P)$ is dagger-dualisable in $\QSet$.
2. The underlying linear map $P:H\otimes A \to B \otimes H$ is bi-invertible.
3. The underlying linear map $P:H\otimes A \to B\otimes H$ is unitary.
4. $(H,P)$ is a quantum bijection as in Definition \[def:quantumbijection\].
If these properties hold, then the quantum function $(H^*,\overline{P}): B\to A$ whose underlying linear map $\overline{P}:H^* \otimes B \to A \otimes H^*$ is defined as follows, is a dagger dual of $(H,P)$: $$\label{eq:daggerdual}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=0.8]
\draw [arrow data={0.2}{<},arrow data={0.8}{<}] (0,0) to (2.25,3);
\draw (2.25,0) to node[zxnode=\zxwhite, pos=0.5] {$\overline{P}$} (0,3);
\end{tz}
~~~:=~~~
\begin{tz}[zx,xscale=0.6,xscale=1]
\draw(0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw [arrow data={0.34}{>}] (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw(1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw[arrow data={0.32}{<}] (1,1) to [out= -45, in= left] (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P^\dagger$};
\end{tz}
~~~=~~~
\begin{tz}[zx,xscale=0.6,xscale=1]
\draw[arrow data={0.68}{>}] (4.75,2.5) to (4.75, 0.5) to [out=down, in=right] (2.65,-0.6) to [out=left, in=down] (0.55,0.5);
\draw (0.55,0.5) to [out=up, in=-135] (1,1);
\draw (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] node[zxvertex=\zxwhite, pos=0] {}(-1.25,1) to (-1.25,-0.5);
\draw[string, arrow data={0.68}{<}] (-2.75, -0.5) to (-2.75,1.5) to [out=up, in=left] (-0.65, 2.6) to [out=right, in=up] (1.45,1.5);
\draw (1.45,1.5) to [out=down, in=45] (1,1);
\draw (1,1) to [out= -45, in= left] node[zxvertex=\zxwhite, pos=1] {} (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P$};
\end{tz}$$ The units and counits witnessing the duality between $(H,P)$ and $(H^*, \conj{P})$ are given by the standard cups and caps in $\Hilb$ . Moreover, $(H^*, \conj{P})$ is the unique dagger dual with these units and counits.
We prove the implications $1.\Rightarrow 2.\Rightarrow 4.\Rightarrow 3.\Rightarrow 1.$
$1. \Rightarrow 2.$ Suppose that $P$ has a dagger dual $Q'$. Since a dagger duality in $\QSet$ induces a dagger duality of the underlying Hilbert spaces and since dagger dualities are unique up to unitary isomorphism, the underlying Hilbert space of $Q'$ is unitarily isomorphic to the dual space $H^*$. Conjugating $Q'$ by this unitary isomorphism leads to a quantum function $Q$, dagger dual to $P$, whose underlying Hilbert space is exactly $H^*$ and such that the underlying linear maps of the unit and counit of the adjunction are the ‘standard’ cups and caps in $\Hilb$ . In other words, a quantum function $P:H\otimes A \to B \otimes H$ has a dagger dual, if and only if there is a quantum function $Q:H^* \otimes B \to A \otimes H^*$ such that the following holds:
\[eq:rightdual\]
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=6.\] (2,0); (-2,0) to \[out=up, in=down\] node\[zxnode=, pos=0.4\] [$P$]{} node\[zxnode=, pos=0.6\] [$Q$]{}(4,4);
=
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=2.\] (2,0); (-2,0) to \[out=up, in=down\] (4,4);
&
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=6.\] (2,0); (-2,0) to \[out=up, in=down\] node\[zxnode=, pos=0.4\] [$P$]{} node\[zxnode=, pos=0.6\] [$Q$]{}(4,4);
=
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=2.\] (2,0); (-2,0) to \[out=up, in=down\] (4,4);
\[eq:leftdual\]
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=6.\] (2,0); (-2,0) to \[out=up, in=down\] node\[zxnode=, pos=0.4\] [$Q$]{} node\[zxnode=, pos=0.6\] [$P$]{}(4,4);
=
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=2.\] (2,0); (-2,0) to \[out=up, in=down\] (4,4);
&
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=6.\] (2,0); (-2,0) to \[out=up, in=down\] node\[zxnode=, pos=0.4\] [$Q$]{} node\[zxnode=, pos=0.6\] [$P$]{}(4,4);
=
(-2.05,-0.05) rectangle (4.05,4.05); (0,0) to \[out=up, in=up, looseness=2.\] (2,0); (-2,0) to \[out=up, in=down\] (4,4);
Equations and can equivalently be expressed as stating the following:
(0.25,-0.5) to (0.25,0) to \[out=up, in=-135\] (1,1); (1,1) to \[out=135, in=right\] (-0.3, 1.7) to \[out=left, in=up\] (-1.25,1) to (-1.25,-0.5); (1.75,2.5) to (1.75,2) to \[out=down, in=45\] (1,1); (1,1) to \[out= -45, in= left\] (2.3,0.3) to \[out=right, in=down\] (3.25,1) to (3.25,2.5); at (1,1) [$Q$]{};
(0,0) to (2.25,3); (2.25,0) to node\[zxnode=, pos=0.5\] [$P$]{} (0,3);
&
(0,0) to (2.25,3); (2.25,0) to node\[zxnode=, pos=0.5\] [$Q$]{} (0,3);
(0.25,-0.5) to (0.25,0) to \[out=up, in=-135\] (1,1); (1,1) to \[out=135, in=right\] (-0.3, 1.7) to \[out=left, in=up\] (-1.25,1) to (-1.25,-0.5); (1.75,2.5) to (1.75,2) to \[out=down, in=45\] (1,1); (1,1) to \[out= -45, in= left\] (2.3,0.3) to \[out=right, in=down\] (3.25,1) to (3.25,2.5); at (1,1) [$P$]{};
We have therefore shown that $P$ is bi-invertible. $2.\Rightarrow 4.$ Suppose that $P:H\otimes A \to B \otimes H$ is a bi-invertible quantum function. We now demonstrate that it fulfills the second equation in :
$$\def\l{2.25}
\def\h{1}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.5}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw[string] (0.75,0.25) to node[zxnode=\zxwhite, pos=0.26] {$P$} (0.75,\top);
\node[zxvertex=\zxwhite] at (0.75,0.25){};
\draw[string,arrow data={0.2}{>}, arrow data={0.8}{>}] (\bottomright, \bottom) to (\topleft, \top);
\end{tz}
~=~~
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\l{2.25}
\def\h{1}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.5}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\node at (0,0) (C){};
\draw (C.center) to [out=down, in=down,looseness=2] node[zxvertex=\zxwhite, pos=0.5] {} (-\l, 0) to [out=up, in=down] (-2.25,\h);
\draw[string] (C.center) to [out=45, in=down] ([xshift=0.75cm, yshift=1cm]C.center) to node[zxnode=\zxwhite, pos=0.16] {$P$} (0.75,\top);
\draw[string] (C.center) to [out=135, in=down] ([xshift=-0.75cm, yshift=1cm]C.center) to (-0.75,\h) to [out=up, in=down] (-2.25, \h+1) to [out=up, in=up] node[zxvertex=\zxwhite, pos=0.5]{} (-0.75,\h+1) to [out=down, in=up] (-2.25,\h);
\node[zxvertex=\zxwhite,zxup] at (C.center){};
\draw[string,arrow data={0.2}{>}, arrow data={0.8}{>}] (\bottomright, \bottom) to (\topleft, \top);
\end{tz}
~~=~
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\l{2.25}
\def\h{3.5}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.2}
\def \middle{2.3}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\node at (0,0) (C){};
\draw (C.center) to [out=down, in=down,looseness=2] node[zxvertex=\zxwhite, pos=0.5] {} (-\l, 0) to [out=up, in=down] (-2.25,\h);
\draw[string] (C.center) to [out=45, in=down] ([xshift=0.75cm, yshift=1cm]C.center) to node[zxnode=\zxwhite, pos=0.0] {$P$} (0.75,\top);
\draw[string] (C.center) to [out=135, in=down] ([xshift=-0.75cm, yshift=1cm]C.center) to node[zxnode=\zxwhite, pos=0.18] {$P$}node[zxnode=\zxwhite, pos=0.85] {$P^{-1}$}(-0.75,\h) to [out=up, in=down] (-2.25, \h+1) to [out=up, in=up] node[zxvertex=\zxwhite, pos=0.5]{} (-0.75,\h+1) to [out=down, in=up] (-2.25,\h);
\node[zxvertex=\zxwhite,zxup] at (C.center){};
\draw[string,arrow data={0.15}{>}, arrow data={0.64}{>},arrow data={0.9}{>}] (\bottomright, \bottom) to (\bottomright, -0.1) to (-1.5, \middle) to (\topleft, \h+0.5) to (\topleft, \top);
\end{tz}
~~=~
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\l{2.25}
\def\h{1.5}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.2}
\def \middle{0.2}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\node at (0,0) (C){};
\draw (C.center) to [out=down, in=down,looseness=2] node[zxnode=\zxwhite, pos=0.15] {$P$} node[zxvertex=\zxwhite, pos=0.5] {} (-\l, 0) to [out=up, in=down] (-2.25,\h);
\draw[string] (C.center) to [out=45, in=down] ([xshift=0.75cm, yshift=1cm]C.center) to (0.75,\top);
\draw[string] (C.center) to [out=135, in=down] ([xshift=-0.75cm, yshift=1cm]C.center) to node[zxnode=\zxwhite, pos=-0.2] {$P^{-1}$}(-0.75,\h) to [out=up, in=down] (-2.25, \h+1) to [out=up, in=up] node[zxvertex=\zxwhite, pos=0.5]{} (-0.75,\h+1) to [out=down, in=up] (-2.25,\h);
\node[zxvertex=\zxwhite,zxup] at (C.center){};
\draw[string,arrow data={0.15}{>}, arrow data={0.4}{>},arrow data={0.9}{>}] (\bottomright, \bottom) to (-1., \middle) to (\topleft, \h+0.5) to (\topleft, \top);
\end{tz}
~~=~
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\l{2.25}
\def\h{3.5}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.2}
\def \middle{2.3}
\def\xgap{-0.9cm}
\def\loose{3}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\node at (0,0) (C){};
\draw (C.center) to [out=down, in=down,looseness=2] node[zxvertex=\zxwhite, pos=0.5] {} (-\l, 0) to [out=up, in=down] (-2.25,\h);
\draw[string] (C.center) to [out=45, in=down] ([xshift=0.75cm, yshift=1cm]C.center) to (0.75,\top);
\draw[string] (C.center) to [out=135, in=down] ([xshift=-0.75cm, yshift=1cm]C.center) to node[zxnode=\zxwhite, pos=0.1](b) {$P^{-1}$}node[zxnode=\zxwhite, pos=0.7](t) {$P$}(-0.75,\h) to [out=up, in=down] (-2.25, \h+1) to [out=up, in=up] node[zxvertex=\zxwhite, pos=0.5]{} (-0.75,\h+1) to [out=down, in=up] (-2.25,\h);
\draw[arrow data={0.5}{>}][string] (t.center) to [looseness=\loose,out=135, in=up]([xshift=\xgap]t.center) to ([xshift=\xgap]b.center) to [looseness=\loose,out=down, in=-135] (b.center);
\node[zxvertex=\zxwhite,zxup] at (C.center){};
\draw[string,arrow data={0.25}{>}] (\bottomright, \bottom) to [in=-45] (t.center);
\draw[string,arrow data = {0.7}{>}] (b.center) to [out=45] (\topleft, \top);
\end{tz}$$
ł[2.25]{} $$=~\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\node at (0,0) (C){};
\draw (C.center) to [out=down, in=down,looseness=2] node[zxvertex=\zxwhite, pos=0.5] {} (-\l, 0) to [out=up, in=down] (-2.25,\h);
\draw[string] (C.center) to [out=45, in=down] ([xshift=0.75cm, yshift=1cm]C.center) to (0.75,\top);
\draw[string] (C.center) to [out=135, in=down] ([xshift=-0.75cm, yshift=1cm]C.center) to (-0.75,\h) to [out=up, in=down] (-2.25, \h+1) to [out=up, in=up] node[zxvertex=\zxwhite, pos=0.5]{} (-0.75,\h+1) to [out=down, in=up] (-2.25,\h);
\node[zxvertex=\zxwhite,zxup] at (C.center){};
\coordinate (loopcenter) at (0.05,2);
\draw[string,arrow data={0.2}{>}, arrow data={0.8}{>}] (\bottomright, \bottom) to [in=-45] (loopcenter) to [looseness=\looploose,out=135, in=up] ([xshift=\loop]loopcenter) to [looseness=\looploose,out=down, in=-135] (loopcenter) to [out=45] (\topleft, \top);
\end{tz}
~~=~
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\def\l{2.25}
\def\h{1}
\def\top{5.5}
\def\bottom{-2}
\def\bottomright{2}
\def\topleft{-0.5}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw[string] (0.75,4) to (0.75,\top);
\node[zxvertex=\zxwhite] at (0.75,4){};
\draw[string,arrow data={0.2}{>}, arrow data={0.8}{>}] (\bottomright, \bottom) to (\topleft, \top);
\end{tz}$$ Here, the first equation follows from the axioms of symmetric special Frobenius algebras, the second equation is invertibility of $P$, the third equation uses the fact that $P$ is a quantum function, the fourth equation follows from the graphical calculus moving $P$ along the right wire onto the top of $P^{-1}$. Finally, in the first equation of the second line, we use bi-invertibility of $P$.
We now show that if $P$ is invertible, then the second equation in implies the first equation. $$\def\scl{0.85}
\begin{tz}[zx,scale=-1,scale=\scl,xscale=-1]
\clip (-3.05,-0.2) rectangle (1.8,4.2);
\draw[arrow data={0.1}{<},arrow data={0.5}{<}, arrow data={0.95}{<}] (1.75,0) to [out=up,in=-45] (1,1) to (-1,3) to [out=135, in=down] (-1.75,4);
\draw (0.25,0) to [out=up, in=-135] (1,1) to [in looseness=2.5,out=45, in=65] node[pos=0.72](c){} (-1,3) to [out=-135, in=right] (-2,2) to [out=left, in=down] (-3,4) ;
\node[zxnode=\zxwhite] at (1,1) {$P$};
\node[zxnode=\zxwhite] at (-1,3) {$P$};
\node[zxvertex=\zxwhite] at (c.center){};
\node[zxvertex=\zxwhite] at (-2,2){};
\end{tz}
~~\superequals{\eqref{eq:quantumfunction}~\&~\eqref{eq:cupcapfrob}}~~~~
\begin{tz}[zx,scale=-1,scale=\scl,xscale=-1]
\draw (1.75,0) to (1.75,1) to[out=up, in=-45] (1,2) to [out=-135, in=up] (0.25,1) to [looseness=1.5,out=down, in=down] node[pos=0.5] (b){}(-1.25,1) to (-1.25,4);
\draw (1,2) to [looseness=0.4,out=up, in=down] (1.75,3.25);
\draw[arrow data={0.4}{<},arrow data={0.9}{<}] (3.5,0) to [looseness=0.5,out=up,in=down]node[zxnode=\zxwhite, pos=0.64] {$P$} (0.5,4);
\node[zxvertex=\zxwhite] at (b.center){};
\node[zxvertex=\zxwhite,zxup] at (1,2){};
\node[zxvertex=\zxwhite] at (1.75,3.25){};
\end{tz}
~~\superequalseq{eq:monadmap}~~
\begin{tz}[zx,scale=-1,scale=\scl,xscale=-1]
\draw (0,0) to (0, 2) to [looseness=1.5,out=up, in=up] node[zxvertex=\zxwhite,pos=0.5]{} (-1.5,2) to [looseness=1.5,out=down, in=down]node[zxvertex=\zxwhite,pos=0.5]{} (-3,2) to (-3,4);
\draw[string,arrow data={0.5}{<}] (1.5,0) to + (0,4);
\end{tz}
~~=~~
\begin{tz}[zx,xscale=1,scale=\scl]
\draw (0,0) to + (0,4);
\draw[string,arrow data={0.5}{>}] (1.5,0) to + (0,4);
\end{tz}
\hspace{0.5cm} \Rightarrow \hspace{0.5cm}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=0.8,scale=1]
\draw [arrow data={0.2}{>},arrow data={0.8}{>}] (0,0) to (2.25,3);
\draw (2.25,0) to node[zxnode=\zxwhite, pos=0.5] {$P^{-1}$} (0,3);
\end{tz}
~~=~~
\begin{tz}[zx,xscale=-0.6,yscale=-1,scale=\scl]
\draw[arrow data={0.5}{<}] (0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw (1,1) to [out=135, in=right] node[zxvertex=\zxwhite, pos=1]{} (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw[arrow data={0.5}{>}] (1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw (1,1) to [out= -45, in= left] node[zxvertex=\zxwhite, pos=1] {} (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P$};
\end{tz}$$ Since $P^{-1}$ is also a left inverse of $P$, this implies the following: $$\label{eq:proofbijection}\begin{tz}[zx,scale=\scl,xscale=1]
\clip (-2.05,-0.05) rectangle (4.05,4.05);
\draw(0,0) to [out=up, in=up, looseness=6.] node[zxvertex=\zxwhite,pos=0.5] {} (2,0);
\draw [string,arrow data={0.1}{>},arrow data={0.9}{>}] (-2,0) to [out=up, in=down] node[zxnode=\zxwhite, pos=0.4] {$P$} node[zxnode=\zxwhite, pos=0.6] {$P$}(4,4);
\end{tz}
=
\begin{tz}[zx,scale=\scl,xscale=1]
\clip (-2.05,-0.05) rectangle (4.05,4.05);
\draw (0,0) to [out=up, in=up, looseness=2.] node[zxvertex=\zxwhite, pos=0.5] {}(2,0);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (-2,0) to [out=up, in=down] (4,4);
\end{tz}$$ We can then prove the first equation as follows: $$\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,2) to [out=45] (0.75,3);
\draw (0,2) to [out=135] (-0.75,3);
\draw[arrow data={0.2}{<}, arrow data={0.8}{<}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxdown] at (0,2){};
\end{tz}
~~\superequalseq{eq:Frobenius}~~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,1.7) to [in=-135] (0.5,2.3) to (0.5,3);
\draw (0.5,2.3) to [out=-45, in=up] (1,1.7) to [out=down, in=down, looseness=1.5] node[zxvertex=\zxwhite, pos=0.5] {}(1.75,1.7) to (1.75,3);
\draw[string,arrow data={0.2}{<}, arrow data={0.8}{<}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0.5,2.3){};
\end{tz}
~~~~~\superequals{\eqref{eq:quantumfunction}~\&~\eqref{eq:proofbijection}}~~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (-1,0) to (-1,0.7) to [in=-135] node[pos=1] (a) {}(-0.5,1.3) to (-0.5,3);
\draw[string] (-0.5,1.3) to [out=-45, in=up] (0,0.7) to [out=down, in=down, looseness=1.5] node[zxvertex=\zxwhite, pos=0.5] {}(0.75,0.7) to (0.75,3);
\draw[string,arrow data={0.2}{<}, arrow data={0.9}{<}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.34] {$P$} node[zxnode=\zxwhite, pos=0.6] {$P$}(-1.75,3);
\node[zxvertex=\zxwhite,zxup] at (a.center){};
\end{tz}
~~\superequalseq{eq:Frobenius}~~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=-1,xscale=-1]
\draw (0,0) to (0,0.75) to [out=45] (0.75,1.75) to (0.75,3);
\draw (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw[arrow data={0.2}{<}, arrow data={0.9}{<}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$P$} node[zxnode=\zxwhite, pos=0.64] {$P$}(-1.75,3);
\node[zxvertex=\zxwhite, zxdown] at (0,0.75){};
\end{tz}$$
$4.\Rightarrow 3.$ $$\begin{tz}[zx,every to/.style={out=up, in=down},scale=1,xscale=-1]
\draw (0.25,0) to [out=up, in=-135] (1,1) to [looseness=1.5,out=45, in=-45] (1,3) to [out=135, in=down] (0.25,4);
\draw[arrow data={0.1}{>}, arrow data={0.5}{>}, arrow data={0.95}{>}] (1.75,0) to [out=up, in=-45] (1,1) to [looseness=1.5, out=135, in=-135] (1,3) to [out=45, in=down] (1.75,4);
\node[zxnode=\zxwhite] at (1,1) {$P$};
\node[zxnode=\zxwhite] at (1,3) {$P^\dagger$};
\end{tz}
~~\superequalseq{eq:quantumfunction}~~
\begin{tz}[zx,xscale=-1]
\clip (-3.05,-0.2) rectangle (1.8,4.2);
\draw[arrow data={0.1}{>},arrow data={0.5}{>}, arrow data={0.95}{>}] (1.75,0) to [out=up,in=-45] (1,1) to (-1,3) to [out=135, in=down] (-1.75,4);
\draw (0.25,0) to [out=up, in=-135] (1,1) to [in looseness=2.5,out=45, in=65] node[pos=0.72](c){} (-1,3) to [out=-135, in=right] (-2,2) to [out=left, in=down] (-3,4) ;
\node[zxnode=\zxwhite] at (1,1) {$P$};
\node[zxnode=\zxwhite] at (-1,3) {$P$};
\node[zxvertex=\zxwhite] at (c.center){};
\node[zxvertex=\zxwhite] at (-2,2){};
\end{tz}
~~\superequalseq{eq:monadmap}~~
\begin{tz}[zx,xscale=-1]
\draw (1.75,0) to (1.75,1) to[out=up, in=-45] (1,2) to [out=-135, in=up] (0.25,1) to [looseness=1.5,out=down, in=down] node[pos=0.5] (b){}(-1.25,1) to (-1.25,4);
\draw (1,2) to [looseness=0.4,out=up, in=down] (1.75,3.25);
\draw[arrow data={0.4}{>},arrow data={0.9}{>}] (3.5,0) to [looseness=0.5,out=up,in=down]node[zxnode=\zxwhite, pos=0.64] {$P$} (0.5,4);
\node[zxvertex=\zxwhite] at (b.center){};
\node[zxvertex=\zxwhite,zxdown] at (1,2){};
\node[zxvertex=\zxwhite] at (1.75,3.25){};
\end{tz}
~~\superequalseq{eq:quantumfunction}~~
\begin{tz}[zx,xscale=-1]
\draw (0,0) to (0, 2) to [looseness=1.5,out=up, in=up] node[zxvertex=\zxwhite,pos=0.5]{} (-1.5,2) to [looseness=1.5,out=down, in=down]node[zxvertex=\zxwhite,pos=0.5]{} (-3,2) to (-3,4);
\draw[string,arrow data={0.5}{>}] (1.5,0) to + (0,4);
\end{tz}
~~=~~
\begin{tz}[zx,xscale=-1]
\draw (0,0) to + (0,4);
\draw[string,arrow data={0.5}{>}] (1.5,0) to + (0,4);
\end{tz}$$ $$\begin{tz}[zx,every to/.style={out=up, in=down},xscale=1]
\draw (0.25,0) to [out=up, in=-135] (1,1) to [looseness=1.5,out=45, in=-45] (1,3) to [out=135, in=down] (0.25,4);
\draw[arrow data={0.1}{>}, arrow data={0.5}{>}, arrow data={0.95}{>}] (1.75,0) to [out=up, in=-45] (1,1) to [looseness=1.5, out=135, in=-135] (1,3) to [out=45, in=down] (1.75,4);
\node[zxnode=\zxwhite] at (1,1) {$P^\dagger$};
\node[zxnode=\zxwhite] at (1,3) {$P$};
\end{tz}
~~\superequalseq{eq:quantumfunction}~~
\begin{tz}[zx,scale=-1,xscale=-1]
\clip (-3.05,-0.2) rectangle (1.8,4.2);
\draw[arrow data={0.1}{<},arrow data={0.5}{<}, arrow data={0.95}{<}] (1.75,0) to [out=up,in=-45] (1,1) to (-1,3) to [out=135, in=down] (-1.75,4);
\draw (0.25,0) to [out=up, in=-135] (1,1) to [in looseness=2.5,out=45, in=65] node[pos=0.72](c){} (-1,3) to [out=-135, in=right] (-2,2) to [out=left, in=down] (-3,4) ;
\node[zxnode=\zxwhite] at (1,1) {$P$};
\node[zxnode=\zxwhite] at (-1,3) {$P$};
\node[zxvertex=\zxwhite] at (c.center){};
\node[zxvertex=\zxwhite] at (-2,2){};
\end{tz}
~~\superequalseq{eq:quantumfunction}~~
\begin{tz}[zx,scale=-1,xscale=-1]
\draw (1.75,0) to (1.75,1) to[out=up, in=-45] (1,2) to [out=-135, in=up] (0.25,1) to [looseness=1.5,out=down, in=down] node[pos=0.5] (b){}(-1.25,1) to (-1.25,4);
\draw (1,2) to [looseness=0.4,out=up, in=down] (1.75,3.25);
\draw[arrow data={0.4}{<},arrow data={0.9}{<}] (3.5,0) to [looseness=0.5,out=up,in=down]node[zxnode=\zxwhite, pos=0.64] {$P$} (0.5,4);
\node[zxvertex=\zxwhite] at (b.center){};
\node[zxvertex=\zxwhite,zxup] at (1,2){};
\node[zxvertex=\zxwhite] at (1.75,3.25){};
\end{tz}
~~\superequalseq{eq:monadmap}~~
\begin{tz}[zx,scale=-1,xscale=-1]
\draw (0,0) to (0, 2) to [looseness=1.5,out=up, in=up] node[zxvertex=\zxwhite,pos=0.5]{} (-1.5,2) to [looseness=1.5,out=down, in=down]node[zxvertex=\zxwhite,pos=0.5]{} (-3,2) to (-3,4);
\draw[string,arrow data={0.5}{<}] (1.5,0) to + (0,4);
\end{tz}
~~=~~
\begin{tz}[zx]
\draw (0,0) to + (0,4);
\draw[string,arrow data={0.5}{>}] (1.5,0) to + (0,4);
\end{tz}$$ $3. \Rightarrow 1.$ Suppose that $P:H \otimes A \to B \otimes H$ is unitary. Then, the following linear map is a quantum function: $$\begin{tz}[zx,every to/.style={out=up, in=down},xscale=0.8,scale=0.8]
\draw [arrow data={0.2}{<},arrow data={0.8}{<}] (0,0) to (2.25,3);
\draw (2.25,0) to node[zxnode=\zxwhite, pos=0.5] {$Q$} (0,3);
\end{tz}
~:=~
\begin{tz}[zx,xscale=0.6,yscale=1,scale=0.8]
\draw(0.25,-0.5) to (0.25,0) to [out=up, in=-135] (1,1);
\draw [arrow data={0.34}{>}] (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] (-1.25,1) to (-1.25,-0.5);
\draw(1.75,2.5) to (1.75,2) to [out=down, in=45] (1,1);
\draw[arrow data={0.32}{<}] (1,1) to [out= -45, in= left] (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P^{\dagger}$};
\end{tz}
~=~
\begin{tz}[zx,xscale=0.6,xscale=1,scale=0.8]
\draw[arrow data={0.68}{>}] (4.75,2.5) to (4.75, 0.5) to [out=down, in=right] (2.65,-0.6) to [out=left, in=down] (0.55,0.5);
\draw (0.55,0.5) to [out=up, in=-135] (1,1);
\draw (1,1) to [out=135, in=right] (-0.3, 1.7) to [out=left, in=up] node[zxvertex=\zxwhite, pos=0] {}(-1.25,1) to (-1.25,-0.5);
\draw[string, arrow data={0.68}{<}] (-2.75, -0.5) to (-2.75,1.5) to [out=up, in=left] (-0.65, 2.6) to [out=right, in=up] (1.45,1.5);
\draw (1.45,1.5) to [out=down, in=45] (1,1);
\draw (1,1) to [out= -45, in= left] node[zxvertex=\zxwhite, pos=1] {} (2.3,0.3) to [out=right, in=down] (3.25,1) to (3.25,2.5);
\node [zxnode=\zxwhite] at (1,1) {$P$};
\end{tz}$$ For example, the counit condition of $Q$ follows from precomposing the counit condition of the quantum function $P$ with $P^\dagger$: $$\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2.25);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1,2.5) to (-1,3);
\node[zxvertex=\zxwhite] at (0,2.25){};
\end{tz}
~~\superequalseq{eq:quantumfunction}~~
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.,0) to (1.,0.75) to (-1,3);
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\end{tz}
\hspace{0.75cm}\super{\text{$P$ unitary}}{\Leftrightarrow} \hspace{0.75cm}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=1]
\draw (0,0) to (0,0.75);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.,0) to (1.,3) ;
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\end{tz}
~~=~~
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw[arrow data ={0.15}{>}, arrow data={0.8}{>}] (0,0) to (1,2) to (1,3);
\draw (1,0) to (0,2);
\node[zxnode=\zxwhite] at (0.5,1) {$P^\dagger$};
\node[zxvertex=\zxwhite] at (0,2){};
\end{tz}
\hspace{0.75cm}\Leftrightarrow \hspace{0.75cm}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75);
\draw[arrow data={0.2}{<}, arrow data={0.9}{<}] (1.,0) to (1.,0.75) to (-1,3);
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\end{tz}
~~=~~
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2.25);
\draw[arrow data={0.2}{<}, arrow data={0.8}{<}] (1,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$Q$} (-1,2.5) to (-1,3);
\node[zxvertex=\zxwhite] at (0,2.25){};
\end{tz}$$ The comultiplication condition of $Q$ can be proven similarly. The last equation in follows from a direct graphical argument, proving that $Q$ is indeed a quantum function.
Moreover, unitarity of $P$ implies that $P$ and $Q$ fulfill equations . We note that equation is the dagger of equation and therefore redundant. This follows directly from the fact that $P$ and $Q$ are quantum functions and $\QSet$ is a dagger 2-category (see equation ). Therefore, $Q$ is a dagger dual of $P$ in $\QSet$.
\[rem:lemma42fromlemma46\] Proposition \[prop:Frobeniusinvertible\] follows as the one-dimensional case of Theorem \[thm:dual2\]. In fact, following our philosophy outlined in Section \[sec:quantumset\], several of the steps of the proof of Theorem \[thm:dual2\] were first devised for the simpler case of Proposition \[prop:Frobeniusinvertible\]. The oriented wire was added in later, in an essentially unique way.
It is a direct consequence of Theorem \[thm:dual2\] that a quantum function $(H,P)$ is dagger-dualisable (or equivalently bi-invertible) if and only if it is biunitary as defined in [@Reutter2016].
We denote the 2-category of quantum sets, quantum bijections and intertwiners by $\QBij$.=-2
\[rem:monoidalsubcats\] Composition of quantum functions makes the category $\QBij(B,B)$ of *quantum permutations* on a quantum set $B$ into a monoidal dagger category with dualisable objects
(see Theorem \[thm:dual2\]). In particular, the categories of quantum permutations $\QBij([n], [n])$ on an $n$-element set $[n]$ can be thought of as a quantum version of the symmetric group $S_n$.
Quantum bijections in noncommutative topology {#sec:Wang}
---------------------------------------------
The monoidal categories $\QBij(B,B)$ of quantum bijections on a quantum set $B$ have been considered in finite noncommutative topology.
In [@Wang1998], Wang introduced ‘quantum symmetry groups of finite quantum spaces’ as non-commutative variants of the symmetric groups $S_n$. We now show that our category of quantum permutations on a quantum set is precisely the category of finite-dimensional representations of the Hopf $C^*$-algebra corresponding to Wang’s quantum symmetry group. In other words, our quantum permutations are quantum elements of this quantum group (cf. Theorem \[thm:universalprop\]).
\[prop:Wang\] For a quantum set $B$, $\QBij(B,B)$ is the category of finite-dimensional representations of Wang’s ‘quantum symmetry group’ algebra $A_{aut}(B)$.
We only prove the proposition for classical sets $[n]$; the proof for general quantum sets is completely analogous, only involving more indices.
In [@Wang1998 Theorem 3.1], Wang defines quantum permutation groups of classical finite sets $[n]$ as $C^*$-algebras $A_{aut}([n])$ with generators $a_{i,j}$ $(i,j=1,\ldots n)$ and relations:
\[eq:Wangcondition\] a\_[i,j]{}\^2=a\_[i,j]{} = a\_[i,j]{}\^\* & \_[i=1]{}\^n a\_[i,j]{} = 1, 1jn & \_[j=1]{}\^n a\_[i,j]{} =1, 1in
Noting that projectors summing to the identity are mutually orthogonal and comparing the relations with Theorem \[thm:quantumfunctionsinprojectors\] shows that our categories $\QBij([n], [n])$ are the categories of finite-dimensional representations of $A_{aut}([n])$.
Proposition \[prop:Wang\] is not surprising: In Section \[sec:universal\], we showed that $\QSet(B,B)$ is the category of finite-dimensional representations of the internal hom $[B,B]$ and in Section \[sec:quantumbijection\] we prove that $\QBij(B,B)$ is the full subcategory of dagger-dualisable objects in $\QSet(B,B)$. We therefore expect $\QBij(B,B)$ to arise as the category of finite-dimensional representations of some internal ‘automorphism Hopf $C^*$-algebra’, which is exactly how Wang defines his compact quantum group algebra $A_{aut}(B)$.
\[rem:Wangnotdiscrete\] As in Remark \[rem:problemTannaka\], we cannot recover $A_{aut}(B)$ from the fibre functor $\QBij(B,B) \to \Hilb$. Indeed, from the perspective of locally compact quantum groups (see e.g. [@Maes1998; @Neshveyev2013]), we can think of $\QBij(B,B)$ as the category of finite-dimensional representations
of the discrete quantum group dual to Wang’s compact quantum symmetry group. Similarly to the case of ordinary discrete groups this discrete quantum group cannot generally be recovered from its (naive) representation category.
Quantum bijections between classical sets {#sec:quantumperm}
-----------------------------------------
We now focus on quantum functions and quantum bijections between classical sets and show that we recover the *magic unitaries* of Banica et al. [@Banica2007] and the *projective permutation matrices* of Atserias et al. [@Atserias2016]. We first show that we can express quantum functions between classical sets as families of projectors satisfying certain =-2
\[thm:quantumfunctionsinprojectors\] A quantum function $X\to Y$ between classical sets $X$ and $Y$ is exactly a family of projectors $\{P_{x,y}\}_{x\in X, y \in Y}$ on a Hilbert space $H$ such that the following holds, for all $x\in X$ and $y_1,y_2\in Y$:
\[eq:PPM1\]P\_[x,y\_1]{} P\_[x,y\_2]{} = \_[y\_1,y\_2]{} P\_[x,y\_1]{} & \_[yY]{} P\_[x,y]{} = \_H
A quantum bijection $X\to Y$ between classical sets $X$ and $Y$ is exactly a family of projectors $\{P_{x,y}\}_{x\in X, y \in Y}$ satisfying and the following additional conditions, for all $x_1,x_2\in X$ and $y\in Y$:
\[eq:PPM2\] P\_[x\_1,y]{} P\_[x\_2,y]{} = \_[x\_1,x\_2]{} P\_[x\_1,y]{} & \_[xX]{} P\_[x,y]{} = \_H
Every classical set $X$ corresponds to a commutative Frobenius algebra defined in Example \[exm:Frob\], the elements of $X$ form a basis of copyable elements of this algebra. The linear map $P:H\otimes X \to Y \otimes H$ may therefore be expressed in this basis: :$$\begin{tz}[zx]
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (0,0) to (0,3);
\node[zxnode=\zxwhite] at (0,1.5) {$P_{x,y}$};
\end{tz}
~~:=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=0.8]
\draw [arrow data={0.2}{>},arrow data={0.8}{>}] (0,0) to (2.25,3);
\draw (1.75,0.5) to node[zxnode=\zxwhite, pos=0.5] {$P$} (0.5,2.5);
\node[zxnode=\zxwhite] at (1.75,0.5){$x$};
\node[zxnode=\zxwhite] at (0.5,2.5){$y$};
\end{tz}$$ As an example, the first equation of , expanded in the classical basis, becomes $$\delta_{y,y'}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2.5);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxnode=\zxwhite] at (0,2.5){$y$};
\node[zxnode=\zxwhite] at (0,0) {$x$};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,2) to [out=45] (0.75,3);
\draw (0,2) to [out=135] (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.8}{>}] (1.75,0) to [looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$P$} (-1.75,2.5) to (-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0,2){};
\node[zxnode=\zxwhite] at (-0.75,3) {$y$};
\node[zxnode=\zxwhite] at (0.75,3) {$y'$};
\node[zxnode=\zxwhite] at (0,0) {$x$};
\end{tz}
\superequalseq{eq:quantumfunction}
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (0,0.75) to [out=45] (0.75,1.75) to (0.75,3);
\draw (0,0.75) to [out=135] (-0.75,1.75) to (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$P$} node[zxnode=\zxwhite, pos=0.64] {$P$}(-1.75,3);
\node[zxvertex=\zxwhite, zxup] at (0,0.75){};
\node[zxnode=\zxwhite] at (-0.75,3) {$y$};
\node[zxnode=\zxwhite] at (0.75,3) {$y'$};
\node[zxnode=\zxwhite] at (0,0) {$x$};
\end{tz}
=
\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0.75,0.85) to (0.75,3);
\draw (-0.75,0.85) to (-0.75,3);
\draw[arrow data={0.2}{>}, arrow data={0.9}{>}] (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.36] {$P$} node[zxnode=\zxwhite, pos=0.64] {$P$}(-1.75,3);
\node[zxnode=\zxwhite] at (-0.75,3) {$y$};
\node[zxnode=\zxwhite] at (0.75,3) {$y'$};
\node[zxnode=\zxwhite] at (0.75,0.85) {$x$};
\node[zxnode=\zxwhite] at (-0.75,0.85) {$x$};
\end{tz}$$ All other expressions are similar translations of and .
It is natural to arrange these projectors in an $|X| \times |Y|$ matrix. From this perspective, equation states that the projectors along each row form a complete orthogonal family, while equation requires this for each column. In the work of Banica et al, matrices of projectors obeying both the row and the column equations are called *magic unitaries* [@Banica2007], while in the work of Atserias et al they are called *projective permutation matrices* [@Atserias2016]. In this paper, we adopt the latter terminology.
Theorem \[thm:quantumfunctionsinprojectors\] has the following immediate corollary.
\[cor:quantumelementmeasurement\]A quantum function between sets $X\to Y$ is a family of projective measurements with outcomes in $Y$, controlled by the set $X$. In particular, a quantum element of a set $X$ is a projective measurement with outcomes in $X$.
\[rem:ppmcomposition\] At the level of matrices of projectors, the composition of quantum functions $P:X\to Y$ and $Q:Y \to Z$ (see equation ) takes the following form:
(QP)\_[x,z]{} =\_[y Y]{} Q\_[y,z]{}P\_[x,y]{}
This coincides with composition of quantum functions in the Kleisli category of Abramsky et al [@Abramsky2017].
Examples of projective permutation matrices are provided by the first author’s *quantum Latin squares* [@Musto2015; @Musto2016]; more examples are given in Section \[sec:semisimple\].
We now show that quantum bijections only exist between classical sets of the
If there is a quantum bijection from a classical set $X$ to a classical set $Y$, then $|X| = |Y|$. In other words, every projective permutation matrix is square.
By Theorem \[thm:dual2\], a quantum bijection $P: H\otimes X \to Y \otimes H$ is unitary. In particular, $|X|\dim(H) = \dim(X\otimes H) = \dim(H\otimes Y) = |Y| \dim(H)$.
\[rem:teleportation\] We remark that there are quantum functions and quantum bijections between noncommutative algebras. We suggest that these structures might also play a role in quantum information theory. For example, unitary error bases[^13] — providing the basic data for quantum teleportation and dense coding schemes [@Werner2001] — give rise to quantum bijections from the matrix algebra $\mathrm{Mat}_n$ to the set $[n^2]$ with $n^2$ elements. Similar correspondences have been noted in the work of Stahlke [@Stahlke2016].
Quantum graph theory {#sec:qgt}
====================
In the first sections of this paper, we have presented a quantisation of the theory of finite sets and functions. In the following, we continue this general approach and quantise the theory of finite undirected graphs, unifying and extending the work of various authors on nonlocal games [@Cameron2007; @Mancinska2016; @Atserias2016], noncommutative topology [@Banica2005; @Banica2009; @Banica2007_2; @Bichon2003], operator algebras [@Weaver2010; @Kuperberg2012; @Weaver2015] and quantum information [@Duan2013; @Stahlke2016].
By analogy with classical graphs, we define a quantum graph as a quantum set of vertices together with a quantum adjacency matrix.[^14] We define a quantum graph homomorphism as a quantum function preserving the quantum adjacency matrix, and show that quantum graphs and quantum graph homomorphisms naturally form a concrete dagger $2$-category $\QGraph$. We then define quantum graph isomorphisms, and show that these are precisely the dagger-dualisable $1$-morphisms in $\QGraph$.
For a classical graph $G$, we denote its set of vertices by $V_G$; for vertices $v,w\in V_G$ we write $v\sim_G w$ if $v$ and $w$ are connected in $G$. We consider only undirected graphs.
Quantum graphs via adjacency matrices {#sec:qgraphsbyadjmats}
-------------------------------------
A classical graph $G$ can be described by its vertex set $V_G$ and its *adjacency matrix*, which is a linear map $G:V_G\to V_G$ (see Terminology \[not:setalgebra\]) with coefficients $G_{v,w}=1$ if $v\sim_G w$ and $G_{v,w} = 0$ otherwise.
This suggests the following definition of a quantum graph as a quantum set equipped with a quantum adjacency matrix.
\[def:quantumgraphsbyadjmats\] A *quantum graph* is a pair $(A,G)$ of a quantum set $A$ (the *quantum set of vertices*) and a self-adjoint linear map $G:A \to A$ (the *quantum adjacency matrix*) satisfying the following equations:
\[eq:propadjacency\]
(0,0) to (0,0.5) to \[out=135, in=-135,looseness=1.5\] node\[zxnode=, pos=0.5\] [$G$]{} (0,2.5) to (0,3); (0,0.5) to \[out=45, in=-45,looseness=1.5\] node\[zxnode=, pos=0.5\] [$G$]{} (0,2.5); at (0,0.5); at (0,2.5);
=
(0,0) to node\[zxnode=, pos=0.5\] [$G$]{} (0,3);
&
(1,0) to (1,1.5) to \[out=up, in=up, looseness=2.5\]node\[zxvertex=, pos=0.5\] (0,1.5) to \[out=down, in=down, looseness=2.5\]node\[zxvertex=, pos=0.5\] (-1,1.5) to (-1,3); at (0,1.5) [$G$]{};
=
(0,0) to (0,3); at (0,1.5) [$G$]{};
For a classical set $A=V_G$, this reduces to the definition of an adjacency matrix $\{G_{v,w}\}_{v,w\in V_G}$; from left to right, the conditions state that $G_{v,w}^2 = G_{v,w}$, and that $G_{v,w} = G_{w,v}$.
A quantum graph is *reflexive* or *irreflexive* if one of the following additional equations holds: $$\includegraphics[trim={1cm 0 1cm .35cm},valign=c]{reflexivity.pdf}$$ For classical graphs this corresponds to $G_{v,v}=1$ or $G_{v,v}=0$, respectively. For a classical set, the definition of an irreflexive quantum graph therefore reduces to the standard definition of an adjacency matrix of a simple graph.
\[rem:literatureqgraph\]There are many related definitions of noncommutative or quantum graphs in the literature [@Kuperberg2012; @Weaver2010; @Weaver2015; @Duan2013], with applications in quantum error correction [@Weaver2015] and zero-error communication [@Duan2013]. In Section \[app:quantumrelation\], we comment on how our quantum graphs correspond to these previous definitions. In particular, we prove the following:=-2
- Our reflexive quantum graphs coincide with Weaver’s finite-dimensional quantum graphs [@Weaver2015], defined in terms of symmetric and reflexive quantum relations [@Kuperberg2012; @Weaver2010].
- Our reflexive quantum graphs $(\mathrm{Mat}_n, G)$ on matrix algebras coincide with Duan, Severini and Winter’s noncommutative graphs [@Duan2013].
Quantum graph homomorphisms {#sec:quantumgraphhomandiso}
---------------------------
Having defined quantum graphs, we now quantise graph homomorphisms. As with quantum functions and quantum sets, we have classical homomorphisms between quantum graphs, quantum homomorphisms between classical graphs and quantum homomorphisms between quantum graphs. We show that our quantum homomorphisms and quantum isomorphisms between classical graphs coincide with the quantum graph homomorphisms of Mančinska et al [@Mancinska2016] and the quantum graph isomorphisms of Atserias et al [@Atserias2016]. We also show that our classical and quantum homomorphisms between quantum graphs are ‘pure’ versions of Stahlke’s quantum graph homomorphisms [@Stahlke2016].
To quantise graph homomorphisms, we first express them via Gelfand duality in terms of string diagrams in the category of finite-dimensional Hilbert spaces and linear maps. For a quantum graph — a quantum set $A$ with quantum adjacency matrix $G:A\to A$ — we introduce the following notation: $$\label{eq:graphprojector1}\begin{tz}[zx]
\draw[string] (0, 0) to (0,3);
\draw (2.5,0) to (2.5, 3);
\draw (0, 1.5) to (2.5, 1.5) ;
\node[zxvertex=\zxwhite] at (0,1.5){};
\node[zxvertex=\zxwhite] at (2.5,1.5){};
\node[zxnode=\zxwhite] at (1.25, 1.5 ) {$G$};
\end{tz}
~~~:=~~~
\begin{tz}[zx,xscale=2.5/3]
\draw (-1.5,0) to (-1.5,1.5) to [out=up, in=-135] (-0.75, 2.5) to (-0.75,3);
\draw (-0.75, 2.5) to [out=-45, in=up] (0, 1.5) to [out=down, in=135] (0.75,0.5) to (0.75,0);
\draw (0.75,0.5) to [out=45, in=down] (1.5,1.5) to (1.5,3);
\node[zxnode=\zxwhite] at (0, 1.5) {$G$};
\node[zxvertex=\zxwhite, zxdown] at (-0.75, 2.5) {};
\node[zxvertex=\zxwhite, zxup] at (0.75, 0.5) {};
\end{tz}
~~~\superequalseq{eq:propadjacency}~~~
\begin{tz}[zx,xscale=-2.5/3]
\draw (-1.5,0) to (-1.5,1.5) to [out=up, in=-135] (-0.75, 2.5) to (-0.75,3);
\draw (-0.75, 2.5) to [out=-45, in=up] (0, 1.5) to [out=down, in=135] (0.75,0.5) to (0.75,0);
\draw (0.75,0.5) to [out=45, in=down] (1.5,1.5) to (1.5,3);
\node[zxnode=\zxwhite] at (0, 1.5) {$G$};
\node[zxvertex=\zxwhite, zxdown] at (-0.75, 2.5) {};
\node[zxvertex=\zxwhite, zxup] at (0.75, 0.5) {};
\end{tz}$$ If $(V_G,G)$ is a classical graph, it can easily be verified that this map is a projector onto the following subset (or equivalently onto the commutative subalgebra corresponding to this subset): $$\label{eq:projectorsubset1}\left\{ (v,w) ~|~ v \sim_G w \right\} \subseteq V_G \times V_G$$ We can use these projectors to express the notion of a classical graph homomorphism diagrammatically:
Let $(V_G,G)$ and $(V_H,H)$ be classical graphs. Under Gelfand duality, a graph homomorphism $G\to H$ corresponds to a $*$-cohomomorphism $f:V_G\to V_H$ fulfilling the following equation: $$\label{eq:classicalgraphhomomorphism}\begin{tz}[zx,yscale=0.8]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.5] {$f$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.5] {$f$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\end{tz}
~~~=~~~
\begin{tz}[zx,yscale=0.8]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.5] {$f$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.5] {$f$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$H$};
\end{tz}
~~
\begin{tz}[zx]
\end{tz}$$
Since is a projector onto the subset , equation is simply expressing that if $v\sim_G w$, then $f(v) \sim_H f(w)$.
We quantise this notion following our usual philosophy.
\[def:quantumgraphhom\] Let $(A,G)$ and $(B,H)$ be quantum graphs. A *quantum homomorphism* $(H,P) : (A,G) \to (B,H)$ is a quantum function $(H,P): A \to B$ fulfilling the following additional equation: $$\label{eq:quantumgraphhomomorphism}\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string, arrow data={0.08}{>},arrow data={0.5}{>}, arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
~=~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$H$};
\draw[string, arrow data={0.08}{>}, arrow data ={0.5}{>},arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}$$
It is possible to have ordinary homomorphisms between quantum graphs (that is, [$*$-cohomomorphisms]{} obeying or equivalently one-dimensional quantum homomorphisms), quantum homomorphisms between classical graphs and quantum homomorphisms between quantum graphs.
Expressing a quantum homomorphism between classical graphs in terms of its family of projectors $\{P_{v,w}\!\}_{v\in V_G, w\in V_H}$, fulfilling , this condition becomes:=-2 $$\label{eq:qgraphhomprojcond} v\sim_G v' \text{ and }w\not\sim_H w'\hspace{0.75cm} \Rightarrow \hspace{0.75cm} P_{v',w'} P_{v,w} = 0$$
\[rem:quantumhomMancinska\]Mančinska and Roberson [@Mancinska2016] define quantum graph homomorphisms between classical graphs $G$ and $H$ as perfect quantum strategies — strategies involving measurements on a shared entangled resource — for a certain bipartite nonlocal *graph homomorphism game*. This work generalises the earlier quantum graph colouring game [@Avis2006] and the corresponding notion of the quantum chromatic number of a graph [@Cameron2007].[^15]
In [@Mancinska2016], the existence of such a perfect strategy is expressed in terms of a family of projectors $\{P_{v,w}\}_{v\in V_G, w\in V_H}$ which form a quantum homomorphism in the sense of Definition \[def:quantumgraphhom\].
\[prop:mancinskagraphhom\] Given graphs $G$ and $H$, there is a perfect quantum strategy for the graph homomorphism game defined in [@Mancinska2016], if and only if there is a nonzero family of projectors $\{P_{v,w}\}_{v\in V_G, w\in V_H}$ fulfilling equation and — or equivalently, if and only if there is a nonzero quantum homomorphism $ (H, P): (V_G,G)\to (V_H,H)$.
Quantum functions are themselves perfect quantum strategies for a *function game*, whose deterministic classical strategies correspond to classical functions between finite sets. In this game, a verifier sends an element of a finite set $X$ to Alice and Bob, who in turn return an element of another set $Y$. A strategy is perfect if Alice and Bob return the same element $y\in Y $ whenever they receive the same element $x\in X$. Of course, every such game admits a classical strategy.
\[rem:stahlkegraphhoms\] Stahlke [@Stahlke2016] defines homomorphisms [@Stahlke2016 Definition 7] and quantum homomorphisms [@Stahlke2016 Definition 15] (there called entanglement assisted homomorphisms) between quantum graphs on endomorphism algebras (cf. Definition \[def:operatorsystem\] and Proposition \[prop:quantumgraphDuanthesame\]), and relates them to zero-error strategies for quantum source-channel coding in various scenarios.
Stahlke’s notions of homomorphism and quantum homomorphism are defined in terms of ‘mixed’ completely positive trace preserving (CPTP) maps which, if restricted to the ‘pure’ setting of $*$-homomorphisms, agree with our definitions.[^16]
Quantum graph isomorphisms {#sec:qisosofgraphs}
--------------------------
Recall that a classical *graph isomorphism* is an invertible graph homomorphism whose inverse is also a graph homomorphism. This condition can be more concisely expressed as follows.
Let $G$ and $H$ be classical graphs. Under Gelfand duality, a graph isomorphism $G\to H$ corresponds to a $*$-isomorphism $f:V_G \to V_H$ fulfilling the $$\label{eq:classicalgraphiso}
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1]
\draw (0,0) to (0,3);
\path (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$f$}(-1.75,3);
\node[zxnode=\zxwhite] at (0,0.9){$G$};
\end{tz}
~~~= ~~~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=1]
\draw (0,0) to (0,3);
\path (1.75,0) to (1.75,0.75) to [looseness=1.1, in looseness=0.9] node[zxnode=\zxwhite, pos=0.5] {$H$}(-1.75,3);
\node[zxnode=\zxwhite] at (0,0.9){$f$};
\end{tz}$$
This is well-known, but will also follow as a corollary of Theorem \[thm:dualisablequantumgraph\].
As in Section \[sec:twocharacterisations\], we can define a quantum graph isomorphism either as a quantum bijection fulfilling a quantised version of , or as a dualisable quantum graph homomorphism in the 2-category of quantum graphs, quantum graph homomomorphisms and intertwiners, which we will shortly define. In fact, we will show in Theorem \[thm:dualisablequantumgraph\] that, as with quantum bijections, both these approaches lead to the same notion of quantum graph isomorphism.
\[def:quantumgraphiso\] Let $(A,G)$ and $(A',G')$ be quantum graphs. A *quantum isomorphism* $(H,P): (A,G)\to (A',G')$ is a quantum bijection $(H,P): A \to A'$ fulfilling the following additional equation:
\[eq:quantumgraphisomorphism\]
(0,0) to (0,3); (1.75,0) to (1.75,0.75) to \[looseness=1.1, in looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{}(-1.75,3); at (0,0.9)[$G$]{};
=
(0,0) to (0,3); (1.75,0) to \[looseness=0.9\] node\[zxnode=, pos=0.5\] [$P$]{} (-1.75,2.5) to (-1.75,3); at (0,2.35) [$G'$]{};
Expressing quantum isomorphisms between classical graphs in terms of their underlying projective permutation matrix $\{P_{v,w}\}_{v\in V_G, w\in V_{G'}}$, which fulfils and , the quantum isomorphism condition becomes: $$\label{eq:ppmquantumgraphcondition1}
\forall~a\in V_{G},~b \in V_{G'}:\hspace{1cm}\sum_{i \in \mathrm{nbh}_{G}(a)} P_{i,b} = \sum_{j \in \mathrm{nbh}_{G'}(b)} P_{a,j}$$ Here, $\mathrm{nbh}_{G}(a)$ and $\mathrm{nbh}_{G'}(b)$ denote the set of neighboring vertices of $a$ and $b$ in $G$ and $G'$, respectively. Equivalently, this condition can be expressed as follows: $$\label{eq:ppmquantumgraphcondition2}\text{If }\left(v\sim_{G}v'\text{ and }w\not\sim_{G'}w'\right)\text{ or }\left(v\not\sim_{G} v'\text{ and }w\sim_{G'} w'\right)\hspace{0.5cm} \Rightarrow\hspace{0.5cm} P_{v',w'} P_{v,w} = 0$$
The equivalence between and was proven by Atserias et al. [@Atserias2016 Theorem 5.8], but will also follow from Theorem \[thm:dualisablequantumgraph\] and the fact that a quantum isomorphism is a dualisable quantum homomorphism whose dual $\overline{P}$ is also a quantum homomorphism.
\[rem:Atseriasquantumiso\] Atserias et al [@Atserias2016] define quantum graph isomorphisms between classical graphs as perfect quantum strategies for the *graph isomorphism game*, which is a symmetric version of the graph homomorphism game of Mančinska and Roberson (see Remark \[rem:quantumhomMancinska\]).
As with quantum graph homomorphisms, Atserias et al. show that such perfect quantum strategies exist if and only if there is a projective permutation matrix forming a quantum isomorphism in the sense of Definition \[def:quantumgraphiso\].
\[prop:Atserias\]Given classical graphs $G$ and $G'$, there is a perfect quantum strategy for the graph isomorphism game defined in [@Atserias2016], if and only if there is a nonzero family of projectors $\{P_{v,w}\}_{v\in V_G, w\in V_{G'}}$ fulfilling equations , and — or equivalently, if and only if there is a nonzero quantum isomorphism $(H,P):(V_G,G) \to (V_{G'},G')$.=-2
The 2-category $\QGraph$ {#sec:qgraph}
------------------------
We now show that these new structures again form a 2-category under composition, which we call $\QGraph$.
\[def:2catqgraph\] The 2-category $\QGraph$ is built from the following structures:
- **objects** are quantum graphs $(A,G),(A',G')$, ...;
- **1-morphisms** $(A,G)\!\begin{tikzpicture}\node at (0,0){};\draw[->](0,0)--(0.25,0);\end{tikzpicture}\, (\!A',G')$ are quantum homomorphisms $(H,P)\!:\!\!(A,G)\!\begin{tikzpicture}\node at (0,0){};\draw[->](0,0)--(0.25,0);\end{tikzpicture}(\!A',G')$;=-5
- **2-morphisms** $(H,P)\!\begin{tikzpicture}\node at (0,0){};\draw[->](0,0)--(0.25,0);\end{tikzpicture}\, (\!H', P')$ are intertwiners of the underlying quantum functions (see Definition \[def:intertwiner\]).
As with $\QBij \subset \QSet$, we may define a subcategory $\QIso \subset \QGraph $ of quantum graphs, quantum graph isomorphisms and their intertwiners.
As with $\QSet$, there is a forgetful dagger $2$-functor $F:\QGraph \to \BHilb$ mapping each quantum homomorphism to its underlying Hilbert space and each intertwiner to its corresponding linear map. This $2$-functor makes $\QGraph$ a concrete dagger $2$-category $(\QGraph, F)$ (see Definition \[def:Hilb2category\]).
\[rem:last\] Every quantum homomorphism between quantum graphs $(A,G)$ and $(B,H)$ is a quantum function between the underlying quantum sets of vertices $A$ and $B$. Since an intertwiner of quantum homomorphisms is exactly an intertwiner of the underlying quantum functions, it follows that the category $\QGraph((A,G),(B,H))$ is a full subcategory of the category $\QSet(A,B)$. In other words, there is a forgetful $2$-functor $\QGraph \to \QSet$ which is locally fully faithful.
Dualisable morphisms in
------------------------
We now show that a quantum isomorphism is nothing other than a dagger-dualisable (see Definition \[def:daggerdualisable\]) quantum homomorphism.
\[thm:dualisablequantumgraph\] A quantum homomorphism is dagger-dualisable in $\QGraph$ if and only if it is a quantum isomorphism.
Let $(A,G)$ and $(A',G')$ be quantum graphs and let $(H,P)$ be a dagger-dualisable quantum homomorphism between them. In particular, $(H,P)$ is a dagger-dualisable quantum function — and therefore by Theorem \[thm:dual2\] a quantum bijection. Moreover, the dual $(H^*,\conj{P})$, defined in , is a quantum homomorphism from $(A',G')$ to $(A,G)$, i.e. it fulfills the following equation: $$\label{eq:dualisquantumhom}
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$\conj{P}$} (0,5);
\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$\conj{P}$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G'$};
\draw[string, arrow data={0.08}{<}, arrow data ={0.5}{<},arrow data ={0.95}{<}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
~~~=~~~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$\conj{P}$} (0,5);
\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$\conj{P}$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G'$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$G$};
\draw[string, arrow data={0.08}{<}, arrow data ={0.5}{<},arrow data ={0.95}{<}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}$$ Recalling and using , we rotate this equation by 180 degree to the left and obtain: $$\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$G'$};
\draw[string, arrow data={0.08}{>}, arrow data ={0.5}{>},arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
~~~=~~~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string] (0, 4) to (\d,4);
\node[zxvertex=\zxwhite] at (0,4){};
\node[zxvertex=\zxwhite] at (\d,4){};
\node[zxnode=\zxwhite] at (\d/2, 4 ) {$G'$};
\draw[string, arrow data={0.08}{>}, arrow data ={0.5}{>},arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}
~~~\superequalseq{eq:quantumgraphhomomorphism}~~~
\begin{tz}[zx,yscale=0.8,xscale=-1]
\draw[string] (0,-0.5) to node[front,zxnode=\zxwhite,pos=0.57] {$P$} (0,5);\draw[string] (\d,-0.5) to node[front,zxnode=\zxwhite, pos=0.43] {$P$} (\d,5);
\draw[string] (0, 0.5) to (\d,0.5);
\node[zxvertex=\zxwhite] at (0,0.5){};
\node[zxvertex=\zxwhite] at (\d,0.5){};
\node[zxnode=\zxwhite] at (\d/2, 0.5 ) {$G$};
\draw[string, arrow data={0.08}{>},arrow data={0.5}{>}, arrow data ={0.95}{>}] (3, -0.5) to (3,0.5) to [out=up, in=down] (-1.,4) to (-1.,5);
\end{tz}$$ Contracting with the unit on the bottom left and the counit on the top right recovers condition .
For the converse, we note that a quantum isomorphism in the sense of Definition \[def:quantumgraphiso\] is in particular a quantum bijection and therefore commutes both with comultiplication , multiplication and the graphs . Therefore, it is clearly a quantum homomorphism whose dual is also a quantum homomorphism .
Following Remark \[rem:monoidalsubcats\], we can think of the monoidal categories\
$\QIso((A,G),(A,G))$ of quantum automorphisms of a quantum graph $(A,G)$ as a noncommutative or quantum version of the automorphism group of this quantum graph.=-2
\[rem:Banicagraph\] Based on Wang’s ‘quantum symmetry groups’ [@Wang1998] (see Section \[sec:Wang\]), Banica [@Banica2005] introduced[^17] ‘quantum automorphism groups’ of finite classical graphs as noncommutative variants of their automorphism groups.
We now show that our quantum graph automorphism categories are the categories of finite-dimensional representations of the Hopf $C^*$-algebras corresponding to these quantum automorphism groups. This can again be understood as asserting that our quantum graph automorphisms are quantum elements of these quantum groups (cf. Section \[sec:Wang\] and Theorem \[thm:universalprop\]).
\[prop:Banica\] Let $(V_G,G)$ be a classical graph. The category $\QIso((V_G,G), (V_G,G))$ is the category of finite-dimensional representations of Banica’s ‘quantum automorphism group’ algebra $A_{aut}(G)$ of the graph $G$ (see e.g. [@Banica2007_2 Definition 2.1]).
The algebra $A_{aut}(G)$ is defined as the $C^*$-algebra with generators $a_{i,j}~(i,j=1,\ldots,n)$ and relations (cf. Theorem 2.1 and Example 2.2 in [@Banica2005]):
a\_[i,j]{}\^2=a\_[i,j]{} = a\_[i,j]{}\^\* & \_[i=1]{}\^n a\_[i,j]{} = 1, 1jn & \_[j=1]{}\^n a\_[i,j]{} =1, 1in
\_[k=1]{}\^n G\_[i,k]{} a\_[k,j]{} = \_[k=1]{}\^n a\_[i,k]{} G\_[k,j]{}
Here $n$ is the number of vertices of the graph $G$ and $\{G_{i,j}\}_{1\leq i,j \leq n}$ is the adjacency matrix of $G$.=-2
Similar to the proof of Proposition \[prop:Wang\], comparing these relations with equations , and shows that $\QIso((V_G,G),(V_G,G))$ is the category of finite-dimensional representations of $A_{aut}(G)$.
The semisimple structure of quantum functions {#sec:semisimple}
=============================================
In this section, we investigate more closely the categories of quantum functions $\QSet(A,B)$ between quantum sets $A$ and $B$ and show that they are *semisimple*. This sheds light on the role played by the intertwiners and is crucial to our understanding of the structure of quantum functions, particularly with a view towards distinguishing genuinely quantum functions from essentially classical ones.
Semisimplicity is a categorical structure which commonly arises in representation theory. The paradigmatic example of a semisimple category is the category of finite-dimensional representations of a finite group; in this category there are a number of *irreducible* representations and every representation splits as a direct sum of these ‘simple objects’. As we will now show, categories of quantum functions are similar; every quantum function from $A$ to $B$ may be decomposed into a direct sum of irreducible quantum functions.
The direct sum of quantum functions {#sec:dirsumqfcts}
-----------------------------------
The *direct sum* of two quantum functions $(H,Q)$ and $(H',P)$ is defined as $(H\oplus H', Q\oplus P)$, where $Q\oplus P$ is the following direct sum of the underlying linear maps:=-2
\[eq:directsum\]
(0,1) to (2,3); (2,1) to node\[zxnode=, pos=0.5\][$Q\oplus P$]{} (0,3); at (0,1) [$A$]{}; at (0,3) [$H\oplus H'$]{}; at (2,3) [$B$]{}; at (2,1) [$H\oplus H'$]{};
=
(0,1) to (2,3); (2,1) to node\[zxnode=, pos=0.5\][$Q$]{} (0,3); at (0,1) [$A$]{}; at (0,3) [$H$]{}; at (2,3) [$B$]{}; at (2,1) [$H$]{};
(0,1) to (2,3); (2,1) to node\[zxnode=, pos=0.5\][$P$]{} (0,3); at (0,1) [$A$]{}; at (0,3) [$H'$]{}; at (2,3) [$B$]{}; at (2,1) [$H'$]{};
In particular, if $A$ and $B$ are sets, and $(H,P)$ and $(H',Q)$ are quantum functions from $A$ to $B$, then the direct sum of the corresponding matrices of projectors has underlying Hilbert space $H\oplus H'$ and projectors: $$\left(P \oplus Q\right)_{a,b} = P_{a,b} \oplus Q_{a,b}$$ We now remark on the operational interpretation of the direct sum of quantum =-2
\[rem:opintofdirsum\] In Remark \[rem:opintofquantumfunctions\], we observed that a quantum function between classical sets is a projective measurement controlled by an input element, whose result determines an output element. Operationally, a quantum function is a direct sum precisely when the following procedures coincide:
- Perform the quantum measurement $(H \oplus H'\!, [P \oplus Q]_{a,-})$ depending on a received input $a$.=-2
- Before receiving any input, perform a projective measurement onto $H$ and $H'$, and then, depending on the outcome, perform the quantum measurement $(H,P_{a,-})$ or $(H',Q_{a,-})$ upon receiving the input $a$.
In the setting of nonlocal games [@Mancinska2016; @Atserias2016], this can be understood as corresponding to a probabilistic mixture of quantum strategies.
On the other hand, a quantum function $P$ is simple if it cannot be decomposed in this way; or equivalently, if there are no non-trivial intertwiners $P \to P$.
\[def:simpleobject\] A quantum function $P$ is *simple* if $\QSet(P,P) \cong \mathbb{C}$.
For a quantum function $P$ between classical sets, simplicity translates into the following condition on the corresponding matrix of projectors:
\[eq:simplePPM\]X (H): P\_[a,b]{}X = XP\_[a,b]{} aA, bB X\_[H]{}
The category of quantum functions $\QSet(A,B)$ is completely determined by the simple quantum functions, as the following theorem shows.
\[thm:semisimple\] For quantum sets $A$ and $B$, $\QSet(A,B)$ is a semisimple dagger category.
We defer the proof of this theorem to Section \[app:semisimplicity\]. It is completely independent of the following Sections \[sec:dirsumqfcts\] and \[sec:fusionstructure\].
The definition of semisimplicity is given in Section \[app:semisimplicity\]; it implies that every quantum function is a finite direct sum of simple quantum functions, which cannot be decomposed any further.
In particular, every classical function is a simple quantum function (cf. Remark \[rem:setsubset\]).
In contrast to many semisimple categories considered in the , most of the semisimple categories we consider have an infinite number of =-2
The only exceptions are the categories of quantum elements (and of course the categories of quantum functions to a one-element set which have a unique simple object).
Every quantum set has a finite number of simple quantum elements.
The category $\QEl(A) = \QSet(*,A)$ is the category of comodules of $A$ and as such the opposite of the category of modules of $A$. Since every symmetric special Frobenius algebra is semisimple (in the algebraic sense) there is only a finite number of inequivalent irreducible representations — these are the simple quantum elements.
\[prop:simpleelsofclasssetareordinaryels\] Simple quantum elements of a classical set are ordinary elements.
A classical set is a commutative $C^*$-algebra. In particular, all its irreducible representations are one-dimensional, i.e. classical elements.
There is a close connection between semisimplicity and classicality of quantum functions, which we capture by the following definition.
\[def:classical\] A quantum function $Q:X\to Y$ between classical sets $X$ and $Y$ is *essentially classical* if it is a direct sum of classical functions, i.e. of one-dimensional quantum functions. A quantum element of a classical set is *essentially classical* if it is essentially classical as a quantum function from the one-element set.
In other words, an essentially classical quantum function is of the following form, where $\{\ket{i}\}$ is an orthonormal basis corresponding to the decomposition of the Hilbert space $H$ into one-dimensional subspaces $H\cong \bigoplus_{i} \mathbb{C} \ket{i}$ and $f_i:X\to Y$ are classical functions:$$\label{eq:classicalquantumfunction}\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (2,3);
\draw[arrow data ={0.2}{>}, arrow data={0.8}{>}] (2,0) to (0,3);
\node[zxnode=\zxwhite] at (1,1.5) {$P$};
\node[dimension,left] at (0,0) {$X$};
\node[dimension,right] at (2,3) {$Y$};
\node[dimension,right] at (2,0) {$H$};
\node[dimension,left] at (0,3){$H$};
\end{tz}
~=~\sum_i~\begin{tz}[zx,every to/.style={out=up, in=down},xscale=-1]
\draw (0,0) to (2,3);
\draw[arrow data ={0.8}{>}] (0,2.) to (0,3);
\draw[arrow data ={0.4}{>}] (2,0) to (2,1);
\node[zxnode=\zxwhite] at (1,1.5) {$f_i$};
\node[zxnode=\zxwhite] at (0,2.) {$i$};
\node[zxnode=\zxwhite] at (2,1) {$i^\dagger$};
\end{tz}$$ The following proposition justifies Definition \[def:classical\].
\[prop:classicalcommute\]A quantum function $P:X\to Y$ between classical sets $X$ and $Y$ is essentially classical if and only if all corresponding projectors $P_{xy}$ commute with each other.
Commuting projectors are simultaneously diagonalizable; all the projectors therefore split as a direct sum over the shared eigenspaces. Conversely, every essentially classical quantum function decomposes into a direct sum of classical functions; all projectors $P_{xy}$ are therefore sums of projectors onto the orthonormal basis vectors corresponding to this decomposition.
In the terminology of Definition \[def:classical\], every projective measurement — thought of as a quantum function from the one-element set to its outcome set — is essentially classical. Indeed, non-classical behaviour only becomes apparent if there are at least two projective measurements; that is, when the quantum function goes from a set with two or more elements to an outcome set. Only in these cases does it make sense to distinguish between essentially classical quantum functions, which correspond to measurements of commuting observables, and non-classical quantum functions. From the nonlocal game perspective [@Mancinska2016], essentially classical quantum graph homomorphisms are mixtures of classical homomorphisms that only use the available quantum ressource as a generator of shared classical randomness.
In this terminology, every Latin square and all quantum elements of a classical set are essentially classical, while the quantum Latin square in and the quantum elements of certain quantum sets are not. In fact, we can prove the following simple lemma.
Quantum sets whose quantum elements are all essentially classical are classical sets.
All simple quantum elements of such a quantum set are one-dimensional; it follows that the corresponding algebra is commutative, i.e. an ordinary set.
The categories of quantum graph homomorphisms $\QGraph((A,G),(A',G'))$ are also semisimple (see Corollary \[cor:semisimplegraph\]). The following proposition shows that a quantum graph homomorphism is simple precisely when its underlying quantum function is.
\[prop:graphhomsimpleifffunctionsimple\] A quantum graph homomorphism $Q:(A,G)\to (A',G')$ is simple in $\QGraph((A,G),(A',G'))$ if and only if its underlying quantum function $Q:A\to A'$ is simple in $\QSet(A,A')$. Moreover, suppose that $Q:(A,G)\to (A',G')$ is a quantum homomorphism and that $Q$ has a decomposition $Q\cong \bigoplus_i f_i$ into simple quantum functions $f_i$ in $\QSet(A,A')$. Then, each $f_i$ is a quantum graph homomorphism $(A,G)\to (A,G')$ and the decomposition is a decomposition in $\QGraph((A,G),(A',G'))$.
The proposition follows from the fact that $\QGraph((A,G),(A',G')) \subseteq \QSet(A,A')$ is a full inclusion of semisimple categories (see Remark \[rem:last\]).
Therefore, the question of semisimplicity of a quantum graph homomorphism is simply a question of semisimplicity of the underlying quantum function and does not depend on the graphs. Despite this result, finding the simple objects in the category of quantum homomorphisms between two quantum graphs is quite involved.
The fusion structure of quantum bijections {#sec:fusionstructure}
------------------------------------------
The category $\QSet(A,A)$ of quantum functions from a quantum set $A$ to itself possesses various other structures which interact with the semisimple structure. In particular, under composition of quantum functions, it becomes a *monoidal* semisimple dagger category. If we restrict further to quantum bijections $\QBij(A,A)$, then by the results of Section \[sec:quantumbijection\] we obtain a monoidal semisimple dagger category with dualisable objects[^18] (see Theorem \[thm:dual2\]). For a finite number of simple objects such a structure is known as a *unitary fusion category*; such structures generalise finite groups, and have been the subject of considerable research in recent years [@Etingof2015].
As noted in Section \[sec:Wang\], these quantum permutation categories have been considered in finite noncommutative topology. We restate some of the results of this work now.=-2
All quantum bijections on an $n$-element classical sets with $n\leq 3$ are essentially classical.
By Proposition \[prop:classicalcommute\], we need to show that all projectors in a projective permutation matrix of size $1,2$ and $3$ commute. For $n=1$, this is trivial. For $n=2$, this follows from the fact that every projective permutation matrix is of the form $$\left(\begin{array}{cc} p& 1-p\\ 1-p & p \end{array} \right)$$ for some projector $p$. The proof for $n=3$ is rather involved and can be found in [@Banica2007].
Note that there are quantum bijections between 4-element classical sets that are not essentially classical; these include the quantum Latin square of dimension 4 in Example \[ex:latinsquaresexample\], and the following projective permutation matrix where $p$ and $q$ are non-commuting projectors on a 2-dimensional Hilbert space: $$\begin{pmatrix}p&1-p&0&0\\1-p&p&0&0\\0&0&q&1-q\\0&0&1-q&q\end{pmatrix}$$ In fact, the general structure of the category of quantum bijections between 4-element sets has been investigated by Banica and Bichon [@Banica2009], leading in particular to the following surprising result regarding the dimension of quantum bijections (recall Definition \[def:qfctdimension\]).
Every simple quantum bijection between 4-element sets must have dimension 1,2 or 4.
Similarly, the monoidal semisimple dagger categories of quantum graph automorphisms $\QIso(G,G)$ have been studied [@Banica2005; @Banica2007_2; @Banica2007_3; @Bichon2003], and several graphs with only essentially classical quantum automorphisms have been identified.
Essentially classical quantum bijections (in the sense of Definition \[def:classical\]) on the $n$-element set $[n]$ form a full fusion subcategory of $\QBij([n], [n])$ equivalent to the category of $S_n$-graded vector spaces [@Etingof2015], where $S_n$ is the symmetric group (also cf. Remark \[rem:setsubset\]). Similarly, essentially classical quantum graph automorphisms on a classical graph $G$ form a full fusion subcategory of $\QIso(G,G)$ equivalent to the category of $\Aut(G)$-graded vector spaces, where $\Aut(G)$ is the automorphism group of $G$.
Proof of semisimplicity {#app:semisimplicity}
-----------------------
In the following section, we show that the hom-categories $\QSet(A,B)$ for quantum sets $A$ and $B$ are semisimple. Recall that a linear category (that is, a category enriched in vector spaces over some field) is semisimple [@Mueger2003], if it has direct sums, if all idempotents split and if there are objects $X_i$ (called *simple objects*) labeled by an indexing set $I$ such that $\Hom(X_i,X_j) \cong \delta_{i,j} \mathbb{C}$ and such that every object $V$ is isomorphic to a finite direct sum of simple objects. For dagger categories such as those in this paper, this notion of semisimplicity is subsumed by more fundamental properties.
A *linear dagger category* is a dagger category enriched in finite-dimensional complex vector spaces such that for objects $X$ and $Y$ the induced function $\dagger: \Hom(X,Y) \to \Hom(Y,X)$ is antilinear. A linear dagger category is *positive*, if for every morphism $A\to[r]B$:$$r^\dagger r= 0 \hspace{1cm}\Rightarrow \hspace{1cm}r=0$$
\[thm:daggerpositive\]For two quantum sets $A$ and $B$, the category $\QSet(A,B)$ is a positive linear dagger category.
By Theorem \[thm:QSetdagger2category\], $\QSet(A,B)$ is a dagger category. Linearity and positivity follow from the existence of the forgetful dagger functor
Following Müger, Roberts and Tuset, we define semisimplicity of such a dagger category as follows [@Mueger2004 Definition 2.2].
A *semisimple dagger category* is a linear dagger category for which the following hold:
1. It is positive.
2. It has a zero object; that is an object such that every object has a unique morphism into and out of it.
3. It has binary direct sums (cf. [@Vicary2011], there called dagger biproducts): For objects $X_1$ and $X_2$, there is an object $X_1\oplus X_2$ and morphisms $s_i:X_i \to X_1\oplus X_2$ ($i=1,2$) such that the following hold:
s\_i\^s\_i = \_[X\_i]{} i=1,2& s\_1s\_1\^+s\_2s\_2\^=\_[X\_1X\_2]{}
4. All dagger idempotents split (cf. [@Selinger2008]): A dagger idempotent is an endomorphism $p\in \End(X)$ on some object $X$ such that $p^2=p =p^\dagger$. It splits if there is a morphism $i: A\to X$ out of some object $A$ such that the following hold:
i\^i = \_A & i i\^= p
An object $X$ in a semisimple dagger category is *simple*, if $\Hom(X,X)\cong \mathbb{C}$.
It is well known that semisimple dagger categories are semisimple in the usual sense (for a proof for monoidal semisimple dagger categories, see e.g. [@Mueger2007 Lemma 1.35]). For completeness, we sketch a version of this argument.
In a semisimple dagger category, non-isomorphic simple objects $X$ and $Y$ are disjoint ($\Hom(X,Y) = \{0\}$) and every object is isomorphic to a finite direct sum of simple objects.
It is a straightforward consequence of positivity that non-isomorphic simple objects are disjoint. Moreover, given an object $X$, then the dagger makes $\Hom(X,X)$ into a positive $*$-algebra. Since every positive $*$-algebra is semisimple, it follows that if $X$ is not simple, there is a non-trivial dagger idempotent in $\Hom(X,X)$ which we can split to obtain an isometry $i:A\to X$ and use this to decompose $X\cong A \oplus A^\bot$. Continuing inductively, every non-simple object can be decomposed into a finite sum of simple objects.
We are now ready to restate and prove Theorem \[thm:semisimple\].
**Theorem \[thm:semisimple\].** *For quantum sets $A$ and $B$, $\QSet(A,B)$ is a semisimple dagger category.*
1\. Positivity of $\QSet(A,B)$ is proven in Proposition \[thm:daggerpositive\].
2\. The zero object is the quantum function $(\mathbf{0}, 0)$, where $\mathbf{0}$ is the zero-dimensional Hilbert space and $0$ is the unique morphism $0: \mathbf{0} \otimes A \to B \otimes \mathbf{0}$.
3\. The direct sum of two quantum functions is defined in and inherits all its structural properties from $\Hilb$.
4\. A dagger idempotent in $\QSet(A,B)$ is a self-adjoint idempotent endomorphism $r: H \to H$ on a quantum function $Q:H\otimes A \to B \otimes H$:
\[eq:idempotentsplit\]
(0,0) to (2,2); (2,0) to node\[zxnode=,pos=0.5\] [$Q$]{} node\[zxnode=, pos=0.2\] [$r$]{} (0,2);
=
(0,0) to (2,2); (2,0) to node\[zxnode=,pos=0.5\] [$Q$]{} node\[zxnode=, pos=0.8\] [$r$]{} (0,2);
In particular, we can split this idempotent in , obtaining an isometry $i:V\to H$ such that $ii^\dagger = r$. It then follows from that
(0,0) to (2,2); (2,0) to node\[zxnode=,pos=0.5\] [$Q'$]{} (0,2); at (2,0) [$V$]{}; at (0,2) [$V$]{};
:=
(0,0) to (2,2); (2,0) to node\[zxnode=,pos=0.5\] [$Q$]{} node\[zxnode=, pos=0.2\] [$i$]{}node\[zxnode=, pos=0.8\] [$i^\dagger$]{} (0,2); at (2,0) [$V$]{}; at (0,2) [$V$]{};
is itself a quantum function and the isometry $i:V\to H$ is an intertwiner of quantum functions $ Q' \to Q$. Therefore, $r$ splits as an idempotent in $\QSet(A,B)$.
Essentially the same proof works for the 2-category $\QGraph$.
\[cor:semisimplegraph\]For quantum graphs $(A,G)$ and $(B,H)$, the category $\QGraph((A,G),(B,H))$ is semisimple.
Quantum graphs and quantum relations {#app:quantumrelation}
====================================
In this final section, we discuss the relational approach to quantum graphs taken by Kuperberg, Weaver, Duan, Severini, Winter, and others [@Kuperberg2012; @Weaver2010; @Weaver2015; @Duan2013; @Ortiz2016] and compare it to our Definition \[def:quantumgraphsbyadjmats\]. We show that our quantum graphs can indeed be understood as symmetric and reflexive quantum relations as defined by Kuperberg and Weaver [@Kuperberg2012; @Weaver2015] and therefore generalise the noncommutative graphs of Duan, Severini and Winter [@Duan2013].
In this section, to fit with the work of other authors, all graphs and quantum graphs will be reflexive (see Definition \[def:quantumgraphsbyadjmats\]); wherever ‘graph’ is written it should be taken to mean ‘reflexive graph’.
The relational approach starts from the observation that a classical graph $G$ may be described as a reflexive and symmetric relation on its vertex set $V_{G}$; that is, a subset $S\subseteq V_G \times V_{G}$ containing both the diagonal $\Delta=\{(x,x)~|~x\in V_G\}$ and such that $(x,y)\in S\Leftrightarrow (y,x) \in S$ for all $x,y\in V_{G}$.
Quantising this description leads to a notion of quantum graph equivalent to Definition \[def:quantumgraphsbyadjmats\] which is closer in spirit to previous definitions [@Duan2013; @Weaver2010; @Weaver2015]. We first quantise relations, recovering the finite-dimensional quantum relations of Kuperberg and Weaver [@Weaver2010; @Kuperberg2012].
Under Gelfand duality, a binary relation between two finite sets $X$ and $Y$ can be expressed as a projector $\!P\!:X\! \otimes Y \!\!\to \!X \otimes Y$ satisfying the following equation:=-2 $$\label{eq:bimodulerelation}
\def\scl{1.25}
\begin{tz}[zx,every to/.style={out=up, in=down},scale=\scl]
\draw (0,0) to (0,2);
\draw (1,0) to (1,2);
\draw (-0.5,0) to[in=-135] (0,0.6);
\draw (1.5,0) to [in=-45] (1,0.6);
\boxwidth{1.25}
\node[widebox] at (0.5,1.25) {$P$};
\node[zxvertex=\zxwhite] at (0,0.6){};
\node[zxvertex=\zxwhite] at (1,0.6){};
\node[dimension,left] at (0,-0.1) {$X$};
\node[dimension, left] at (-0.5,-0.1) {$X$};
\node[dimension,right] at (1.5,-0.1) {$Y$};
\node[dimension, right] at (1,-0.1) {$Y$};
\node[dimension, left] at (0., 2) {$X$};
\node[dimension, right] at (1, 2) {$Y$};
\end{tz}
~~=~~
\begin{tz}[zx,every to/.style={out=up, in=down},scale=\scl]
\draw (0,0) to (0,2);
\draw (1,0) to (1,2);
\draw (-0.5,0) to[in=-150] (0,1.4);
\draw (1.5,0) to [in=-30] (1,1.4);
\boxwidth{1.25}
\node[widebox] at (0.5,0.75) {$P$};
\node[zxvertex=\zxwhite] at (0,1.4){};
\node[zxvertex=\zxwhite] at (1,1.4){};
\node[dimension,left] at (0,-0.1) {$X$};
\node[dimension, left] at (-0.5,-0.1) {$X$};
\node[dimension,right] at (1.5,-0.1) {$Y$};
\node[dimension, right] at (1,-0.1) {$Y$};
\node[dimension, left] at (0., 2) {$X$};
\node[dimension, right] at (1, 2) {$Y$};
\end{tz}$$
Projectors on $X\otimes Y$ encode subspaces of $X\otimes Y$; the bimodule condition guarantees that the corresponding subspace is the linear span of a *subset* of $X\times Y$.
We define quantum relations analogously.
\[def:quantumrelations\] A *quantum relation* between quantum sets $A$ and $B$ is a projector\
${P: A \otimes B \to A \otimes B}$ satisfying equation . A *quantum relation on a quantum set $A$* is a quantum relation between $A$ and itself.
\[rem:quantumrelationprojection\]Let $A$ and $B$ be finite-dimensional $C^*$-algebras. Given a projector\
$P\in \End(A\otimes B)$ fulfilling equation , we obtain a projection[^19] $p:=P(e_A\otimes e_B)\in A^{op}\otimes B$; conversely, any projection $p\in A^{op}\otimes B$ gives rise to a projector $P:=L_p \in \End(A\otimes B)$ fulfilling . (Here, $A^{op}$ denotes the opposite algebra of $A$, $e_A\in A,~e_B\in B$ denote the units of $A$ and $B$, and $L_p$ denotes left multiplication with $p$ in the algebra $A^{op}\otimes B$.) A quantum relation between $A$ and $B$ may therefore be equivalently defined as a projection in the algebra $A^{op} \otimes B$. Plugging units into the second and third slot of equation gives a simple diagrammatic proof of this fact.
Kuperberg and Weaver [@Weaver2010; @Kuperberg2012] define a quantum relation on a von Neumann algebra $\mathcal{M}\subseteq \mathcal{B}(H)$ to be a weak$^*$-closed operator bimodule over the commutant $\mathcal{M}'$ (see Definition 2.1 in [@Weaver2010]). In the finite-dimensional case (see Definition 5.1 in [@Weaver2015]), this definition reduces to the following:
\[KuperbergWeaverQreldef\]Let $A$ be a finite-dimensional $C^*$-algebra, let $H$ be some finite-dimensional Hilbert space such that $A\subseteq \End(H)$ and let $A'=\{b \in \End(H)~|~ ba=ab~\forall a\in A\}$ be the commutant with respect to this embedding. A *quantum relation in the sense of Kuperberg and Weaver* is a subspace $V\subseteq \End(H)$ which fulfills $A' V A'\subseteq V$.
It is shown in Theorem 2.7 of [@Weaver2010] that this definition is independent[^20] of the embedding $A\subseteq \End(H)$. We now show that, in the finite-dimensional case, this definition is equivalent to our own.
\[prop:WeaverKuperberg\] Given a finite-dimensional $C^*$-algebra $A$, there is a Hilbert space $H$ and an embedding $A\subseteq \End(H)$ such that our notion of quantum relation on $A$ (Definition \[def:quantumrelations\]) coincides with that of Kuperberg and Weaver (Definition \[KuperbergWeaverQreldef\]).
As noted in Remark \[rem:quantumrelationprojection\], our quantum relations on a finite-dimensional $C^*$-algebra $A$ can be expressed as projections in $A^{op}\otimes A$; Weaver proves in [@Weaver2010 Proposition 2.23] that quantum relations in the sense of Kuperberg-Weaver on finite-dimensional $C^*$-algebras correspond precisely to such projections.
For concreteness, we will now provide an alternative proof using the Frobenius structure on the $C^*$-algebra. We have seen in Section \[sec:diagramsforC\*\] that a finite-dimensional $C^*$-algebra $A$ admits the structure of a symmetric special dagger Frobenius algebra. In particular, $A$ admits the structure of a Hilbert space, and there is an embedding of $C^*$-algebras $A\subseteq \End(A)$ given as follows:
A(A) & aL\_a (:bA ab)
We now show that the corresponding commutant $A'$ is $*$-isomorphic to the opposite algebra $A^{op}$, defined on the vector space $A$ with multiplication $a\star b:= ba$. The following linear map is clearly a faithful $*$-homomorphism:
A\^[op]{} A’ & aR\_a(:bA ba)
We now show that it is also surjective and therefore a $*$-isomorphism. Let $X:A\to A$ be an element of $A'$, and denote the unit of $A$ by $e\in A$. Then $X(e)\in A$ and we will show that $R_{X(e)}=X$. For all $b\in A$ the following holds: $$R_{X(e)} (b) = b X(e) = L_b X(e) ~~\superequals{$X\in A'$}~~ X L_b(e) = X(b)$$ A quantum relation in the sense of Kuperberg and Weaver is therefore a subspace $\mathcal{V} \subseteq \End(A)$ such that $R_a\mathcal{V} R_b \subseteq \mathcal{V}$ for all $a,b\in A$. Since $A$ is a symmetric Frobenius algebra, there is a canonical isomorphism $A\cong A^*$ which induces an isomorphism $\End(A) \cong A \otimes A$. Under this isomorphism, left and right composition with $R_a$ in $\End(A)$ can be identified with left and right multiplication with $(a\otimes e) $ and $(e\otimes a)$ in $A\otimes A$, respectively:
R\_a - : (A) (A) & -R\_b: (A) (A)\
(ae) - : AAAA & -(eb) : AA AA
Therefore, a quantum relation in the sense of Kuperberg and Weaver can be understood as a subspace $\mathcal{V} \subseteq A\otimes A $ such that $(a\otimes e)\mathcal{V}(e\otimes b)\subseteq \mathcal{V}$ for all $a,b\in A$.
This coincides with our Definition \[def:quantumrelations\].
Following Weaver [@Weaver2015], we now consider additional properties of quantum relations, such as symmetry and reflexivity.
\[def:quantumgraphsbyrelations\] A quantum relation $P:A\otimes A\to A \otimes A$ is *symmetric* or *reflexive* if one of the following additional equations holds:
\[eq:symrefrelation\]
(0,0.5) to (0,1.) to \[out=up, in=up,looseness=1.75\] node\[zxvertex=, pos=0.5\] (2.5,1.) to (2.5,-1); (1,0.5) to (1,1.) to \[out=up, in=up,looseness=3\] node\[zxvertex=, pos=0.5\](1.5,1.) to (1.5,-1); (-0.5,2.5) to (-0.5,0.5) to \[out=down, in=down, looseness=3\] node\[zxvertex=, pos=0.5\] (0,0.5); (-1.5,2.5) to (-1.5,0.5) to \[out=down, in=down, looseness=1.75\]node\[zxvertex=, pos=0.5\] (1,0.5); at (0.5,0.75) [$P$]{};
=
(0,0) to (0, 3); (1,0) to (1,3); at (0.5, 1.5)[$P$]{};
&
(-0.3,0) rectangle (1.3,3); (0,3) to (0,1) to \[out=down, in=down\] node\[zxvertex=, pos=0.5\] (1,1) to (1,3); at (0.5,2)[$P$]{};
=
(-0.2,0) rectangle (1.2,3); (0,3) to (0,2) to \[out=down, in=down\]node\[zxvertex=, pos=0.5\] (1,2) to (1,3);
\
&
We now show that quantum graphs as in Definition \[def:quantumgraphsbyadjmats\] are indeed symmetric and reflexive quantum relations. For a quantum graph $(A,G)$ as in Definition \[def:quantumgraphsbyadjmats\], we introduce the following linear map $P_G:A\otimes A \to A \otimes A$: $$\label{eq:graphprojector}
\begin{tz}[zx]
\draw[string] (0, 0) to (0,3);
\draw (1.5,0) to (1.5, 3);
\boxwidth{1.5}
\node[widebox] at (0.75, 1.5 ) {$P_G$};
\end{tz}
~~~=~~~
\begin{tz}[zx]
\draw[string] (0, 0) to (0,3);
\draw (2.5,0) to (2.5, 3);
\draw (0, 1.5) to (2.5, 1.5) ;
\node[zxvertex=\zxwhite] at (0,1.5){};
\node[zxvertex=\zxwhite] at (2.5,1.5){};
\node[zxnode=\zxwhite] at (1.25, 1.5 ) {$G$};
\end{tz}
~~~:=~~~
\begin{tz}[zx,xscale=2.5/3]
\draw (-1.5,0) to (-1.5,1.5) to [out=up, in=-135] (-0.75, 2.5) to (-0.75,3);
\draw (-0.75, 2.5) to [out=-45, in=up] (0, 1.5) to [out=down, in=135] (0.75,0.5) to (0.75,0);
\draw (0.75,0.5) to [out=45, in=down] (1.5,1.5) to (1.5,3);
\node[zxnode=\zxwhite] at (0, 1.5) {$G$};
\node[zxvertex=\zxwhite, zxdown] at (-0.75, 2.5) {};
\node[zxvertex=\zxwhite, zxup] at (0.75, 0.5) {};
\end{tz}
~~~\superequalseq{eq:propadjacency}~~~
\begin{tz}[zx,xscale=-2.5/3]
\draw (-1.5,0) to (-1.5,1.5) to [out=up, in=-135] (-0.75, 2.5) to (-0.75,3);
\draw (-0.75, 2.5) to [out=-45, in=up] (0, 1.5) to [out=down, in=135] (0.75,0.5) to (0.75,0);
\draw (0.75,0.5) to [out=45, in=down] (1.5,1.5) to (1.5,3);
\node[zxnode=\zxwhite] at (0, 1.5) {$G$};
\node[zxvertex=\zxwhite, zxdown] at (-0.75, 2.5) {};
\node[zxvertex=\zxwhite, zxup] at (0.75, 0.5) {};
\end{tz}$$ For a classical graph $(V_G,G)$, it is easily verified that this map is the projector onto the subspace of the symmetric and reflexive relation defining the graph: $$\label{eq:projectorsubset}\left\{ (v,w) ~|~ v \sim_G w \right\} \subseteq V_G \times V_G$$ Conversely, the adjacency matrix $G$ can be recovered from this projector as follows: $$\label{eq:recoveradjacency}
\begin{tz}[zx,scale=\scl]
\draw (0,0) to (0,2);
\node[zxnode=\zxwhite] at (0,1) {$G$};
\end{tz}
~~~:=~~~
\begin{tz}[zx,scale=\scl]
\draw (0,0) to (0,1) to [out=up, in=up,looseness=2.5] node[zxvertex=\zxwhite, pos=0.5] {} (-0.5,1) to (-0.5,-0.25);
\draw (0.75,0) to (0.75,1.75);
\boxwidth{1}
\node[widebox] at (0.375,0.65){$P_G$};
\node[zxvertex=\zxwhite] at (0,0){};
\node[zxvertex=\zxwhite] at (0.75,0){};
\end{tz}
~~~\superequalseq{eq:bimodulerelation}~~~
\begin{tz}[zx,scale=\scl]
\draw (0,-0.25) to (0,1.3);
\draw (0.75,0) to (0.75,1.75);
\boxwidth{1}
\node[widebox] at (0.375,0.65){$P_G$};
\node[zxvertex=\zxwhite] at (0,1.3){};
\node[zxvertex=\zxwhite] at (0.75,0){};
\end{tz}
~~~\superequals{$(-)^\dagger$}~~~
\begin{tz}[zx,xscale=-1,scale=\scl]
\draw (0,-0.25) to (0,1.3);
\draw (0.75,0) to (0.75,1.75);
\boxwidth{1}
\node[widebox] at (0.375,0.65){$P_G$};
\node[zxvertex=\zxwhite] at (0,1.3){};
\node[zxvertex=\zxwhite] at (0.75,0){};
\end{tz}
~~~\superequalseq{eq:bimodulerelation}~~~
\begin{tz}[zx,xscale=-1,scale=\scl]
\draw (0,0) to (0,1) to [out=up, in=up,looseness=2.5] node[zxvertex=\zxwhite, pos=0.5] {} (-0.5,1) to (-0.5,-0.25);
\draw (0.75,0) to (0.75,1.75);
\boxwidth{1}
\node[widebox] at (0.375,0.65){$P_G$};
\node[zxvertex=\zxwhite] at (0,0){};
\node[zxvertex=\zxwhite] at (0.75,0){};
\end{tz}$$ The same correspondence holds for general quantum graphs.
\[thm:relationadjacency\]Given a quantum graph $(A,G)$ as in Definition \[def:quantumgraphsbyadjmats\], the projector\
[$P_G:A\otimes A\to A \otimes A$]{} defined in equation is a symmetric and reflexive quantum relation as in Definition \[def:quantumrelations\]. Conversely, given a symmetric and reflexive quantum relation $P$ on $A$, the map defines a quantum adjacency matrix. These two constructions are mutually inverse.
Given an arbitrary linear map $G:A\to A$, we define the following linear map $P_G:A\otimes A\to A \otimes A$ fulfilling :$$\begin{tz}[zx]
\draw[string] (0, 0) to (0,3);
\draw (1.5,0) to (1.5, 3);
\boxwidth{1.5}
\node[widebox] at (0.75, 1.5 ) {$P_G$};
\end{tz}
~~~:=~~~
\begin{tz}[zx,xscale=2.5/3]
\draw (-1.5,0) to (-1.5,1.5) to [out=up, in=-135] (-0.75, 2.5) to (-0.75,3);
\draw (-0.75, 2.5) to [out=-45, in=up] (0, 1.5) to [out=down, in=135] (0.75,0.5) to (0.75,0);
\draw (0.75,0.5) to [out=45, in=down] (1.5,1.5) to (1.5,3);
\node[zxnode=\zxwhite] at (0, 1.5) {$G$};
\node[zxvertex=\zxwhite, zxdown] at (-0.75, 2.5) {};
\node[zxvertex=\zxwhite, zxup] at (0.75, 0.5) {};
\end{tz}$$ Conversely, given a linear map $P_G:A\otimes A \to A\otimes A$ fulfilling , we define $$\begin{tz}[zx,scale=\scl]
\draw (0,0) to (0,2);
\node[zxnode=\zxwhite] at (0,1) {$G$};
\end{tz}
~~~:=~~~
\begin{tz}[zx,xscale=-1,scale=\scl]
\draw (0,-0.25) to (0,1.3);
\draw (0.75,0) to (0.75,1.75);
\boxwidth{1}
\node[widebox] at (0.375,0.65){$P_G$};
\node[zxvertex=\zxwhite] at (0,1.3){};
\node[zxvertex=\zxwhite] at (0.75,0){};
\end{tz}$$ It follows easily from equation that these two constructions are mutually inverse. In the following, we say that $G$ is *real* if the following holds: $$\begin{tz}[zx]
\draw (0,0) to (0,3);
\node[zxnode=\zxwhite] at (0,1.5) {$G^\dagger$};
\end{tz}
~~~=~~~
\begin{tz}[zx,xscale=-1]
\draw (1,0) to (1,1.5) to [out=up, in=up, looseness=2.5]node[zxvertex=\zxwhite, pos=0.5] {} (0,1.5) to [out=down, in=down, looseness=2.5]node[zxvertex=\zxwhite, pos=0.5] {} (-1,1.5) to (-1,3);
\node[zxnode=\zxwhite] at (0,1.5) {$G$};
\end{tz}$$ Simple graphical arguments then establish the following:
- $G$ is real if and only if $P_G$ is self-adjoint.
- $G$ fulfills the first equation of if and only if $P_G^2=P_G$.
- $G$ fulfills the second equation of if and only if $P_G$ is symmetric.
- $G$ fulfills the last equation of if and only if $P_G$ is reflexive.
We observed in Remark \[rem:quantumrelationprojection\] that a quantum relation on $A$ can be understood as a projection $p$ in $A^{op}\otimes A $. In terms of the adjacency matrix $G$, the projection corresponding to a quantum graph $(A,G)$ is the following element $p$ of $A\otimes A$: $$\begin{tz}[zx]
\clip (-0.2, -0.8) rectangle (1.2,1.5);
\draw (0,0) to (0,1.5);
\draw (1,0) to (1,1.5);
\boxwidth{1}
\node[widebox] at (0.5,0){$p$};
\end{tz}
~~~=~~~
\begin{tz}[zx]
\clip (-0.2, -0.8) rectangle (1.2,1.5);
\draw (1,1.5) to (1,0) to [out=down, in=down, looseness=2] node[zxvertex=\zxwhite, pos=0.5]{} (0,0) to (0,1.5);
\boxwidth{1}
\node[zxnode=\zxwhite] at (1,0.5){$G$};
\end{tz}$$ In particular, a quantum relation on $A$ given by a projection $p\in A^{op}\otimes A$ is symmetric if $\sigma(p) = p$, where $\sigma:A\otimes A \to A\otimes A$ is the swap map, and reflexive if $m(p)=e_A$, where $m:A\otimes A \to A$ is the multiplication of $A$ and $e_A$ is the unit of $A$.
A related notion of quantum graph was introduced and studied by Duan, Severini and Winter [@Duan2013] as a noncommutative quantum version of the *confusability graphs* in classical zero-error communication. These noncommutative graphs are defined as operator systems on finite-dimensional Hilbert spaces. We now show that such quantum graphs correspond to our quantum graphs on matrix algebras.
\[def:operatorsystem\] Let $H$ be a finite-dimensional Hilbert space. An *operator system* on $H$ is a subspace $\mathcal{V} \subseteq \End(H)$ such that the following hold:
X X\^ & \_H
We prove that every operator system on $H$ gives rise to a quantum graph $(\End(H), G)$ on the endomorphism algebra $\End(H)$ and vice versa.
\[prop:quantumgraphDuanthesame\]Let $H$ be a finite-dimensional Hilbert space. There is a canonical correspondence between quantum graphs $(\End(H),G)$ on the endomorphism algebra of $H$ and operator systems on $H$.
It is shown in [@Weaver2015] that symmetric and reflexive quantum relations (in the sense of Kuperberg and Weaver) on $\End(H)$ coincide with operator systems on $H$. The proposition therefore follows from Proposition \[prop:WeaverKuperberg\] and Theorem \[thm:relationadjacency\], in which we prove that our notion of quantum graph coincides with Kuperberg and Weaver’s symmetric and reflexive quantum relations. For concreteness, we explicitly construct the correspondence between quantum graphs on endomorphism algebras and operator systems.
An operator system $\mathcal{V}\subseteq \End(H)$ can be described in terms of the projector $P_{\mathcal{V}}:H^*\otimes H \to H^*\otimes H$ onto the subspace $\mathcal{V}\subseteq \End(H)\cong H^* \otimes H$. This projector fulfills the following equations:
\[eq:operatorsystemgraphics\]
(0,0.25) to (0,1.25) to \[out=up, in=up,looseness=1.75\] (2.5,1.25) to (2.5,-1); (1,0.25) to (1,1.25) to \[out=up, in=up,looseness=3\] (1.5,1.25) to (1.5,-1); (-0.5,2.5) to (-0.5,0.25) to \[out=down, in=down, looseness=3\] (0,0.25); (-1.5,2.5) to (-1.5,0.25) to \[out=down, in=down, looseness=1.75\] (1,0.25); at (0.5,0.75) [$P_{\mathcal{V}}$]{};
=
(0,0) to (0, 3); (1,0) to (1,3); at (0.5, 1.5)[$P_{\mathcal{V}}$]{};
&
(-0.3,0) rectangle (1.3,3); (0,3) to (0,1) to \[out=down, in=down\] (1,1) to (1,3); at (0.5,2)[$P_{\mathcal{V}}$]{};
=
(-0.2,0) rectangle (1.2,3); (0,3) to (0,2) to \[out=down, in=down\] (1,2) to (1,3);
The first equation corresponds to the property that $X\in \mathcal{V}\Rightarrow X^\dagger \in \mathcal{V}$, the second equation encodes that $\mathbbm{1}_H \in \mathcal{V}$.
Let $(\End(H), G)$ be a quantum graph, expressed as a symmetric, reflexive quantum relation $P_G: \End(H) \otimes \End(H) \to \End(H) \otimes \End(H)$ (see Theorem \[thm:relationadjacency\] and Definition \[def:quantumgraphsbyrelations\]). The bimodule equation can now be expressed as the following first equation; the second equation is obtained from contracting the second wire at the bottom with the third and the sixth with the seventh (here we identify $\End(H) \cong H \otimes H^*$):$$\label{eq:bimodulematrix}
\begin{tz}[zx]
\draw[arrow data={0.15}{<}, arrow data={0.9}{<}] (0,-1.5) to +(0,3);
\draw[arrow data={0.16}{>}, arrow data={0.91}{>}] (1,-1.5) to +(0,3);
\draw[arrow data={0.15}{>}, arrow data={0.91}{>}] (-1.2,-1.5) to [out=up, in=down] (-0.4,0) to (-0.4,1.5);
\draw[arrow data={0.8}{<}] (-0.8, -1.5) to [out=up, in=up, looseness=3] (-0.4,-1.5);
\draw [arrow data={0.15}{<}, arrow data={0.91}{<}](2.2, -1.5) to[out=up, in=down] (1.4,0) to (1.4,1.5);
\draw[arrow data={0.8}{<}] (1.4, -1.5) to [out=up, in=up, looseness=3] (1.8,-1.5);
\boxwidth{1.8}
\node[widebox] at (0.5, 0.5) {$P_G$};
\end{tz}
~~~=~~~
\begin{tz}[zx]
\draw[arrow data={0.15}{<}, arrow data={0.9}{<}] (0,-1.5) to +(0,3);
\draw[arrow data={0.16}{>}, arrow data={0.91}{>}] (1,-1.5) to +(0,3);
\draw[arrow data={0.15}{>}, arrow data={0.91}{>}] (-1.2,-1.5) to (-1.2,0) to [out=up, in=down] (-0.4,1.5);
\draw[arrow data={0.12}{<}, arrow data={0.62}{<}, arrow data={0.91}{<}] (-0.8, -1.5) to (-0.8, 0) to [out=up, in=up, looseness=3] (-0.4,0) to (-0.4, -1.5);
\draw[arrow data={0.15}{<}, arrow data={0.91}{<}] (2.2, -1.5) to (2.2,0) to [out=up, in=down] (1.4,1.5);
\draw[arrow data={0.12}{<}, arrow data={0.41}{<}, arrow data={0.91}{<}] (1.4, -1.5) to (1.4, 0) to [out=up, in=up, looseness=3] (1.8, 0) to (1.8, -1.5);
\boxwidth{1.8}
\node[widebox] at (0.5, -0.5) {$P_G$};
\end{tz}
\hspace{1.25cm} \Rightarrow \hspace{1.25cm}
\dim(H)^2~~
\begin{tz}[zx]
\draw[arrow data={0.15}{<}, arrow data={0.85}{<}] (0,-1.5) to (0,1.5);
\draw[arrow data={0.15}{>}, arrow data={0.85}{>}] (1,-1.5) to +(0,3);
\draw[arrow data={0.15}{>}, arrow data={0.85}{>}] (-0.4,-1.5) to + (0,3);
\draw[arrow data={0.15}{<}, arrow data={0.85}{<}] (1.4,-1.5) to + (0,3);
\boxwidth{1.8}
\node[widebox] at (0.5, 0){$P_G$};
\end{tz}
~~~=~~~
\begin{tz}[zx]
\draw[arrow data={0.15}{>}, arrow data={0.85}{>}] (-1.4, -1.5) to + (0,3);
\draw[arrow data={0.15}{<}, arrow data={0.85}{<}] (2.4,-1.5) to + (0,3);
\draw[arrow data={0.15}{<}, arrow data={0.85}{<}] (0,-1.5) to (0,1.5);
\draw[arrow data={0.15}{>}, arrow data={0.85}{>}] (1,-1.5) to +(0,3);
\draw[arrow data={0.18}{>}, arrow data={0.88}{>}] (-0.4,0) to [out=up, in=up, looseness=5] (-1,0) to [out=down, in=down, looseness=5] (-0.4,0);
\draw[arrow data={0.18}{<}, arrow data={0.88}{<}] (1.4,0) to [out=up, in=up, looseness=5] (2,0) to [out=down, in=down, looseness=5] (1.4,0);
\boxwidth{1.8}
\node[widebox] at (0.5, 0){$P_G$};
\end{tz}$$ Therefore, quantum relations $P_G: \End(H)\otimes \End(H) \to \End(H) \otimes \End(H)$ can be interconverted into projectors $P_\mathcal{V}:H^*\otimes H \to H^*\otimes H$ as follows:
(0,-1.5) to (0,1.5); (1,-1.5) to +(0,3); at (0.5, 0)[$P_{\mathcal{V}}$]{};
=
(0,-1.5) to (0,1.5); (1,-1.5) to +(0,3); (-0.4,0) to \[out=up, in=up, looseness=5\] (-1,0) to \[out=down, in=down, looseness=5\] (-0.4,0); (1.4,0) to \[out=up, in=up, looseness=5\] (2,0) to \[out=down, in=down, looseness=5\] (1.4,0); at (0.5, 0)[$P_G$]{};
&
(0,-1.5) to (0,1.5); (1,-1.5) to +(0,3); (-0.4,-1.5) to + (0,3); (1.4,-1.5) to + (0,3); at (0.5, 0)[$P_G$]{};
=
(0,-1.5) to (0,1.5); (1,-1.5) to +(0,3); (-0.4,-1.5) to + (0,3); (1.4,-1.5) to + (0,3); at (0.5, 0)[$P_{\mathcal{V}}$]{};
It follows from equation that these constructions are mutually inverse. Under this correspondence, the map $P_G$ is a projector if and only if $P_{\mathcal{V}}$ is a projector, $P_G$ is a symmetric quantum relation if and only if $P_\mathcal{V}$ fulfills the first equation of , and $P_G$ is reflexive if and only if $P_{\mathcal{V}}$ fulfills the second equation of .
[^1]: An extensive discussion of related work can be found in Section \[sec:relatedwork\].
[^2]: Here, noncommutative topology refers to the study of $C^*$-algebras in light of Gelfand duality. The term noncommutative geometry is usually reserved for the study of spectral triples (see e.g. [@Connes2000]), a noncommutative version of the theory of Riemannian manifolds.
[^3]: This observation is a version of the famous *Eckmann-Hilton argument* [@Eckmann1962] playing a central role in algebraic topology and higher category theory.
[^4]: The two equations in express nothing more than the fact that the algebra $\mathbb{C}$ is commutative, while the algebra $\End(H)$ of endomorphisms of a finite-dimensional Hilbert space $H$ is not.
[^5]: Weak dagger 2-categories are the obvious generalisation, with unitary associators and unitors.
[^6]: In , the last equation on the right is redundant, as it follows from the other equations; we include it for completeness.
[^7]: Here, we use $*$-cohomomorphisms instead of the more conventional $*$-homomorphisms of Corollary \[cor:finiteGelfandduality\] to obtain an equivalence with the category of finite sets and not with its opposite. This is a recurring theme throughout this work; we use copyable elements instead of the equivalent characters and comodules instead of modules.
[^8]: If we use a strict version of the monoidal category $\Hilb$, then $\QSet$ is a strict 2-category.
[^9]: All finite-dimensional $C^*$-algebras $A$ are *nuclear*; that is for all $C^*$-algebras $C$, there is a unique norm on the algebraic tensor product; the $C^*$-tensor product $C\otimes A$ is defined as its completion.
[^10]: A $C^*$-representation of $A$ is a $*$-homomorphism $A\to \mathcal{B}(H)$, where $\mathcal{B}(H)$ is the $C^*$-algebra of bounded operators on a Hilbert space $H$.
[^11]: It is straightforward to show directly that a $*$-homomorphism which is also a $*$-cohomomorphism is invertible. However, it is harder to show that an invertible $*$-cohomomorphism is a $*$-homomorphism; this may be a new result.
[^12]: The triangle equations given here are strict but can straightforwardly be weakened to apply to weak 2-categories.
[^13]: A *unitary error basis* is a basis of unitary operators on a finite-dimensional Hilbert space, orthogonal with respect to the trace inner product.
[^14]: In Section \[app:quantumrelation\], we show that these quantum graphs also arise from quantising the edge relation on the vertex set of a graph. This relational approach is closer to the work of previous authors, whereas the adjacency matrix approach has a stronger resemblance to classical graph theory.
[^15]: The quantum chromatic number of a graph $G$ is the smallest $n$ for which there exists a quantum graph homomorphism $G$ to $K_n$, the complete graph on $n$ vertices.
[^16]: More precisely, the pure $*$-homomorphism version of Stahlke’s entanglement assisted homomorphisms agrees with our quantum homomorphisms if the entangled resource is a maximally entangled state; or, in other words, if the positive operator $\Lambda$ used to define entanglement assisted homomorphisms (see [@Stahlke2016 Definition 15]) is the identity on some finite-dimensional Hilbert space $V$.
[^17]: An earlier, related but different definition is due to Bichon [@Bichon2003].
[^18]: It is clear that the tensor unit of $\QBij(A,A)$, i.e. the quantum bijection $\mathrm{id}_A:A\otimes \mathbb{C} \to \mathbb{C} \otimes A$, is simple. Therefore, $\QBij(A,A)$ is a tensor $*$-category as defined in [@Mueger2007 Definition 1.33].
[^19]: A projection $p$ in a $C^*$-algebra $A$ is an element $p\in A$ such that $p^*=p=p^2$.
[^20]: More precisely, it can be shown that if $H$ and $H'$ are finite-dimensional Hilbert spaces such that there are embeddings $A\subseteq \End(H)$ and $A\subseteq \End(H')$, then there is a canonical correspondence between the correspondingly defined quantum relations.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.